Update SwiftShader to April code dump.
April code dump from Transgaming. Adds new shader compiler.
diff --git a/src/LLVM/lib/Analysis/AliasAnalysis.cpp b/src/LLVM/lib/Analysis/AliasAnalysis.cpp
index 1d2488a..bd132c0 100644
--- a/src/LLVM/lib/Analysis/AliasAnalysis.cpp
+++ b/src/LLVM/lib/Analysis/AliasAnalysis.cpp
@@ -30,12 +30,13 @@
#include "llvm/Function.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/Instructions.h"
+#include "llvm/LLVMContext.h"
#include "llvm/Type.h"
#include "llvm/Target/TargetData.h"
using namespace llvm;
// Register the AliasAnalysis interface, providing a nice name to refer to.
-static RegisterAnalysisGroup<AliasAnalysis> Z("Alias Analysis");
+INITIALIZE_ANALYSIS_GROUP(AliasAnalysis, "Alias Analysis", NoAA)
char AliasAnalysis::ID = 0;
//===----------------------------------------------------------------------===//
@@ -43,15 +44,15 @@
//===----------------------------------------------------------------------===//
AliasAnalysis::AliasResult
-AliasAnalysis::alias(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
+AliasAnalysis::alias(const Location &LocA, const Location &LocB) {
assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
- return AA->alias(V1, V1Size, V2, V2Size);
+ return AA->alias(LocA, LocB);
}
-bool AliasAnalysis::pointsToConstantMemory(const Value *P) {
+bool AliasAnalysis::pointsToConstantMemory(const Location &Loc,
+ bool OrLocal) {
assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
- return AA->pointsToConstantMemory(P);
+ return AA->pointsToConstantMemory(Loc, OrLocal);
}
void AliasAnalysis::deleteValue(Value *V) {
@@ -64,49 +65,61 @@
AA->copyValue(From, To);
}
+void AliasAnalysis::addEscapingUse(Use &U) {
+ assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
+ AA->addEscapingUse(U);
+}
+
+
AliasAnalysis::ModRefResult
AliasAnalysis::getModRefInfo(ImmutableCallSite CS,
- const Value *P, unsigned Size) {
- // Don't assert AA because BasicAA calls us in order to make use of the
- // logic here.
+ const Location &Loc) {
+ assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
ModRefBehavior MRB = getModRefBehavior(CS);
if (MRB == DoesNotAccessMemory)
return NoModRef;
ModRefResult Mask = ModRef;
- if (MRB == OnlyReadsMemory)
+ if (onlyReadsMemory(MRB))
Mask = Ref;
- else if (MRB == AliasAnalysis::AccessesArguments) {
- bool doesAlias = false;
- for (ImmutableCallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
- AI != AE; ++AI)
- if (!isNoAlias(*AI, ~0U, P, Size)) {
- doesAlias = true;
- break;
- }
+ if (onlyAccessesArgPointees(MRB)) {
+ bool doesAlias = false;
+ if (doesAccessArgPointees(MRB)) {
+ MDNode *CSTag = CS.getInstruction()->getMetadata(LLVMContext::MD_tbaa);
+ for (ImmutableCallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
+ AI != AE; ++AI) {
+ const Value *Arg = *AI;
+ if (!Arg->getType()->isPointerTy())
+ continue;
+ Location CSLoc(Arg, UnknownSize, CSTag);
+ if (!isNoAlias(CSLoc, Loc)) {
+ doesAlias = true;
+ break;
+ }
+ }
+ }
if (!doesAlias)
return NoModRef;
}
- // If P points to a constant memory location, the call definitely could not
+ // If Loc is a constant memory location, the call definitely could not
// modify the memory location.
- if ((Mask & Mod) && pointsToConstantMemory(P))
+ if ((Mask & Mod) && pointsToConstantMemory(Loc))
Mask = ModRefResult(Mask & ~Mod);
- // If this is BasicAA, don't forward.
+ // If this is the end of the chain, don't forward.
if (!AA) return Mask;
// Otherwise, fall back to the next AA in the chain. But we can merge
// in any mask we've managed to compute.
- return ModRefResult(AA->getModRefInfo(CS, P, Size) & Mask);
+ return ModRefResult(AA->getModRefInfo(CS, Loc) & Mask);
}
AliasAnalysis::ModRefResult
AliasAnalysis::getModRefInfo(ImmutableCallSite CS1, ImmutableCallSite CS2) {
- // Don't assert AA because BasicAA calls us in order to make use of the
- // logic here.
+ assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
// If CS1 or CS2 are readnone, they don't interact.
ModRefBehavior CS1B = getModRefBehavior(CS1);
@@ -116,45 +129,60 @@
if (CS2B == DoesNotAccessMemory) return NoModRef;
// If they both only read from memory, there is no dependence.
- if (CS1B == OnlyReadsMemory && CS2B == OnlyReadsMemory)
+ if (onlyReadsMemory(CS1B) && onlyReadsMemory(CS2B))
return NoModRef;
AliasAnalysis::ModRefResult Mask = ModRef;
// If CS1 only reads memory, the only dependence on CS2 can be
// from CS1 reading memory written by CS2.
- if (CS1B == OnlyReadsMemory)
+ if (onlyReadsMemory(CS1B))
Mask = ModRefResult(Mask & Ref);
// If CS2 only access memory through arguments, accumulate the mod/ref
// information from CS1's references to the memory referenced by
// CS2's arguments.
- if (CS2B == AccessesArguments) {
+ if (onlyAccessesArgPointees(CS2B)) {
AliasAnalysis::ModRefResult R = NoModRef;
- for (ImmutableCallSite::arg_iterator
- I = CS2.arg_begin(), E = CS2.arg_end(); I != E; ++I) {
- R = ModRefResult((R | getModRefInfo(CS1, *I, UnknownSize)) & Mask);
- if (R == Mask)
- break;
+ if (doesAccessArgPointees(CS2B)) {
+ MDNode *CS2Tag = CS2.getInstruction()->getMetadata(LLVMContext::MD_tbaa);
+ for (ImmutableCallSite::arg_iterator
+ I = CS2.arg_begin(), E = CS2.arg_end(); I != E; ++I) {
+ const Value *Arg = *I;
+ if (!Arg->getType()->isPointerTy())
+ continue;
+ Location CS2Loc(Arg, UnknownSize, CS2Tag);
+ R = ModRefResult((R | getModRefInfo(CS1, CS2Loc)) & Mask);
+ if (R == Mask)
+ break;
+ }
}
return R;
}
// If CS1 only accesses memory through arguments, check if CS2 references
// any of the memory referenced by CS1's arguments. If not, return NoModRef.
- if (CS1B == AccessesArguments) {
+ if (onlyAccessesArgPointees(CS1B)) {
AliasAnalysis::ModRefResult R = NoModRef;
- for (ImmutableCallSite::arg_iterator
- I = CS1.arg_begin(), E = CS1.arg_end(); I != E; ++I)
- if (getModRefInfo(CS2, *I, UnknownSize) != NoModRef) {
- R = Mask;
- break;
+ if (doesAccessArgPointees(CS1B)) {
+ MDNode *CS1Tag = CS1.getInstruction()->getMetadata(LLVMContext::MD_tbaa);
+ for (ImmutableCallSite::arg_iterator
+ I = CS1.arg_begin(), E = CS1.arg_end(); I != E; ++I) {
+ const Value *Arg = *I;
+ if (!Arg->getType()->isPointerTy())
+ continue;
+ Location CS1Loc(Arg, UnknownSize, CS1Tag);
+ if (getModRefInfo(CS2, CS1Loc) != NoModRef) {
+ R = Mask;
+ break;
+ }
}
+ }
if (R == NoModRef)
return R;
}
- // If this is BasicAA, don't forward.
+ // If this is the end of the chain, don't forward.
if (!AA) return Mask;
// Otherwise, fall back to the next AA in the chain. But we can merge
@@ -164,8 +192,7 @@
AliasAnalysis::ModRefBehavior
AliasAnalysis::getModRefBehavior(ImmutableCallSite CS) {
- // Don't assert AA because BasicAA calls us in order to make use of the
- // logic here.
+ assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
ModRefBehavior Min = UnknownModRefBehavior;
@@ -174,12 +201,12 @@
if (const Function *F = CS.getCalledFunction())
Min = getModRefBehavior(F);
- // If this is BasicAA, don't forward.
+ // If this is the end of the chain, don't forward.
if (!AA) return Min;
// Otherwise, fall back to the next AA in the chain. But we can merge
// in any result we've managed to compute.
- return std::min(AA->getModRefBehavior(CS), Min);
+ return ModRefBehavior(AA->getModRefBehavior(CS) & Min);
}
AliasAnalysis::ModRefBehavior
@@ -188,20 +215,79 @@
return AA->getModRefBehavior(F);
}
-
//===----------------------------------------------------------------------===//
// AliasAnalysis non-virtual helper method implementation
//===----------------------------------------------------------------------===//
+AliasAnalysis::Location AliasAnalysis::getLocation(const LoadInst *LI) {
+ return Location(LI->getPointerOperand(),
+ getTypeStoreSize(LI->getType()),
+ LI->getMetadata(LLVMContext::MD_tbaa));
+}
+
+AliasAnalysis::Location AliasAnalysis::getLocation(const StoreInst *SI) {
+ return Location(SI->getPointerOperand(),
+ getTypeStoreSize(SI->getValueOperand()->getType()),
+ SI->getMetadata(LLVMContext::MD_tbaa));
+}
+
+AliasAnalysis::Location AliasAnalysis::getLocation(const VAArgInst *VI) {
+ return Location(VI->getPointerOperand(),
+ UnknownSize,
+ VI->getMetadata(LLVMContext::MD_tbaa));
+}
+
+AliasAnalysis::Location
+AliasAnalysis::getLocation(const AtomicCmpXchgInst *CXI) {
+ return Location(CXI->getPointerOperand(),
+ getTypeStoreSize(CXI->getCompareOperand()->getType()),
+ CXI->getMetadata(LLVMContext::MD_tbaa));
+}
+
+AliasAnalysis::Location
+AliasAnalysis::getLocation(const AtomicRMWInst *RMWI) {
+ return Location(RMWI->getPointerOperand(),
+ getTypeStoreSize(RMWI->getValOperand()->getType()),
+ RMWI->getMetadata(LLVMContext::MD_tbaa));
+}
+
+AliasAnalysis::Location
+AliasAnalysis::getLocationForSource(const MemTransferInst *MTI) {
+ uint64_t Size = UnknownSize;
+ if (ConstantInt *C = dyn_cast<ConstantInt>(MTI->getLength()))
+ Size = C->getValue().getZExtValue();
+
+ // memcpy/memmove can have TBAA tags. For memcpy, they apply
+ // to both the source and the destination.
+ MDNode *TBAATag = MTI->getMetadata(LLVMContext::MD_tbaa);
+
+ return Location(MTI->getRawSource(), Size, TBAATag);
+}
+
+AliasAnalysis::Location
+AliasAnalysis::getLocationForDest(const MemIntrinsic *MTI) {
+ uint64_t Size = UnknownSize;
+ if (ConstantInt *C = dyn_cast<ConstantInt>(MTI->getLength()))
+ Size = C->getValue().getZExtValue();
+
+ // memcpy/memmove can have TBAA tags. For memcpy, they apply
+ // to both the source and the destination.
+ MDNode *TBAATag = MTI->getMetadata(LLVMContext::MD_tbaa);
+
+ return Location(MTI->getRawDest(), Size, TBAATag);
+}
+
+
+
AliasAnalysis::ModRefResult
-AliasAnalysis::getModRefInfo(const LoadInst *L, const Value *P, unsigned Size) {
- // Be conservative in the face of volatile.
- if (L->isVolatile())
+AliasAnalysis::getModRefInfo(const LoadInst *L, const Location &Loc) {
+ // Be conservative in the face of volatile/atomic.
+ if (!L->isUnordered())
return ModRef;
// If the load address doesn't alias the given address, it doesn't read
// or write the specified memory.
- if (!alias(L->getOperand(0), getTypeStoreSize(L->getType()), P, Size))
+ if (!alias(getLocation(L), Loc))
return NoModRef;
// Otherwise, a load just reads.
@@ -209,33 +295,68 @@
}
AliasAnalysis::ModRefResult
-AliasAnalysis::getModRefInfo(const StoreInst *S, const Value *P, unsigned Size) {
- // Be conservative in the face of volatile.
- if (S->isVolatile())
+AliasAnalysis::getModRefInfo(const StoreInst *S, const Location &Loc) {
+ // Be conservative in the face of volatile/atomic.
+ if (!S->isUnordered())
return ModRef;
// If the store address cannot alias the pointer in question, then the
// specified memory cannot be modified by the store.
- if (!alias(S->getOperand(1),
- getTypeStoreSize(S->getOperand(0)->getType()), P, Size))
+ if (!alias(getLocation(S), Loc))
return NoModRef;
// If the pointer is a pointer to constant memory, then it could not have been
// modified by this store.
- if (pointsToConstantMemory(P))
+ if (pointsToConstantMemory(Loc))
return NoModRef;
// Otherwise, a store just writes.
return Mod;
}
-AliasAnalysis::ModRefBehavior
-AliasAnalysis::getIntrinsicModRefBehavior(unsigned iid) {
-#define GET_INTRINSIC_MODREF_BEHAVIOR
-#include "llvm/Intrinsics.gen"
-#undef GET_INTRINSIC_MODREF_BEHAVIOR
+AliasAnalysis::ModRefResult
+AliasAnalysis::getModRefInfo(const VAArgInst *V, const Location &Loc) {
+ // If the va_arg address cannot alias the pointer in question, then the
+ // specified memory cannot be accessed by the va_arg.
+ if (!alias(getLocation(V), Loc))
+ return NoModRef;
+
+ // If the pointer is a pointer to constant memory, then it could not have been
+ // modified by this va_arg.
+ if (pointsToConstantMemory(Loc))
+ return NoModRef;
+
+ // Otherwise, a va_arg reads and writes.
+ return ModRef;
}
+AliasAnalysis::ModRefResult
+AliasAnalysis::getModRefInfo(const AtomicCmpXchgInst *CX, const Location &Loc) {
+ // Acquire/Release cmpxchg has properties that matter for arbitrary addresses.
+ if (CX->getOrdering() > Monotonic)
+ return ModRef;
+
+ // If the cmpxchg address does not alias the location, it does not access it.
+ if (!alias(getLocation(CX), Loc))
+ return NoModRef;
+
+ return ModRef;
+}
+
+AliasAnalysis::ModRefResult
+AliasAnalysis::getModRefInfo(const AtomicRMWInst *RMW, const Location &Loc) {
+ // Acquire/Release atomicrmw has properties that matter for arbitrary addresses.
+ if (RMW->getOrdering() > Monotonic)
+ return ModRef;
+
+ // If the atomicrmw address does not alias the location, it does not access it.
+ if (!alias(getLocation(RMW), Loc))
+ return NoModRef;
+
+ return ModRef;
+}
+
+
// AliasAnalysis destructor: DO NOT move this to the header file for
// AliasAnalysis or else clients of the AliasAnalysis class may not depend on
// the AliasAnalysis.o file in the current .a file, causing alias analysis
@@ -260,16 +381,16 @@
/// getTypeStoreSize - Return the TargetData store size for the given type,
/// if known, or a conservative value otherwise.
///
-unsigned AliasAnalysis::getTypeStoreSize(const Type *Ty) {
- return TD ? TD->getTypeStoreSize(Ty) : ~0u;
+uint64_t AliasAnalysis::getTypeStoreSize(Type *Ty) {
+ return TD ? TD->getTypeStoreSize(Ty) : UnknownSize;
}
/// canBasicBlockModify - Return true if it is possible for execution of the
/// specified basic block to modify the value pointed to by Ptr.
///
bool AliasAnalysis::canBasicBlockModify(const BasicBlock &BB,
- const Value *Ptr, unsigned Size) {
- return canInstructionRangeModify(BB.front(), BB.back(), Ptr, Size);
+ const Location &Loc) {
+ return canInstructionRangeModify(BB.front(), BB.back(), Loc);
}
/// canInstructionRangeModify - Return true if it is possible for the execution
@@ -279,7 +400,7 @@
///
bool AliasAnalysis::canInstructionRangeModify(const Instruction &I1,
const Instruction &I2,
- const Value *Ptr, unsigned Size) {
+ const Location &Loc) {
assert(I1.getParent() == I2.getParent() &&
"Instructions not in same basic block!");
BasicBlock::const_iterator I = &I1;
@@ -287,7 +408,7 @@
++E; // Convert from inclusive to exclusive range.
for (; I != E; ++I) // Check every instruction in range
- if (getModRefInfo(I, Ptr, Size) & Mod)
+ if (getModRefInfo(I, Loc) & Mod)
return true;
return false;
}
@@ -295,7 +416,7 @@
/// isNoAliasCall - Return true if this pointer is returned by a noalias
/// function.
bool llvm::isNoAliasCall(const Value *V) {
- if (isa<CallInst>(V) || ISA_INVOKE_INST(V))
+ if (isa<CallInst>(V) || isa<InvokeInst>(V))
return ImmutableCallSite(cast<Instruction>(V))
.paramHasAttr(0, Attribute::NoAlias);
return false;
@@ -319,9 +440,3 @@
return A->hasNoAliasAttr() || A->hasByValAttr();
return false;
}
-
-// Because of the way .a files work, we must force the BasicAA implementation to
-// be pulled in if the AliasAnalysis classes are pulled in. Otherwise we run
-// the risk of AliasAnalysis being used, but the default implementation not
-// being linked into the tool that uses it.
-DEFINING_FILE_FOR(AliasAnalysis)
diff --git a/src/LLVM/lib/Analysis/AliasAnalysisCounter.cpp b/src/LLVM/lib/Analysis/AliasAnalysisCounter.cpp
new file mode 100644
index 0000000..d947220
--- /dev/null
+++ b/src/LLVM/lib/Analysis/AliasAnalysisCounter.cpp
@@ -0,0 +1,173 @@
+//===- AliasAnalysisCounter.cpp - Alias Analysis Query Counter ------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a pass which can be used to count how many alias queries
+// are being made and how the alias analysis implementation being used responds.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+static cl::opt<bool>
+PrintAll("count-aa-print-all-queries", cl::ReallyHidden, cl::init(true));
+static cl::opt<bool>
+PrintAllFailures("count-aa-print-all-failed-queries", cl::ReallyHidden);
+
+namespace {
+ class AliasAnalysisCounter : public ModulePass, public AliasAnalysis {
+ unsigned No, May, Partial, Must;
+ unsigned NoMR, JustRef, JustMod, MR;
+ Module *M;
+ public:
+ static char ID; // Class identification, replacement for typeinfo
+ AliasAnalysisCounter() : ModulePass(ID) {
+ initializeAliasAnalysisCounterPass(*PassRegistry::getPassRegistry());
+ No = May = Partial = Must = 0;
+ NoMR = JustRef = JustMod = MR = 0;
+ }
+
+ void printLine(const char *Desc, unsigned Val, unsigned Sum) {
+ errs() << " " << Val << " " << Desc << " responses ("
+ << Val*100/Sum << "%)\n";
+ }
+ ~AliasAnalysisCounter() {
+ unsigned AASum = No+May+Partial+Must;
+ unsigned MRSum = NoMR+JustRef+JustMod+MR;
+ if (AASum + MRSum) { // Print a report if any counted queries occurred...
+ errs() << "\n===== Alias Analysis Counter Report =====\n"
+ << " Analysis counted:\n"
+ << " " << AASum << " Total Alias Queries Performed\n";
+ if (AASum) {
+ printLine("no alias", No, AASum);
+ printLine("may alias", May, AASum);
+ printLine("partial alias", Partial, AASum);
+ printLine("must alias", Must, AASum);
+ errs() << " Alias Analysis Counter Summary: " << No*100/AASum << "%/"
+ << May*100/AASum << "%/"
+ << Partial*100/AASum << "%/"
+ << Must*100/AASum<<"%\n\n";
+ }
+
+ errs() << " " << MRSum << " Total Mod/Ref Queries Performed\n";
+ if (MRSum) {
+ printLine("no mod/ref", NoMR, MRSum);
+ printLine("ref", JustRef, MRSum);
+ printLine("mod", JustMod, MRSum);
+ printLine("mod/ref", MR, MRSum);
+ errs() << " Mod/Ref Analysis Counter Summary: " <<NoMR*100/MRSum
+ << "%/" << JustRef*100/MRSum << "%/" << JustMod*100/MRSum
+ << "%/" << MR*100/MRSum <<"%\n\n";
+ }
+ }
+ }
+
+ bool runOnModule(Module &M) {
+ this->M = &M;
+ InitializeAliasAnalysis(this);
+ return false;
+ }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AliasAnalysis::getAnalysisUsage(AU);
+ AU.addRequired<AliasAnalysis>();
+ AU.setPreservesAll();
+ }
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ virtual void *getAdjustedAnalysisPointer(AnalysisID PI) {
+ if (PI == &AliasAnalysis::ID)
+ return (AliasAnalysis*)this;
+ return this;
+ }
+
+ // FIXME: We could count these too...
+ bool pointsToConstantMemory(const Location &Loc, bool OrLocal) {
+ return getAnalysis<AliasAnalysis>().pointsToConstantMemory(Loc, OrLocal);
+ }
+
+ // Forwarding functions: just delegate to a real AA implementation, counting
+ // the number of responses...
+ AliasResult alias(const Location &LocA, const Location &LocB);
+
+ ModRefResult getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc);
+ ModRefResult getModRefInfo(ImmutableCallSite CS1,
+ ImmutableCallSite CS2) {
+ return AliasAnalysis::getModRefInfo(CS1,CS2);
+ }
+ };
+}
+
+char AliasAnalysisCounter::ID = 0;
+INITIALIZE_AG_PASS(AliasAnalysisCounter, AliasAnalysis, "count-aa",
+ "Count Alias Analysis Query Responses", false, true, false)
+
+ModulePass *llvm::createAliasAnalysisCounterPass() {
+ return new AliasAnalysisCounter();
+}
+
+AliasAnalysis::AliasResult
+AliasAnalysisCounter::alias(const Location &LocA, const Location &LocB) {
+ AliasResult R = getAnalysis<AliasAnalysis>().alias(LocA, LocB);
+
+ const char *AliasString;
+ switch (R) {
+ default: llvm_unreachable("Unknown alias type!");
+ case NoAlias: No++; AliasString = "No alias"; break;
+ case MayAlias: May++; AliasString = "May alias"; break;
+ case PartialAlias: Partial++; AliasString = "Partial alias"; break;
+ case MustAlias: Must++; AliasString = "Must alias"; break;
+ }
+
+ if (PrintAll || (PrintAllFailures && R == MayAlias)) {
+ errs() << AliasString << ":\t";
+ errs() << "[" << LocA.Size << "B] ";
+ WriteAsOperand(errs(), LocA.Ptr, true, M);
+ errs() << ", ";
+ errs() << "[" << LocB.Size << "B] ";
+ WriteAsOperand(errs(), LocB.Ptr, true, M);
+ errs() << "\n";
+ }
+
+ return R;
+}
+
+AliasAnalysis::ModRefResult
+AliasAnalysisCounter::getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc) {
+ ModRefResult R = getAnalysis<AliasAnalysis>().getModRefInfo(CS, Loc);
+
+ const char *MRString;
+ switch (R) {
+ default: llvm_unreachable("Unknown mod/ref type!");
+ case NoModRef: NoMR++; MRString = "NoModRef"; break;
+ case Ref: JustRef++; MRString = "JustRef"; break;
+ case Mod: JustMod++; MRString = "JustMod"; break;
+ case ModRef: MR++; MRString = "ModRef"; break;
+ }
+
+ if (PrintAll || (PrintAllFailures && R == ModRef)) {
+ errs() << MRString << ": Ptr: ";
+ errs() << "[" << Loc.Size << "B] ";
+ WriteAsOperand(errs(), Loc.Ptr, true, M);
+ errs() << "\t<->" << *CS.getInstruction() << '\n';
+ }
+ return R;
+}
diff --git a/src/LLVM/lib/Analysis/AliasAnalysisEvaluator.cpp b/src/LLVM/lib/Analysis/AliasAnalysisEvaluator.cpp
new file mode 100644
index 0000000..37271b9
--- /dev/null
+++ b/src/LLVM/lib/Analysis/AliasAnalysisEvaluator.cpp
@@ -0,0 +1,304 @@
+//===- AliasAnalysisEvaluator.cpp - Alias Analysis Accuracy Evaluator -----===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a simple N^2 alias analysis accuracy evaluator.
+// Basically, for each function in the program, it simply queries to see how the
+// alias analysis implementation answers alias queries between each pair of
+// pointers in the function.
+//
+// This is inspired and adapted from code by: Naveen Neelakantam, Francesco
+// Spadini, and Wojciech Stryjewski.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Constants.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Function.h"
+#include "llvm/Instructions.h"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/InstIterator.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/SetVector.h"
+using namespace llvm;
+
+static cl::opt<bool> PrintAll("print-all-alias-modref-info", cl::ReallyHidden);
+
+static cl::opt<bool> PrintNoAlias("print-no-aliases", cl::ReallyHidden);
+static cl::opt<bool> PrintMayAlias("print-may-aliases", cl::ReallyHidden);
+static cl::opt<bool> PrintPartialAlias("print-partial-aliases", cl::ReallyHidden);
+static cl::opt<bool> PrintMustAlias("print-must-aliases", cl::ReallyHidden);
+
+static cl::opt<bool> PrintNoModRef("print-no-modref", cl::ReallyHidden);
+static cl::opt<bool> PrintMod("print-mod", cl::ReallyHidden);
+static cl::opt<bool> PrintRef("print-ref", cl::ReallyHidden);
+static cl::opt<bool> PrintModRef("print-modref", cl::ReallyHidden);
+
+namespace {
+ class AAEval : public FunctionPass {
+ unsigned NoAlias, MayAlias, PartialAlias, MustAlias;
+ unsigned NoModRef, Mod, Ref, ModRef;
+
+ public:
+ static char ID; // Pass identification, replacement for typeid
+ AAEval() : FunctionPass(ID) {
+ initializeAAEvalPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.addRequired<AliasAnalysis>();
+ AU.setPreservesAll();
+ }
+
+ bool doInitialization(Module &M) {
+ NoAlias = MayAlias = PartialAlias = MustAlias = 0;
+ NoModRef = Mod = Ref = ModRef = 0;
+
+ if (PrintAll) {
+ PrintNoAlias = PrintMayAlias = true;
+ PrintPartialAlias = PrintMustAlias = true;
+ PrintNoModRef = PrintMod = PrintRef = PrintModRef = true;
+ }
+ return false;
+ }
+
+ bool runOnFunction(Function &F);
+ bool doFinalization(Module &M);
+ };
+}
+
+char AAEval::ID = 0;
+INITIALIZE_PASS_BEGIN(AAEval, "aa-eval",
+ "Exhaustive Alias Analysis Precision Evaluator", false, true)
+INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_END(AAEval, "aa-eval",
+ "Exhaustive Alias Analysis Precision Evaluator", false, true)
+
+FunctionPass *llvm::createAAEvalPass() { return new AAEval(); }
+
+static void PrintResults(const char *Msg, bool P, const Value *V1,
+ const Value *V2, const Module *M) {
+ if (P) {
+ std::string o1, o2;
+ {
+ raw_string_ostream os1(o1), os2(o2);
+ WriteAsOperand(os1, V1, true, M);
+ WriteAsOperand(os2, V2, true, M);
+ }
+
+ if (o2 < o1)
+ std::swap(o1, o2);
+ errs() << " " << Msg << ":\t"
+ << o1 << ", "
+ << o2 << "\n";
+ }
+}
+
+static inline void
+PrintModRefResults(const char *Msg, bool P, Instruction *I, Value *Ptr,
+ Module *M) {
+ if (P) {
+ errs() << " " << Msg << ": Ptr: ";
+ WriteAsOperand(errs(), Ptr, true, M);
+ errs() << "\t<->" << *I << '\n';
+ }
+}
+
+static inline void
+PrintModRefResults(const char *Msg, bool P, CallSite CSA, CallSite CSB,
+ Module *M) {
+ if (P) {
+ errs() << " " << Msg << ": " << *CSA.getInstruction()
+ << " <-> " << *CSB.getInstruction() << '\n';
+ }
+}
+
+static inline bool isInterestingPointer(Value *V) {
+ return V->getType()->isPointerTy()
+ && !isa<ConstantPointerNull>(V);
+}
+
+bool AAEval::runOnFunction(Function &F) {
+ AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
+
+ SetVector<Value *> Pointers;
+ SetVector<CallSite> CallSites;
+
+ for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E; ++I)
+ if (I->getType()->isPointerTy()) // Add all pointer arguments.
+ Pointers.insert(I);
+
+ for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) {
+ if (I->getType()->isPointerTy()) // Add all pointer instructions.
+ Pointers.insert(&*I);
+ Instruction &Inst = *I;
+ if (CallSite CS = cast<Value>(&Inst)) {
+ Value *Callee = CS.getCalledValue();
+ // Skip actual functions for direct function calls.
+ if (!isa<Function>(Callee) && isInterestingPointer(Callee))
+ Pointers.insert(Callee);
+ // Consider formals.
+ for (CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
+ AI != AE; ++AI)
+ if (isInterestingPointer(*AI))
+ Pointers.insert(*AI);
+ CallSites.insert(CS);
+ } else {
+ // Consider all operands.
+ for (Instruction::op_iterator OI = Inst.op_begin(), OE = Inst.op_end();
+ OI != OE; ++OI)
+ if (isInterestingPointer(*OI))
+ Pointers.insert(*OI);
+ }
+ }
+
+ if (PrintNoAlias || PrintMayAlias || PrintPartialAlias || PrintMustAlias ||
+ PrintNoModRef || PrintMod || PrintRef || PrintModRef)
+ errs() << "Function: " << F.getName() << ": " << Pointers.size()
+ << " pointers, " << CallSites.size() << " call sites\n";
+
+ // iterate over the worklist, and run the full (n^2)/2 disambiguations
+ for (SetVector<Value *>::iterator I1 = Pointers.begin(), E = Pointers.end();
+ I1 != E; ++I1) {
+ uint64_t I1Size = AliasAnalysis::UnknownSize;
+ Type *I1ElTy = cast<PointerType>((*I1)->getType())->getElementType();
+ if (I1ElTy->isSized()) I1Size = AA.getTypeStoreSize(I1ElTy);
+
+ for (SetVector<Value *>::iterator I2 = Pointers.begin(); I2 != I1; ++I2) {
+ uint64_t I2Size = AliasAnalysis::UnknownSize;
+ Type *I2ElTy =cast<PointerType>((*I2)->getType())->getElementType();
+ if (I2ElTy->isSized()) I2Size = AA.getTypeStoreSize(I2ElTy);
+
+ switch (AA.alias(*I1, I1Size, *I2, I2Size)) {
+ case AliasAnalysis::NoAlias:
+ PrintResults("NoAlias", PrintNoAlias, *I1, *I2, F.getParent());
+ ++NoAlias; break;
+ case AliasAnalysis::MayAlias:
+ PrintResults("MayAlias", PrintMayAlias, *I1, *I2, F.getParent());
+ ++MayAlias; break;
+ case AliasAnalysis::PartialAlias:
+ PrintResults("PartialAlias", PrintPartialAlias, *I1, *I2,
+ F.getParent());
+ ++PartialAlias; break;
+ case AliasAnalysis::MustAlias:
+ PrintResults("MustAlias", PrintMustAlias, *I1, *I2, F.getParent());
+ ++MustAlias; break;
+ default:
+ errs() << "Unknown alias query result!\n";
+ }
+ }
+ }
+
+ // Mod/ref alias analysis: compare all pairs of calls and values
+ for (SetVector<CallSite>::iterator C = CallSites.begin(),
+ Ce = CallSites.end(); C != Ce; ++C) {
+ Instruction *I = C->getInstruction();
+
+ for (SetVector<Value *>::iterator V = Pointers.begin(), Ve = Pointers.end();
+ V != Ve; ++V) {
+ uint64_t Size = AliasAnalysis::UnknownSize;
+ Type *ElTy = cast<PointerType>((*V)->getType())->getElementType();
+ if (ElTy->isSized()) Size = AA.getTypeStoreSize(ElTy);
+
+ switch (AA.getModRefInfo(*C, *V, Size)) {
+ case AliasAnalysis::NoModRef:
+ PrintModRefResults("NoModRef", PrintNoModRef, I, *V, F.getParent());
+ ++NoModRef; break;
+ case AliasAnalysis::Mod:
+ PrintModRefResults("Just Mod", PrintMod, I, *V, F.getParent());
+ ++Mod; break;
+ case AliasAnalysis::Ref:
+ PrintModRefResults("Just Ref", PrintRef, I, *V, F.getParent());
+ ++Ref; break;
+ case AliasAnalysis::ModRef:
+ PrintModRefResults("Both ModRef", PrintModRef, I, *V, F.getParent());
+ ++ModRef; break;
+ default:
+ errs() << "Unknown alias query result!\n";
+ }
+ }
+ }
+
+ // Mod/ref alias analysis: compare all pairs of calls
+ for (SetVector<CallSite>::iterator C = CallSites.begin(),
+ Ce = CallSites.end(); C != Ce; ++C) {
+ for (SetVector<CallSite>::iterator D = CallSites.begin(); D != Ce; ++D) {
+ if (D == C)
+ continue;
+ switch (AA.getModRefInfo(*C, *D)) {
+ case AliasAnalysis::NoModRef:
+ PrintModRefResults("NoModRef", PrintNoModRef, *C, *D, F.getParent());
+ ++NoModRef; break;
+ case AliasAnalysis::Mod:
+ PrintModRefResults("Just Mod", PrintMod, *C, *D, F.getParent());
+ ++Mod; break;
+ case AliasAnalysis::Ref:
+ PrintModRefResults("Just Ref", PrintRef, *C, *D, F.getParent());
+ ++Ref; break;
+ case AliasAnalysis::ModRef:
+ PrintModRefResults("Both ModRef", PrintModRef, *C, *D, F.getParent());
+ ++ModRef; break;
+ }
+ }
+ }
+
+ return false;
+}
+
+static void PrintPercent(unsigned Num, unsigned Sum) {
+ errs() << "(" << Num*100ULL/Sum << "."
+ << ((Num*1000ULL/Sum) % 10) << "%)\n";
+}
+
+bool AAEval::doFinalization(Module &M) {
+ unsigned AliasSum = NoAlias + MayAlias + PartialAlias + MustAlias;
+ errs() << "===== Alias Analysis Evaluator Report =====\n";
+ if (AliasSum == 0) {
+ errs() << " Alias Analysis Evaluator Summary: No pointers!\n";
+ } else {
+ errs() << " " << AliasSum << " Total Alias Queries Performed\n";
+ errs() << " " << NoAlias << " no alias responses ";
+ PrintPercent(NoAlias, AliasSum);
+ errs() << " " << MayAlias << " may alias responses ";
+ PrintPercent(MayAlias, AliasSum);
+ errs() << " " << PartialAlias << " partial alias responses ";
+ PrintPercent(PartialAlias, AliasSum);
+ errs() << " " << MustAlias << " must alias responses ";
+ PrintPercent(MustAlias, AliasSum);
+ errs() << " Alias Analysis Evaluator Pointer Alias Summary: "
+ << NoAlias*100/AliasSum << "%/" << MayAlias*100/AliasSum << "%/"
+ << PartialAlias*100/AliasSum << "%/"
+ << MustAlias*100/AliasSum << "%\n";
+ }
+
+ // Display the summary for mod/ref analysis
+ unsigned ModRefSum = NoModRef + Mod + Ref + ModRef;
+ if (ModRefSum == 0) {
+ errs() << " Alias Analysis Mod/Ref Evaluator Summary: no mod/ref!\n";
+ } else {
+ errs() << " " << ModRefSum << " Total ModRef Queries Performed\n";
+ errs() << " " << NoModRef << " no mod/ref responses ";
+ PrintPercent(NoModRef, ModRefSum);
+ errs() << " " << Mod << " mod responses ";
+ PrintPercent(Mod, ModRefSum);
+ errs() << " " << Ref << " ref responses ";
+ PrintPercent(Ref, ModRefSum);
+ errs() << " " << ModRef << " mod & ref responses ";
+ PrintPercent(ModRef, ModRefSum);
+ errs() << " Alias Analysis Evaluator Mod/Ref Summary: "
+ << NoModRef*100/ModRefSum << "%/" << Mod*100/ModRefSum << "%/"
+ << Ref*100/ModRefSum << "%/" << ModRef*100/ModRefSum << "%\n";
+ }
+
+ return false;
+}
diff --git a/src/LLVM/lib/Analysis/AliasDebugger.cpp b/src/LLVM/lib/Analysis/AliasDebugger.cpp
new file mode 100644
index 0000000..f15c051
--- /dev/null
+++ b/src/LLVM/lib/Analysis/AliasDebugger.cpp
@@ -0,0 +1,138 @@
+//===- AliasDebugger.cpp - Simple Alias Analysis Use Checker --------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This simple pass checks alias analysis users to ensure that if they
+// create a new value, they do not query AA without informing it of the value.
+// It acts as a shim over any other AA pass you want.
+//
+// Yes keeping track of every value in the program is expensive, but this is
+// a debugging pass.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Module.h"
+#include "llvm/Pass.h"
+#include "llvm/Instructions.h"
+#include "llvm/Constants.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include <set>
+using namespace llvm;
+
+namespace {
+
+ class AliasDebugger : public ModulePass, public AliasAnalysis {
+
+ //What we do is simple. Keep track of every value the AA could
+ //know about, and verify that queries are one of those.
+ //A query to a value that didn't exist when the AA was created
+ //means someone forgot to update the AA when creating new values
+
+ std::set<const Value*> Vals;
+
+ public:
+ static char ID; // Class identification, replacement for typeinfo
+ AliasDebugger() : ModulePass(ID) {
+ initializeAliasDebuggerPass(*PassRegistry::getPassRegistry());
+ }
+
+ bool runOnModule(Module &M) {
+ InitializeAliasAnalysis(this); // set up super class
+
+ for(Module::global_iterator I = M.global_begin(),
+ E = M.global_end(); I != E; ++I) {
+ Vals.insert(&*I);
+ for (User::const_op_iterator OI = I->op_begin(),
+ OE = I->op_end(); OI != OE; ++OI)
+ Vals.insert(*OI);
+ }
+
+ for(Module::iterator I = M.begin(),
+ E = M.end(); I != E; ++I){
+ Vals.insert(&*I);
+ if(!I->isDeclaration()) {
+ for (Function::arg_iterator AI = I->arg_begin(), AE = I->arg_end();
+ AI != AE; ++AI)
+ Vals.insert(&*AI);
+ for (Function::const_iterator FI = I->begin(), FE = I->end();
+ FI != FE; ++FI)
+ for (BasicBlock::const_iterator BI = FI->begin(), BE = FI->end();
+ BI != BE; ++BI) {
+ Vals.insert(&*BI);
+ for (User::const_op_iterator OI = BI->op_begin(),
+ OE = BI->op_end(); OI != OE; ++OI)
+ Vals.insert(*OI);
+ }
+ }
+
+ }
+ return false;
+ }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AliasAnalysis::getAnalysisUsage(AU);
+ AU.setPreservesAll(); // Does not transform code
+ }
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ virtual void *getAdjustedAnalysisPointer(AnalysisID PI) {
+ if (PI == &AliasAnalysis::ID)
+ return (AliasAnalysis*)this;
+ return this;
+ }
+
+ //------------------------------------------------
+ // Implement the AliasAnalysis API
+ //
+ AliasResult alias(const Location &LocA, const Location &LocB) {
+ assert(Vals.find(LocA.Ptr) != Vals.end() &&
+ "Never seen value in AA before");
+ assert(Vals.find(LocB.Ptr) != Vals.end() &&
+ "Never seen value in AA before");
+ return AliasAnalysis::alias(LocA, LocB);
+ }
+
+ ModRefResult getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc) {
+ assert(Vals.find(Loc.Ptr) != Vals.end() && "Never seen value in AA before");
+ return AliasAnalysis::getModRefInfo(CS, Loc);
+ }
+
+ ModRefResult getModRefInfo(ImmutableCallSite CS1,
+ ImmutableCallSite CS2) {
+ return AliasAnalysis::getModRefInfo(CS1,CS2);
+ }
+
+ bool pointsToConstantMemory(const Location &Loc, bool OrLocal) {
+ assert(Vals.find(Loc.Ptr) != Vals.end() && "Never seen value in AA before");
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+ }
+
+ virtual void deleteValue(Value *V) {
+ assert(Vals.find(V) != Vals.end() && "Never seen value in AA before");
+ AliasAnalysis::deleteValue(V);
+ }
+ virtual void copyValue(Value *From, Value *To) {
+ Vals.insert(To);
+ AliasAnalysis::copyValue(From, To);
+ }
+
+ };
+}
+
+char AliasDebugger::ID = 0;
+INITIALIZE_AG_PASS(AliasDebugger, AliasAnalysis, "debug-aa",
+ "AA use debugger", false, true, false)
+
+Pass *llvm::createAliasDebugger() { return new AliasDebugger(); }
+
diff --git a/src/LLVM/lib/Analysis/AliasSetTracker.cpp b/src/LLVM/lib/Analysis/AliasSetTracker.cpp
index 87d6fe5..3fcd3b5 100644
--- a/src/LLVM/lib/Analysis/AliasSetTracker.cpp
+++ b/src/LLVM/lib/Analysis/AliasSetTracker.cpp
@@ -15,6 +15,7 @@
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
+#include "llvm/LLVMContext.h"
#include "llvm/Pass.h"
#include "llvm/Type.h"
#include "llvm/Target/TargetData.h"
@@ -22,7 +23,6 @@
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/InstIterator.h"
-#include "llvm/Support/Format.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
@@ -35,6 +35,7 @@
// Update the alias and access types of this set...
AccessTy |= AS.AccessTy;
AliasTy |= AS.AliasTy;
+ Volatile |= AS.Volatile;
if (AliasTy == MustAlias) {
// Check that these two merged sets really are must aliases. Since both
@@ -45,17 +46,22 @@
PointerRec *R = AS.getSomePointer();
// If the pointers are not a must-alias pair, this set becomes a may alias.
- if (AA.alias(L->getValue(), L->getSize(), R->getValue(), R->getSize())
+ if (AA.alias(AliasAnalysis::Location(L->getValue(),
+ L->getSize(),
+ L->getTBAAInfo()),
+ AliasAnalysis::Location(R->getValue(),
+ R->getSize(),
+ R->getTBAAInfo()))
!= AliasAnalysis::MustAlias)
AliasTy = MayAlias;
}
- if (CallSites.empty()) { // Merge call sites...
- if (!AS.CallSites.empty())
- std::swap(CallSites, AS.CallSites);
- } else if (!AS.CallSites.empty()) {
- CallSites.insert(CallSites.end(), AS.CallSites.begin(), AS.CallSites.end());
- AS.CallSites.clear();
+ if (UnknownInsts.empty()) { // Merge call sites...
+ if (!AS.UnknownInsts.empty())
+ std::swap(UnknownInsts, AS.UnknownInsts);
+ } else if (!AS.UnknownInsts.empty()) {
+ UnknownInsts.insert(UnknownInsts.end(), AS.UnknownInsts.begin(), AS.UnknownInsts.end());
+ AS.UnknownInsts.clear();
}
AS.Forward = this; // Forward across AS now...
@@ -87,7 +93,8 @@
}
void AliasSet::addPointer(AliasSetTracker &AST, PointerRec &Entry,
- unsigned Size, bool KnownMustAlias) {
+ uint64_t Size, const MDNode *TBAAInfo,
+ bool KnownMustAlias) {
assert(!Entry.hasAliasSet() && "Entry already in set!");
// Check to see if we have to downgrade to _may_ alias.
@@ -95,32 +102,31 @@
if (PointerRec *P = getSomePointer()) {
AliasAnalysis &AA = AST.getAliasAnalysis();
AliasAnalysis::AliasResult Result =
- AA.alias(P->getValue(), P->getSize(), Entry.getValue(), Size);
- if (Result == AliasAnalysis::MayAlias)
+ AA.alias(AliasAnalysis::Location(P->getValue(), P->getSize(),
+ P->getTBAAInfo()),
+ AliasAnalysis::Location(Entry.getValue(), Size, TBAAInfo));
+ if (Result != AliasAnalysis::MustAlias)
AliasTy = MayAlias;
else // First entry of must alias must have maximum size!
- P->updateSize(Size);
+ P->updateSizeAndTBAAInfo(Size, TBAAInfo);
assert(Result != AliasAnalysis::NoAlias && "Cannot be part of must set!");
}
Entry.setAliasSet(this);
- Entry.updateSize(Size);
+ Entry.updateSizeAndTBAAInfo(Size, TBAAInfo);
// Add it to the end of the list...
assert(*PtrListEnd == 0 && "End of list is not null?");
*PtrListEnd = &Entry;
PtrListEnd = Entry.setPrevInList(PtrListEnd);
assert(*PtrListEnd == 0 && "End of list is not null?");
- addRef(); // Entry points to alias set...
+ addRef(); // Entry points to alias set.
}
-void AliasSet::addCallSite(CallSite CS, AliasAnalysis &AA) {
- CallSites.push_back(CS);
+void AliasSet::addUnknownInst(Instruction *I, AliasAnalysis &AA) {
+ UnknownInsts.push_back(I);
- AliasAnalysis::ModRefBehavior Behavior = AA.getModRefBehavior(CS);
- if (Behavior == AliasAnalysis::DoesNotAccessMemory)
- return;
- else if (Behavior == AliasAnalysis::OnlyReadsMemory) {
+ if (!I->mayWriteToMemory()) {
AliasTy = MayAlias;
AccessTy |= Refs;
return;
@@ -134,46 +140,56 @@
/// aliasesPointer - Return true if the specified pointer "may" (or must)
/// alias one of the members in the set.
///
-bool AliasSet::aliasesPointer(const Value *Ptr, unsigned Size,
+bool AliasSet::aliasesPointer(const Value *Ptr, uint64_t Size,
+ const MDNode *TBAAInfo,
AliasAnalysis &AA) const {
if (AliasTy == MustAlias) {
- assert(CallSites.empty() && "Illegal must alias set!");
+ assert(UnknownInsts.empty() && "Illegal must alias set!");
// If this is a set of MustAliases, only check to see if the pointer aliases
- // SOME value in the set...
+ // SOME value in the set.
PointerRec *SomePtr = getSomePointer();
assert(SomePtr && "Empty must-alias set??");
- return AA.alias(SomePtr->getValue(), SomePtr->getSize(), Ptr, Size);
+ return AA.alias(AliasAnalysis::Location(SomePtr->getValue(),
+ SomePtr->getSize(),
+ SomePtr->getTBAAInfo()),
+ AliasAnalysis::Location(Ptr, Size, TBAAInfo));
}
// If this is a may-alias set, we have to check all of the pointers in the set
// to be sure it doesn't alias the set...
for (iterator I = begin(), E = end(); I != E; ++I)
- if (AA.alias(Ptr, Size, I.getPointer(), I.getSize()))
+ if (AA.alias(AliasAnalysis::Location(Ptr, Size, TBAAInfo),
+ AliasAnalysis::Location(I.getPointer(), I.getSize(),
+ I.getTBAAInfo())))
return true;
- // Check the call sites list and invoke list...
- if (!CallSites.empty()) {
- for (unsigned i = 0, e = CallSites.size(); i != e; ++i)
- if (AA.getModRefInfo(CallSites[i], Ptr, Size)
- != AliasAnalysis::NoModRef)
+ // Check the unknown instructions...
+ if (!UnknownInsts.empty()) {
+ for (unsigned i = 0, e = UnknownInsts.size(); i != e; ++i)
+ if (AA.getModRefInfo(UnknownInsts[i],
+ AliasAnalysis::Location(Ptr, Size, TBAAInfo)) !=
+ AliasAnalysis::NoModRef)
return true;
}
return false;
}
-bool AliasSet::aliasesCallSite(CallSite CS, AliasAnalysis &AA) const {
- if (AA.doesNotAccessMemory(CS))
+bool AliasSet::aliasesUnknownInst(Instruction *Inst, AliasAnalysis &AA) const {
+ if (!Inst->mayReadOrWriteMemory())
return false;
- for (unsigned i = 0, e = CallSites.size(); i != e; ++i)
- if (AA.getModRefInfo(CallSites[i], CS) != AliasAnalysis::NoModRef ||
- AA.getModRefInfo(CS, CallSites[i]) != AliasAnalysis::NoModRef)
+ for (unsigned i = 0, e = UnknownInsts.size(); i != e; ++i) {
+ CallSite C1 = getUnknownInst(i), C2 = Inst;
+ if (!C1 || !C2 ||
+ AA.getModRefInfo(C1, C2) != AliasAnalysis::NoModRef ||
+ AA.getModRefInfo(C2, C1) != AliasAnalysis::NoModRef)
return true;
+ }
for (iterator I = begin(), E = end(); I != E; ++I)
- if (AA.getModRefInfo(CS, I.getPointer(), I.getSize()) !=
+ if (AA.getModRefInfo(Inst, I.getPointer(), I.getSize()) !=
AliasAnalysis::NoModRef)
return true;
@@ -198,16 +214,18 @@
/// that may alias the pointer, merge them together and return the unified set.
///
AliasSet *AliasSetTracker::findAliasSetForPointer(const Value *Ptr,
- unsigned Size) {
+ uint64_t Size,
+ const MDNode *TBAAInfo) {
AliasSet *FoundSet = 0;
- for (iterator I = begin(), E = end(); I != E; ++I)
- if (!I->Forward && I->aliasesPointer(Ptr, Size, AA)) {
- if (FoundSet == 0) { // If this is the first alias set ptr can go into.
- FoundSet = I; // Remember it.
- } else { // Otherwise, we must merge the sets.
- FoundSet->mergeSetIn(*I, *this); // Merge in contents.
- }
+ for (iterator I = begin(), E = end(); I != E; ++I) {
+ if (I->Forward || !I->aliasesPointer(Ptr, Size, TBAAInfo, AA)) continue;
+
+ if (FoundSet == 0) { // If this is the first alias set ptr can go into.
+ FoundSet = I; // Remember it.
+ } else { // Otherwise, we must merge the sets.
+ FoundSet->mergeSetIn(*I, *this); // Merge in contents.
}
+ }
return FoundSet;
}
@@ -215,26 +233,27 @@
/// containsPointer - Return true if the specified location is represented by
/// this alias set, false otherwise. This does not modify the AST object or
/// alias sets.
-bool AliasSetTracker::containsPointer(Value *Ptr, unsigned Size) const {
+bool AliasSetTracker::containsPointer(Value *Ptr, uint64_t Size,
+ const MDNode *TBAAInfo) const {
for (const_iterator I = begin(), E = end(); I != E; ++I)
- if (!I->Forward && I->aliasesPointer(Ptr, Size, AA))
+ if (!I->Forward && I->aliasesPointer(Ptr, Size, TBAAInfo, AA))
return true;
return false;
}
-AliasSet *AliasSetTracker::findAliasSetForCallSite(CallSite CS) {
+AliasSet *AliasSetTracker::findAliasSetForUnknownInst(Instruction *Inst) {
AliasSet *FoundSet = 0;
- for (iterator I = begin(), E = end(); I != E; ++I)
- if (!I->Forward && I->aliasesCallSite(CS, AA)) {
- if (FoundSet == 0) { // If this is the first alias set ptr can go into.
- FoundSet = I; // Remember it.
- } else if (!I->Forward) { // Otherwise, we must merge the sets.
- FoundSet->mergeSetIn(*I, *this); // Merge in contents.
- }
- }
-
+ for (iterator I = begin(), E = end(); I != E; ++I) {
+ if (I->Forward || !I->aliasesUnknownInst(Inst, AA))
+ continue;
+
+ if (FoundSet == 0) // If this is the first alias set ptr can go into.
+ FoundSet = I; // Remember it.
+ else if (!I->Forward) // Otherwise, we must merge the sets.
+ FoundSet->mergeSetIn(*I, *this); // Merge in contents.
+ }
return FoundSet;
}
@@ -243,81 +262,100 @@
/// getAliasSetForPointer - Return the alias set that the specified pointer
/// lives in.
-AliasSet &AliasSetTracker::getAliasSetForPointer(Value *Pointer, unsigned Size,
+AliasSet &AliasSetTracker::getAliasSetForPointer(Value *Pointer, uint64_t Size,
+ const MDNode *TBAAInfo,
bool *New) {
AliasSet::PointerRec &Entry = getEntryFor(Pointer);
- // Check to see if the pointer is already known...
+ // Check to see if the pointer is already known.
if (Entry.hasAliasSet()) {
- Entry.updateSize(Size);
+ Entry.updateSizeAndTBAAInfo(Size, TBAAInfo);
// Return the set!
return *Entry.getAliasSet(*this)->getForwardedTarget(*this);
- } else if (AliasSet *AS = findAliasSetForPointer(Pointer, Size)) {
- // Add it to the alias set it aliases...
- AS->addPointer(*this, Entry, Size);
- return *AS;
- } else {
- if (New) *New = true;
- // Otherwise create a new alias set to hold the loaded pointer...
- AliasSets.push_back(new AliasSet());
- AliasSets.back().addPointer(*this, Entry, Size);
- return AliasSets.back();
}
+
+ if (AliasSet *AS = findAliasSetForPointer(Pointer, Size, TBAAInfo)) {
+ // Add it to the alias set it aliases.
+ AS->addPointer(*this, Entry, Size, TBAAInfo);
+ return *AS;
+ }
+
+ if (New) *New = true;
+ // Otherwise create a new alias set to hold the loaded pointer.
+ AliasSets.push_back(new AliasSet());
+ AliasSets.back().addPointer(*this, Entry, Size, TBAAInfo);
+ return AliasSets.back();
}
-bool AliasSetTracker::add(Value *Ptr, unsigned Size) {
+bool AliasSetTracker::add(Value *Ptr, uint64_t Size, const MDNode *TBAAInfo) {
bool NewPtr;
- addPointer(Ptr, Size, AliasSet::NoModRef, NewPtr);
+ addPointer(Ptr, Size, TBAAInfo, AliasSet::NoModRef, NewPtr);
return NewPtr;
}
bool AliasSetTracker::add(LoadInst *LI) {
+ if (LI->getOrdering() > Monotonic) return addUnknown(LI);
+ AliasSet::AccessType ATy = AliasSet::Refs;
+ if (!LI->isUnordered()) ATy = AliasSet::ModRef;
bool NewPtr;
AliasSet &AS = addPointer(LI->getOperand(0),
AA.getTypeStoreSize(LI->getType()),
- AliasSet::Refs, NewPtr);
+ LI->getMetadata(LLVMContext::MD_tbaa),
+ ATy, NewPtr);
if (LI->isVolatile()) AS.setVolatile();
return NewPtr;
}
bool AliasSetTracker::add(StoreInst *SI) {
+ if (SI->getOrdering() > Monotonic) return addUnknown(SI);
+ AliasSet::AccessType ATy = AliasSet::Mods;
+ if (!SI->isUnordered()) ATy = AliasSet::ModRef;
bool NewPtr;
Value *Val = SI->getOperand(0);
AliasSet &AS = addPointer(SI->getOperand(1),
AA.getTypeStoreSize(Val->getType()),
- AliasSet::Mods, NewPtr);
+ SI->getMetadata(LLVMContext::MD_tbaa),
+ ATy, NewPtr);
if (SI->isVolatile()) AS.setVolatile();
return NewPtr;
}
-bool AliasSetTracker::add(CallSite CS) {
- if (ISA_DEBUG_INFO_INTRINSIC(CS.getInstruction()))
+bool AliasSetTracker::add(VAArgInst *VAAI) {
+ bool NewPtr;
+ addPointer(VAAI->getOperand(0), AliasAnalysis::UnknownSize,
+ VAAI->getMetadata(LLVMContext::MD_tbaa),
+ AliasSet::ModRef, NewPtr);
+ return NewPtr;
+}
+
+
+bool AliasSetTracker::addUnknown(Instruction *Inst) {
+ if (isa<DbgInfoIntrinsic>(Inst))
return true; // Ignore DbgInfo Intrinsics.
- if (AA.doesNotAccessMemory(CS))
+ if (!Inst->mayReadOrWriteMemory())
return true; // doesn't alias anything
- AliasSet *AS = findAliasSetForCallSite(CS);
- if (!AS) {
- AliasSets.push_back(new AliasSet());
- AS = &AliasSets.back();
- AS->addCallSite(CS, AA);
- return true;
- } else {
- AS->addCallSite(CS, AA);
+ AliasSet *AS = findAliasSetForUnknownInst(Inst);
+ if (AS) {
+ AS->addUnknownInst(Inst, AA);
return false;
}
+ AliasSets.push_back(new AliasSet());
+ AS = &AliasSets.back();
+ AS->addUnknownInst(Inst, AA);
+ return true;
}
bool AliasSetTracker::add(Instruction *I) {
- // Dispatch to one of the other add methods...
+ // Dispatch to one of the other add methods.
if (LoadInst *LI = dyn_cast<LoadInst>(I))
return add(LI);
- else if (StoreInst *SI = dyn_cast<StoreInst>(I))
+ if (StoreInst *SI = dyn_cast<StoreInst>(I))
return add(SI);
- else if (CallInst *CI = dyn_cast<CallInst>(I))
- return add(CI);
- return true;
+ if (VAArgInst *VAAI = dyn_cast<VAArgInst>(I))
+ return add(VAAI);
+ return addUnknown(I);
}
void AliasSetTracker::add(BasicBlock &BB) {
@@ -332,30 +370,31 @@
// Loop over all of the alias sets in AST, adding the pointers contained
// therein into the current alias sets. This can cause alias sets to be
// merged together in the current AST.
- for (const_iterator I = AST.begin(), E = AST.end(); I != E; ++I)
- if (!I->Forward) { // Ignore forwarding alias sets
- AliasSet &AS = const_cast<AliasSet&>(*I);
+ for (const_iterator I = AST.begin(), E = AST.end(); I != E; ++I) {
+ if (I->Forward) continue; // Ignore forwarding alias sets
+
+ AliasSet &AS = const_cast<AliasSet&>(*I);
- // If there are any call sites in the alias set, add them to this AST.
- for (unsigned i = 0, e = AS.CallSites.size(); i != e; ++i)
- add(AS.CallSites[i]);
+ // If there are any call sites in the alias set, add them to this AST.
+ for (unsigned i = 0, e = AS.UnknownInsts.size(); i != e; ++i)
+ add(AS.UnknownInsts[i]);
- // Loop over all of the pointers in this alias set...
- AliasSet::iterator I = AS.begin(), E = AS.end();
- bool X;
- for (; I != E; ++I) {
- AliasSet &NewAS = addPointer(I.getPointer(), I.getSize(),
- (AliasSet::AccessType)AS.AccessTy, X);
- if (AS.isVolatile()) NewAS.setVolatile();
- }
+ // Loop over all of the pointers in this alias set.
+ bool X;
+ for (AliasSet::iterator ASI = AS.begin(), E = AS.end(); ASI != E; ++ASI) {
+ AliasSet &NewAS = addPointer(ASI.getPointer(), ASI.getSize(),
+ ASI.getTBAAInfo(),
+ (AliasSet::AccessType)AS.AccessTy, X);
+ if (AS.isVolatile()) NewAS.setVolatile();
}
+ }
}
/// remove - Remove the specified (potentially non-empty) alias set from the
/// tracker.
void AliasSetTracker::remove(AliasSet &AS) {
// Drop all call sites.
- AS.CallSites.clear();
+ AS.UnknownInsts.clear();
// Clear the alias set.
unsigned NumRefs = 0;
@@ -380,34 +419,46 @@
AS.removeFromTracker(*this);
}
-bool AliasSetTracker::remove(Value *Ptr, unsigned Size) {
- AliasSet *AS = findAliasSetForPointer(Ptr, Size);
+bool
+AliasSetTracker::remove(Value *Ptr, uint64_t Size, const MDNode *TBAAInfo) {
+ AliasSet *AS = findAliasSetForPointer(Ptr, Size, TBAAInfo);
if (!AS) return false;
remove(*AS);
return true;
}
bool AliasSetTracker::remove(LoadInst *LI) {
- unsigned Size = AA.getTypeStoreSize(LI->getType());
- AliasSet *AS = findAliasSetForPointer(LI->getOperand(0), Size);
+ uint64_t Size = AA.getTypeStoreSize(LI->getType());
+ const MDNode *TBAAInfo = LI->getMetadata(LLVMContext::MD_tbaa);
+ AliasSet *AS = findAliasSetForPointer(LI->getOperand(0), Size, TBAAInfo);
if (!AS) return false;
remove(*AS);
return true;
}
bool AliasSetTracker::remove(StoreInst *SI) {
- unsigned Size = AA.getTypeStoreSize(SI->getOperand(0)->getType());
- AliasSet *AS = findAliasSetForPointer(SI->getOperand(1), Size);
+ uint64_t Size = AA.getTypeStoreSize(SI->getOperand(0)->getType());
+ const MDNode *TBAAInfo = SI->getMetadata(LLVMContext::MD_tbaa);
+ AliasSet *AS = findAliasSetForPointer(SI->getOperand(1), Size, TBAAInfo);
if (!AS) return false;
remove(*AS);
return true;
}
-bool AliasSetTracker::remove(CallSite CS) {
- if (AA.doesNotAccessMemory(CS))
+bool AliasSetTracker::remove(VAArgInst *VAAI) {
+ AliasSet *AS = findAliasSetForPointer(VAAI->getOperand(0),
+ AliasAnalysis::UnknownSize,
+ VAAI->getMetadata(LLVMContext::MD_tbaa));
+ if (!AS) return false;
+ remove(*AS);
+ return true;
+}
+
+bool AliasSetTracker::removeUnknown(Instruction *I) {
+ if (!I->mayReadOrWriteMemory())
return false; // doesn't alias anything
- AliasSet *AS = findAliasSetForCallSite(CS);
+ AliasSet *AS = findAliasSetForUnknownInst(I);
if (!AS) return false;
remove(*AS);
return true;
@@ -417,11 +468,11 @@
// Dispatch to one of the other remove methods...
if (LoadInst *LI = dyn_cast<LoadInst>(I))
return remove(LI);
- else if (StoreInst *SI = dyn_cast<StoreInst>(I))
+ if (StoreInst *SI = dyn_cast<StoreInst>(I))
return remove(SI);
- else if (CallInst *CI = dyn_cast<CallInst>(I))
- return remove(CI);
- return true;
+ if (VAArgInst *VAAI = dyn_cast<VAArgInst>(I))
+ return remove(VAAI);
+ return removeUnknown(I);
}
@@ -435,11 +486,17 @@
AA.deleteValue(PtrVal);
// If this is a call instruction, remove the callsite from the appropriate
- // AliasSet.
- if (CallSite CS = PtrVal)
- if (!AA.doesNotAccessMemory(CS))
- if (AliasSet *AS = findAliasSetForCallSite(CS))
- AS->removeCallSite(CS);
+ // AliasSet (if present).
+ if (Instruction *Inst = dyn_cast<Instruction>(PtrVal)) {
+ if (Inst->mayReadOrWriteMemory()) {
+ // Scan all the alias sets to see if this call site is contained.
+ for (iterator I = begin(), E = end(); I != E; ++I) {
+ if (I->Forward) continue;
+
+ I->removeUnknownInst(Inst);
+ }
+ }
+ }
// First, look up the PointerRec for this pointer.
PointerMapType::iterator I = PointerMap.find(PtrVal);
@@ -479,7 +536,9 @@
// Add it to the alias set it aliases...
I = PointerMap.find(From);
AliasSet *AS = I->second->getAliasSet(*this);
- AS->addPointer(*this, Entry, I->second->getSize(), true);
+ AS->addPointer(*this, Entry, I->second->getSize(),
+ I->second->getTBAAInfo(),
+ true);
}
@@ -489,7 +548,7 @@
//===----------------------------------------------------------------------===//
void AliasSet::print(raw_ostream &OS) const {
- OS << " AliasSet[" << format("0x%p", (void*)this) << "," << RefCount << "] ";
+ OS << " AliasSet[" << (void*)this << ", " << RefCount << "] ";
OS << (AliasTy == MustAlias ? "must" : "may") << " alias, ";
switch (AccessTy) {
case NoModRef: OS << "No access "; break;
@@ -511,11 +570,11 @@
OS << ", " << I.getSize() << ")";
}
}
- if (!CallSites.empty()) {
- OS << "\n " << CallSites.size() << " Call Sites: ";
- for (unsigned i = 0, e = CallSites.size(); i != e; ++i) {
+ if (!UnknownInsts.empty()) {
+ OS << "\n " << UnknownInsts.size() << " Unknown instructions: ";
+ for (unsigned i = 0, e = UnknownInsts.size(); i != e; ++i) {
if (i) OS << ", ";
- WriteAsOperand(OS, CallSites[i].getCalledValue());
+ WriteAsOperand(OS, UnknownInsts[i]);
}
}
OS << "\n";
@@ -542,6 +601,10 @@
// this now dangles!
}
+void AliasSetTracker::ASTCallbackVH::allUsesReplacedWith(Value *V) {
+ AST->copyValue(getValPtr(), V);
+}
+
AliasSetTracker::ASTCallbackVH::ASTCallbackVH(Value *V, AliasSetTracker *ast)
: CallbackVH(V), AST(ast) {}
@@ -559,7 +622,9 @@
AliasSetTracker *Tracker;
public:
static char ID; // Pass identification, replacement for typeid
- AliasSetPrinter() : FunctionPass(ID) {}
+ AliasSetPrinter() : FunctionPass(ID) {
+ initializeAliasSetPrinterPass(*PassRegistry::getPassRegistry());
+ }
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesAll();
@@ -579,5 +644,8 @@
}
char AliasSetPrinter::ID = 0;
-INITIALIZE_PASS(AliasSetPrinter, "print-alias-sets",
- "Alias Set Printer", false, true);
+INITIALIZE_PASS_BEGIN(AliasSetPrinter, "print-alias-sets",
+ "Alias Set Printer", false, true)
+INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_END(AliasSetPrinter, "print-alias-sets",
+ "Alias Set Printer", false, true)
diff --git a/src/LLVM/lib/Analysis/Analysis.cpp b/src/LLVM/lib/Analysis/Analysis.cpp
new file mode 100644
index 0000000..0ba6af9
--- /dev/null
+++ b/src/LLVM/lib/Analysis/Analysis.cpp
@@ -0,0 +1,105 @@
+//===-- Analysis.cpp ------------------------------------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm-c/Analysis.h"
+#include "llvm-c/Initialization.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Analysis/Verifier.h"
+#include <cstring>
+
+using namespace llvm;
+
+/// initializeAnalysis - Initialize all passes linked into the Analysis library.
+void llvm::initializeAnalysis(PassRegistry &Registry) {
+ initializeAliasAnalysisAnalysisGroup(Registry);
+ initializeAliasAnalysisCounterPass(Registry);
+ initializeAAEvalPass(Registry);
+ initializeAliasDebuggerPass(Registry);
+ initializeAliasSetPrinterPass(Registry);
+ initializeNoAAPass(Registry);
+ initializeBasicAliasAnalysisPass(Registry);
+ initializeBlockFrequencyInfoPass(Registry);
+ initializeBranchProbabilityInfoPass(Registry);
+ initializeCFGViewerPass(Registry);
+ initializeCFGPrinterPass(Registry);
+ initializeCFGOnlyViewerPass(Registry);
+ initializeCFGOnlyPrinterPass(Registry);
+ initializePrintDbgInfoPass(Registry);
+ initializeDominanceFrontierPass(Registry);
+ initializeDomViewerPass(Registry);
+ initializeDomPrinterPass(Registry);
+ initializeDomOnlyViewerPass(Registry);
+ initializePostDomViewerPass(Registry);
+ initializeDomOnlyPrinterPass(Registry);
+ initializePostDomPrinterPass(Registry);
+ initializePostDomOnlyViewerPass(Registry);
+ initializePostDomOnlyPrinterPass(Registry);
+ initializeIVUsersPass(Registry);
+ initializeInstCountPass(Registry);
+ initializeIntervalPartitionPass(Registry);
+ initializeLazyValueInfoPass(Registry);
+ initializeLibCallAliasAnalysisPass(Registry);
+ initializeLintPass(Registry);
+ initializeLoopDependenceAnalysisPass(Registry);
+ initializeLoopInfoPass(Registry);
+ initializeMemDepPrinterPass(Registry);
+ initializeMemoryDependenceAnalysisPass(Registry);
+ initializeModuleDebugInfoPrinterPass(Registry);
+ initializePostDominatorTreePass(Registry);
+ initializeProfileEstimatorPassPass(Registry);
+ initializeNoProfileInfoPass(Registry);
+ initializeNoPathProfileInfoPass(Registry);
+ initializeProfileInfoAnalysisGroup(Registry);
+ initializePathProfileInfoAnalysisGroup(Registry);
+ initializeLoaderPassPass(Registry);
+ initializePathProfileLoaderPassPass(Registry);
+ initializeProfileVerifierPassPass(Registry);
+ initializePathProfileVerifierPass(Registry);
+ initializeRegionInfoPass(Registry);
+ initializeRegionViewerPass(Registry);
+ initializeRegionPrinterPass(Registry);
+ initializeRegionOnlyViewerPass(Registry);
+ initializeRegionOnlyPrinterPass(Registry);
+ initializeScalarEvolutionPass(Registry);
+ initializeScalarEvolutionAliasAnalysisPass(Registry);
+ initializeTypeBasedAliasAnalysisPass(Registry);
+}
+
+void LLVMInitializeAnalysis(LLVMPassRegistryRef R) {
+ initializeAnalysis(*unwrap(R));
+}
+
+LLVMBool LLVMVerifyModule(LLVMModuleRef M, LLVMVerifierFailureAction Action,
+ char **OutMessages) {
+ std::string Messages;
+
+ LLVMBool Result = verifyModule(*unwrap(M),
+ static_cast<VerifierFailureAction>(Action),
+ OutMessages? &Messages : 0);
+
+ if (OutMessages)
+ *OutMessages = strdup(Messages.c_str());
+
+ return Result;
+}
+
+LLVMBool LLVMVerifyFunction(LLVMValueRef Fn, LLVMVerifierFailureAction Action) {
+ return verifyFunction(*unwrap<Function>(Fn),
+ static_cast<VerifierFailureAction>(Action));
+}
+
+void LLVMViewFunctionCFG(LLVMValueRef Fn) {
+ Function *F = unwrap<Function>(Fn);
+ F->viewCFG();
+}
+
+void LLVMViewFunctionCFGOnly(LLVMValueRef Fn) {
+ Function *F = unwrap<Function>(Fn);
+ F->viewCFGOnly();
+}
diff --git a/src/LLVM/lib/Analysis/BasicAliasAnalysis.cpp b/src/LLVM/lib/Analysis/BasicAliasAnalysis.cpp
index ed97fa5..7a2fb2f 100644
--- a/src/LLVM/lib/Analysis/BasicAliasAnalysis.cpp
+++ b/src/LLVM/lib/Analysis/BasicAliasAnalysis.cpp
@@ -1,4 +1,4 @@
-//===- BasicAliasAnalysis.cpp - Local Alias Analysis Impl -----------------===//
+//===- BasicAliasAnalysis.cpp - Stateless Alias Analysis Impl -------------===//
//
// The LLVM Compiler Infrastructure
//
@@ -7,9 +7,9 @@
//
//===----------------------------------------------------------------------===//
//
-// This file defines the default implementation of the Alias Analysis interface
-// that simply implements a few identities (two different globals cannot alias,
-// etc), but otherwise does no analysis.
+// This file defines the primary stateless implementation of the
+// Alias Analysis interface that implements identities (two different
+// globals cannot alias, etc), but does no stateful analysis.
//
//===----------------------------------------------------------------------===//
@@ -18,17 +18,23 @@
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
+#include "llvm/GlobalAlias.h"
#include "llvm/GlobalVariable.h"
#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
+#include "llvm/LLVMContext.h"
#include "llvm/Operator.h"
#include "llvm/Pass.h"
#include "llvm/Analysis/CaptureTracking.h"
+#include "llvm/Analysis/MemoryBuiltins.h"
+#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/GetElementPtrTypeIterator.h"
#include <algorithm>
using namespace llvm;
@@ -80,7 +86,7 @@
/// isEscapeSource - Return true if the pointer is one which would have
/// been considered an escape by isNonEscapingLocalObject.
static bool isEscapeSource(const Value *V) {
- if (isa<CallInst>(V) || ISA_INVOKE_INST(V) || isa<Argument>(V))
+ if (isa<CallInst>(V) || isa<InvokeInst>(V) || isa<Argument>(V))
return true;
// The load case works because isNonEscapingLocalObject considers all
@@ -92,98 +98,332 @@
return false;
}
-/// isObjectSmallerThan - Return true if we can prove that the object specified
-/// by V is smaller than Size.
-static bool isObjectSmallerThan(const Value *V, unsigned Size,
- const TargetData &TD) {
- const Type *AccessTy;
+/// getObjectSize - Return the size of the object specified by V, or
+/// UnknownSize if unknown.
+static uint64_t getObjectSize(const Value *V, const TargetData &TD) {
+ Type *AccessTy;
if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
+ if (!GV->hasDefinitiveInitializer())
+ return AliasAnalysis::UnknownSize;
AccessTy = GV->getType()->getElementType();
} else if (const AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
if (!AI->isArrayAllocation())
AccessTy = AI->getType()->getElementType();
else
- return false;
+ return AliasAnalysis::UnknownSize;
+ } else if (const CallInst* CI = extractMallocCall(V)) {
+ if (!isArrayMalloc(V, &TD))
+ // The size is the argument to the malloc call.
+ if (const ConstantInt* C = dyn_cast<ConstantInt>(CI->getArgOperand(0)))
+ return C->getZExtValue();
+ return AliasAnalysis::UnknownSize;
} else if (const Argument *A = dyn_cast<Argument>(V)) {
if (A->hasByValAttr())
AccessTy = cast<PointerType>(A->getType())->getElementType();
else
- return false;
+ return AliasAnalysis::UnknownSize;
} else {
- return false;
+ return AliasAnalysis::UnknownSize;
}
if (AccessTy->isSized())
- return TD.getTypeAllocSize(AccessTy) < Size;
- return false;
+ return TD.getTypeAllocSize(AccessTy);
+ return AliasAnalysis::UnknownSize;
+}
+
+/// isObjectSmallerThan - Return true if we can prove that the object specified
+/// by V is smaller than Size.
+static bool isObjectSmallerThan(const Value *V, uint64_t Size,
+ const TargetData &TD) {
+ uint64_t ObjectSize = getObjectSize(V, TD);
+ return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize < Size;
+}
+
+/// isObjectSize - Return true if we can prove that the object specified
+/// by V has size Size.
+static bool isObjectSize(const Value *V, uint64_t Size,
+ const TargetData &TD) {
+ uint64_t ObjectSize = getObjectSize(V, TD);
+ return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize == Size;
}
//===----------------------------------------------------------------------===//
-// NoAA Pass
+// GetElementPtr Instruction Decomposition and Analysis
//===----------------------------------------------------------------------===//
namespace {
- /// NoAA - This class implements the -no-aa pass, which always returns "I
- /// don't know" for alias queries. NoAA is unlike other alias analysis
- /// implementations, in that it does not chain to a previous analysis. As
- /// such it doesn't follow many of the rules that other alias analyses must.
- ///
- struct NoAA : public ImmutablePass, public AliasAnalysis {
- static char ID; // Class identification, replacement for typeinfo
- NoAA() : ImmutablePass(ID) {}
- explicit NoAA(char &PID) : ImmutablePass(PID) { }
-
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
- }
-
- virtual void initializePass() {
- TD = getAnalysisIfAvailable<TargetData>();
- }
-
- virtual AliasResult alias(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
- return MayAlias;
- }
-
- virtual ModRefBehavior getModRefBehavior(ImmutableCallSite CS) {
- return UnknownModRefBehavior;
- }
- virtual ModRefBehavior getModRefBehavior(const Function *F) {
- return UnknownModRefBehavior;
- }
-
- virtual bool pointsToConstantMemory(const Value *P) { return false; }
- virtual ModRefResult getModRefInfo(ImmutableCallSite CS,
- const Value *P, unsigned Size) {
- return ModRef;
- }
- virtual ModRefResult getModRefInfo(ImmutableCallSite CS1,
- ImmutableCallSite CS2) {
- return ModRef;
- }
-
- virtual void deleteValue(Value *V) {}
- virtual void copyValue(Value *From, Value *To) {}
-
- /// getAdjustedAnalysisPointer - This method is used when a pass implements
- /// an analysis interface through multiple inheritance. If needed, it
- /// should override this to adjust the this pointer as needed for the
- /// specified pass info.
- virtual void *getAdjustedAnalysisPointer(const void *ID) {
- if (ID == &AliasAnalysis::ID)
- return (AliasAnalysis*)this;
- return this;
- }
+ enum ExtensionKind {
+ EK_NotExtended,
+ EK_SignExt,
+ EK_ZeroExt
};
-} // End of anonymous namespace
+
+ struct VariableGEPIndex {
+ const Value *V;
+ ExtensionKind Extension;
+ int64_t Scale;
+ };
+}
-// Register this pass...
-char NoAA::ID = 0;
-INITIALIZE_AG_PASS(NoAA, AliasAnalysis, "no-aa",
- "No Alias Analysis (always returns 'may' alias)",
- true, true, false);
-ImmutablePass *llvm::createNoAAPass() { return new NoAA(); }
+/// GetLinearExpression - Analyze the specified value as a linear expression:
+/// "A*V + B", where A and B are constant integers. Return the scale and offset
+/// values as APInts and return V as a Value*, and return whether we looked
+/// through any sign or zero extends. The incoming Value is known to have
+/// IntegerType and it may already be sign or zero extended.
+///
+/// Note that this looks through extends, so the high bits may not be
+/// represented in the result.
+static Value *GetLinearExpression(Value *V, APInt &Scale, APInt &Offset,
+ ExtensionKind &Extension,
+ const TargetData &TD, unsigned Depth) {
+ assert(V->getType()->isIntegerTy() && "Not an integer value");
+
+ // Limit our recursion depth.
+ if (Depth == 6) {
+ Scale = 1;
+ Offset = 0;
+ return V;
+ }
+
+ if (BinaryOperator *BOp = dyn_cast<BinaryOperator>(V)) {
+ if (ConstantInt *RHSC = dyn_cast<ConstantInt>(BOp->getOperand(1))) {
+ switch (BOp->getOpcode()) {
+ default: break;
+ case Instruction::Or:
+ // X|C == X+C if all the bits in C are unset in X. Otherwise we can't
+ // analyze it.
+ if (!MaskedValueIsZero(BOp->getOperand(0), RHSC->getValue(), &TD))
+ break;
+ // FALL THROUGH.
+ case Instruction::Add:
+ V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension,
+ TD, Depth+1);
+ Offset += RHSC->getValue();
+ return V;
+ case Instruction::Mul:
+ V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension,
+ TD, Depth+1);
+ Offset *= RHSC->getValue();
+ Scale *= RHSC->getValue();
+ return V;
+ case Instruction::Shl:
+ V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension,
+ TD, Depth+1);
+ Offset <<= RHSC->getValue().getLimitedValue();
+ Scale <<= RHSC->getValue().getLimitedValue();
+ return V;
+ }
+ }
+ }
+
+ // Since GEP indices are sign extended anyway, we don't care about the high
+ // bits of a sign or zero extended value - just scales and offsets. The
+ // extensions have to be consistent though.
+ if ((isa<SExtInst>(V) && Extension != EK_ZeroExt) ||
+ (isa<ZExtInst>(V) && Extension != EK_SignExt)) {
+ Value *CastOp = cast<CastInst>(V)->getOperand(0);
+ unsigned OldWidth = Scale.getBitWidth();
+ unsigned SmallWidth = CastOp->getType()->getPrimitiveSizeInBits();
+ Scale = Scale.trunc(SmallWidth);
+ Offset = Offset.trunc(SmallWidth);
+ Extension = isa<SExtInst>(V) ? EK_SignExt : EK_ZeroExt;
+
+ Value *Result = GetLinearExpression(CastOp, Scale, Offset, Extension,
+ TD, Depth+1);
+ Scale = Scale.zext(OldWidth);
+ Offset = Offset.zext(OldWidth);
+
+ return Result;
+ }
+
+ Scale = 1;
+ Offset = 0;
+ return V;
+}
+
+/// DecomposeGEPExpression - If V is a symbolic pointer expression, decompose it
+/// into a base pointer with a constant offset and a number of scaled symbolic
+/// offsets.
+///
+/// The scaled symbolic offsets (represented by pairs of a Value* and a scale in
+/// the VarIndices vector) are Value*'s that are known to be scaled by the
+/// specified amount, but which may have other unrepresented high bits. As such,
+/// the gep cannot necessarily be reconstructed from its decomposed form.
+///
+/// When TargetData is around, this function is capable of analyzing everything
+/// that GetUnderlyingObject can look through. When not, it just looks
+/// through pointer casts.
+///
+static const Value *
+DecomposeGEPExpression(const Value *V, int64_t &BaseOffs,
+ SmallVectorImpl<VariableGEPIndex> &VarIndices,
+ const TargetData *TD) {
+ // Limit recursion depth to limit compile time in crazy cases.
+ unsigned MaxLookup = 6;
+
+ BaseOffs = 0;
+ do {
+ // See if this is a bitcast or GEP.
+ const Operator *Op = dyn_cast<Operator>(V);
+ if (Op == 0) {
+ // The only non-operator case we can handle are GlobalAliases.
+ if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
+ if (!GA->mayBeOverridden()) {
+ V = GA->getAliasee();
+ continue;
+ }
+ }
+ return V;
+ }
+
+ if (Op->getOpcode() == Instruction::BitCast) {
+ V = Op->getOperand(0);
+ continue;
+ }
+
+ const GEPOperator *GEPOp = dyn_cast<GEPOperator>(Op);
+ if (GEPOp == 0) {
+ // If it's not a GEP, hand it off to SimplifyInstruction to see if it
+ // can come up with something. This matches what GetUnderlyingObject does.
+ if (const Instruction *I = dyn_cast<Instruction>(V))
+ // TODO: Get a DominatorTree and use it here.
+ if (const Value *Simplified =
+ SimplifyInstruction(const_cast<Instruction *>(I), TD)) {
+ V = Simplified;
+ continue;
+ }
+
+ return V;
+ }
+
+ // Don't attempt to analyze GEPs over unsized objects.
+ if (!cast<PointerType>(GEPOp->getOperand(0)->getType())
+ ->getElementType()->isSized())
+ return V;
+
+ // If we are lacking TargetData information, we can't compute the offets of
+ // elements computed by GEPs. However, we can handle bitcast equivalent
+ // GEPs.
+ if (TD == 0) {
+ if (!GEPOp->hasAllZeroIndices())
+ return V;
+ V = GEPOp->getOperand(0);
+ continue;
+ }
+
+ // Walk the indices of the GEP, accumulating them into BaseOff/VarIndices.
+ gep_type_iterator GTI = gep_type_begin(GEPOp);
+ for (User::const_op_iterator I = GEPOp->op_begin()+1,
+ E = GEPOp->op_end(); I != E; ++I) {
+ Value *Index = *I;
+ // Compute the (potentially symbolic) offset in bytes for this index.
+ if (StructType *STy = dyn_cast<StructType>(*GTI++)) {
+ // For a struct, add the member offset.
+ unsigned FieldNo = cast<ConstantInt>(Index)->getZExtValue();
+ if (FieldNo == 0) continue;
+
+ BaseOffs += TD->getStructLayout(STy)->getElementOffset(FieldNo);
+ continue;
+ }
+
+ // For an array/pointer, add the element offset, explicitly scaled.
+ if (ConstantInt *CIdx = dyn_cast<ConstantInt>(Index)) {
+ if (CIdx->isZero()) continue;
+ BaseOffs += TD->getTypeAllocSize(*GTI)*CIdx->getSExtValue();
+ continue;
+ }
+
+ uint64_t Scale = TD->getTypeAllocSize(*GTI);
+ ExtensionKind Extension = EK_NotExtended;
+
+ // If the integer type is smaller than the pointer size, it is implicitly
+ // sign extended to pointer size.
+ unsigned Width = cast<IntegerType>(Index->getType())->getBitWidth();
+ if (TD->getPointerSizeInBits() > Width)
+ Extension = EK_SignExt;
+
+ // Use GetLinearExpression to decompose the index into a C1*V+C2 form.
+ APInt IndexScale(Width, 0), IndexOffset(Width, 0);
+ Index = GetLinearExpression(Index, IndexScale, IndexOffset, Extension,
+ *TD, 0);
+
+ // The GEP index scale ("Scale") scales C1*V+C2, yielding (C1*V+C2)*Scale.
+ // This gives us an aggregate computation of (C1*Scale)*V + C2*Scale.
+ BaseOffs += IndexOffset.getSExtValue()*Scale;
+ Scale *= IndexScale.getSExtValue();
+
+
+ // If we already had an occurrence of this index variable, merge this
+ // scale into it. For example, we want to handle:
+ // A[x][x] -> x*16 + x*4 -> x*20
+ // This also ensures that 'x' only appears in the index list once.
+ for (unsigned i = 0, e = VarIndices.size(); i != e; ++i) {
+ if (VarIndices[i].V == Index &&
+ VarIndices[i].Extension == Extension) {
+ Scale += VarIndices[i].Scale;
+ VarIndices.erase(VarIndices.begin()+i);
+ break;
+ }
+ }
+
+ // Make sure that we have a scale that makes sense for this target's
+ // pointer size.
+ if (unsigned ShiftBits = 64-TD->getPointerSizeInBits()) {
+ Scale <<= ShiftBits;
+ Scale = (int64_t)Scale >> ShiftBits;
+ }
+
+ if (Scale) {
+ VariableGEPIndex Entry = {Index, Extension,
+ static_cast<int64_t>(Scale)};
+ VarIndices.push_back(Entry);
+ }
+ }
+
+ // Analyze the base pointer next.
+ V = GEPOp->getOperand(0);
+ } while (--MaxLookup);
+
+ // If the chain of expressions is too deep, just return early.
+ return V;
+}
+
+/// GetIndexDifference - Dest and Src are the variable indices from two
+/// decomposed GetElementPtr instructions GEP1 and GEP2 which have common base
+/// pointers. Subtract the GEP2 indices from GEP1 to find the symbolic
+/// difference between the two pointers.
+static void GetIndexDifference(SmallVectorImpl<VariableGEPIndex> &Dest,
+ const SmallVectorImpl<VariableGEPIndex> &Src) {
+ if (Src.empty()) return;
+
+ for (unsigned i = 0, e = Src.size(); i != e; ++i) {
+ const Value *V = Src[i].V;
+ ExtensionKind Extension = Src[i].Extension;
+ int64_t Scale = Src[i].Scale;
+
+ // Find V in Dest. This is N^2, but pointer indices almost never have more
+ // than a few variable indexes.
+ for (unsigned j = 0, e = Dest.size(); j != e; ++j) {
+ if (Dest[j].V != V || Dest[j].Extension != Extension) continue;
+
+ // If we found it, subtract off Scale V's from the entry in Dest. If it
+ // goes to zero, remove the entry.
+ if (Dest[j].Scale != Scale)
+ Dest[j].Scale -= Scale;
+ else
+ Dest.erase(Dest.begin()+j);
+ Scale = 0;
+ break;
+ }
+
+ // If we didn't consume this entry, add it to the end of the Dest list.
+ if (Scale) {
+ VariableGEPIndex Entry = { V, Extension, -Scale };
+ Dest.push_back(Entry);
+ }
+ }
+}
//===----------------------------------------------------------------------===//
// BasicAliasAnalysis Pass
@@ -210,25 +450,41 @@
#endif
namespace {
- /// BasicAliasAnalysis - This is the default alias analysis implementation.
- /// Because it doesn't chain to a previous alias analysis (like -no-aa), it
- /// derives from the NoAA class.
- struct BasicAliasAnalysis : public NoAA {
+ /// BasicAliasAnalysis - This is the primary alias analysis implementation.
+ struct BasicAliasAnalysis : public ImmutablePass, public AliasAnalysis {
static char ID; // Class identification, replacement for typeinfo
- BasicAliasAnalysis() : NoAA(ID) {}
+ BasicAliasAnalysis() : ImmutablePass(ID),
+ // AliasCache rarely has more than 1 or 2 elements,
+ // so start it off fairly small so that clear()
+ // doesn't have to tromp through 64 (the default)
+ // elements on each alias query. This really wants
+ // something like a SmallDenseMap.
+ AliasCache(8) {
+ initializeBasicAliasAnalysisPass(*PassRegistry::getPassRegistry());
+ }
- virtual AliasResult alias(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
- assert(Visited.empty() && "Visited must be cleared after use!");
- assert(notDifferentParent(V1, V2) &&
+ virtual void initializePass() {
+ InitializeAliasAnalysis(this);
+ }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.addRequired<AliasAnalysis>();
+ AU.addRequired<TargetLibraryInfo>();
+ }
+
+ virtual AliasResult alias(const Location &LocA,
+ const Location &LocB) {
+ assert(AliasCache.empty() && "AliasCache must be cleared after use!");
+ assert(notDifferentParent(LocA.Ptr, LocB.Ptr) &&
"BasicAliasAnalysis doesn't support interprocedural queries.");
- AliasResult Alias = aliasCheck(V1, V1Size, V2, V2Size);
- Visited.clear();
+ AliasResult Alias = aliasCheck(LocA.Ptr, LocA.Size, LocA.TBAATag,
+ LocB.Ptr, LocB.Size, LocB.TBAATag);
+ AliasCache.clear();
return Alias;
}
virtual ModRefResult getModRefInfo(ImmutableCallSite CS,
- const Value *P, unsigned Size);
+ const Location &Loc);
virtual ModRefResult getModRefInfo(ImmutableCallSite CS1,
ImmutableCallSite CS2) {
@@ -238,7 +494,7 @@
/// pointsToConstantMemory - Chase pointers until we find a (constant
/// global) or not.
- virtual bool pointsToConstantMemory(const Value *P);
+ virtual bool pointsToConstantMemory(const Location &Loc, bool OrLocal);
/// getModRefBehavior - Return the behavior when calling the given
/// call site.
@@ -259,51 +515,117 @@
}
private:
- // Visited - Track instructions visited by a aliasPHI, aliasSelect(), and aliasGEP().
+ // AliasCache - Track alias queries to guard against recursion.
+ typedef std::pair<Location, Location> LocPair;
+ typedef DenseMap<LocPair, AliasResult> AliasCacheTy;
+ AliasCacheTy AliasCache;
+
+ // Visited - Track instructions visited by pointsToConstantMemory.
SmallPtrSet<const Value*, 16> Visited;
// aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP
// instruction against another.
- AliasResult aliasGEP(const GEPOperator *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size,
+ AliasResult aliasGEP(const GEPOperator *V1, uint64_t V1Size,
+ const Value *V2, uint64_t V2Size,
+ const MDNode *V2TBAAInfo,
const Value *UnderlyingV1, const Value *UnderlyingV2);
// aliasPHI - Provide a bunch of ad-hoc rules to disambiguate a PHI
// instruction against another.
- AliasResult aliasPHI(const PHINode *PN, unsigned PNSize,
- const Value *V2, unsigned V2Size);
+ AliasResult aliasPHI(const PHINode *PN, uint64_t PNSize,
+ const MDNode *PNTBAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const MDNode *V2TBAAInfo);
/// aliasSelect - Disambiguate a Select instruction against another value.
- AliasResult aliasSelect(const SelectInst *SI, unsigned SISize,
- const Value *V2, unsigned V2Size);
+ AliasResult aliasSelect(const SelectInst *SI, uint64_t SISize,
+ const MDNode *SITBAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const MDNode *V2TBAAInfo);
- AliasResult aliasCheck(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size);
+ AliasResult aliasCheck(const Value *V1, uint64_t V1Size,
+ const MDNode *V1TBAATag,
+ const Value *V2, uint64_t V2Size,
+ const MDNode *V2TBAATag);
};
} // End of anonymous namespace
// Register this pass...
char BasicAliasAnalysis::ID = 0;
-INITIALIZE_AG_PASS(BasicAliasAnalysis, AliasAnalysis, "basicaa",
- "Basic Alias Analysis (default AA impl)",
- false, true, true);
+INITIALIZE_AG_PASS_BEGIN(BasicAliasAnalysis, AliasAnalysis, "basicaa",
+ "Basic Alias Analysis (stateless AA impl)",
+ false, true, false)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
+INITIALIZE_AG_PASS_END(BasicAliasAnalysis, AliasAnalysis, "basicaa",
+ "Basic Alias Analysis (stateless AA impl)",
+ false, true, false)
+
ImmutablePass *llvm::createBasicAliasAnalysisPass() {
return new BasicAliasAnalysis();
}
+/// pointsToConstantMemory - Returns whether the given pointer value
+/// points to memory that is local to the function, with global constants being
+/// considered local to all functions.
+bool
+BasicAliasAnalysis::pointsToConstantMemory(const Location &Loc, bool OrLocal) {
+ assert(Visited.empty() && "Visited must be cleared after use!");
-/// pointsToConstantMemory - Chase pointers until we find a (constant
-/// global) or not.
-bool BasicAliasAnalysis::pointsToConstantMemory(const Value *P) {
- if (const GlobalVariable *GV =
- dyn_cast<GlobalVariable>(P->getUnderlyingObject()))
- // Note: this doesn't require GV to be "ODR" because it isn't legal for a
- // global to be marked constant in some modules and non-constant in others.
- // GV may even be a declaration, not a definition.
- return GV->isConstant();
+ unsigned MaxLookup = 8;
+ SmallVector<const Value *, 16> Worklist;
+ Worklist.push_back(Loc.Ptr);
+ do {
+ const Value *V = GetUnderlyingObject(Worklist.pop_back_val(), TD);
+ if (!Visited.insert(V)) {
+ Visited.clear();
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+ }
- return NoAA::pointsToConstantMemory(P);
+ // An alloca instruction defines local memory.
+ if (OrLocal && isa<AllocaInst>(V))
+ continue;
+
+ // A global constant counts as local memory for our purposes.
+ if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
+ // Note: this doesn't require GV to be "ODR" because it isn't legal for a
+ // global to be marked constant in some modules and non-constant in
+ // others. GV may even be a declaration, not a definition.
+ if (!GV->isConstant()) {
+ Visited.clear();
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+ }
+ continue;
+ }
+
+ // If both select values point to local memory, then so does the select.
+ if (const SelectInst *SI = dyn_cast<SelectInst>(V)) {
+ Worklist.push_back(SI->getTrueValue());
+ Worklist.push_back(SI->getFalseValue());
+ continue;
+ }
+
+ // If all values incoming to a phi node point to local memory, then so does
+ // the phi.
+ if (const PHINode *PN = dyn_cast<PHINode>(V)) {
+ // Don't bother inspecting phi nodes with many operands.
+ if (PN->getNumIncomingValues() > MaxLookup) {
+ Visited.clear();
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+ }
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ Worklist.push_back(PN->getIncomingValue(i));
+ continue;
+ }
+
+ // Otherwise be conservative.
+ Visited.clear();
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+
+ } while (!Worklist.empty() && --MaxLookup);
+
+ Visited.clear();
+ return Worklist.empty();
}
/// getModRefBehavior - Return the behavior when calling the given call site.
@@ -321,22 +643,32 @@
Min = OnlyReadsMemory;
// The AliasAnalysis base class has some smarts, lets use them.
- return std::min(AliasAnalysis::getModRefBehavior(CS), Min);
+ return ModRefBehavior(AliasAnalysis::getModRefBehavior(CS) & Min);
}
/// getModRefBehavior - Return the behavior when calling the given function.
/// For use when the call site is not known.
AliasAnalysis::ModRefBehavior
BasicAliasAnalysis::getModRefBehavior(const Function *F) {
+ // If the function declares it doesn't access memory, we can't do better.
if (F->doesNotAccessMemory())
- // Can't do better than this.
return DoesNotAccessMemory;
- if (F->onlyReadsMemory())
- return OnlyReadsMemory;
- if (unsigned id = F->getIntrinsicID())
- return getIntrinsicModRefBehavior(id);
- return NoAA::getModRefBehavior(F);
+ // For intrinsics, we can check the table.
+ if (unsigned iid = F->getIntrinsicID()) {
+#define GET_INTRINSIC_MODREF_BEHAVIOR
+#include "llvm/Intrinsics.gen"
+#undef GET_INTRINSIC_MODREF_BEHAVIOR
+ }
+
+ ModRefBehavior Min = UnknownModRefBehavior;
+
+ // If the function declares it only reads memory, go with that.
+ if (F->onlyReadsMemory())
+ Min = OnlyReadsMemory;
+
+ // Otherwise be conservative.
+ return ModRefBehavior(AliasAnalysis::getModRefBehavior(F) & Min);
}
/// getModRefInfo - Check to see if the specified callsite can clobber the
@@ -345,13 +677,13 @@
/// simple "address taken" analysis on local objects.
AliasAnalysis::ModRefResult
BasicAliasAnalysis::getModRefInfo(ImmutableCallSite CS,
- const Value *P, unsigned Size) {
- assert(notDifferentParent(CS.getInstruction(), P) &&
+ const Location &Loc) {
+ assert(notDifferentParent(CS.getInstruction(), Loc.Ptr) &&
"AliasAnalysis query involving multiple functions!");
- const Value *Object = P->getUnderlyingObject();
+ const Value *Object = GetUnderlyingObject(Loc.Ptr, TD);
- // If this is a tail call and P points to a stack location, we know that
+ // If this is a tail call and Loc.Ptr points to a stack location, we know that
// the tail call cannot access or modify the local stack.
// We cannot exclude byval arguments here; these belong to the caller of
// the current function not to the current function, and a tail callee
@@ -370,16 +702,19 @@
unsigned ArgNo = 0;
for (ImmutableCallSite::arg_iterator CI = CS.arg_begin(), CE = CS.arg_end();
CI != CE; ++CI, ++ArgNo) {
- // Only look at the no-capture pointer arguments.
+ // Only look at the no-capture or byval pointer arguments. If this
+ // pointer were passed to arguments that were neither of these, then it
+ // couldn't be no-capture.
if (!(*CI)->getType()->isPointerTy() ||
- !CS.paramHasAttr(ArgNo+1, Attribute::NoCapture))
+ (!CS.paramHasAttr(ArgNo+1, Attribute::NoCapture) &&
+ !CS.paramHasAttr(ArgNo+1, Attribute::ByVal)))
continue;
- // If this is a no-capture pointer argument, see if we can tell that it
+ // If this is a no-capture pointer argument, see if we can tell that it
// is impossible to alias the pointer we're checking. If not, we have to
// assume that the call could touch the pointer, even though it doesn't
// escape.
- if (!isNoAlias(cast<Value>(CI), UnknownSize, P, UnknownSize)) {
+ if (!isNoAlias(Location(*CI), Location(Object))) {
PassedAsArg = true;
break;
}
@@ -389,6 +724,9 @@
return NoModRef;
}
+ const TargetLibraryInfo &TLI = getAnalysis<TargetLibraryInfo>();
+ ModRefResult Min = ModRef;
+
// Finally, handle specific knowledge of intrinsics.
const IntrinsicInst *II = dyn_cast<IntrinsicInst>(CS.getInstruction());
if (II != 0)
@@ -396,15 +734,20 @@
default: break;
case Intrinsic::memcpy:
case Intrinsic::memmove: {
- unsigned Len = UnknownSize;
+ uint64_t Len = UnknownSize;
if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getArgOperand(2)))
Len = LenCI->getZExtValue();
Value *Dest = II->getArgOperand(0);
Value *Src = II->getArgOperand(1);
- if (isNoAlias(Dest, Len, P, Size)) {
- if (isNoAlias(Src, Len, P, Size))
+ // If it can't overlap the source dest, then it doesn't modref the loc.
+ if (isNoAlias(Location(Dest, Len), Loc)) {
+ if (isNoAlias(Location(Src, Len), Loc))
return NoModRef;
- return Ref;
+ // If it can't overlap the dest, then worst case it reads the loc.
+ Min = Ref;
+ } else if (isNoAlias(Location(Src, Len), Loc)) {
+ // If it can't overlap the source, then worst case it mutates the loc.
+ Min = Mod;
}
break;
}
@@ -412,114 +755,115 @@
// Since memset is 'accesses arguments' only, the AliasAnalysis base class
// will handle it for the variable length case.
if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getArgOperand(2))) {
- unsigned Len = LenCI->getZExtValue();
+ uint64_t Len = LenCI->getZExtValue();
Value *Dest = II->getArgOperand(0);
- if (isNoAlias(Dest, Len, P, Size))
+ if (isNoAlias(Location(Dest, Len), Loc))
return NoModRef;
}
- break;
- case Intrinsic::atomic_cmp_swap:
- case Intrinsic::atomic_swap:
- case Intrinsic::atomic_load_add:
- case Intrinsic::atomic_load_sub:
- case Intrinsic::atomic_load_and:
- case Intrinsic::atomic_load_nand:
- case Intrinsic::atomic_load_or:
- case Intrinsic::atomic_load_xor:
- case Intrinsic::atomic_load_max:
- case Intrinsic::atomic_load_min:
- case Intrinsic::atomic_load_umax:
- case Intrinsic::atomic_load_umin:
- if (TD) {
- Value *Op1 = II->getArgOperand(0);
- unsigned Op1Size = TD->getTypeStoreSize(Op1->getType());
- if (isNoAlias(Op1, Op1Size, P, Size))
- return NoModRef;
- }
+ // We know that memset doesn't load anything.
+ Min = Mod;
break;
case Intrinsic::lifetime_start:
case Intrinsic::lifetime_end:
case Intrinsic::invariant_start: {
- unsigned PtrSize =
+ uint64_t PtrSize =
cast<ConstantInt>(II->getArgOperand(0))->getZExtValue();
- if (isNoAlias(II->getArgOperand(1), PtrSize, P, Size))
+ if (isNoAlias(Location(II->getArgOperand(1),
+ PtrSize,
+ II->getMetadata(LLVMContext::MD_tbaa)),
+ Loc))
return NoModRef;
break;
}
case Intrinsic::invariant_end: {
- unsigned PtrSize =
+ uint64_t PtrSize =
cast<ConstantInt>(II->getArgOperand(1))->getZExtValue();
- if (isNoAlias(II->getArgOperand(2), PtrSize, P, Size))
+ if (isNoAlias(Location(II->getArgOperand(2),
+ PtrSize,
+ II->getMetadata(LLVMContext::MD_tbaa)),
+ Loc))
return NoModRef;
break;
}
+ //case Intrinsic::arm_neon_vld1: {
+ // // LLVM's vld1 and vst1 intrinsics currently only support a single
+ // // vector register.
+ // uint64_t Size =
+ // TD ? TD->getTypeStoreSize(II->getType()) : UnknownSize;
+ // if (isNoAlias(Location(II->getArgOperand(0), Size,
+ // II->getMetadata(LLVMContext::MD_tbaa)),
+ // Loc))
+ // return NoModRef;
+ // break;
+ //}
+ //case Intrinsic::arm_neon_vst1: {
+ // uint64_t Size =
+ // TD ? TD->getTypeStoreSize(II->getArgOperand(1)->getType()) : UnknownSize;
+ // if (isNoAlias(Location(II->getArgOperand(0), Size,
+ // II->getMetadata(LLVMContext::MD_tbaa)),
+ // Loc))
+ // return NoModRef;
+ // break;
+ //}
}
+ // We can bound the aliasing properties of memset_pattern16 just as we can
+ // for memcpy/memset. This is particularly important because the
+ // LoopIdiomRecognizer likes to turn loops into calls to memset_pattern16
+ // whenever possible.
+ else if (TLI.has(LibFunc::memset_pattern16) &&
+ CS.getCalledFunction() &&
+ CS.getCalledFunction()->getName() == "memset_pattern16") {
+ const Function *MS = CS.getCalledFunction();
+ FunctionType *MemsetType = MS->getFunctionType();
+ if (!MemsetType->isVarArg() && MemsetType->getNumParams() == 3 &&
+ isa<PointerType>(MemsetType->getParamType(0)) &&
+ isa<PointerType>(MemsetType->getParamType(1)) &&
+ isa<IntegerType>(MemsetType->getParamType(2))) {
+ uint64_t Len = UnknownSize;
+ if (const ConstantInt *LenCI = dyn_cast<ConstantInt>(CS.getArgument(2)))
+ Len = LenCI->getZExtValue();
+ const Value *Dest = CS.getArgument(0);
+ const Value *Src = CS.getArgument(1);
+ // If it can't overlap the source dest, then it doesn't modref the loc.
+ if (isNoAlias(Location(Dest, Len), Loc)) {
+ // Always reads 16 bytes of the source.
+ if (isNoAlias(Location(Src, 16), Loc))
+ return NoModRef;
+ // If it can't overlap the dest, then worst case it reads the loc.
+ Min = Ref;
+ // Always reads 16 bytes of the source.
+ } else if (isNoAlias(Location(Src, 16), Loc)) {
+ // If it can't overlap the source, then worst case it mutates the loc.
+ Min = Mod;
+ }
+ }
+ }
+
// The AliasAnalysis base class has some smarts, lets use them.
- return AliasAnalysis::getModRefInfo(CS, P, Size);
-}
-
-
-/// GetIndexDifference - Dest and Src are the variable indices from two
-/// decomposed GetElementPtr instructions GEP1 and GEP2 which have common base
-/// pointers. Subtract the GEP2 indices from GEP1 to find the symbolic
-/// difference between the two pointers.
-static void GetIndexDifference(
- SmallVectorImpl<std::pair<const Value*, int64_t> > &Dest,
- const SmallVectorImpl<std::pair<const Value*, int64_t> > &Src) {
- if (Src.empty()) return;
-
- for (unsigned i = 0, e = Src.size(); i != e; ++i) {
- const Value *V = Src[i].first;
- int64_t Scale = Src[i].second;
-
- // Find V in Dest. This is N^2, but pointer indices almost never have more
- // than a few variable indexes.
- for (unsigned j = 0, e = Dest.size(); j != e; ++j) {
- if (Dest[j].first != V) continue;
-
- // If we found it, subtract off Scale V's from the entry in Dest. If it
- // goes to zero, remove the entry.
- if (Dest[j].second != Scale)
- Dest[j].second -= Scale;
- else
- Dest.erase(Dest.begin()+j);
- Scale = 0;
- break;
- }
-
- // If we didn't consume this entry, add it to the end of the Dest list.
- if (Scale)
- Dest.push_back(std::make_pair(V, -Scale));
- }
+ return ModRefResult(AliasAnalysis::getModRefInfo(CS, Loc) & Min);
}
/// aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP instruction
/// against another pointer. We know that V1 is a GEP, but we don't know
-/// anything about V2. UnderlyingV1 is GEP1->getUnderlyingObject(),
+/// anything about V2. UnderlyingV1 is GetUnderlyingObject(GEP1, TD),
/// UnderlyingV2 is the same for V2.
///
AliasAnalysis::AliasResult
-BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, unsigned V1Size,
- const Value *V2, unsigned V2Size,
+BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size,
+ const Value *V2, uint64_t V2Size,
+ const MDNode *V2TBAAInfo,
const Value *UnderlyingV1,
const Value *UnderlyingV2) {
- // If this GEP has been visited before, we're on a use-def cycle.
- // Such cycles are only valid when PHI nodes are involved or in unreachable
- // code. The visitPHI function catches cycles containing PHIs, but there
- // could still be a cycle without PHIs in unreachable code.
- if (!Visited.insert(GEP1))
- return MayAlias;
-
int64_t GEP1BaseOffset;
- SmallVector<std::pair<const Value*, int64_t>, 4> GEP1VariableIndices;
+ SmallVector<VariableGEPIndex, 4> GEP1VariableIndices;
// If we have two gep instructions with must-alias'ing base pointers, figure
// out if the indexes to the GEP tell us anything about the derived pointer.
if (const GEPOperator *GEP2 = dyn_cast<GEPOperator>(V2)) {
// Do the base pointers alias?
- AliasResult BaseAlias = aliasCheck(UnderlyingV1, UnknownSize,
- UnderlyingV2, UnknownSize);
+ AliasResult BaseAlias = aliasCheck(UnderlyingV1, UnknownSize, 0,
+ UnderlyingV2, UnknownSize, 0);
// If we get a No or May, then return it immediately, no amount of analysis
// will improve this situation.
@@ -532,7 +876,7 @@
DecomposeGEPExpression(GEP1, GEP1BaseOffset, GEP1VariableIndices, TD);
int64_t GEP2BaseOffset;
- SmallVector<std::pair<const Value*, int64_t>, 4> GEP2VariableIndices;
+ SmallVector<VariableGEPIndex, 4> GEP2VariableIndices;
const Value *GEP2BasePtr =
DecomposeGEPExpression(GEP2, GEP2BaseOffset, GEP2VariableIndices, TD);
@@ -541,7 +885,7 @@
// to handle without it.
if (GEP1BasePtr != UnderlyingV1 || GEP2BasePtr != UnderlyingV2) {
assert(TD == 0 &&
- "DecomposeGEPExpression and getUnderlyingObject disagree!");
+ "DecomposeGEPExpression and GetUnderlyingObject disagree!");
return MayAlias;
}
@@ -559,7 +903,8 @@
if (V1Size == UnknownSize && V2Size == UnknownSize)
return MayAlias;
- AliasResult R = aliasCheck(UnderlyingV1, UnknownSize, V2, V2Size);
+ AliasResult R = aliasCheck(UnderlyingV1, UnknownSize, 0,
+ V2, V2Size, V2TBAAInfo);
if (R != MustAlias)
// If V2 may alias GEP base pointer, conservatively returns MayAlias.
// If V2 is known not to alias GEP base pointer, then the two values
@@ -576,7 +921,7 @@
// to handle without it.
if (GEP1BasePtr != UnderlyingV1) {
assert(TD == 0 &&
- "DecomposeGEPExpression and getUnderlyingObject disagree!");
+ "DecomposeGEPExpression and GetUnderlyingObject disagree!");
return MayAlias;
}
}
@@ -590,108 +935,129 @@
if (GEP1BaseOffset == 0 && GEP1VariableIndices.empty())
return MustAlias;
- // If we have a known constant offset, see if this offset is larger than the
- // access size being queried. If so, and if no variable indices can remove
- // pieces of this constant, then we know we have a no-alias. For example,
- // &A[100] != &A.
-
- // In order to handle cases like &A[100][i] where i is an out of range
- // subscript, we have to ignore all constant offset pieces that are a multiple
- // of a scaled index. Do this by removing constant offsets that are a
- // multiple of any of our variable indices. This allows us to transform
- // things like &A[i][1] because i has a stride of (e.g.) 8 bytes but the 1
- // provides an offset of 4 bytes (assuming a <= 4 byte access).
- for (unsigned i = 0, e = GEP1VariableIndices.size();
- i != e && GEP1BaseOffset;++i)
- if (int64_t RemovedOffset = GEP1BaseOffset/GEP1VariableIndices[i].second)
- GEP1BaseOffset -= RemovedOffset*GEP1VariableIndices[i].second;
-
- // If our known offset is bigger than the access size, we know we don't have
- // an alias.
- if (GEP1BaseOffset) {
- if (GEP1BaseOffset >= (int64_t)V2Size ||
- GEP1BaseOffset <= -(int64_t)V1Size)
+ // If there is a constant difference between the pointers, but the difference
+ // is less than the size of the associated memory object, then we know
+ // that the objects are partially overlapping. If the difference is
+ // greater, we know they do not overlap.
+ if (GEP1BaseOffset != 0 && GEP1VariableIndices.empty()) {
+ if (GEP1BaseOffset >= 0) {
+ if (V2Size != UnknownSize) {
+ if ((uint64_t)GEP1BaseOffset < V2Size)
+ return PartialAlias;
+ return NoAlias;
+ }
+ } else {
+ if (V1Size != UnknownSize) {
+ if (-(uint64_t)GEP1BaseOffset < V1Size)
+ return PartialAlias;
+ return NoAlias;
+ }
+ }
+ }
+
+ // Try to distinguish something like &A[i][1] against &A[42][0].
+ // Grab the least significant bit set in any of the scales.
+ if (!GEP1VariableIndices.empty()) {
+ uint64_t Modulo = 0;
+ for (unsigned i = 0, e = GEP1VariableIndices.size(); i != e; ++i)
+ Modulo |= (uint64_t)GEP1VariableIndices[i].Scale;
+ Modulo = Modulo ^ (Modulo & (Modulo - 1));
+
+ // We can compute the difference between the two addresses
+ // mod Modulo. Check whether that difference guarantees that the
+ // two locations do not alias.
+ uint64_t ModOffset = (uint64_t)GEP1BaseOffset & (Modulo - 1);
+ if (V1Size != UnknownSize && V2Size != UnknownSize &&
+ ModOffset >= V2Size && V1Size <= Modulo - ModOffset)
return NoAlias;
}
-
- return MayAlias;
+
+ // Statically, we can see that the base objects are the same, but the
+ // pointers have dynamic offsets which we can't resolve. And none of our
+ // little tricks above worked.
+ //
+ // TODO: Returning PartialAlias instead of MayAlias is a mild hack; the
+ // practical effect of this is protecting TBAA in the case of dynamic
+ // indices into arrays of unions. An alternative way to solve this would
+ // be to have clang emit extra metadata for unions and/or union accesses.
+ // A union-specific solution wouldn't handle the problem for malloc'd
+ // memory however.
+ return PartialAlias;
+}
+
+static AliasAnalysis::AliasResult
+MergeAliasResults(AliasAnalysis::AliasResult A, AliasAnalysis::AliasResult B) {
+ // If the results agree, take it.
+ if (A == B)
+ return A;
+ // A mix of PartialAlias and MustAlias is PartialAlias.
+ if ((A == AliasAnalysis::PartialAlias && B == AliasAnalysis::MustAlias) ||
+ (B == AliasAnalysis::PartialAlias && A == AliasAnalysis::MustAlias))
+ return AliasAnalysis::PartialAlias;
+ // Otherwise, we don't know anything.
+ return AliasAnalysis::MayAlias;
}
/// aliasSelect - Provide a bunch of ad-hoc rules to disambiguate a Select
/// instruction against another.
AliasAnalysis::AliasResult
-BasicAliasAnalysis::aliasSelect(const SelectInst *SI, unsigned SISize,
- const Value *V2, unsigned V2Size) {
- // If this select has been visited before, we're on a use-def cycle.
- // Such cycles are only valid when PHI nodes are involved or in unreachable
- // code. The visitPHI function catches cycles containing PHIs, but there
- // could still be a cycle without PHIs in unreachable code.
- if (!Visited.insert(SI))
- return MayAlias;
-
+BasicAliasAnalysis::aliasSelect(const SelectInst *SI, uint64_t SISize,
+ const MDNode *SITBAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const MDNode *V2TBAAInfo) {
// If the values are Selects with the same condition, we can do a more precise
// check: just check for aliases between the values on corresponding arms.
if (const SelectInst *SI2 = dyn_cast<SelectInst>(V2))
if (SI->getCondition() == SI2->getCondition()) {
AliasResult Alias =
- aliasCheck(SI->getTrueValue(), SISize,
- SI2->getTrueValue(), V2Size);
+ aliasCheck(SI->getTrueValue(), SISize, SITBAAInfo,
+ SI2->getTrueValue(), V2Size, V2TBAAInfo);
if (Alias == MayAlias)
return MayAlias;
AliasResult ThisAlias =
- aliasCheck(SI->getFalseValue(), SISize,
- SI2->getFalseValue(), V2Size);
- if (ThisAlias != Alias)
- return MayAlias;
- return Alias;
+ aliasCheck(SI->getFalseValue(), SISize, SITBAAInfo,
+ SI2->getFalseValue(), V2Size, V2TBAAInfo);
+ return MergeAliasResults(ThisAlias, Alias);
}
// If both arms of the Select node NoAlias or MustAlias V2, then returns
// NoAlias / MustAlias. Otherwise, returns MayAlias.
AliasResult Alias =
- aliasCheck(V2, V2Size, SI->getTrueValue(), SISize);
+ aliasCheck(V2, V2Size, V2TBAAInfo, SI->getTrueValue(), SISize, SITBAAInfo);
if (Alias == MayAlias)
return MayAlias;
- // If V2 is visited, the recursive case will have been caught in the
- // above aliasCheck call, so these subsequent calls to aliasCheck
- // don't need to assume that V2 is being visited recursively.
- Visited.erase(V2);
-
AliasResult ThisAlias =
- aliasCheck(V2, V2Size, SI->getFalseValue(), SISize);
- if (ThisAlias != Alias)
- return MayAlias;
- return Alias;
+ aliasCheck(V2, V2Size, V2TBAAInfo, SI->getFalseValue(), SISize, SITBAAInfo);
+ return MergeAliasResults(ThisAlias, Alias);
}
// aliasPHI - Provide a bunch of ad-hoc rules to disambiguate a PHI instruction
// against another.
AliasAnalysis::AliasResult
-BasicAliasAnalysis::aliasPHI(const PHINode *PN, unsigned PNSize,
- const Value *V2, unsigned V2Size) {
- // The PHI node has already been visited, avoid recursion any further.
- if (!Visited.insert(PN))
- return MayAlias;
-
+BasicAliasAnalysis::aliasPHI(const PHINode *PN, uint64_t PNSize,
+ const MDNode *PNTBAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const MDNode *V2TBAAInfo) {
// If the values are PHIs in the same block, we can do a more precise
// as well as efficient check: just check for aliases between the values
// on corresponding edges.
if (const PHINode *PN2 = dyn_cast<PHINode>(V2))
if (PN2->getParent() == PN->getParent()) {
AliasResult Alias =
- aliasCheck(PN->getIncomingValue(0), PNSize,
+ aliasCheck(PN->getIncomingValue(0), PNSize, PNTBAAInfo,
PN2->getIncomingValueForBlock(PN->getIncomingBlock(0)),
- V2Size);
+ V2Size, V2TBAAInfo);
if (Alias == MayAlias)
return MayAlias;
for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) {
AliasResult ThisAlias =
- aliasCheck(PN->getIncomingValue(i), PNSize,
+ aliasCheck(PN->getIncomingValue(i), PNSize, PNTBAAInfo,
PN2->getIncomingValueForBlock(PN->getIncomingBlock(i)),
- V2Size);
- if (ThisAlias != Alias)
- return MayAlias;
+ V2Size, V2TBAAInfo);
+ Alias = MergeAliasResults(ThisAlias, Alias);
+ if (Alias == MayAlias)
+ break;
}
return Alias;
}
@@ -710,7 +1076,8 @@
V1Srcs.push_back(PV1);
}
- AliasResult Alias = aliasCheck(V2, V2Size, V1Srcs[0], PNSize);
+ AliasResult Alias = aliasCheck(V2, V2Size, V2TBAAInfo,
+ V1Srcs[0], PNSize, PNTBAAInfo);
// Early exit if the check of the first PHI source against V2 is MayAlias.
// Other results are not possible.
if (Alias == MayAlias)
@@ -721,14 +1088,11 @@
for (unsigned i = 1, e = V1Srcs.size(); i != e; ++i) {
Value *V = V1Srcs[i];
- // If V2 is visited, the recursive case will have been caught in the
- // above aliasCheck call, so these subsequent calls to aliasCheck
- // don't need to assume that V2 is being visited recursively.
- Visited.erase(V2);
-
- AliasResult ThisAlias = aliasCheck(V2, V2Size, V, PNSize);
- if (ThisAlias != Alias || ThisAlias == MayAlias)
- return MayAlias;
+ AliasResult ThisAlias = aliasCheck(V2, V2Size, V2TBAAInfo,
+ V, PNSize, PNTBAAInfo);
+ Alias = MergeAliasResults(ThisAlias, Alias);
+ if (Alias == MayAlias)
+ break;
}
return Alias;
@@ -738,8 +1102,10 @@
// such as array references.
//
AliasAnalysis::AliasResult
-BasicAliasAnalysis::aliasCheck(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
+BasicAliasAnalysis::aliasCheck(const Value *V1, uint64_t V1Size,
+ const MDNode *V1TBAAInfo,
+ const Value *V2, uint64_t V2Size,
+ const MDNode *V2TBAAInfo) {
// If either of the memory references is empty, it doesn't matter what the
// pointer values are.
if (V1Size == 0 || V2Size == 0)
@@ -756,8 +1122,8 @@
return NoAlias; // Scalars cannot alias each other
// Figure out what objects these things are pointing to if we can.
- const Value *O1 = V1->getUnderlyingObject();
- const Value *O2 = V2->getUnderlyingObject();
+ const Value *O1 = GetUnderlyingObject(V1, TD);
+ const Value *O2 = GetUnderlyingObject(V2, TD);
// Null values in the default address space don't point to any object, so they
// don't alias any other pointer.
@@ -811,6 +1177,17 @@
(V2Size != UnknownSize && isObjectSmallerThan(O1, V2Size, *TD)))
return NoAlias;
+ // Check the cache before climbing up use-def chains. This also terminates
+ // otherwise infinitely recursive queries.
+ LocPair Locs(Location(V1, V1Size, V1TBAAInfo),
+ Location(V2, V2Size, V2TBAAInfo));
+ if (V1 > V2)
+ std::swap(Locs.first, Locs.second);
+ std::pair<AliasCacheTy::iterator, bool> Pair =
+ AliasCache.insert(std::make_pair(Locs, MayAlias));
+ if (!Pair.second)
+ return Pair.first->second;
+
// FIXME: This isn't aggressively handling alias(GEP, PHI) for example: if the
// GEP can't simplify, we don't even look at the PHI cases.
if (!isa<GEPOperator>(V1) && isa<GEPOperator>(V2)) {
@@ -818,25 +1195,41 @@
std::swap(V1Size, V2Size);
std::swap(O1, O2);
}
- if (const GEPOperator *GV1 = dyn_cast<GEPOperator>(V1))
- return aliasGEP(GV1, V1Size, V2, V2Size, O1, O2);
+ if (const GEPOperator *GV1 = dyn_cast<GEPOperator>(V1)) {
+ AliasResult Result = aliasGEP(GV1, V1Size, V2, V2Size, V2TBAAInfo, O1, O2);
+ if (Result != MayAlias) return AliasCache[Locs] = Result;
+ }
if (isa<PHINode>(V2) && !isa<PHINode>(V1)) {
std::swap(V1, V2);
std::swap(V1Size, V2Size);
}
- if (const PHINode *PN = dyn_cast<PHINode>(V1))
- return aliasPHI(PN, V1Size, V2, V2Size);
+ if (const PHINode *PN = dyn_cast<PHINode>(V1)) {
+ AliasResult Result = aliasPHI(PN, V1Size, V1TBAAInfo,
+ V2, V2Size, V2TBAAInfo);
+ if (Result != MayAlias) return AliasCache[Locs] = Result;
+ }
if (isa<SelectInst>(V2) && !isa<SelectInst>(V1)) {
std::swap(V1, V2);
std::swap(V1Size, V2Size);
}
- if (const SelectInst *S1 = dyn_cast<SelectInst>(V1))
- return aliasSelect(S1, V1Size, V2, V2Size);
+ if (const SelectInst *S1 = dyn_cast<SelectInst>(V1)) {
+ AliasResult Result = aliasSelect(S1, V1Size, V1TBAAInfo,
+ V2, V2Size, V2TBAAInfo);
+ if (Result != MayAlias) return AliasCache[Locs] = Result;
+ }
- return NoAA::alias(V1, V1Size, V2, V2Size);
+ // If both pointers are pointing into the same object and one of them
+ // accesses is accessing the entire object, then the accesses must
+ // overlap in some way.
+ if (TD && O1 == O2)
+ if ((V1Size != UnknownSize && isObjectSize(O1, V1Size, *TD)) ||
+ (V2Size != UnknownSize && isObjectSize(O2, V2Size, *TD)))
+ return AliasCache[Locs] = PartialAlias;
+
+ AliasResult Result =
+ AliasAnalysis::alias(Location(V1, V1Size, V1TBAAInfo),
+ Location(V2, V2Size, V2TBAAInfo));
+ return AliasCache[Locs] = Result;
}
-
-// Make sure that anything that uses AliasAnalysis pulls in this file.
-DEFINING_FILE_FOR(BasicAliasAnalysis)
diff --git a/src/LLVM/lib/Analysis/BlockFrequencyInfo.cpp b/src/LLVM/lib/Analysis/BlockFrequencyInfo.cpp
new file mode 100644
index 0000000..d16665f
--- /dev/null
+++ b/src/LLVM/lib/Analysis/BlockFrequencyInfo.cpp
@@ -0,0 +1,63 @@
+//=======-------- BlockFrequencyInfo.cpp - Block Frequency Analysis -------=======//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Loops should be simplified before this analysis.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/InitializePasses.h"
+#include "llvm/Analysis/BlockFrequencyImpl.h"
+#include "llvm/Analysis/BlockFrequencyInfo.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/BranchProbabilityInfo.h"
+
+using namespace llvm;
+
+INITIALIZE_PASS_BEGIN(BlockFrequencyInfo, "block-freq", "Block Frequency Analysis",
+ true, true)
+INITIALIZE_PASS_DEPENDENCY(BranchProbabilityInfo)
+INITIALIZE_PASS_END(BlockFrequencyInfo, "block-freq", "Block Frequency Analysis",
+ true, true)
+
+char BlockFrequencyInfo::ID = 0;
+
+
+BlockFrequencyInfo::BlockFrequencyInfo() : FunctionPass(ID) {
+ initializeBlockFrequencyInfoPass(*PassRegistry::getPassRegistry());
+ BFI = new BlockFrequencyImpl<BasicBlock, Function, BranchProbabilityInfo>();
+}
+
+BlockFrequencyInfo::~BlockFrequencyInfo() {
+ delete BFI;
+}
+
+void BlockFrequencyInfo::getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.addRequired<BranchProbabilityInfo>();
+ AU.setPreservesAll();
+}
+
+bool BlockFrequencyInfo::runOnFunction(Function &F) {
+ BranchProbabilityInfo &BPI = getAnalysis<BranchProbabilityInfo>();
+ BFI->doFunction(&F, &BPI);
+ return false;
+}
+
+void BlockFrequencyInfo::print(raw_ostream &O, const Module *) const {
+ if (BFI) BFI->print(O);
+}
+
+/// getblockFreq - Return block frequency. Return 0 if we don't have the
+/// information. Please note that initial frequency is equal to 1024. It means
+/// that we should not rely on the value itself, but only on the comparison to
+/// the other block frequencies. We do this to avoid using of floating points.
+///
+BlockFrequency BlockFrequencyInfo::getBlockFreq(BasicBlock *BB) const {
+ return BFI->getBlockFreq(BB);
+}
diff --git a/src/LLVM/lib/Analysis/BranchProbabilityInfo.cpp b/src/LLVM/lib/Analysis/BranchProbabilityInfo.cpp
new file mode 100644
index 0000000..bde3b76
--- /dev/null
+++ b/src/LLVM/lib/Analysis/BranchProbabilityInfo.cpp
@@ -0,0 +1,511 @@
+//===-- BranchProbabilityInfo.cpp - Branch Probability Analysis -*- C++ -*-===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Loops should be simplified before this analysis.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Constants.h"
+#include "llvm/Instructions.h"
+#include "llvm/LLVMContext.h"
+#include "llvm/Metadata.h"
+#include "llvm/Analysis/BranchProbabilityInfo.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Support/Debug.h"
+
+using namespace llvm;
+
+INITIALIZE_PASS_BEGIN(BranchProbabilityInfo, "branch-prob",
+ "Branch Probability Analysis", false, true)
+INITIALIZE_PASS_DEPENDENCY(LoopInfo)
+INITIALIZE_PASS_END(BranchProbabilityInfo, "branch-prob",
+ "Branch Probability Analysis", false, true)
+
+char BranchProbabilityInfo::ID = 0;
+
+namespace {
+// Please note that BranchProbabilityAnalysis is not a FunctionPass.
+// It is created by BranchProbabilityInfo (which is a FunctionPass), which
+// provides a clear interface. Thanks to that, all heuristics and other
+// private methods are hidden in the .cpp file.
+class BranchProbabilityAnalysis {
+
+ typedef std::pair<const BasicBlock *, const BasicBlock *> Edge;
+
+ DenseMap<Edge, uint32_t> *Weights;
+
+ BranchProbabilityInfo *BP;
+
+ LoopInfo *LI;
+
+
+ // Weights are for internal use only. They are used by heuristics to help to
+ // estimate edges' probability. Example:
+ //
+ // Using "Loop Branch Heuristics" we predict weights of edges for the
+ // block BB2.
+ // ...
+ // |
+ // V
+ // BB1<-+
+ // | |
+ // | | (Weight = 124)
+ // V |
+ // BB2--+
+ // |
+ // | (Weight = 4)
+ // V
+ // BB3
+ //
+ // Probability of the edge BB2->BB1 = 124 / (124 + 4) = 0.96875
+ // Probability of the edge BB2->BB3 = 4 / (124 + 4) = 0.03125
+
+ static const uint32_t LBH_TAKEN_WEIGHT = 124;
+ static const uint32_t LBH_NONTAKEN_WEIGHT = 4;
+
+ static const uint32_t RH_TAKEN_WEIGHT = 24;
+ static const uint32_t RH_NONTAKEN_WEIGHT = 8;
+
+ static const uint32_t PH_TAKEN_WEIGHT = 20;
+ static const uint32_t PH_NONTAKEN_WEIGHT = 12;
+
+ static const uint32_t ZH_TAKEN_WEIGHT = 20;
+ static const uint32_t ZH_NONTAKEN_WEIGHT = 12;
+
+ // Standard weight value. Used when none of the heuristics set weight for
+ // the edge.
+ static const uint32_t NORMAL_WEIGHT = 16;
+
+ // Minimum weight of an edge. Please note, that weight is NEVER 0.
+ static const uint32_t MIN_WEIGHT = 1;
+
+ // Return TRUE if BB leads directly to a Return Instruction.
+ static bool isReturningBlock(BasicBlock *BB) {
+ SmallPtrSet<BasicBlock *, 8> Visited;
+
+ while (true) {
+ TerminatorInst *TI = BB->getTerminator();
+ if (isa<ReturnInst>(TI))
+ return true;
+
+ if (TI->getNumSuccessors() > 1)
+ break;
+
+ // It is unreachable block which we can consider as a return instruction.
+ if (TI->getNumSuccessors() == 0)
+ return true;
+
+ Visited.insert(BB);
+ BB = TI->getSuccessor(0);
+
+ // Stop if cycle is detected.
+ if (Visited.count(BB))
+ return false;
+ }
+
+ return false;
+ }
+
+ uint32_t getMaxWeightFor(BasicBlock *BB) const {
+ return UINT32_MAX / BB->getTerminator()->getNumSuccessors();
+ }
+
+public:
+ BranchProbabilityAnalysis(DenseMap<Edge, uint32_t> *W,
+ BranchProbabilityInfo *BP, LoopInfo *LI)
+ : Weights(W), BP(BP), LI(LI) {
+ }
+
+ // Metadata Weights
+ bool calcMetadataWeights(BasicBlock *BB);
+
+ // Return Heuristics
+ bool calcReturnHeuristics(BasicBlock *BB);
+
+ // Pointer Heuristics
+ bool calcPointerHeuristics(BasicBlock *BB);
+
+ // Loop Branch Heuristics
+ bool calcLoopBranchHeuristics(BasicBlock *BB);
+
+ // Zero Heurestics
+ bool calcZeroHeuristics(BasicBlock *BB);
+
+ bool runOnFunction(Function &F);
+};
+} // end anonymous namespace
+
+// Propagate existing explicit probabilities from either profile data or
+// 'expect' intrinsic processing.
+bool BranchProbabilityAnalysis::calcMetadataWeights(BasicBlock *BB) {
+ TerminatorInst *TI = BB->getTerminator();
+ if (TI->getNumSuccessors() == 1)
+ return false;
+ if (!isa<BranchInst>(TI) && !isa<SwitchInst>(TI))
+ return false;
+
+ MDNode *WeightsNode = TI->getMetadata(LLVMContext::MD_prof);
+ if (!WeightsNode)
+ return false;
+
+ // Ensure there are weights for all of the successors. Note that the first
+ // operand to the metadata node is a name, not a weight.
+ if (WeightsNode->getNumOperands() != TI->getNumSuccessors() + 1)
+ return false;
+
+ // Build up the final weights that will be used in a temporary buffer, but
+ // don't add them until all weihts are present. Each weight value is clamped
+ // to [1, getMaxWeightFor(BB)].
+ uint32_t WeightLimit = getMaxWeightFor(BB);
+ SmallVector<uint32_t, 2> Weights;
+ Weights.reserve(TI->getNumSuccessors());
+ for (unsigned i = 1, e = WeightsNode->getNumOperands(); i != e; ++i) {
+ ConstantInt *Weight = dyn_cast<ConstantInt>(WeightsNode->getOperand(i));
+ if (!Weight)
+ return false;
+ Weights.push_back(
+ std::max<uint32_t>(1, Weight->getLimitedValue(WeightLimit)));
+ }
+ assert(Weights.size() == TI->getNumSuccessors() && "Checked above");
+ for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
+ BP->setEdgeWeight(BB, TI->getSuccessor(i), Weights[i]);
+
+ return true;
+}
+
+// Calculate Edge Weights using "Return Heuristics". Predict a successor which
+// leads directly to Return Instruction will not be taken.
+bool BranchProbabilityAnalysis::calcReturnHeuristics(BasicBlock *BB){
+ if (BB->getTerminator()->getNumSuccessors() == 1)
+ return false;
+
+ SmallPtrSet<BasicBlock *, 4> ReturningEdges;
+ SmallPtrSet<BasicBlock *, 4> StayEdges;
+
+ for (succ_iterator I = succ_begin(BB), E = succ_end(BB); I != E; ++I) {
+ BasicBlock *Succ = *I;
+ if (isReturningBlock(Succ))
+ ReturningEdges.insert(Succ);
+ else
+ StayEdges.insert(Succ);
+ }
+
+ if (uint32_t numStayEdges = StayEdges.size()) {
+ uint32_t stayWeight = RH_TAKEN_WEIGHT / numStayEdges;
+ if (stayWeight < NORMAL_WEIGHT)
+ stayWeight = NORMAL_WEIGHT;
+
+ for (SmallPtrSet<BasicBlock *, 4>::iterator I = StayEdges.begin(),
+ E = StayEdges.end(); I != E; ++I)
+ BP->setEdgeWeight(BB, *I, stayWeight);
+ }
+
+ if (uint32_t numRetEdges = ReturningEdges.size()) {
+ uint32_t retWeight = RH_NONTAKEN_WEIGHT / numRetEdges;
+ if (retWeight < MIN_WEIGHT)
+ retWeight = MIN_WEIGHT;
+ for (SmallPtrSet<BasicBlock *, 4>::iterator I = ReturningEdges.begin(),
+ E = ReturningEdges.end(); I != E; ++I) {
+ BP->setEdgeWeight(BB, *I, retWeight);
+ }
+ }
+
+ return ReturningEdges.size() > 0;
+}
+
+// Calculate Edge Weights using "Pointer Heuristics". Predict a comparsion
+// between two pointer or pointer and NULL will fail.
+bool BranchProbabilityAnalysis::calcPointerHeuristics(BasicBlock *BB) {
+ BranchInst * BI = dyn_cast<BranchInst>(BB->getTerminator());
+ if (!BI || !BI->isConditional())
+ return false;
+
+ Value *Cond = BI->getCondition();
+ ICmpInst *CI = dyn_cast<ICmpInst>(Cond);
+ if (!CI || !CI->isEquality())
+ return false;
+
+ Value *LHS = CI->getOperand(0);
+
+ if (!LHS->getType()->isPointerTy())
+ return false;
+
+ assert(CI->getOperand(1)->getType()->isPointerTy());
+
+ BasicBlock *Taken = BI->getSuccessor(0);
+ BasicBlock *NonTaken = BI->getSuccessor(1);
+
+ // p != 0 -> isProb = true
+ // p == 0 -> isProb = false
+ // p != q -> isProb = true
+ // p == q -> isProb = false;
+ bool isProb = CI->getPredicate() == ICmpInst::ICMP_NE;
+ if (!isProb)
+ std::swap(Taken, NonTaken);
+
+ BP->setEdgeWeight(BB, Taken, PH_TAKEN_WEIGHT);
+ BP->setEdgeWeight(BB, NonTaken, PH_NONTAKEN_WEIGHT);
+ return true;
+}
+
+// Calculate Edge Weights using "Loop Branch Heuristics". Predict backedges
+// as taken, exiting edges as not-taken.
+bool BranchProbabilityAnalysis::calcLoopBranchHeuristics(BasicBlock *BB) {
+ uint32_t numSuccs = BB->getTerminator()->getNumSuccessors();
+
+ Loop *L = LI->getLoopFor(BB);
+ if (!L)
+ return false;
+
+ SmallPtrSet<BasicBlock *, 8> BackEdges;
+ SmallPtrSet<BasicBlock *, 8> ExitingEdges;
+ SmallPtrSet<BasicBlock *, 8> InEdges; // Edges from header to the loop.
+
+ bool isHeader = BB == L->getHeader();
+
+ for (succ_iterator I = succ_begin(BB), E = succ_end(BB); I != E; ++I) {
+ BasicBlock *Succ = *I;
+ Loop *SuccL = LI->getLoopFor(Succ);
+ if (SuccL != L)
+ ExitingEdges.insert(Succ);
+ else if (Succ == L->getHeader())
+ BackEdges.insert(Succ);
+ else if (isHeader)
+ InEdges.insert(Succ);
+ }
+
+ if (uint32_t numBackEdges = BackEdges.size()) {
+ uint32_t backWeight = LBH_TAKEN_WEIGHT / numBackEdges;
+ if (backWeight < NORMAL_WEIGHT)
+ backWeight = NORMAL_WEIGHT;
+
+ for (SmallPtrSet<BasicBlock *, 8>::iterator EI = BackEdges.begin(),
+ EE = BackEdges.end(); EI != EE; ++EI) {
+ BasicBlock *Back = *EI;
+ BP->setEdgeWeight(BB, Back, backWeight);
+ }
+ }
+
+ if (uint32_t numInEdges = InEdges.size()) {
+ uint32_t inWeight = LBH_TAKEN_WEIGHT / numInEdges;
+ if (inWeight < NORMAL_WEIGHT)
+ inWeight = NORMAL_WEIGHT;
+
+ for (SmallPtrSet<BasicBlock *, 8>::iterator EI = InEdges.begin(),
+ EE = InEdges.end(); EI != EE; ++EI) {
+ BasicBlock *Back = *EI;
+ BP->setEdgeWeight(BB, Back, inWeight);
+ }
+ }
+
+ uint32_t numExitingEdges = ExitingEdges.size();
+ if (uint32_t numNonExitingEdges = numSuccs - numExitingEdges) {
+ uint32_t exitWeight = LBH_NONTAKEN_WEIGHT / numNonExitingEdges;
+ if (exitWeight < MIN_WEIGHT)
+ exitWeight = MIN_WEIGHT;
+
+ for (SmallPtrSet<BasicBlock *, 8>::iterator EI = ExitingEdges.begin(),
+ EE = ExitingEdges.end(); EI != EE; ++EI) {
+ BasicBlock *Exiting = *EI;
+ BP->setEdgeWeight(BB, Exiting, exitWeight);
+ }
+ }
+
+ return true;
+}
+
+bool BranchProbabilityAnalysis::calcZeroHeuristics(BasicBlock *BB) {
+ BranchInst * BI = dyn_cast<BranchInst>(BB->getTerminator());
+ if (!BI || !BI->isConditional())
+ return false;
+
+ Value *Cond = BI->getCondition();
+ ICmpInst *CI = dyn_cast<ICmpInst>(Cond);
+ if (!CI)
+ return false;
+
+ Value *RHS = CI->getOperand(1);
+ ConstantInt *CV = dyn_cast<ConstantInt>(RHS);
+ if (!CV)
+ return false;
+
+ bool isProb;
+ if (CV->isZero()) {
+ switch (CI->getPredicate()) {
+ case CmpInst::ICMP_EQ:
+ // X == 0 -> Unlikely
+ isProb = false;
+ break;
+ case CmpInst::ICMP_NE:
+ // X != 0 -> Likely
+ isProb = true;
+ break;
+ case CmpInst::ICMP_SLT:
+ // X < 0 -> Unlikely
+ isProb = false;
+ break;
+ case CmpInst::ICMP_SGT:
+ // X > 0 -> Likely
+ isProb = true;
+ break;
+ default:
+ return false;
+ }
+ } else if (CV->isOne() && CI->getPredicate() == CmpInst::ICMP_SLT) {
+ // InstCombine canonicalizes X <= 0 into X < 1.
+ // X <= 0 -> Unlikely
+ isProb = false;
+ } else if (CV->isAllOnesValue() && CI->getPredicate() == CmpInst::ICMP_SGT) {
+ // InstCombine canonicalizes X >= 0 into X > -1.
+ // X >= 0 -> Likely
+ isProb = true;
+ } else {
+ return false;
+ }
+
+ BasicBlock *Taken = BI->getSuccessor(0);
+ BasicBlock *NonTaken = BI->getSuccessor(1);
+
+ if (!isProb)
+ std::swap(Taken, NonTaken);
+
+ BP->setEdgeWeight(BB, Taken, ZH_TAKEN_WEIGHT);
+ BP->setEdgeWeight(BB, NonTaken, ZH_NONTAKEN_WEIGHT);
+
+ return true;
+}
+
+
+bool BranchProbabilityAnalysis::runOnFunction(Function &F) {
+
+ for (Function::iterator I = F.begin(), E = F.end(); I != E; ) {
+ BasicBlock *BB = I++;
+
+ if (calcMetadataWeights(BB))
+ continue;
+
+ if (calcLoopBranchHeuristics(BB))
+ continue;
+
+ if (calcReturnHeuristics(BB))
+ continue;
+
+ if (calcPointerHeuristics(BB))
+ continue;
+
+ calcZeroHeuristics(BB);
+ }
+
+ return false;
+}
+
+void BranchProbabilityInfo::getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.addRequired<LoopInfo>();
+ AU.setPreservesAll();
+}
+
+bool BranchProbabilityInfo::runOnFunction(Function &F) {
+ LoopInfo &LI = getAnalysis<LoopInfo>();
+ BranchProbabilityAnalysis BPA(&Weights, this, &LI);
+ return BPA.runOnFunction(F);
+}
+
+uint32_t BranchProbabilityInfo::getSumForBlock(const BasicBlock *BB) const {
+ uint32_t Sum = 0;
+
+ for (succ_const_iterator I = succ_begin(BB), E = succ_end(BB); I != E; ++I) {
+ const BasicBlock *Succ = *I;
+ uint32_t Weight = getEdgeWeight(BB, Succ);
+ uint32_t PrevSum = Sum;
+
+ Sum += Weight;
+ assert(Sum > PrevSum); (void) PrevSum;
+ }
+
+ return Sum;
+}
+
+bool BranchProbabilityInfo::
+isEdgeHot(const BasicBlock *Src, const BasicBlock *Dst) const {
+ // Hot probability is at least 4/5 = 80%
+ uint32_t Weight = getEdgeWeight(Src, Dst);
+ uint32_t Sum = getSumForBlock(Src);
+
+ // FIXME: Implement BranchProbability::compare then change this code to
+ // compare this BranchProbability against a static "hot" BranchProbability.
+ return (uint64_t)Weight * 5 > (uint64_t)Sum * 4;
+}
+
+BasicBlock *BranchProbabilityInfo::getHotSucc(BasicBlock *BB) const {
+ uint32_t Sum = 0;
+ uint32_t MaxWeight = 0;
+ BasicBlock *MaxSucc = 0;
+
+ for (succ_iterator I = succ_begin(BB), E = succ_end(BB); I != E; ++I) {
+ BasicBlock *Succ = *I;
+ uint32_t Weight = getEdgeWeight(BB, Succ);
+ uint32_t PrevSum = Sum;
+
+ Sum += Weight;
+ assert(Sum > PrevSum); (void) PrevSum;
+
+ if (Weight > MaxWeight) {
+ MaxWeight = Weight;
+ MaxSucc = Succ;
+ }
+ }
+
+ // FIXME: Use BranchProbability::compare.
+ if ((uint64_t)MaxWeight * 5 > (uint64_t)Sum * 4)
+ return MaxSucc;
+
+ return 0;
+}
+
+// Return edge's weight. If can't find it, return DEFAULT_WEIGHT value.
+uint32_t BranchProbabilityInfo::
+getEdgeWeight(const BasicBlock *Src, const BasicBlock *Dst) const {
+ Edge E(Src, Dst);
+ DenseMap<Edge, uint32_t>::const_iterator I = Weights.find(E);
+
+ if (I != Weights.end())
+ return I->second;
+
+ return DEFAULT_WEIGHT;
+}
+
+void BranchProbabilityInfo::
+setEdgeWeight(const BasicBlock *Src, const BasicBlock *Dst, uint32_t Weight) {
+ Weights[std::make_pair(Src, Dst)] = Weight;
+ DEBUG(dbgs() << "set edge " << Src->getNameStr() << " -> "
+ << Dst->getNameStr() << " weight to " << Weight
+ << (isEdgeHot(Src, Dst) ? " [is HOT now]\n" : "\n"));
+}
+
+
+BranchProbability BranchProbabilityInfo::
+getEdgeProbability(const BasicBlock *Src, const BasicBlock *Dst) const {
+
+ uint32_t N = getEdgeWeight(Src, Dst);
+ uint32_t D = getSumForBlock(Src);
+
+ return BranchProbability(N, D);
+}
+
+raw_ostream &
+BranchProbabilityInfo::printEdgeProbability(raw_ostream &OS, BasicBlock *Src,
+ BasicBlock *Dst) const {
+
+ const BranchProbability Prob = getEdgeProbability(Src, Dst);
+ OS << "edge " << Src->getNameStr() << " -> " << Dst->getNameStr()
+ << " probability is " << Prob
+ << (isEdgeHot(Src, Dst) ? " [HOT edge]\n" : "\n");
+
+ return OS;
+}
diff --git a/src/LLVM/lib/Analysis/CFGPrinter.cpp b/src/LLVM/lib/Analysis/CFGPrinter.cpp
new file mode 100644
index 0000000..7bb063f
--- /dev/null
+++ b/src/LLVM/lib/Analysis/CFGPrinter.cpp
@@ -0,0 +1,165 @@
+//===- CFGPrinter.cpp - DOT printer for the control flow graph ------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines a '-dot-cfg' analysis pass, which emits the
+// cfg.<fnname>.dot file for each function in the program, with a graph of the
+// CFG for that function.
+//
+// The other main feature of this file is that it implements the
+// Function::viewCFG method, which is useful for debugging passes which operate
+// on the CFG.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/CFGPrinter.h"
+
+#include "llvm/Pass.h"
+using namespace llvm;
+
+namespace {
+ struct CFGViewer : public FunctionPass {
+ static char ID; // Pass identifcation, replacement for typeid
+ CFGViewer() : FunctionPass(ID) {
+ initializeCFGOnlyViewerPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnFunction(Function &F) {
+ F.viewCFG();
+ return false;
+ }
+
+ void print(raw_ostream &OS, const Module* = 0) const {}
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+ };
+}
+
+char CFGViewer::ID = 0;
+INITIALIZE_PASS(CFGViewer, "view-cfg", "View CFG of function", false, true)
+
+namespace {
+ struct CFGOnlyViewer : public FunctionPass {
+ static char ID; // Pass identifcation, replacement for typeid
+ CFGOnlyViewer() : FunctionPass(ID) {
+ initializeCFGOnlyViewerPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnFunction(Function &F) {
+ F.viewCFGOnly();
+ return false;
+ }
+
+ void print(raw_ostream &OS, const Module* = 0) const {}
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+ };
+}
+
+char CFGOnlyViewer::ID = 0;
+INITIALIZE_PASS(CFGOnlyViewer, "view-cfg-only",
+ "View CFG of function (with no function bodies)", false, true)
+
+namespace {
+ struct CFGPrinter : public FunctionPass {
+ static char ID; // Pass identification, replacement for typeid
+ CFGPrinter() : FunctionPass(ID) {
+ initializeCFGPrinterPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnFunction(Function &F) {
+ std::string Filename = "cfg." + F.getNameStr() + ".dot";
+ errs() << "Writing '" << Filename << "'...";
+
+ std::string ErrorInfo;
+ raw_fd_ostream File(Filename.c_str(), ErrorInfo);
+
+ if (ErrorInfo.empty())
+ WriteGraph(File, (const Function*)&F);
+ else
+ errs() << " error opening file for writing!";
+ errs() << "\n";
+ return false;
+ }
+
+ void print(raw_ostream &OS, const Module* = 0) const {}
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+ };
+}
+
+char CFGPrinter::ID = 0;
+INITIALIZE_PASS(CFGPrinter, "dot-cfg", "Print CFG of function to 'dot' file",
+ false, true)
+
+namespace {
+ struct CFGOnlyPrinter : public FunctionPass {
+ static char ID; // Pass identification, replacement for typeid
+ CFGOnlyPrinter() : FunctionPass(ID) {
+ initializeCFGOnlyPrinterPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnFunction(Function &F) {
+ std::string Filename = "cfg." + F.getNameStr() + ".dot";
+ errs() << "Writing '" << Filename << "'...";
+
+ std::string ErrorInfo;
+ raw_fd_ostream File(Filename.c_str(), ErrorInfo);
+
+ if (ErrorInfo.empty())
+ WriteGraph(File, (const Function*)&F, true);
+ else
+ errs() << " error opening file for writing!";
+ errs() << "\n";
+ return false;
+ }
+ void print(raw_ostream &OS, const Module* = 0) const {}
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+ };
+}
+
+char CFGOnlyPrinter::ID = 0;
+INITIALIZE_PASS(CFGOnlyPrinter, "dot-cfg-only",
+ "Print CFG of function to 'dot' file (with no function bodies)",
+ false, true)
+
+/// viewCFG - This function is meant for use from the debugger. You can just
+/// say 'call F->viewCFG()' and a ghostview window should pop up from the
+/// program, displaying the CFG of the current function. This depends on there
+/// being a 'dot' and 'gv' program in your path.
+///
+void Function::viewCFG() const {
+ ViewGraph(this, "cfg" + getNameStr());
+}
+
+/// viewCFGOnly - This function is meant for use from the debugger. It works
+/// just like viewCFG, but it does not include the contents of basic blocks
+/// into the nodes, just the label. If you are only interested in the CFG t
+/// his can make the graph smaller.
+///
+void Function::viewCFGOnly() const {
+ ViewGraph(this, "cfg" + getNameStr(), true);
+}
+
+FunctionPass *llvm::createCFGPrinterPass () {
+ return new CFGPrinter();
+}
+
+FunctionPass *llvm::createCFGOnlyPrinterPass () {
+ return new CFGOnlyPrinter();
+}
+
diff --git a/src/LLVM/lib/Analysis/CaptureTracking.cpp b/src/LLVM/lib/Analysis/CaptureTracking.cpp
index 1928d46..b2c27d1 100644
--- a/src/LLVM/lib/Analysis/CaptureTracking.cpp
+++ b/src/LLVM/lib/Analysis/CaptureTracking.cpp
@@ -17,6 +17,7 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/CaptureTracking.h"
+#include "llvm/Constants.h"
#include "llvm/Instructions.h"
#include "llvm/Value.h"
#include "llvm/Analysis/AliasAnalysis.h"
@@ -68,9 +69,36 @@
switch (I->getOpcode()) {
case Instruction::Call:
+ case Instruction::Invoke: {
+ CallSite CS(I);
+ // Not captured if the callee is readonly, doesn't return a copy through
+ // its return value and doesn't unwind (a readonly function can leak bits
+ // by throwing an exception or not depending on the input value).
+ if (CS.onlyReadsMemory() && CS.doesNotThrow() && I->getType()->isVoidTy())
+ break;
+
+ // Not captured if only passed via 'nocapture' arguments. Note that
+ // calling a function pointer does not in itself cause the pointer to
+ // be captured. This is a subtle point considering that (for example)
+ // the callee might return its own address. It is analogous to saying
+ // that loading a value from a pointer does not cause the pointer to be
+ // captured, even though the loaded value might be the pointer itself
+ // (think of self-referential objects).
+ CallSite::arg_iterator B = CS.arg_begin(), E = CS.arg_end();
+ for (CallSite::arg_iterator A = B; A != E; ++A)
+ if (A->get() == V && !CS.paramHasAttr(A - B + 1, Attribute::NoCapture))
+ // The parameter is not marked 'nocapture' - captured.
+ return true;
+ // Only passed via 'nocapture' arguments, or is the called function - not
+ // captured.
+ break;
+ }
case Instruction::Load:
// Loading from a pointer does not cause it to be captured.
break;
+ case Instruction::VAArg:
+ // "va-arg" from a pointer does not cause it to be captured.
+ break;
case Instruction::Ret:
if (ReturnCaptures)
return true;
diff --git a/src/LLVM/lib/Analysis/ConstantFolding.cpp b/src/LLVM/lib/Analysis/ConstantFolding.cpp
index 0bf7967..df79849 100644
--- a/src/LLVM/lib/Analysis/ConstantFolding.cpp
+++ b/src/LLVM/lib/Analysis/ConstantFolding.cpp
@@ -23,6 +23,7 @@
#include "llvm/GlobalVariable.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
+#include "llvm/Operator.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Target/TargetData.h"
#include "llvm/ADT/SmallVector.h"
@@ -30,6 +31,7 @@
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
+#include "llvm/Support/FEnv.h"
#include <cerrno>
#include <cmath>
using namespace llvm;
@@ -41,11 +43,16 @@
/// FoldBitCast - Constant fold bitcast, symbolically evaluating it with
/// TargetData. This always returns a non-null constant, but it may be a
/// ConstantExpr if unfoldable.
-static Constant *FoldBitCast(Constant *C, const Type *DestTy,
+static Constant *FoldBitCast(Constant *C, Type *DestTy,
const TargetData &TD) {
-
- // This only handles casts to vectors currently.
- const VectorType *DestVTy = dyn_cast<VectorType>(DestTy);
+ // Catch the obvious splat cases.
+ if (C->isNullValue() && !DestTy->isX86_MMXTy())
+ return Constant::getNullValue(DestTy);
+ if (C->isAllOnesValue() && !DestTy->isX86_MMXTy())
+ return Constant::getAllOnesValue(DestTy);
+
+ // The code below only handles casts to vectors currently.
+ VectorType *DestVTy = dyn_cast<VectorType>(DestTy);
if (DestVTy == 0)
return ConstantExpr::getBitCast(C, DestTy);
@@ -53,7 +60,7 @@
// vector so the code below can handle it uniformly.
if (isa<ConstantFP>(C) || isa<ConstantInt>(C)) {
Constant *Ops = C; // don't take the address of C!
- return FoldBitCast(ConstantVector::get(&Ops, 1), DestTy, TD);
+ return FoldBitCast(ConstantVector::get(Ops), DestTy, TD);
}
// If this is a bitcast from constant vector -> vector, fold it.
@@ -67,8 +74,8 @@
if (NumDstElt == NumSrcElt)
return ConstantExpr::getBitCast(C, DestTy);
- const Type *SrcEltTy = CV->getType()->getElementType();
- const Type *DstEltTy = DestVTy->getElementType();
+ Type *SrcEltTy = CV->getType()->getElementType();
+ Type *DstEltTy = DestVTy->getElementType();
// Otherwise, we're changing the number of elements in a vector, which
// requires endianness information to do the right thing. For example,
@@ -83,7 +90,7 @@
if (DstEltTy->isFloatingPointTy()) {
// Fold to an vector of integers with same size as our FP type.
unsigned FPWidth = DstEltTy->getPrimitiveSizeInBits();
- const Type *DestIVTy =
+ Type *DestIVTy =
VectorType::get(IntegerType::get(C->getContext(), FPWidth), NumDstElt);
// Recursively handle this integer conversion, if possible.
C = FoldBitCast(C, DestIVTy, TD);
@@ -97,7 +104,7 @@
// it to integer first.
if (SrcEltTy->isFloatingPointTy()) {
unsigned FPWidth = SrcEltTy->getPrimitiveSizeInBits();
- const Type *SrcIVTy =
+ Type *SrcIVTy =
VectorType::get(IntegerType::get(C->getContext(), FPWidth), NumSrcElt);
// Ask VMCore to do the conversion now that #elts line up.
C = ConstantExpr::getBitCast(C, SrcIVTy);
@@ -166,7 +173,7 @@
}
}
- return ConstantVector::get(Result.data(), Result.size());
+ return ConstantVector::get(Result);
}
@@ -210,11 +217,11 @@
if (!CI) return false; // Index isn't a simple constant?
if (CI->isZero()) continue; // Not adding anything.
- if (const StructType *ST = dyn_cast<StructType>(*GTI)) {
+ if (StructType *ST = dyn_cast<StructType>(*GTI)) {
// N = N + Offset
Offset += TD.getStructLayout(ST)->getElementOffset(CI->getZExtValue());
} else {
- const SequentialType *SQT = cast<SequentialType>(*GTI);
+ SequentialType *SQT = cast<SequentialType>(*GTI);
Offset += TD.getTypeAllocSize(SQT->getElementType())*CI->getSExtValue();
}
}
@@ -339,14 +346,21 @@
return true;
}
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
+ if (CE->getOpcode() == Instruction::IntToPtr &&
+ CE->getOperand(0)->getType() == TD.getIntPtrType(CE->getContext()))
+ return ReadDataFromGlobal(CE->getOperand(0), ByteOffset, CurPtr,
+ BytesLeft, TD);
+ }
+
// Otherwise, unknown initializer type.
return false;
}
static Constant *FoldReinterpretLoadFromConstPtr(Constant *C,
const TargetData &TD) {
- const Type *LoadTy = cast<PointerType>(C->getType())->getElementType();
- const IntegerType *IntType = dyn_cast<IntegerType>(LoadTy);
+ Type *LoadTy = cast<PointerType>(C->getType())->getElementType();
+ IntegerType *IntType = dyn_cast<IntegerType>(LoadTy);
// If this isn't an integer load we can't fold it directly.
if (!IntType) {
@@ -354,7 +368,7 @@
// and then bitcast the result. This can be useful for union cases. Note
// that address spaces don't matter here since we're not going to result in
// an actual new load.
- const Type *MapTy;
+ Type *MapTy;
if (LoadTy->isFloatTy())
MapTy = Type::getInt32PtrTy(C->getContext());
else if (LoadTy->isDoubleTy())
@@ -434,7 +448,7 @@
std::string Str;
if (TD && GetConstantStringInfo(CE, Str) && !Str.empty()) {
unsigned StrLen = Str.length();
- const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
+ Type *Ty = cast<PointerType>(CE->getType())->getElementType();
unsigned NumBits = Ty->getPrimitiveSizeInBits();
// Replace load with immediate integer if the result is an integer or fp
// value.
@@ -466,9 +480,10 @@
// If this load comes from anywhere in a constant global, and if the global
// is all undef or zero, we know what it loads.
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(CE->getUnderlyingObject())){
+ if (GlobalVariable *GV =
+ dyn_cast<GlobalVariable>(GetUnderlyingObject(CE, TD))) {
if (GV->isConstant() && GV->hasDefinitiveInitializer()) {
- const Type *ResTy = cast<PointerType>(C->getType())->getElementType();
+ Type *ResTy = cast<PointerType>(C->getType())->getElementType();
if (GV->getInitializer()->isNullValue())
return Constant::getNullValue(ResTy);
if (isa<UndefValue>(GV->getInitializer()))
@@ -526,19 +541,18 @@
/// CastGEPIndices - If array indices are not pointer-sized integers,
/// explicitly cast them so that they aren't implicitly casted by the
/// getelementptr.
-static Constant *CastGEPIndices(Constant *const *Ops, unsigned NumOps,
- const Type *ResultTy,
+static Constant *CastGEPIndices(ArrayRef<Constant *> Ops,
+ Type *ResultTy,
const TargetData *TD) {
if (!TD) return 0;
- const Type *IntPtrTy = TD->getIntPtrType(ResultTy->getContext());
+ Type *IntPtrTy = TD->getIntPtrType(ResultTy->getContext());
bool Any = false;
SmallVector<Constant*, 32> NewIdxs;
- for (unsigned i = 1; i != NumOps; ++i) {
+ for (unsigned i = 1, e = Ops.size(); i != e; ++i) {
if ((i == 1 ||
!isa<StructType>(GetElementPtrInst::getIndexedType(Ops[0]->getType(),
- reinterpret_cast<Value *const *>(Ops+1),
- i-1))) &&
+ Ops.slice(1, i-1)))) &&
Ops[i]->getType() != IntPtrTy) {
Any = true;
NewIdxs.push_back(ConstantExpr::getCast(CastInst::getCastOpcode(Ops[i],
@@ -552,7 +566,7 @@
if (!Any) return 0;
Constant *C =
- ConstantExpr::getGetElementPtr(Ops[0], &NewIdxs[0], NewIdxs.size());
+ ConstantExpr::getGetElementPtr(Ops[0], NewIdxs);
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
if (Constant *Folded = ConstantFoldConstantExpression(CE, TD))
C = Folded;
@@ -561,25 +575,45 @@
/// SymbolicallyEvaluateGEP - If we can symbolically evaluate the specified GEP
/// constant expression, do so.
-static Constant *SymbolicallyEvaluateGEP(Constant *const *Ops, unsigned NumOps,
- const Type *ResultTy,
+static Constant *SymbolicallyEvaluateGEP(ArrayRef<Constant *> Ops,
+ Type *ResultTy,
const TargetData *TD) {
Constant *Ptr = Ops[0];
if (!TD || !cast<PointerType>(Ptr->getType())->getElementType()->isSized())
return 0;
-
- unsigned BitWidth =
- TD->getTypeSizeInBits(TD->getIntPtrType(Ptr->getContext()));
+
+ Type *IntPtrTy = TD->getIntPtrType(Ptr->getContext());
// If this is a constant expr gep that is effectively computing an
// "offsetof", fold it into 'cast int Size to T*' instead of 'gep 0, 0, 12'
- for (unsigned i = 1; i != NumOps; ++i)
- if (!isa<ConstantInt>(Ops[i]))
+ for (unsigned i = 1, e = Ops.size(); i != e; ++i)
+ if (!isa<ConstantInt>(Ops[i])) {
+
+ // If this is "gep i8* Ptr, (sub 0, V)", fold this as:
+ // "inttoptr (sub (ptrtoint Ptr), V)"
+ if (Ops.size() == 2 &&
+ cast<PointerType>(ResultTy)->getElementType()->isIntegerTy(8)) {
+ ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[1]);
+ assert((CE == 0 || CE->getType() == IntPtrTy) &&
+ "CastGEPIndices didn't canonicalize index types!");
+ if (CE && CE->getOpcode() == Instruction::Sub &&
+ CE->getOperand(0)->isNullValue()) {
+ Constant *Res = ConstantExpr::getPtrToInt(Ptr, CE->getType());
+ Res = ConstantExpr::getSub(Res, CE->getOperand(1));
+ Res = ConstantExpr::getIntToPtr(Res, ResultTy);
+ if (ConstantExpr *ResCE = dyn_cast<ConstantExpr>(Res))
+ Res = ConstantFoldConstantExpression(ResCE, TD);
+ return Res;
+ }
+ }
return 0;
+ }
- APInt Offset = APInt(BitWidth,
- TD->getIndexedOffset(Ptr->getType(),
- (Value**)Ops+1, NumOps-1));
+ unsigned BitWidth = TD->getTypeSizeInBits(IntPtrTy);
+ APInt Offset =
+ APInt(BitWidth, TD->getIndexedOffset(Ptr->getType(),
+ makeArrayRef((Value **)Ops.data() + 1,
+ Ops.size() - 1)));
Ptr = cast<Constant>(Ptr->stripPointerCasts());
// If this is a GEP of a GEP, fold it all into a single GEP.
@@ -598,9 +632,7 @@
Ptr = cast<Constant>(GEP->getOperand(0));
Offset += APInt(BitWidth,
- TD->getIndexedOffset(Ptr->getType(),
- (Value**)NestedOps.data(),
- NestedOps.size()));
+ TD->getIndexedOffset(Ptr->getType(), NestedOps));
Ptr = cast<Constant>(Ptr->stripPointerCasts());
}
@@ -609,10 +641,8 @@
APInt BasePtr(BitWidth, 0);
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr))
if (CE->getOpcode() == Instruction::IntToPtr)
- if (ConstantInt *Base = dyn_cast<ConstantInt>(CE->getOperand(0))) {
- BasePtr = Base->getValue();
- BasePtr.zextOrTrunc(BitWidth);
- }
+ if (ConstantInt *Base = dyn_cast<ConstantInt>(CE->getOperand(0)))
+ BasePtr = Base->getValue().zextOrTrunc(BitWidth);
if (Ptr->isNullValue() || BasePtr != 0) {
Constant *C = ConstantInt::get(Ptr->getContext(), Offset+BasePtr);
return ConstantExpr::getIntToPtr(C, ResultTy);
@@ -622,10 +652,10 @@
// we eliminate over-indexing of the notional static type array bounds.
// This makes it easy to determine if the getelementptr is "inbounds".
// Also, this helps GlobalOpt do SROA on GlobalVariables.
- const Type *Ty = Ptr->getType();
+ Type *Ty = Ptr->getType();
SmallVector<Constant*, 32> NewIdxs;
do {
- if (const SequentialType *ATy = dyn_cast<SequentialType>(Ty)) {
+ if (SequentialType *ATy = dyn_cast<SequentialType>(Ty)) {
if (ATy->isPointerTy()) {
// The only pointer indexing we'll do is on the first index of the GEP.
if (!NewIdxs.empty())
@@ -638,14 +668,21 @@
// Determine which element of the array the offset points into.
APInt ElemSize(BitWidth, TD->getTypeAllocSize(ATy->getElementType()));
+ IntegerType *IntPtrTy = TD->getIntPtrType(Ty->getContext());
if (ElemSize == 0)
- return 0;
- APInt NewIdx = Offset.udiv(ElemSize);
- Offset -= NewIdx * ElemSize;
- NewIdxs.push_back(ConstantInt::get(TD->getIntPtrType(Ty->getContext()),
- NewIdx));
+ // The element size is 0. This may be [0 x Ty]*, so just use a zero
+ // index for this level and proceed to the next level to see if it can
+ // accommodate the offset.
+ NewIdxs.push_back(ConstantInt::get(IntPtrTy, 0));
+ else {
+ // The element size is non-zero divide the offset by the element
+ // size (rounding down), to compute the index at this level.
+ APInt NewIdx = Offset.udiv(ElemSize);
+ Offset -= NewIdx * ElemSize;
+ NewIdxs.push_back(ConstantInt::get(IntPtrTy, NewIdx));
+ }
Ty = ATy->getElementType();
- } else if (const StructType *STy = dyn_cast<StructType>(Ty)) {
+ } else if (StructType *STy = dyn_cast<StructType>(Ty)) {
// Determine which field of the struct the offset points into. The
// getZExtValue is at least as safe as the StructLayout API because we
// know the offset is within the struct at this point.
@@ -669,7 +706,7 @@
// Create a GEP.
Constant *C =
- ConstantExpr::getGetElementPtr(Ptr, &NewIdxs[0], NewIdxs.size());
+ ConstantExpr::getGetElementPtr(Ptr, NewIdxs);
assert(cast<PointerType>(C->getType())->getElementType() == Ty &&
"Computed GetElementPtr has unexpected type!");
@@ -687,27 +724,34 @@
// Constant Folding public APIs
//===----------------------------------------------------------------------===//
-
-/// ConstantFoldInstruction - Attempt to constant fold the specified
-/// instruction. If successful, the constant result is returned, if not, null
-/// is returned. Note that this function can only fail when attempting to fold
-/// instructions like loads and stores, which have no constant expression form.
-///
+/// ConstantFoldInstruction - Try to constant fold the specified instruction.
+/// If successful, the constant result is returned, if not, null is returned.
+/// Note that this fails if not all of the operands are constant. Otherwise,
+/// this function can only fail when attempting to fold instructions like loads
+/// and stores, which have no constant expression form.
Constant *llvm::ConstantFoldInstruction(Instruction *I, const TargetData *TD) {
+ // Handle PHI nodes quickly here...
if (PHINode *PN = dyn_cast<PHINode>(I)) {
- if (PN->getNumIncomingValues() == 0)
- return UndefValue::get(PN->getType());
+ Constant *CommonValue = 0;
- Constant *Result = dyn_cast<Constant>(PN->getIncomingValue(0));
- if (Result == 0) return 0;
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
+ Value *Incoming = PN->getIncomingValue(i);
+ // If the incoming value is undef then skip it. Note that while we could
+ // skip the value if it is equal to the phi node itself we choose not to
+ // because that would break the rule that constant folding only applies if
+ // all operands are constants.
+ if (isa<UndefValue>(Incoming))
+ continue;
+ // If the incoming value is not a constant, or is a different constant to
+ // the one we saw previously, then give up.
+ Constant *C = dyn_cast<Constant>(Incoming);
+ if (!C || (CommonValue && C != CommonValue))
+ return 0;
+ CommonValue = C;
+ }
- // Handle PHI nodes specially here...
- for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i)
- if (PN->getIncomingValue(i) != Result && PN->getIncomingValue(i) != PN)
- return 0; // Not all the same incoming constants...
-
- // If we reach here, all incoming values are the same constant.
- return Result;
+ // If we reach here, all incoming values are the same constant or undef.
+ return CommonValue ? CommonValue : UndefValue::get(PN->getType());
}
// Scan the operand list, checking to see if they are all constants, if so,
@@ -725,9 +769,19 @@
if (const LoadInst *LI = dyn_cast<LoadInst>(I))
return ConstantFoldLoadInst(LI, TD);
-
- return ConstantFoldInstOperands(I->getOpcode(), I->getType(),
- Ops.data(), Ops.size(), TD);
+
+ if (InsertValueInst *IVI = dyn_cast<InsertValueInst>(I))
+ return ConstantExpr::getInsertValue(
+ cast<Constant>(IVI->getAggregateOperand()),
+ cast<Constant>(IVI->getInsertedValueOperand()),
+ IVI->getIndices());
+
+ if (ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I))
+ return ConstantExpr::getExtractValue(
+ cast<Constant>(EVI->getAggregateOperand()),
+ EVI->getIndices());
+
+ return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Ops, TD);
}
/// ConstantFoldConstantExpression - Attempt to fold the constant expression
@@ -736,7 +790,8 @@
Constant *llvm::ConstantFoldConstantExpression(const ConstantExpr *CE,
const TargetData *TD) {
SmallVector<Constant*, 8> Ops;
- for (User::const_op_iterator i = CE->op_begin(), e = CE->op_end(); i != e; ++i) {
+ for (User::const_op_iterator i = CE->op_begin(), e = CE->op_end();
+ i != e; ++i) {
Constant *NewC = cast<Constant>(*i);
// Recursively fold the ConstantExpr's operands.
if (ConstantExpr *NewCE = dyn_cast<ConstantExpr>(NewC))
@@ -747,8 +802,7 @@
if (CE->isCompare())
return ConstantFoldCompareInstOperands(CE->getPredicate(), Ops[0], Ops[1],
TD);
- return ConstantFoldInstOperands(CE->getOpcode(), CE->getType(),
- Ops.data(), Ops.size(), TD);
+ return ConstantFoldInstOperands(CE->getOpcode(), CE->getType(), Ops, TD);
}
/// ConstantFoldInstOperands - Attempt to constant fold an instruction with the
@@ -761,8 +815,8 @@
/// information, due to only being passed an opcode and operands. Constant
/// folding using this function strips this information.
///
-Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, const Type *DestTy,
- Constant* const* Ops, unsigned NumOps,
+Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy,
+ ArrayRef<Constant *> Ops,
const TargetData *TD) {
// Handle easy binops first.
if (Instruction::isBinaryOp(Opcode)) {
@@ -778,9 +832,9 @@
case Instruction::ICmp:
case Instruction::FCmp: assert(0 && "Invalid for compares");
case Instruction::Call:
- if (Function *F = dyn_cast<Function>(Ops[NumOps - 1]))
+ if (Function *F = dyn_cast<Function>(Ops.back()))
if (canConstantFoldCallTo(F))
- return ConstantFoldCall(F, Ops, NumOps - 1);
+ return ConstantFoldCall(F, Ops.slice(0, Ops.size() - 1));
return 0;
case Instruction::PtrToInt:
// If the input is a inttoptr, eliminate the pair. This requires knowing
@@ -834,12 +888,12 @@
case Instruction::ShuffleVector:
return ConstantExpr::getShuffleVector(Ops[0], Ops[1], Ops[2]);
case Instruction::GetElementPtr:
- if (Constant *C = CastGEPIndices(Ops, NumOps, DestTy, TD))
+ if (Constant *C = CastGEPIndices(Ops, DestTy, TD))
return C;
- if (Constant *C = SymbolicallyEvaluateGEP(Ops, NumOps, DestTy, TD))
+ if (Constant *C = SymbolicallyEvaluateGEP(Ops, DestTy, TD))
return C;
- return ConstantExpr::getGetElementPtr(Ops[0], Ops+1, NumOps-1);
+ return ConstantExpr::getGetElementPtr(Ops[0], Ops.slice(1));
}
}
@@ -859,7 +913,7 @@
// around to know if bit truncation is happening.
if (ConstantExpr *CE0 = dyn_cast<ConstantExpr>(Ops0)) {
if (TD && Ops1->isNullValue()) {
- const Type *IntPtrTy = TD->getIntPtrType(CE0->getContext());
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getContext());
if (CE0->getOpcode() == Instruction::IntToPtr) {
// Convert the integer value to the right size to ensure we get the
// proper extension or truncation.
@@ -881,7 +935,7 @@
if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(Ops1)) {
if (TD && CE0->getOpcode() == CE1->getOpcode()) {
- const Type *IntPtrTy = TD->getIntPtrType(CE0->getContext());
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getContext());
if (CE0->getOpcode() == Instruction::IntToPtr) {
// Convert the integer value to the right size to ensure we get the
@@ -914,7 +968,7 @@
unsigned OpC =
Predicate == ICmpInst::ICMP_EQ ? Instruction::And : Instruction::Or;
Constant *Ops[] = { LHS, RHS };
- return ConstantFoldInstOperands(OpC, LHS->getType(), Ops, 2, TD);
+ return ConstantFoldInstOperands(OpC, LHS->getType(), Ops, TD);
}
}
@@ -934,7 +988,7 @@
// addressing...
gep_type_iterator I = gep_type_begin(CE), E = gep_type_end(CE);
for (++I; I != E; ++I)
- if (const StructType *STy = dyn_cast<StructType>(*I)) {
+ if (StructType *STy = dyn_cast<StructType>(*I)) {
ConstantInt *CU = cast<ConstantInt>(I.getOperand());
assert(CU->getZExtValue() < STy->getNumElements() &&
"Struct index out of range!");
@@ -949,7 +1003,7 @@
return 0;
}
} else if (ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand())) {
- if (const ArrayType *ATy = dyn_cast<ArrayType>(*I)) {
+ if (ArrayType *ATy = dyn_cast<ArrayType>(*I)) {
if (CI->getZExtValue() >= ATy->getNumElements())
return 0;
if (ConstantArray *CA = dyn_cast<ConstantArray>(C))
@@ -960,7 +1014,7 @@
C = UndefValue::get(ATy->getElementType());
else
return 0;
- } else if (const VectorType *VTy = dyn_cast<VectorType>(*I)) {
+ } else if (VectorType *VTy = dyn_cast<VectorType>(*I)) {
if (CI->getZExtValue() >= VTy->getNumElements())
return 0;
if (ConstantVector *CP = dyn_cast<ConstantVector>(C))
@@ -996,12 +1050,22 @@
case Intrinsic::ctpop:
case Intrinsic::ctlz:
case Intrinsic::cttz:
- case Intrinsic::uadd_with_overflow:
- case Intrinsic::usub_with_overflow:
case Intrinsic::sadd_with_overflow:
+ case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
+ case Intrinsic::usub_with_overflow:
+ case Intrinsic::smul_with_overflow:
+ case Intrinsic::umul_with_overflow:
case Intrinsic::convert_from_fp16:
case Intrinsic::convert_to_fp16:
+ case Intrinsic::x86_sse_cvtss2si:
+ case Intrinsic::x86_sse_cvtss2si64:
+ case Intrinsic::x86_sse_cvttss2si:
+ case Intrinsic::x86_sse_cvttss2si64:
+ case Intrinsic::x86_sse2_cvtsd2si:
+ case Intrinsic::x86_sse2_cvtsd2si64:
+ case Intrinsic::x86_sse2_cvttsd2si:
+ case Intrinsic::x86_sse2_cvttsd2si64:
return true;
default:
return false;
@@ -1022,7 +1086,7 @@
case 'c':
return Name == "cos" || Name == "ceil" || Name == "cosf" || Name == "cosh";
case 'e':
- return Name == "exp";
+ return Name == "exp" || Name == "exp2";
case 'f':
return Name == "fabs" || Name == "fmod" || Name == "floor";
case 'l':
@@ -1038,11 +1102,11 @@
}
static Constant *ConstantFoldFP(double (*NativeFP)(double), double V,
- const Type *Ty) {
- errno = 0;
+ Type *Ty) {
+ sys::llvm_fenv_clearexcept();
V = NativeFP(V);
- if (errno != 0) {
- errno = 0;
+ if (sys::llvm_fenv_testexcept()) {
+ sys::llvm_fenv_clearexcept();
return 0;
}
@@ -1055,11 +1119,11 @@
}
static Constant *ConstantFoldBinaryFP(double (*NativeFP)(double, double),
- double V, double W, const Type *Ty) {
- errno = 0;
+ double V, double W, Type *Ty) {
+ sys::llvm_fenv_clearexcept();
V = NativeFP(V, W);
- if (errno != 0) {
- errno = 0;
+ if (sys::llvm_fenv_testexcept()) {
+ sys::llvm_fenv_clearexcept();
return 0;
}
@@ -1071,18 +1135,47 @@
return 0; // dummy return to suppress warning
}
+/// ConstantFoldConvertToInt - Attempt to an SSE floating point to integer
+/// conversion of a constant floating point. If roundTowardZero is false, the
+/// default IEEE rounding is used (toward nearest, ties to even). This matches
+/// the behavior of the non-truncating SSE instructions in the default rounding
+/// mode. The desired integer type Ty is used to select how many bits are
+/// available for the result. Returns null if the conversion cannot be
+/// performed, otherwise returns the Constant value resulting from the
+/// conversion.
+static Constant *ConstantFoldConvertToInt(ConstantFP *Op, bool roundTowardZero,
+ Type *Ty) {
+ assert(Op && "Called with NULL operand");
+ APFloat Val(Op->getValueAPF());
+
+ // All of these conversion intrinsics form an integer of at most 64bits.
+ unsigned ResultWidth = cast<IntegerType>(Ty)->getBitWidth();
+ assert(ResultWidth <= 64 &&
+ "Can only constant fold conversions to 64 and 32 bit ints");
+
+ uint64_t UIntVal;
+ bool isExact = false;
+ APFloat::roundingMode mode = roundTowardZero? APFloat::rmTowardZero
+ : APFloat::rmNearestTiesToEven;
+ APFloat::opStatus status = Val.convertToInteger(&UIntVal, ResultWidth,
+ /*isSigned=*/true, mode,
+ &isExact);
+ if (status != APFloat::opOK && status != APFloat::opInexact)
+ return 0;
+ return ConstantInt::get(Ty, UIntVal, /*isSigned=*/true);
+}
+
/// ConstantFoldCall - Attempt to constant fold a call to the specified function
/// with the specified arguments, returning null if unsuccessful.
Constant *
-llvm::ConstantFoldCall(Function *F,
- Constant *const *Operands, unsigned NumOperands) {
+llvm::ConstantFoldCall(Function *F, ArrayRef<Constant *> Operands) {
if (!F->hasName()) return 0;
StringRef Name = F->getName();
- const Type *Ty = F->getReturnType();
- if (NumOperands == 1) {
+ Type *Ty = F->getReturnType();
+ if (Operands.size() == 1) {
if (ConstantFP *Op = dyn_cast<ConstantFP>(Operands[0])) {
- if (Name == "llvm.convert.to.fp16") {
+ if (F->getIntrinsicID() == Intrinsic::convert_to_fp16) {
APFloat Val(Op->getValueAPF());
bool lost = false;
@@ -1093,6 +1186,13 @@
if (!Ty->isFloatTy() && !Ty->isDoubleTy())
return 0;
+
+ /// We only fold functions with finite arguments. Folding NaN and inf is
+ /// likely to be aborted with an exception anyway, and some host libms
+ /// have known errors raising exceptions.
+ if (Op->getValueAPF().isNaN() || Op->getValueAPF().isInfinity())
+ return 0;
+
/// Currently APFloat versions of these functions do not exist, so we use
/// the host native double versions. Float versions are not called
/// directly but for all these it is true (float)(f((double)arg)) ==
@@ -1121,6 +1221,12 @@
case 'e':
if (Name == "exp")
return ConstantFoldFP(exp, V, Ty);
+
+ if (Name == "exp2") {
+ // Constant fold exp2(x) as pow(2,x) in case the host doesn't have a
+ // C99 library.
+ return ConstantFoldBinaryFP(pow, 2.0, V, Ty);
+ }
break;
case 'f':
if (Name == "fabs")
@@ -1133,8 +1239,8 @@
return ConstantFoldFP(log, V, Ty);
else if (Name == "log10" && V > 0)
return ConstantFoldFP(log10, V, Ty);
- else if (Name == "llvm.sqrt.f32" ||
- Name == "llvm.sqrt.f64") {
+ else if (F->getIntrinsicID() == Intrinsic::sqrt &&
+ (Ty->isFloatTy() || Ty->isDoubleTy())) {
if (V >= -0.0)
return ConstantFoldFP(sqrt, V, Ty);
else // Undefined
@@ -1164,18 +1270,18 @@
}
return 0;
}
-
-
+
if (ConstantInt *Op = dyn_cast<ConstantInt>(Operands[0])) {
- if (Name.startswith("llvm.bswap"))
+ switch (F->getIntrinsicID()) {
+ case Intrinsic::bswap:
return ConstantInt::get(F->getContext(), Op->getValue().byteSwap());
- else if (Name.startswith("llvm.ctpop"))
+ case Intrinsic::ctpop:
return ConstantInt::get(Ty, Op->getValue().countPopulation());
- else if (Name.startswith("llvm.cttz"))
+ case Intrinsic::cttz:
return ConstantInt::get(Ty, Op->getValue().countTrailingZeros());
- else if (Name.startswith("llvm.ctlz"))
+ case Intrinsic::ctlz:
return ConstantInt::get(Ty, Op->getValue().countLeadingZeros());
- else if (Name == "llvm.convert.from.fp16") {
+ case Intrinsic::convert_from_fp16: {
APFloat Val(Op->getValue());
bool lost = false;
@@ -1183,25 +1289,45 @@
Val.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven, &lost);
// Conversion is always precise.
- status = status;
+ (void)status;
assert(status == APFloat::opOK && !lost &&
"Precision lost during fp16 constfolding");
return ConstantFP::get(F->getContext(), Val);
}
- return 0;
+ default:
+ return 0;
+ }
}
-
+
+ if (ConstantVector *Op = dyn_cast<ConstantVector>(Operands[0])) {
+ switch (F->getIntrinsicID()) {
+ default: break;
+ case Intrinsic::x86_sse_cvtss2si:
+ case Intrinsic::x86_sse_cvtss2si64:
+ case Intrinsic::x86_sse2_cvtsd2si:
+ case Intrinsic::x86_sse2_cvtsd2si64:
+ if (ConstantFP *FPOp = dyn_cast<ConstantFP>(Op->getOperand(0)))
+ return ConstantFoldConvertToInt(FPOp, /*roundTowardZero=*/false, Ty);
+ case Intrinsic::x86_sse_cvttss2si:
+ case Intrinsic::x86_sse_cvttss2si64:
+ case Intrinsic::x86_sse2_cvttsd2si:
+ case Intrinsic::x86_sse2_cvttsd2si64:
+ if (ConstantFP *FPOp = dyn_cast<ConstantFP>(Op->getOperand(0)))
+ return ConstantFoldConvertToInt(FPOp, /*roundTowardZero=*/true, Ty);
+ }
+ }
+
if (isa<UndefValue>(Operands[0])) {
- if (Name.startswith("llvm.bswap"))
+ if (F->getIntrinsicID() == Intrinsic::bswap)
return Operands[0];
return 0;
}
return 0;
}
-
- if (NumOperands == 2) {
+
+ if (Operands.size() == 2) {
if (ConstantFP *Op1 = dyn_cast<ConstantFP>(Operands[0])) {
if (!Ty->isFloatTy() && !Ty->isDoubleTy())
return 0;
@@ -1223,11 +1349,11 @@
if (Name == "atan2")
return ConstantFoldBinaryFP(atan2, Op1V, Op2V, Ty);
} else if (ConstantInt *Op2C = dyn_cast<ConstantInt>(Operands[1])) {
- if (Name == "llvm.powi.f32")
+ if (F->getIntrinsicID() == Intrinsic::powi && Ty->isFloatTy())
return ConstantFP::get(F->getContext(),
APFloat((float)std::pow((float)Op1V,
(int)Op2C->getZExtValue())));
- if (Name == "llvm.powi.f64")
+ if (F->getIntrinsicID() == Intrinsic::powi && Ty->isDoubleTy())
return ConstantFP::get(F->getContext(),
APFloat((double)std::pow((double)Op1V,
(int)Op2C->getZExtValue())));
@@ -1240,41 +1366,40 @@
if (ConstantInt *Op2 = dyn_cast<ConstantInt>(Operands[1])) {
switch (F->getIntrinsicID()) {
default: break;
- case Intrinsic::uadd_with_overflow: {
- Constant *Res = ConstantExpr::getAdd(Op1, Op2); // result.
+ case Intrinsic::sadd_with_overflow:
+ case Intrinsic::uadd_with_overflow:
+ case Intrinsic::ssub_with_overflow:
+ case Intrinsic::usub_with_overflow:
+ case Intrinsic::smul_with_overflow:
+ case Intrinsic::umul_with_overflow: {
+ APInt Res;
+ bool Overflow;
+ switch (F->getIntrinsicID()) {
+ default: assert(0 && "Invalid case");
+ case Intrinsic::sadd_with_overflow:
+ Res = Op1->getValue().sadd_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::uadd_with_overflow:
+ Res = Op1->getValue().uadd_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::ssub_with_overflow:
+ Res = Op1->getValue().ssub_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::usub_with_overflow:
+ Res = Op1->getValue().usub_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::smul_with_overflow:
+ Res = Op1->getValue().smul_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::umul_with_overflow:
+ Res = Op1->getValue().umul_ov(Op2->getValue(), Overflow);
+ break;
+ }
Constant *Ops[] = {
- Res, ConstantExpr::getICmp(CmpInst::ICMP_ULT, Res, Op1) // overflow.
+ ConstantInt::get(F->getContext(), Res),
+ ConstantInt::get(Type::getInt1Ty(F->getContext()), Overflow)
};
- return ConstantStruct::get(F->getContext(), Ops, 2, false);
- }
- case Intrinsic::usub_with_overflow: {
- Constant *Res = ConstantExpr::getSub(Op1, Op2); // result.
- Constant *Ops[] = {
- Res, ConstantExpr::getICmp(CmpInst::ICMP_UGT, Res, Op1) // overflow.
- };
- return ConstantStruct::get(F->getContext(), Ops, 2, false);
- }
- case Intrinsic::sadd_with_overflow: {
- Constant *Res = ConstantExpr::getAdd(Op1, Op2); // result.
- Constant *Overflow = ConstantExpr::getSelect(
- ConstantExpr::getICmp(CmpInst::ICMP_SGT,
- ConstantInt::get(Op1->getType(), 0), Op1),
- ConstantExpr::getICmp(CmpInst::ICMP_SGT, Res, Op2),
- ConstantExpr::getICmp(CmpInst::ICMP_SLT, Res, Op2)); // overflow.
-
- Constant *Ops[] = { Res, Overflow };
- return ConstantStruct::get(F->getContext(), Ops, 2, false);
- }
- case Intrinsic::ssub_with_overflow: {
- Constant *Res = ConstantExpr::getSub(Op1, Op2); // result.
- Constant *Overflow = ConstantExpr::getSelect(
- ConstantExpr::getICmp(CmpInst::ICMP_SGT,
- ConstantInt::get(Op2->getType(), 0), Op2),
- ConstantExpr::getICmp(CmpInst::ICMP_SLT, Res, Op1),
- ConstantExpr::getICmp(CmpInst::ICMP_SGT, Res, Op1)); // overflow.
-
- Constant *Ops[] = { Res, Overflow };
- return ConstantStruct::get(F->getContext(), Ops, 2, false);
+ return ConstantStruct::get(cast<StructType>(F->getReturnType()), Ops);
}
}
}
@@ -1285,4 +1410,3 @@
}
return 0;
}
-
diff --git a/src/LLVM/lib/Analysis/DIBuilder.cpp b/src/LLVM/lib/Analysis/DIBuilder.cpp
new file mode 100644
index 0000000..bfa429d
--- /dev/null
+++ b/src/LLVM/lib/Analysis/DIBuilder.cpp
@@ -0,0 +1,939 @@
+//===--- DIBuilder.cpp - Debug Information Builder ------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the DIBuilder.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/DIBuilder.h"
+#include "llvm/Analysis/DebugInfo.h"
+#include "llvm/Constants.h"
+#include "llvm/IntrinsicInst.h"
+#include "llvm/Module.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/Support/Dwarf.h"
+
+using namespace llvm;
+using namespace llvm::dwarf;
+
+static Constant *GetTagConstant(LLVMContext &VMContext, unsigned Tag) {
+ assert((Tag & LLVMDebugVersionMask) == 0 &&
+ "Tag too large for debug encoding!");
+ return ConstantInt::get(Type::getInt32Ty(VMContext), Tag | LLVMDebugVersion);
+}
+
+DIBuilder::DIBuilder(Module &m)
+ : M(m), VMContext(M.getContext()), TheCU(0), TempEnumTypes(0),
+ TempRetainTypes(0), TempSubprograms(0), TempGVs(0), DeclareFn(0),
+ ValueFn(0)
+{}
+
+/// finalize - Construct any deferred debug info descriptors.
+void DIBuilder::finalize() {
+ DIArray Enums = getOrCreateArray(AllEnumTypes);
+ DIType(TempEnumTypes).replaceAllUsesWith(Enums);
+
+ DIArray RetainTypes = getOrCreateArray(AllRetainTypes);
+ DIType(TempRetainTypes).replaceAllUsesWith(RetainTypes);
+
+ DIArray SPs = getOrCreateArray(AllSubprograms);
+ DIType(TempSubprograms).replaceAllUsesWith(SPs);
+ for (unsigned i = 0, e = SPs.getNumElements(); i != e; ++i) {
+ DISubprogram SP(SPs.getElement(i));
+ if (NamedMDNode *NMD = getFnSpecificMDNode(M, SP)) {
+ SmallVector<Value *, 4> Variables;
+ for (unsigned ii = 0, ee = NMD->getNumOperands(); ii != ee; ++ii)
+ Variables.push_back(NMD->getOperand(ii));
+ if (MDNode *Temp = SP.getVariablesNodes()) {
+ DIArray AV = getOrCreateArray(Variables);
+ DIType(Temp).replaceAllUsesWith(AV);
+ }
+ NMD->eraseFromParent();
+ }
+ }
+
+ DIArray GVs = getOrCreateArray(AllGVs);
+ DIType(TempGVs).replaceAllUsesWith(GVs);
+}
+
+/// getNonCompileUnitScope - If N is compile unit return NULL otherwise return
+/// N.
+static MDNode *getNonCompileUnitScope(MDNode *N) {
+ if (DIDescriptor(N).isCompileUnit())
+ return NULL;
+ return N;
+}
+
+/// createCompileUnit - A CompileUnit provides an anchor for all debugging
+/// information generated during this instance of compilation.
+void DIBuilder::createCompileUnit(unsigned Lang, StringRef Filename,
+ StringRef Directory, StringRef Producer,
+ bool isOptimized, StringRef Flags,
+ unsigned RunTimeVer) {
+ assert (Lang <= dwarf::DW_LANG_D && Lang >= dwarf::DW_LANG_C89
+ && "Invalid Language tag");
+ assert (!Filename.empty()
+ && "Unable to create compile unit without filename");
+ Value *TElts[] = { GetTagConstant(VMContext, DW_TAG_base_type) };
+ TempEnumTypes = MDNode::getTemporary(VMContext, TElts);
+ Value *THElts[] = { TempEnumTypes };
+ MDNode *EnumHolder = MDNode::get(VMContext, THElts);
+
+ TempRetainTypes = MDNode::getTemporary(VMContext, TElts);
+ Value *TRElts[] = { TempRetainTypes };
+ MDNode *RetainHolder = MDNode::get(VMContext, TRElts);
+
+ TempSubprograms = MDNode::getTemporary(VMContext, TElts);
+ Value *TSElts[] = { TempSubprograms };
+ MDNode *SPHolder = MDNode::get(VMContext, TSElts);
+
+ TempGVs = MDNode::getTemporary(VMContext, TElts);
+ Value *TVElts[] = { TempGVs };
+ MDNode *GVHolder = MDNode::get(VMContext, TVElts);
+
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_compile_unit),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Lang),
+ MDString::get(VMContext, Filename),
+ MDString::get(VMContext, Directory),
+ MDString::get(VMContext, Producer),
+ // Deprecate isMain field.
+ ConstantInt::get(Type::getInt1Ty(VMContext), true), // isMain
+ ConstantInt::get(Type::getInt1Ty(VMContext), isOptimized),
+ MDString::get(VMContext, Flags),
+ ConstantInt::get(Type::getInt32Ty(VMContext), RunTimeVer),
+ EnumHolder,
+ RetainHolder,
+ SPHolder,
+ GVHolder
+ };
+ TheCU = DICompileUnit(MDNode::get(VMContext, Elts));
+
+ // Create a named metadata so that it is easier to find cu in a module.
+ NamedMDNode *NMD = M.getOrInsertNamedMetadata("llvm.dbg.cu");
+ NMD->addOperand(TheCU);
+}
+
+/// createFile - Create a file descriptor to hold debugging information
+/// for a file.
+DIFile DIBuilder::createFile(StringRef Filename, StringRef Directory) {
+ assert(TheCU && "Unable to create DW_TAG_file_type without CompileUnit");
+ assert(!Filename.empty() && "Unable to create file without name");
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_file_type),
+ MDString::get(VMContext, Filename),
+ MDString::get(VMContext, Directory),
+ NULL // TheCU
+ };
+ return DIFile(MDNode::get(VMContext, Elts));
+}
+
+/// createEnumerator - Create a single enumerator value.
+DIEnumerator DIBuilder::createEnumerator(StringRef Name, uint64_t Val) {
+ assert(!Name.empty() && "Unable to create enumerator without name");
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_enumerator),
+ MDString::get(VMContext, Name),
+ ConstantInt::get(Type::getInt64Ty(VMContext), Val)
+ };
+ return DIEnumerator(MDNode::get(VMContext, Elts));
+}
+
+/// createNullPtrType - Create C++0x nullptr type.
+DIType DIBuilder::createNullPtrType(StringRef Name) {
+ assert(!Name.empty() && "Unable to create type without name");
+ // nullptr is encoded in DIBasicType format. Line number, filename,
+ // ,size, alignment, offset and flags are always empty here.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_unspecified_type),
+ NULL, //TheCU,
+ MDString::get(VMContext, Name),
+ NULL, // Filename
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Line
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Size
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Align
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Offset
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Flags;
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Encoding
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createBasicType - Create debugging information entry for a basic
+/// type, e.g 'char'.
+DIType DIBuilder::createBasicType(StringRef Name, uint64_t SizeInBits,
+ uint64_t AlignInBits,
+ unsigned Encoding) {
+ assert(!Name.empty() && "Unable to create type without name");
+ // Basic types are encoded in DIBasicType format. Line number, filename,
+ // offset and flags are always empty here.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_base_type),
+ NULL, //TheCU,
+ MDString::get(VMContext, Name),
+ NULL, // Filename
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Line
+ ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Offset
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Flags;
+ ConstantInt::get(Type::getInt32Ty(VMContext), Encoding)
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createQaulifiedType - Create debugging information entry for a qualified
+/// type, e.g. 'const int'.
+DIType DIBuilder::createQualifiedType(unsigned Tag, DIType FromTy) {
+ // Qualified types are encoded in DIDerivedType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, Tag),
+ NULL, //TheCU,
+ MDString::get(VMContext, StringRef()), // Empty name.
+ NULL, // Filename
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Line
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Size
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Align
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Offset
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Flags
+ FromTy
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createPointerType - Create debugging information entry for a pointer.
+DIType DIBuilder::createPointerType(DIType PointeeTy, uint64_t SizeInBits,
+ uint64_t AlignInBits, StringRef Name) {
+ // Pointer types are encoded in DIDerivedType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_pointer_type),
+ NULL, //TheCU,
+ MDString::get(VMContext, Name),
+ NULL, // Filename
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Line
+ ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Offset
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Flags
+ PointeeTy
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createReferenceType - Create debugging information entry for a reference.
+DIType DIBuilder::createReferenceType(DIType RTy) {
+ assert(RTy.Verify() && "Unable to create reference type");
+ // References are encoded in DIDerivedType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_reference_type),
+ NULL, // TheCU,
+ NULL, // Name
+ NULL, // Filename
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Line
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Size
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Align
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Offset
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Flags
+ RTy
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createTypedef - Create debugging information entry for a typedef.
+DIType DIBuilder::createTypedef(DIType Ty, StringRef Name, DIFile File,
+ unsigned LineNo, DIDescriptor Context) {
+ // typedefs are encoded in DIDerivedType format.
+ assert(Ty.Verify() && "Invalid typedef type!");
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_typedef),
+ getNonCompileUnitScope(Context),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Size
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Align
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Offset
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Flags
+ Ty
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createFriend - Create debugging information entry for a 'friend'.
+DIType DIBuilder::createFriend(DIType Ty, DIType FriendTy) {
+ // typedefs are encoded in DIDerivedType format.
+ assert(Ty.Verify() && "Invalid type!");
+ assert(FriendTy.Verify() && "Invalid friend type!");
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_friend),
+ Ty,
+ NULL, // Name
+ Ty.getFile(),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Line
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Size
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Align
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Offset
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Flags
+ FriendTy
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createInheritance - Create debugging information entry to establish
+/// inheritance relationship between two types.
+DIType DIBuilder::createInheritance(DIType Ty, DIType BaseTy,
+ uint64_t BaseOffset, unsigned Flags) {
+ assert(Ty.Verify() && "Unable to create inheritance");
+ // TAG_inheritance is encoded in DIDerivedType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_inheritance),
+ Ty,
+ NULL, // Name
+ Ty.getFile(),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0), // Line
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Size
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0), // Align
+ ConstantInt::get(Type::getInt64Ty(VMContext), BaseOffset),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ BaseTy
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createMemberType - Create debugging information entry for a member.
+DIType DIBuilder::createMemberType(DIDescriptor Scope, StringRef Name,
+ DIFile File, unsigned LineNumber,
+ uint64_t SizeInBits, uint64_t AlignInBits,
+ uint64_t OffsetInBits, unsigned Flags,
+ DIType Ty) {
+ // TAG_member is encoded in DIDerivedType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_member),
+ getNonCompileUnitScope(Scope),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
+ ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), OffsetInBits),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ Ty
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createObjCIVar - Create debugging information entry for Objective-C
+/// instance variable.
+DIType DIBuilder::createObjCIVar(StringRef Name,
+ DIFile File, unsigned LineNumber,
+ uint64_t SizeInBits, uint64_t AlignInBits,
+ uint64_t OffsetInBits, unsigned Flags,
+ DIType Ty, StringRef PropertyName,
+ StringRef GetterName, StringRef SetterName,
+ unsigned PropertyAttributes) {
+ // TAG_member is encoded in DIDerivedType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_member),
+ getNonCompileUnitScope(File),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
+ ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), OffsetInBits),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ Ty,
+ MDString::get(VMContext, PropertyName),
+ MDString::get(VMContext, GetterName),
+ MDString::get(VMContext, SetterName),
+ ConstantInt::get(Type::getInt32Ty(VMContext), PropertyAttributes)
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createClassType - Create debugging information entry for a class.
+DIType DIBuilder::createClassType(DIDescriptor Context, StringRef Name,
+ DIFile File, unsigned LineNumber,
+ uint64_t SizeInBits, uint64_t AlignInBits,
+ uint64_t OffsetInBits, unsigned Flags,
+ DIType DerivedFrom, DIArray Elements,
+ MDNode *VTableHoder, MDNode *TemplateParams) {
+ // TAG_class_type is encoded in DICompositeType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_class_type),
+ getNonCompileUnitScope(Context),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
+ ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt32Ty(VMContext), OffsetInBits),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ DerivedFrom,
+ Elements,
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ VTableHoder,
+ TemplateParams
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createTemplateTypeParameter - Create debugging information for template
+/// type parameter.
+DITemplateTypeParameter
+DIBuilder::createTemplateTypeParameter(DIDescriptor Context, StringRef Name,
+ DIType Ty, MDNode *File, unsigned LineNo,
+ unsigned ColumnNo) {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_template_type_parameter),
+ getNonCompileUnitScope(Context),
+ MDString::get(VMContext, Name),
+ Ty,
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
+ ConstantInt::get(Type::getInt32Ty(VMContext), ColumnNo)
+ };
+ return DITemplateTypeParameter(MDNode::get(VMContext, Elts));
+}
+
+/// createTemplateValueParameter - Create debugging information for template
+/// value parameter.
+DITemplateValueParameter
+DIBuilder::createTemplateValueParameter(DIDescriptor Context, StringRef Name,
+ DIType Ty, uint64_t Val,
+ MDNode *File, unsigned LineNo,
+ unsigned ColumnNo) {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_template_value_parameter),
+ getNonCompileUnitScope(Context),
+ MDString::get(VMContext, Name),
+ Ty,
+ ConstantInt::get(Type::getInt64Ty(VMContext), Val),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
+ ConstantInt::get(Type::getInt32Ty(VMContext), ColumnNo)
+ };
+ return DITemplateValueParameter(MDNode::get(VMContext, Elts));
+}
+
+/// createStructType - Create debugging information entry for a struct.
+DIType DIBuilder::createStructType(DIDescriptor Context, StringRef Name,
+ DIFile File, unsigned LineNumber,
+ uint64_t SizeInBits, uint64_t AlignInBits,
+ unsigned Flags, DIArray Elements,
+ unsigned RunTimeLang) {
+ // TAG_structure_type is encoded in DICompositeType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_structure_type),
+ getNonCompileUnitScope(Context),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
+ ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ Elements,
+ ConstantInt::get(Type::getInt32Ty(VMContext), RunTimeLang),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createUnionType - Create debugging information entry for an union.
+DIType DIBuilder::createUnionType(DIDescriptor Scope, StringRef Name,
+ DIFile File,
+ unsigned LineNumber, uint64_t SizeInBits,
+ uint64_t AlignInBits, unsigned Flags,
+ DIArray Elements, unsigned RunTimeLang) {
+ // TAG_union_type is encoded in DICompositeType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_union_type),
+ getNonCompileUnitScope(Scope),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
+ ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ Elements,
+ ConstantInt::get(Type::getInt32Ty(VMContext), RunTimeLang),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createSubroutineType - Create subroutine type.
+DIType DIBuilder::createSubroutineType(DIFile File, DIArray ParameterTypes) {
+ // TAG_subroutine_type is encoded in DICompositeType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_subroutine_type),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ MDString::get(VMContext, ""),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt64Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ ParameterTypes,
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createEnumerationType - Create debugging information entry for an
+/// enumeration.
+DIType DIBuilder::createEnumerationType(DIDescriptor Scope, StringRef Name,
+ DIFile File, unsigned LineNumber,
+ uint64_t SizeInBits,
+ uint64_t AlignInBits,
+ DIArray Elements) {
+ // TAG_enumeration_type is encoded in DICompositeType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_enumeration_type),
+ getNonCompileUnitScope(Scope),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
+ ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ Elements,
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ };
+ MDNode *Node = MDNode::get(VMContext, Elts);
+ AllEnumTypes.push_back(Node);
+ return DIType(Node);
+}
+
+/// createArrayType - Create debugging information entry for an array.
+DIType DIBuilder::createArrayType(uint64_t Size, uint64_t AlignInBits,
+ DIType Ty, DIArray Subscripts) {
+ // TAG_array_type is encoded in DICompositeType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_array_type),
+ NULL, //TheCU,
+ MDString::get(VMContext, ""),
+ NULL, //TheCU,
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt64Ty(VMContext), Size),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ Ty,
+ Subscripts,
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createVectorType - Create debugging information entry for a vector.
+DIType DIBuilder::createVectorType(uint64_t Size, uint64_t AlignInBits,
+ DIType Ty, DIArray Subscripts) {
+ // TAG_vector_type is encoded in DICompositeType format.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_vector_type),
+ NULL, //TheCU,
+ MDString::get(VMContext, ""),
+ NULL, //TheCU,
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt64Ty(VMContext), Size),
+ ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ Ty,
+ Subscripts,
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ };
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// createArtificialType - Create a new DIType with "artificial" flag set.
+DIType DIBuilder::createArtificialType(DIType Ty) {
+ if (Ty.isArtificial())
+ return Ty;
+
+ SmallVector<Value *, 9> Elts;
+ MDNode *N = Ty;
+ assert (N && "Unexpected input DIType!");
+ for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
+ if (Value *V = N->getOperand(i))
+ Elts.push_back(V);
+ else
+ Elts.push_back(Constant::getNullValue(Type::getInt32Ty(VMContext)));
+ }
+
+ unsigned CurFlags = Ty.getFlags();
+ CurFlags = CurFlags | DIType::FlagArtificial;
+
+ // Flags are stored at this slot.
+ Elts[8] = ConstantInt::get(Type::getInt32Ty(VMContext), CurFlags);
+
+ return DIType(MDNode::get(VMContext, Elts));
+}
+
+/// retainType - Retain DIType in a module even if it is not referenced
+/// through debug info anchors.
+void DIBuilder::retainType(DIType T) {
+ AllRetainTypes.push_back(T);
+}
+
+/// createUnspecifiedParameter - Create unspeicified type descriptor
+/// for the subroutine type.
+DIDescriptor DIBuilder::createUnspecifiedParameter() {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_unspecified_parameters)
+ };
+ return DIDescriptor(MDNode::get(VMContext, Elts));
+}
+
+/// createTemporaryType - Create a temporary forward-declared type.
+DIType DIBuilder::createTemporaryType() {
+ // Give the temporary MDNode a tag. It doesn't matter what tag we
+ // use here as long as DIType accepts it.
+ Value *Elts[] = { GetTagConstant(VMContext, DW_TAG_base_type) };
+ MDNode *Node = MDNode::getTemporary(VMContext, Elts);
+ return DIType(Node);
+}
+
+/// createTemporaryType - Create a temporary forward-declared type.
+DIType DIBuilder::createTemporaryType(DIFile F) {
+ // Give the temporary MDNode a tag. It doesn't matter what tag we
+ // use here as long as DIType accepts it.
+ Value *Elts[] = {
+ GetTagConstant(VMContext, DW_TAG_base_type),
+ TheCU,
+ NULL,
+ F
+ };
+ MDNode *Node = MDNode::getTemporary(VMContext, Elts);
+ return DIType(Node);
+}
+
+/// getOrCreateArray - Get a DIArray, create one if required.
+DIArray DIBuilder::getOrCreateArray(ArrayRef<Value *> Elements) {
+ if (Elements.empty()) {
+ Value *Null = llvm::Constant::getNullValue(Type::getInt32Ty(VMContext));
+ return DIArray(MDNode::get(VMContext, Null));
+ }
+ return DIArray(MDNode::get(VMContext, Elements));
+}
+
+/// getOrCreateSubrange - Create a descriptor for a value range. This
+/// implicitly uniques the values returned.
+DISubrange DIBuilder::getOrCreateSubrange(int64_t Lo, int64_t Hi) {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_subrange_type),
+ ConstantInt::get(Type::getInt64Ty(VMContext), Lo),
+ ConstantInt::get(Type::getInt64Ty(VMContext), Hi)
+ };
+
+ return DISubrange(MDNode::get(VMContext, Elts));
+}
+
+/// createGlobalVariable - Create a new descriptor for the specified global.
+DIGlobalVariable DIBuilder::
+createGlobalVariable(StringRef Name, DIFile F, unsigned LineNumber,
+ DIType Ty, bool isLocalToUnit, llvm::Value *Val) {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_variable),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ NULL, // TheCU,
+ MDString::get(VMContext, Name),
+ MDString::get(VMContext, Name),
+ MDString::get(VMContext, Name),
+ F,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
+ Ty,
+ ConstantInt::get(Type::getInt32Ty(VMContext), isLocalToUnit),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 1), /* isDefinition*/
+ Val
+ };
+ MDNode *Node = MDNode::get(VMContext, Elts);
+ AllGVs.push_back(Node);
+ return DIGlobalVariable(Node);
+}
+
+/// createStaticVariable - Create a new descriptor for the specified static
+/// variable.
+DIGlobalVariable DIBuilder::
+createStaticVariable(DIDescriptor Context, StringRef Name,
+ StringRef LinkageName, DIFile F, unsigned LineNumber,
+ DIType Ty, bool isLocalToUnit, llvm::Value *Val) {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_variable),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ getNonCompileUnitScope(Context),
+ MDString::get(VMContext, Name),
+ MDString::get(VMContext, Name),
+ MDString::get(VMContext, LinkageName),
+ F,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
+ Ty,
+ ConstantInt::get(Type::getInt32Ty(VMContext), isLocalToUnit),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 1), /* isDefinition*/
+ Val
+ };
+ MDNode *Node = MDNode::get(VMContext, Elts);
+ AllGVs.push_back(Node);
+ return DIGlobalVariable(Node);
+}
+
+/// createVariable - Create a new descriptor for the specified variable.
+DIVariable DIBuilder::createLocalVariable(unsigned Tag, DIDescriptor Scope,
+ StringRef Name, DIFile File,
+ unsigned LineNo, DIType Ty,
+ bool AlwaysPreserve, unsigned Flags,
+ unsigned ArgNo) {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, Tag),
+ getNonCompileUnitScope(Scope),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), (LineNo | (ArgNo << 24))),
+ Ty,
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ };
+ MDNode *Node = MDNode::get(VMContext, Elts);
+ if (AlwaysPreserve) {
+ // The optimizer may remove local variable. If there is an interest
+ // to preserve variable info in such situation then stash it in a
+ // named mdnode.
+ DISubprogram Fn(getDISubprogram(Scope));
+ NamedMDNode *FnLocals = getOrInsertFnSpecificMDNode(M, Fn);
+ FnLocals->addOperand(Node);
+ }
+ return DIVariable(Node);
+}
+
+/// createComplexVariable - Create a new descriptor for the specified variable
+/// which has a complex address expression for its address.
+DIVariable DIBuilder::createComplexVariable(unsigned Tag, DIDescriptor Scope,
+ StringRef Name, DIFile F,
+ unsigned LineNo,
+ DIType Ty, ArrayRef<Value *> Addr,
+ unsigned ArgNo) {
+ SmallVector<Value *, 15> Elts;
+ Elts.push_back(GetTagConstant(VMContext, Tag));
+ Elts.push_back(getNonCompileUnitScope(Scope)),
+ Elts.push_back(MDString::get(VMContext, Name));
+ Elts.push_back(F);
+ Elts.push_back(ConstantInt::get(Type::getInt32Ty(VMContext),
+ (LineNo | (ArgNo << 24))));
+ Elts.push_back(Ty);
+ Elts.push_back(llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)));
+ Elts.push_back(llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)));
+ Elts.append(Addr.begin(), Addr.end());
+
+ return DIVariable(MDNode::get(VMContext, Elts));
+}
+
+/// createFunction - Create a new descriptor for the specified function.
+DISubprogram DIBuilder::createFunction(DIDescriptor Context,
+ StringRef Name,
+ StringRef LinkageName,
+ DIFile File, unsigned LineNo,
+ DIType Ty,
+ bool isLocalToUnit, bool isDefinition,
+ unsigned Flags, bool isOptimized,
+ Function *Fn,
+ MDNode *TParams,
+ MDNode *Decl) {
+ Value *TElts[] = { GetTagConstant(VMContext, DW_TAG_base_type) };
+ MDNode *Temp = MDNode::getTemporary(VMContext, TElts);
+ Value *TVElts[] = { Temp };
+ MDNode *THolder = MDNode::get(VMContext, TVElts);
+
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_subprogram),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ getNonCompileUnitScope(Context),
+ MDString::get(VMContext, Name),
+ MDString::get(VMContext, Name),
+ MDString::get(VMContext, LinkageName),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
+ Ty,
+ ConstantInt::get(Type::getInt1Ty(VMContext), isLocalToUnit),
+ ConstantInt::get(Type::getInt1Ty(VMContext), isDefinition),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ ConstantInt::get(Type::getInt32Ty(VMContext), 0),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ ConstantInt::get(Type::getInt1Ty(VMContext), isOptimized),
+ Fn,
+ TParams,
+ Decl,
+ THolder
+ };
+ MDNode *Node = MDNode::get(VMContext, Elts);
+
+ // Create a named metadata so that we do not lose this mdnode.
+ AllSubprograms.push_back(Node);
+ return DISubprogram(Node);
+}
+
+/// createMethod - Create a new descriptor for the specified C++ method.
+DISubprogram DIBuilder::createMethod(DIDescriptor Context,
+ StringRef Name,
+ StringRef LinkageName,
+ DIFile F,
+ unsigned LineNo, DIType Ty,
+ bool isLocalToUnit,
+ bool isDefinition,
+ unsigned VK, unsigned VIndex,
+ MDNode *VTableHolder,
+ unsigned Flags,
+ bool isOptimized,
+ Function *Fn,
+ MDNode *TParam) {
+ Value *TElts[] = { GetTagConstant(VMContext, DW_TAG_base_type) };
+ MDNode *Temp = MDNode::getTemporary(VMContext, TElts);
+ Value *TVElts[] = { Temp };
+ MDNode *THolder = MDNode::get(VMContext, TVElts);
+
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_subprogram),
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ getNonCompileUnitScope(Context),
+ MDString::get(VMContext, Name),
+ MDString::get(VMContext, Name),
+ MDString::get(VMContext, LinkageName),
+ F,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
+ Ty,
+ ConstantInt::get(Type::getInt1Ty(VMContext), isLocalToUnit),
+ ConstantInt::get(Type::getInt1Ty(VMContext), isDefinition),
+ ConstantInt::get(Type::getInt32Ty(VMContext), (unsigned)VK),
+ ConstantInt::get(Type::getInt32Ty(VMContext), VIndex),
+ VTableHolder,
+ ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
+ ConstantInt::get(Type::getInt1Ty(VMContext), isOptimized),
+ Fn,
+ TParam,
+ llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
+ THolder
+ };
+ MDNode *Node = MDNode::get(VMContext, Elts);
+ return DISubprogram(Node);
+}
+
+/// createNameSpace - This creates new descriptor for a namespace
+/// with the specified parent scope.
+DINameSpace DIBuilder::createNameSpace(DIDescriptor Scope, StringRef Name,
+ DIFile File, unsigned LineNo) {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_namespace),
+ getNonCompileUnitScope(Scope),
+ MDString::get(VMContext, Name),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), LineNo)
+ };
+ return DINameSpace(MDNode::get(VMContext, Elts));
+}
+
+/// createLexicalBlockFile - This creates a new MDNode that encapsulates
+/// an existing scope with a new filename.
+DILexicalBlockFile DIBuilder::createLexicalBlockFile(DIDescriptor Scope,
+ DIFile File) {
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_lexical_block),
+ Scope,
+ File
+ };
+ return DILexicalBlockFile(MDNode::get(VMContext, Elts));
+}
+
+DILexicalBlock DIBuilder::createLexicalBlock(DIDescriptor Scope, DIFile File,
+ unsigned Line, unsigned Col) {
+ // Defeat MDNode uniqing for lexical blocks by using unique id.
+ static unsigned int unique_id = 0;
+ Value *Elts[] = {
+ GetTagConstant(VMContext, dwarf::DW_TAG_lexical_block),
+ getNonCompileUnitScope(Scope),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Line),
+ ConstantInt::get(Type::getInt32Ty(VMContext), Col),
+ File,
+ ConstantInt::get(Type::getInt32Ty(VMContext), unique_id++)
+ };
+ return DILexicalBlock(MDNode::get(VMContext, Elts));
+}
+
+/// insertDeclare - Insert a new llvm.dbg.declare intrinsic call.
+Instruction *DIBuilder::insertDeclare(Value *Storage, DIVariable VarInfo,
+ Instruction *InsertBefore) {
+ assert(Storage && "no storage passed to dbg.declare");
+ assert(VarInfo.Verify() && "empty DIVariable passed to dbg.declare");
+ if (!DeclareFn)
+ DeclareFn = Intrinsic::getDeclaration(&M, Intrinsic::dbg_declare);
+
+ Value *Args[] = { MDNode::get(Storage->getContext(), Storage), VarInfo };
+ return CallInst::Create(DeclareFn, Args, "", InsertBefore);
+}
+
+/// insertDeclare - Insert a new llvm.dbg.declare intrinsic call.
+Instruction *DIBuilder::insertDeclare(Value *Storage, DIVariable VarInfo,
+ BasicBlock *InsertAtEnd) {
+ assert(Storage && "no storage passed to dbg.declare");
+ assert(VarInfo.Verify() && "invalid DIVariable passed to dbg.declare");
+ if (!DeclareFn)
+ DeclareFn = Intrinsic::getDeclaration(&M, Intrinsic::dbg_declare);
+
+ Value *Args[] = { MDNode::get(Storage->getContext(), Storage), VarInfo };
+
+ // If this block already has a terminator then insert this intrinsic
+ // before the terminator.
+ if (TerminatorInst *T = InsertAtEnd->getTerminator())
+ return CallInst::Create(DeclareFn, Args, "", T);
+ else
+ return CallInst::Create(DeclareFn, Args, "", InsertAtEnd);
+}
+
+/// insertDbgValueIntrinsic - Insert a new llvm.dbg.value intrinsic call.
+Instruction *DIBuilder::insertDbgValueIntrinsic(Value *V, uint64_t Offset,
+ DIVariable VarInfo,
+ Instruction *InsertBefore) {
+ assert(V && "no value passed to dbg.value");
+ assert(VarInfo.Verify() && "invalid DIVariable passed to dbg.value");
+ if (!ValueFn)
+ ValueFn = Intrinsic::getDeclaration(&M, Intrinsic::dbg_value);
+
+ Value *Args[] = { MDNode::get(V->getContext(), V),
+ ConstantInt::get(Type::getInt64Ty(V->getContext()), Offset),
+ VarInfo };
+ return CallInst::Create(ValueFn, Args, "", InsertBefore);
+}
+
+/// insertDbgValueIntrinsic - Insert a new llvm.dbg.value intrinsic call.
+Instruction *DIBuilder::insertDbgValueIntrinsic(Value *V, uint64_t Offset,
+ DIVariable VarInfo,
+ BasicBlock *InsertAtEnd) {
+ assert(V && "no value passed to dbg.value");
+ assert(VarInfo.Verify() && "invalid DIVariable passed to dbg.value");
+ if (!ValueFn)
+ ValueFn = Intrinsic::getDeclaration(&M, Intrinsic::dbg_value);
+
+ Value *Args[] = { MDNode::get(V->getContext(), V),
+ ConstantInt::get(Type::getInt64Ty(V->getContext()), Offset),
+ VarInfo };
+ return CallInst::Create(ValueFn, Args, "", InsertAtEnd);
+}
diff --git a/src/LLVM/lib/Analysis/DbgInfoPrinter.cpp b/src/LLVM/lib/Analysis/DbgInfoPrinter.cpp
new file mode 100644
index 0000000..cd832ab
--- /dev/null
+++ b/src/LLVM/lib/Analysis/DbgInfoPrinter.cpp
@@ -0,0 +1,224 @@
+//===- DbgInfoPrinter.cpp - Print debug info in a human readable form ------==//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a pass that prints instructions, and associated debug
+// info:
+//
+// - source/line/col information
+// - original variable name
+// - original type name
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Pass.h"
+#include "llvm/Function.h"
+#include "llvm/IntrinsicInst.h"
+#include "llvm/Metadata.h"
+#include "llvm/Module.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Analysis/DebugInfo.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+static cl::opt<bool>
+PrintDirectory("print-fullpath",
+ cl::desc("Print fullpath when printing debug info"),
+ cl::Hidden);
+
+namespace {
+ class PrintDbgInfo : public FunctionPass {
+ raw_ostream &Out;
+ void printVariableDeclaration(const Value *V);
+ public:
+ static char ID; // Pass identification
+ PrintDbgInfo() : FunctionPass(ID), Out(errs()) {
+ initializePrintDbgInfoPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnFunction(Function &F);
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+ };
+ char PrintDbgInfo::ID = 0;
+}
+
+INITIALIZE_PASS(PrintDbgInfo, "print-dbginfo",
+ "Print debug info in human readable form", false, false)
+
+FunctionPass *llvm::createDbgInfoPrinterPass() { return new PrintDbgInfo(); }
+
+/// Find the debug info descriptor corresponding to this global variable.
+static Value *findDbgGlobalDeclare(GlobalVariable *V) {
+ const Module *M = V->getParent();
+ NamedMDNode *NMD = M->getNamedMetadata("llvm.dbg.gv");
+ if (!NMD)
+ return 0;
+
+ for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i) {
+ DIDescriptor DIG(cast<MDNode>(NMD->getOperand(i)));
+ if (!DIG.isGlobalVariable())
+ continue;
+ if (DIGlobalVariable(DIG).getGlobal() == V)
+ return DIG;
+ }
+ return 0;
+}
+
+/// Find the debug info descriptor corresponding to this function.
+static Value *findDbgSubprogramDeclare(Function *V) {
+ const Module *M = V->getParent();
+ NamedMDNode *NMD = M->getNamedMetadata("llvm.dbg.sp");
+ if (!NMD)
+ return 0;
+
+ for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i) {
+ DIDescriptor DIG(cast<MDNode>(NMD->getOperand(i)));
+ if (!DIG.isSubprogram())
+ continue;
+ if (DISubprogram(DIG).getFunction() == V)
+ return DIG;
+ }
+ return 0;
+}
+
+/// Finds the llvm.dbg.declare intrinsic corresponding to this value if any.
+/// It looks through pointer casts too.
+static const DbgDeclareInst *findDbgDeclare(const Value *V) {
+ V = V->stripPointerCasts();
+
+ if (!isa<Instruction>(V) && !isa<Argument>(V))
+ return 0;
+
+ const Function *F = NULL;
+ if (const Instruction *I = dyn_cast<Instruction>(V))
+ F = I->getParent()->getParent();
+ else if (const Argument *A = dyn_cast<Argument>(V))
+ F = A->getParent();
+
+ for (Function::const_iterator FI = F->begin(), FE = F->end(); FI != FE; ++FI)
+ for (BasicBlock::const_iterator BI = (*FI).begin(), BE = (*FI).end();
+ BI != BE; ++BI)
+ if (const DbgDeclareInst *DDI = dyn_cast<DbgDeclareInst>(BI))
+ if (DDI->getAddress() == V)
+ return DDI;
+
+ return 0;
+}
+
+static bool getLocationInfo(const Value *V, std::string &DisplayName,
+ std::string &Type, unsigned &LineNo,
+ std::string &File, std::string &Dir) {
+ DICompileUnit Unit;
+ DIType TypeD;
+
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(const_cast<Value*>(V))) {
+ Value *DIGV = findDbgGlobalDeclare(GV);
+ if (!DIGV) return false;
+ DIGlobalVariable Var(cast<MDNode>(DIGV));
+
+ StringRef D = Var.getDisplayName();
+ if (!D.empty())
+ DisplayName = D;
+ LineNo = Var.getLineNumber();
+ Unit = Var.getCompileUnit();
+ TypeD = Var.getType();
+ } else if (Function *F = dyn_cast<Function>(const_cast<Value*>(V))){
+ Value *DIF = findDbgSubprogramDeclare(F);
+ if (!DIF) return false;
+ DISubprogram Var(cast<MDNode>(DIF));
+
+ StringRef D = Var.getDisplayName();
+ if (!D.empty())
+ DisplayName = D;
+ LineNo = Var.getLineNumber();
+ Unit = Var.getCompileUnit();
+ TypeD = Var.getType();
+ } else {
+ const DbgDeclareInst *DDI = findDbgDeclare(V);
+ if (!DDI) return false;
+ DIVariable Var(cast<MDNode>(DDI->getVariable()));
+
+ StringRef D = Var.getName();
+ if (!D.empty())
+ DisplayName = D;
+ LineNo = Var.getLineNumber();
+ Unit = Var.getCompileUnit();
+ TypeD = Var.getType();
+ }
+
+ StringRef T = TypeD.getName();
+ if (!T.empty())
+ Type = T;
+ StringRef F = Unit.getFilename();
+ if (!F.empty())
+ File = F;
+ StringRef D = Unit.getDirectory();
+ if (!D.empty())
+ Dir = D;
+ return true;
+}
+
+void PrintDbgInfo::printVariableDeclaration(const Value *V) {
+ std::string DisplayName, File, Directory, Type;
+ unsigned LineNo = 0;
+
+ if (!getLocationInfo(V, DisplayName, Type, LineNo, File, Directory))
+ return;
+
+ Out << "; ";
+ WriteAsOperand(Out, V, false, 0);
+ if (isa<Function>(V))
+ Out << " is function " << DisplayName
+ << " of type " << Type << " declared at ";
+ else
+ Out << " is variable " << DisplayName
+ << " of type " << Type << " declared at ";
+
+ if (PrintDirectory)
+ Out << Directory << "/";
+
+ Out << File << ":" << LineNo << "\n";
+}
+
+bool PrintDbgInfo::runOnFunction(Function &F) {
+ if (F.isDeclaration())
+ return false;
+
+ Out << "function " << F.getName() << "\n\n";
+
+ for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I) {
+ BasicBlock *BB = I;
+
+ if (I != F.begin() && (pred_begin(BB) == pred_end(BB)))
+ // Skip dead blocks.
+ continue;
+
+ Out << BB->getName();
+ Out << ":";
+
+ Out << "\n";
+
+ for (BasicBlock::const_iterator i = BB->begin(), e = BB->end();
+ i != e; ++i) {
+
+ printVariableDeclaration(i);
+
+ if (const User *U = dyn_cast<User>(i)) {
+ for(unsigned i=0;i<U->getNumOperands();i++)
+ printVariableDeclaration(U->getOperand(i));
+ }
+ }
+ }
+ return false;
+}
diff --git a/src/LLVM/lib/Analysis/DebugInfo.cpp b/src/LLVM/lib/Analysis/DebugInfo.cpp
index 78f0dee..44457d3 100644
--- a/src/LLVM/lib/Analysis/DebugInfo.cpp
+++ b/src/LLVM/lib/Analysis/DebugInfo.cpp
@@ -22,6 +22,7 @@
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/Dwarf.h"
#include "llvm/Support/raw_ostream.h"
@@ -38,6 +39,9 @@
DIDescriptor::DIDescriptor(const DISubprogram F) : DbgNode(F.DbgNode) {
}
+DIDescriptor::DIDescriptor(const DILexicalBlockFile F) : DbgNode(F.DbgNode) {
+}
+
DIDescriptor::DIDescriptor(const DILexicalBlock F) : DbgNode(F.DbgNode) {
}
@@ -108,9 +112,19 @@
}
unsigned DIVariable::getNumAddrElements() const {
- return DbgNode->getNumOperands()-6;
+ if (getVersion() <= llvm::LLVMDebugVersion8)
+ return DbgNode->getNumOperands()-6;
+ if (getVersion() == llvm::LLVMDebugVersion9)
+ return DbgNode->getNumOperands()-7;
+ return DbgNode->getNumOperands()-8;
}
+/// getInlinedAt - If this variable is inlined then return inline location.
+MDNode *DIVariable::getInlinedAt() const {
+ if (getVersion() <= llvm::LLVMDebugVersion9)
+ return NULL;
+ return dyn_cast_or_null<MDNode>(DbgNode->getOperand(7));
+}
//===----------------------------------------------------------------------===//
// Predicates
@@ -119,7 +133,14 @@
/// isBasicType - Return true if the specified tag is legal for
/// DIBasicType.
bool DIDescriptor::isBasicType() const {
- return DbgNode && getTag() == dwarf::DW_TAG_base_type;
+ if (!DbgNode) return false;
+ switch (getTag()) {
+ case dwarf::DW_TAG_base_type:
+ case dwarf::DW_TAG_unspecified_type:
+ return true;
+ default:
+ return false;
+ }
}
/// isDerivedType - Return true if the specified tag is legal for DIDerivedType.
@@ -134,6 +155,7 @@
case dwarf::DW_TAG_restrict_type:
case dwarf::DW_TAG_member:
case dwarf::DW_TAG_inheritance:
+ case dwarf::DW_TAG_friend:
return true;
default:
// CompositeTypes are currently modelled as DerivedTypes.
@@ -195,6 +217,12 @@
return isGlobalVariable();
}
+/// isUnspecifiedParmeter - Return true if the specified tag is
+/// DW_TAG_unspecified_parameters.
+bool DIDescriptor::isUnspecifiedParameter() const {
+ return DbgNode && getTag() == dwarf::DW_TAG_unspecified_parameters;
+}
+
/// isScope - Return true if the specified tag is one of the scope
/// related tag.
bool DIDescriptor::isScope() const {
@@ -211,6 +239,18 @@
return false;
}
+/// isTemplateTypeParameter - Return true if the specified tag is
+/// DW_TAG_template_type_parameter.
+bool DIDescriptor::isTemplateTypeParameter() const {
+ return DbgNode && getTag() == dwarf::DW_TAG_template_type_parameter;
+}
+
+/// isTemplateValueParameter - Return true if the specified tag is
+/// DW_TAG_template_value_parameter.
+bool DIDescriptor::isTemplateValueParameter() const {
+ return DbgNode && getTag() == dwarf::DW_TAG_template_value_parameter;
+}
+
/// isCompileUnit - Return true if the specified tag is DW_TAG_compile_unit.
bool DIDescriptor::isCompileUnit() const {
return DbgNode && getTag() == dwarf::DW_TAG_compile_unit;
@@ -226,9 +266,17 @@
return DbgNode && getTag() == dwarf::DW_TAG_namespace;
}
+/// isLexicalBlockFile - Return true if the specified descriptor is a
+/// lexical block with an extra file.
+bool DIDescriptor::isLexicalBlockFile() const {
+ return DbgNode && getTag() == dwarf::DW_TAG_lexical_block &&
+ (DbgNode->getNumOperands() == 3);
+}
+
/// isLexicalBlock - Return true if the specified tag is DW_TAG_lexical_block.
bool DIDescriptor::isLexicalBlock() const {
- return DbgNode && getTag() == dwarf::DW_TAG_lexical_block;
+ return DbgNode && getTag() == dwarf::DW_TAG_lexical_block &&
+ (DbgNode->getNumOperands() > 3);
}
/// isSubrange - Return true if the specified tag is DW_TAG_subrange_type.
@@ -260,7 +308,7 @@
/// replaceAllUsesWith - Replace all uses of debug info referenced by
/// this descriptor.
-void DIDerivedType::replaceAllUsesWith(DIDescriptor &D) {
+void DIType::replaceAllUsesWith(DIDescriptor &D) {
if (!DbgNode)
return;
@@ -274,9 +322,46 @@
const MDNode *DN = D;
const Value *V = cast_or_null<Value>(DN);
Node->replaceAllUsesWith(const_cast<Value*>(V));
+ MDNode::deleteTemporary(Node);
}
}
+/// replaceAllUsesWith - Replace all uses of debug info referenced by
+/// this descriptor.
+void DIType::replaceAllUsesWith(MDNode *D) {
+ if (!DbgNode)
+ return;
+
+ // Since we use a TrackingVH for the node, its easy for clients to manufacture
+ // legitimate situations where they want to replaceAllUsesWith() on something
+ // which, due to uniquing, has merged with the source. We shield clients from
+ // this detail by allowing a value to be replaced with replaceAllUsesWith()
+ // itself.
+ if (DbgNode != D) {
+ MDNode *Node = const_cast<MDNode*>(DbgNode);
+ const MDNode *DN = D;
+ const Value *V = cast_or_null<Value>(DN);
+ Node->replaceAllUsesWith(const_cast<Value*>(V));
+ MDNode::deleteTemporary(Node);
+ }
+}
+
+/// isUnsignedDIType - Return true if type encoding is unsigned.
+bool DIType::isUnsignedDIType() {
+ DIDerivedType DTy(DbgNode);
+ if (DTy.Verify())
+ return DTy.getTypeDerivedFrom().isUnsignedDIType();
+
+ DIBasicType BTy(DbgNode);
+ if (BTy.Verify()) {
+ unsigned Encoding = BTy.getEncoding();
+ if (Encoding == dwarf::DW_ATE_unsigned ||
+ Encoding == dwarf::DW_ATE_unsigned_char)
+ return true;
+ }
+ return false;
+}
+
/// Verify - Verify that a compile unit is well formed.
bool DICompileUnit::Verify() const {
if (!DbgNode)
@@ -292,25 +377,37 @@
bool DIType::Verify() const {
if (!DbgNode)
return false;
- if (!getContext().Verify())
+ if (getContext() && !getContext().Verify())
return false;
-
- DICompileUnit CU = getCompileUnit();
- if (!CU.Verify())
+ unsigned Tag = getTag();
+ if (!isBasicType() && Tag != dwarf::DW_TAG_const_type &&
+ Tag != dwarf::DW_TAG_volatile_type && Tag != dwarf::DW_TAG_pointer_type &&
+ Tag != dwarf::DW_TAG_reference_type && Tag != dwarf::DW_TAG_restrict_type
+ && Tag != dwarf::DW_TAG_vector_type && Tag != dwarf::DW_TAG_array_type
+ && Tag != dwarf::DW_TAG_enumeration_type
+ && Tag != dwarf::DW_TAG_subroutine_type
+ && getFilename().empty())
return false;
return true;
}
+/// Verify - Verify that a basic type descriptor is well formed.
+bool DIBasicType::Verify() const {
+ return isBasicType();
+}
+
+/// Verify - Verify that a derived type descriptor is well formed.
+bool DIDerivedType::Verify() const {
+ return isDerivedType();
+}
+
/// Verify - Verify that a composite type descriptor is well formed.
bool DICompositeType::Verify() const {
if (!DbgNode)
return false;
- if (!getContext().Verify())
+ if (getContext() && !getContext().Verify())
return false;
- DICompileUnit CU = getCompileUnit();
- if (!CU.Verify())
- return false;
return true;
}
@@ -319,11 +416,7 @@
if (!DbgNode)
return false;
- if (!getContext().Verify())
- return false;
-
- DICompileUnit CU = getCompileUnit();
- if (!CU.Verify())
+ if (getContext() && !getContext().Verify())
return false;
DICompositeType Ty = getType();
@@ -340,11 +433,7 @@
if (getDisplayName().empty())
return false;
- if (!getContext().Verify())
- return false;
-
- DICompileUnit CU = getCompileUnit();
- if (!CU.Verify())
+ if (getContext() && !getContext().Verify())
return false;
DIType Ty = getType();
@@ -362,10 +451,7 @@
if (!DbgNode)
return false;
- if (!getContext().Verify())
- return false;
-
- if (!getCompileUnit().Verify())
+ if (getContext() && !getContext().Verify())
return false;
DIType Ty = getType();
@@ -389,8 +475,6 @@
return false;
if (getName().empty())
return false;
- if (!getCompileUnit().Verify())
- return false;
return true;
}
@@ -447,9 +531,28 @@
return 0;
}
+MDNode *DISubprogram::getVariablesNodes() const {
+ if (!DbgNode || DbgNode->getNumOperands() <= 19)
+ return NULL;
+ if (MDNode *Temp = dyn_cast_or_null<MDNode>(DbgNode->getOperand(19)))
+ return dyn_cast_or_null<MDNode>(Temp->getOperand(0));
+ return NULL;
+}
+
+DIArray DISubprogram::getVariables() const {
+ if (!DbgNode || DbgNode->getNumOperands() <= 19)
+ return DIArray();
+ if (MDNode *T = dyn_cast_or_null<MDNode>(DbgNode->getOperand(19)))
+ if (MDNode *A = dyn_cast_or_null<MDNode>(T->getOperand(0)))
+ return DIArray(A);
+ return DIArray();
+}
+
StringRef DIScope::getFilename() const {
if (!DbgNode)
return StringRef();
+ if (isLexicalBlockFile())
+ return DILexicalBlockFile(DbgNode).getFilename();
if (isLexicalBlock())
return DILexicalBlock(DbgNode).getFilename();
if (isSubprogram())
@@ -469,6 +572,8 @@
StringRef DIScope::getDirectory() const {
if (!DbgNode)
return StringRef();
+ if (isLexicalBlockFile())
+ return DILexicalBlockFile(DbgNode).getDirectory();
if (isLexicalBlock())
return DILexicalBlock(DbgNode).getDirectory();
if (isSubprogram())
@@ -485,6 +590,47 @@
return StringRef();
}
+DIArray DICompileUnit::getEnumTypes() const {
+ if (!DbgNode || DbgNode->getNumOperands() < 14)
+ return DIArray();
+
+ if (MDNode *N = dyn_cast_or_null<MDNode>(DbgNode->getOperand(10)))
+ if (MDNode *A = dyn_cast_or_null<MDNode>(N->getOperand(0)))
+ return DIArray(A);
+ return DIArray();
+}
+
+DIArray DICompileUnit::getRetainedTypes() const {
+ if (!DbgNode || DbgNode->getNumOperands() < 14)
+ return DIArray();
+
+ if (MDNode *N = dyn_cast_or_null<MDNode>(DbgNode->getOperand(11)))
+ if (MDNode *A = dyn_cast_or_null<MDNode>(N->getOperand(0)))
+ return DIArray(A);
+ return DIArray();
+}
+
+DIArray DICompileUnit::getSubprograms() const {
+ if (!DbgNode || DbgNode->getNumOperands() < 14)
+ return DIArray();
+
+ if (MDNode *N = dyn_cast_or_null<MDNode>(DbgNode->getOperand(12)))
+ if (MDNode *A = dyn_cast_or_null<MDNode>(N->getOperand(0)))
+ return DIArray(A);
+ return DIArray();
+}
+
+
+DIArray DICompileUnit::getGlobalVariables() const {
+ if (!DbgNode || DbgNode->getNumOperands() < 14)
+ return DIArray();
+
+ if (MDNode *N = dyn_cast_or_null<MDNode>(DbgNode->getOperand(13)))
+ if (MDNode *A = dyn_cast_or_null<MDNode>(N->getOperand(0)))
+ return DIArray(A);
+ return DIArray();
+}
+
//===----------------------------------------------------------------------===//
// DIDescriptor: dump routines for all descriptors.
//===----------------------------------------------------------------------===//
@@ -516,7 +662,6 @@
OS << " [" << dwarf::TagString(Tag) << "] ";
// TODO : Print context
- getCompileUnit().print(OS);
OS << " ["
<< "line " << getLineNumber() << ", "
<< getSizeInBits() << " bits, "
@@ -572,7 +717,6 @@
OS << " [" << dwarf::TagString(Tag) << "] ";
// TODO : Print context
- getCompileUnit().print(OS);
OS << " [" << getLineNumber() << "] ";
if (isLocalToUnit())
@@ -595,7 +739,6 @@
OS << " [" << dwarf::TagString(Tag) << "] ";
// TODO : Print context
- getCompileUnit().print(OS);
OS << " [" << getLineNumber() << "] ";
if (isLocalToUnit())
@@ -609,13 +752,48 @@
OS << "]\n";
}
+static void printDebugLoc(DebugLoc DL, raw_ostream &CommentOS,
+ const LLVMContext &Ctx) {
+ if (!DL.isUnknown()) { // Print source line info.
+ DIScope Scope(DL.getScope(Ctx));
+ // Omit the directory, because it's likely to be long and uninteresting.
+ if (Scope.Verify())
+ CommentOS << Scope.getFilename();
+ else
+ CommentOS << "<unknown>";
+ CommentOS << ':' << DL.getLine();
+ if (DL.getCol() != 0)
+ CommentOS << ':' << DL.getCol();
+ DebugLoc InlinedAtDL = DebugLoc::getFromDILocation(DL.getInlinedAt(Ctx));
+ if (!InlinedAtDL.isUnknown()) {
+ CommentOS << " @[ ";
+ printDebugLoc(InlinedAtDL, CommentOS, Ctx);
+ CommentOS << " ]";
+ }
+ }
+}
+
+void DIVariable::printExtendedName(raw_ostream &OS) const {
+ const LLVMContext &Ctx = DbgNode->getContext();
+ StringRef Res = getName();
+ if (!Res.empty())
+ OS << Res << "," << getLineNumber();
+ if (MDNode *InlinedAt = getInlinedAt()) {
+ DebugLoc InlinedAtDL = DebugLoc::getFromDILocation(InlinedAt);
+ if (!InlinedAtDL.isUnknown()) {
+ OS << " @[";
+ printDebugLoc(InlinedAtDL, OS, Ctx);
+ OS << "]";
+ }
+ }
+}
+
/// print - Print variable.
void DIVariable::print(raw_ostream &OS) const {
StringRef Res = getName();
if (!Res.empty())
OS << " [" << Res << "] ";
- getCompileUnit().print(OS);
OS << " [" << getLineNumber() << "] ";
getType().print(OS);
OS << "\n";
@@ -668,555 +846,79 @@
print(dbgs()); dbgs() << '\n';
}
-//===----------------------------------------------------------------------===//
-// DIFactory: Basic Helpers
-//===----------------------------------------------------------------------===//
+/// fixupObjcLikeName - Replace contains special characters used
+/// in a typical Objective-C names with '.' in a given string.
+static void fixupObjcLikeName(StringRef Str, SmallVectorImpl<char> &Out) {
+ bool isObjCLike = false;
+ for (size_t i = 0, e = Str.size(); i < e; ++i) {
+ char C = Str[i];
+ if (C == '[')
+ isObjCLike = true;
-DIFactory::DIFactory(Module &m)
- : M(m), VMContext(M.getContext()), DeclareFn(0), ValueFn(0) {}
-
-Constant *DIFactory::GetTagConstant(unsigned TAG) {
- assert((TAG & LLVMDebugVersionMask) == 0 &&
- "Tag too large for debug encoding!");
- return ConstantInt::get(Type::getInt32Ty(VMContext), TAG | LLVMDebugVersion);
-}
-
-//===----------------------------------------------------------------------===//
-// DIFactory: Primary Constructors
-//===----------------------------------------------------------------------===//
-
-/// GetOrCreateArray - Create an descriptor for an array of descriptors.
-/// This implicitly uniques the arrays created.
-DIArray DIFactory::GetOrCreateArray(DIDescriptor *Tys, unsigned NumTys) {
- SmallVector<Value*, 16> Elts;
-
- if (NumTys == 0)
- Elts.push_back(llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)));
- else
- for (unsigned i = 0; i != NumTys; ++i)
- Elts.push_back(Tys[i]);
-
- return DIArray(MDNode::get(VMContext,Elts.data(), Elts.size()));
-}
-
-/// GetOrCreateSubrange - Create a descriptor for a value range. This
-/// implicitly uniques the values returned.
-DISubrange DIFactory::GetOrCreateSubrange(int64_t Lo, int64_t Hi) {
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_subrange_type),
- ConstantInt::get(Type::getInt64Ty(VMContext), Lo),
- ConstantInt::get(Type::getInt64Ty(VMContext), Hi)
- };
-
- return DISubrange(MDNode::get(VMContext, &Elts[0], 3));
-}
-
-
-
-/// CreateCompileUnit - Create a new descriptor for the specified compile
-/// unit. Note that this does not unique compile units within the module.
-DICompileUnit DIFactory::CreateCompileUnit(unsigned LangID,
- StringRef Filename,
- StringRef Directory,
- StringRef Producer,
- bool isMain,
- bool isOptimized,
- StringRef Flags,
- unsigned RunTimeVer) {
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_compile_unit),
- llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
- ConstantInt::get(Type::getInt32Ty(VMContext), LangID),
- MDString::get(VMContext, Filename),
- MDString::get(VMContext, Directory),
- MDString::get(VMContext, Producer),
- ConstantInt::get(Type::getInt1Ty(VMContext), isMain),
- ConstantInt::get(Type::getInt1Ty(VMContext), isOptimized),
- MDString::get(VMContext, Flags),
- ConstantInt::get(Type::getInt32Ty(VMContext), RunTimeVer)
- };
-
- return DICompileUnit(MDNode::get(VMContext, &Elts[0], 10));
-}
-
-/// CreateFile - Create a new descriptor for the specified file.
-DIFile DIFactory::CreateFile(StringRef Filename,
- StringRef Directory,
- DICompileUnit CU) {
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_file_type),
- MDString::get(VMContext, Filename),
- MDString::get(VMContext, Directory),
- CU
- };
-
- return DIFile(MDNode::get(VMContext, &Elts[0], 4));
-}
-
-/// CreateEnumerator - Create a single enumerator value.
-DIEnumerator DIFactory::CreateEnumerator(StringRef Name, uint64_t Val){
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_enumerator),
- MDString::get(VMContext, Name),
- ConstantInt::get(Type::getInt64Ty(VMContext), Val)
- };
- return DIEnumerator(MDNode::get(VMContext, &Elts[0], 3));
-}
-
-
-/// CreateBasicType - Create a basic type like int, float, etc.
-DIBasicType DIFactory::CreateBasicType(DIDescriptor Context,
- StringRef Name,
- DIFile F,
- unsigned LineNumber,
- uint64_t SizeInBits,
- uint64_t AlignInBits,
- uint64_t OffsetInBits, unsigned Flags,
- unsigned Encoding) {
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_base_type),
- Context,
- MDString::get(VMContext, Name),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
- ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
- ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
- ConstantInt::get(Type::getInt64Ty(VMContext), OffsetInBits),
- ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
- ConstantInt::get(Type::getInt32Ty(VMContext), Encoding)
- };
- return DIBasicType(MDNode::get(VMContext, &Elts[0], 10));
-}
-
-
-/// CreateBasicType - Create a basic type like int, float, etc.
-DIBasicType DIFactory::CreateBasicTypeEx(DIDescriptor Context,
- StringRef Name,
- DIFile F,
- unsigned LineNumber,
- Constant *SizeInBits,
- Constant *AlignInBits,
- Constant *OffsetInBits, unsigned Flags,
- unsigned Encoding) {
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_base_type),
- Context,
- MDString::get(VMContext, Name),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
- SizeInBits,
- AlignInBits,
- OffsetInBits,
- ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
- ConstantInt::get(Type::getInt32Ty(VMContext), Encoding)
- };
- return DIBasicType(MDNode::get(VMContext, &Elts[0], 10));
-}
-
-/// CreateArtificialType - Create a new DIType with "artificial" flag set.
-DIType DIFactory::CreateArtificialType(DIType Ty) {
- if (Ty.isArtificial())
- return Ty;
-
- SmallVector<Value *, 9> Elts;
- MDNode *N = Ty;
- assert (N && "Unexpected input DIType!");
- for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
- if (Value *V = N->getOperand(i))
- Elts.push_back(V);
+ if (isObjCLike && (C == '[' || C == ']' || C == ' ' || C == ':' ||
+ C == '+' || C == '(' || C == ')'))
+ Out.push_back('.');
else
- Elts.push_back(Constant::getNullValue(Type::getInt32Ty(VMContext)));
+ Out.push_back(C);
}
-
- unsigned CurFlags = Ty.getFlags();
- CurFlags = CurFlags | DIType::FlagArtificial;
-
- // Flags are stored at this slot.
- Elts[8] = ConstantInt::get(Type::getInt32Ty(VMContext), CurFlags);
-
- return DIType(MDNode::get(VMContext, Elts.data(), Elts.size()));
}
-/// CreateDerivedType - Create a derived type like const qualified type,
-/// pointer, typedef, etc.
-DIDerivedType DIFactory::CreateDerivedType(unsigned Tag,
- DIDescriptor Context,
- StringRef Name,
- DIFile F,
- unsigned LineNumber,
- uint64_t SizeInBits,
- uint64_t AlignInBits,
- uint64_t OffsetInBits,
- unsigned Flags,
- DIType DerivedFrom) {
- Value *Elts[] = {
- GetTagConstant(Tag),
- Context,
- MDString::get(VMContext, Name),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
- ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
- ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
- ConstantInt::get(Type::getInt64Ty(VMContext), OffsetInBits),
- ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
- DerivedFrom,
- };
- return DIDerivedType(MDNode::get(VMContext, &Elts[0], 10));
+/// getFnSpecificMDNode - Return a NameMDNode, if available, that is
+/// suitable to hold function specific information.
+NamedMDNode *llvm::getFnSpecificMDNode(const Module &M, DISubprogram Fn) {
+ SmallString<32> Name = StringRef("llvm.dbg.lv.");
+ StringRef FName = "fn";
+ if (Fn.getFunction())
+ FName = Fn.getFunction()->getName();
+ else
+ FName = Fn.getName();
+ char One = '\1';
+ if (FName.startswith(StringRef(&One, 1)))
+ FName = FName.substr(1);
+ fixupObjcLikeName(FName, Name);
+ return M.getNamedMetadata(Name.str());
}
-
-/// CreateDerivedType - Create a derived type like const qualified type,
-/// pointer, typedef, etc.
-DIDerivedType DIFactory::CreateDerivedTypeEx(unsigned Tag,
- DIDescriptor Context,
- StringRef Name,
- DIFile F,
- unsigned LineNumber,
- Constant *SizeInBits,
- Constant *AlignInBits,
- Constant *OffsetInBits,
- unsigned Flags,
- DIType DerivedFrom) {
- Value *Elts[] = {
- GetTagConstant(Tag),
- Context,
- MDString::get(VMContext, Name),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
- SizeInBits,
- AlignInBits,
- OffsetInBits,
- ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
- DerivedFrom,
- };
- return DIDerivedType(MDNode::get(VMContext, &Elts[0], 10));
+/// getOrInsertFnSpecificMDNode - Return a NameMDNode that is suitable
+/// to hold function specific information.
+NamedMDNode *llvm::getOrInsertFnSpecificMDNode(Module &M, DISubprogram Fn) {
+ SmallString<32> Name = StringRef("llvm.dbg.lv.");
+ StringRef FName = "fn";
+ if (Fn.getFunction())
+ FName = Fn.getFunction()->getName();
+ else
+ FName = Fn.getName();
+ char One = '\1';
+ if (FName.startswith(StringRef(&One, 1)))
+ FName = FName.substr(1);
+ fixupObjcLikeName(FName, Name);
+
+ return M.getOrInsertNamedMetadata(Name.str());
}
-
-/// CreateCompositeType - Create a composite type like array, struct, etc.
-DICompositeType DIFactory::CreateCompositeType(unsigned Tag,
- DIDescriptor Context,
- StringRef Name,
- DIFile F,
- unsigned LineNumber,
- uint64_t SizeInBits,
- uint64_t AlignInBits,
- uint64_t OffsetInBits,
- unsigned Flags,
- DIType DerivedFrom,
- DIArray Elements,
- unsigned RuntimeLang,
- MDNode *ContainingType) {
-
- Value *Elts[] = {
- GetTagConstant(Tag),
- Context,
- MDString::get(VMContext, Name),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
- ConstantInt::get(Type::getInt64Ty(VMContext), SizeInBits),
- ConstantInt::get(Type::getInt64Ty(VMContext), AlignInBits),
- ConstantInt::get(Type::getInt64Ty(VMContext), OffsetInBits),
- ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
- DerivedFrom,
- Elements,
- ConstantInt::get(Type::getInt32Ty(VMContext), RuntimeLang),
- ContainingType
- };
-
- MDNode *Node = MDNode::get(VMContext, &Elts[0], 13);
- // Create a named metadata so that we do not lose this enum info.
- if (Tag == dwarf::DW_TAG_enumeration_type) {
- NamedMDNode *NMD = M.getOrInsertNamedMetadata("llvm.dbg.enum");
- NMD->addOperand(Node);
- }
- return DICompositeType(Node);
+/// createInlinedVariable - Create a new inlined variable based on current
+/// variable.
+/// @param DV Current Variable.
+/// @param InlinedScope Location at current variable is inlined.
+DIVariable llvm::createInlinedVariable(MDNode *DV, MDNode *InlinedScope,
+ LLVMContext &VMContext) {
+ SmallVector<Value *, 16> Elts;
+ // Insert inlined scope as 7th element.
+ for (unsigned i = 0, e = DV->getNumOperands(); i != e; ++i)
+ i == 7 ? Elts.push_back(InlinedScope) :
+ Elts.push_back(DV->getOperand(i));
+ return DIVariable(MDNode::get(VMContext, Elts));
}
-
-/// CreateCompositeType - Create a composite type like array, struct, etc.
-DICompositeType DIFactory::CreateCompositeTypeEx(unsigned Tag,
- DIDescriptor Context,
- StringRef Name,
- DIFile F,
- unsigned LineNumber,
- Constant *SizeInBits,
- Constant *AlignInBits,
- Constant *OffsetInBits,
- unsigned Flags,
- DIType DerivedFrom,
- DIArray Elements,
- unsigned RuntimeLang,
- MDNode *ContainingType) {
- Value *Elts[] = {
- GetTagConstant(Tag),
- Context,
- MDString::get(VMContext, Name),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNumber),
- SizeInBits,
- AlignInBits,
- OffsetInBits,
- ConstantInt::get(Type::getInt32Ty(VMContext), Flags),
- DerivedFrom,
- Elements,
- ConstantInt::get(Type::getInt32Ty(VMContext), RuntimeLang),
- ContainingType
- };
- MDNode *Node = MDNode::get(VMContext, &Elts[0], 13);
- // Create a named metadata so that we do not lose this enum info.
- if (Tag == dwarf::DW_TAG_enumeration_type) {
- NamedMDNode *NMD = M.getOrInsertNamedMetadata("llvm.dbg.enum");
- NMD->addOperand(Node);
- }
- return DICompositeType(Node);
-}
-
-
-/// CreateSubprogram - Create a new descriptor for the specified subprogram.
-/// See comments in DISubprogram for descriptions of these fields. This
-/// method does not unique the generated descriptors.
-DISubprogram DIFactory::CreateSubprogram(DIDescriptor Context,
- StringRef Name,
- StringRef DisplayName,
- StringRef LinkageName,
- DIFile F,
- unsigned LineNo, DIType Ty,
- bool isLocalToUnit,
- bool isDefinition,
- unsigned VK, unsigned VIndex,
- DIType ContainingType,
- bool isArtificial,
- bool isOptimized,
- Function *Fn) {
-
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_subprogram),
- llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
- Context,
- MDString::get(VMContext, Name),
- MDString::get(VMContext, DisplayName),
- MDString::get(VMContext, LinkageName),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
- Ty,
- ConstantInt::get(Type::getInt1Ty(VMContext), isLocalToUnit),
- ConstantInt::get(Type::getInt1Ty(VMContext), isDefinition),
- ConstantInt::get(Type::getInt32Ty(VMContext), (unsigned)VK),
- ConstantInt::get(Type::getInt32Ty(VMContext), VIndex),
- ContainingType,
- ConstantInt::get(Type::getInt1Ty(VMContext), isArtificial),
- ConstantInt::get(Type::getInt1Ty(VMContext), isOptimized),
- Fn
- };
- MDNode *Node = MDNode::get(VMContext, &Elts[0], 17);
-
- // Create a named metadata so that we do not lose this mdnode.
- NamedMDNode *NMD = M.getOrInsertNamedMetadata("llvm.dbg.sp");
- NMD->addOperand(Node);
- return DISubprogram(Node);
-}
-
-/// CreateSubprogramDefinition - Create new subprogram descriptor for the
-/// given declaration.
-DISubprogram DIFactory::CreateSubprogramDefinition(DISubprogram &SPDeclaration){
- if (SPDeclaration.isDefinition())
- return DISubprogram(SPDeclaration);
-
- MDNode *DeclNode = SPDeclaration;
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_subprogram),
- llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
- DeclNode->getOperand(2), // Context
- DeclNode->getOperand(3), // Name
- DeclNode->getOperand(4), // DisplayName
- DeclNode->getOperand(5), // LinkageName
- DeclNode->getOperand(6), // CompileUnit
- DeclNode->getOperand(7), // LineNo
- DeclNode->getOperand(8), // Type
- DeclNode->getOperand(9), // isLocalToUnit
- ConstantInt::get(Type::getInt1Ty(VMContext), true),
- DeclNode->getOperand(11), // Virtuality
- DeclNode->getOperand(12), // VIndex
- DeclNode->getOperand(13), // Containting Type
- DeclNode->getOperand(14), // isArtificial
- DeclNode->getOperand(15), // isOptimized
- SPDeclaration.getFunction()
- };
- MDNode *Node =MDNode::get(VMContext, &Elts[0], 16);
-
- // Create a named metadata so that we do not lose this mdnode.
- NamedMDNode *NMD = M.getOrInsertNamedMetadata("llvm.dbg.sp");
- NMD->addOperand(Node);
- return DISubprogram(Node);
-}
-
-/// CreateGlobalVariable - Create a new descriptor for the specified global.
-DIGlobalVariable
-DIFactory::CreateGlobalVariable(DIDescriptor Context, StringRef Name,
- StringRef DisplayName,
- StringRef LinkageName,
- DIFile F,
- unsigned LineNo, DIType Ty,bool isLocalToUnit,
- bool isDefinition, llvm::GlobalVariable *Val) {
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_variable),
- llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
- Context,
- MDString::get(VMContext, Name),
- MDString::get(VMContext, DisplayName),
- MDString::get(VMContext, LinkageName),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
- Ty,
- ConstantInt::get(Type::getInt1Ty(VMContext), isLocalToUnit),
- ConstantInt::get(Type::getInt1Ty(VMContext), isDefinition),
- Val
- };
-
- Value *const *Vs = &Elts[0];
- MDNode *Node = MDNode::get(VMContext,Vs, 12);
-
- // Create a named metadata so that we do not lose this mdnode.
- NamedMDNode *NMD = M.getOrInsertNamedMetadata("llvm.dbg.gv");
- NMD->addOperand(Node);
-
- return DIGlobalVariable(Node);
-}
-
-/// CreateGlobalVariable - Create a new descriptor for the specified constant.
-DIGlobalVariable
-DIFactory::CreateGlobalVariable(DIDescriptor Context, StringRef Name,
- StringRef DisplayName,
- StringRef LinkageName,
- DIFile F,
- unsigned LineNo, DIType Ty,bool isLocalToUnit,
- bool isDefinition, llvm::Constant *Val) {
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_constant),
- llvm::Constant::getNullValue(Type::getInt32Ty(VMContext)),
- Context,
- MDString::get(VMContext, Name),
- MDString::get(VMContext, DisplayName),
- MDString::get(VMContext, LinkageName),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
- Ty,
- ConstantInt::get(Type::getInt1Ty(VMContext), isLocalToUnit),
- ConstantInt::get(Type::getInt1Ty(VMContext), isDefinition),
- Val
- };
-
- Value *const *Vs = &Elts[0];
- MDNode *Node = MDNode::get(VMContext,Vs, 12);
-
- // Create a named metadata so that we do not lose this mdnode.
- NamedMDNode *NMD = M.getOrInsertNamedMetadata("llvm.dbg.gv");
- NMD->addOperand(Node);
-
- return DIGlobalVariable(Node);
-}
-
-/// CreateVariable - Create a new descriptor for the specified variable.
-DIVariable DIFactory::CreateVariable(unsigned Tag, DIDescriptor Context,
- StringRef Name,
- DIFile F,
- unsigned LineNo,
- DIType Ty, bool AlwaysPreserve) {
- Value *Elts[] = {
- GetTagConstant(Tag),
- Context,
- MDString::get(VMContext, Name),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
- Ty,
- };
- MDNode *Node = MDNode::get(VMContext, &Elts[0], 6);
- if (AlwaysPreserve) {
- // The optimizer may remove local variable. If there is an interest
- // to preserve variable info in such situation then stash it in a
- // named mdnode.
- DISubprogram Fn(getDISubprogram(Context));
- StringRef FName = "fn";
- if (Fn.getFunction())
- FName = Fn.getFunction()->getName();
- char One = '\1';
- if (FName.startswith(StringRef(&One, 1)))
- FName = FName.substr(1);
-
- SmallString<32> Out;
- NamedMDNode *FnLocals =
- M.getOrInsertNamedMetadata(Twine("llvm.dbg.lv.", FName).toStringRef(Out));
- FnLocals->addOperand(Node);
- }
- return DIVariable(Node);
-}
-
-
-/// CreateComplexVariable - Create a new descriptor for the specified variable
-/// which has a complex address expression for its address.
-DIVariable DIFactory::CreateComplexVariable(unsigned Tag, DIDescriptor Context,
- const std::string &Name,
- DIFile F,
- unsigned LineNo,
- DIType Ty,
- SmallVector<Value *, 9> &addr) {
- SmallVector<Value *, 9> Elts;
- Elts.push_back(GetTagConstant(Tag));
- Elts.push_back(Context);
- Elts.push_back(MDString::get(VMContext, Name));
- Elts.push_back(F);
- Elts.push_back(ConstantInt::get(Type::getInt32Ty(VMContext), LineNo));
- Elts.push_back(Ty);
- Elts.insert(Elts.end(), addr.begin(), addr.end());
-
- return DIVariable(MDNode::get(VMContext, &Elts[0], 6+addr.size()));
-}
-
-
-/// CreateBlock - This creates a descriptor for a lexical block with the
-/// specified parent VMContext.
-DILexicalBlock DIFactory::CreateLexicalBlock(DIDescriptor Context,
- DIFile F, unsigned LineNo,
- unsigned Col) {
- // Defeat MDNode uniqing for lexical blocks.
- static unsigned int unique_id = 0;
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_lexical_block),
- Context,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
- ConstantInt::get(Type::getInt32Ty(VMContext), Col),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), unique_id++)
- };
- return DILexicalBlock(MDNode::get(VMContext, &Elts[0], 6));
-}
-
-/// CreateNameSpace - This creates new descriptor for a namespace
-/// with the specified parent context.
-DINameSpace DIFactory::CreateNameSpace(DIDescriptor Context, StringRef Name,
- DIFile F,
- unsigned LineNo) {
- Value *Elts[] = {
- GetTagConstant(dwarf::DW_TAG_namespace),
- Context,
- MDString::get(VMContext, Name),
- F,
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNo)
- };
- return DINameSpace(MDNode::get(VMContext, &Elts[0], 5));
-}
-
-/// CreateLocation - Creates a debug info location.
-DILocation DIFactory::CreateLocation(unsigned LineNo, unsigned ColumnNo,
- DIScope S, DILocation OrigLoc) {
- Value *Elts[] = {
- ConstantInt::get(Type::getInt32Ty(VMContext), LineNo),
- ConstantInt::get(Type::getInt32Ty(VMContext), ColumnNo),
- S,
- OrigLoc,
- };
- return DILocation(MDNode::get(VMContext, &Elts[0], 4));
+/// cleanseInlinedVariable - Remove inlined scope from the variable.
+DIVariable llvm::cleanseInlinedVariable(MDNode *DV, LLVMContext &VMContext) {
+ SmallVector<Value *, 16> Elts;
+ // Insert inlined scope as 7th element.
+ for (unsigned i = 0, e = DV->getNumOperands(); i != e; ++i)
+ i == 7 ?
+ Elts.push_back(llvm::Constant::getNullValue(Type::getInt32Ty(VMContext))):
+ Elts.push_back(DV->getOperand(i));
+ return DIVariable(MDNode::get(VMContext, Elts));
}
//===----------------------------------------------------------------------===//
@@ -1225,10 +927,17 @@
/// processModule - Process entire module and collect debug info.
void DebugInfoFinder::processModule(Module &M) {
+ if (NamedMDNode *CU_Nodes = M.getNamedMetadata("llvm.dbg.cu"))
+ for (unsigned i = 0, e = CU_Nodes->getNumOperands(); i != e; ++i)
+ addCompileUnit(DICompileUnit(CU_Nodes->getOperand(i)));
+
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
for (Function::iterator FI = (*I).begin(), FE = (*I).end(); FI != FE; ++FI)
for (BasicBlock::iterator BI = (*FI).begin(), BE = (*FI).end(); BI != BE;
++BI) {
+ if (DbgDeclareInst *DDI = dyn_cast<DbgDeclareInst>(BI))
+ processDeclare(DDI);
+
DebugLoc Loc = BI->getDebugLoc();
if (Loc.isUnknown())
continue;
@@ -1240,6 +949,10 @@
addCompileUnit(DICompileUnit(Scope));
else if (Scope.isSubprogram())
processSubprogram(DISubprogram(Scope));
+ else if (Scope.isLexicalBlockFile()) {
+ DILexicalBlockFile DBF = DILexicalBlockFile(Scope);
+ processLexicalBlock(DILexicalBlock(DBF.getScope()));
+ }
else if (Scope.isLexicalBlock())
processLexicalBlock(DILexicalBlock(Scope));
@@ -1251,7 +964,8 @@
for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i) {
DIGlobalVariable DIG(cast<MDNode>(NMD->getOperand(i)));
if (addGlobalVariable(DIG)) {
- addCompileUnit(DIG.getCompileUnit());
+ if (DIG.getVersion() <= LLVMDebugVersion10)
+ addCompileUnit(DIG.getCompileUnit());
processType(DIG.getType());
}
}
@@ -1272,6 +986,10 @@
processSubprogram(DISubprogram(S));
else if (S.isLexicalBlock())
processLexicalBlock(DILexicalBlock(S));
+ else if (S.isLexicalBlockFile()) {
+ DILexicalBlockFile DBF = DILexicalBlockFile(S);
+ processLexicalBlock(DILexicalBlock(DBF.getScope()));
+ }
processLocation(Loc.getOrigLocation());
}
@@ -1279,8 +997,8 @@
void DebugInfoFinder::processType(DIType DT) {
if (!addType(DT))
return;
-
- addCompileUnit(DT.getCompileUnit());
+ if (DT.getVersion() <= LLVMDebugVersion10)
+ addCompileUnit(DT.getCompileUnit());
if (DT.isCompositeType()) {
DICompositeType DCT(DT);
processType(DCT.getTypeDerivedFrom());
@@ -1303,6 +1021,10 @@
DIScope Context = LB.getContext();
if (Context.isLexicalBlock())
return processLexicalBlock(DILexicalBlock(Context));
+ else if (Context.isLexicalBlockFile()) {
+ DILexicalBlockFile DBF = DILexicalBlockFile(Context);
+ return processLexicalBlock(DILexicalBlock(DBF.getScope()));
+ }
else
return processSubprogram(DISubprogram(Context));
}
@@ -1311,10 +1033,27 @@
void DebugInfoFinder::processSubprogram(DISubprogram SP) {
if (!addSubprogram(SP))
return;
- addCompileUnit(SP.getCompileUnit());
+ if (SP.getVersion() <= LLVMDebugVersion10)
+ addCompileUnit(SP.getCompileUnit());
processType(SP.getType());
}
+/// processDeclare - Process DbgDeclareInst.
+void DebugInfoFinder::processDeclare(DbgDeclareInst *DDI) {
+ MDNode *N = dyn_cast<MDNode>(DDI->getVariable());
+ if (!N) return;
+
+ DIDescriptor DV(N);
+ if (!DV.isVariable())
+ return;
+
+ if (!NodesSeen.insert(DV))
+ return;
+ if (DIVariable(N).getVersion() <= LLVMDebugVersion10)
+ addCompileUnit(DIVariable(N).getCompileUnit());
+ processType(DIVariable(N).getType());
+}
+
/// addType - Add type into Tys.
bool DebugInfoFinder::addType(DIType DT) {
if (!DT.isValid())
@@ -1363,62 +1102,15 @@
return true;
}
-/// Find the debug info descriptor corresponding to this global variable.
-static Value *findDbgGlobalDeclare(GlobalVariable *V) {
- const Module *M = V->getParent();
- NamedMDNode *NMD = M->getNamedMetadata("llvm.dbg.gv");
- if (!NMD)
- return 0;
-
- for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i) {
- DIDescriptor DIG(cast<MDNode>(NMD->getOperand(i)));
- if (!DIG.isGlobalVariable())
- continue;
- if (DIGlobalVariable(DIG).getGlobal() == V)
- return DIG;
- }
- return 0;
-}
-
-bool llvm::getLocationInfo(const Value *V, std::string &DisplayName,
- std::string &Type, unsigned &LineNo,
- std::string &File, std::string &Dir) {
- DICompileUnit Unit;
- DIType TypeD;
-
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(const_cast<Value*>(V))) {
- Value *DIGV = findDbgGlobalDeclare(GV);
- if (!DIGV) return false;
- DIGlobalVariable Var(cast<MDNode>(DIGV));
-
- StringRef D = Var.getDisplayName();
- if (!D.empty())
- DisplayName = D;
- LineNo = Var.getLineNumber();
- Unit = Var.getCompileUnit();
- TypeD = Var.getType();
- } else {
- return false;
- }
-
- StringRef T = TypeD.getName();
- if (!T.empty())
- Type = T;
- StringRef F = Unit.getFilename();
- if (!F.empty())
- File = F;
- StringRef D = Unit.getDirectory();
- if (!D.empty())
- Dir = D;
- return true;
-}
-
/// getDISubprogram - Find subprogram that is enclosing this scope.
DISubprogram llvm::getDISubprogram(const MDNode *Scope) {
DIDescriptor D(Scope);
if (D.isSubprogram())
return DISubprogram(Scope);
+ if (D.isLexicalBlockFile())
+ return getDISubprogram(DILexicalBlockFile(Scope).getContext());
+
if (D.isLexicalBlock())
return getDISubprogram(DILexicalBlock(Scope).getContext());
@@ -1435,3 +1127,17 @@
return DICompositeType();
}
+
+/// isSubprogramContext - Return true if Context is either a subprogram
+/// or another context nested inside a subprogram.
+bool llvm::isSubprogramContext(const MDNode *Context) {
+ if (!Context)
+ return false;
+ DIDescriptor D(Context);
+ if (D.isSubprogram())
+ return true;
+ if (D.isType())
+ return isSubprogramContext(DIType(Context).getContext());
+ return false;
+}
+
diff --git a/src/LLVM/lib/Analysis/DomPrinter.cpp b/src/LLVM/lib/Analysis/DomPrinter.cpp
new file mode 100644
index 0000000..cde4314
--- /dev/null
+++ b/src/LLVM/lib/Analysis/DomPrinter.cpp
@@ -0,0 +1,232 @@
+//===- DomPrinter.cpp - DOT printer for the dominance trees ------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines '-dot-dom' and '-dot-postdom' analysis passes, which emit
+// a dom.<fnname>.dot or postdom.<fnname>.dot file for each function in the
+// program, with a graph of the dominance/postdominance tree of that
+// function.
+//
+// There are also passes available to directly call dotty ('-view-dom' or
+// '-view-postdom'). By appending '-only' like '-dot-dom-only' only the
+// names of the bbs are printed, but the content is hidden.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/DomPrinter.h"
+#include "llvm/Analysis/DOTGraphTraitsPass.h"
+#include "llvm/Analysis/PostDominators.h"
+
+using namespace llvm;
+
+namespace llvm {
+template<>
+struct DOTGraphTraits<DomTreeNode*> : public DefaultDOTGraphTraits {
+
+ DOTGraphTraits (bool isSimple=false)
+ : DefaultDOTGraphTraits(isSimple) {}
+
+ std::string getNodeLabel(DomTreeNode *Node, DomTreeNode *Graph) {
+
+ BasicBlock *BB = Node->getBlock();
+
+ if (!BB)
+ return "Post dominance root node";
+
+
+ if (isSimple())
+ return DOTGraphTraits<const Function*>
+ ::getSimpleNodeLabel(BB, BB->getParent());
+ else
+ return DOTGraphTraits<const Function*>
+ ::getCompleteNodeLabel(BB, BB->getParent());
+ }
+};
+
+template<>
+struct DOTGraphTraits<DominatorTree*> : public DOTGraphTraits<DomTreeNode*> {
+
+ DOTGraphTraits (bool isSimple=false)
+ : DOTGraphTraits<DomTreeNode*>(isSimple) {}
+
+ static std::string getGraphName(DominatorTree *DT) {
+ return "Dominator tree";
+ }
+
+ std::string getNodeLabel(DomTreeNode *Node, DominatorTree *G) {
+ return DOTGraphTraits<DomTreeNode*>::getNodeLabel(Node, G->getRootNode());
+ }
+};
+
+template<>
+struct DOTGraphTraits<PostDominatorTree*>
+ : public DOTGraphTraits<DomTreeNode*> {
+
+ DOTGraphTraits (bool isSimple=false)
+ : DOTGraphTraits<DomTreeNode*>(isSimple) {}
+
+ static std::string getGraphName(PostDominatorTree *DT) {
+ return "Post dominator tree";
+ }
+
+ std::string getNodeLabel(DomTreeNode *Node, PostDominatorTree *G ) {
+ return DOTGraphTraits<DomTreeNode*>::getNodeLabel(Node, G->getRootNode());
+ }
+};
+}
+
+namespace {
+struct DomViewer
+ : public DOTGraphTraitsViewer<DominatorTree, false> {
+ static char ID;
+ DomViewer() : DOTGraphTraitsViewer<DominatorTree, false>("dom", ID){
+ initializeDomViewerPass(*PassRegistry::getPassRegistry());
+ }
+};
+
+struct DomOnlyViewer
+ : public DOTGraphTraitsViewer<DominatorTree, true> {
+ static char ID;
+ DomOnlyViewer() : DOTGraphTraitsViewer<DominatorTree, true>("domonly", ID){
+ initializeDomOnlyViewerPass(*PassRegistry::getPassRegistry());
+ }
+};
+
+struct PostDomViewer
+ : public DOTGraphTraitsViewer<PostDominatorTree, false> {
+ static char ID;
+ PostDomViewer() :
+ DOTGraphTraitsViewer<PostDominatorTree, false>("postdom", ID){
+ initializePostDomViewerPass(*PassRegistry::getPassRegistry());
+ }
+};
+
+struct PostDomOnlyViewer
+ : public DOTGraphTraitsViewer<PostDominatorTree, true> {
+ static char ID;
+ PostDomOnlyViewer() :
+ DOTGraphTraitsViewer<PostDominatorTree, true>("postdomonly", ID){
+ initializePostDomOnlyViewerPass(*PassRegistry::getPassRegistry());
+ }
+};
+} // end anonymous namespace
+
+char DomViewer::ID = 0;
+INITIALIZE_PASS(DomViewer, "view-dom",
+ "View dominance tree of function", false, false)
+
+char DomOnlyViewer::ID = 0;
+INITIALIZE_PASS(DomOnlyViewer, "view-dom-only",
+ "View dominance tree of function (with no function bodies)",
+ false, false)
+
+char PostDomViewer::ID = 0;
+INITIALIZE_PASS(PostDomViewer, "view-postdom",
+ "View postdominance tree of function", false, false)
+
+char PostDomOnlyViewer::ID = 0;
+INITIALIZE_PASS(PostDomOnlyViewer, "view-postdom-only",
+ "View postdominance tree of function "
+ "(with no function bodies)",
+ false, false)
+
+namespace {
+struct DomPrinter
+ : public DOTGraphTraitsPrinter<DominatorTree, false> {
+ static char ID;
+ DomPrinter() : DOTGraphTraitsPrinter<DominatorTree, false>("dom", ID) {
+ initializeDomPrinterPass(*PassRegistry::getPassRegistry());
+ }
+};
+
+struct DomOnlyPrinter
+ : public DOTGraphTraitsPrinter<DominatorTree, true> {
+ static char ID;
+ DomOnlyPrinter() : DOTGraphTraitsPrinter<DominatorTree, true>("domonly", ID) {
+ initializeDomOnlyPrinterPass(*PassRegistry::getPassRegistry());
+ }
+};
+
+struct PostDomPrinter
+ : public DOTGraphTraitsPrinter<PostDominatorTree, false> {
+ static char ID;
+ PostDomPrinter() :
+ DOTGraphTraitsPrinter<PostDominatorTree, false>("postdom", ID) {
+ initializePostDomPrinterPass(*PassRegistry::getPassRegistry());
+ }
+};
+
+struct PostDomOnlyPrinter
+ : public DOTGraphTraitsPrinter<PostDominatorTree, true> {
+ static char ID;
+ PostDomOnlyPrinter() :
+ DOTGraphTraitsPrinter<PostDominatorTree, true>("postdomonly", ID) {
+ initializePostDomOnlyPrinterPass(*PassRegistry::getPassRegistry());
+ }
+};
+} // end anonymous namespace
+
+
+
+char DomPrinter::ID = 0;
+INITIALIZE_PASS(DomPrinter, "dot-dom",
+ "Print dominance tree of function to 'dot' file",
+ false, false)
+
+char DomOnlyPrinter::ID = 0;
+INITIALIZE_PASS(DomOnlyPrinter, "dot-dom-only",
+ "Print dominance tree of function to 'dot' file "
+ "(with no function bodies)",
+ false, false)
+
+char PostDomPrinter::ID = 0;
+INITIALIZE_PASS(PostDomPrinter, "dot-postdom",
+ "Print postdominance tree of function to 'dot' file",
+ false, false)
+
+char PostDomOnlyPrinter::ID = 0;
+INITIALIZE_PASS(PostDomOnlyPrinter, "dot-postdom-only",
+ "Print postdominance tree of function to 'dot' file "
+ "(with no function bodies)",
+ false, false)
+
+// Create methods available outside of this file, to use them
+// "include/llvm/LinkAllPasses.h". Otherwise the pass would be deleted by
+// the link time optimization.
+
+FunctionPass *llvm::createDomPrinterPass() {
+ return new DomPrinter();
+}
+
+FunctionPass *llvm::createDomOnlyPrinterPass() {
+ return new DomOnlyPrinter();
+}
+
+FunctionPass *llvm::createDomViewerPass() {
+ return new DomViewer();
+}
+
+FunctionPass *llvm::createDomOnlyViewerPass() {
+ return new DomOnlyViewer();
+}
+
+FunctionPass *llvm::createPostDomPrinterPass() {
+ return new PostDomPrinter();
+}
+
+FunctionPass *llvm::createPostDomOnlyPrinterPass() {
+ return new PostDomOnlyPrinter();
+}
+
+FunctionPass *llvm::createPostDomViewerPass() {
+ return new PostDomViewer();
+}
+
+FunctionPass *llvm::createPostDomOnlyViewerPass() {
+ return new PostDomOnlyViewer();
+}
diff --git a/src/LLVM/lib/Analysis/DominanceFrontier.cpp b/src/LLVM/lib/Analysis/DominanceFrontier.cpp
new file mode 100644
index 0000000..6de4e1e
--- /dev/null
+++ b/src/LLVM/lib/Analysis/DominanceFrontier.cpp
@@ -0,0 +1,137 @@
+//===- DominanceFrontier.cpp - Dominance Frontier Calculation -------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/DominanceFrontier.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+char DominanceFrontier::ID = 0;
+INITIALIZE_PASS_BEGIN(DominanceFrontier, "domfrontier",
+ "Dominance Frontier Construction", true, true)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_PASS_END(DominanceFrontier, "domfrontier",
+ "Dominance Frontier Construction", true, true)
+
+namespace {
+ class DFCalculateWorkObject {
+ public:
+ DFCalculateWorkObject(BasicBlock *B, BasicBlock *P,
+ const DomTreeNode *N,
+ const DomTreeNode *PN)
+ : currentBB(B), parentBB(P), Node(N), parentNode(PN) {}
+ BasicBlock *currentBB;
+ BasicBlock *parentBB;
+ const DomTreeNode *Node;
+ const DomTreeNode *parentNode;
+ };
+}
+
+const DominanceFrontier::DomSetType &
+DominanceFrontier::calculate(const DominatorTree &DT,
+ const DomTreeNode *Node) {
+ BasicBlock *BB = Node->getBlock();
+ DomSetType *Result = NULL;
+
+ std::vector<DFCalculateWorkObject> workList;
+ SmallPtrSet<BasicBlock *, 32> visited;
+
+ workList.push_back(DFCalculateWorkObject(BB, NULL, Node, NULL));
+ do {
+ DFCalculateWorkObject *currentW = &workList.back();
+ assert (currentW && "Missing work object.");
+
+ BasicBlock *currentBB = currentW->currentBB;
+ BasicBlock *parentBB = currentW->parentBB;
+ const DomTreeNode *currentNode = currentW->Node;
+ const DomTreeNode *parentNode = currentW->parentNode;
+ assert (currentBB && "Invalid work object. Missing current Basic Block");
+ assert (currentNode && "Invalid work object. Missing current Node");
+ DomSetType &S = Frontiers[currentBB];
+
+ // Visit each block only once.
+ if (visited.count(currentBB) == 0) {
+ visited.insert(currentBB);
+
+ // Loop over CFG successors to calculate DFlocal[currentNode]
+ for (succ_iterator SI = succ_begin(currentBB), SE = succ_end(currentBB);
+ SI != SE; ++SI) {
+ // Does Node immediately dominate this successor?
+ if (DT[*SI]->getIDom() != currentNode)
+ S.insert(*SI);
+ }
+ }
+
+ // At this point, S is DFlocal. Now we union in DFup's of our children...
+ // Loop through and visit the nodes that Node immediately dominates (Node's
+ // children in the IDomTree)
+ bool visitChild = false;
+ for (DomTreeNode::const_iterator NI = currentNode->begin(),
+ NE = currentNode->end(); NI != NE; ++NI) {
+ DomTreeNode *IDominee = *NI;
+ BasicBlock *childBB = IDominee->getBlock();
+ if (visited.count(childBB) == 0) {
+ workList.push_back(DFCalculateWorkObject(childBB, currentBB,
+ IDominee, currentNode));
+ visitChild = true;
+ }
+ }
+
+ // If all children are visited or there is any child then pop this block
+ // from the workList.
+ if (!visitChild) {
+
+ if (!parentBB) {
+ Result = &S;
+ break;
+ }
+
+ DomSetType::const_iterator CDFI = S.begin(), CDFE = S.end();
+ DomSetType &parentSet = Frontiers[parentBB];
+ for (; CDFI != CDFE; ++CDFI) {
+ if (!DT.properlyDominates(parentNode, DT[*CDFI]))
+ parentSet.insert(*CDFI);
+ }
+ workList.pop_back();
+ }
+
+ } while (!workList.empty());
+
+ return *Result;
+}
+
+void DominanceFrontierBase::print(raw_ostream &OS, const Module* ) const {
+ for (const_iterator I = begin(), E = end(); I != E; ++I) {
+ OS << " DomFrontier for BB ";
+ if (I->first)
+ WriteAsOperand(OS, I->first, false);
+ else
+ OS << " <<exit node>>";
+ OS << " is:\t";
+
+ const std::set<BasicBlock*> &BBs = I->second;
+
+ for (std::set<BasicBlock*>::const_iterator I = BBs.begin(), E = BBs.end();
+ I != E; ++I) {
+ OS << ' ';
+ if (*I)
+ WriteAsOperand(OS, *I, false);
+ else
+ OS << "<<exit node>>";
+ }
+ OS << "\n";
+ }
+}
+
+void DominanceFrontierBase::dump() const {
+ print(dbgs());
+}
+
diff --git a/src/LLVM/lib/Analysis/INSTALL.vcxproj b/src/LLVM/lib/Analysis/INSTALL.vcxproj
new file mode 100644
index 0000000..ef2c2c1
--- /dev/null
+++ b/src/LLVM/lib/Analysis/INSTALL.vcxproj
@@ -0,0 +1,261 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup Label="ProjectConfigurations">
+ <ProjectConfiguration Include="Debug|Win32">
+ <Configuration>Debug</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Release|Win32">
+ <Configuration>Release</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="MinSizeRel|Win32">
+ <Configuration>MinSizeRel</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="RelWithDebInfo|Win32">
+ <Configuration>RelWithDebInfo</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ </ItemGroup>
+ <PropertyGroup Label="Globals">
+ <ProjectGUID>{048BB775-7681-4EE1-AACF-5A067ACEEEA5}</ProjectGUID>
+ <Keyword>Win32Proj</Keyword>
+ <Platform>Win32</Platform>
+ <ProjectName>INSTALL</ProjectName>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+ <ImportGroup Label="ExtensionSettings">
+ </ImportGroup>
+ <ImportGroup Label="PropertySheets">
+ <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" /> </ImportGroup>
+ <PropertyGroup Label="UserMacros" />
+ <PropertyGroup>
+ <_ProjectFileVersion>10.0.20506.1</_ProjectFileVersion>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ </PropertyGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -DBUILD_TYPE=$(Configuration) -P cmake_install.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -DBUILD_TYPE=$(Configuration) -P cmake_install.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -DBUILD_TYPE=$(Configuration) -P cmake_install.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -DBUILD_TYPE=$(Configuration) -P cmake_install.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeFiles\INSTALL_force.rule">
+ <Message Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles/INSTALL_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles\INSTALL_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='Release|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles/INSTALL_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles\INSTALL_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles/INSTALL_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles\INSTALL_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles/INSTALL_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles\INSTALL_force</Outputs>
+ </CustomBuild>
+ </ItemGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeLists.txt">
+ <Message Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../.. -B../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../.. -B../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../.. -B../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../.. -B../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles\generate.stamp</Outputs>
+ </CustomBuild>
+ </ItemGroup>
+ <ItemGroup>
+ <ProjectReference Include="..\..\ALL_BUILD.vcxproj">
+ <Project>17AECBCF-B2AE-4524-9010-9A175A8F6BFE</Project>
+ </ProjectReference>
+ </ItemGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+ <ImportGroup Label="ExtensionTargets">
+ </ImportGroup>
+</Project>
\ No newline at end of file
diff --git a/src/LLVM/lib/Analysis/INSTALL.vcxproj.filters b/src/LLVM/lib/Analysis/INSTALL.vcxproj.filters
new file mode 100644
index 0000000..251dd1d
--- /dev/null
+++ b/src/LLVM/lib/Analysis/INSTALL.vcxproj.filters
@@ -0,0 +1,24 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeFiles\INSTALL_force.rule">
+ <Filter>CMake Rules</Filter>
+ </CustomBuild>
+ <CustomBuild Include="CMakeLists.txt" />
+ </ItemGroup>
+ <ItemGroup>
+ <Filter Include="CMake Rules">
+ <UniqueIdentifier>{71794486-B3CB-4A48-93CC-DE95557E96E1}</UniqueIdentifier>
+ </Filter>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+</Project>
diff --git a/src/LLVM/lib/Analysis/IPA/CallGraph.cpp b/src/LLVM/lib/Analysis/IPA/CallGraph.cpp
index b363528..2e79eab 100644
--- a/src/LLVM/lib/Analysis/IPA/CallGraph.cpp
+++ b/src/LLVM/lib/Analysis/IPA/CallGraph.cpp
@@ -43,7 +43,9 @@
public:
static char ID; // Class identification, replacement for typeinfo
BasicCallGraph() : ModulePass(ID), Root(0),
- ExternalCallingNode(0), CallsExternalNode(0) {}
+ ExternalCallingNode(0), CallsExternalNode(0) {
+ initializeBasicCallGraphPass(*PassRegistry::getPassRegistry());
+ }
// runOnModule - Compute the call graph for the specified module.
virtual bool runOnModule(Module &M) {
@@ -146,7 +148,7 @@
for (BasicBlock::iterator II = BB->begin(), IE = BB->end();
II != IE; ++II) {
CallSite CS(cast<Value>(II));
- if (CS && !isa<DbgInfoIntrinsic>(II)) {
+ if (CS && !isa<IntrinsicInst>(II)) {
const Function *Callee = CS.getCalledFunction();
if (Callee)
Node->addCalledFunction(CS, getOrInsertFunction(Callee));
@@ -171,9 +173,9 @@
} //End anonymous namespace
-static RegisterAnalysisGroup<CallGraph> X("Call Graph");
+INITIALIZE_ANALYSIS_GROUP(CallGraph, "Call Graph", BasicCallGraph)
INITIALIZE_AG_PASS(BasicCallGraph, CallGraph, "basiccg",
- "Basic CallGraph Construction", false, true, true);
+ "Basic CallGraph Construction", false, true, true)
char CallGraph::ID = 0;
char BasicCallGraph::ID = 0;
@@ -228,6 +230,21 @@
return F;
}
+/// spliceFunction - Replace the function represented by this node by another.
+/// This does not rescan the body of the function, so it is suitable when
+/// splicing the body of the old function to the new while also updating all
+/// callers from old to new.
+///
+void CallGraph::spliceFunction(const Function *From, const Function *To) {
+ assert(FunctionMap.count(From) && "No CallGraphNode for function!");
+ assert(!FunctionMap.count(To) &&
+ "Pointing CallGraphNode at a function that already exists");
+ FunctionMapTy::iterator I = FunctionMap.find(From);
+ I->second->F = const_cast<Function*>(To);
+ FunctionMap[To] = I->second;
+ FunctionMap.erase(I);
+}
+
// getOrInsertFunction - This method is identical to calling operator[], but
// it will insert a new CallGraphNode for the specified function if one does
// not already exist.
@@ -274,7 +291,6 @@
}
}
-
// removeAnyCallEdgeTo - This method removes any call edges from this node to
// the specified callee function. This takes more time to execute than
// removeCallEdgeTo, so it should not be used unless necessary.
diff --git a/src/LLVM/lib/Analysis/IPA/CallGraphSCCPass.cpp b/src/LLVM/lib/Analysis/IPA/CallGraphSCCPass.cpp
index b7a27cb..963da75 100644
--- a/src/LLVM/lib/Analysis/IPA/CallGraphSCCPass.cpp
+++ b/src/LLVM/lib/Analysis/IPA/CallGraphSCCPass.cpp
@@ -44,8 +44,8 @@
class CGPassManager : public ModulePass, public PMDataManager {
public:
static char ID;
- explicit CGPassManager(int Depth)
- : ModulePass(ID), PMDataManager(Depth) { }
+ explicit CGPassManager()
+ : ModulePass(ID), PMDataManager() { }
/// run - Execute all of the passes scheduled for execution. Keep track of
/// whether any of the passes modifies the module, and if so, return true.
@@ -245,8 +245,8 @@
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
- CallSite CS(cast<Value>(I));
- if (!CS || isa<DbgInfoIntrinsic>(I)) continue;
+ CallSite CS(cast<Value>(I));
+ if (!CS || isa<IntrinsicInst>(I)) continue;
// If this call site already existed in the callgraph, just verify it
// matches up to expectations and remove it from CallSites.
@@ -350,6 +350,7 @@
dbgs() << "CGSCCPASSMGR: SCC Refresh didn't change call graph.\n";
}
);
+ (void)MadeChange;
return DevirtualizedCall;
}
@@ -542,7 +543,7 @@
PMDataManager *PMD = PMS.top();
// [1] Create new Call Graph Pass Manager
- CGP = new CGPassManager(PMD->getDepth() + 1);
+ CGP = new CGPassManager();
// [2] Set up new manager's top level manager
PMTopLevelManager *TPM = PMD->getTopLevelManager();
@@ -582,7 +583,6 @@
public:
static char ID;
- PrintCallGraphPass() : CallGraphSCCPass(ID), Out(dbgs()) {}
PrintCallGraphPass(const std::string &B, raw_ostream &o)
: CallGraphSCCPass(ID), Banner(B), Out(o) {}
diff --git a/src/LLVM/lib/Analysis/IPA/FindUsedTypes.cpp b/src/LLVM/lib/Analysis/IPA/FindUsedTypes.cpp
index 8eed9d6..e9df3ca 100644
--- a/src/LLVM/lib/Analysis/IPA/FindUsedTypes.cpp
+++ b/src/LLVM/lib/Analysis/IPA/FindUsedTypes.cpp
@@ -24,15 +24,15 @@
char FindUsedTypes::ID = 0;
INITIALIZE_PASS(FindUsedTypes, "print-used-types",
- "Find Used Types", false, true);
+ "Find Used Types", false, true)
// IncorporateType - Incorporate one type and all of its subtypes into the
// collection of used types.
//
-void FindUsedTypes::IncorporateType(const Type *Ty) {
+void FindUsedTypes::IncorporateType(Type *Ty) {
// If ty doesn't already exist in the used types map, add it now, otherwise
// return.
- if (!UsedTypes.insert(Ty).second) return; // Already contain Ty.
+ if (!UsedTypes.insert(Ty)) return; // Already contain Ty.
// Make sure to add any types this type references now.
//
@@ -94,10 +94,8 @@
//
void FindUsedTypes::print(raw_ostream &OS, const Module *M) const {
OS << "Types in use by this module:\n";
- for (std::set<const Type *>::const_iterator I = UsedTypes.begin(),
+ for (SetVector<Type *>::const_iterator I = UsedTypes.begin(),
E = UsedTypes.end(); I != E; ++I) {
- OS << " ";
- WriteTypeSymbolic(OS, *I, M);
- OS << '\n';
+ OS << " " << **I << '\n';
}
}
diff --git a/src/LLVM/lib/Analysis/IPA/GlobalsModRef.cpp b/src/LLVM/lib/Analysis/IPA/GlobalsModRef.cpp
index 43a61d0..b226d66 100644
--- a/src/LLVM/lib/Analysis/IPA/GlobalsModRef.cpp
+++ b/src/LLVM/lib/Analysis/IPA/GlobalsModRef.cpp
@@ -24,6 +24,7 @@
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/CallGraph.h"
#include "llvm/Analysis/MemoryBuiltins.h"
+#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/InstIterator.h"
#include "llvm/ADT/Statistic.h"
@@ -88,7 +89,9 @@
public:
static char ID;
- GlobalsModRef() : ModulePass(ID) {}
+ GlobalsModRef() : ModulePass(ID) {
+ initializeGlobalsModRefPass(*PassRegistry::getPassRegistry());
+ }
bool runOnModule(Module &M) {
InitializeAliasAnalysis(this); // set up super class
@@ -106,10 +109,9 @@
//------------------------------------------------
// Implement the AliasAnalysis API
//
- AliasResult alias(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size);
+ AliasResult alias(const Location &LocA, const Location &LocB);
ModRefResult getModRefInfo(ImmutableCallSite CS,
- const Value *P, unsigned Size);
+ const Location &Loc);
ModRefResult getModRefInfo(ImmutableCallSite CS1,
ImmutableCallSite CS2) {
return AliasAnalysis::getModRefInfo(CS1, CS2);
@@ -119,32 +121,38 @@
/// called from the specified call site. The call site may be null in which
/// case the most generic behavior of this function should be returned.
ModRefBehavior getModRefBehavior(const Function *F) {
+ ModRefBehavior Min = UnknownModRefBehavior;
+
if (FunctionRecord *FR = getFunctionInfo(F)) {
if (FR->FunctionEffect == 0)
- return DoesNotAccessMemory;
+ Min = DoesNotAccessMemory;
else if ((FR->FunctionEffect & Mod) == 0)
- return OnlyReadsMemory;
+ Min = OnlyReadsMemory;
}
- return AliasAnalysis::getModRefBehavior(F);
+
+ return ModRefBehavior(AliasAnalysis::getModRefBehavior(F) & Min);
}
/// getModRefBehavior - Return the behavior of the specified function if
/// called from the specified call site. The call site may be null in which
/// case the most generic behavior of this function should be returned.
ModRefBehavior getModRefBehavior(ImmutableCallSite CS) {
- const Function* F = CS.getCalledFunction();
- if (!F) return AliasAnalysis::getModRefBehavior(CS);
- if (FunctionRecord *FR = getFunctionInfo(F)) {
- if (FR->FunctionEffect == 0)
- return DoesNotAccessMemory;
- else if ((FR->FunctionEffect & Mod) == 0)
- return OnlyReadsMemory;
- }
- return AliasAnalysis::getModRefBehavior(CS);
+ ModRefBehavior Min = UnknownModRefBehavior;
+
+ if (const Function* F = CS.getCalledFunction())
+ if (FunctionRecord *FR = getFunctionInfo(F)) {
+ if (FR->FunctionEffect == 0)
+ Min = DoesNotAccessMemory;
+ else if ((FR->FunctionEffect & Mod) == 0)
+ Min = OnlyReadsMemory;
+ }
+
+ return ModRefBehavior(AliasAnalysis::getModRefBehavior(CS) & Min);
}
virtual void deleteValue(Value *V);
virtual void copyValue(Value *From, Value *To);
+ virtual void addEscapingUse(Use &U);
/// getAdjustedAnalysisPointer - This method is used when a pass implements
/// an analysis interface through multiple inheritance. If needed, it
@@ -177,9 +185,13 @@
}
char GlobalsModRef::ID = 0;
-static RegisterPass<GlobalsModRef>
-X("globalsmodref-aa", "Simple mod/ref analysis for globals", false, true);
-static RegisterAnalysisGroup<AliasAnalysis> Y(X);
+INITIALIZE_AG_PASS_BEGIN(GlobalsModRef, AliasAnalysis,
+ "globalsmodref-aa", "Simple mod/ref analysis for globals",
+ false, true, false)
+INITIALIZE_AG_DEPENDENCY(CallGraph)
+INITIALIZE_AG_PASS_END(GlobalsModRef, AliasAnalysis,
+ "globalsmodref-aa", "Simple mod/ref analysis for globals",
+ false, true, false)
Pass *llvm::createGlobalsModRefPass() { return new GlobalsModRef(); }
@@ -314,7 +326,7 @@
continue;
// Check the value being stored.
- Value *Ptr = SI->getOperand(0)->getUnderlyingObject();
+ Value *Ptr = GetUnderlyingObject(SI->getOperand(0));
if (isMalloc(Ptr)) {
// Okay, easy case.
@@ -476,11 +488,11 @@
/// other is some random pointer, we know there cannot be an alias, because the
/// address of the global isn't taken.
AliasAnalysis::AliasResult
-GlobalsModRef::alias(const Value *V1, unsigned V1Size,
- const Value *V2, unsigned V2Size) {
+GlobalsModRef::alias(const Location &LocA,
+ const Location &LocB) {
// Get the base object these pointers point to.
- const Value *UV1 = V1->getUnderlyingObject();
- const Value *UV2 = V2->getUnderlyingObject();
+ const Value *UV1 = GetUnderlyingObject(LocA.Ptr);
+ const Value *UV2 = GetUnderlyingObject(LocB.Ptr);
// If either of the underlying values is a global, they may be non-addr-taken
// globals, which we can answer queries about.
@@ -528,17 +540,18 @@
if ((GV1 || GV2) && GV1 != GV2)
return NoAlias;
- return AliasAnalysis::alias(V1, V1Size, V2, V2Size);
+ return AliasAnalysis::alias(LocA, LocB);
}
AliasAnalysis::ModRefResult
GlobalsModRef::getModRefInfo(ImmutableCallSite CS,
- const Value *P, unsigned Size) {
+ const Location &Loc) {
unsigned Known = ModRef;
// If we are asking for mod/ref info of a direct call with a pointer to a
// global we are tracking, return information if we have it.
- if (const GlobalValue *GV = dyn_cast<GlobalValue>(P->getUnderlyingObject()))
+ if (const GlobalValue *GV =
+ dyn_cast<GlobalValue>(GetUnderlyingObject(Loc.Ptr)))
if (GV->hasLocalLinkage())
if (const Function *F = CS.getCalledFunction())
if (NonAddressTakenGlobals.count(GV))
@@ -547,7 +560,7 @@
if (Known == NoModRef)
return NoModRef; // No need to query other mod/ref analyses
- return ModRefResult(Known & AliasAnalysis::getModRefInfo(CS, P, Size));
+ return ModRefResult(Known & AliasAnalysis::getModRefInfo(CS, Loc));
}
@@ -584,3 +597,13 @@
void GlobalsModRef::copyValue(Value *From, Value *To) {
AliasAnalysis::copyValue(From, To);
}
+
+void GlobalsModRef::addEscapingUse(Use &U) {
+ // For the purposes of this analysis, it is conservatively correct to treat
+ // a newly escaping value equivalently to a deleted one. We could perhaps
+ // be more precise by processing the new use and attempting to update our
+ // saved analysis results to accommodate it.
+ deleteValue(U);
+
+ AliasAnalysis::addEscapingUse(U);
+}
diff --git a/src/LLVM/lib/Analysis/IPA/INSTALL.vcxproj b/src/LLVM/lib/Analysis/IPA/INSTALL.vcxproj
new file mode 100644
index 0000000..c52c7b0
--- /dev/null
+++ b/src/LLVM/lib/Analysis/IPA/INSTALL.vcxproj
@@ -0,0 +1,261 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup Label="ProjectConfigurations">
+ <ProjectConfiguration Include="Debug|Win32">
+ <Configuration>Debug</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Release|Win32">
+ <Configuration>Release</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="MinSizeRel|Win32">
+ <Configuration>MinSizeRel</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="RelWithDebInfo|Win32">
+ <Configuration>RelWithDebInfo</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ </ItemGroup>
+ <PropertyGroup Label="Globals">
+ <ProjectGUID>{048BB775-7681-4EE1-AACF-5A067ACEEEA5}</ProjectGUID>
+ <Keyword>Win32Proj</Keyword>
+ <Platform>Win32</Platform>
+ <ProjectName>INSTALL</ProjectName>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+ <ImportGroup Label="ExtensionSettings">
+ </ImportGroup>
+ <ImportGroup Label="PropertySheets">
+ <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" /> </ImportGroup>
+ <PropertyGroup Label="UserMacros" />
+ <PropertyGroup>
+ <_ProjectFileVersion>10.0.20506.1</_ProjectFileVersion>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ </PropertyGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -DBUILD_TYPE=$(Configuration) -P cmake_install.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -DBUILD_TYPE=$(Configuration) -P cmake_install.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -DBUILD_TYPE=$(Configuration) -P cmake_install.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -DBUILD_TYPE=$(Configuration) -P cmake_install.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeFiles\INSTALL_force.rule">
+ <Message Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles/INSTALL_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles\INSTALL_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='Release|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles/INSTALL_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles\INSTALL_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles/INSTALL_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles\INSTALL_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles/INSTALL_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles\INSTALL_force</Outputs>
+ </CustomBuild>
+ </ItemGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeLists.txt">
+ <Message Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../../.. -B../../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../../.. -B../../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../../.. -B../../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../../.. -B../../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles\generate.stamp</Outputs>
+ </CustomBuild>
+ </ItemGroup>
+ <ItemGroup>
+ <ProjectReference Include="..\..\..\ALL_BUILD.vcxproj">
+ <Project>17AECBCF-B2AE-4524-9010-9A175A8F6BFE</Project>
+ </ProjectReference>
+ </ItemGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+ <ImportGroup Label="ExtensionTargets">
+ </ImportGroup>
+</Project>
\ No newline at end of file
diff --git a/src/LLVM/lib/Analysis/IPA/INSTALL.vcxproj.filters b/src/LLVM/lib/Analysis/IPA/INSTALL.vcxproj.filters
new file mode 100644
index 0000000..251dd1d
--- /dev/null
+++ b/src/LLVM/lib/Analysis/IPA/INSTALL.vcxproj.filters
@@ -0,0 +1,24 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeFiles\INSTALL_force.rule">
+ <Filter>CMake Rules</Filter>
+ </CustomBuild>
+ <CustomBuild Include="CMakeLists.txt" />
+ </ItemGroup>
+ <ItemGroup>
+ <Filter Include="CMake Rules">
+ <UniqueIdentifier>{71794486-B3CB-4A48-93CC-DE95557E96E1}</UniqueIdentifier>
+ </Filter>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+</Project>
diff --git a/src/LLVM/lib/Analysis/IPA/IPA.cpp b/src/LLVM/lib/Analysis/IPA/IPA.cpp
new file mode 100644
index 0000000..0ba2e04
--- /dev/null
+++ b/src/LLVM/lib/Analysis/IPA/IPA.cpp
@@ -0,0 +1,29 @@
+//===-- IPA.cpp -----------------------------------------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the common initialization routines for the IPA library.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/InitializePasses.h"
+#include "llvm-c/Initialization.h"
+
+using namespace llvm;
+
+/// initializeIPA - Initialize all passes linked into the IPA library.
+void llvm::initializeIPA(PassRegistry &Registry) {
+ initializeBasicCallGraphPass(Registry);
+ initializeCallGraphAnalysisGroup(Registry);
+ initializeFindUsedTypesPass(Registry);
+ initializeGlobalsModRefPass(Registry);
+}
+
+void LLVMInitializeIPA(LLVMPassRegistryRef R) {
+ initializeIPA(*unwrap(R));
+}
diff --git a/src/LLVM/lib/Analysis/IPA/LLVMipa.vcxproj b/src/LLVM/lib/Analysis/IPA/LLVMipa.vcxproj
new file mode 100644
index 0000000..a77a751
--- /dev/null
+++ b/src/LLVM/lib/Analysis/IPA/LLVMipa.vcxproj
@@ -0,0 +1,360 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup Label="ProjectConfigurations">
+ <ProjectConfiguration Include="Debug|Win32">
+ <Configuration>Debug</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Debug|x64">
+ <Configuration>Debug</Configuration>
+ <Platform>x64</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Profile|Win32">
+ <Configuration>Profile</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Profile|x64">
+ <Configuration>Profile</Configuration>
+ <Platform>x64</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Release|Win32">
+ <Configuration>Release</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Release|x64">
+ <Configuration>Release</Configuration>
+ <Platform>x64</Platform>
+ </ProjectConfiguration>
+ </ItemGroup>
+ <PropertyGroup Label="Globals">
+ <ProjectGUID>{32FFFFE6-D5B8-43F6-8E23-362A0D0252A1}</ProjectGUID>
+ <Keyword>Win32Proj</Keyword>
+ <Platform>Win32</Platform>
+ <ProjectName>LLVMipa</ProjectName>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>NotSet</CharacterSet>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>NotSet</CharacterSet>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>NotSet</CharacterSet>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Profile|x64'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>NotSet</CharacterSet>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+ <ImportGroup Label="ExtensionSettings">
+ </ImportGroup>
+ <ImportGroup Label="PropertySheets">
+ <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+ </ImportGroup>
+ <PropertyGroup Label="UserMacros" />
+ <PropertyGroup>
+ <_ProjectFileVersion>10.0.20506.1</_ProjectFileVersion>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(Platform)\$(Configuration)\</OutDir>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">$(Platform)\$(Configuration)\</OutDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(Platform)\$(Configuration)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">$(Platform)\$(Configuration)\</IntDir>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(ProjectName)</TargetName>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">$(ProjectName)</TargetName>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">.lib</TargetExt>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">.lib</TargetExt>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(Platform)\$(Configuration)\</OutDir>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">$(Platform)\$(Configuration)\</OutDir>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Release|x64'">$(Platform)\$(Configuration)\</OutDir>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">$(Platform)\$(Configuration)\</OutDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(Platform)\$(Configuration)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">$(Platform)\$(Configuration)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Release|x64'">$(Platform)\$(Configuration)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">$(Platform)\$(Configuration)\</IntDir>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(ProjectName)</TargetName>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">$(ProjectName)</TargetName>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Release|x64'">$(ProjectName)</TargetName>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">$(ProjectName)</TargetName>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">.lib</TargetExt>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">.lib</TargetExt>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Release|x64'">.lib</TargetExt>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">.lib</TargetExt>
+ </PropertyGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <BasicRuntimeChecks>EnableFastChecks</BasicRuntimeChecks>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>
+ </ExceptionHandling>
+ <InlineFunctionExpansion>Disabled</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Disabled</Optimization>
+ <RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;_DEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Debug";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Debug</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\..\Debug/LLVMipa.pdb</ProgramDataBaseFileName>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;_DEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Debug\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <BasicRuntimeChecks>EnableFastChecks</BasicRuntimeChecks>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>
+ </ExceptionHandling>
+ <InlineFunctionExpansion>Disabled</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Disabled</Optimization>
+ <RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;_DEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Debug";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Debug</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\..\Debug/LLVMipa.pdb</ProgramDataBaseFileName>
+ <OmitFramePointers>false</OmitFramePointers>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;_DEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Debug\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>false</ExceptionHandling>
+ <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Full</Optimization>
+ <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <DebugInformationFormat>
+ </DebugInformationFormat>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Release";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Release</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\..\Release/LLVMipa.pdb</ProgramDataBaseFileName>
+ <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
+ <OmitFramePointers>true</OmitFramePointers>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ <StringPooling>true</StringPooling>
+ <BufferSecurityCheck>false</BufferSecurityCheck>
+ <FunctionLevelLinking>false</FunctionLevelLinking>
+ <FloatingPointExceptions>false</FloatingPointExceptions>
+ <CreateHotpatchableImage>false</CreateHotpatchableImage>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Release\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>false</ExceptionHandling>
+ <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Full</Optimization>
+ <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Release";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Release</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\..\Release/LLVMipa.pdb</ProgramDataBaseFileName>
+ <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
+ <OmitFramePointers>false</OmitFramePointers>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ <StringPooling>true</StringPooling>
+ <BufferSecurityCheck>false</BufferSecurityCheck>
+ <FunctionLevelLinking>false</FunctionLevelLinking>
+ <FloatingPointExceptions>false</FloatingPointExceptions>
+ <CreateHotpatchableImage>false</CreateHotpatchableImage>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Release\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>false</ExceptionHandling>
+ <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Full</Optimization>
+ <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <DebugInformationFormat>
+ </DebugInformationFormat>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Release";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Release</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\..\Release/LLVMipa.pdb</ProgramDataBaseFileName>
+ <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
+ <OmitFramePointers>true</OmitFramePointers>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ <StringPooling>true</StringPooling>
+ <BufferSecurityCheck>false</BufferSecurityCheck>
+ <FunctionLevelLinking>false</FunctionLevelLinking>
+ <FloatingPointExceptions>false</FloatingPointExceptions>
+ <CreateHotpatchableImage>false</CreateHotpatchableImage>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Release\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>false</ExceptionHandling>
+ <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Full</Optimization>
+ <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <DebugInformationFormat>
+ </DebugInformationFormat>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Release";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Release</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\..\Release/LLVMipa.pdb</ProgramDataBaseFileName>
+ <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
+ <OmitFramePointers>false</OmitFramePointers>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ <StringPooling>true</StringPooling>
+ <BufferSecurityCheck>false</BufferSecurityCheck>
+ <FunctionLevelLinking>false</FunctionLevelLinking>
+ <FloatingPointExceptions>false</FloatingPointExceptions>
+ <CreateHotpatchableImage>false</CreateHotpatchableImage>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Release\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemGroup>
+ <ClCompile Include="CallGraph.cpp" />
+ <ClCompile Include="CallGraphSCCPass.cpp" />
+ <ClCompile Include="FindUsedTypes.cpp" />
+ <ClCompile Include="GlobalsModRef.cpp" />
+ <ClCompile Include="IPA.cpp" />
+ </ItemGroup>
+ <ItemGroup>
+ <ProjectReference Include="..\LLVMAnalysis.vcxproj">
+ <Project>97EDF19C-6360-4770-9255-EBA2F1A13E9B</Project>
+ </ProjectReference>
+ <ProjectReference Include="..\..\VMCore/LLVMCore.vcxproj">
+ <Project>00F3295C-F7A0-43D3-BD0B-1BC0515B30E1</Project>
+ </ProjectReference>
+ <ProjectReference Include="..\..\Support/LLVMSupport.vcxproj">
+ <Project>C688DD59-C6CB-4B33-B56F-A7D6F3761524</Project>
+ </ProjectReference>
+ <ProjectReference Include="..\..\..\include/llvm/intrinsics_gen.vcxproj">
+ <Project>E9B87B46-1EB0-4D95-9049-41B148FBADCD</Project>
+ </ProjectReference>
+ </ItemGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+ <ImportGroup Label="ExtensionTargets">
+ </ImportGroup>
+</Project>
\ No newline at end of file
diff --git a/src/LLVM/lib/Analysis/IPA/LLVMipa.vcxproj.filters b/src/LLVM/lib/Analysis/IPA/LLVMipa.vcxproj.filters
new file mode 100644
index 0000000..3b36969
--- /dev/null
+++ b/src/LLVM/lib/Analysis/IPA/LLVMipa.vcxproj.filters
@@ -0,0 +1,33 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup>
+ <ClCompile Include="CallGraph.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="CallGraphSCCPass.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="FindUsedTypes.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="GlobalsModRef.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="IPA.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ <Filter Include="Source Files">
+ <UniqueIdentifier>{1733179C-6FE4-462E-9EA5-4A29A1ACFE25}</UniqueIdentifier>
+ </Filter>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+</Project>
\ No newline at end of file
diff --git a/src/LLVM/lib/Analysis/IPA/PACKAGE.vcxproj b/src/LLVM/lib/Analysis/IPA/PACKAGE.vcxproj
new file mode 100644
index 0000000..9433fa3
--- /dev/null
+++ b/src/LLVM/lib/Analysis/IPA/PACKAGE.vcxproj
@@ -0,0 +1,277 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup Label="ProjectConfigurations">
+ <ProjectConfiguration Include="Debug|Win32">
+ <Configuration>Debug</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Release|Win32">
+ <Configuration>Release</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="MinSizeRel|Win32">
+ <Configuration>MinSizeRel</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="RelWithDebInfo|Win32">
+ <Configuration>RelWithDebInfo</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ </ItemGroup>
+ <PropertyGroup Label="Globals">
+ <ProjectGUID>{1B050569-3318-48D9-8BB0-4DE9EF58B202}</ProjectGUID>
+ <Keyword>Win32Proj</Keyword>
+ <Platform>Win32</Platform>
+ <ProjectName>PACKAGE</ProjectName>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+ <ImportGroup Label="ExtensionSettings">
+ </ImportGroup>
+ <ImportGroup Label="PropertySheets">
+ <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" /> </ImportGroup>
+ <PropertyGroup Label="UserMacros" />
+ <PropertyGroup>
+ <_ProjectFileVersion>10.0.20506.1</_ProjectFileVersion>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ </PropertyGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+cd ..\..\..\..\LLVM
+if %errorlevel% neq 0 goto :cmEnd
+D:
+if %errorlevel% neq 0 goto :cmEnd
+"C:\Program Files (x86)\CMake 2.8\bin\cpack.exe" -C $(Configuration) --config ./CPackConfig.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+cd ..\..\..\..\LLVM
+if %errorlevel% neq 0 goto :cmEnd
+D:
+if %errorlevel% neq 0 goto :cmEnd
+"C:\Program Files (x86)\CMake 2.8\bin\cpack.exe" -C $(Configuration) --config ./CPackConfig.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+cd ..\..\..\..\LLVM
+if %errorlevel% neq 0 goto :cmEnd
+D:
+if %errorlevel% neq 0 goto :cmEnd
+"C:\Program Files (x86)\CMake 2.8\bin\cpack.exe" -C $(Configuration) --config ./CPackConfig.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\IPA;..\..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+cd ..\..\..\..\LLVM
+if %errorlevel% neq 0 goto :cmEnd
+D:
+if %errorlevel% neq 0 goto :cmEnd
+"C:\Program Files (x86)\CMake 2.8\bin\cpack.exe" -C $(Configuration) --config ./CPackConfig.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeFiles\PACKAGE_force.rule">
+ <Message Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles/PACKAGE_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles\PACKAGE_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='Release|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles/PACKAGE_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles\PACKAGE_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles/PACKAGE_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles\PACKAGE_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles/PACKAGE_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles\PACKAGE_force</Outputs>
+ </CustomBuild>
+ </ItemGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeLists.txt">
+ <Message Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../../.. -B../../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../../.. -B../../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../../.. -B../../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../../.. -B../../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles\generate.stamp</Outputs>
+ </CustomBuild>
+ </ItemGroup>
+ <ItemGroup>
+ <ProjectReference Include="..\..\..\ALL_BUILD.vcxproj">
+ <Project>17AECBCF-B2AE-4524-9010-9A175A8F6BFE</Project>
+ </ProjectReference>
+ </ItemGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+ <ImportGroup Label="ExtensionTargets">
+ </ImportGroup>
+</Project>
\ No newline at end of file
diff --git a/src/LLVM/lib/Analysis/IPA/PACKAGE.vcxproj.filters b/src/LLVM/lib/Analysis/IPA/PACKAGE.vcxproj.filters
new file mode 100644
index 0000000..a570359
--- /dev/null
+++ b/src/LLVM/lib/Analysis/IPA/PACKAGE.vcxproj.filters
@@ -0,0 +1,24 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeFiles\PACKAGE_force.rule">
+ <Filter>CMake Rules</Filter>
+ </CustomBuild>
+ <CustomBuild Include="CMakeLists.txt" />
+ </ItemGroup>
+ <ItemGroup>
+ <Filter Include="CMake Rules">
+ <UniqueIdentifier>{71794486-B3CB-4A48-93CC-DE95557E96E1}</UniqueIdentifier>
+ </Filter>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+</Project>
diff --git a/src/LLVM/lib/Analysis/IVUsers.cpp b/src/LLVM/lib/Analysis/IVUsers.cpp
index bd43026..d0ca892 100644
--- a/src/LLVM/lib/Analysis/IVUsers.cpp
+++ b/src/LLVM/lib/Analysis/IVUsers.cpp
@@ -21,7 +21,8 @@
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/ScalarEvolutionExpressions.h"
-#include "llvm/Assembly/AsmAnnotationWriter.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Assembly/Writer.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
@@ -29,7 +30,13 @@
using namespace llvm;
char IVUsers::ID = 0;
-INITIALIZE_PASS(IVUsers, "iv-users", "Induction Variable Users", false, true);
+INITIALIZE_PASS_BEGIN(IVUsers, "iv-users",
+ "Induction Variable Users", false, true)
+INITIALIZE_PASS_DEPENDENCY(LoopInfo)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
+INITIALIZE_PASS_END(IVUsers, "iv-users",
+ "Induction Variable Users", false, true)
Pass *llvm::createIVUsersPass() {
return new IVUsers();
@@ -38,27 +45,34 @@
/// isInteresting - Test whether the given expression is "interesting" when
/// used by the given expression, within the context of analyzing the
/// given loop.
-static bool isInteresting(const SCEV *S, const Instruction *I, const Loop *L) {
- // Anything loop-invariant is interesting.
- if (!isa<SCEVUnknown>(S) && S->isLoopInvariant(L))
- return true;
-
+static bool isInteresting(const SCEV *S, const Instruction *I, const Loop *L,
+ ScalarEvolution *SE, LoopInfo *LI) {
// An addrec is interesting if it's affine or if it has an interesting start.
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S)) {
- // Keep things simple. Don't touch loop-variant strides.
+ // Keep things simple. Don't touch loop-variant strides unless they're
+ // only used outside the loop and we can simplify them.
if (AR->getLoop() == L)
- return AR->isAffine() || !L->contains(I);
- // Otherwise recurse to see if the start value is interesting.
- return isInteresting(AR->getStart(), I, L);
+ return AR->isAffine() ||
+ (!L->contains(I) &&
+ SE->getSCEVAtScope(AR, LI->getLoopFor(I->getParent())) != AR);
+ // Otherwise recurse to see if the start value is interesting, and that
+ // the step value is not interesting, since we don't yet know how to
+ // do effective SCEV expansions for addrecs with interesting steps.
+ return isInteresting(AR->getStart(), I, L, SE, LI) &&
+ !isInteresting(AR->getStepRecurrence(*SE), I, L, SE, LI);
}
- // An add is interesting if any of its operands is.
+ // An add is interesting if exactly one of its operands is interesting.
if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
+ bool AnyInterestingYet = false;
for (SCEVAddExpr::op_iterator OI = Add->op_begin(), OE = Add->op_end();
OI != OE; ++OI)
- if (isInteresting(*OI, I, L))
- return true;
- return false;
+ if (isInteresting(*OI, I, L, SE, LI)) {
+ if (AnyInterestingYet)
+ return false;
+ AnyInterestingYet = true;
+ }
+ return AnyInterestingYet;
}
// Nothing else is interesting here.
@@ -73,7 +87,10 @@
return false; // Void and FP expressions cannot be reduced.
// LSR is not APInt clean, do not touch integers bigger than 64-bits.
- if (SE->getTypeSizeInBits(I->getType()) > 64)
+ // Also avoid creating IVs of non-native types. For example, we don't want a
+ // 64-bit IV in 32-bit code just because the loop has one 64-bit cast.
+ uint64_t Width = SE->getTypeSizeInBits(I->getType());
+ if (Width > 64 || (TD && !TD->isLegalInteger(Width)))
return false;
if (!Processed.insert(I))
@@ -84,7 +101,7 @@
// If we've come to an uninteresting expression, stop the traversal and
// call this a user.
- if (!isInteresting(ISE, I, L))
+ if (!isInteresting(ISE, I, L, SE, LI))
return false;
SmallPtrSet<Instruction *, 4> UniqueUsers;
@@ -112,8 +129,7 @@
<< " OF SCEV: " << *ISE << '\n');
AddUserToIVUsers = true;
}
- } else if (Processed.count(User) ||
- !AddUsersIfInteresting(User)) {
+ } else if (Processed.count(User) || !AddUsersIfInteresting(User)) {
DEBUG(dbgs() << "FOUND USER: " << *User << '\n'
<< " OF SCEV: " << *ISE << '\n');
AddUserToIVUsers = true;
@@ -123,12 +139,15 @@
// Okay, we found a user that we cannot reduce.
IVUses.push_back(new IVStrideUse(this, User, I));
IVStrideUse &NewUse = IVUses.back();
- // Transform the expression into a normalized form.
+ // Autodetect the post-inc loop set, populating NewUse.PostIncLoops.
+ // The regular return value here is discarded; instead of recording
+ // it, we just recompute it when we need it.
ISE = TransformForPostIncUse(NormalizeAutodetect,
ISE, User, I,
NewUse.PostIncLoops,
*SE, *DT);
- DEBUG(dbgs() << " NORMALIZED TO: " << *ISE << '\n');
+ DEBUG(if (SE->getSCEV(I) != ISE)
+ dbgs() << " NORMALIZED TO: " << *ISE << '\n');
}
}
return true;
@@ -140,7 +159,8 @@
}
IVUsers::IVUsers()
- : LoopPass(ID) {
+ : LoopPass(ID) {
+ initializeIVUsersPass(*PassRegistry::getPassRegistry());
}
void IVUsers::getAnalysisUsage(AnalysisUsage &AU) const {
@@ -156,6 +176,7 @@
LI = &getAnalysis<LoopInfo>();
DT = &getAnalysis<DominatorTree>();
SE = &getAnalysis<ScalarEvolution>();
+ TD = getAnalysisIfAvailable<TargetData>();
// Find all uses of induction variables in this loop, and categorize
// them by stride. Start by finding all of the PHI nodes in the header for
@@ -175,9 +196,6 @@
}
OS << ":\n";
- // Use a default AssemblyAnnotationWriter to suppress the default info
- // comments, which aren't relevant here.
- AssemblyAnnotationWriter Annotator;
for (ilist<IVStrideUse>::const_iterator UI = IVUses.begin(),
E = IVUses.end(); UI != E; ++UI) {
OS << " ";
@@ -191,7 +209,7 @@
OS << ")";
}
OS << " in ";
- UI->getUser()->print(OS, &Annotator);
+ UI->getUser()->print(OS);
OS << '\n';
}
}
diff --git a/src/LLVM/lib/Analysis/InlineCost.cpp b/src/LLVM/lib/Analysis/InlineCost.cpp
new file mode 100644
index 0000000..e12e322
--- /dev/null
+++ b/src/LLVM/lib/Analysis/InlineCost.cpp
@@ -0,0 +1,655 @@
+//===- InlineCost.cpp - Cost analysis for inliner -------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements inline cost analysis.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/InlineCost.h"
+#include "llvm/Support/CallSite.h"
+#include "llvm/CallingConv.h"
+#include "llvm/IntrinsicInst.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/ADT/SmallPtrSet.h"
+
+using namespace llvm;
+
+/// callIsSmall - If a call is likely to lower to a single target instruction,
+/// or is otherwise deemed small return true.
+/// TODO: Perhaps calls like memcpy, strcpy, etc?
+bool llvm::callIsSmall(const Function *F) {
+ if (!F) return false;
+
+ if (F->hasLocalLinkage()) return false;
+
+ if (!F->hasName()) return false;
+
+ StringRef Name = F->getName();
+
+ // These will all likely lower to a single selection DAG node.
+ if (Name == "copysign" || Name == "copysignf" || Name == "copysignl" ||
+ Name == "fabs" || Name == "fabsf" || Name == "fabsl" ||
+ Name == "sin" || Name == "sinf" || Name == "sinl" ||
+ Name == "cos" || Name == "cosf" || Name == "cosl" ||
+ Name == "sqrt" || Name == "sqrtf" || Name == "sqrtl" )
+ return true;
+
+ // These are all likely to be optimized into something smaller.
+ if (Name == "pow" || Name == "powf" || Name == "powl" ||
+ Name == "exp2" || Name == "exp2l" || Name == "exp2f" ||
+ Name == "floor" || Name == "floorf" || Name == "ceil" ||
+ Name == "round" || Name == "ffs" || Name == "ffsl" ||
+ Name == "abs" || Name == "labs" || Name == "llabs")
+ return true;
+
+ return false;
+}
+
+/// analyzeBasicBlock - Fill in the current structure with information gleaned
+/// from the specified block.
+void CodeMetrics::analyzeBasicBlock(const BasicBlock *BB,
+ const TargetData *TD) {
+ ++NumBlocks;
+ unsigned NumInstsBeforeThisBB = NumInsts;
+ for (BasicBlock::const_iterator II = BB->begin(), E = BB->end();
+ II != E; ++II) {
+ if (isa<PHINode>(II)) continue; // PHI nodes don't count.
+
+ // Special handling for calls.
+ if (isa<CallInst>(II) || isa<InvokeInst>(II)) {
+ if (isa<DbgInfoIntrinsic>(II))
+ continue; // Debug intrinsics don't count as size.
+
+ ImmutableCallSite CS(cast<Instruction>(II));
+
+ if (const Function *F = CS.getCalledFunction()) {
+ // If a function is both internal and has a single use, then it is
+ // extremely likely to get inlined in the future (it was probably
+ // exposed by an interleaved devirtualization pass).
+ if (F->hasInternalLinkage() && F->hasOneUse())
+ ++NumInlineCandidates;
+
+ // If this call is to function itself, then the function is recursive.
+ // Inlining it into other functions is a bad idea, because this is
+ // basically just a form of loop peeling, and our metrics aren't useful
+ // for that case.
+ if (F == BB->getParent())
+ isRecursive = true;
+ }
+
+ if (!isa<IntrinsicInst>(II) && !callIsSmall(CS.getCalledFunction())) {
+ // Each argument to a call takes on average one instruction to set up.
+ NumInsts += CS.arg_size();
+
+ // We don't want inline asm to count as a call - that would prevent loop
+ // unrolling. The argument setup cost is still real, though.
+ if (!isa<InlineAsm>(CS.getCalledValue()))
+ ++NumCalls;
+ }
+ }
+
+ if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
+ if (!AI->isStaticAlloca())
+ this->usesDynamicAlloca = true;
+ }
+
+ if (isa<ExtractElementInst>(II) || II->getType()->isVectorTy())
+ ++NumVectorInsts;
+
+ if (const CastInst *CI = dyn_cast<CastInst>(II)) {
+ // Noop casts, including ptr <-> int, don't count.
+ if (CI->isLosslessCast() || isa<IntToPtrInst>(CI) ||
+ isa<PtrToIntInst>(CI))
+ continue;
+ // trunc to a native type is free (assuming the target has compare and
+ // shift-right of the same width).
+ if (isa<TruncInst>(CI) && TD &&
+ TD->isLegalInteger(TD->getTypeSizeInBits(CI->getType())))
+ continue;
+ // Result of a cmp instruction is often extended (to be used by other
+ // cmp instructions, logical or return instructions). These are usually
+ // nop on most sane targets.
+ if (isa<CmpInst>(CI->getOperand(0)))
+ continue;
+ } else if (const GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(II)){
+ // If a GEP has all constant indices, it will probably be folded with
+ // a load/store.
+ if (GEPI->hasAllConstantIndices())
+ continue;
+ }
+
+ ++NumInsts;
+ }
+
+ if (isa<ReturnInst>(BB->getTerminator()))
+ ++NumRets;
+
+ // We never want to inline functions that contain an indirectbr. This is
+ // incorrect because all the blockaddress's (in static global initializers
+ // for example) would be referring to the original function, and this indirect
+ // jump would jump from the inlined copy of the function into the original
+ // function which is extremely undefined behavior.
+ if (isa<IndirectBrInst>(BB->getTerminator()))
+ containsIndirectBr = true;
+
+ // Remember NumInsts for this BB.
+ NumBBInsts[BB] = NumInsts - NumInstsBeforeThisBB;
+}
+
+// CountCodeReductionForConstant - Figure out an approximation for how many
+// instructions will be constant folded if the specified value is constant.
+//
+unsigned CodeMetrics::CountCodeReductionForConstant(Value *V) {
+ unsigned Reduction = 0;
+ for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
+ User *U = *UI;
+ if (isa<BranchInst>(U) || isa<SwitchInst>(U)) {
+ // We will be able to eliminate all but one of the successors.
+ const TerminatorInst &TI = cast<TerminatorInst>(*U);
+ const unsigned NumSucc = TI.getNumSuccessors();
+ unsigned Instrs = 0;
+ for (unsigned I = 0; I != NumSucc; ++I)
+ Instrs += NumBBInsts[TI.getSuccessor(I)];
+ // We don't know which blocks will be eliminated, so use the average size.
+ Reduction += InlineConstants::InstrCost*Instrs*(NumSucc-1)/NumSucc;
+ } else {
+ // Figure out if this instruction will be removed due to simple constant
+ // propagation.
+ Instruction &Inst = cast<Instruction>(*U);
+
+ // We can't constant propagate instructions which have effects or
+ // read memory.
+ //
+ // FIXME: It would be nice to capture the fact that a load from a
+ // pointer-to-constant-global is actually a *really* good thing to zap.
+ // Unfortunately, we don't know the pointer that may get propagated here,
+ // so we can't make this decision.
+ if (Inst.mayReadFromMemory() || Inst.mayHaveSideEffects() ||
+ isa<AllocaInst>(Inst))
+ continue;
+
+ bool AllOperandsConstant = true;
+ for (unsigned i = 0, e = Inst.getNumOperands(); i != e; ++i)
+ if (!isa<Constant>(Inst.getOperand(i)) && Inst.getOperand(i) != V) {
+ AllOperandsConstant = false;
+ break;
+ }
+
+ if (AllOperandsConstant) {
+ // We will get to remove this instruction...
+ Reduction += InlineConstants::InstrCost;
+
+ // And any other instructions that use it which become constants
+ // themselves.
+ Reduction += CountCodeReductionForConstant(&Inst);
+ }
+ }
+ }
+ return Reduction;
+}
+
+// CountCodeReductionForAlloca - Figure out an approximation of how much smaller
+// the function will be if it is inlined into a context where an argument
+// becomes an alloca.
+//
+unsigned CodeMetrics::CountCodeReductionForAlloca(Value *V) {
+ if (!V->getType()->isPointerTy()) return 0; // Not a pointer
+ unsigned Reduction = 0;
+ for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
+ Instruction *I = cast<Instruction>(*UI);
+ if (isa<LoadInst>(I) || isa<StoreInst>(I))
+ Reduction += InlineConstants::InstrCost;
+ else if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
+ // If the GEP has variable indices, we won't be able to do much with it.
+ if (GEP->hasAllConstantIndices())
+ Reduction += CountCodeReductionForAlloca(GEP);
+ } else if (BitCastInst *BCI = dyn_cast<BitCastInst>(I)) {
+ // Track pointer through bitcasts.
+ Reduction += CountCodeReductionForAlloca(BCI);
+ } else {
+ // If there is some other strange instruction, we're not going to be able
+ // to do much if we inline this.
+ return 0;
+ }
+ }
+
+ return Reduction;
+}
+
+/// analyzeFunction - Fill in the current structure with information gleaned
+/// from the specified function.
+void CodeMetrics::analyzeFunction(Function *F, const TargetData *TD) {
+ // If this function contains a call to setjmp or _setjmp, never inline
+ // it. This is a hack because we depend on the user marking their local
+ // variables as volatile if they are live across a setjmp call, and they
+ // probably won't do this in callers.
+ if (F->callsFunctionThatReturnsTwice())
+ callsSetJmp = true;
+
+ // Look at the size of the callee.
+ for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
+ analyzeBasicBlock(&*BB, TD);
+}
+
+/// analyzeFunction - Fill in the current structure with information gleaned
+/// from the specified function.
+void InlineCostAnalyzer::FunctionInfo::analyzeFunction(Function *F,
+ const TargetData *TD) {
+ Metrics.analyzeFunction(F, TD);
+
+ // A function with exactly one return has it removed during the inlining
+ // process (see InlineFunction), so don't count it.
+ // FIXME: This knowledge should really be encoded outside of FunctionInfo.
+ if (Metrics.NumRets==1)
+ --Metrics.NumInsts;
+
+ // Check out all of the arguments to the function, figuring out how much
+ // code can be eliminated if one of the arguments is a constant.
+ ArgumentWeights.reserve(F->arg_size());
+ for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I)
+ ArgumentWeights.push_back(ArgInfo(Metrics.CountCodeReductionForConstant(I),
+ Metrics.CountCodeReductionForAlloca(I)));
+}
+
+/// NeverInline - returns true if the function should never be inlined into
+/// any caller
+bool InlineCostAnalyzer::FunctionInfo::NeverInline() {
+ return (Metrics.callsSetJmp || Metrics.isRecursive ||
+ Metrics.containsIndirectBr);
+}
+// getSpecializationBonus - The heuristic used to determine the per-call
+// performance boost for using a specialization of Callee with argument
+// specializedArgNo replaced by a constant.
+int InlineCostAnalyzer::getSpecializationBonus(Function *Callee,
+ SmallVectorImpl<unsigned> &SpecializedArgNos)
+{
+ if (Callee->mayBeOverridden())
+ return 0;
+
+ int Bonus = 0;
+ // If this function uses the coldcc calling convention, prefer not to
+ // specialize it.
+ if (Callee->getCallingConv() == CallingConv::Cold)
+ Bonus -= InlineConstants::ColdccPenalty;
+
+ // Get information about the callee.
+ FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
+
+ // If we haven't calculated this information yet, do so now.
+ if (CalleeFI->Metrics.NumBlocks == 0)
+ CalleeFI->analyzeFunction(Callee, TD);
+
+ unsigned ArgNo = 0;
+ unsigned i = 0;
+ for (Function::arg_iterator I = Callee->arg_begin(), E = Callee->arg_end();
+ I != E; ++I, ++ArgNo)
+ if (ArgNo == SpecializedArgNos[i]) {
+ ++i;
+ Bonus += CountBonusForConstant(I);
+ }
+
+ // Calls usually take a long time, so they make the specialization gain
+ // smaller.
+ Bonus -= CalleeFI->Metrics.NumCalls * InlineConstants::CallPenalty;
+
+ return Bonus;
+}
+
+// ConstantFunctionBonus - Figure out how much of a bonus we can get for
+// possibly devirtualizing a function. We'll subtract the size of the function
+// we may wish to inline from the indirect call bonus providing a limit on
+// growth. Leave an upper limit of 0 for the bonus - we don't want to penalize
+// inlining because we decide we don't want to give a bonus for
+// devirtualizing.
+int InlineCostAnalyzer::ConstantFunctionBonus(CallSite CS, Constant *C) {
+
+ // This could just be NULL.
+ if (!C) return 0;
+
+ Function *F = dyn_cast<Function>(C);
+ if (!F) return 0;
+
+ int Bonus = InlineConstants::IndirectCallBonus + getInlineSize(CS, F);
+ return (Bonus > 0) ? 0 : Bonus;
+}
+
+// CountBonusForConstant - Figure out an approximation for how much per-call
+// performance boost we can expect if the specified value is constant.
+int InlineCostAnalyzer::CountBonusForConstant(Value *V, Constant *C) {
+ unsigned Bonus = 0;
+ for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
+ User *U = *UI;
+ if (CallInst *CI = dyn_cast<CallInst>(U)) {
+ // Turning an indirect call into a direct call is a BIG win
+ if (CI->getCalledValue() == V)
+ Bonus += ConstantFunctionBonus(CallSite(CI), C);
+ } else if (InvokeInst *II = dyn_cast<InvokeInst>(U)) {
+ // Turning an indirect call into a direct call is a BIG win
+ if (II->getCalledValue() == V)
+ Bonus += ConstantFunctionBonus(CallSite(II), C);
+ }
+ // FIXME: Eliminating conditional branches and switches should
+ // also yield a per-call performance boost.
+ else {
+ // Figure out the bonuses that wll accrue due to simple constant
+ // propagation.
+ Instruction &Inst = cast<Instruction>(*U);
+
+ // We can't constant propagate instructions which have effects or
+ // read memory.
+ //
+ // FIXME: It would be nice to capture the fact that a load from a
+ // pointer-to-constant-global is actually a *really* good thing to zap.
+ // Unfortunately, we don't know the pointer that may get propagated here,
+ // so we can't make this decision.
+ if (Inst.mayReadFromMemory() || Inst.mayHaveSideEffects() ||
+ isa<AllocaInst>(Inst))
+ continue;
+
+ bool AllOperandsConstant = true;
+ for (unsigned i = 0, e = Inst.getNumOperands(); i != e; ++i)
+ if (!isa<Constant>(Inst.getOperand(i)) && Inst.getOperand(i) != V) {
+ AllOperandsConstant = false;
+ break;
+ }
+
+ if (AllOperandsConstant)
+ Bonus += CountBonusForConstant(&Inst);
+ }
+ }
+
+ return Bonus;
+}
+
+int InlineCostAnalyzer::getInlineSize(CallSite CS, Function *Callee) {
+ // Get information about the callee.
+ FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
+
+ // If we haven't calculated this information yet, do so now.
+ if (CalleeFI->Metrics.NumBlocks == 0)
+ CalleeFI->analyzeFunction(Callee, TD);
+
+ // InlineCost - This value measures how good of an inline candidate this call
+ // site is to inline. A lower inline cost make is more likely for the call to
+ // be inlined. This value may go negative.
+ //
+ int InlineCost = 0;
+
+ // Compute any size reductions we can expect due to arguments being passed into
+ // the function.
+ //
+ unsigned ArgNo = 0;
+ CallSite::arg_iterator I = CS.arg_begin();
+ for (Function::arg_iterator FI = Callee->arg_begin(), FE = Callee->arg_end();
+ FI != FE; ++I, ++FI, ++ArgNo) {
+
+ // If an alloca is passed in, inlining this function is likely to allow
+ // significant future optimization possibilities (like scalar promotion, and
+ // scalarization), so encourage the inlining of the function.
+ //
+ if (isa<AllocaInst>(I))
+ InlineCost -= CalleeFI->ArgumentWeights[ArgNo].AllocaWeight;
+
+ // If this is a constant being passed into the function, use the argument
+ // weights calculated for the callee to determine how much will be folded
+ // away with this information.
+ else if (isa<Constant>(I))
+ InlineCost -= CalleeFI->ArgumentWeights[ArgNo].ConstantWeight;
+ }
+
+ // Each argument passed in has a cost at both the caller and the callee
+ // sides. Measurements show that each argument costs about the same as an
+ // instruction.
+ InlineCost -= (CS.arg_size() * InlineConstants::InstrCost);
+
+ // Now that we have considered all of the factors that make the call site more
+ // likely to be inlined, look at factors that make us not want to inline it.
+
+ // Calls usually take a long time, so they make the inlining gain smaller.
+ InlineCost += CalleeFI->Metrics.NumCalls * InlineConstants::CallPenalty;
+
+ // Look at the size of the callee. Each instruction counts as 5.
+ InlineCost += CalleeFI->Metrics.NumInsts*InlineConstants::InstrCost;
+
+ return InlineCost;
+}
+
+int InlineCostAnalyzer::getInlineBonuses(CallSite CS, Function *Callee) {
+ // Get information about the callee.
+ FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
+
+ // If we haven't calculated this information yet, do so now.
+ if (CalleeFI->Metrics.NumBlocks == 0)
+ CalleeFI->analyzeFunction(Callee, TD);
+
+ bool isDirectCall = CS.getCalledFunction() == Callee;
+ Instruction *TheCall = CS.getInstruction();
+ int Bonus = 0;
+
+ // If there is only one call of the function, and it has internal linkage,
+ // make it almost guaranteed to be inlined.
+ //
+ if (Callee->hasLocalLinkage() && Callee->hasOneUse() && isDirectCall)
+ Bonus += InlineConstants::LastCallToStaticBonus;
+
+ // If the instruction after the call, or if the normal destination of the
+ // invoke is an unreachable instruction, the function is noreturn. As such,
+ // there is little point in inlining this.
+ if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
+ if (isa<UnreachableInst>(II->getNormalDest()->begin()))
+ Bonus += InlineConstants::NoreturnPenalty;
+ } else if (isa<UnreachableInst>(++BasicBlock::iterator(TheCall)))
+ Bonus += InlineConstants::NoreturnPenalty;
+
+ // If this function uses the coldcc calling convention, prefer not to inline
+ // it.
+ if (Callee->getCallingConv() == CallingConv::Cold)
+ Bonus += InlineConstants::ColdccPenalty;
+
+ // Add to the inline quality for properties that make the call valuable to
+ // inline. This includes factors that indicate that the result of inlining
+ // the function will be optimizable. Currently this just looks at arguments
+ // passed into the function.
+ //
+ CallSite::arg_iterator I = CS.arg_begin();
+ for (Function::arg_iterator FI = Callee->arg_begin(), FE = Callee->arg_end();
+ FI != FE; ++I, ++FI)
+ // Compute any constant bonus due to inlining we want to give here.
+ if (isa<Constant>(I))
+ Bonus += CountBonusForConstant(FI, cast<Constant>(I));
+
+ return Bonus;
+}
+
+// getInlineCost - The heuristic used to determine if we should inline the
+// function call or not.
+//
+InlineCost InlineCostAnalyzer::getInlineCost(CallSite CS,
+ SmallPtrSet<const Function*, 16> &NeverInline) {
+ return getInlineCost(CS, CS.getCalledFunction(), NeverInline);
+}
+
+InlineCost InlineCostAnalyzer::getInlineCost(CallSite CS,
+ Function *Callee,
+ SmallPtrSet<const Function*, 16> &NeverInline) {
+ Instruction *TheCall = CS.getInstruction();
+ Function *Caller = TheCall->getParent()->getParent();
+
+ // Don't inline functions which can be redefined at link-time to mean
+ // something else. Don't inline functions marked noinline or call sites
+ // marked noinline.
+ if (Callee->mayBeOverridden() ||
+ Callee->hasFnAttr(Attribute::NoInline) || NeverInline.count(Callee) ||
+ CS.isNoInline())
+ return llvm::InlineCost::getNever();
+
+ // Get information about the callee.
+ FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
+
+ // If we haven't calculated this information yet, do so now.
+ if (CalleeFI->Metrics.NumBlocks == 0)
+ CalleeFI->analyzeFunction(Callee, TD);
+
+ // If we should never inline this, return a huge cost.
+ if (CalleeFI->NeverInline())
+ return InlineCost::getNever();
+
+ // FIXME: It would be nice to kill off CalleeFI->NeverInline. Then we
+ // could move this up and avoid computing the FunctionInfo for
+ // things we are going to just return always inline for. This
+ // requires handling setjmp somewhere else, however.
+ if (!Callee->isDeclaration() && Callee->hasFnAttr(Attribute::AlwaysInline))
+ return InlineCost::getAlways();
+
+ if (CalleeFI->Metrics.usesDynamicAlloca) {
+ // Get information about the caller.
+ FunctionInfo &CallerFI = CachedFunctionInfo[Caller];
+
+ // If we haven't calculated this information yet, do so now.
+ if (CallerFI.Metrics.NumBlocks == 0) {
+ CallerFI.analyzeFunction(Caller, TD);
+
+ // Recompute the CalleeFI pointer, getting Caller could have invalidated
+ // it.
+ CalleeFI = &CachedFunctionInfo[Callee];
+ }
+
+ // Don't inline a callee with dynamic alloca into a caller without them.
+ // Functions containing dynamic alloca's are inefficient in various ways;
+ // don't create more inefficiency.
+ if (!CallerFI.Metrics.usesDynamicAlloca)
+ return InlineCost::getNever();
+ }
+
+ // InlineCost - This value measures how good of an inline candidate this call
+ // site is to inline. A lower inline cost make is more likely for the call to
+ // be inlined. This value may go negative due to the fact that bonuses
+ // are negative numbers.
+ //
+ int InlineCost = getInlineSize(CS, Callee) + getInlineBonuses(CS, Callee);
+ return llvm::InlineCost::get(InlineCost);
+}
+
+// getSpecializationCost - The heuristic used to determine the code-size
+// impact of creating a specialized version of Callee with argument
+// SpecializedArgNo replaced by a constant.
+InlineCost InlineCostAnalyzer::getSpecializationCost(Function *Callee,
+ SmallVectorImpl<unsigned> &SpecializedArgNos)
+{
+ // Don't specialize functions which can be redefined at link-time to mean
+ // something else.
+ if (Callee->mayBeOverridden())
+ return llvm::InlineCost::getNever();
+
+ // Get information about the callee.
+ FunctionInfo *CalleeFI = &CachedFunctionInfo[Callee];
+
+ // If we haven't calculated this information yet, do so now.
+ if (CalleeFI->Metrics.NumBlocks == 0)
+ CalleeFI->analyzeFunction(Callee, TD);
+
+ int Cost = 0;
+
+ // Look at the original size of the callee. Each instruction counts as 5.
+ Cost += CalleeFI->Metrics.NumInsts * InlineConstants::InstrCost;
+
+ // Offset that with the amount of code that can be constant-folded
+ // away with the given arguments replaced by constants.
+ for (SmallVectorImpl<unsigned>::iterator an = SpecializedArgNos.begin(),
+ ae = SpecializedArgNos.end(); an != ae; ++an)
+ Cost -= CalleeFI->ArgumentWeights[*an].ConstantWeight;
+
+ return llvm::InlineCost::get(Cost);
+}
+
+// getInlineFudgeFactor - Return a > 1.0 factor if the inliner should use a
+// higher threshold to determine if the function call should be inlined.
+float InlineCostAnalyzer::getInlineFudgeFactor(CallSite CS) {
+ Function *Callee = CS.getCalledFunction();
+
+ // Get information about the callee.
+ FunctionInfo &CalleeFI = CachedFunctionInfo[Callee];
+
+ // If we haven't calculated this information yet, do so now.
+ if (CalleeFI.Metrics.NumBlocks == 0)
+ CalleeFI.analyzeFunction(Callee, TD);
+
+ float Factor = 1.0f;
+ // Single BB functions are often written to be inlined.
+ if (CalleeFI.Metrics.NumBlocks == 1)
+ Factor += 0.5f;
+
+ // Be more aggressive if the function contains a good chunk (if it mades up
+ // at least 10% of the instructions) of vector instructions.
+ if (CalleeFI.Metrics.NumVectorInsts > CalleeFI.Metrics.NumInsts/2)
+ Factor += 2.0f;
+ else if (CalleeFI.Metrics.NumVectorInsts > CalleeFI.Metrics.NumInsts/10)
+ Factor += 1.5f;
+ return Factor;
+}
+
+/// growCachedCostInfo - update the cached cost info for Caller after Callee has
+/// been inlined.
+void
+InlineCostAnalyzer::growCachedCostInfo(Function *Caller, Function *Callee) {
+ CodeMetrics &CallerMetrics = CachedFunctionInfo[Caller].Metrics;
+
+ // For small functions we prefer to recalculate the cost for better accuracy.
+ if (CallerMetrics.NumBlocks < 10 && CallerMetrics.NumInsts < 1000) {
+ resetCachedCostInfo(Caller);
+ return;
+ }
+
+ // For large functions, we can save a lot of computation time by skipping
+ // recalculations.
+ if (CallerMetrics.NumCalls > 0)
+ --CallerMetrics.NumCalls;
+
+ if (Callee == 0) return;
+
+ CodeMetrics &CalleeMetrics = CachedFunctionInfo[Callee].Metrics;
+
+ // If we don't have metrics for the callee, don't recalculate them just to
+ // update an approximation in the caller. Instead, just recalculate the
+ // caller info from scratch.
+ if (CalleeMetrics.NumBlocks == 0) {
+ resetCachedCostInfo(Caller);
+ return;
+ }
+
+ // Since CalleeMetrics were already calculated, we know that the CallerMetrics
+ // reference isn't invalidated: both were in the DenseMap.
+ CallerMetrics.usesDynamicAlloca |= CalleeMetrics.usesDynamicAlloca;
+
+ // FIXME: If any of these three are true for the callee, the callee was
+ // not inlined into the caller, so I think they're redundant here.
+ CallerMetrics.callsSetJmp |= CalleeMetrics.callsSetJmp;
+ CallerMetrics.isRecursive |= CalleeMetrics.isRecursive;
+ CallerMetrics.containsIndirectBr |= CalleeMetrics.containsIndirectBr;
+
+ CallerMetrics.NumInsts += CalleeMetrics.NumInsts;
+ CallerMetrics.NumBlocks += CalleeMetrics.NumBlocks;
+ CallerMetrics.NumCalls += CalleeMetrics.NumCalls;
+ CallerMetrics.NumVectorInsts += CalleeMetrics.NumVectorInsts;
+ CallerMetrics.NumRets += CalleeMetrics.NumRets;
+
+ // analyzeBasicBlock counts each function argument as an inst.
+ if (CallerMetrics.NumInsts >= Callee->arg_size())
+ CallerMetrics.NumInsts -= Callee->arg_size();
+ else
+ CallerMetrics.NumInsts = 0;
+
+ // We are not updating the argument weights. We have already determined that
+ // Caller is a fairly large function, so we accept the loss of precision.
+}
+
+/// clear - empty the cache of inline costs
+void InlineCostAnalyzer::clear() {
+ CachedFunctionInfo.clear();
+}
diff --git a/src/LLVM/lib/Analysis/InstCount.cpp b/src/LLVM/lib/Analysis/InstCount.cpp
new file mode 100644
index 0000000..3b385d2
--- /dev/null
+++ b/src/LLVM/lib/Analysis/InstCount.cpp
@@ -0,0 +1,87 @@
+//===-- InstCount.cpp - Collects the count of all instructions ------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass collects the count of all instructions and reports them
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "instcount"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Pass.h"
+#include "llvm/Function.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/InstVisitor.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/Statistic.h"
+using namespace llvm;
+
+STATISTIC(TotalInsts , "Number of instructions (of all types)");
+STATISTIC(TotalBlocks, "Number of basic blocks");
+STATISTIC(TotalFuncs , "Number of non-external functions");
+STATISTIC(TotalMemInst, "Number of memory instructions");
+
+#define HANDLE_INST(N, OPCODE, CLASS) \
+ STATISTIC(Num ## OPCODE ## Inst, "Number of " #OPCODE " insts");
+
+#include "llvm/Instruction.def"
+
+
+namespace {
+ class InstCount : public FunctionPass, public InstVisitor<InstCount> {
+ friend class InstVisitor<InstCount>;
+
+ void visitFunction (Function &F) { ++TotalFuncs; }
+ void visitBasicBlock(BasicBlock &BB) { ++TotalBlocks; }
+
+#define HANDLE_INST(N, OPCODE, CLASS) \
+ void visit##OPCODE(CLASS &) { ++Num##OPCODE##Inst; ++TotalInsts; }
+
+#include "llvm/Instruction.def"
+
+ void visitInstruction(Instruction &I) {
+ errs() << "Instruction Count does not know about " << I;
+ llvm_unreachable(0);
+ }
+ public:
+ static char ID; // Pass identification, replacement for typeid
+ InstCount() : FunctionPass(ID) {
+ initializeInstCountPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnFunction(Function &F);
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+ virtual void print(raw_ostream &O, const Module *M) const {}
+
+ };
+}
+
+char InstCount::ID = 0;
+INITIALIZE_PASS(InstCount, "instcount",
+ "Counts the various types of Instructions", false, true)
+
+FunctionPass *llvm::createInstCountPass() { return new InstCount(); }
+
+// InstCount::run - This is the main Analysis entry point for a
+// function.
+//
+bool InstCount::runOnFunction(Function &F) {
+ unsigned StartMemInsts =
+ NumGetElementPtrInst + NumLoadInst + NumStoreInst + NumCallInst +
+ NumInvokeInst + NumAllocaInst;
+ visit(F);
+ unsigned EndMemInsts =
+ NumGetElementPtrInst + NumLoadInst + NumStoreInst + NumCallInst +
+ NumInvokeInst + NumAllocaInst;
+ TotalMemInst += EndMemInsts-StartMemInsts;
+ return false;
+}
diff --git a/src/LLVM/lib/Analysis/InstructionSimplify.cpp b/src/LLVM/lib/Analysis/InstructionSimplify.cpp
index 24cd343..131cc97 100644
--- a/src/LLVM/lib/Analysis/InstructionSimplify.cpp
+++ b/src/LLVM/lib/Analysis/InstructionSimplify.cpp
@@ -8,179 +8,1422 @@
//===----------------------------------------------------------------------===//
//
// This file implements routines for folding instructions into simpler forms
-// that do not require creating new instructions. For example, this does
-// constant folding, and can handle identities like (X&0)->0.
+// that do not require creating new instructions. This does constant folding
+// ("add i32 1, 1" -> "2") but can also handle non-constant operands, either
+// returning a constant ("and i32 %x, 0" -> "0") or an already existing value
+// ("and i32 %x, %x" -> "%x"). All operands are assumed to have already been
+// simplified: This is usually true and assuming it simplifies the logic (if
+// they have not been simplified then results are correct but maybe suboptimal).
//
//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "instsimplify"
+#include "llvm/Operator.h"
+#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/ConstantFolding.h"
-#include "llvm/Support/ValueHandle.h"
-#include "llvm/Instructions.h"
+#include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Support/ConstantRange.h"
#include "llvm/Support/PatternMatch.h"
+#include "llvm/Support/ValueHandle.h"
+#include "llvm/Target/TargetData.h"
using namespace llvm;
using namespace llvm::PatternMatch;
+enum { RecursionLimit = 3 };
+
+STATISTIC(NumExpand, "Number of expansions");
+STATISTIC(NumFactor , "Number of factorizations");
+STATISTIC(NumReassoc, "Number of reassociations");
+
+static Value *SimplifyAndInst(Value *, Value *, const TargetData *,
+ const DominatorTree *, unsigned);
+static Value *SimplifyBinOp(unsigned, Value *, Value *, const TargetData *,
+ const DominatorTree *, unsigned);
+static Value *SimplifyCmpInst(unsigned, Value *, Value *, const TargetData *,
+ const DominatorTree *, unsigned);
+static Value *SimplifyOrInst(Value *, Value *, const TargetData *,
+ const DominatorTree *, unsigned);
+static Value *SimplifyXorInst(Value *, Value *, const TargetData *,
+ const DominatorTree *, unsigned);
+
+/// getFalse - For a boolean type, or a vector of boolean type, return false, or
+/// a vector with every element false, as appropriate for the type.
+static Constant *getFalse(Type *Ty) {
+ assert((Ty->isIntegerTy(1) ||
+ (Ty->isVectorTy() &&
+ cast<VectorType>(Ty)->getElementType()->isIntegerTy(1))) &&
+ "Expected i1 type or a vector of i1!");
+ return Constant::getNullValue(Ty);
+}
+
+/// getTrue - For a boolean type, or a vector of boolean type, return true, or
+/// a vector with every element true, as appropriate for the type.
+static Constant *getTrue(Type *Ty) {
+ assert((Ty->isIntegerTy(1) ||
+ (Ty->isVectorTy() &&
+ cast<VectorType>(Ty)->getElementType()->isIntegerTy(1))) &&
+ "Expected i1 type or a vector of i1!");
+ return Constant::getAllOnesValue(Ty);
+}
+
+/// ValueDominatesPHI - Does the given value dominate the specified phi node?
+static bool ValueDominatesPHI(Value *V, PHINode *P, const DominatorTree *DT) {
+ Instruction *I = dyn_cast<Instruction>(V);
+ if (!I)
+ // Arguments and constants dominate all instructions.
+ return true;
+
+ // If we have a DominatorTree then do a precise test.
+ if (DT)
+ return DT->dominates(I, P);
+
+ // Otherwise, if the instruction is in the entry block, and is not an invoke,
+ // then it obviously dominates all phi nodes.
+ if (I->getParent() == &I->getParent()->getParent()->getEntryBlock() &&
+ !isa<InvokeInst>(I))
+ return true;
+
+ return false;
+}
+
+/// ExpandBinOp - Simplify "A op (B op' C)" by distributing op over op', turning
+/// it into "(A op B) op' (A op C)". Here "op" is given by Opcode and "op'" is
+/// given by OpcodeToExpand, while "A" corresponds to LHS and "B op' C" to RHS.
+/// Also performs the transform "(A op' B) op C" -> "(A op C) op' (B op C)".
+/// Returns the simplified value, or null if no simplification was performed.
+static Value *ExpandBinOp(unsigned Opcode, Value *LHS, Value *RHS,
+ unsigned OpcToExpand, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
+ Instruction::BinaryOps OpcodeToExpand = (Instruction::BinaryOps)OpcToExpand;
+ // Recursion is always used, so bail out at once if we already hit the limit.
+ if (!MaxRecurse--)
+ return 0;
+
+ // Check whether the expression has the form "(A op' B) op C".
+ if (BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS))
+ if (Op0->getOpcode() == OpcodeToExpand) {
+ // It does! Try turning it into "(A op C) op' (B op C)".
+ Value *A = Op0->getOperand(0), *B = Op0->getOperand(1), *C = RHS;
+ // Do "A op C" and "B op C" both simplify?
+ if (Value *L = SimplifyBinOp(Opcode, A, C, TD, DT, MaxRecurse))
+ if (Value *R = SimplifyBinOp(Opcode, B, C, TD, DT, MaxRecurse)) {
+ // They do! Return "L op' R" if it simplifies or is already available.
+ // If "L op' R" equals "A op' B" then "L op' R" is just the LHS.
+ if ((L == A && R == B) || (Instruction::isCommutative(OpcodeToExpand)
+ && L == B && R == A)) {
+ ++NumExpand;
+ return LHS;
+ }
+ // Otherwise return "L op' R" if it simplifies.
+ if (Value *V = SimplifyBinOp(OpcodeToExpand, L, R, TD, DT,
+ MaxRecurse)) {
+ ++NumExpand;
+ return V;
+ }
+ }
+ }
+
+ // Check whether the expression has the form "A op (B op' C)".
+ if (BinaryOperator *Op1 = dyn_cast<BinaryOperator>(RHS))
+ if (Op1->getOpcode() == OpcodeToExpand) {
+ // It does! Try turning it into "(A op B) op' (A op C)".
+ Value *A = LHS, *B = Op1->getOperand(0), *C = Op1->getOperand(1);
+ // Do "A op B" and "A op C" both simplify?
+ if (Value *L = SimplifyBinOp(Opcode, A, B, TD, DT, MaxRecurse))
+ if (Value *R = SimplifyBinOp(Opcode, A, C, TD, DT, MaxRecurse)) {
+ // They do! Return "L op' R" if it simplifies or is already available.
+ // If "L op' R" equals "B op' C" then "L op' R" is just the RHS.
+ if ((L == B && R == C) || (Instruction::isCommutative(OpcodeToExpand)
+ && L == C && R == B)) {
+ ++NumExpand;
+ return RHS;
+ }
+ // Otherwise return "L op' R" if it simplifies.
+ if (Value *V = SimplifyBinOp(OpcodeToExpand, L, R, TD, DT,
+ MaxRecurse)) {
+ ++NumExpand;
+ return V;
+ }
+ }
+ }
+
+ return 0;
+}
+
+/// FactorizeBinOp - Simplify "LHS Opcode RHS" by factorizing out a common term
+/// using the operation OpCodeToExtract. For example, when Opcode is Add and
+/// OpCodeToExtract is Mul then this tries to turn "(A*B)+(A*C)" into "A*(B+C)".
+/// Returns the simplified value, or null if no simplification was performed.
+static Value *FactorizeBinOp(unsigned Opcode, Value *LHS, Value *RHS,
+ unsigned OpcToExtract, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
+ Instruction::BinaryOps OpcodeToExtract = (Instruction::BinaryOps)OpcToExtract;
+ // Recursion is always used, so bail out at once if we already hit the limit.
+ if (!MaxRecurse--)
+ return 0;
+
+ BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS);
+ BinaryOperator *Op1 = dyn_cast<BinaryOperator>(RHS);
+
+ if (!Op0 || Op0->getOpcode() != OpcodeToExtract ||
+ !Op1 || Op1->getOpcode() != OpcodeToExtract)
+ return 0;
+
+ // The expression has the form "(A op' B) op (C op' D)".
+ Value *A = Op0->getOperand(0), *B = Op0->getOperand(1);
+ Value *C = Op1->getOperand(0), *D = Op1->getOperand(1);
+
+ // Use left distributivity, i.e. "X op' (Y op Z) = (X op' Y) op (X op' Z)".
+ // Does the instruction have the form "(A op' B) op (A op' D)" or, in the
+ // commutative case, "(A op' B) op (C op' A)"?
+ if (A == C || (Instruction::isCommutative(OpcodeToExtract) && A == D)) {
+ Value *DD = A == C ? D : C;
+ // Form "A op' (B op DD)" if it simplifies completely.
+ // Does "B op DD" simplify?
+ if (Value *V = SimplifyBinOp(Opcode, B, DD, TD, DT, MaxRecurse)) {
+ // It does! Return "A op' V" if it simplifies or is already available.
+ // If V equals B then "A op' V" is just the LHS. If V equals DD then
+ // "A op' V" is just the RHS.
+ if (V == B || V == DD) {
+ ++NumFactor;
+ return V == B ? LHS : RHS;
+ }
+ // Otherwise return "A op' V" if it simplifies.
+ if (Value *W = SimplifyBinOp(OpcodeToExtract, A, V, TD, DT, MaxRecurse)) {
+ ++NumFactor;
+ return W;
+ }
+ }
+ }
+
+ // Use right distributivity, i.e. "(X op Y) op' Z = (X op' Z) op (Y op' Z)".
+ // Does the instruction have the form "(A op' B) op (C op' B)" or, in the
+ // commutative case, "(A op' B) op (B op' D)"?
+ if (B == D || (Instruction::isCommutative(OpcodeToExtract) && B == C)) {
+ Value *CC = B == D ? C : D;
+ // Form "(A op CC) op' B" if it simplifies completely..
+ // Does "A op CC" simplify?
+ if (Value *V = SimplifyBinOp(Opcode, A, CC, TD, DT, MaxRecurse)) {
+ // It does! Return "V op' B" if it simplifies or is already available.
+ // If V equals A then "V op' B" is just the LHS. If V equals CC then
+ // "V op' B" is just the RHS.
+ if (V == A || V == CC) {
+ ++NumFactor;
+ return V == A ? LHS : RHS;
+ }
+ // Otherwise return "V op' B" if it simplifies.
+ if (Value *W = SimplifyBinOp(OpcodeToExtract, V, B, TD, DT, MaxRecurse)) {
+ ++NumFactor;
+ return W;
+ }
+ }
+ }
+
+ return 0;
+}
+
+/// SimplifyAssociativeBinOp - Generic simplifications for associative binary
+/// operations. Returns the simpler value, or null if none was found.
+static Value *SimplifyAssociativeBinOp(unsigned Opc, Value *LHS, Value *RHS,
+ const TargetData *TD,
+ const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ Instruction::BinaryOps Opcode = (Instruction::BinaryOps)Opc;
+ assert(Instruction::isAssociative(Opcode) && "Not an associative operation!");
+
+ // Recursion is always used, so bail out at once if we already hit the limit.
+ if (!MaxRecurse--)
+ return 0;
+
+ BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS);
+ BinaryOperator *Op1 = dyn_cast<BinaryOperator>(RHS);
+
+ // Transform: "(A op B) op C" ==> "A op (B op C)" if it simplifies completely.
+ if (Op0 && Op0->getOpcode() == Opcode) {
+ Value *A = Op0->getOperand(0);
+ Value *B = Op0->getOperand(1);
+ Value *C = RHS;
+
+ // Does "B op C" simplify?
+ if (Value *V = SimplifyBinOp(Opcode, B, C, TD, DT, MaxRecurse)) {
+ // It does! Return "A op V" if it simplifies or is already available.
+ // If V equals B then "A op V" is just the LHS.
+ if (V == B) return LHS;
+ // Otherwise return "A op V" if it simplifies.
+ if (Value *W = SimplifyBinOp(Opcode, A, V, TD, DT, MaxRecurse)) {
+ ++NumReassoc;
+ return W;
+ }
+ }
+ }
+
+ // Transform: "A op (B op C)" ==> "(A op B) op C" if it simplifies completely.
+ if (Op1 && Op1->getOpcode() == Opcode) {
+ Value *A = LHS;
+ Value *B = Op1->getOperand(0);
+ Value *C = Op1->getOperand(1);
+
+ // Does "A op B" simplify?
+ if (Value *V = SimplifyBinOp(Opcode, A, B, TD, DT, MaxRecurse)) {
+ // It does! Return "V op C" if it simplifies or is already available.
+ // If V equals B then "V op C" is just the RHS.
+ if (V == B) return RHS;
+ // Otherwise return "V op C" if it simplifies.
+ if (Value *W = SimplifyBinOp(Opcode, V, C, TD, DT, MaxRecurse)) {
+ ++NumReassoc;
+ return W;
+ }
+ }
+ }
+
+ // The remaining transforms require commutativity as well as associativity.
+ if (!Instruction::isCommutative(Opcode))
+ return 0;
+
+ // Transform: "(A op B) op C" ==> "(C op A) op B" if it simplifies completely.
+ if (Op0 && Op0->getOpcode() == Opcode) {
+ Value *A = Op0->getOperand(0);
+ Value *B = Op0->getOperand(1);
+ Value *C = RHS;
+
+ // Does "C op A" simplify?
+ if (Value *V = SimplifyBinOp(Opcode, C, A, TD, DT, MaxRecurse)) {
+ // It does! Return "V op B" if it simplifies or is already available.
+ // If V equals A then "V op B" is just the LHS.
+ if (V == A) return LHS;
+ // Otherwise return "V op B" if it simplifies.
+ if (Value *W = SimplifyBinOp(Opcode, V, B, TD, DT, MaxRecurse)) {
+ ++NumReassoc;
+ return W;
+ }
+ }
+ }
+
+ // Transform: "A op (B op C)" ==> "B op (C op A)" if it simplifies completely.
+ if (Op1 && Op1->getOpcode() == Opcode) {
+ Value *A = LHS;
+ Value *B = Op1->getOperand(0);
+ Value *C = Op1->getOperand(1);
+
+ // Does "C op A" simplify?
+ if (Value *V = SimplifyBinOp(Opcode, C, A, TD, DT, MaxRecurse)) {
+ // It does! Return "B op V" if it simplifies or is already available.
+ // If V equals C then "B op V" is just the RHS.
+ if (V == C) return RHS;
+ // Otherwise return "B op V" if it simplifies.
+ if (Value *W = SimplifyBinOp(Opcode, B, V, TD, DT, MaxRecurse)) {
+ ++NumReassoc;
+ return W;
+ }
+ }
+ }
+
+ return 0;
+}
+
+/// ThreadBinOpOverSelect - In the case of a binary operation with a select
+/// instruction as an operand, try to simplify the binop by seeing whether
+/// evaluating it on both branches of the select results in the same value.
+/// Returns the common value if so, otherwise returns null.
+static Value *ThreadBinOpOverSelect(unsigned Opcode, Value *LHS, Value *RHS,
+ const TargetData *TD,
+ const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ // Recursion is always used, so bail out at once if we already hit the limit.
+ if (!MaxRecurse--)
+ return 0;
+
+ SelectInst *SI;
+ if (isa<SelectInst>(LHS)) {
+ SI = cast<SelectInst>(LHS);
+ } else {
+ assert(isa<SelectInst>(RHS) && "No select instruction operand!");
+ SI = cast<SelectInst>(RHS);
+ }
+
+ // Evaluate the BinOp on the true and false branches of the select.
+ Value *TV;
+ Value *FV;
+ if (SI == LHS) {
+ TV = SimplifyBinOp(Opcode, SI->getTrueValue(), RHS, TD, DT, MaxRecurse);
+ FV = SimplifyBinOp(Opcode, SI->getFalseValue(), RHS, TD, DT, MaxRecurse);
+ } else {
+ TV = SimplifyBinOp(Opcode, LHS, SI->getTrueValue(), TD, DT, MaxRecurse);
+ FV = SimplifyBinOp(Opcode, LHS, SI->getFalseValue(), TD, DT, MaxRecurse);
+ }
+
+ // If they simplified to the same value, then return the common value.
+ // If they both failed to simplify then return null.
+ if (TV == FV)
+ return TV;
+
+ // If one branch simplified to undef, return the other one.
+ if (TV && isa<UndefValue>(TV))
+ return FV;
+ if (FV && isa<UndefValue>(FV))
+ return TV;
+
+ // If applying the operation did not change the true and false select values,
+ // then the result of the binop is the select itself.
+ if (TV == SI->getTrueValue() && FV == SI->getFalseValue())
+ return SI;
+
+ // If one branch simplified and the other did not, and the simplified
+ // value is equal to the unsimplified one, return the simplified value.
+ // For example, select (cond, X, X & Z) & Z -> X & Z.
+ if ((FV && !TV) || (TV && !FV)) {
+ // Check that the simplified value has the form "X op Y" where "op" is the
+ // same as the original operation.
+ Instruction *Simplified = dyn_cast<Instruction>(FV ? FV : TV);
+ if (Simplified && Simplified->getOpcode() == Opcode) {
+ // The value that didn't simplify is "UnsimplifiedLHS op UnsimplifiedRHS".
+ // We already know that "op" is the same as for the simplified value. See
+ // if the operands match too. If so, return the simplified value.
+ Value *UnsimplifiedBranch = FV ? SI->getTrueValue() : SI->getFalseValue();
+ Value *UnsimplifiedLHS = SI == LHS ? UnsimplifiedBranch : LHS;
+ Value *UnsimplifiedRHS = SI == LHS ? RHS : UnsimplifiedBranch;
+ if (Simplified->getOperand(0) == UnsimplifiedLHS &&
+ Simplified->getOperand(1) == UnsimplifiedRHS)
+ return Simplified;
+ if (Simplified->isCommutative() &&
+ Simplified->getOperand(1) == UnsimplifiedLHS &&
+ Simplified->getOperand(0) == UnsimplifiedRHS)
+ return Simplified;
+ }
+ }
+
+ return 0;
+}
+
+/// ThreadCmpOverSelect - In the case of a comparison with a select instruction,
+/// try to simplify the comparison by seeing whether both branches of the select
+/// result in the same value. Returns the common value if so, otherwise returns
+/// null.
+static Value *ThreadCmpOverSelect(CmpInst::Predicate Pred, Value *LHS,
+ Value *RHS, const TargetData *TD,
+ const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ // Recursion is always used, so bail out at once if we already hit the limit.
+ if (!MaxRecurse--)
+ return 0;
+
+ // Make sure the select is on the LHS.
+ if (!isa<SelectInst>(LHS)) {
+ std::swap(LHS, RHS);
+ Pred = CmpInst::getSwappedPredicate(Pred);
+ }
+ assert(isa<SelectInst>(LHS) && "Not comparing with a select instruction!");
+ SelectInst *SI = cast<SelectInst>(LHS);
+
+ // Now that we have "cmp select(Cond, TV, FV), RHS", analyse it.
+ // Does "cmp TV, RHS" simplify?
+ if (Value *TCmp = SimplifyCmpInst(Pred, SI->getTrueValue(), RHS, TD, DT,
+ MaxRecurse)) {
+ // It does! Does "cmp FV, RHS" simplify?
+ if (Value *FCmp = SimplifyCmpInst(Pred, SI->getFalseValue(), RHS, TD, DT,
+ MaxRecurse)) {
+ // It does! If they simplified to the same value, then use it as the
+ // result of the original comparison.
+ if (TCmp == FCmp)
+ return TCmp;
+ Value *Cond = SI->getCondition();
+ // If the false value simplified to false, then the result of the compare
+ // is equal to "Cond && TCmp". This also catches the case when the false
+ // value simplified to false and the true value to true, returning "Cond".
+ if (match(FCmp, m_Zero()))
+ if (Value *V = SimplifyAndInst(Cond, TCmp, TD, DT, MaxRecurse))
+ return V;
+ // If the true value simplified to true, then the result of the compare
+ // is equal to "Cond || FCmp".
+ if (match(TCmp, m_One()))
+ if (Value *V = SimplifyOrInst(Cond, FCmp, TD, DT, MaxRecurse))
+ return V;
+ // Finally, if the false value simplified to true and the true value to
+ // false, then the result of the compare is equal to "!Cond".
+ if (match(FCmp, m_One()) && match(TCmp, m_Zero()))
+ if (Value *V =
+ SimplifyXorInst(Cond, Constant::getAllOnesValue(Cond->getType()),
+ TD, DT, MaxRecurse))
+ return V;
+ }
+ }
+
+ return 0;
+}
+
+/// ThreadBinOpOverPHI - In the case of a binary operation with an operand that
+/// is a PHI instruction, try to simplify the binop by seeing whether evaluating
+/// it on the incoming phi values yields the same result for every value. If so
+/// returns the common value, otherwise returns null.
+static Value *ThreadBinOpOverPHI(unsigned Opcode, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ // Recursion is always used, so bail out at once if we already hit the limit.
+ if (!MaxRecurse--)
+ return 0;
+
+ PHINode *PI;
+ if (isa<PHINode>(LHS)) {
+ PI = cast<PHINode>(LHS);
+ // Bail out if RHS and the phi may be mutually interdependent due to a loop.
+ if (!ValueDominatesPHI(RHS, PI, DT))
+ return 0;
+ } else {
+ assert(isa<PHINode>(RHS) && "No PHI instruction operand!");
+ PI = cast<PHINode>(RHS);
+ // Bail out if LHS and the phi may be mutually interdependent due to a loop.
+ if (!ValueDominatesPHI(LHS, PI, DT))
+ return 0;
+ }
+
+ // Evaluate the BinOp on the incoming phi values.
+ Value *CommonValue = 0;
+ for (unsigned i = 0, e = PI->getNumIncomingValues(); i != e; ++i) {
+ Value *Incoming = PI->getIncomingValue(i);
+ // If the incoming value is the phi node itself, it can safely be skipped.
+ if (Incoming == PI) continue;
+ Value *V = PI == LHS ?
+ SimplifyBinOp(Opcode, Incoming, RHS, TD, DT, MaxRecurse) :
+ SimplifyBinOp(Opcode, LHS, Incoming, TD, DT, MaxRecurse);
+ // If the operation failed to simplify, or simplified to a different value
+ // to previously, then give up.
+ if (!V || (CommonValue && V != CommonValue))
+ return 0;
+ CommonValue = V;
+ }
+
+ return CommonValue;
+}
+
+/// ThreadCmpOverPHI - In the case of a comparison with a PHI instruction, try
+/// try to simplify the comparison by seeing whether comparing with all of the
+/// incoming phi values yields the same result every time. If so returns the
+/// common result, otherwise returns null.
+static Value *ThreadCmpOverPHI(CmpInst::Predicate Pred, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ // Recursion is always used, so bail out at once if we already hit the limit.
+ if (!MaxRecurse--)
+ return 0;
+
+ // Make sure the phi is on the LHS.
+ if (!isa<PHINode>(LHS)) {
+ std::swap(LHS, RHS);
+ Pred = CmpInst::getSwappedPredicate(Pred);
+ }
+ assert(isa<PHINode>(LHS) && "Not comparing with a phi instruction!");
+ PHINode *PI = cast<PHINode>(LHS);
+
+ // Bail out if RHS and the phi may be mutually interdependent due to a loop.
+ if (!ValueDominatesPHI(RHS, PI, DT))
+ return 0;
+
+ // Evaluate the BinOp on the incoming phi values.
+ Value *CommonValue = 0;
+ for (unsigned i = 0, e = PI->getNumIncomingValues(); i != e; ++i) {
+ Value *Incoming = PI->getIncomingValue(i);
+ // If the incoming value is the phi node itself, it can safely be skipped.
+ if (Incoming == PI) continue;
+ Value *V = SimplifyCmpInst(Pred, Incoming, RHS, TD, DT, MaxRecurse);
+ // If the operation failed to simplify, or simplified to a different value
+ // to previously, then give up.
+ if (!V || (CommonValue && V != CommonValue))
+ return 0;
+ CommonValue = V;
+ }
+
+ return CommonValue;
+}
+
/// SimplifyAddInst - Given operands for an Add, see if we can
/// fold the result. If not, this returns null.
-Value *llvm::SimplifyAddInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
- const TargetData *TD) {
+static Value *SimplifyAddInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
if (Constant *CLHS = dyn_cast<Constant>(Op0)) {
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::Add, CLHS->getType(),
- Ops, 2, TD);
+ Ops, TD);
}
-
+
// Canonicalize the constant to the RHS.
std::swap(Op0, Op1);
}
-
- if (Constant *Op1C = dyn_cast<Constant>(Op1)) {
- // X + undef -> undef
- if (isa<UndefValue>(Op1C))
- return Op1C;
-
- // X + 0 --> X
- if (Op1C->isNullValue())
- return Op0;
- }
-
- // FIXME: Could pull several more out of instcombine.
+
+ // X + undef -> undef
+ if (match(Op1, m_Undef()))
+ return Op1;
+
+ // X + 0 -> X
+ if (match(Op1, m_Zero()))
+ return Op0;
+
+ // X + (Y - X) -> Y
+ // (Y - X) + X -> Y
+ // Eg: X + -X -> 0
+ Value *Y = 0;
+ if (match(Op1, m_Sub(m_Value(Y), m_Specific(Op0))) ||
+ match(Op0, m_Sub(m_Value(Y), m_Specific(Op1))))
+ return Y;
+
+ // X + ~X -> -1 since ~X = -X-1
+ if (match(Op0, m_Not(m_Specific(Op1))) ||
+ match(Op1, m_Not(m_Specific(Op0))))
+ return Constant::getAllOnesValue(Op0->getType());
+
+ /// i1 add -> xor.
+ if (MaxRecurse && Op0->getType()->isIntegerTy(1))
+ if (Value *V = SimplifyXorInst(Op0, Op1, TD, DT, MaxRecurse-1))
+ return V;
+
+ // Try some generic simplifications for associative operations.
+ if (Value *V = SimplifyAssociativeBinOp(Instruction::Add, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // Mul distributes over Add. Try some generic simplifications based on this.
+ if (Value *V = FactorizeBinOp(Instruction::Add, Op0, Op1, Instruction::Mul,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // Threading Add over selects and phi nodes is pointless, so don't bother.
+ // Threading over the select in "A + select(cond, B, C)" means evaluating
+ // "A+B" and "A+C" and seeing if they are equal; but they are equal if and
+ // only if B and C are equal. If B and C are equal then (since we assume
+ // that operands have already been simplified) "select(cond, B, C)" should
+ // have been simplified to the common value of B and C already. Analysing
+ // "A+B" and "A+C" thus gains nothing, but costs compile time. Similarly
+ // for threading over phi nodes.
+
return 0;
}
+Value *llvm::SimplifyAddInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifyAddInst(Op0, Op1, isNSW, isNUW, TD, DT, RecursionLimit);
+}
+
+/// SimplifySubInst - Given operands for a Sub, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifySubInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ if (Constant *CLHS = dyn_cast<Constant>(Op0))
+ if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
+ Constant *Ops[] = { CLHS, CRHS };
+ return ConstantFoldInstOperands(Instruction::Sub, CLHS->getType(),
+ Ops, TD);
+ }
+
+ // X - undef -> undef
+ // undef - X -> undef
+ if (match(Op0, m_Undef()) || match(Op1, m_Undef()))
+ return UndefValue::get(Op0->getType());
+
+ // X - 0 -> X
+ if (match(Op1, m_Zero()))
+ return Op0;
+
+ // X - X -> 0
+ if (Op0 == Op1)
+ return Constant::getNullValue(Op0->getType());
+
+ // (X*2) - X -> X
+ // (X<<1) - X -> X
+ Value *X = 0;
+ if (match(Op0, m_Mul(m_Specific(Op1), m_ConstantInt<2>())) ||
+ match(Op0, m_Shl(m_Specific(Op1), m_One())))
+ return Op1;
+
+ // (X + Y) - Z -> X + (Y - Z) or Y + (X - Z) if everything simplifies.
+ // For example, (X + Y) - Y -> X; (Y + X) - Y -> X
+ Value *Y = 0, *Z = Op1;
+ if (MaxRecurse && match(Op0, m_Add(m_Value(X), m_Value(Y)))) { // (X + Y) - Z
+ // See if "V === Y - Z" simplifies.
+ if (Value *V = SimplifyBinOp(Instruction::Sub, Y, Z, TD, DT, MaxRecurse-1))
+ // It does! Now see if "X + V" simplifies.
+ if (Value *W = SimplifyBinOp(Instruction::Add, X, V, TD, DT,
+ MaxRecurse-1)) {
+ // It does, we successfully reassociated!
+ ++NumReassoc;
+ return W;
+ }
+ // See if "V === X - Z" simplifies.
+ if (Value *V = SimplifyBinOp(Instruction::Sub, X, Z, TD, DT, MaxRecurse-1))
+ // It does! Now see if "Y + V" simplifies.
+ if (Value *W = SimplifyBinOp(Instruction::Add, Y, V, TD, DT,
+ MaxRecurse-1)) {
+ // It does, we successfully reassociated!
+ ++NumReassoc;
+ return W;
+ }
+ }
+
+ // X - (Y + Z) -> (X - Y) - Z or (X - Z) - Y if everything simplifies.
+ // For example, X - (X + 1) -> -1
+ X = Op0;
+ if (MaxRecurse && match(Op1, m_Add(m_Value(Y), m_Value(Z)))) { // X - (Y + Z)
+ // See if "V === X - Y" simplifies.
+ if (Value *V = SimplifyBinOp(Instruction::Sub, X, Y, TD, DT, MaxRecurse-1))
+ // It does! Now see if "V - Z" simplifies.
+ if (Value *W = SimplifyBinOp(Instruction::Sub, V, Z, TD, DT,
+ MaxRecurse-1)) {
+ // It does, we successfully reassociated!
+ ++NumReassoc;
+ return W;
+ }
+ // See if "V === X - Z" simplifies.
+ if (Value *V = SimplifyBinOp(Instruction::Sub, X, Z, TD, DT, MaxRecurse-1))
+ // It does! Now see if "V - Y" simplifies.
+ if (Value *W = SimplifyBinOp(Instruction::Sub, V, Y, TD, DT,
+ MaxRecurse-1)) {
+ // It does, we successfully reassociated!
+ ++NumReassoc;
+ return W;
+ }
+ }
+
+ // Z - (X - Y) -> (Z - X) + Y if everything simplifies.
+ // For example, X - (X - Y) -> Y.
+ Z = Op0;
+ if (MaxRecurse && match(Op1, m_Sub(m_Value(X), m_Value(Y)))) // Z - (X - Y)
+ // See if "V === Z - X" simplifies.
+ if (Value *V = SimplifyBinOp(Instruction::Sub, Z, X, TD, DT, MaxRecurse-1))
+ // It does! Now see if "V + Y" simplifies.
+ if (Value *W = SimplifyBinOp(Instruction::Add, V, Y, TD, DT,
+ MaxRecurse-1)) {
+ // It does, we successfully reassociated!
+ ++NumReassoc;
+ return W;
+ }
+
+ // Mul distributes over Sub. Try some generic simplifications based on this.
+ if (Value *V = FactorizeBinOp(Instruction::Sub, Op0, Op1, Instruction::Mul,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // i1 sub -> xor.
+ if (MaxRecurse && Op0->getType()->isIntegerTy(1))
+ if (Value *V = SimplifyXorInst(Op0, Op1, TD, DT, MaxRecurse-1))
+ return V;
+
+ // Threading Sub over selects and phi nodes is pointless, so don't bother.
+ // Threading over the select in "A - select(cond, B, C)" means evaluating
+ // "A-B" and "A-C" and seeing if they are equal; but they are equal if and
+ // only if B and C are equal. If B and C are equal then (since we assume
+ // that operands have already been simplified) "select(cond, B, C)" should
+ // have been simplified to the common value of B and C already. Analysing
+ // "A-B" and "A-C" thus gains nothing, but costs compile time. Similarly
+ // for threading over phi nodes.
+
+ return 0;
+}
+
+Value *llvm::SimplifySubInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifySubInst(Op0, Op1, isNSW, isNUW, TD, DT, RecursionLimit);
+}
+
+/// SimplifyMulInst - Given operands for a Mul, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyMulInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
+ if (Constant *CLHS = dyn_cast<Constant>(Op0)) {
+ if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
+ Constant *Ops[] = { CLHS, CRHS };
+ return ConstantFoldInstOperands(Instruction::Mul, CLHS->getType(),
+ Ops, TD);
+ }
+
+ // Canonicalize the constant to the RHS.
+ std::swap(Op0, Op1);
+ }
+
+ // X * undef -> 0
+ if (match(Op1, m_Undef()))
+ return Constant::getNullValue(Op0->getType());
+
+ // X * 0 -> 0
+ if (match(Op1, m_Zero()))
+ return Op1;
+
+ // X * 1 -> X
+ if (match(Op1, m_One()))
+ return Op0;
+
+ // (X / Y) * Y -> X if the division is exact.
+ Value *X = 0, *Y = 0;
+ if ((match(Op0, m_IDiv(m_Value(X), m_Value(Y))) && Y == Op1) || // (X / Y) * Y
+ (match(Op1, m_IDiv(m_Value(X), m_Value(Y))) && Y == Op0)) { // Y * (X / Y)
+ BinaryOperator *Div = cast<BinaryOperator>(Y == Op1 ? Op0 : Op1);
+ if (Div->isExact())
+ return X;
+ }
+
+ // i1 mul -> and.
+ if (MaxRecurse && Op0->getType()->isIntegerTy(1))
+ if (Value *V = SimplifyAndInst(Op0, Op1, TD, DT, MaxRecurse-1))
+ return V;
+
+ // Try some generic simplifications for associative operations.
+ if (Value *V = SimplifyAssociativeBinOp(Instruction::Mul, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // Mul distributes over Add. Try some generic simplifications based on this.
+ if (Value *V = ExpandBinOp(Instruction::Mul, Op0, Op1, Instruction::Add,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a select instruction, check whether
+ // operating on either branch of the select always yields the same value.
+ if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
+ if (Value *V = ThreadBinOpOverSelect(Instruction::Mul, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a phi instruction, check whether
+ // operating on all incoming values of the phi always yields the same value.
+ if (isa<PHINode>(Op0) || isa<PHINode>(Op1))
+ if (Value *V = ThreadBinOpOverPHI(Instruction::Mul, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ return 0;
+}
+
+Value *llvm::SimplifyMulInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifyMulInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
+/// SimplifyDiv - Given operands for an SDiv or UDiv, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyDiv(Instruction::BinaryOps Opcode, Value *Op0, Value *Op1,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ if (Constant *C0 = dyn_cast<Constant>(Op0)) {
+ if (Constant *C1 = dyn_cast<Constant>(Op1)) {
+ Constant *Ops[] = { C0, C1 };
+ return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, TD);
+ }
+ }
+
+ bool isSigned = Opcode == Instruction::SDiv;
+
+ // X / undef -> undef
+ if (match(Op1, m_Undef()))
+ return Op1;
+
+ // undef / X -> 0
+ if (match(Op0, m_Undef()))
+ return Constant::getNullValue(Op0->getType());
+
+ // 0 / X -> 0, we don't need to preserve faults!
+ if (match(Op0, m_Zero()))
+ return Op0;
+
+ // X / 1 -> X
+ if (match(Op1, m_One()))
+ return Op0;
+
+ if (Op0->getType()->isIntegerTy(1))
+ // It can't be division by zero, hence it must be division by one.
+ return Op0;
+
+ // X / X -> 1
+ if (Op0 == Op1)
+ return ConstantInt::get(Op0->getType(), 1);
+
+ // (X * Y) / Y -> X if the multiplication does not overflow.
+ Value *X = 0, *Y = 0;
+ if (match(Op0, m_Mul(m_Value(X), m_Value(Y))) && (X == Op1 || Y == Op1)) {
+ if (Y != Op1) std::swap(X, Y); // Ensure expression is (X * Y) / Y, Y = Op1
+ BinaryOperator *Mul = cast<BinaryOperator>(Op0);
+ // If the Mul knows it does not overflow, then we are good to go.
+ if ((isSigned && Mul->hasNoSignedWrap()) ||
+ (!isSigned && Mul->hasNoUnsignedWrap()))
+ return X;
+ // If X has the form X = A / Y then X * Y cannot overflow.
+ if (BinaryOperator *Div = dyn_cast<BinaryOperator>(X))
+ if (Div->getOpcode() == Opcode && Div->getOperand(1) == Y)
+ return X;
+ }
+
+ // (X rem Y) / Y -> 0
+ if ((isSigned && match(Op0, m_SRem(m_Value(), m_Specific(Op1)))) ||
+ (!isSigned && match(Op0, m_URem(m_Value(), m_Specific(Op1)))))
+ return Constant::getNullValue(Op0->getType());
+
+ // If the operation is with the result of a select instruction, check whether
+ // operating on either branch of the select always yields the same value.
+ if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
+ if (Value *V = ThreadBinOpOverSelect(Opcode, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a phi instruction, check whether
+ // operating on all incoming values of the phi always yields the same value.
+ if (isa<PHINode>(Op0) || isa<PHINode>(Op1))
+ if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ return 0;
+}
+
+/// SimplifySDivInst - Given operands for an SDiv, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifySDivInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
+ if (Value *V = SimplifyDiv(Instruction::SDiv, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ return 0;
+}
+
+Value *llvm::SimplifySDivInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifySDivInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
+/// SimplifyUDivInst - Given operands for a UDiv, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyUDivInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
+ if (Value *V = SimplifyDiv(Instruction::UDiv, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ return 0;
+}
+
+Value *llvm::SimplifyUDivInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifyUDivInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
+static Value *SimplifyFDivInst(Value *Op0, Value *Op1, const TargetData *,
+ const DominatorTree *, unsigned) {
+ // undef / X -> undef (the undef could be a snan).
+ if (match(Op0, m_Undef()))
+ return Op0;
+
+ // X / undef -> undef
+ if (match(Op1, m_Undef()))
+ return Op1;
+
+ return 0;
+}
+
+Value *llvm::SimplifyFDivInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifyFDivInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
+/// SimplifyRem - Given operands for an SRem or URem, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyRem(Instruction::BinaryOps Opcode, Value *Op0, Value *Op1,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ if (Constant *C0 = dyn_cast<Constant>(Op0)) {
+ if (Constant *C1 = dyn_cast<Constant>(Op1)) {
+ Constant *Ops[] = { C0, C1 };
+ return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, TD);
+ }
+ }
+
+ // X % undef -> undef
+ if (match(Op1, m_Undef()))
+ return Op1;
+
+ // undef % X -> 0
+ if (match(Op0, m_Undef()))
+ return Constant::getNullValue(Op0->getType());
+
+ // 0 % X -> 0, we don't need to preserve faults!
+ if (match(Op0, m_Zero()))
+ return Op0;
+
+ // X % 0 -> undef, we don't need to preserve faults!
+ if (match(Op1, m_Zero()))
+ return UndefValue::get(Op0->getType());
+
+ // X % 1 -> 0
+ if (match(Op1, m_One()))
+ return Constant::getNullValue(Op0->getType());
+
+ if (Op0->getType()->isIntegerTy(1))
+ // It can't be remainder by zero, hence it must be remainder by one.
+ return Constant::getNullValue(Op0->getType());
+
+ // X % X -> 0
+ if (Op0 == Op1)
+ return Constant::getNullValue(Op0->getType());
+
+ // If the operation is with the result of a select instruction, check whether
+ // operating on either branch of the select always yields the same value.
+ if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
+ if (Value *V = ThreadBinOpOverSelect(Opcode, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a phi instruction, check whether
+ // operating on all incoming values of the phi always yields the same value.
+ if (isa<PHINode>(Op0) || isa<PHINode>(Op1))
+ if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ return 0;
+}
+
+/// SimplifySRemInst - Given operands for an SRem, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifySRemInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
+ if (Value *V = SimplifyRem(Instruction::SRem, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ return 0;
+}
+
+Value *llvm::SimplifySRemInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifySRemInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
+/// SimplifyURemInst - Given operands for a URem, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyURemInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
+ if (Value *V = SimplifyRem(Instruction::URem, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ return 0;
+}
+
+Value *llvm::SimplifyURemInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifyURemInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
+static Value *SimplifyFRemInst(Value *Op0, Value *Op1, const TargetData *,
+ const DominatorTree *, unsigned) {
+ // undef % X -> undef (the undef could be a snan).
+ if (match(Op0, m_Undef()))
+ return Op0;
+
+ // X % undef -> undef
+ if (match(Op1, m_Undef()))
+ return Op1;
+
+ return 0;
+}
+
+Value *llvm::SimplifyFRemInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifyFRemInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
+/// SimplifyShift - Given operands for an Shl, LShr or AShr, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyShift(unsigned Opcode, Value *Op0, Value *Op1,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ if (Constant *C0 = dyn_cast<Constant>(Op0)) {
+ if (Constant *C1 = dyn_cast<Constant>(Op1)) {
+ Constant *Ops[] = { C0, C1 };
+ return ConstantFoldInstOperands(Opcode, C0->getType(), Ops, TD);
+ }
+ }
+
+ // 0 shift by X -> 0
+ if (match(Op0, m_Zero()))
+ return Op0;
+
+ // X shift by 0 -> X
+ if (match(Op1, m_Zero()))
+ return Op0;
+
+ // X shift by undef -> undef because it may shift by the bitwidth.
+ if (match(Op1, m_Undef()))
+ return Op1;
+
+ // Shifting by the bitwidth or more is undefined.
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1))
+ if (CI->getValue().getLimitedValue() >=
+ Op0->getType()->getScalarSizeInBits())
+ return UndefValue::get(Op0->getType());
+
+ // If the operation is with the result of a select instruction, check whether
+ // operating on either branch of the select always yields the same value.
+ if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
+ if (Value *V = ThreadBinOpOverSelect(Opcode, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a phi instruction, check whether
+ // operating on all incoming values of the phi always yields the same value.
+ if (isa<PHINode>(Op0) || isa<PHINode>(Op1))
+ if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ return 0;
+}
+
+/// SimplifyShlInst - Given operands for an Shl, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyShlInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ if (Value *V = SimplifyShift(Instruction::Shl, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ // undef << X -> 0
+ if (match(Op0, m_Undef()))
+ return Constant::getNullValue(Op0->getType());
+
+ // (X >> A) << A -> X
+ Value *X;
+ if (match(Op0, m_Shr(m_Value(X), m_Specific(Op1))) &&
+ cast<PossiblyExactOperator>(Op0)->isExact())
+ return X;
+ return 0;
+}
+
+Value *llvm::SimplifyShlInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifyShlInst(Op0, Op1, isNSW, isNUW, TD, DT, RecursionLimit);
+}
+
+/// SimplifyLShrInst - Given operands for an LShr, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyLShrInst(Value *Op0, Value *Op1, bool isExact,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ if (Value *V = SimplifyShift(Instruction::LShr, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ // undef >>l X -> 0
+ if (match(Op0, m_Undef()))
+ return Constant::getNullValue(Op0->getType());
+
+ // (X << A) >> A -> X
+ Value *X;
+ if (match(Op0, m_Shl(m_Value(X), m_Specific(Op1))) &&
+ cast<OverflowingBinaryOperator>(Op0)->hasNoUnsignedWrap())
+ return X;
+
+ return 0;
+}
+
+Value *llvm::SimplifyLShrInst(Value *Op0, Value *Op1, bool isExact,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifyLShrInst(Op0, Op1, isExact, TD, DT, RecursionLimit);
+}
+
+/// SimplifyAShrInst - Given operands for an AShr, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyAShrInst(Value *Op0, Value *Op1, bool isExact,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ if (Value *V = SimplifyShift(Instruction::AShr, Op0, Op1, TD, DT, MaxRecurse))
+ return V;
+
+ // all ones >>a X -> all ones
+ if (match(Op0, m_AllOnes()))
+ return Op0;
+
+ // undef >>a X -> all ones
+ if (match(Op0, m_Undef()))
+ return Constant::getAllOnesValue(Op0->getType());
+
+ // (X << A) >> A -> X
+ Value *X;
+ if (match(Op0, m_Shl(m_Value(X), m_Specific(Op1))) &&
+ cast<OverflowingBinaryOperator>(Op0)->hasNoSignedWrap())
+ return X;
+
+ return 0;
+}
+
+Value *llvm::SimplifyAShrInst(Value *Op0, Value *Op1, bool isExact,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifyAShrInst(Op0, Op1, isExact, TD, DT, RecursionLimit);
+}
+
/// SimplifyAndInst - Given operands for an And, see if we can
/// fold the result. If not, this returns null.
-Value *llvm::SimplifyAndInst(Value *Op0, Value *Op1, const TargetData *TD) {
+static Value *SimplifyAndInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
if (Constant *CLHS = dyn_cast<Constant>(Op0)) {
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::And, CLHS->getType(),
- Ops, 2, TD);
+ Ops, TD);
}
-
+
// Canonicalize the constant to the RHS.
std::swap(Op0, Op1);
}
-
+
// X & undef -> 0
- if (isa<UndefValue>(Op1))
+ if (match(Op1, m_Undef()))
return Constant::getNullValue(Op0->getType());
-
+
// X & X = X
if (Op0 == Op1)
return Op0;
-
- // X & <0,0> = <0,0>
- if (isa<ConstantAggregateZero>(Op1))
+
+ // X & 0 = 0
+ if (match(Op1, m_Zero()))
return Op1;
-
- // X & <-1,-1> = X
- if (ConstantVector *CP = dyn_cast<ConstantVector>(Op1))
- if (CP->isAllOnesValue())
- return Op0;
-
- if (ConstantInt *Op1CI = dyn_cast<ConstantInt>(Op1)) {
- // X & 0 = 0
- if (Op1CI->isZero())
- return Op1CI;
- // X & -1 = X
- if (Op1CI->isAllOnesValue())
- return Op0;
- }
-
+
+ // X & -1 = X
+ if (match(Op1, m_AllOnes()))
+ return Op0;
+
// A & ~A = ~A & A = 0
- Value *A, *B;
- if ((match(Op0, m_Not(m_Value(A))) && A == Op1) ||
- (match(Op1, m_Not(m_Value(A))) && A == Op0))
+ if (match(Op0, m_Not(m_Specific(Op1))) ||
+ match(Op1, m_Not(m_Specific(Op0))))
return Constant::getNullValue(Op0->getType());
-
+
// (A | ?) & A = A
+ Value *A = 0, *B = 0;
if (match(Op0, m_Or(m_Value(A), m_Value(B))) &&
(A == Op1 || B == Op1))
return Op1;
-
+
// A & (A | ?) = A
if (match(Op1, m_Or(m_Value(A), m_Value(B))) &&
(A == Op0 || B == Op0))
return Op0;
-
+
+ // Try some generic simplifications for associative operations.
+ if (Value *V = SimplifyAssociativeBinOp(Instruction::And, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // And distributes over Or. Try some generic simplifications based on this.
+ if (Value *V = ExpandBinOp(Instruction::And, Op0, Op1, Instruction::Or,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // And distributes over Xor. Try some generic simplifications based on this.
+ if (Value *V = ExpandBinOp(Instruction::And, Op0, Op1, Instruction::Xor,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // Or distributes over And. Try some generic simplifications based on this.
+ if (Value *V = FactorizeBinOp(Instruction::And, Op0, Op1, Instruction::Or,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a select instruction, check whether
+ // operating on either branch of the select always yields the same value.
+ if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
+ if (Value *V = ThreadBinOpOverSelect(Instruction::And, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a phi instruction, check whether
+ // operating on all incoming values of the phi always yields the same value.
+ if (isa<PHINode>(Op0) || isa<PHINode>(Op1))
+ if (Value *V = ThreadBinOpOverPHI(Instruction::And, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
return 0;
}
+Value *llvm::SimplifyAndInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifyAndInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
/// SimplifyOrInst - Given operands for an Or, see if we can
/// fold the result. If not, this returns null.
-Value *llvm::SimplifyOrInst(Value *Op0, Value *Op1, const TargetData *TD) {
+static Value *SimplifyOrInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
if (Constant *CLHS = dyn_cast<Constant>(Op0)) {
if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
Constant *Ops[] = { CLHS, CRHS };
return ConstantFoldInstOperands(Instruction::Or, CLHS->getType(),
- Ops, 2, TD);
+ Ops, TD);
}
-
+
// Canonicalize the constant to the RHS.
std::swap(Op0, Op1);
}
-
+
// X | undef -> -1
- if (isa<UndefValue>(Op1))
+ if (match(Op1, m_Undef()))
return Constant::getAllOnesValue(Op0->getType());
-
+
// X | X = X
if (Op0 == Op1)
return Op0;
- // X | <0,0> = X
- if (isa<ConstantAggregateZero>(Op1))
+ // X | 0 = X
+ if (match(Op1, m_Zero()))
return Op0;
-
- // X | <-1,-1> = <-1,-1>
- if (ConstantVector *CP = dyn_cast<ConstantVector>(Op1))
- if (CP->isAllOnesValue())
- return Op1;
-
- if (ConstantInt *Op1CI = dyn_cast<ConstantInt>(Op1)) {
- // X | 0 = X
- if (Op1CI->isZero())
- return Op0;
- // X | -1 = -1
- if (Op1CI->isAllOnesValue())
- return Op1CI;
- }
-
+
+ // X | -1 = -1
+ if (match(Op1, m_AllOnes()))
+ return Op1;
+
// A | ~A = ~A | A = -1
- Value *A, *B;
- if ((match(Op0, m_Not(m_Value(A))) && A == Op1) ||
- (match(Op1, m_Not(m_Value(A))) && A == Op0))
+ if (match(Op0, m_Not(m_Specific(Op1))) ||
+ match(Op1, m_Not(m_Specific(Op0))))
return Constant::getAllOnesValue(Op0->getType());
-
+
// (A & ?) | A = A
+ Value *A = 0, *B = 0;
if (match(Op0, m_And(m_Value(A), m_Value(B))) &&
(A == Op1 || B == Op1))
return Op1;
-
+
// A | (A & ?) = A
if (match(Op1, m_And(m_Value(A), m_Value(B))) &&
(A == Op0 || B == Op0))
return Op0;
-
+
+ // ~(A & ?) | A = -1
+ if (match(Op0, m_Not(m_And(m_Value(A), m_Value(B)))) &&
+ (A == Op1 || B == Op1))
+ return Constant::getAllOnesValue(Op1->getType());
+
+ // A | ~(A & ?) = -1
+ if (match(Op1, m_Not(m_And(m_Value(A), m_Value(B)))) &&
+ (A == Op0 || B == Op0))
+ return Constant::getAllOnesValue(Op0->getType());
+
+ // Try some generic simplifications for associative operations.
+ if (Value *V = SimplifyAssociativeBinOp(Instruction::Or, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // Or distributes over And. Try some generic simplifications based on this.
+ if (Value *V = ExpandBinOp(Instruction::Or, Op0, Op1, Instruction::And,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // And distributes over Or. Try some generic simplifications based on this.
+ if (Value *V = FactorizeBinOp(Instruction::Or, Op0, Op1, Instruction::And,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a select instruction, check whether
+ // operating on either branch of the select always yields the same value.
+ if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1))
+ if (Value *V = ThreadBinOpOverSelect(Instruction::Or, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a phi instruction, check whether
+ // operating on all incoming values of the phi always yields the same value.
+ if (isa<PHINode>(Op0) || isa<PHINode>(Op1))
+ if (Value *V = ThreadBinOpOverPHI(Instruction::Or, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
return 0;
}
+Value *llvm::SimplifyOrInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifyOrInst(Op0, Op1, TD, DT, RecursionLimit);
+}
-static const Type *GetCompareTy(Value *Op) {
+/// SimplifyXorInst - Given operands for a Xor, see if we can
+/// fold the result. If not, this returns null.
+static Value *SimplifyXorInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT, unsigned MaxRecurse) {
+ if (Constant *CLHS = dyn_cast<Constant>(Op0)) {
+ if (Constant *CRHS = dyn_cast<Constant>(Op1)) {
+ Constant *Ops[] = { CLHS, CRHS };
+ return ConstantFoldInstOperands(Instruction::Xor, CLHS->getType(),
+ Ops, TD);
+ }
+
+ // Canonicalize the constant to the RHS.
+ std::swap(Op0, Op1);
+ }
+
+ // A ^ undef -> undef
+ if (match(Op1, m_Undef()))
+ return Op1;
+
+ // A ^ 0 = A
+ if (match(Op1, m_Zero()))
+ return Op0;
+
+ // A ^ A = 0
+ if (Op0 == Op1)
+ return Constant::getNullValue(Op0->getType());
+
+ // A ^ ~A = ~A ^ A = -1
+ if (match(Op0, m_Not(m_Specific(Op1))) ||
+ match(Op1, m_Not(m_Specific(Op0))))
+ return Constant::getAllOnesValue(Op0->getType());
+
+ // Try some generic simplifications for associative operations.
+ if (Value *V = SimplifyAssociativeBinOp(Instruction::Xor, Op0, Op1, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // And distributes over Xor. Try some generic simplifications based on this.
+ if (Value *V = FactorizeBinOp(Instruction::Xor, Op0, Op1, Instruction::And,
+ TD, DT, MaxRecurse))
+ return V;
+
+ // Threading Xor over selects and phi nodes is pointless, so don't bother.
+ // Threading over the select in "A ^ select(cond, B, C)" means evaluating
+ // "A^B" and "A^C" and seeing if they are equal; but they are equal if and
+ // only if B and C are equal. If B and C are equal then (since we assume
+ // that operands have already been simplified) "select(cond, B, C)" should
+ // have been simplified to the common value of B and C already. Analysing
+ // "A^B" and "A^C" thus gains nothing, but costs compile time. Similarly
+ // for threading over phi nodes.
+
+ return 0;
+}
+
+Value *llvm::SimplifyXorInst(Value *Op0, Value *Op1, const TargetData *TD,
+ const DominatorTree *DT) {
+ return ::SimplifyXorInst(Op0, Op1, TD, DT, RecursionLimit);
+}
+
+static Type *GetCompareTy(Value *Op) {
return CmpInst::makeCmpResultType(Op->getType());
}
+/// ExtractEquivalentCondition - Rummage around inside V looking for something
+/// equivalent to the comparison "LHS Pred RHS". Return such a value if found,
+/// otherwise return null. Helper function for analyzing max/min idioms.
+static Value *ExtractEquivalentCondition(Value *V, CmpInst::Predicate Pred,
+ Value *LHS, Value *RHS) {
+ SelectInst *SI = dyn_cast<SelectInst>(V);
+ if (!SI)
+ return 0;
+ CmpInst *Cmp = dyn_cast<CmpInst>(SI->getCondition());
+ if (!Cmp)
+ return 0;
+ Value *CmpLHS = Cmp->getOperand(0), *CmpRHS = Cmp->getOperand(1);
+ if (Pred == Cmp->getPredicate() && LHS == CmpLHS && RHS == CmpRHS)
+ return Cmp;
+ if (Pred == CmpInst::getSwappedPredicate(Cmp->getPredicate()) &&
+ LHS == CmpRHS && RHS == CmpLHS)
+ return Cmp;
+ return 0;
+}
/// SimplifyICmpInst - Given operands for an ICmpInst, see if we can
/// fold the result. If not, this returns null.
-Value *llvm::SimplifyICmpInst(unsigned Predicate, Value *LHS, Value *RHS,
- const TargetData *TD) {
+static Value *SimplifyICmpInst(unsigned Predicate, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
CmpInst::Predicate Pred = (CmpInst::Predicate)Predicate;
assert(CmpInst::isIntPredicate(Pred) && "Not an integer compare!");
-
+
if (Constant *CLHS = dyn_cast<Constant>(LHS)) {
if (Constant *CRHS = dyn_cast<Constant>(RHS))
return ConstantFoldCompareInstOperands(Pred, CLHS, CRHS, TD);
@@ -189,70 +1432,701 @@
std::swap(LHS, RHS);
Pred = CmpInst::getSwappedPredicate(Pred);
}
-
- // ITy - This is the return type of the compare we're considering.
- const Type *ITy = GetCompareTy(LHS);
-
+
+ Type *ITy = GetCompareTy(LHS); // The return type.
+ Type *OpTy = LHS->getType(); // The operand type.
+
// icmp X, X -> true/false
// X icmp undef -> true/false. For example, icmp ugt %X, undef -> false
// because X could be 0.
if (LHS == RHS || isa<UndefValue>(RHS))
return ConstantInt::get(ITy, CmpInst::isTrueWhenEqual(Pred));
-
- // icmp <global/alloca*/null>, <global/alloca*/null> - Global/Stack value
- // addresses never equal each other! We already know that Op0 != Op1.
- if ((isa<GlobalValue>(LHS) || isa<AllocaInst>(LHS) ||
- isa<ConstantPointerNull>(LHS)) &&
- (isa<GlobalValue>(RHS) || isa<AllocaInst>(RHS) ||
- isa<ConstantPointerNull>(RHS)))
- return ConstantInt::get(ITy, CmpInst::isFalseWhenEqual(Pred));
-
- // See if we are doing a comparison with a constant.
- if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
- // If we have an icmp le or icmp ge instruction, turn it into the
- // appropriate icmp lt or icmp gt instruction. This allows us to rely on
- // them being folded in the code below.
+
+ // Special case logic when the operands have i1 type.
+ if (OpTy->isIntegerTy(1) || (OpTy->isVectorTy() &&
+ cast<VectorType>(OpTy)->getElementType()->isIntegerTy(1))) {
switch (Pred) {
default: break;
- case ICmpInst::ICMP_ULE:
- if (CI->isMaxValue(false)) // A <=u MAX -> TRUE
- return ConstantInt::getTrue(CI->getContext());
+ case ICmpInst::ICMP_EQ:
+ // X == 1 -> X
+ if (match(RHS, m_One()))
+ return LHS;
break;
- case ICmpInst::ICMP_SLE:
- if (CI->isMaxValue(true)) // A <=s MAX -> TRUE
- return ConstantInt::getTrue(CI->getContext());
+ case ICmpInst::ICMP_NE:
+ // X != 0 -> X
+ if (match(RHS, m_Zero()))
+ return LHS;
+ break;
+ case ICmpInst::ICMP_UGT:
+ // X >u 0 -> X
+ if (match(RHS, m_Zero()))
+ return LHS;
break;
case ICmpInst::ICMP_UGE:
- if (CI->isMinValue(false)) // A >=u MIN -> TRUE
- return ConstantInt::getTrue(CI->getContext());
+ // X >=u 1 -> X
+ if (match(RHS, m_One()))
+ return LHS;
break;
- case ICmpInst::ICMP_SGE:
- if (CI->isMinValue(true)) // A >=s MIN -> TRUE
- return ConstantInt::getTrue(CI->getContext());
+ case ICmpInst::ICMP_SLT:
+ // X <s 0 -> X
+ if (match(RHS, m_Zero()))
+ return LHS;
+ break;
+ case ICmpInst::ICMP_SLE:
+ // X <=s -1 -> X
+ if (match(RHS, m_One()))
+ return LHS;
break;
}
}
-
-
+
+ // icmp <alloca*>, <global/alloca*/null> - Different stack variables have
+ // different addresses, and what's more the address of a stack variable is
+ // never null or equal to the address of a global. Note that generalizing
+ // to the case where LHS is a global variable address or null is pointless,
+ // since if both LHS and RHS are constants then we already constant folded
+ // the compare, and if only one of them is then we moved it to RHS already.
+ if (isa<AllocaInst>(LHS) && (isa<GlobalValue>(RHS) || isa<AllocaInst>(RHS) ||
+ isa<ConstantPointerNull>(RHS)))
+ // We already know that LHS != RHS.
+ return ConstantInt::get(ITy, CmpInst::isFalseWhenEqual(Pred));
+
+ // If we are comparing with zero then try hard since this is a common case.
+ if (match(RHS, m_Zero())) {
+ bool LHSKnownNonNegative, LHSKnownNegative;
+ switch (Pred) {
+ default:
+ assert(false && "Unknown ICmp predicate!");
+ case ICmpInst::ICMP_ULT:
+ return getFalse(ITy);
+ case ICmpInst::ICMP_UGE:
+ return getTrue(ITy);
+ case ICmpInst::ICMP_EQ:
+ case ICmpInst::ICMP_ULE:
+ if (isKnownNonZero(LHS, TD))
+ return getFalse(ITy);
+ break;
+ case ICmpInst::ICMP_NE:
+ case ICmpInst::ICMP_UGT:
+ if (isKnownNonZero(LHS, TD))
+ return getTrue(ITy);
+ break;
+ case ICmpInst::ICMP_SLT:
+ ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, TD);
+ if (LHSKnownNegative)
+ return getTrue(ITy);
+ if (LHSKnownNonNegative)
+ return getFalse(ITy);
+ break;
+ case ICmpInst::ICMP_SLE:
+ ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, TD);
+ if (LHSKnownNegative)
+ return getTrue(ITy);
+ if (LHSKnownNonNegative && isKnownNonZero(LHS, TD))
+ return getFalse(ITy);
+ break;
+ case ICmpInst::ICMP_SGE:
+ ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, TD);
+ if (LHSKnownNegative)
+ return getFalse(ITy);
+ if (LHSKnownNonNegative)
+ return getTrue(ITy);
+ break;
+ case ICmpInst::ICMP_SGT:
+ ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, TD);
+ if (LHSKnownNegative)
+ return getFalse(ITy);
+ if (LHSKnownNonNegative && isKnownNonZero(LHS, TD))
+ return getTrue(ITy);
+ break;
+ }
+ }
+
+ // See if we are doing a comparison with a constant integer.
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
+ // Rule out tautological comparisons (eg., ult 0 or uge 0).
+ ConstantRange RHS_CR = ICmpInst::makeConstantRange(Pred, CI->getValue());
+ if (RHS_CR.isEmptySet())
+ return ConstantInt::getFalse(CI->getContext());
+ if (RHS_CR.isFullSet())
+ return ConstantInt::getTrue(CI->getContext());
+
+ // Many binary operators with constant RHS have easy to compute constant
+ // range. Use them to check whether the comparison is a tautology.
+ uint32_t Width = CI->getBitWidth();
+ APInt Lower = APInt(Width, 0);
+ APInt Upper = APInt(Width, 0);
+ ConstantInt *CI2;
+ if (match(LHS, m_URem(m_Value(), m_ConstantInt(CI2)))) {
+ // 'urem x, CI2' produces [0, CI2).
+ Upper = CI2->getValue();
+ } else if (match(LHS, m_SRem(m_Value(), m_ConstantInt(CI2)))) {
+ // 'srem x, CI2' produces (-|CI2|, |CI2|).
+ Upper = CI2->getValue().abs();
+ Lower = (-Upper) + 1;
+ } else if (match(LHS, m_UDiv(m_Value(), m_ConstantInt(CI2)))) {
+ // 'udiv x, CI2' produces [0, UINT_MAX / CI2].
+ APInt NegOne = APInt::getAllOnesValue(Width);
+ if (!CI2->isZero())
+ Upper = NegOne.udiv(CI2->getValue()) + 1;
+ } else if (match(LHS, m_SDiv(m_Value(), m_ConstantInt(CI2)))) {
+ // 'sdiv x, CI2' produces [INT_MIN / CI2, INT_MAX / CI2].
+ APInt IntMin = APInt::getSignedMinValue(Width);
+ APInt IntMax = APInt::getSignedMaxValue(Width);
+ APInt Val = CI2->getValue().abs();
+ if (!Val.isMinValue()) {
+ Lower = IntMin.sdiv(Val);
+ Upper = IntMax.sdiv(Val) + 1;
+ }
+ } else if (match(LHS, m_LShr(m_Value(), m_ConstantInt(CI2)))) {
+ // 'lshr x, CI2' produces [0, UINT_MAX >> CI2].
+ APInt NegOne = APInt::getAllOnesValue(Width);
+ if (CI2->getValue().ult(Width))
+ Upper = NegOne.lshr(CI2->getValue()) + 1;
+ } else if (match(LHS, m_AShr(m_Value(), m_ConstantInt(CI2)))) {
+ // 'ashr x, CI2' produces [INT_MIN >> CI2, INT_MAX >> CI2].
+ APInt IntMin = APInt::getSignedMinValue(Width);
+ APInt IntMax = APInt::getSignedMaxValue(Width);
+ if (CI2->getValue().ult(Width)) {
+ Lower = IntMin.ashr(CI2->getValue());
+ Upper = IntMax.ashr(CI2->getValue()) + 1;
+ }
+ } else if (match(LHS, m_Or(m_Value(), m_ConstantInt(CI2)))) {
+ // 'or x, CI2' produces [CI2, UINT_MAX].
+ Lower = CI2->getValue();
+ } else if (match(LHS, m_And(m_Value(), m_ConstantInt(CI2)))) {
+ // 'and x, CI2' produces [0, CI2].
+ Upper = CI2->getValue() + 1;
+ }
+ if (Lower != Upper) {
+ ConstantRange LHS_CR = ConstantRange(Lower, Upper);
+ if (RHS_CR.contains(LHS_CR))
+ return ConstantInt::getTrue(RHS->getContext());
+ if (RHS_CR.inverse().contains(LHS_CR))
+ return ConstantInt::getFalse(RHS->getContext());
+ }
+ }
+
+ // Compare of cast, for example (zext X) != 0 -> X != 0
+ if (isa<CastInst>(LHS) && (isa<Constant>(RHS) || isa<CastInst>(RHS))) {
+ Instruction *LI = cast<CastInst>(LHS);
+ Value *SrcOp = LI->getOperand(0);
+ Type *SrcTy = SrcOp->getType();
+ Type *DstTy = LI->getType();
+
+ // Turn icmp (ptrtoint x), (ptrtoint/constant) into a compare of the input
+ // if the integer type is the same size as the pointer type.
+ if (MaxRecurse && TD && isa<PtrToIntInst>(LI) &&
+ TD->getPointerSizeInBits() == DstTy->getPrimitiveSizeInBits()) {
+ if (Constant *RHSC = dyn_cast<Constant>(RHS)) {
+ // Transfer the cast to the constant.
+ if (Value *V = SimplifyICmpInst(Pred, SrcOp,
+ ConstantExpr::getIntToPtr(RHSC, SrcTy),
+ TD, DT, MaxRecurse-1))
+ return V;
+ } else if (PtrToIntInst *RI = dyn_cast<PtrToIntInst>(RHS)) {
+ if (RI->getOperand(0)->getType() == SrcTy)
+ // Compare without the cast.
+ if (Value *V = SimplifyICmpInst(Pred, SrcOp, RI->getOperand(0),
+ TD, DT, MaxRecurse-1))
+ return V;
+ }
+ }
+
+ if (isa<ZExtInst>(LHS)) {
+ // Turn icmp (zext X), (zext Y) into a compare of X and Y if they have the
+ // same type.
+ if (ZExtInst *RI = dyn_cast<ZExtInst>(RHS)) {
+ if (MaxRecurse && SrcTy == RI->getOperand(0)->getType())
+ // Compare X and Y. Note that signed predicates become unsigned.
+ if (Value *V = SimplifyICmpInst(ICmpInst::getUnsignedPredicate(Pred),
+ SrcOp, RI->getOperand(0), TD, DT,
+ MaxRecurse-1))
+ return V;
+ }
+ // Turn icmp (zext X), Cst into a compare of X and Cst if Cst is extended
+ // too. If not, then try to deduce the result of the comparison.
+ else if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
+ // Compute the constant that would happen if we truncated to SrcTy then
+ // reextended to DstTy.
+ Constant *Trunc = ConstantExpr::getTrunc(CI, SrcTy);
+ Constant *RExt = ConstantExpr::getCast(CastInst::ZExt, Trunc, DstTy);
+
+ // If the re-extended constant didn't change then this is effectively
+ // also a case of comparing two zero-extended values.
+ if (RExt == CI && MaxRecurse)
+ if (Value *V = SimplifyICmpInst(ICmpInst::getUnsignedPredicate(Pred),
+ SrcOp, Trunc, TD, DT, MaxRecurse-1))
+ return V;
+
+ // Otherwise the upper bits of LHS are zero while RHS has a non-zero bit
+ // there. Use this to work out the result of the comparison.
+ if (RExt != CI) {
+ switch (Pred) {
+ default:
+ assert(false && "Unknown ICmp predicate!");
+ // LHS <u RHS.
+ case ICmpInst::ICMP_EQ:
+ case ICmpInst::ICMP_UGT:
+ case ICmpInst::ICMP_UGE:
+ return ConstantInt::getFalse(CI->getContext());
+
+ case ICmpInst::ICMP_NE:
+ case ICmpInst::ICMP_ULT:
+ case ICmpInst::ICMP_ULE:
+ return ConstantInt::getTrue(CI->getContext());
+
+ // LHS is non-negative. If RHS is negative then LHS >s LHS. If RHS
+ // is non-negative then LHS <s RHS.
+ case ICmpInst::ICMP_SGT:
+ case ICmpInst::ICMP_SGE:
+ return CI->getValue().isNegative() ?
+ ConstantInt::getTrue(CI->getContext()) :
+ ConstantInt::getFalse(CI->getContext());
+
+ case ICmpInst::ICMP_SLT:
+ case ICmpInst::ICMP_SLE:
+ return CI->getValue().isNegative() ?
+ ConstantInt::getFalse(CI->getContext()) :
+ ConstantInt::getTrue(CI->getContext());
+ }
+ }
+ }
+ }
+
+ if (isa<SExtInst>(LHS)) {
+ // Turn icmp (sext X), (sext Y) into a compare of X and Y if they have the
+ // same type.
+ if (SExtInst *RI = dyn_cast<SExtInst>(RHS)) {
+ if (MaxRecurse && SrcTy == RI->getOperand(0)->getType())
+ // Compare X and Y. Note that the predicate does not change.
+ if (Value *V = SimplifyICmpInst(Pred, SrcOp, RI->getOperand(0),
+ TD, DT, MaxRecurse-1))
+ return V;
+ }
+ // Turn icmp (sext X), Cst into a compare of X and Cst if Cst is extended
+ // too. If not, then try to deduce the result of the comparison.
+ else if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
+ // Compute the constant that would happen if we truncated to SrcTy then
+ // reextended to DstTy.
+ Constant *Trunc = ConstantExpr::getTrunc(CI, SrcTy);
+ Constant *RExt = ConstantExpr::getCast(CastInst::SExt, Trunc, DstTy);
+
+ // If the re-extended constant didn't change then this is effectively
+ // also a case of comparing two sign-extended values.
+ if (RExt == CI && MaxRecurse)
+ if (Value *V = SimplifyICmpInst(Pred, SrcOp, Trunc, TD, DT,
+ MaxRecurse-1))
+ return V;
+
+ // Otherwise the upper bits of LHS are all equal, while RHS has varying
+ // bits there. Use this to work out the result of the comparison.
+ if (RExt != CI) {
+ switch (Pred) {
+ default:
+ assert(false && "Unknown ICmp predicate!");
+ case ICmpInst::ICMP_EQ:
+ return ConstantInt::getFalse(CI->getContext());
+ case ICmpInst::ICMP_NE:
+ return ConstantInt::getTrue(CI->getContext());
+
+ // If RHS is non-negative then LHS <s RHS. If RHS is negative then
+ // LHS >s RHS.
+ case ICmpInst::ICMP_SGT:
+ case ICmpInst::ICMP_SGE:
+ return CI->getValue().isNegative() ?
+ ConstantInt::getTrue(CI->getContext()) :
+ ConstantInt::getFalse(CI->getContext());
+ case ICmpInst::ICMP_SLT:
+ case ICmpInst::ICMP_SLE:
+ return CI->getValue().isNegative() ?
+ ConstantInt::getFalse(CI->getContext()) :
+ ConstantInt::getTrue(CI->getContext());
+
+ // If LHS is non-negative then LHS <u RHS. If LHS is negative then
+ // LHS >u RHS.
+ case ICmpInst::ICMP_UGT:
+ case ICmpInst::ICMP_UGE:
+ // Comparison is true iff the LHS <s 0.
+ if (MaxRecurse)
+ if (Value *V = SimplifyICmpInst(ICmpInst::ICMP_SLT, SrcOp,
+ Constant::getNullValue(SrcTy),
+ TD, DT, MaxRecurse-1))
+ return V;
+ break;
+ case ICmpInst::ICMP_ULT:
+ case ICmpInst::ICMP_ULE:
+ // Comparison is true iff the LHS >=s 0.
+ if (MaxRecurse)
+ if (Value *V = SimplifyICmpInst(ICmpInst::ICMP_SGE, SrcOp,
+ Constant::getNullValue(SrcTy),
+ TD, DT, MaxRecurse-1))
+ return V;
+ break;
+ }
+ }
+ }
+ }
+ }
+
+ // Special logic for binary operators.
+ BinaryOperator *LBO = dyn_cast<BinaryOperator>(LHS);
+ BinaryOperator *RBO = dyn_cast<BinaryOperator>(RHS);
+ if (MaxRecurse && (LBO || RBO)) {
+ // Analyze the case when either LHS or RHS is an add instruction.
+ Value *A = 0, *B = 0, *C = 0, *D = 0;
+ // LHS = A + B (or A and B are null); RHS = C + D (or C and D are null).
+ bool NoLHSWrapProblem = false, NoRHSWrapProblem = false;
+ if (LBO && LBO->getOpcode() == Instruction::Add) {
+ A = LBO->getOperand(0); B = LBO->getOperand(1);
+ NoLHSWrapProblem = ICmpInst::isEquality(Pred) ||
+ (CmpInst::isUnsigned(Pred) && LBO->hasNoUnsignedWrap()) ||
+ (CmpInst::isSigned(Pred) && LBO->hasNoSignedWrap());
+ }
+ if (RBO && RBO->getOpcode() == Instruction::Add) {
+ C = RBO->getOperand(0); D = RBO->getOperand(1);
+ NoRHSWrapProblem = ICmpInst::isEquality(Pred) ||
+ (CmpInst::isUnsigned(Pred) && RBO->hasNoUnsignedWrap()) ||
+ (CmpInst::isSigned(Pred) && RBO->hasNoSignedWrap());
+ }
+
+ // icmp (X+Y), X -> icmp Y, 0 for equalities or if there is no overflow.
+ if ((A == RHS || B == RHS) && NoLHSWrapProblem)
+ if (Value *V = SimplifyICmpInst(Pred, A == RHS ? B : A,
+ Constant::getNullValue(RHS->getType()),
+ TD, DT, MaxRecurse-1))
+ return V;
+
+ // icmp X, (X+Y) -> icmp 0, Y for equalities or if there is no overflow.
+ if ((C == LHS || D == LHS) && NoRHSWrapProblem)
+ if (Value *V = SimplifyICmpInst(Pred,
+ Constant::getNullValue(LHS->getType()),
+ C == LHS ? D : C, TD, DT, MaxRecurse-1))
+ return V;
+
+ // icmp (X+Y), (X+Z) -> icmp Y,Z for equalities or if there is no overflow.
+ if (A && C && (A == C || A == D || B == C || B == D) &&
+ NoLHSWrapProblem && NoRHSWrapProblem) {
+ // Determine Y and Z in the form icmp (X+Y), (X+Z).
+ Value *Y = (A == C || A == D) ? B : A;
+ Value *Z = (C == A || C == B) ? D : C;
+ if (Value *V = SimplifyICmpInst(Pred, Y, Z, TD, DT, MaxRecurse-1))
+ return V;
+ }
+ }
+
+ if (LBO && match(LBO, m_URem(m_Value(), m_Specific(RHS)))) {
+ bool KnownNonNegative, KnownNegative;
+ switch (Pred) {
+ default:
+ break;
+ case ICmpInst::ICMP_SGT:
+ case ICmpInst::ICMP_SGE:
+ ComputeSignBit(LHS, KnownNonNegative, KnownNegative, TD);
+ if (!KnownNonNegative)
+ break;
+ // fall-through
+ case ICmpInst::ICMP_EQ:
+ case ICmpInst::ICMP_UGT:
+ case ICmpInst::ICMP_UGE:
+ return getFalse(ITy);
+ case ICmpInst::ICMP_SLT:
+ case ICmpInst::ICMP_SLE:
+ ComputeSignBit(LHS, KnownNonNegative, KnownNegative, TD);
+ if (!KnownNonNegative)
+ break;
+ // fall-through
+ case ICmpInst::ICMP_NE:
+ case ICmpInst::ICMP_ULT:
+ case ICmpInst::ICMP_ULE:
+ return getTrue(ITy);
+ }
+ }
+ if (RBO && match(RBO, m_URem(m_Value(), m_Specific(LHS)))) {
+ bool KnownNonNegative, KnownNegative;
+ switch (Pred) {
+ default:
+ break;
+ case ICmpInst::ICMP_SGT:
+ case ICmpInst::ICMP_SGE:
+ ComputeSignBit(RHS, KnownNonNegative, KnownNegative, TD);
+ if (!KnownNonNegative)
+ break;
+ // fall-through
+ case ICmpInst::ICMP_NE:
+ case ICmpInst::ICMP_UGT:
+ case ICmpInst::ICMP_UGE:
+ return getTrue(ITy);
+ case ICmpInst::ICMP_SLT:
+ case ICmpInst::ICMP_SLE:
+ ComputeSignBit(RHS, KnownNonNegative, KnownNegative, TD);
+ if (!KnownNonNegative)
+ break;
+ // fall-through
+ case ICmpInst::ICMP_EQ:
+ case ICmpInst::ICMP_ULT:
+ case ICmpInst::ICMP_ULE:
+ return getFalse(ITy);
+ }
+ }
+
+ if (MaxRecurse && LBO && RBO && LBO->getOpcode() == RBO->getOpcode() &&
+ LBO->getOperand(1) == RBO->getOperand(1)) {
+ switch (LBO->getOpcode()) {
+ default: break;
+ case Instruction::UDiv:
+ case Instruction::LShr:
+ if (ICmpInst::isSigned(Pred))
+ break;
+ // fall-through
+ case Instruction::SDiv:
+ case Instruction::AShr:
+ if (!LBO->isExact() || !RBO->isExact())
+ break;
+ if (Value *V = SimplifyICmpInst(Pred, LBO->getOperand(0),
+ RBO->getOperand(0), TD, DT, MaxRecurse-1))
+ return V;
+ break;
+ case Instruction::Shl: {
+ bool NUW = LBO->hasNoUnsignedWrap() && RBO->hasNoUnsignedWrap();
+ bool NSW = LBO->hasNoSignedWrap() && RBO->hasNoSignedWrap();
+ if (!NUW && !NSW)
+ break;
+ if (!NSW && ICmpInst::isSigned(Pred))
+ break;
+ if (Value *V = SimplifyICmpInst(Pred, LBO->getOperand(0),
+ RBO->getOperand(0), TD, DT, MaxRecurse-1))
+ return V;
+ break;
+ }
+ }
+ }
+
+ // Simplify comparisons involving max/min.
+ Value *A, *B;
+ CmpInst::Predicate P = CmpInst::BAD_ICMP_PREDICATE;
+ CmpInst::Predicate EqP; // Chosen so that "A == max/min(A,B)" iff "A EqP B".
+
+ // Signed variants on "max(a,b)>=a -> true".
+ if (match(LHS, m_SMax(m_Value(A), m_Value(B))) && (A == RHS || B == RHS)) {
+ if (A != RHS) std::swap(A, B); // smax(A, B) pred A.
+ EqP = CmpInst::ICMP_SGE; // "A == smax(A, B)" iff "A sge B".
+ // We analyze this as smax(A, B) pred A.
+ P = Pred;
+ } else if (match(RHS, m_SMax(m_Value(A), m_Value(B))) &&
+ (A == LHS || B == LHS)) {
+ if (A != LHS) std::swap(A, B); // A pred smax(A, B).
+ EqP = CmpInst::ICMP_SGE; // "A == smax(A, B)" iff "A sge B".
+ // We analyze this as smax(A, B) swapped-pred A.
+ P = CmpInst::getSwappedPredicate(Pred);
+ } else if (match(LHS, m_SMin(m_Value(A), m_Value(B))) &&
+ (A == RHS || B == RHS)) {
+ if (A != RHS) std::swap(A, B); // smin(A, B) pred A.
+ EqP = CmpInst::ICMP_SLE; // "A == smin(A, B)" iff "A sle B".
+ // We analyze this as smax(-A, -B) swapped-pred -A.
+ // Note that we do not need to actually form -A or -B thanks to EqP.
+ P = CmpInst::getSwappedPredicate(Pred);
+ } else if (match(RHS, m_SMin(m_Value(A), m_Value(B))) &&
+ (A == LHS || B == LHS)) {
+ if (A != LHS) std::swap(A, B); // A pred smin(A, B).
+ EqP = CmpInst::ICMP_SLE; // "A == smin(A, B)" iff "A sle B".
+ // We analyze this as smax(-A, -B) pred -A.
+ // Note that we do not need to actually form -A or -B thanks to EqP.
+ P = Pred;
+ }
+ if (P != CmpInst::BAD_ICMP_PREDICATE) {
+ // Cases correspond to "max(A, B) p A".
+ switch (P) {
+ default:
+ break;
+ case CmpInst::ICMP_EQ:
+ case CmpInst::ICMP_SLE:
+ // Equivalent to "A EqP B". This may be the same as the condition tested
+ // in the max/min; if so, we can just return that.
+ if (Value *V = ExtractEquivalentCondition(LHS, EqP, A, B))
+ return V;
+ if (Value *V = ExtractEquivalentCondition(RHS, EqP, A, B))
+ return V;
+ // Otherwise, see if "A EqP B" simplifies.
+ if (MaxRecurse)
+ if (Value *V = SimplifyICmpInst(EqP, A, B, TD, DT, MaxRecurse-1))
+ return V;
+ break;
+ case CmpInst::ICMP_NE:
+ case CmpInst::ICMP_SGT: {
+ CmpInst::Predicate InvEqP = CmpInst::getInversePredicate(EqP);
+ // Equivalent to "A InvEqP B". This may be the same as the condition
+ // tested in the max/min; if so, we can just return that.
+ if (Value *V = ExtractEquivalentCondition(LHS, InvEqP, A, B))
+ return V;
+ if (Value *V = ExtractEquivalentCondition(RHS, InvEqP, A, B))
+ return V;
+ // Otherwise, see if "A InvEqP B" simplifies.
+ if (MaxRecurse)
+ if (Value *V = SimplifyICmpInst(InvEqP, A, B, TD, DT, MaxRecurse-1))
+ return V;
+ break;
+ }
+ case CmpInst::ICMP_SGE:
+ // Always true.
+ return getTrue(ITy);
+ case CmpInst::ICMP_SLT:
+ // Always false.
+ return getFalse(ITy);
+ }
+ }
+
+ // Unsigned variants on "max(a,b)>=a -> true".
+ P = CmpInst::BAD_ICMP_PREDICATE;
+ if (match(LHS, m_UMax(m_Value(A), m_Value(B))) && (A == RHS || B == RHS)) {
+ if (A != RHS) std::swap(A, B); // umax(A, B) pred A.
+ EqP = CmpInst::ICMP_UGE; // "A == umax(A, B)" iff "A uge B".
+ // We analyze this as umax(A, B) pred A.
+ P = Pred;
+ } else if (match(RHS, m_UMax(m_Value(A), m_Value(B))) &&
+ (A == LHS || B == LHS)) {
+ if (A != LHS) std::swap(A, B); // A pred umax(A, B).
+ EqP = CmpInst::ICMP_UGE; // "A == umax(A, B)" iff "A uge B".
+ // We analyze this as umax(A, B) swapped-pred A.
+ P = CmpInst::getSwappedPredicate(Pred);
+ } else if (match(LHS, m_UMin(m_Value(A), m_Value(B))) &&
+ (A == RHS || B == RHS)) {
+ if (A != RHS) std::swap(A, B); // umin(A, B) pred A.
+ EqP = CmpInst::ICMP_ULE; // "A == umin(A, B)" iff "A ule B".
+ // We analyze this as umax(-A, -B) swapped-pred -A.
+ // Note that we do not need to actually form -A or -B thanks to EqP.
+ P = CmpInst::getSwappedPredicate(Pred);
+ } else if (match(RHS, m_UMin(m_Value(A), m_Value(B))) &&
+ (A == LHS || B == LHS)) {
+ if (A != LHS) std::swap(A, B); // A pred umin(A, B).
+ EqP = CmpInst::ICMP_ULE; // "A == umin(A, B)" iff "A ule B".
+ // We analyze this as umax(-A, -B) pred -A.
+ // Note that we do not need to actually form -A or -B thanks to EqP.
+ P = Pred;
+ }
+ if (P != CmpInst::BAD_ICMP_PREDICATE) {
+ // Cases correspond to "max(A, B) p A".
+ switch (P) {
+ default:
+ break;
+ case CmpInst::ICMP_EQ:
+ case CmpInst::ICMP_ULE:
+ // Equivalent to "A EqP B". This may be the same as the condition tested
+ // in the max/min; if so, we can just return that.
+ if (Value *V = ExtractEquivalentCondition(LHS, EqP, A, B))
+ return V;
+ if (Value *V = ExtractEquivalentCondition(RHS, EqP, A, B))
+ return V;
+ // Otherwise, see if "A EqP B" simplifies.
+ if (MaxRecurse)
+ if (Value *V = SimplifyICmpInst(EqP, A, B, TD, DT, MaxRecurse-1))
+ return V;
+ break;
+ case CmpInst::ICMP_NE:
+ case CmpInst::ICMP_UGT: {
+ CmpInst::Predicate InvEqP = CmpInst::getInversePredicate(EqP);
+ // Equivalent to "A InvEqP B". This may be the same as the condition
+ // tested in the max/min; if so, we can just return that.
+ if (Value *V = ExtractEquivalentCondition(LHS, InvEqP, A, B))
+ return V;
+ if (Value *V = ExtractEquivalentCondition(RHS, InvEqP, A, B))
+ return V;
+ // Otherwise, see if "A InvEqP B" simplifies.
+ if (MaxRecurse)
+ if (Value *V = SimplifyICmpInst(InvEqP, A, B, TD, DT, MaxRecurse-1))
+ return V;
+ break;
+ }
+ case CmpInst::ICMP_UGE:
+ // Always true.
+ return getTrue(ITy);
+ case CmpInst::ICMP_ULT:
+ // Always false.
+ return getFalse(ITy);
+ }
+ }
+
+ // Variants on "max(x,y) >= min(x,z)".
+ Value *C, *D;
+ if (match(LHS, m_SMax(m_Value(A), m_Value(B))) &&
+ match(RHS, m_SMin(m_Value(C), m_Value(D))) &&
+ (A == C || A == D || B == C || B == D)) {
+ // max(x, ?) pred min(x, ?).
+ if (Pred == CmpInst::ICMP_SGE)
+ // Always true.
+ return getTrue(ITy);
+ if (Pred == CmpInst::ICMP_SLT)
+ // Always false.
+ return getFalse(ITy);
+ } else if (match(LHS, m_SMin(m_Value(A), m_Value(B))) &&
+ match(RHS, m_SMax(m_Value(C), m_Value(D))) &&
+ (A == C || A == D || B == C || B == D)) {
+ // min(x, ?) pred max(x, ?).
+ if (Pred == CmpInst::ICMP_SLE)
+ // Always true.
+ return getTrue(ITy);
+ if (Pred == CmpInst::ICMP_SGT)
+ // Always false.
+ return getFalse(ITy);
+ } else if (match(LHS, m_UMax(m_Value(A), m_Value(B))) &&
+ match(RHS, m_UMin(m_Value(C), m_Value(D))) &&
+ (A == C || A == D || B == C || B == D)) {
+ // max(x, ?) pred min(x, ?).
+ if (Pred == CmpInst::ICMP_UGE)
+ // Always true.
+ return getTrue(ITy);
+ if (Pred == CmpInst::ICMP_ULT)
+ // Always false.
+ return getFalse(ITy);
+ } else if (match(LHS, m_UMin(m_Value(A), m_Value(B))) &&
+ match(RHS, m_UMax(m_Value(C), m_Value(D))) &&
+ (A == C || A == D || B == C || B == D)) {
+ // min(x, ?) pred max(x, ?).
+ if (Pred == CmpInst::ICMP_ULE)
+ // Always true.
+ return getTrue(ITy);
+ if (Pred == CmpInst::ICMP_UGT)
+ // Always false.
+ return getFalse(ITy);
+ }
+
+ // If the comparison is with the result of a select instruction, check whether
+ // comparing with either branch of the select always yields the same value.
+ if (isa<SelectInst>(LHS) || isa<SelectInst>(RHS))
+ if (Value *V = ThreadCmpOverSelect(Pred, LHS, RHS, TD, DT, MaxRecurse))
+ return V;
+
+ // If the comparison is with the result of a phi instruction, check whether
+ // doing the compare with each incoming phi value yields a common result.
+ if (isa<PHINode>(LHS) || isa<PHINode>(RHS))
+ if (Value *V = ThreadCmpOverPHI(Pred, LHS, RHS, TD, DT, MaxRecurse))
+ return V;
+
return 0;
}
+Value *llvm::SimplifyICmpInst(unsigned Predicate, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifyICmpInst(Predicate, LHS, RHS, TD, DT, RecursionLimit);
+}
+
/// SimplifyFCmpInst - Given operands for an FCmpInst, see if we can
/// fold the result. If not, this returns null.
-Value *llvm::SimplifyFCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
- const TargetData *TD) {
+static Value *SimplifyFCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
CmpInst::Predicate Pred = (CmpInst::Predicate)Predicate;
assert(CmpInst::isFPPredicate(Pred) && "Not an FP compare!");
if (Constant *CLHS = dyn_cast<Constant>(LHS)) {
if (Constant *CRHS = dyn_cast<Constant>(RHS))
return ConstantFoldCompareInstOperands(Pred, CLHS, CRHS, TD);
-
+
// If we have a constant, make sure it is on the RHS.
std::swap(LHS, RHS);
Pred = CmpInst::getSwappedPredicate(Pred);
}
-
+
// Fold trivial predicates.
if (Pred == FCmpInst::FCMP_FALSE)
return ConstantInt::get(GetCompareTy(LHS), 0);
@@ -269,7 +2143,7 @@
if (CmpInst::isFalseWhenEqual(Pred))
return ConstantInt::get(GetCompareTy(LHS), 0);
}
-
+
// Handle fcmp with constant RHS
if (Constant *RHSC = dyn_cast<Constant>(RHS)) {
// If the constant is a nan, see if we can fold the comparison based on it.
@@ -310,64 +2184,154 @@
}
}
}
-
+
+ // If the comparison is with the result of a select instruction, check whether
+ // comparing with either branch of the select always yields the same value.
+ if (isa<SelectInst>(LHS) || isa<SelectInst>(RHS))
+ if (Value *V = ThreadCmpOverSelect(Pred, LHS, RHS, TD, DT, MaxRecurse))
+ return V;
+
+ // If the comparison is with the result of a phi instruction, check whether
+ // doing the compare with each incoming phi value yields a common result.
+ if (isa<PHINode>(LHS) || isa<PHINode>(RHS))
+ if (Value *V = ThreadCmpOverPHI(Pred, LHS, RHS, TD, DT, MaxRecurse))
+ return V;
+
return 0;
}
+Value *llvm::SimplifyFCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifyFCmpInst(Predicate, LHS, RHS, TD, DT, RecursionLimit);
+}
+
/// SimplifySelectInst - Given operands for a SelectInst, see if we can fold
/// the result. If not, this returns null.
Value *llvm::SimplifySelectInst(Value *CondVal, Value *TrueVal, Value *FalseVal,
- const TargetData *TD) {
+ const TargetData *TD, const DominatorTree *) {
// select true, X, Y -> X
// select false, X, Y -> Y
if (ConstantInt *CB = dyn_cast<ConstantInt>(CondVal))
return CB->getZExtValue() ? TrueVal : FalseVal;
-
+
// select C, X, X -> X
if (TrueVal == FalseVal)
return TrueVal;
-
- if (isa<UndefValue>(TrueVal)) // select C, undef, X -> X
- return FalseVal;
- if (isa<UndefValue>(FalseVal)) // select C, X, undef -> X
- return TrueVal;
+
if (isa<UndefValue>(CondVal)) { // select undef, X, Y -> X or Y
if (isa<Constant>(TrueVal))
return TrueVal;
return FalseVal;
}
-
-
-
+ if (isa<UndefValue>(TrueVal)) // select C, undef, X -> X
+ return FalseVal;
+ if (isa<UndefValue>(FalseVal)) // select C, X, undef -> X
+ return TrueVal;
+
return 0;
}
-
/// SimplifyGEPInst - Given operands for an GetElementPtrInst, see if we can
/// fold the result. If not, this returns null.
-Value *llvm::SimplifyGEPInst(Value *const *Ops, unsigned NumOps,
- const TargetData *TD) {
+Value *llvm::SimplifyGEPInst(ArrayRef<Value *> Ops,
+ const TargetData *TD, const DominatorTree *) {
+ // The type of the GEP pointer operand.
+ PointerType *PtrTy = cast<PointerType>(Ops[0]->getType());
+
// getelementptr P -> P.
- if (NumOps == 1)
+ if (Ops.size() == 1)
return Ops[0];
- // TODO.
- //if (isa<UndefValue>(Ops[0]))
- // return UndefValue::get(GEP.getType());
+ if (isa<UndefValue>(Ops[0])) {
+ // Compute the (pointer) type returned by the GEP instruction.
+ Type *LastType = GetElementPtrInst::getIndexedType(PtrTy, Ops.slice(1));
+ Type *GEPTy = PointerType::get(LastType, PtrTy->getAddressSpace());
+ return UndefValue::get(GEPTy);
+ }
- // getelementptr P, 0 -> P.
- if (NumOps == 2)
+ if (Ops.size() == 2) {
+ // getelementptr P, 0 -> P.
if (ConstantInt *C = dyn_cast<ConstantInt>(Ops[1]))
if (C->isZero())
return Ops[0];
-
+ // getelementptr P, N -> P if P points to a type of zero size.
+ if (TD) {
+ Type *Ty = PtrTy->getElementType();
+ if (Ty->isSized() && TD->getTypeAllocSize(Ty) == 0)
+ return Ops[0];
+ }
+ }
+
// Check to see if this is constant foldable.
- for (unsigned i = 0; i != NumOps; ++i)
+ for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (!isa<Constant>(Ops[i]))
return 0;
-
- return ConstantExpr::getGetElementPtr(cast<Constant>(Ops[0]),
- (Constant *const*)Ops+1, NumOps-1);
+
+ return ConstantExpr::getGetElementPtr(cast<Constant>(Ops[0]), Ops.slice(1));
+}
+
+/// SimplifyInsertValueInst - Given operands for an InsertValueInst, see if we
+/// can fold the result. If not, this returns null.
+Value *llvm::SimplifyInsertValueInst(Value *Agg, Value *Val,
+ ArrayRef<unsigned> Idxs,
+ const TargetData *,
+ const DominatorTree *) {
+ if (Constant *CAgg = dyn_cast<Constant>(Agg))
+ if (Constant *CVal = dyn_cast<Constant>(Val))
+ return ConstantFoldInsertValueInstruction(CAgg, CVal, Idxs);
+
+ // insertvalue x, undef, n -> x
+ if (match(Val, m_Undef()))
+ return Agg;
+
+ // insertvalue x, (extractvalue y, n), n
+ if (ExtractValueInst *EV = dyn_cast<ExtractValueInst>(Val))
+ if (EV->getAggregateOperand()->getType() == Agg->getType() &&
+ EV->getIndices() == Idxs) {
+ // insertvalue undef, (extractvalue y, n), n -> y
+ if (match(Agg, m_Undef()))
+ return EV->getAggregateOperand();
+
+ // insertvalue y, (extractvalue y, n), n -> y
+ if (Agg == EV->getAggregateOperand())
+ return Agg;
+ }
+
+ return 0;
+}
+
+/// SimplifyPHINode - See if we can fold the given phi. If not, returns null.
+static Value *SimplifyPHINode(PHINode *PN, const DominatorTree *DT) {
+ // If all of the PHI's incoming values are the same then replace the PHI node
+ // with the common value.
+ Value *CommonValue = 0;
+ bool HasUndefInput = false;
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
+ Value *Incoming = PN->getIncomingValue(i);
+ // If the incoming value is the phi node itself, it can safely be skipped.
+ if (Incoming == PN) continue;
+ if (isa<UndefValue>(Incoming)) {
+ // Remember that we saw an undef value, but otherwise ignore them.
+ HasUndefInput = true;
+ continue;
+ }
+ if (CommonValue && Incoming != CommonValue)
+ return 0; // Not the same, bail out.
+ CommonValue = Incoming;
+ }
+
+ // If CommonValue is null then all of the incoming values were either undef or
+ // equal to the phi node itself.
+ if (!CommonValue)
+ return UndefValue::get(PN->getType());
+
+ // If we have a PHI node like phi(X, undef, X), where X is defined by some
+ // instruction, we cannot return X as the result of the PHI node unless it
+ // dominates the PHI block.
+ if (HasUndefInput)
+ return ValueDominatesPHI(CommonValue, PN, DT) ? CommonValue : 0;
+
+ return CommonValue;
}
@@ -375,59 +2339,184 @@
/// SimplifyBinOp - Given operands for a BinaryOperator, see if we can
/// fold the result. If not, this returns null.
-Value *llvm::SimplifyBinOp(unsigned Opcode, Value *LHS, Value *RHS,
- const TargetData *TD) {
+static Value *SimplifyBinOp(unsigned Opcode, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
switch (Opcode) {
- case Instruction::And: return SimplifyAndInst(LHS, RHS, TD);
- case Instruction::Or: return SimplifyOrInst(LHS, RHS, TD);
+ case Instruction::Add:
+ return SimplifyAddInst(LHS, RHS, /*isNSW*/false, /*isNUW*/false,
+ TD, DT, MaxRecurse);
+ case Instruction::Sub:
+ return SimplifySubInst(LHS, RHS, /*isNSW*/false, /*isNUW*/false,
+ TD, DT, MaxRecurse);
+ case Instruction::Mul: return SimplifyMulInst (LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::SDiv: return SimplifySDivInst(LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::UDiv: return SimplifyUDivInst(LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::FDiv: return SimplifyFDivInst(LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::SRem: return SimplifySRemInst(LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::URem: return SimplifyURemInst(LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::FRem: return SimplifyFRemInst(LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::Shl:
+ return SimplifyShlInst(LHS, RHS, /*isNSW*/false, /*isNUW*/false,
+ TD, DT, MaxRecurse);
+ case Instruction::LShr:
+ return SimplifyLShrInst(LHS, RHS, /*isExact*/false, TD, DT, MaxRecurse);
+ case Instruction::AShr:
+ return SimplifyAShrInst(LHS, RHS, /*isExact*/false, TD, DT, MaxRecurse);
+ case Instruction::And: return SimplifyAndInst(LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::Or: return SimplifyOrInst (LHS, RHS, TD, DT, MaxRecurse);
+ case Instruction::Xor: return SimplifyXorInst(LHS, RHS, TD, DT, MaxRecurse);
default:
if (Constant *CLHS = dyn_cast<Constant>(LHS))
if (Constant *CRHS = dyn_cast<Constant>(RHS)) {
Constant *COps[] = {CLHS, CRHS};
- return ConstantFoldInstOperands(Opcode, LHS->getType(), COps, 2, TD);
+ return ConstantFoldInstOperands(Opcode, LHS->getType(), COps, TD);
}
+
+ // If the operation is associative, try some generic simplifications.
+ if (Instruction::isAssociative(Opcode))
+ if (Value *V = SimplifyAssociativeBinOp(Opcode, LHS, RHS, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a select instruction, check whether
+ // operating on either branch of the select always yields the same value.
+ if (isa<SelectInst>(LHS) || isa<SelectInst>(RHS))
+ if (Value *V = ThreadBinOpOverSelect(Opcode, LHS, RHS, TD, DT,
+ MaxRecurse))
+ return V;
+
+ // If the operation is with the result of a phi instruction, check whether
+ // operating on all incoming values of the phi always yields the same value.
+ if (isa<PHINode>(LHS) || isa<PHINode>(RHS))
+ if (Value *V = ThreadBinOpOverPHI(Opcode, LHS, RHS, TD, DT, MaxRecurse))
+ return V;
+
return 0;
}
}
-/// SimplifyCmpInst - Given operands for a CmpInst, see if we can
-/// fold the result.
-Value *llvm::SimplifyCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
- const TargetData *TD) {
- if (CmpInst::isIntPredicate((CmpInst::Predicate)Predicate))
- return SimplifyICmpInst(Predicate, LHS, RHS, TD);
- return SimplifyFCmpInst(Predicate, LHS, RHS, TD);
+Value *llvm::SimplifyBinOp(unsigned Opcode, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifyBinOp(Opcode, LHS, RHS, TD, DT, RecursionLimit);
}
+/// SimplifyCmpInst - Given operands for a CmpInst, see if we can
+/// fold the result.
+static Value *SimplifyCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT,
+ unsigned MaxRecurse) {
+ if (CmpInst::isIntPredicate((CmpInst::Predicate)Predicate))
+ return SimplifyICmpInst(Predicate, LHS, RHS, TD, DT, MaxRecurse);
+ return SimplifyFCmpInst(Predicate, LHS, RHS, TD, DT, MaxRecurse);
+}
+
+Value *llvm::SimplifyCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
+ const TargetData *TD, const DominatorTree *DT) {
+ return ::SimplifyCmpInst(Predicate, LHS, RHS, TD, DT, RecursionLimit);
+}
/// SimplifyInstruction - See if we can compute a simplified version of this
/// instruction. If not, this returns null.
-Value *llvm::SimplifyInstruction(Instruction *I, const TargetData *TD) {
+Value *llvm::SimplifyInstruction(Instruction *I, const TargetData *TD,
+ const DominatorTree *DT) {
+ Value *Result;
+
switch (I->getOpcode()) {
default:
- return ConstantFoldInstruction(I, TD);
+ Result = ConstantFoldInstruction(I, TD);
+ break;
case Instruction::Add:
- return SimplifyAddInst(I->getOperand(0), I->getOperand(1),
- cast<BinaryOperator>(I)->hasNoSignedWrap(),
- cast<BinaryOperator>(I)->hasNoUnsignedWrap(), TD);
+ Result = SimplifyAddInst(I->getOperand(0), I->getOperand(1),
+ cast<BinaryOperator>(I)->hasNoSignedWrap(),
+ cast<BinaryOperator>(I)->hasNoUnsignedWrap(),
+ TD, DT);
+ break;
+ case Instruction::Sub:
+ Result = SimplifySubInst(I->getOperand(0), I->getOperand(1),
+ cast<BinaryOperator>(I)->hasNoSignedWrap(),
+ cast<BinaryOperator>(I)->hasNoUnsignedWrap(),
+ TD, DT);
+ break;
+ case Instruction::Mul:
+ Result = SimplifyMulInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
+ case Instruction::SDiv:
+ Result = SimplifySDivInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
+ case Instruction::UDiv:
+ Result = SimplifyUDivInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
+ case Instruction::FDiv:
+ Result = SimplifyFDivInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
+ case Instruction::SRem:
+ Result = SimplifySRemInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
+ case Instruction::URem:
+ Result = SimplifyURemInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
+ case Instruction::FRem:
+ Result = SimplifyFRemInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
+ case Instruction::Shl:
+ Result = SimplifyShlInst(I->getOperand(0), I->getOperand(1),
+ cast<BinaryOperator>(I)->hasNoSignedWrap(),
+ cast<BinaryOperator>(I)->hasNoUnsignedWrap(),
+ TD, DT);
+ break;
+ case Instruction::LShr:
+ Result = SimplifyLShrInst(I->getOperand(0), I->getOperand(1),
+ cast<BinaryOperator>(I)->isExact(),
+ TD, DT);
+ break;
+ case Instruction::AShr:
+ Result = SimplifyAShrInst(I->getOperand(0), I->getOperand(1),
+ cast<BinaryOperator>(I)->isExact(),
+ TD, DT);
+ break;
case Instruction::And:
- return SimplifyAndInst(I->getOperand(0), I->getOperand(1), TD);
+ Result = SimplifyAndInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
case Instruction::Or:
- return SimplifyOrInst(I->getOperand(0), I->getOperand(1), TD);
+ Result = SimplifyOrInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
+ case Instruction::Xor:
+ Result = SimplifyXorInst(I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
case Instruction::ICmp:
- return SimplifyICmpInst(cast<ICmpInst>(I)->getPredicate(),
- I->getOperand(0), I->getOperand(1), TD);
+ Result = SimplifyICmpInst(cast<ICmpInst>(I)->getPredicate(),
+ I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
case Instruction::FCmp:
- return SimplifyFCmpInst(cast<FCmpInst>(I)->getPredicate(),
- I->getOperand(0), I->getOperand(1), TD);
+ Result = SimplifyFCmpInst(cast<FCmpInst>(I)->getPredicate(),
+ I->getOperand(0), I->getOperand(1), TD, DT);
+ break;
case Instruction::Select:
- return SimplifySelectInst(I->getOperand(0), I->getOperand(1),
- I->getOperand(2), TD);
+ Result = SimplifySelectInst(I->getOperand(0), I->getOperand(1),
+ I->getOperand(2), TD, DT);
+ break;
case Instruction::GetElementPtr: {
SmallVector<Value*, 8> Ops(I->op_begin(), I->op_end());
- return SimplifyGEPInst(&Ops[0], Ops.size(), TD);
+ Result = SimplifyGEPInst(Ops, TD, DT);
+ break;
}
+ case Instruction::InsertValue: {
+ InsertValueInst *IV = cast<InsertValueInst>(I);
+ Result = SimplifyInsertValueInst(IV->getAggregateOperand(),
+ IV->getInsertedValueOperand(),
+ IV->getIndices(), TD, DT);
+ break;
}
+ case Instruction::PHI:
+ Result = SimplifyPHINode(cast<PHINode>(I), DT);
+ break;
+ }
+
+ /// If called on unreachable code, the above logic may report that the
+ /// instruction simplified to itself. Make life easier for users by
+ /// detecting that case here, returning a safe value instead.
+ return Result == I ? UndefValue::get(I->getType()) : Result;
}
/// ReplaceAndSimplifyAllUses - Perform From->replaceAllUsesWith(To) and then
@@ -437,15 +2526,16 @@
/// simplifies and deletes scalar operations, it does not change the CFG.
///
void llvm::ReplaceAndSimplifyAllUses(Instruction *From, Value *To,
- const TargetData *TD) {
+ const TargetData *TD,
+ const DominatorTree *DT) {
assert(From != To && "ReplaceAndSimplifyAllUses(X,X) is not valid!");
-
+
// FromHandle/ToHandle - This keeps a WeakVH on the from/to values so that
// we can know if it gets deleted out from under us or replaced in a
// recursive simplification.
WeakVH FromHandle(From);
WeakVH ToHandle(To);
-
+
while (!From->use_empty()) {
// Update the instruction to use the new value.
Use &TheUse = From->use_begin().getUse();
@@ -460,27 +2550,26 @@
// Sanity check to make sure 'User' doesn't dangle across
// SimplifyInstruction.
AssertingVH<> UserHandle(User);
-
- SimplifiedVal = SimplifyInstruction(User, TD);
+
+ SimplifiedVal = SimplifyInstruction(User, TD, DT);
if (SimplifiedVal == 0) continue;
}
-
+
// Recursively simplify this user to the new value.
- ReplaceAndSimplifyAllUses(User, SimplifiedVal, TD);
+ ReplaceAndSimplifyAllUses(User, SimplifiedVal, TD, DT);
From = dyn_cast_or_null<Instruction>((Value*)FromHandle);
To = ToHandle;
-
+
assert(ToHandle && "To value deleted by recursive simplification?");
-
+
// If the recursive simplification ended up revisiting and deleting
// 'From' then we're done.
if (From == 0)
return;
}
-
+
// If 'From' has value handles referring to it, do a real RAUW to update them.
From->replaceAllUsesWith(To);
-
+
From->eraseFromParent();
}
-
diff --git a/src/LLVM/lib/Analysis/Interval.cpp b/src/LLVM/lib/Analysis/Interval.cpp
new file mode 100644
index 0000000..ca9cdca
--- /dev/null
+++ b/src/LLVM/lib/Analysis/Interval.cpp
@@ -0,0 +1,58 @@
+//===- Interval.cpp - Interval class code ---------------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the definition of the Interval class, which represents a
+// partition of a control flow graph of some kind.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/Interval.h"
+#include "llvm/BasicBlock.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/raw_ostream.h"
+#include <algorithm>
+
+using namespace llvm;
+
+//===----------------------------------------------------------------------===//
+// Interval Implementation
+//===----------------------------------------------------------------------===//
+
+// isLoop - Find out if there is a back edge in this interval...
+//
+bool Interval::isLoop() const {
+ // There is a loop in this interval iff one of the predecessors of the header
+ // node lives in the interval.
+ for (::pred_iterator I = ::pred_begin(HeaderNode), E = ::pred_end(HeaderNode);
+ I != E; ++I)
+ if (contains(*I))
+ return true;
+ return false;
+}
+
+
+void Interval::print(raw_ostream &OS) const {
+ OS << "-------------------------------------------------------------\n"
+ << "Interval Contents:\n";
+
+ // Print out all of the basic blocks in the interval...
+ for (std::vector<BasicBlock*>::const_iterator I = Nodes.begin(),
+ E = Nodes.end(); I != E; ++I)
+ OS << **I << "\n";
+
+ OS << "Interval Predecessors:\n";
+ for (std::vector<BasicBlock*>::const_iterator I = Predecessors.begin(),
+ E = Predecessors.end(); I != E; ++I)
+ OS << **I << "\n";
+
+ OS << "Interval Successors:\n";
+ for (std::vector<BasicBlock*>::const_iterator I = Successors.begin(),
+ E = Successors.end(); I != E; ++I)
+ OS << **I << "\n";
+}
diff --git a/src/LLVM/lib/Analysis/IntervalPartition.cpp b/src/LLVM/lib/Analysis/IntervalPartition.cpp
new file mode 100644
index 0000000..2e259b1
--- /dev/null
+++ b/src/LLVM/lib/Analysis/IntervalPartition.cpp
@@ -0,0 +1,114 @@
+//===- IntervalPartition.cpp - Interval Partition module code -------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the definition of the IntervalPartition class, which
+// calculates and represent the interval partition of a function.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/IntervalIterator.h"
+using namespace llvm;
+
+char IntervalPartition::ID = 0;
+INITIALIZE_PASS(IntervalPartition, "intervals",
+ "Interval Partition Construction", true, true)
+
+//===----------------------------------------------------------------------===//
+// IntervalPartition Implementation
+//===----------------------------------------------------------------------===//
+
+// releaseMemory - Reset state back to before function was analyzed
+void IntervalPartition::releaseMemory() {
+ for (unsigned i = 0, e = Intervals.size(); i != e; ++i)
+ delete Intervals[i];
+ IntervalMap.clear();
+ Intervals.clear();
+ RootInterval = 0;
+}
+
+void IntervalPartition::print(raw_ostream &O, const Module*) const {
+ for(unsigned i = 0, e = Intervals.size(); i != e; ++i)
+ Intervals[i]->print(O);
+}
+
+// addIntervalToPartition - Add an interval to the internal list of intervals,
+// and then add mappings from all of the basic blocks in the interval to the
+// interval itself (in the IntervalMap).
+//
+void IntervalPartition::addIntervalToPartition(Interval *I) {
+ Intervals.push_back(I);
+
+ // Add mappings for all of the basic blocks in I to the IntervalPartition
+ for (Interval::node_iterator It = I->Nodes.begin(), End = I->Nodes.end();
+ It != End; ++It)
+ IntervalMap.insert(std::make_pair(*It, I));
+}
+
+// updatePredecessors - Interval generation only sets the successor fields of
+// the interval data structures. After interval generation is complete,
+// run through all of the intervals and propagate successor info as
+// predecessor info.
+//
+void IntervalPartition::updatePredecessors(Interval *Int) {
+ BasicBlock *Header = Int->getHeaderNode();
+ for (Interval::succ_iterator I = Int->Successors.begin(),
+ E = Int->Successors.end(); I != E; ++I)
+ getBlockInterval(*I)->Predecessors.push_back(Header);
+}
+
+// IntervalPartition ctor - Build the first level interval partition for the
+// specified function...
+//
+bool IntervalPartition::runOnFunction(Function &F) {
+ // Pass false to intervals_begin because we take ownership of it's memory
+ function_interval_iterator I = intervals_begin(&F, false);
+ assert(I != intervals_end(&F) && "No intervals in function!?!?!");
+
+ addIntervalToPartition(RootInterval = *I);
+
+ ++I; // After the first one...
+
+ // Add the rest of the intervals to the partition.
+ for (function_interval_iterator E = intervals_end(&F); I != E; ++I)
+ addIntervalToPartition(*I);
+
+ // Now that we know all of the successor information, propagate this to the
+ // predecessors for each block.
+ for (unsigned i = 0, e = Intervals.size(); i != e; ++i)
+ updatePredecessors(Intervals[i]);
+ return false;
+}
+
+
+// IntervalPartition ctor - Build a reduced interval partition from an
+// existing interval graph. This takes an additional boolean parameter to
+// distinguish it from a copy constructor. Always pass in false for now.
+//
+IntervalPartition::IntervalPartition(IntervalPartition &IP, bool)
+ : FunctionPass(ID) {
+ assert(IP.getRootInterval() && "Cannot operate on empty IntervalPartitions!");
+
+ // Pass false to intervals_begin because we take ownership of it's memory
+ interval_part_interval_iterator I = intervals_begin(IP, false);
+ assert(I != intervals_end(IP) && "No intervals in interval partition!?!?!");
+
+ addIntervalToPartition(RootInterval = *I);
+
+ ++I; // After the first one...
+
+ // Add the rest of the intervals to the partition.
+ for (interval_part_interval_iterator E = intervals_end(IP); I != E; ++I)
+ addIntervalToPartition(*I);
+
+ // Now that we know all of the successor information, propagate this to the
+ // predecessors for each block.
+ for (unsigned i = 0, e = Intervals.size(); i != e; ++i)
+ updatePredecessors(Intervals[i]);
+}
+
diff --git a/src/LLVM/lib/Analysis/LLVMAnalysis.vcxproj b/src/LLVM/lib/Analysis/LLVMAnalysis.vcxproj
new file mode 100644
index 0000000..3029d51
--- /dev/null
+++ b/src/LLVM/lib/Analysis/LLVMAnalysis.vcxproj
@@ -0,0 +1,379 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup Label="ProjectConfigurations">
+ <ProjectConfiguration Include="Debug|Win32">
+ <Configuration>Debug</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Debug|x64">
+ <Configuration>Debug</Configuration>
+ <Platform>x64</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Profile|Win32">
+ <Configuration>Profile</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Profile|x64">
+ <Configuration>Profile</Configuration>
+ <Platform>x64</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Release|Win32">
+ <Configuration>Release</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Release|x64">
+ <Configuration>Release</Configuration>
+ <Platform>x64</Platform>
+ </ProjectConfiguration>
+ </ItemGroup>
+ <PropertyGroup Label="Globals">
+ <ProjectGUID>{97EDF19C-6360-4770-9255-EBA2F1A13E9B}</ProjectGUID>
+ <Keyword>Win32Proj</Keyword>
+ <Platform>Win32</Platform>
+ <ProjectName>LLVMAnalysis</ProjectName>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>NotSet</CharacterSet>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>NotSet</CharacterSet>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>NotSet</CharacterSet>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Profile|x64'" Label="Configuration">
+ <ConfigurationType>StaticLibrary</ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>NotSet</CharacterSet>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+ <ImportGroup Label="ExtensionSettings">
+ </ImportGroup>
+ <ImportGroup Label="PropertySheets">
+ <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+ </ImportGroup>
+ <PropertyGroup Label="UserMacros" />
+ <PropertyGroup>
+ <_ProjectFileVersion>10.0.20506.1</_ProjectFileVersion>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(Platform)\$(Configuration)\</OutDir>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">$(Platform)\$(Configuration)\</OutDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(Platform)\$(Configuration)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">$(Platform)\$(Configuration)\</IntDir>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(ProjectName)</TargetName>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">$(ProjectName)</TargetName>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">.lib</TargetExt>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">.lib</TargetExt>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(Platform)\$(Configuration)\</OutDir>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">$(Platform)\$(Configuration)\</OutDir>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Release|x64'">$(Platform)\$(Configuration)\</OutDir>
+ <OutDir Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">$(Platform)\$(Configuration)\</OutDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(Platform)\$(Configuration)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">$(Platform)\$(Configuration)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Release|x64'">$(Platform)\$(Configuration)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">$(Platform)\$(Configuration)\</IntDir>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(ProjectName)</TargetName>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">$(ProjectName)</TargetName>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Release|x64'">$(ProjectName)</TargetName>
+ <TargetName Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">$(ProjectName)</TargetName>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">.lib</TargetExt>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">.lib</TargetExt>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Release|x64'">.lib</TargetExt>
+ <TargetExt Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">.lib</TargetExt>
+ </PropertyGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <BasicRuntimeChecks>EnableFastChecks</BasicRuntimeChecks>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>
+ </ExceptionHandling>
+ <InlineFunctionExpansion>Disabled</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Disabled</Optimization>
+ <RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;_DEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Debug";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Debug</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\Debug/LLVMAnalysis.pdb</ProgramDataBaseFileName>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;_DEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Debug\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <BasicRuntimeChecks>EnableFastChecks</BasicRuntimeChecks>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>
+ </ExceptionHandling>
+ <InlineFunctionExpansion>Disabled</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Disabled</Optimization>
+ <RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;_DEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Debug";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Debug</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\Debug/LLVMAnalysis.pdb</ProgramDataBaseFileName>
+ <OmitFramePointers>false</OmitFramePointers>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;_DEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Debug\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>false</ExceptionHandling>
+ <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Full</Optimization>
+ <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <DebugInformationFormat>
+ </DebugInformationFormat>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Release";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Release</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\Release/LLVMAnalysis.pdb</ProgramDataBaseFileName>
+ <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
+ <OmitFramePointers>true</OmitFramePointers>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ <StringPooling>true</StringPooling>
+ <BufferSecurityCheck>false</BufferSecurityCheck>
+ <FunctionLevelLinking>false</FunctionLevelLinking>
+ <FloatingPointExceptions>false</FloatingPointExceptions>
+ <CreateHotpatchableImage>false</CreateHotpatchableImage>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Release\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Profile|Win32'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>false</ExceptionHandling>
+ <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Full</Optimization>
+ <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Release";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Release</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\Release/LLVMAnalysis.pdb</ProgramDataBaseFileName>
+ <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
+ <OmitFramePointers>false</OmitFramePointers>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ <StringPooling>true</StringPooling>
+ <BufferSecurityCheck>false</BufferSecurityCheck>
+ <FunctionLevelLinking>false</FunctionLevelLinking>
+ <FloatingPointExceptions>false</FloatingPointExceptions>
+ <CreateHotpatchableImage>false</CreateHotpatchableImage>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Release\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>false</ExceptionHandling>
+ <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Full</Optimization>
+ <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <DebugInformationFormat>
+ </DebugInformationFormat>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Release";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Release</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\Release/LLVMAnalysis.pdb</ProgramDataBaseFileName>
+ <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
+ <OmitFramePointers>true</OmitFramePointers>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ <StringPooling>true</StringPooling>
+ <BufferSecurityCheck>false</BufferSecurityCheck>
+ <FunctionLevelLinking>false</FunctionLevelLinking>
+ <FloatingPointExceptions>false</FloatingPointExceptions>
+ <CreateHotpatchableImage>false</CreateHotpatchableImage>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Release\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Profile|x64'">
+ <ClCompile>
+ <AdditionalOptions> /Zm1000 /EHs-c- -w14062 %(AdditionalOptions)</AdditionalOptions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <CompileAs>CompileAsCpp</CompileAs>
+ <DisableSpecificWarnings>4146;4180;4224;4244;4267;4275;4291;4345;4351;4355;4503;4551;4624;4715;4800;4065;4181</DisableSpecificWarnings>
+ <ExceptionHandling>false</ExceptionHandling>
+ <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
+ <MultiProcessorCompilation>true</MultiProcessorCompilation>
+ <Optimization>Full</Optimization>
+ <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+ <RuntimeTypeInfo>false</RuntimeTypeInfo>
+ <WarningLevel>Level3</WarningLevel>
+ <DebugInformationFormat>
+ </DebugInformationFormat>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR="Release";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AssemblerListingLocation>Release</AssemblerListingLocation>
+ <ObjectFileName>$(IntDir)</ObjectFileName>
+ <ProgramDataBaseFileName>..\Release/LLVMAnalysis.pdb</ProgramDataBaseFileName>
+ <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
+ <OmitFramePointers>false</OmitFramePointers>
+ <WholeProgramOptimization>true</WholeProgramOptimization>
+ <StringPooling>true</StringPooling>
+ <BufferSecurityCheck>false</BufferSecurityCheck>
+ <FunctionLevelLinking>false</FunctionLevelLinking>
+ <FloatingPointExceptions>false</FloatingPointExceptions>
+ <CreateHotpatchableImage>false</CreateHotpatchableImage>
+ </ClCompile>
+ <ResourceCompile>
+ <PreprocessorDefinitions>WIN32;_WINDOWS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_SECURE_NO_WARNINGS;_CRT_NONSTDC_NO_DEPRECATE;_CRT_NONSTDC_NO_WARNINGS;_SCL_SECURE_NO_DEPRECATE;_SCL_SECURE_NO_WARNINGS;__STDC_CONSTANT_MACROS;__STDC_FORMAT_MACROS;__STDC_LIMIT_MACROS;_HAS_EXCEPTIONS=0;CMAKE_INTDIR=\"Release\";%(PreprocessorDefinitions)</PreprocessorDefinitions>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ </ResourceCompile>
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ </ItemDefinitionGroup>
+ <ItemGroup>
+ <ClCompile Include="AliasAnalysis.cpp" />
+ <ClCompile Include="AliasSetTracker.cpp" />
+ <ClCompile Include="BasicAliasAnalysis.cpp" />
+ <ClCompile Include="BranchProbabilityInfo.cpp" />
+ <ClCompile Include="CaptureTracking.cpp" />
+ <ClCompile Include="ConstantFolding.cpp" />
+ <ClCompile Include="DIBuilder.cpp" />
+ <ClCompile Include="DebugInfo.cpp" />
+ <ClCompile Include="IVUsers.cpp" />
+ <ClCompile Include="InstructionSimplify.cpp" />
+ <ClCompile Include="Loads.cpp" />
+ <ClCompile Include="LoopInfo.cpp" />
+ <ClCompile Include="LoopPass.cpp" />
+ <ClCompile Include="MemoryBuiltins.cpp" />
+ <ClCompile Include="MemoryDependenceAnalysis.cpp" />
+ <ClCompile Include="NoAliasAnalysis.cpp" />
+ <ClCompile Include="PHITransAddr.cpp" />
+ <ClCompile Include="PathNumbering.cpp" />
+ <ClCompile Include="ProfileInfo.cpp" />
+ <ClCompile Include="ScalarEvolution.cpp" />
+ <ClCompile Include="ScalarEvolutionExpander.cpp" />
+ <ClCompile Include="ScalarEvolutionNormalization.cpp" />
+ <ClCompile Include="TypeBasedAliasAnalysis.cpp" />
+ <ClCompile Include="ValueTracking.cpp" />
+ </ItemGroup>
+ <ItemGroup>
+ <ProjectReference Include="..\VMCore/LLVMCore.vcxproj">
+ <Project>00F3295C-F7A0-43D3-BD0B-1BC0515B30E1</Project>
+ </ProjectReference>
+ <ProjectReference Include="..\Support/LLVMSupport.vcxproj">
+ <Project>C688DD59-C6CB-4B33-B56F-A7D6F3761524</Project>
+ </ProjectReference>
+ <ProjectReference Include="..\Target/LLVMTarget.vcxproj">
+ <Project>76F7B8C2-C825-40DC-BB68-9D987275E320</Project>
+ </ProjectReference>
+ <ProjectReference Include="..\..\include/llvm/intrinsics_gen.vcxproj">
+ <Project>E9B87B46-1EB0-4D95-9049-41B148FBADCD</Project>
+ </ProjectReference>
+ </ItemGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+ <ImportGroup Label="ExtensionTargets">
+ </ImportGroup>
+</Project>
\ No newline at end of file
diff --git a/src/LLVM/lib/Analysis/LLVMAnalysis.vcxproj.filters b/src/LLVM/lib/Analysis/LLVMAnalysis.vcxproj.filters
new file mode 100644
index 0000000..3aba59b
--- /dev/null
+++ b/src/LLVM/lib/Analysis/LLVMAnalysis.vcxproj.filters
@@ -0,0 +1,90 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup>
+ <ClCompile Include="AliasAnalysis.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="AliasSetTracker.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="BasicAliasAnalysis.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="BranchProbabilityInfo.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="CaptureTracking.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="ConstantFolding.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="DIBuilder.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="DebugInfo.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="IVUsers.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="InstructionSimplify.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="Loads.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="LoopInfo.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="LoopPass.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="MemoryBuiltins.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="MemoryDependenceAnalysis.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="NoAliasAnalysis.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="PHITransAddr.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="PathNumbering.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="ProfileInfo.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="ScalarEvolution.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="ScalarEvolutionExpander.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="ScalarEvolutionNormalization.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="TypeBasedAliasAnalysis.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ <ClCompile Include="ValueTracking.cpp">
+ <Filter>Source Files</Filter>
+ </ClCompile>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ <Filter Include="Source Files">
+ <UniqueIdentifier>{1733179C-6FE4-462E-9EA5-4A29A1ACFE25}</UniqueIdentifier>
+ </Filter>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+</Project>
\ No newline at end of file
diff --git a/src/LLVM/lib/Analysis/LazyValueInfo.cpp b/src/LLVM/lib/Analysis/LazyValueInfo.cpp
new file mode 100644
index 0000000..f80595c
--- /dev/null
+++ b/src/LLVM/lib/Analysis/LazyValueInfo.cpp
@@ -0,0 +1,1128 @@
+//===- LazyValueInfo.cpp - Value constraint analysis ----------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the interface for lazy computation of value constraint
+// information.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "lazy-value-info"
+#include "llvm/Analysis/LazyValueInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Constants.h"
+#include "llvm/Instructions.h"
+#include "llvm/IntrinsicInst.h"
+#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/ConstantRange.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/ValueHandle.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/DenseSet.h"
+#include "llvm/ADT/STLExtras.h"
+#include <map>
+#include <stack>
+using namespace llvm;
+
+char LazyValueInfo::ID = 0;
+INITIALIZE_PASS(LazyValueInfo, "lazy-value-info",
+ "Lazy Value Information Analysis", false, true)
+
+namespace llvm {
+ FunctionPass *createLazyValueInfoPass() { return new LazyValueInfo(); }
+}
+
+
+//===----------------------------------------------------------------------===//
+// LVILatticeVal
+//===----------------------------------------------------------------------===//
+
+/// LVILatticeVal - This is the information tracked by LazyValueInfo for each
+/// value.
+///
+/// FIXME: This is basically just for bringup, this can be made a lot more rich
+/// in the future.
+///
+namespace {
+class LVILatticeVal {
+ enum LatticeValueTy {
+ /// undefined - This Value has no known value yet.
+ undefined,
+
+ /// constant - This Value has a specific constant value.
+ constant,
+ /// notconstant - This Value is known to not have the specified value.
+ notconstant,
+
+ /// constantrange - The Value falls within this range.
+ constantrange,
+
+ /// overdefined - This value is not known to be constant, and we know that
+ /// it has a value.
+ overdefined
+ };
+
+ /// Val: This stores the current lattice value along with the Constant* for
+ /// the constant if this is a 'constant' or 'notconstant' value.
+ LatticeValueTy Tag;
+ Constant *Val;
+ ConstantRange Range;
+
+public:
+ LVILatticeVal() : Tag(undefined), Val(0), Range(1, true) {}
+
+ static LVILatticeVal get(Constant *C) {
+ LVILatticeVal Res;
+ if (!isa<UndefValue>(C))
+ Res.markConstant(C);
+ return Res;
+ }
+ static LVILatticeVal getNot(Constant *C) {
+ LVILatticeVal Res;
+ if (!isa<UndefValue>(C))
+ Res.markNotConstant(C);
+ return Res;
+ }
+ static LVILatticeVal getRange(ConstantRange CR) {
+ LVILatticeVal Res;
+ Res.markConstantRange(CR);
+ return Res;
+ }
+
+ bool isUndefined() const { return Tag == undefined; }
+ bool isConstant() const { return Tag == constant; }
+ bool isNotConstant() const { return Tag == notconstant; }
+ bool isConstantRange() const { return Tag == constantrange; }
+ bool isOverdefined() const { return Tag == overdefined; }
+
+ Constant *getConstant() const {
+ assert(isConstant() && "Cannot get the constant of a non-constant!");
+ return Val;
+ }
+
+ Constant *getNotConstant() const {
+ assert(isNotConstant() && "Cannot get the constant of a non-notconstant!");
+ return Val;
+ }
+
+ ConstantRange getConstantRange() const {
+ assert(isConstantRange() &&
+ "Cannot get the constant-range of a non-constant-range!");
+ return Range;
+ }
+
+ /// markOverdefined - Return true if this is a change in status.
+ bool markOverdefined() {
+ if (isOverdefined())
+ return false;
+ Tag = overdefined;
+ return true;
+ }
+
+ /// markConstant - Return true if this is a change in status.
+ bool markConstant(Constant *V) {
+ assert(V && "Marking constant with NULL");
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(V))
+ return markConstantRange(ConstantRange(CI->getValue()));
+ if (isa<UndefValue>(V))
+ return false;
+
+ assert((!isConstant() || getConstant() == V) &&
+ "Marking constant with different value");
+ assert(isUndefined());
+ Tag = constant;
+ Val = V;
+ return true;
+ }
+
+ /// markNotConstant - Return true if this is a change in status.
+ bool markNotConstant(Constant *V) {
+ assert(V && "Marking constant with NULL");
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(V))
+ return markConstantRange(ConstantRange(CI->getValue()+1, CI->getValue()));
+ if (isa<UndefValue>(V))
+ return false;
+
+ assert((!isConstant() || getConstant() != V) &&
+ "Marking constant !constant with same value");
+ assert((!isNotConstant() || getNotConstant() == V) &&
+ "Marking !constant with different value");
+ assert(isUndefined() || isConstant());
+ Tag = notconstant;
+ Val = V;
+ return true;
+ }
+
+ /// markConstantRange - Return true if this is a change in status.
+ bool markConstantRange(const ConstantRange NewR) {
+ if (isConstantRange()) {
+ if (NewR.isEmptySet())
+ return markOverdefined();
+
+ bool changed = Range == NewR;
+ Range = NewR;
+ return changed;
+ }
+
+ assert(isUndefined());
+ if (NewR.isEmptySet())
+ return markOverdefined();
+
+ Tag = constantrange;
+ Range = NewR;
+ return true;
+ }
+
+ /// mergeIn - Merge the specified lattice value into this one, updating this
+ /// one and returning true if anything changed.
+ bool mergeIn(const LVILatticeVal &RHS) {
+ if (RHS.isUndefined() || isOverdefined()) return false;
+ if (RHS.isOverdefined()) return markOverdefined();
+
+ if (isUndefined()) {
+ Tag = RHS.Tag;
+ Val = RHS.Val;
+ Range = RHS.Range;
+ return true;
+ }
+
+ if (isConstant()) {
+ if (RHS.isConstant()) {
+ if (Val == RHS.Val)
+ return false;
+ return markOverdefined();
+ }
+
+ if (RHS.isNotConstant()) {
+ if (Val == RHS.Val)
+ return markOverdefined();
+
+ // Unless we can prove that the two Constants are different, we must
+ // move to overdefined.
+ // FIXME: use TargetData for smarter constant folding.
+ if (ConstantInt *Res = dyn_cast<ConstantInt>(
+ ConstantFoldCompareInstOperands(CmpInst::ICMP_NE,
+ getConstant(),
+ RHS.getNotConstant())))
+ if (Res->isOne())
+ return markNotConstant(RHS.getNotConstant());
+
+ return markOverdefined();
+ }
+
+ // RHS is a ConstantRange, LHS is a non-integer Constant.
+
+ // FIXME: consider the case where RHS is a range [1, 0) and LHS is
+ // a function. The correct result is to pick up RHS.
+
+ return markOverdefined();
+ }
+
+ if (isNotConstant()) {
+ if (RHS.isConstant()) {
+ if (Val == RHS.Val)
+ return markOverdefined();
+
+ // Unless we can prove that the two Constants are different, we must
+ // move to overdefined.
+ // FIXME: use TargetData for smarter constant folding.
+ if (ConstantInt *Res = dyn_cast<ConstantInt>(
+ ConstantFoldCompareInstOperands(CmpInst::ICMP_NE,
+ getNotConstant(),
+ RHS.getConstant())))
+ if (Res->isOne())
+ return false;
+
+ return markOverdefined();
+ }
+
+ if (RHS.isNotConstant()) {
+ if (Val == RHS.Val)
+ return false;
+ return markOverdefined();
+ }
+
+ return markOverdefined();
+ }
+
+ assert(isConstantRange() && "New LVILattice type?");
+ if (!RHS.isConstantRange())
+ return markOverdefined();
+
+ ConstantRange NewR = Range.unionWith(RHS.getConstantRange());
+ if (NewR.isFullSet())
+ return markOverdefined();
+ return markConstantRange(NewR);
+ }
+};
+
+} // end anonymous namespace.
+
+namespace llvm {
+raw_ostream &operator<<(raw_ostream &OS, const LVILatticeVal &Val)
+ LLVM_ATTRIBUTE_USED;
+raw_ostream &operator<<(raw_ostream &OS, const LVILatticeVal &Val) {
+ if (Val.isUndefined())
+ return OS << "undefined";
+ if (Val.isOverdefined())
+ return OS << "overdefined";
+
+ if (Val.isNotConstant())
+ return OS << "notconstant<" << *Val.getNotConstant() << '>';
+ else if (Val.isConstantRange())
+ return OS << "constantrange<" << Val.getConstantRange().getLower() << ", "
+ << Val.getConstantRange().getUpper() << '>';
+ return OS << "constant<" << *Val.getConstant() << '>';
+}
+}
+
+//===----------------------------------------------------------------------===//
+// LazyValueInfoCache Decl
+//===----------------------------------------------------------------------===//
+
+namespace {
+ /// LVIValueHandle - A callback value handle update the cache when
+ /// values are erased.
+ class LazyValueInfoCache;
+ struct LVIValueHandle : public CallbackVH {
+ LazyValueInfoCache *Parent;
+
+ LVIValueHandle(Value *V, LazyValueInfoCache *P)
+ : CallbackVH(V), Parent(P) { }
+
+ void deleted();
+ void allUsesReplacedWith(Value *V) {
+ deleted();
+ }
+ };
+}
+
+namespace llvm {
+ template<>
+ struct DenseMapInfo<LVIValueHandle> {
+ typedef DenseMapInfo<Value*> PointerInfo;
+ static inline LVIValueHandle getEmptyKey() {
+ return LVIValueHandle(PointerInfo::getEmptyKey(),
+ static_cast<LazyValueInfoCache*>(0));
+ }
+ static inline LVIValueHandle getTombstoneKey() {
+ return LVIValueHandle(PointerInfo::getTombstoneKey(),
+ static_cast<LazyValueInfoCache*>(0));
+ }
+ static unsigned getHashValue(const LVIValueHandle &Val) {
+ return PointerInfo::getHashValue(Val);
+ }
+ static bool isEqual(const LVIValueHandle &LHS, const LVIValueHandle &RHS) {
+ return LHS == RHS;
+ }
+ };
+
+ template<>
+ struct DenseMapInfo<std::pair<AssertingVH<BasicBlock>, Value*> > {
+ typedef std::pair<AssertingVH<BasicBlock>, Value*> PairTy;
+ typedef DenseMapInfo<AssertingVH<BasicBlock> > APointerInfo;
+ typedef DenseMapInfo<Value*> BPointerInfo;
+ static inline PairTy getEmptyKey() {
+ return std::make_pair(APointerInfo::getEmptyKey(),
+ BPointerInfo::getEmptyKey());
+ }
+ static inline PairTy getTombstoneKey() {
+ return std::make_pair(APointerInfo::getTombstoneKey(),
+ BPointerInfo::getTombstoneKey());
+ }
+ static unsigned getHashValue( const PairTy &Val) {
+ return APointerInfo::getHashValue(Val.first) ^
+ BPointerInfo::getHashValue(Val.second);
+ }
+ static bool isEqual(const PairTy &LHS, const PairTy &RHS) {
+ return APointerInfo::isEqual(LHS.first, RHS.first) &&
+ BPointerInfo::isEqual(LHS.second, RHS.second);
+ }
+ };
+}
+
+namespace {
+ /// LazyValueInfoCache - This is the cache kept by LazyValueInfo which
+ /// maintains information about queries across the clients' queries.
+ class LazyValueInfoCache {
+ /// ValueCacheEntryTy - This is all of the cached block information for
+ /// exactly one Value*. The entries are sorted by the BasicBlock* of the
+ /// entries, allowing us to do a lookup with a binary search.
+ typedef std::map<AssertingVH<BasicBlock>, LVILatticeVal> ValueCacheEntryTy;
+
+ /// ValueCache - This is all of the cached information for all values,
+ /// mapped from Value* to key information.
+ DenseMap<LVIValueHandle, ValueCacheEntryTy> ValueCache;
+
+ /// OverDefinedCache - This tracks, on a per-block basis, the set of
+ /// values that are over-defined at the end of that block. This is required
+ /// for cache updating.
+ typedef std::pair<AssertingVH<BasicBlock>, Value*> OverDefinedPairTy;
+ DenseSet<OverDefinedPairTy> OverDefinedCache;
+
+ /// BlockValueStack - This stack holds the state of the value solver
+ /// during a query. It basically emulates the callstack of the naive
+ /// recursive value lookup process.
+ std::stack<std::pair<BasicBlock*, Value*> > BlockValueStack;
+
+ friend struct LVIValueHandle;
+
+ /// OverDefinedCacheUpdater - A helper object that ensures that the
+ /// OverDefinedCache is updated whenever solveBlockValue returns.
+ struct OverDefinedCacheUpdater {
+ LazyValueInfoCache *Parent;
+ Value *Val;
+ BasicBlock *BB;
+ LVILatticeVal &BBLV;
+
+ OverDefinedCacheUpdater(Value *V, BasicBlock *B, LVILatticeVal &LV,
+ LazyValueInfoCache *P)
+ : Parent(P), Val(V), BB(B), BBLV(LV) { }
+
+ bool markResult(bool changed) {
+ if (changed && BBLV.isOverdefined())
+ Parent->OverDefinedCache.insert(std::make_pair(BB, Val));
+ return changed;
+ }
+ };
+
+
+
+ LVILatticeVal getBlockValue(Value *Val, BasicBlock *BB);
+ bool getEdgeValue(Value *V, BasicBlock *F, BasicBlock *T,
+ LVILatticeVal &Result);
+ bool hasBlockValue(Value *Val, BasicBlock *BB);
+
+ // These methods process one work item and may add more. A false value
+ // returned means that the work item was not completely processed and must
+ // be revisited after going through the new items.
+ bool solveBlockValue(Value *Val, BasicBlock *BB);
+ bool solveBlockValueNonLocal(LVILatticeVal &BBLV,
+ Value *Val, BasicBlock *BB);
+ bool solveBlockValuePHINode(LVILatticeVal &BBLV,
+ PHINode *PN, BasicBlock *BB);
+ bool solveBlockValueConstantRange(LVILatticeVal &BBLV,
+ Instruction *BBI, BasicBlock *BB);
+
+ void solve();
+
+ ValueCacheEntryTy &lookup(Value *V) {
+ return ValueCache[LVIValueHandle(V, this)];
+ }
+
+ public:
+ /// getValueInBlock - This is the query interface to determine the lattice
+ /// value for the specified Value* at the end of the specified block.
+ LVILatticeVal getValueInBlock(Value *V, BasicBlock *BB);
+
+ /// getValueOnEdge - This is the query interface to determine the lattice
+ /// value for the specified Value* that is true on the specified edge.
+ LVILatticeVal getValueOnEdge(Value *V, BasicBlock *FromBB,BasicBlock *ToBB);
+
+ /// threadEdge - This is the update interface to inform the cache that an
+ /// edge from PredBB to OldSucc has been threaded to be from PredBB to
+ /// NewSucc.
+ void threadEdge(BasicBlock *PredBB,BasicBlock *OldSucc,BasicBlock *NewSucc);
+
+ /// eraseBlock - This is part of the update interface to inform the cache
+ /// that a block has been deleted.
+ void eraseBlock(BasicBlock *BB);
+
+ /// clear - Empty the cache.
+ void clear() {
+ ValueCache.clear();
+ OverDefinedCache.clear();
+ }
+ };
+} // end anonymous namespace
+
+void LVIValueHandle::deleted() {
+ typedef std::pair<AssertingVH<BasicBlock>, Value*> OverDefinedPairTy;
+
+ SmallVector<OverDefinedPairTy, 4> ToErase;
+ for (DenseSet<OverDefinedPairTy>::iterator
+ I = Parent->OverDefinedCache.begin(),
+ E = Parent->OverDefinedCache.end();
+ I != E; ++I) {
+ if (I->second == getValPtr())
+ ToErase.push_back(*I);
+ }
+
+ for (SmallVector<OverDefinedPairTy, 4>::iterator I = ToErase.begin(),
+ E = ToErase.end(); I != E; ++I)
+ Parent->OverDefinedCache.erase(*I);
+
+ // This erasure deallocates *this, so it MUST happen after we're done
+ // using any and all members of *this.
+ Parent->ValueCache.erase(*this);
+}
+
+void LazyValueInfoCache::eraseBlock(BasicBlock *BB) {
+ SmallVector<OverDefinedPairTy, 4> ToErase;
+ for (DenseSet<OverDefinedPairTy>::iterator I = OverDefinedCache.begin(),
+ E = OverDefinedCache.end(); I != E; ++I) {
+ if (I->first == BB)
+ ToErase.push_back(*I);
+ }
+
+ for (SmallVector<OverDefinedPairTy, 4>::iterator I = ToErase.begin(),
+ E = ToErase.end(); I != E; ++I)
+ OverDefinedCache.erase(*I);
+
+ for (DenseMap<LVIValueHandle, ValueCacheEntryTy>::iterator
+ I = ValueCache.begin(), E = ValueCache.end(); I != E; ++I)
+ I->second.erase(BB);
+}
+
+void LazyValueInfoCache::solve() {
+ while (!BlockValueStack.empty()) {
+ std::pair<BasicBlock*, Value*> &e = BlockValueStack.top();
+ if (solveBlockValue(e.second, e.first))
+ BlockValueStack.pop();
+ }
+}
+
+bool LazyValueInfoCache::hasBlockValue(Value *Val, BasicBlock *BB) {
+ // If already a constant, there is nothing to compute.
+ if (isa<Constant>(Val))
+ return true;
+
+ LVIValueHandle ValHandle(Val, this);
+ if (!ValueCache.count(ValHandle)) return false;
+ return ValueCache[ValHandle].count(BB);
+}
+
+LVILatticeVal LazyValueInfoCache::getBlockValue(Value *Val, BasicBlock *BB) {
+ // If already a constant, there is nothing to compute.
+ if (Constant *VC = dyn_cast<Constant>(Val))
+ return LVILatticeVal::get(VC);
+
+ return lookup(Val)[BB];
+}
+
+bool LazyValueInfoCache::solveBlockValue(Value *Val, BasicBlock *BB) {
+ if (isa<Constant>(Val))
+ return true;
+
+ ValueCacheEntryTy &Cache = lookup(Val);
+ LVILatticeVal &BBLV = Cache[BB];
+
+ // OverDefinedCacheUpdater is a helper object that will update
+ // the OverDefinedCache for us when this method exits. Make sure to
+ // call markResult on it as we exist, passing a bool to indicate if the
+ // cache needs updating, i.e. if we have solve a new value or not.
+ OverDefinedCacheUpdater ODCacheUpdater(Val, BB, BBLV, this);
+
+ // If we've already computed this block's value, return it.
+ if (!BBLV.isUndefined()) {
+ DEBUG(dbgs() << " reuse BB '" << BB->getName() << "' val=" << BBLV <<'\n');
+
+ // Since we're reusing a cached value here, we don't need to update the
+ // OverDefinedCahce. The cache will have been properly updated
+ // whenever the cached value was inserted.
+ ODCacheUpdater.markResult(false);
+ return true;
+ }
+
+ // Otherwise, this is the first time we're seeing this block. Reset the
+ // lattice value to overdefined, so that cycles will terminate and be
+ // conservatively correct.
+ BBLV.markOverdefined();
+
+ Instruction *BBI = dyn_cast<Instruction>(Val);
+ if (BBI == 0 || BBI->getParent() != BB) {
+ return ODCacheUpdater.markResult(solveBlockValueNonLocal(BBLV, Val, BB));
+ }
+
+ if (PHINode *PN = dyn_cast<PHINode>(BBI)) {
+ return ODCacheUpdater.markResult(solveBlockValuePHINode(BBLV, PN, BB));
+ }
+
+ if (AllocaInst *AI = dyn_cast<AllocaInst>(BBI)) {
+ BBLV = LVILatticeVal::getNot(ConstantPointerNull::get(AI->getType()));
+ return ODCacheUpdater.markResult(true);
+ }
+
+ // We can only analyze the definitions of certain classes of instructions
+ // (integral binops and casts at the moment), so bail if this isn't one.
+ LVILatticeVal Result;
+ if ((!isa<BinaryOperator>(BBI) && !isa<CastInst>(BBI)) ||
+ !BBI->getType()->isIntegerTy()) {
+ DEBUG(dbgs() << " compute BB '" << BB->getName()
+ << "' - overdefined because inst def found.\n");
+ BBLV.markOverdefined();
+ return ODCacheUpdater.markResult(true);
+ }
+
+ // FIXME: We're currently limited to binops with a constant RHS. This should
+ // be improved.
+ BinaryOperator *BO = dyn_cast<BinaryOperator>(BBI);
+ if (BO && !isa<ConstantInt>(BO->getOperand(1))) {
+ DEBUG(dbgs() << " compute BB '" << BB->getName()
+ << "' - overdefined because inst def found.\n");
+
+ BBLV.markOverdefined();
+ return ODCacheUpdater.markResult(true);
+ }
+
+ return ODCacheUpdater.markResult(solveBlockValueConstantRange(BBLV, BBI, BB));
+}
+
+static bool InstructionDereferencesPointer(Instruction *I, Value *Ptr) {
+ if (LoadInst *L = dyn_cast<LoadInst>(I)) {
+ return L->getPointerAddressSpace() == 0 &&
+ GetUnderlyingObject(L->getPointerOperand()) ==
+ GetUnderlyingObject(Ptr);
+ }
+ if (StoreInst *S = dyn_cast<StoreInst>(I)) {
+ return S->getPointerAddressSpace() == 0 &&
+ GetUnderlyingObject(S->getPointerOperand()) ==
+ GetUnderlyingObject(Ptr);
+ }
+ if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(I)) {
+ if (MI->isVolatile()) return false;
+
+ // FIXME: check whether it has a valuerange that excludes zero?
+ ConstantInt *Len = dyn_cast<ConstantInt>(MI->getLength());
+ if (!Len || Len->isZero()) return false;
+
+ if (MI->getDestAddressSpace() == 0)
+ if (MI->getRawDest() == Ptr || MI->getDest() == Ptr)
+ return true;
+ if (MemTransferInst *MTI = dyn_cast<MemTransferInst>(MI))
+ if (MTI->getSourceAddressSpace() == 0)
+ if (MTI->getRawSource() == Ptr || MTI->getSource() == Ptr)
+ return true;
+ }
+ return false;
+}
+
+bool LazyValueInfoCache::solveBlockValueNonLocal(LVILatticeVal &BBLV,
+ Value *Val, BasicBlock *BB) {
+ LVILatticeVal Result; // Start Undefined.
+
+ // If this is a pointer, and there's a load from that pointer in this BB,
+ // then we know that the pointer can't be NULL.
+ bool NotNull = false;
+ if (Val->getType()->isPointerTy()) {
+ if (isa<AllocaInst>(Val)) {
+ NotNull = true;
+ } else {
+ for (BasicBlock::iterator BI = BB->begin(), BE = BB->end();BI != BE;++BI){
+ if (InstructionDereferencesPointer(BI, Val)) {
+ NotNull = true;
+ break;
+ }
+ }
+ }
+ }
+
+ // If this is the entry block, we must be asking about an argument. The
+ // value is overdefined.
+ if (BB == &BB->getParent()->getEntryBlock()) {
+ assert(isa<Argument>(Val) && "Unknown live-in to the entry block");
+ if (NotNull) {
+ PointerType *PTy = cast<PointerType>(Val->getType());
+ Result = LVILatticeVal::getNot(ConstantPointerNull::get(PTy));
+ } else {
+ Result.markOverdefined();
+ }
+ BBLV = Result;
+ return true;
+ }
+
+ // Loop over all of our predecessors, merging what we know from them into
+ // result.
+ bool EdgesMissing = false;
+ for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
+ LVILatticeVal EdgeResult;
+ EdgesMissing |= !getEdgeValue(Val, *PI, BB, EdgeResult);
+ if (EdgesMissing)
+ continue;
+
+ Result.mergeIn(EdgeResult);
+
+ // If we hit overdefined, exit early. The BlockVals entry is already set
+ // to overdefined.
+ if (Result.isOverdefined()) {
+ DEBUG(dbgs() << " compute BB '" << BB->getName()
+ << "' - overdefined because of pred.\n");
+ // If we previously determined that this is a pointer that can't be null
+ // then return that rather than giving up entirely.
+ if (NotNull) {
+ PointerType *PTy = cast<PointerType>(Val->getType());
+ Result = LVILatticeVal::getNot(ConstantPointerNull::get(PTy));
+ }
+
+ BBLV = Result;
+ return true;
+ }
+ }
+ if (EdgesMissing)
+ return false;
+
+ // Return the merged value, which is more precise than 'overdefined'.
+ assert(!Result.isOverdefined());
+ BBLV = Result;
+ return true;
+}
+
+bool LazyValueInfoCache::solveBlockValuePHINode(LVILatticeVal &BBLV,
+ PHINode *PN, BasicBlock *BB) {
+ LVILatticeVal Result; // Start Undefined.
+
+ // Loop over all of our predecessors, merging what we know from them into
+ // result.
+ bool EdgesMissing = false;
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
+ BasicBlock *PhiBB = PN->getIncomingBlock(i);
+ Value *PhiVal = PN->getIncomingValue(i);
+ LVILatticeVal EdgeResult;
+ EdgesMissing |= !getEdgeValue(PhiVal, PhiBB, BB, EdgeResult);
+ if (EdgesMissing)
+ continue;
+
+ Result.mergeIn(EdgeResult);
+
+ // If we hit overdefined, exit early. The BlockVals entry is already set
+ // to overdefined.
+ if (Result.isOverdefined()) {
+ DEBUG(dbgs() << " compute BB '" << BB->getName()
+ << "' - overdefined because of pred.\n");
+
+ BBLV = Result;
+ return true;
+ }
+ }
+ if (EdgesMissing)
+ return false;
+
+ // Return the merged value, which is more precise than 'overdefined'.
+ assert(!Result.isOverdefined() && "Possible PHI in entry block?");
+ BBLV = Result;
+ return true;
+}
+
+bool LazyValueInfoCache::solveBlockValueConstantRange(LVILatticeVal &BBLV,
+ Instruction *BBI,
+ BasicBlock *BB) {
+ // Figure out the range of the LHS. If that fails, bail.
+ if (!hasBlockValue(BBI->getOperand(0), BB)) {
+ BlockValueStack.push(std::make_pair(BB, BBI->getOperand(0)));
+ return false;
+ }
+
+ LVILatticeVal LHSVal = getBlockValue(BBI->getOperand(0), BB);
+ if (!LHSVal.isConstantRange()) {
+ BBLV.markOverdefined();
+ return true;
+ }
+
+ ConstantRange LHSRange = LHSVal.getConstantRange();
+ ConstantRange RHSRange(1);
+ IntegerType *ResultTy = cast<IntegerType>(BBI->getType());
+ if (isa<BinaryOperator>(BBI)) {
+ if (ConstantInt *RHS = dyn_cast<ConstantInt>(BBI->getOperand(1))) {
+ RHSRange = ConstantRange(RHS->getValue());
+ } else {
+ BBLV.markOverdefined();
+ return true;
+ }
+ }
+
+ // NOTE: We're currently limited by the set of operations that ConstantRange
+ // can evaluate symbolically. Enhancing that set will allows us to analyze
+ // more definitions.
+ LVILatticeVal Result;
+ switch (BBI->getOpcode()) {
+ case Instruction::Add:
+ Result.markConstantRange(LHSRange.add(RHSRange));
+ break;
+ case Instruction::Sub:
+ Result.markConstantRange(LHSRange.sub(RHSRange));
+ break;
+ case Instruction::Mul:
+ Result.markConstantRange(LHSRange.multiply(RHSRange));
+ break;
+ case Instruction::UDiv:
+ Result.markConstantRange(LHSRange.udiv(RHSRange));
+ break;
+ case Instruction::Shl:
+ Result.markConstantRange(LHSRange.shl(RHSRange));
+ break;
+ case Instruction::LShr:
+ Result.markConstantRange(LHSRange.lshr(RHSRange));
+ break;
+ case Instruction::Trunc:
+ Result.markConstantRange(LHSRange.truncate(ResultTy->getBitWidth()));
+ break;
+ case Instruction::SExt:
+ Result.markConstantRange(LHSRange.signExtend(ResultTy->getBitWidth()));
+ break;
+ case Instruction::ZExt:
+ Result.markConstantRange(LHSRange.zeroExtend(ResultTy->getBitWidth()));
+ break;
+ case Instruction::BitCast:
+ Result.markConstantRange(LHSRange);
+ break;
+ case Instruction::And:
+ Result.markConstantRange(LHSRange.binaryAnd(RHSRange));
+ break;
+ case Instruction::Or:
+ Result.markConstantRange(LHSRange.binaryOr(RHSRange));
+ break;
+
+ // Unhandled instructions are overdefined.
+ default:
+ DEBUG(dbgs() << " compute BB '" << BB->getName()
+ << "' - overdefined because inst def found.\n");
+ Result.markOverdefined();
+ break;
+ }
+
+ BBLV = Result;
+ return true;
+}
+
+/// getEdgeValue - This method attempts to infer more complex
+bool LazyValueInfoCache::getEdgeValue(Value *Val, BasicBlock *BBFrom,
+ BasicBlock *BBTo, LVILatticeVal &Result) {
+ // If already a constant, there is nothing to compute.
+ if (Constant *VC = dyn_cast<Constant>(Val)) {
+ Result = LVILatticeVal::get(VC);
+ return true;
+ }
+
+ // TODO: Handle more complex conditionals. If (v == 0 || v2 < 1) is false, we
+ // know that v != 0.
+ if (BranchInst *BI = dyn_cast<BranchInst>(BBFrom->getTerminator())) {
+ // If this is a conditional branch and only one successor goes to BBTo, then
+ // we maybe able to infer something from the condition.
+ if (BI->isConditional() &&
+ BI->getSuccessor(0) != BI->getSuccessor(1)) {
+ bool isTrueDest = BI->getSuccessor(0) == BBTo;
+ assert(BI->getSuccessor(!isTrueDest) == BBTo &&
+ "BBTo isn't a successor of BBFrom");
+
+ // If V is the condition of the branch itself, then we know exactly what
+ // it is.
+ if (BI->getCondition() == Val) {
+ Result = LVILatticeVal::get(ConstantInt::get(
+ Type::getInt1Ty(Val->getContext()), isTrueDest));
+ return true;
+ }
+
+ // If the condition of the branch is an equality comparison, we may be
+ // able to infer the value.
+ ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition());
+ if (ICI && ICI->getOperand(0) == Val &&
+ isa<Constant>(ICI->getOperand(1))) {
+ if (ICI->isEquality()) {
+ // We know that V has the RHS constant if this is a true SETEQ or
+ // false SETNE.
+ if (isTrueDest == (ICI->getPredicate() == ICmpInst::ICMP_EQ))
+ Result = LVILatticeVal::get(cast<Constant>(ICI->getOperand(1)));
+ else
+ Result = LVILatticeVal::getNot(cast<Constant>(ICI->getOperand(1)));
+ return true;
+ }
+
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(ICI->getOperand(1))) {
+ // Calculate the range of values that would satisfy the comparison.
+ ConstantRange CmpRange(CI->getValue(), CI->getValue()+1);
+ ConstantRange TrueValues =
+ ConstantRange::makeICmpRegion(ICI->getPredicate(), CmpRange);
+
+ // If we're interested in the false dest, invert the condition.
+ if (!isTrueDest) TrueValues = TrueValues.inverse();
+
+ // Figure out the possible values of the query BEFORE this branch.
+ if (!hasBlockValue(Val, BBFrom)) {
+ BlockValueStack.push(std::make_pair(BBFrom, Val));
+ return false;
+ }
+
+ LVILatticeVal InBlock = getBlockValue(Val, BBFrom);
+ if (!InBlock.isConstantRange()) {
+ Result = LVILatticeVal::getRange(TrueValues);
+ return true;
+ }
+
+ // Find all potential values that satisfy both the input and output
+ // conditions.
+ ConstantRange PossibleValues =
+ TrueValues.intersectWith(InBlock.getConstantRange());
+
+ Result = LVILatticeVal::getRange(PossibleValues);
+ return true;
+ }
+ }
+ }
+ }
+
+ // If the edge was formed by a switch on the value, then we may know exactly
+ // what it is.
+ if (SwitchInst *SI = dyn_cast<SwitchInst>(BBFrom->getTerminator())) {
+ if (SI->getCondition() == Val) {
+ // We don't know anything in the default case.
+ if (SI->getDefaultDest() == BBTo) {
+ Result.markOverdefined();
+ return true;
+ }
+
+ // We only know something if there is exactly one value that goes from
+ // BBFrom to BBTo.
+ unsigned NumEdges = 0;
+ ConstantInt *EdgeVal = 0;
+ for (unsigned i = 1, e = SI->getNumSuccessors(); i != e; ++i) {
+ if (SI->getSuccessor(i) != BBTo) continue;
+ if (NumEdges++) break;
+ EdgeVal = SI->getCaseValue(i);
+ }
+ assert(EdgeVal && "Missing successor?");
+ if (NumEdges == 1) {
+ Result = LVILatticeVal::get(EdgeVal);
+ return true;
+ }
+ }
+ }
+
+ // Otherwise see if the value is known in the block.
+ if (hasBlockValue(Val, BBFrom)) {
+ Result = getBlockValue(Val, BBFrom);
+ return true;
+ }
+ BlockValueStack.push(std::make_pair(BBFrom, Val));
+ return false;
+}
+
+LVILatticeVal LazyValueInfoCache::getValueInBlock(Value *V, BasicBlock *BB) {
+ DEBUG(dbgs() << "LVI Getting block end value " << *V << " at '"
+ << BB->getName() << "'\n");
+
+ BlockValueStack.push(std::make_pair(BB, V));
+ solve();
+ LVILatticeVal Result = getBlockValue(V, BB);
+
+ DEBUG(dbgs() << " Result = " << Result << "\n");
+ return Result;
+}
+
+LVILatticeVal LazyValueInfoCache::
+getValueOnEdge(Value *V, BasicBlock *FromBB, BasicBlock *ToBB) {
+ DEBUG(dbgs() << "LVI Getting edge value " << *V << " from '"
+ << FromBB->getName() << "' to '" << ToBB->getName() << "'\n");
+
+ LVILatticeVal Result;
+ if (!getEdgeValue(V, FromBB, ToBB, Result)) {
+ solve();
+ bool WasFastQuery = getEdgeValue(V, FromBB, ToBB, Result);
+ (void)WasFastQuery;
+ assert(WasFastQuery && "More work to do after problem solved?");
+ }
+
+ DEBUG(dbgs() << " Result = " << Result << "\n");
+ return Result;
+}
+
+void LazyValueInfoCache::threadEdge(BasicBlock *PredBB, BasicBlock *OldSucc,
+ BasicBlock *NewSucc) {
+ // When an edge in the graph has been threaded, values that we could not
+ // determine a value for before (i.e. were marked overdefined) may be possible
+ // to solve now. We do NOT try to proactively update these values. Instead,
+ // we clear their entries from the cache, and allow lazy updating to recompute
+ // them when needed.
+
+ // The updating process is fairly simple: we need to dropped cached info
+ // for all values that were marked overdefined in OldSucc, and for those same
+ // values in any successor of OldSucc (except NewSucc) in which they were
+ // also marked overdefined.
+ std::vector<BasicBlock*> worklist;
+ worklist.push_back(OldSucc);
+
+ DenseSet<Value*> ClearSet;
+ for (DenseSet<OverDefinedPairTy>::iterator I = OverDefinedCache.begin(),
+ E = OverDefinedCache.end(); I != E; ++I) {
+ if (I->first == OldSucc)
+ ClearSet.insert(I->second);
+ }
+
+ // Use a worklist to perform a depth-first search of OldSucc's successors.
+ // NOTE: We do not need a visited list since any blocks we have already
+ // visited will have had their overdefined markers cleared already, and we
+ // thus won't loop to their successors.
+ while (!worklist.empty()) {
+ BasicBlock *ToUpdate = worklist.back();
+ worklist.pop_back();
+
+ // Skip blocks only accessible through NewSucc.
+ if (ToUpdate == NewSucc) continue;
+
+ bool changed = false;
+ for (DenseSet<Value*>::iterator I = ClearSet.begin(), E = ClearSet.end();
+ I != E; ++I) {
+ // If a value was marked overdefined in OldSucc, and is here too...
+ DenseSet<OverDefinedPairTy>::iterator OI =
+ OverDefinedCache.find(std::make_pair(ToUpdate, *I));
+ if (OI == OverDefinedCache.end()) continue;
+
+ // Remove it from the caches.
+ ValueCacheEntryTy &Entry = ValueCache[LVIValueHandle(*I, this)];
+ ValueCacheEntryTy::iterator CI = Entry.find(ToUpdate);
+
+ assert(CI != Entry.end() && "Couldn't find entry to update?");
+ Entry.erase(CI);
+ OverDefinedCache.erase(OI);
+
+ // If we removed anything, then we potentially need to update
+ // blocks successors too.
+ changed = true;
+ }
+
+ if (!changed) continue;
+
+ worklist.insert(worklist.end(), succ_begin(ToUpdate), succ_end(ToUpdate));
+ }
+}
+
+//===----------------------------------------------------------------------===//
+// LazyValueInfo Impl
+//===----------------------------------------------------------------------===//
+
+/// getCache - This lazily constructs the LazyValueInfoCache.
+static LazyValueInfoCache &getCache(void *&PImpl) {
+ if (!PImpl)
+ PImpl = new LazyValueInfoCache();
+ return *static_cast<LazyValueInfoCache*>(PImpl);
+}
+
+bool LazyValueInfo::runOnFunction(Function &F) {
+ if (PImpl)
+ getCache(PImpl).clear();
+
+ TD = getAnalysisIfAvailable<TargetData>();
+ // Fully lazy.
+ return false;
+}
+
+void LazyValueInfo::releaseMemory() {
+ // If the cache was allocated, free it.
+ if (PImpl) {
+ delete &getCache(PImpl);
+ PImpl = 0;
+ }
+}
+
+Constant *LazyValueInfo::getConstant(Value *V, BasicBlock *BB) {
+ LVILatticeVal Result = getCache(PImpl).getValueInBlock(V, BB);
+
+ if (Result.isConstant())
+ return Result.getConstant();
+ if (Result.isConstantRange()) {
+ ConstantRange CR = Result.getConstantRange();
+ if (const APInt *SingleVal = CR.getSingleElement())
+ return ConstantInt::get(V->getContext(), *SingleVal);
+ }
+ return 0;
+}
+
+/// getConstantOnEdge - Determine whether the specified value is known to be a
+/// constant on the specified edge. Return null if not.
+Constant *LazyValueInfo::getConstantOnEdge(Value *V, BasicBlock *FromBB,
+ BasicBlock *ToBB) {
+ LVILatticeVal Result = getCache(PImpl).getValueOnEdge(V, FromBB, ToBB);
+
+ if (Result.isConstant())
+ return Result.getConstant();
+ if (Result.isConstantRange()) {
+ ConstantRange CR = Result.getConstantRange();
+ if (const APInt *SingleVal = CR.getSingleElement())
+ return ConstantInt::get(V->getContext(), *SingleVal);
+ }
+ return 0;
+}
+
+/// getPredicateOnEdge - Determine whether the specified value comparison
+/// with a constant is known to be true or false on the specified CFG edge.
+/// Pred is a CmpInst predicate.
+LazyValueInfo::Tristate
+LazyValueInfo::getPredicateOnEdge(unsigned Pred, Value *V, Constant *C,
+ BasicBlock *FromBB, BasicBlock *ToBB) {
+ LVILatticeVal Result = getCache(PImpl).getValueOnEdge(V, FromBB, ToBB);
+
+ // If we know the value is a constant, evaluate the conditional.
+ Constant *Res = 0;
+ if (Result.isConstant()) {
+ Res = ConstantFoldCompareInstOperands(Pred, Result.getConstant(), C, TD);
+ if (ConstantInt *ResCI = dyn_cast<ConstantInt>(Res))
+ return ResCI->isZero() ? False : True;
+ return Unknown;
+ }
+
+ if (Result.isConstantRange()) {
+ ConstantInt *CI = dyn_cast<ConstantInt>(C);
+ if (!CI) return Unknown;
+
+ ConstantRange CR = Result.getConstantRange();
+ if (Pred == ICmpInst::ICMP_EQ) {
+ if (!CR.contains(CI->getValue()))
+ return False;
+
+ if (CR.isSingleElement() && CR.contains(CI->getValue()))
+ return True;
+ } else if (Pred == ICmpInst::ICMP_NE) {
+ if (!CR.contains(CI->getValue()))
+ return True;
+
+ if (CR.isSingleElement() && CR.contains(CI->getValue()))
+ return False;
+ }
+
+ // Handle more complex predicates.
+ ConstantRange TrueValues =
+ ICmpInst::makeConstantRange((ICmpInst::Predicate)Pred, CI->getValue());
+ if (TrueValues.contains(CR))
+ return True;
+ if (TrueValues.inverse().contains(CR))
+ return False;
+ return Unknown;
+ }
+
+ if (Result.isNotConstant()) {
+ // If this is an equality comparison, we can try to fold it knowing that
+ // "V != C1".
+ if (Pred == ICmpInst::ICMP_EQ) {
+ // !C1 == C -> false iff C1 == C.
+ Res = ConstantFoldCompareInstOperands(ICmpInst::ICMP_NE,
+ Result.getNotConstant(), C, TD);
+ if (Res->isNullValue())
+ return False;
+ } else if (Pred == ICmpInst::ICMP_NE) {
+ // !C1 != C -> true iff C1 == C.
+ Res = ConstantFoldCompareInstOperands(ICmpInst::ICMP_NE,
+ Result.getNotConstant(), C, TD);
+ if (Res->isNullValue())
+ return True;
+ }
+ return Unknown;
+ }
+
+ return Unknown;
+}
+
+void LazyValueInfo::threadEdge(BasicBlock *PredBB, BasicBlock *OldSucc,
+ BasicBlock *NewSucc) {
+ if (PImpl) getCache(PImpl).threadEdge(PredBB, OldSucc, NewSucc);
+}
+
+void LazyValueInfo::eraseBlock(BasicBlock *BB) {
+ if (PImpl) getCache(PImpl).eraseBlock(BB);
+}
diff --git a/src/LLVM/lib/Analysis/LibCallAliasAnalysis.cpp b/src/LLVM/lib/Analysis/LibCallAliasAnalysis.cpp
new file mode 100644
index 0000000..efb722b
--- /dev/null
+++ b/src/LLVM/lib/Analysis/LibCallAliasAnalysis.cpp
@@ -0,0 +1,137 @@
+//===- LibCallAliasAnalysis.cpp - Implement AliasAnalysis for libcalls ----===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the LibCallAliasAnalysis class.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/LibCallAliasAnalysis.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/LibCallSemantics.h"
+#include "llvm/Function.h"
+#include "llvm/Pass.h"
+using namespace llvm;
+
+// Register this pass...
+char LibCallAliasAnalysis::ID = 0;
+INITIALIZE_AG_PASS(LibCallAliasAnalysis, AliasAnalysis, "libcall-aa",
+ "LibCall Alias Analysis", false, true, false)
+
+FunctionPass *llvm::createLibCallAliasAnalysisPass(LibCallInfo *LCI) {
+ return new LibCallAliasAnalysis(LCI);
+}
+
+LibCallAliasAnalysis::~LibCallAliasAnalysis() {
+ delete LCI;
+}
+
+void LibCallAliasAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
+ AliasAnalysis::getAnalysisUsage(AU);
+ AU.setPreservesAll(); // Does not transform code
+}
+
+
+
+/// AnalyzeLibCallDetails - Given a call to a function with the specified
+/// LibCallFunctionInfo, see if we can improve the mod/ref footprint of the call
+/// vs the specified pointer/size.
+AliasAnalysis::ModRefResult
+LibCallAliasAnalysis::AnalyzeLibCallDetails(const LibCallFunctionInfo *FI,
+ ImmutableCallSite CS,
+ const Location &Loc) {
+ // If we have a function, check to see what kind of mod/ref effects it
+ // has. Start by including any info globally known about the function.
+ AliasAnalysis::ModRefResult MRInfo = FI->UniversalBehavior;
+ if (MRInfo == NoModRef) return MRInfo;
+
+ // If that didn't tell us that the function is 'readnone', check to see
+ // if we have detailed info and if 'P' is any of the locations we know
+ // about.
+ const LibCallFunctionInfo::LocationMRInfo *Details = FI->LocationDetails;
+ if (Details == 0)
+ return MRInfo;
+
+ // If the details array is of the 'DoesNot' kind, we only know something if
+ // the pointer is a match for one of the locations in 'Details'. If we find a
+ // match, we can prove some interactions cannot happen.
+ //
+ if (FI->DetailsType == LibCallFunctionInfo::DoesNot) {
+ // Find out if the pointer refers to a known location.
+ for (unsigned i = 0; Details[i].LocationID != ~0U; ++i) {
+ const LibCallLocationInfo &LocInfo =
+ LCI->getLocationInfo(Details[i].LocationID);
+ LibCallLocationInfo::LocResult Res = LocInfo.isLocation(CS, Loc);
+ if (Res != LibCallLocationInfo::Yes) continue;
+
+ // If we find a match against a location that we 'do not' interact with,
+ // learn this info into MRInfo.
+ return ModRefResult(MRInfo & ~Details[i].MRInfo);
+ }
+ return MRInfo;
+ }
+
+ // If the details are of the 'DoesOnly' sort, we know something if the pointer
+ // is a match for one of the locations in 'Details'. Also, if we can prove
+ // that the pointers is *not* one of the locations in 'Details', we know that
+ // the call is NoModRef.
+ assert(FI->DetailsType == LibCallFunctionInfo::DoesOnly);
+
+ // Find out if the pointer refers to a known location.
+ bool NoneMatch = true;
+ for (unsigned i = 0; Details[i].LocationID != ~0U; ++i) {
+ const LibCallLocationInfo &LocInfo =
+ LCI->getLocationInfo(Details[i].LocationID);
+ LibCallLocationInfo::LocResult Res = LocInfo.isLocation(CS, Loc);
+ if (Res == LibCallLocationInfo::No) continue;
+
+ // If we don't know if this pointer points to the location, then we have to
+ // assume it might alias in some case.
+ if (Res == LibCallLocationInfo::Unknown) {
+ NoneMatch = false;
+ continue;
+ }
+
+ // If we know that this pointer definitely is pointing into the location,
+ // merge in this information.
+ return ModRefResult(MRInfo & Details[i].MRInfo);
+ }
+
+ // If we found that the pointer is guaranteed to not match any of the
+ // locations in our 'DoesOnly' rule, then we know that the pointer must point
+ // to some other location. Since the libcall doesn't mod/ref any other
+ // locations, return NoModRef.
+ if (NoneMatch)
+ return NoModRef;
+
+ // Otherwise, return any other info gained so far.
+ return MRInfo;
+}
+
+// getModRefInfo - Check to see if the specified callsite can clobber the
+// specified memory object.
+//
+AliasAnalysis::ModRefResult
+LibCallAliasAnalysis::getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc) {
+ ModRefResult MRInfo = ModRef;
+
+ // If this is a direct call to a function that LCI knows about, get the
+ // information about the runtime function.
+ if (LCI) {
+ if (const Function *F = CS.getCalledFunction()) {
+ if (const LibCallFunctionInfo *FI = LCI->getFunctionInfo(F)) {
+ MRInfo = ModRefResult(MRInfo & AnalyzeLibCallDetails(FI, CS, Loc));
+ if (MRInfo == NoModRef) return NoModRef;
+ }
+ }
+ }
+
+ // The AliasAnalysis base class has some smarts, lets use them.
+ return (ModRefResult)(MRInfo | AliasAnalysis::getModRefInfo(CS, Loc));
+}
diff --git a/src/LLVM/lib/Analysis/LibCallSemantics.cpp b/src/LLVM/lib/Analysis/LibCallSemantics.cpp
new file mode 100644
index 0000000..81b0f46
--- /dev/null
+++ b/src/LLVM/lib/Analysis/LibCallSemantics.cpp
@@ -0,0 +1,63 @@
+//===- LibCallSemantics.cpp - Describe library semantics ------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements interfaces that can be used to describe language
+// specific runtime library interfaces (e.g. libc, libm, etc) to LLVM
+// optimizers.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/LibCallSemantics.h"
+#include "llvm/ADT/StringMap.h"
+#include "llvm/Function.h"
+using namespace llvm;
+
+/// getMap - This impl pointer in ~LibCallInfo is actually a StringMap. This
+/// helper does the cast.
+static StringMap<const LibCallFunctionInfo*> *getMap(void *Ptr) {
+ return static_cast<StringMap<const LibCallFunctionInfo*> *>(Ptr);
+}
+
+LibCallInfo::~LibCallInfo() {
+ delete getMap(Impl);
+}
+
+const LibCallLocationInfo &LibCallInfo::getLocationInfo(unsigned LocID) const {
+ // Get location info on the first call.
+ if (NumLocations == 0)
+ NumLocations = getLocationInfo(Locations);
+
+ assert(LocID < NumLocations && "Invalid location ID!");
+ return Locations[LocID];
+}
+
+
+/// getFunctionInfo - Return the LibCallFunctionInfo object corresponding to
+/// the specified function if we have it. If not, return null.
+const LibCallFunctionInfo *
+LibCallInfo::getFunctionInfo(const Function *F) const {
+ StringMap<const LibCallFunctionInfo*> *Map = getMap(Impl);
+
+ /// If this is the first time we are querying for this info, lazily construct
+ /// the StringMap to index it.
+ if (Map == 0) {
+ Impl = Map = new StringMap<const LibCallFunctionInfo*>();
+
+ const LibCallFunctionInfo *Array = getFunctionInfoArray();
+ if (Array == 0) return 0;
+
+ // We now have the array of entries. Populate the StringMap.
+ for (unsigned i = 0; Array[i].Name; ++i)
+ (*Map)[Array[i].Name] = Array+i;
+ }
+
+ // Look up this function in the string map.
+ return Map->lookup(F->getName());
+}
+
diff --git a/src/LLVM/lib/Analysis/Lint.cpp b/src/LLVM/lib/Analysis/Lint.cpp
new file mode 100644
index 0000000..38d677d
--- /dev/null
+++ b/src/LLVM/lib/Analysis/Lint.cpp
@@ -0,0 +1,655 @@
+//===-- Lint.cpp - Check for common errors in LLVM IR ---------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass statically checks for common and easily-identified constructs
+// which produce undefined or likely unintended behavior in LLVM IR.
+//
+// It is not a guarantee of correctness, in two ways. First, it isn't
+// comprehensive. There are checks which could be done statically which are
+// not yet implemented. Some of these are indicated by TODO comments, but
+// those aren't comprehensive either. Second, many conditions cannot be
+// checked statically. This pass does no dynamic instrumentation, so it
+// can't check for all possible problems.
+//
+// Another limitation is that it assumes all code will be executed. A store
+// through a null pointer in a basic block which is never reached is harmless,
+// but this pass will warn about it anyway. This is the main reason why most
+// of these checks live here instead of in the Verifier pass.
+//
+// Optimization passes may make conditions that this pass checks for more or
+// less obvious. If an optimization pass appears to be introducing a warning,
+// it may be that the optimization pass is merely exposing an existing
+// condition in the code.
+//
+// This code may be run before instcombine. In many cases, instcombine checks
+// for the same kinds of things and turns instructions with undefined behavior
+// into unreachable (or equivalent). Because of this, this pass makes some
+// effort to look through bitcasts and so on.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/InstructionSimplify.h"
+#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/Lint.h"
+#include "llvm/Analysis/Loads.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Pass.h"
+#include "llvm/PassManager.h"
+#include "llvm/IntrinsicInst.h"
+#include "llvm/Function.h"
+#include "llvm/Support/CallSite.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/InstVisitor.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/STLExtras.h"
+using namespace llvm;
+
+namespace {
+ namespace MemRef {
+ static unsigned Read = 1;
+ static unsigned Write = 2;
+ static unsigned Callee = 4;
+ static unsigned Branchee = 8;
+ }
+
+ class Lint : public FunctionPass, public InstVisitor<Lint> {
+ friend class InstVisitor<Lint>;
+
+ void visitFunction(Function &F);
+
+ void visitCallSite(CallSite CS);
+ void visitMemoryReference(Instruction &I, Value *Ptr,
+ uint64_t Size, unsigned Align,
+ Type *Ty, unsigned Flags);
+
+ void visitCallInst(CallInst &I);
+ void visitInvokeInst(InvokeInst &I);
+ void visitReturnInst(ReturnInst &I);
+ void visitLoadInst(LoadInst &I);
+ void visitStoreInst(StoreInst &I);
+ void visitXor(BinaryOperator &I);
+ void visitSub(BinaryOperator &I);
+ void visitLShr(BinaryOperator &I);
+ void visitAShr(BinaryOperator &I);
+ void visitShl(BinaryOperator &I);
+ void visitSDiv(BinaryOperator &I);
+ void visitUDiv(BinaryOperator &I);
+ void visitSRem(BinaryOperator &I);
+ void visitURem(BinaryOperator &I);
+ void visitAllocaInst(AllocaInst &I);
+ void visitVAArgInst(VAArgInst &I);
+ void visitIndirectBrInst(IndirectBrInst &I);
+ void visitExtractElementInst(ExtractElementInst &I);
+ void visitInsertElementInst(InsertElementInst &I);
+ void visitUnreachableInst(UnreachableInst &I);
+
+ Value *findValue(Value *V, bool OffsetOk) const;
+ Value *findValueImpl(Value *V, bool OffsetOk,
+ SmallPtrSet<Value *, 4> &Visited) const;
+
+ public:
+ Module *Mod;
+ AliasAnalysis *AA;
+ DominatorTree *DT;
+ TargetData *TD;
+
+ std::string Messages;
+ raw_string_ostream MessagesStr;
+
+ static char ID; // Pass identification, replacement for typeid
+ Lint() : FunctionPass(ID), MessagesStr(Messages) {
+ initializeLintPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnFunction(Function &F);
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ AU.addRequired<AliasAnalysis>();
+ AU.addRequired<DominatorTree>();
+ }
+ virtual void print(raw_ostream &O, const Module *M) const {}
+
+ void WriteValue(const Value *V) {
+ if (!V) return;
+ if (isa<Instruction>(V)) {
+ MessagesStr << *V << '\n';
+ } else {
+ WriteAsOperand(MessagesStr, V, true, Mod);
+ MessagesStr << '\n';
+ }
+ }
+
+ // CheckFailed - A check failed, so print out the condition and the message
+ // that failed. This provides a nice place to put a breakpoint if you want
+ // to see why something is not correct.
+ void CheckFailed(const Twine &Message,
+ const Value *V1 = 0, const Value *V2 = 0,
+ const Value *V3 = 0, const Value *V4 = 0) {
+ MessagesStr << Message.str() << "\n";
+ WriteValue(V1);
+ WriteValue(V2);
+ WriteValue(V3);
+ WriteValue(V4);
+ }
+ };
+}
+
+char Lint::ID = 0;
+INITIALIZE_PASS_BEGIN(Lint, "lint", "Statically lint-checks LLVM IR",
+ false, true)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_END(Lint, "lint", "Statically lint-checks LLVM IR",
+ false, true)
+
+// Assert - We know that cond should be true, if not print an error message.
+#define Assert(C, M) \
+ do { if (!(C)) { CheckFailed(M); return; } } while (0)
+#define Assert1(C, M, V1) \
+ do { if (!(C)) { CheckFailed(M, V1); return; } } while (0)
+#define Assert2(C, M, V1, V2) \
+ do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0)
+#define Assert3(C, M, V1, V2, V3) \
+ do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0)
+#define Assert4(C, M, V1, V2, V3, V4) \
+ do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0)
+
+// Lint::run - This is the main Analysis entry point for a
+// function.
+//
+bool Lint::runOnFunction(Function &F) {
+ Mod = F.getParent();
+ AA = &getAnalysis<AliasAnalysis>();
+ DT = &getAnalysis<DominatorTree>();
+ TD = getAnalysisIfAvailable<TargetData>();
+ visit(F);
+ dbgs() << MessagesStr.str();
+ Messages.clear();
+ return false;
+}
+
+void Lint::visitFunction(Function &F) {
+ // This isn't undefined behavior, it's just a little unusual, and it's a
+ // fairly common mistake to neglect to name a function.
+ Assert1(F.hasName() || F.hasLocalLinkage(),
+ "Unusual: Unnamed function with non-local linkage", &F);
+
+ // TODO: Check for irreducible control flow.
+}
+
+void Lint::visitCallSite(CallSite CS) {
+ Instruction &I = *CS.getInstruction();
+ Value *Callee = CS.getCalledValue();
+
+ visitMemoryReference(I, Callee, AliasAnalysis::UnknownSize,
+ 0, 0, MemRef::Callee);
+
+ if (Function *F = dyn_cast<Function>(findValue(Callee, /*OffsetOk=*/false))) {
+ Assert1(CS.getCallingConv() == F->getCallingConv(),
+ "Undefined behavior: Caller and callee calling convention differ",
+ &I);
+
+ FunctionType *FT = F->getFunctionType();
+ unsigned NumActualArgs = unsigned(CS.arg_end()-CS.arg_begin());
+
+ Assert1(FT->isVarArg() ?
+ FT->getNumParams() <= NumActualArgs :
+ FT->getNumParams() == NumActualArgs,
+ "Undefined behavior: Call argument count mismatches callee "
+ "argument count", &I);
+
+ Assert1(FT->getReturnType() == I.getType(),
+ "Undefined behavior: Call return type mismatches "
+ "callee return type", &I);
+
+ // Check argument types (in case the callee was casted) and attributes.
+ // TODO: Verify that caller and callee attributes are compatible.
+ Function::arg_iterator PI = F->arg_begin(), PE = F->arg_end();
+ CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
+ for (; AI != AE; ++AI) {
+ Value *Actual = *AI;
+ if (PI != PE) {
+ Argument *Formal = PI++;
+ Assert1(Formal->getType() == Actual->getType(),
+ "Undefined behavior: Call argument type mismatches "
+ "callee parameter type", &I);
+
+ // Check that noalias arguments don't alias other arguments. This is
+ // not fully precise because we don't know the sizes of the dereferenced
+ // memory regions.
+ if (Formal->hasNoAliasAttr() && Actual->getType()->isPointerTy())
+ for (CallSite::arg_iterator BI = CS.arg_begin(); BI != AE; ++BI)
+ if (AI != BI && (*BI)->getType()->isPointerTy()) {
+ AliasAnalysis::AliasResult Result = AA->alias(*AI, *BI);
+ Assert1(Result != AliasAnalysis::MustAlias &&
+ Result != AliasAnalysis::PartialAlias,
+ "Unusual: noalias argument aliases another argument", &I);
+ }
+
+ // Check that an sret argument points to valid memory.
+ if (Formal->hasStructRetAttr() && Actual->getType()->isPointerTy()) {
+ Type *Ty =
+ cast<PointerType>(Formal->getType())->getElementType();
+ visitMemoryReference(I, Actual, AA->getTypeStoreSize(Ty),
+ TD ? TD->getABITypeAlignment(Ty) : 0,
+ Ty, MemRef::Read | MemRef::Write);
+ }
+ }
+ }
+ }
+
+ if (CS.isCall() && cast<CallInst>(CS.getInstruction())->isTailCall())
+ for (CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
+ AI != AE; ++AI) {
+ Value *Obj = findValue(*AI, /*OffsetOk=*/true);
+ Assert1(!isa<AllocaInst>(Obj),
+ "Undefined behavior: Call with \"tail\" keyword references "
+ "alloca", &I);
+ }
+
+
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(&I))
+ switch (II->getIntrinsicID()) {
+ default: break;
+
+ // TODO: Check more intrinsics
+
+ case Intrinsic::memcpy: {
+ MemCpyInst *MCI = cast<MemCpyInst>(&I);
+ // TODO: If the size is known, use it.
+ visitMemoryReference(I, MCI->getDest(), AliasAnalysis::UnknownSize,
+ MCI->getAlignment(), 0,
+ MemRef::Write);
+ visitMemoryReference(I, MCI->getSource(), AliasAnalysis::UnknownSize,
+ MCI->getAlignment(), 0,
+ MemRef::Read);
+
+ // Check that the memcpy arguments don't overlap. The AliasAnalysis API
+ // isn't expressive enough for what we really want to do. Known partial
+ // overlap is not distinguished from the case where nothing is known.
+ uint64_t Size = 0;
+ if (const ConstantInt *Len =
+ dyn_cast<ConstantInt>(findValue(MCI->getLength(),
+ /*OffsetOk=*/false)))
+ if (Len->getValue().isIntN(32))
+ Size = Len->getValue().getZExtValue();
+ Assert1(AA->alias(MCI->getSource(), Size, MCI->getDest(), Size) !=
+ AliasAnalysis::MustAlias,
+ "Undefined behavior: memcpy source and destination overlap", &I);
+ break;
+ }
+ case Intrinsic::memmove: {
+ MemMoveInst *MMI = cast<MemMoveInst>(&I);
+ // TODO: If the size is known, use it.
+ visitMemoryReference(I, MMI->getDest(), AliasAnalysis::UnknownSize,
+ MMI->getAlignment(), 0,
+ MemRef::Write);
+ visitMemoryReference(I, MMI->getSource(), AliasAnalysis::UnknownSize,
+ MMI->getAlignment(), 0,
+ MemRef::Read);
+ break;
+ }
+ case Intrinsic::memset: {
+ MemSetInst *MSI = cast<MemSetInst>(&I);
+ // TODO: If the size is known, use it.
+ visitMemoryReference(I, MSI->getDest(), AliasAnalysis::UnknownSize,
+ MSI->getAlignment(), 0,
+ MemRef::Write);
+ break;
+ }
+
+ case Intrinsic::vastart:
+ Assert1(I.getParent()->getParent()->isVarArg(),
+ "Undefined behavior: va_start called in a non-varargs function",
+ &I);
+
+ visitMemoryReference(I, CS.getArgument(0), AliasAnalysis::UnknownSize,
+ 0, 0, MemRef::Read | MemRef::Write);
+ break;
+ case Intrinsic::vacopy:
+ visitMemoryReference(I, CS.getArgument(0), AliasAnalysis::UnknownSize,
+ 0, 0, MemRef::Write);
+ visitMemoryReference(I, CS.getArgument(1), AliasAnalysis::UnknownSize,
+ 0, 0, MemRef::Read);
+ break;
+ case Intrinsic::vaend:
+ visitMemoryReference(I, CS.getArgument(0), AliasAnalysis::UnknownSize,
+ 0, 0, MemRef::Read | MemRef::Write);
+ break;
+
+ case Intrinsic::stackrestore:
+ // Stackrestore doesn't read or write memory, but it sets the
+ // stack pointer, which the compiler may read from or write to
+ // at any time, so check it for both readability and writeability.
+ visitMemoryReference(I, CS.getArgument(0), AliasAnalysis::UnknownSize,
+ 0, 0, MemRef::Read | MemRef::Write);
+ break;
+ }
+}
+
+void Lint::visitCallInst(CallInst &I) {
+ return visitCallSite(&I);
+}
+
+void Lint::visitInvokeInst(InvokeInst &I) {
+ return visitCallSite(&I);
+}
+
+void Lint::visitReturnInst(ReturnInst &I) {
+ Function *F = I.getParent()->getParent();
+ Assert1(!F->doesNotReturn(),
+ "Unusual: Return statement in function with noreturn attribute",
+ &I);
+
+ if (Value *V = I.getReturnValue()) {
+ Value *Obj = findValue(V, /*OffsetOk=*/true);
+ Assert1(!isa<AllocaInst>(Obj),
+ "Unusual: Returning alloca value", &I);
+ }
+}
+
+// TODO: Check that the reference is in bounds.
+// TODO: Check readnone/readonly function attributes.
+void Lint::visitMemoryReference(Instruction &I,
+ Value *Ptr, uint64_t Size, unsigned Align,
+ Type *Ty, unsigned Flags) {
+ // If no memory is being referenced, it doesn't matter if the pointer
+ // is valid.
+ if (Size == 0)
+ return;
+
+ Value *UnderlyingObject = findValue(Ptr, /*OffsetOk=*/true);
+ Assert1(!isa<ConstantPointerNull>(UnderlyingObject),
+ "Undefined behavior: Null pointer dereference", &I);
+ Assert1(!isa<UndefValue>(UnderlyingObject),
+ "Undefined behavior: Undef pointer dereference", &I);
+ Assert1(!isa<ConstantInt>(UnderlyingObject) ||
+ !cast<ConstantInt>(UnderlyingObject)->isAllOnesValue(),
+ "Unusual: All-ones pointer dereference", &I);
+ Assert1(!isa<ConstantInt>(UnderlyingObject) ||
+ !cast<ConstantInt>(UnderlyingObject)->isOne(),
+ "Unusual: Address one pointer dereference", &I);
+
+ if (Flags & MemRef::Write) {
+ if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(UnderlyingObject))
+ Assert1(!GV->isConstant(),
+ "Undefined behavior: Write to read-only memory", &I);
+ Assert1(!isa<Function>(UnderlyingObject) &&
+ !isa<BlockAddress>(UnderlyingObject),
+ "Undefined behavior: Write to text section", &I);
+ }
+ if (Flags & MemRef::Read) {
+ Assert1(!isa<Function>(UnderlyingObject),
+ "Unusual: Load from function body", &I);
+ Assert1(!isa<BlockAddress>(UnderlyingObject),
+ "Undefined behavior: Load from block address", &I);
+ }
+ if (Flags & MemRef::Callee) {
+ Assert1(!isa<BlockAddress>(UnderlyingObject),
+ "Undefined behavior: Call to block address", &I);
+ }
+ if (Flags & MemRef::Branchee) {
+ Assert1(!isa<Constant>(UnderlyingObject) ||
+ isa<BlockAddress>(UnderlyingObject),
+ "Undefined behavior: Branch to non-blockaddress", &I);
+ }
+
+ if (TD) {
+ if (Align == 0 && Ty) Align = TD->getABITypeAlignment(Ty);
+
+ if (Align != 0) {
+ unsigned BitWidth = TD->getTypeSizeInBits(Ptr->getType());
+ APInt Mask = APInt::getAllOnesValue(BitWidth),
+ KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
+ ComputeMaskedBits(Ptr, Mask, KnownZero, KnownOne, TD);
+ Assert1(!(KnownOne & APInt::getLowBitsSet(BitWidth, Log2_32(Align))),
+ "Undefined behavior: Memory reference address is misaligned", &I);
+ }
+ }
+}
+
+void Lint::visitLoadInst(LoadInst &I) {
+ visitMemoryReference(I, I.getPointerOperand(),
+ AA->getTypeStoreSize(I.getType()), I.getAlignment(),
+ I.getType(), MemRef::Read);
+}
+
+void Lint::visitStoreInst(StoreInst &I) {
+ visitMemoryReference(I, I.getPointerOperand(),
+ AA->getTypeStoreSize(I.getOperand(0)->getType()),
+ I.getAlignment(),
+ I.getOperand(0)->getType(), MemRef::Write);
+}
+
+void Lint::visitXor(BinaryOperator &I) {
+ Assert1(!isa<UndefValue>(I.getOperand(0)) ||
+ !isa<UndefValue>(I.getOperand(1)),
+ "Undefined result: xor(undef, undef)", &I);
+}
+
+void Lint::visitSub(BinaryOperator &I) {
+ Assert1(!isa<UndefValue>(I.getOperand(0)) ||
+ !isa<UndefValue>(I.getOperand(1)),
+ "Undefined result: sub(undef, undef)", &I);
+}
+
+void Lint::visitLShr(BinaryOperator &I) {
+ if (ConstantInt *CI =
+ dyn_cast<ConstantInt>(findValue(I.getOperand(1), /*OffsetOk=*/false)))
+ Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
+ "Undefined result: Shift count out of range", &I);
+}
+
+void Lint::visitAShr(BinaryOperator &I) {
+ if (ConstantInt *CI =
+ dyn_cast<ConstantInt>(findValue(I.getOperand(1), /*OffsetOk=*/false)))
+ Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
+ "Undefined result: Shift count out of range", &I);
+}
+
+void Lint::visitShl(BinaryOperator &I) {
+ if (ConstantInt *CI =
+ dyn_cast<ConstantInt>(findValue(I.getOperand(1), /*OffsetOk=*/false)))
+ Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
+ "Undefined result: Shift count out of range", &I);
+}
+
+static bool isZero(Value *V, TargetData *TD) {
+ // Assume undef could be zero.
+ if (isa<UndefValue>(V)) return true;
+
+ unsigned BitWidth = cast<IntegerType>(V->getType())->getBitWidth();
+ APInt Mask = APInt::getAllOnesValue(BitWidth),
+ KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
+ ComputeMaskedBits(V, Mask, KnownZero, KnownOne, TD);
+ return KnownZero.isAllOnesValue();
+}
+
+void Lint::visitSDiv(BinaryOperator &I) {
+ Assert1(!isZero(I.getOperand(1), TD),
+ "Undefined behavior: Division by zero", &I);
+}
+
+void Lint::visitUDiv(BinaryOperator &I) {
+ Assert1(!isZero(I.getOperand(1), TD),
+ "Undefined behavior: Division by zero", &I);
+}
+
+void Lint::visitSRem(BinaryOperator &I) {
+ Assert1(!isZero(I.getOperand(1), TD),
+ "Undefined behavior: Division by zero", &I);
+}
+
+void Lint::visitURem(BinaryOperator &I) {
+ Assert1(!isZero(I.getOperand(1), TD),
+ "Undefined behavior: Division by zero", &I);
+}
+
+void Lint::visitAllocaInst(AllocaInst &I) {
+ if (isa<ConstantInt>(I.getArraySize()))
+ // This isn't undefined behavior, it's just an obvious pessimization.
+ Assert1(&I.getParent()->getParent()->getEntryBlock() == I.getParent(),
+ "Pessimization: Static alloca outside of entry block", &I);
+
+ // TODO: Check for an unusual size (MSB set?)
+}
+
+void Lint::visitVAArgInst(VAArgInst &I) {
+ visitMemoryReference(I, I.getOperand(0), AliasAnalysis::UnknownSize, 0, 0,
+ MemRef::Read | MemRef::Write);
+}
+
+void Lint::visitIndirectBrInst(IndirectBrInst &I) {
+ visitMemoryReference(I, I.getAddress(), AliasAnalysis::UnknownSize, 0, 0,
+ MemRef::Branchee);
+
+ Assert1(I.getNumDestinations() != 0,
+ "Undefined behavior: indirectbr with no destinations", &I);
+}
+
+void Lint::visitExtractElementInst(ExtractElementInst &I) {
+ if (ConstantInt *CI =
+ dyn_cast<ConstantInt>(findValue(I.getIndexOperand(),
+ /*OffsetOk=*/false)))
+ Assert1(CI->getValue().ult(I.getVectorOperandType()->getNumElements()),
+ "Undefined result: extractelement index out of range", &I);
+}
+
+void Lint::visitInsertElementInst(InsertElementInst &I) {
+ if (ConstantInt *CI =
+ dyn_cast<ConstantInt>(findValue(I.getOperand(2),
+ /*OffsetOk=*/false)))
+ Assert1(CI->getValue().ult(I.getType()->getNumElements()),
+ "Undefined result: insertelement index out of range", &I);
+}
+
+void Lint::visitUnreachableInst(UnreachableInst &I) {
+ // This isn't undefined behavior, it's merely suspicious.
+ Assert1(&I == I.getParent()->begin() ||
+ prior(BasicBlock::iterator(&I))->mayHaveSideEffects(),
+ "Unusual: unreachable immediately preceded by instruction without "
+ "side effects", &I);
+}
+
+/// findValue - Look through bitcasts and simple memory reference patterns
+/// to identify an equivalent, but more informative, value. If OffsetOk
+/// is true, look through getelementptrs with non-zero offsets too.
+///
+/// Most analysis passes don't require this logic, because instcombine
+/// will simplify most of these kinds of things away. But it's a goal of
+/// this Lint pass to be useful even on non-optimized IR.
+Value *Lint::findValue(Value *V, bool OffsetOk) const {
+ SmallPtrSet<Value *, 4> Visited;
+ return findValueImpl(V, OffsetOk, Visited);
+}
+
+/// findValueImpl - Implementation helper for findValue.
+Value *Lint::findValueImpl(Value *V, bool OffsetOk,
+ SmallPtrSet<Value *, 4> &Visited) const {
+ // Detect self-referential values.
+ if (!Visited.insert(V))
+ return UndefValue::get(V->getType());
+
+ // TODO: Look through sext or zext cast, when the result is known to
+ // be interpreted as signed or unsigned, respectively.
+ // TODO: Look through eliminable cast pairs.
+ // TODO: Look through calls with unique return values.
+ // TODO: Look through vector insert/extract/shuffle.
+ V = OffsetOk ? GetUnderlyingObject(V, TD) : V->stripPointerCasts();
+ if (LoadInst *L = dyn_cast<LoadInst>(V)) {
+ BasicBlock::iterator BBI = L;
+ BasicBlock *BB = L->getParent();
+ SmallPtrSet<BasicBlock *, 4> VisitedBlocks;
+ for (;;) {
+ if (!VisitedBlocks.insert(BB)) break;
+ if (Value *U = FindAvailableLoadedValue(L->getPointerOperand(),
+ BB, BBI, 6, AA))
+ return findValueImpl(U, OffsetOk, Visited);
+ if (BBI != BB->begin()) break;
+ BB = BB->getUniquePredecessor();
+ if (!BB) break;
+ BBI = BB->end();
+ }
+ } else if (PHINode *PN = dyn_cast<PHINode>(V)) {
+ if (Value *W = PN->hasConstantValue())
+ if (W != V)
+ return findValueImpl(W, OffsetOk, Visited);
+ } else if (CastInst *CI = dyn_cast<CastInst>(V)) {
+ if (CI->isNoopCast(TD ? TD->getIntPtrType(V->getContext()) :
+ Type::getInt64Ty(V->getContext())))
+ return findValueImpl(CI->getOperand(0), OffsetOk, Visited);
+ } else if (ExtractValueInst *Ex = dyn_cast<ExtractValueInst>(V)) {
+ if (Value *W = FindInsertedValue(Ex->getAggregateOperand(),
+ Ex->getIndices()))
+ if (W != V)
+ return findValueImpl(W, OffsetOk, Visited);
+ } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
+ // Same as above, but for ConstantExpr instead of Instruction.
+ if (Instruction::isCast(CE->getOpcode())) {
+ if (CastInst::isNoopCast(Instruction::CastOps(CE->getOpcode()),
+ CE->getOperand(0)->getType(),
+ CE->getType(),
+ TD ? TD->getIntPtrType(V->getContext()) :
+ Type::getInt64Ty(V->getContext())))
+ return findValueImpl(CE->getOperand(0), OffsetOk, Visited);
+ } else if (CE->getOpcode() == Instruction::ExtractValue) {
+ ArrayRef<unsigned> Indices = CE->getIndices();
+ if (Value *W = FindInsertedValue(CE->getOperand(0), Indices))
+ if (W != V)
+ return findValueImpl(W, OffsetOk, Visited);
+ }
+ }
+
+ // As a last resort, try SimplifyInstruction or constant folding.
+ if (Instruction *Inst = dyn_cast<Instruction>(V)) {
+ if (Value *W = SimplifyInstruction(Inst, TD, DT))
+ return findValueImpl(W, OffsetOk, Visited);
+ } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
+ if (Value *W = ConstantFoldConstantExpression(CE, TD))
+ if (W != V)
+ return findValueImpl(W, OffsetOk, Visited);
+ }
+
+ return V;
+}
+
+//===----------------------------------------------------------------------===//
+// Implement the public interfaces to this file...
+//===----------------------------------------------------------------------===//
+
+FunctionPass *llvm::createLintPass() {
+ return new Lint();
+}
+
+/// lintFunction - Check a function for errors, printing messages on stderr.
+///
+void llvm::lintFunction(const Function &f) {
+ Function &F = const_cast<Function&>(f);
+ assert(!F.isDeclaration() && "Cannot lint external functions");
+
+ FunctionPassManager FPM(F.getParent());
+ Lint *V = new Lint();
+ FPM.add(V);
+ FPM.run(F);
+}
+
+/// lintModule - Check a module for errors, printing messages on stderr.
+///
+void llvm::lintModule(const Module &M) {
+ PassManager PM;
+ Lint *V = new Lint();
+ PM.add(V);
+ PM.run(const_cast<Module&>(M));
+}
diff --git a/src/LLVM/lib/Analysis/Loads.cpp b/src/LLVM/lib/Analysis/Loads.cpp
index 45dff8b..0e6bcbf 100644
--- a/src/LLVM/lib/Analysis/Loads.cpp
+++ b/src/LLVM/lib/Analysis/Loads.cpp
@@ -17,6 +17,7 @@
#include "llvm/GlobalAlias.h"
#include "llvm/GlobalVariable.h"
#include "llvm/IntrinsicInst.h"
+#include "llvm/Operator.h"
using namespace llvm;
/// AreEquivalentAddressValues - Test if A and B will obviously have the same
@@ -30,7 +31,7 @@
static bool AreEquivalentAddressValues(const Value *A, const Value *B) {
// Test if the values are trivially equivalent.
if (A == B) return true;
-
+
// Test if the values come from identical arithmetic instructions.
// Use isIdenticalToWhenDefined instead of isIdenticalTo because
// this function is only used when one address use dominates the
@@ -41,7 +42,7 @@
if (const Instruction *BI = dyn_cast<Instruction>(B))
if (cast<Instruction>(A)->isIdenticalToWhenDefined(BI))
return true;
-
+
// Otherwise they may not be equivalent.
return false;
}
@@ -49,7 +50,7 @@
/// getUnderlyingObjectWithOffset - Strip off up to MaxLookup GEPs and
/// bitcasts to get back to the underlying object being addressed, keeping
/// track of the offset in bytes from the GEPs relative to the result.
-/// This is closely related to Value::getUnderlyingObject but is located
+/// This is closely related to GetUnderlyingObject but is located
/// here to avoid making VMCore depend on TargetData.
static Value *getUnderlyingObjectWithOffset(Value *V, const TargetData *TD,
uint64_t &ByteOffset,
@@ -62,7 +63,7 @@
return V;
SmallVector<Value*, 8> Indices(GEP->op_begin() + 1, GEP->op_end());
ByteOffset += TD->getIndexedOffset(GEP->getPointerOperandType(),
- &Indices[0], Indices.size());
+ Indices);
V = GEP->getPointerOperand();
} else if (Operator::getOpcode(V) == Instruction::BitCast) {
V = cast<Operator>(V)->getOperand(0);
@@ -89,7 +90,7 @@
if (TD)
Base = getUnderlyingObjectWithOffset(V, TD, ByteOffset);
- const Type *BaseType = 0;
+ Type *BaseType = 0;
unsigned BaseAlign = 0;
if (const AllocaInst *AI = dyn_cast<AllocaInst>(Base)) {
// An alloca is safe to load from as load as it is suitably aligned.
@@ -113,7 +114,7 @@
return true; // Loading directly from an alloca or global is OK.
// Check if the load is within the bounds of the underlying object.
- const PointerType *AddrTy = cast<PointerType>(V->getType());
+ PointerType *AddrTy = cast<PointerType>(V->getType());
uint64_t LoadSize = TD->getTypeStoreSize(AddrTy->getElementType());
if (ByteOffset + LoadSize <= TD->getTypeAllocSize(BaseType) &&
(Align == 0 || (ByteOffset % Align) == 0))
@@ -134,7 +135,7 @@
// If we see a free or a call which may write to memory (i.e. which might do
// a free) the pointer could be marked invalid.
if (isa<CallInst>(BBI) && BBI->mayWriteToMemory() &&
- !ISA_DEBUG_INFO_INTRINSIC(BBI))
+ !isa<DbgInfoIntrinsic>(BBI))
return false;
if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
@@ -166,9 +167,9 @@
if (MaxInstsToScan == 0) MaxInstsToScan = ~0U;
// If we're using alias analysis to disambiguate get the size of *Ptr.
- unsigned AccessSize = 0;
+ uint64_t AccessSize = 0;
if (AA) {
- const Type *AccessTy = cast<PointerType>(Ptr->getType())->getElementType();
+ Type *AccessTy = cast<PointerType>(Ptr->getType())->getElementType();
AccessSize = AA->getTypeStoreSize(AccessTy);
}
@@ -176,7 +177,7 @@
// We must ignore debug info directives when counting (otherwise they
// would affect codegen).
Instruction *Inst = --ScanFrom;
- if (ISA_DEBUG_INFO_INTRINSIC(Inst))
+ if (isa<DbgInfoIntrinsic>(Inst))
continue;
// Restore ScanFrom to expected value in case next test succeeds
@@ -187,12 +188,16 @@
--ScanFrom;
// If this is a load of Ptr, the loaded value is available.
+ // (This is true even if the load is volatile or atomic, although
+ // those cases are unlikely.)
if (LoadInst *LI = dyn_cast<LoadInst>(Inst))
if (AreEquivalentAddressValues(LI->getOperand(0), Ptr))
return LI;
if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
// If this is a store through Ptr, the value is available!
+ // (This is true even if the store is volatile or atomic, although
+ // those cases are unlikely.)
if (AreEquivalentAddressValues(SI->getOperand(1), Ptr))
return SI->getOperand(0);
diff --git a/src/LLVM/lib/Analysis/LoopDependenceAnalysis.cpp b/src/LLVM/lib/Analysis/LoopDependenceAnalysis.cpp
new file mode 100644
index 0000000..3997ac4
--- /dev/null
+++ b/src/LLVM/lib/Analysis/LoopDependenceAnalysis.cpp
@@ -0,0 +1,364 @@
+//===- LoopDependenceAnalysis.cpp - LDA Implementation ----------*- C++ -*-===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This is the (beginning) of an implementation of a loop dependence analysis
+// framework, which is used to detect dependences in memory accesses in loops.
+//
+// Please note that this is work in progress and the interface is subject to
+// change.
+//
+// TODO: adapt as implementation progresses.
+//
+// TODO: document lingo (pair, subscript, index)
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "lda"
+#include "llvm/ADT/DenseSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/LoopDependenceAnalysis.h"
+#include "llvm/Analysis/LoopPass.h"
+#include "llvm/Analysis/ScalarEvolution.h"
+#include "llvm/Analysis/ScalarEvolutionExpressions.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Instructions.h"
+#include "llvm/Operator.h"
+#include "llvm/Support/Allocator.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetData.h"
+using namespace llvm;
+
+STATISTIC(NumAnswered, "Number of dependence queries answered");
+STATISTIC(NumAnalysed, "Number of distinct dependence pairs analysed");
+STATISTIC(NumDependent, "Number of pairs with dependent accesses");
+STATISTIC(NumIndependent, "Number of pairs with independent accesses");
+STATISTIC(NumUnknown, "Number of pairs with unknown accesses");
+
+LoopPass *llvm::createLoopDependenceAnalysisPass() {
+ return new LoopDependenceAnalysis();
+}
+
+INITIALIZE_PASS_BEGIN(LoopDependenceAnalysis, "lda",
+ "Loop Dependence Analysis", false, true)
+INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
+INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_END(LoopDependenceAnalysis, "lda",
+ "Loop Dependence Analysis", false, true)
+char LoopDependenceAnalysis::ID = 0;
+
+//===----------------------------------------------------------------------===//
+// Utility Functions
+//===----------------------------------------------------------------------===//
+
+static inline bool IsMemRefInstr(const Value *V) {
+ const Instruction *I = dyn_cast<const Instruction>(V);
+ return I && (I->mayReadFromMemory() || I->mayWriteToMemory());
+}
+
+static void GetMemRefInstrs(const Loop *L,
+ SmallVectorImpl<Instruction*> &Memrefs) {
+ for (Loop::block_iterator b = L->block_begin(), be = L->block_end();
+ b != be; ++b)
+ for (BasicBlock::iterator i = (*b)->begin(), ie = (*b)->end();
+ i != ie; ++i)
+ if (IsMemRefInstr(i))
+ Memrefs.push_back(i);
+}
+
+static bool IsLoadOrStoreInst(Value *I) {
+ // Returns true if the load or store can be analyzed. Atomic and volatile
+ // operations have properties which this analysis does not understand.
+ if (LoadInst *LI = dyn_cast<LoadInst>(I))
+ return LI->isUnordered();
+ else if (StoreInst *SI = dyn_cast<StoreInst>(I))
+ return SI->isUnordered();
+ return false;
+}
+
+static Value *GetPointerOperand(Value *I) {
+ if (LoadInst *i = dyn_cast<LoadInst>(I))
+ return i->getPointerOperand();
+ if (StoreInst *i = dyn_cast<StoreInst>(I))
+ return i->getPointerOperand();
+ llvm_unreachable("Value is no load or store instruction!");
+ // Never reached.
+ return 0;
+}
+
+static AliasAnalysis::AliasResult UnderlyingObjectsAlias(AliasAnalysis *AA,
+ const Value *A,
+ const Value *B) {
+ const Value *aObj = GetUnderlyingObject(A);
+ const Value *bObj = GetUnderlyingObject(B);
+ return AA->alias(aObj, AA->getTypeStoreSize(aObj->getType()),
+ bObj, AA->getTypeStoreSize(bObj->getType()));
+}
+
+static inline const SCEV *GetZeroSCEV(ScalarEvolution *SE) {
+ return SE->getConstant(Type::getInt32Ty(SE->getContext()), 0L);
+}
+
+//===----------------------------------------------------------------------===//
+// Dependence Testing
+//===----------------------------------------------------------------------===//
+
+bool LoopDependenceAnalysis::isDependencePair(const Value *A,
+ const Value *B) const {
+ return IsMemRefInstr(A) &&
+ IsMemRefInstr(B) &&
+ (cast<const Instruction>(A)->mayWriteToMemory() ||
+ cast<const Instruction>(B)->mayWriteToMemory());
+}
+
+bool LoopDependenceAnalysis::findOrInsertDependencePair(Value *A,
+ Value *B,
+ DependencePair *&P) {
+ void *insertPos = 0;
+ FoldingSetNodeID id;
+ id.AddPointer(A);
+ id.AddPointer(B);
+
+ P = Pairs.FindNodeOrInsertPos(id, insertPos);
+ if (P) return true;
+
+ P = new (PairAllocator) DependencePair(id, A, B);
+ Pairs.InsertNode(P, insertPos);
+ return false;
+}
+
+void LoopDependenceAnalysis::getLoops(const SCEV *S,
+ DenseSet<const Loop*>* Loops) const {
+ // Refactor this into an SCEVVisitor, if efficiency becomes a concern.
+ for (const Loop *L = this->L; L != 0; L = L->getParentLoop())
+ if (!SE->isLoopInvariant(S, L))
+ Loops->insert(L);
+}
+
+bool LoopDependenceAnalysis::isLoopInvariant(const SCEV *S) const {
+ DenseSet<const Loop*> loops;
+ getLoops(S, &loops);
+ return loops.empty();
+}
+
+bool LoopDependenceAnalysis::isAffine(const SCEV *S) const {
+ const SCEVAddRecExpr *rec = dyn_cast<SCEVAddRecExpr>(S);
+ return isLoopInvariant(S) || (rec && rec->isAffine());
+}
+
+bool LoopDependenceAnalysis::isZIVPair(const SCEV *A, const SCEV *B) const {
+ return isLoopInvariant(A) && isLoopInvariant(B);
+}
+
+bool LoopDependenceAnalysis::isSIVPair(const SCEV *A, const SCEV *B) const {
+ DenseSet<const Loop*> loops;
+ getLoops(A, &loops);
+ getLoops(B, &loops);
+ return loops.size() == 1;
+}
+
+LoopDependenceAnalysis::DependenceResult
+LoopDependenceAnalysis::analyseZIV(const SCEV *A,
+ const SCEV *B,
+ Subscript *S) const {
+ assert(isZIVPair(A, B) && "Attempted to ZIV-test non-ZIV SCEVs!");
+ return A == B ? Dependent : Independent;
+}
+
+LoopDependenceAnalysis::DependenceResult
+LoopDependenceAnalysis::analyseSIV(const SCEV *A,
+ const SCEV *B,
+ Subscript *S) const {
+ return Unknown; // TODO: Implement.
+}
+
+LoopDependenceAnalysis::DependenceResult
+LoopDependenceAnalysis::analyseMIV(const SCEV *A,
+ const SCEV *B,
+ Subscript *S) const {
+ return Unknown; // TODO: Implement.
+}
+
+LoopDependenceAnalysis::DependenceResult
+LoopDependenceAnalysis::analyseSubscript(const SCEV *A,
+ const SCEV *B,
+ Subscript *S) const {
+ DEBUG(dbgs() << " Testing subscript: " << *A << ", " << *B << "\n");
+
+ if (A == B) {
+ DEBUG(dbgs() << " -> [D] same SCEV\n");
+ return Dependent;
+ }
+
+ if (!isAffine(A) || !isAffine(B)) {
+ DEBUG(dbgs() << " -> [?] not affine\n");
+ return Unknown;
+ }
+
+ if (isZIVPair(A, B))
+ return analyseZIV(A, B, S);
+
+ if (isSIVPair(A, B))
+ return analyseSIV(A, B, S);
+
+ return analyseMIV(A, B, S);
+}
+
+LoopDependenceAnalysis::DependenceResult
+LoopDependenceAnalysis::analysePair(DependencePair *P) const {
+ DEBUG(dbgs() << "Analysing:\n" << *P->A << "\n" << *P->B << "\n");
+
+ // We only analyse loads and stores but no possible memory accesses by e.g.
+ // free, call, or invoke instructions.
+ if (!IsLoadOrStoreInst(P->A) || !IsLoadOrStoreInst(P->B)) {
+ DEBUG(dbgs() << "--> [?] no load/store\n");
+ return Unknown;
+ }
+
+ Value *aPtr = GetPointerOperand(P->A);
+ Value *bPtr = GetPointerOperand(P->B);
+
+ switch (UnderlyingObjectsAlias(AA, aPtr, bPtr)) {
+ case AliasAnalysis::MayAlias:
+ case AliasAnalysis::PartialAlias:
+ // We can not analyse objects if we do not know about their aliasing.
+ DEBUG(dbgs() << "---> [?] may alias\n");
+ return Unknown;
+
+ case AliasAnalysis::NoAlias:
+ // If the objects noalias, they are distinct, accesses are independent.
+ DEBUG(dbgs() << "---> [I] no alias\n");
+ return Independent;
+
+ case AliasAnalysis::MustAlias:
+ break; // The underlying objects alias, test accesses for dependence.
+ }
+
+ const GEPOperator *aGEP = dyn_cast<GEPOperator>(aPtr);
+ const GEPOperator *bGEP = dyn_cast<GEPOperator>(bPtr);
+
+ if (!aGEP || !bGEP)
+ return Unknown;
+
+ // FIXME: Is filtering coupled subscripts necessary?
+
+ // Collect GEP operand pairs (FIXME: use GetGEPOperands from BasicAA), adding
+ // trailing zeroes to the smaller GEP, if needed.
+ typedef SmallVector<std::pair<const SCEV*, const SCEV*>, 4> GEPOpdPairsTy;
+ GEPOpdPairsTy opds;
+ for(GEPOperator::const_op_iterator aIdx = aGEP->idx_begin(),
+ aEnd = aGEP->idx_end(),
+ bIdx = bGEP->idx_begin(),
+ bEnd = bGEP->idx_end();
+ aIdx != aEnd && bIdx != bEnd;
+ aIdx += (aIdx != aEnd), bIdx += (bIdx != bEnd)) {
+ const SCEV* aSCEV = (aIdx != aEnd) ? SE->getSCEV(*aIdx) : GetZeroSCEV(SE);
+ const SCEV* bSCEV = (bIdx != bEnd) ? SE->getSCEV(*bIdx) : GetZeroSCEV(SE);
+ opds.push_back(std::make_pair(aSCEV, bSCEV));
+ }
+
+ if (!opds.empty() && opds[0].first != opds[0].second) {
+ // We cannot (yet) handle arbitrary GEP pointer offsets. By limiting
+ //
+ // TODO: this could be relaxed by adding the size of the underlying object
+ // to the first subscript. If we have e.g. (GEP x,0,i; GEP x,2,-i) and we
+ // know that x is a [100 x i8]*, we could modify the first subscript to be
+ // (i, 200-i) instead of (i, -i).
+ return Unknown;
+ }
+
+ // Now analyse the collected operand pairs (skipping the GEP ptr offsets).
+ for (GEPOpdPairsTy::const_iterator i = opds.begin() + 1, end = opds.end();
+ i != end; ++i) {
+ Subscript subscript;
+ DependenceResult result = analyseSubscript(i->first, i->second, &subscript);
+ if (result != Dependent) {
+ // We either proved independence or failed to analyse this subscript.
+ // Further subscripts will not improve the situation, so abort early.
+ return result;
+ }
+ P->Subscripts.push_back(subscript);
+ }
+ // We successfully analysed all subscripts but failed to prove independence.
+ return Dependent;
+}
+
+bool LoopDependenceAnalysis::depends(Value *A, Value *B) {
+ assert(isDependencePair(A, B) && "Values form no dependence pair!");
+ ++NumAnswered;
+
+ DependencePair *p;
+ if (!findOrInsertDependencePair(A, B, p)) {
+ // The pair is not cached, so analyse it.
+ ++NumAnalysed;
+ switch (p->Result = analysePair(p)) {
+ case Dependent: ++NumDependent; break;
+ case Independent: ++NumIndependent; break;
+ case Unknown: ++NumUnknown; break;
+ }
+ }
+ return p->Result != Independent;
+}
+
+//===----------------------------------------------------------------------===//
+// LoopDependenceAnalysis Implementation
+//===----------------------------------------------------------------------===//
+
+bool LoopDependenceAnalysis::runOnLoop(Loop *L, LPPassManager &) {
+ this->L = L;
+ AA = &getAnalysis<AliasAnalysis>();
+ SE = &getAnalysis<ScalarEvolution>();
+ return false;
+}
+
+void LoopDependenceAnalysis::releaseMemory() {
+ Pairs.clear();
+ PairAllocator.Reset();
+}
+
+void LoopDependenceAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ AU.addRequiredTransitive<AliasAnalysis>();
+ AU.addRequiredTransitive<ScalarEvolution>();
+}
+
+static void PrintLoopInfo(raw_ostream &OS,
+ LoopDependenceAnalysis *LDA, const Loop *L) {
+ if (!L->empty()) return; // ignore non-innermost loops
+
+ SmallVector<Instruction*, 8> memrefs;
+ GetMemRefInstrs(L, memrefs);
+
+ OS << "Loop at depth " << L->getLoopDepth() << ", header block: ";
+ WriteAsOperand(OS, L->getHeader(), false);
+ OS << "\n";
+
+ OS << " Load/store instructions: " << memrefs.size() << "\n";
+ for (SmallVector<Instruction*, 8>::const_iterator x = memrefs.begin(),
+ end = memrefs.end(); x != end; ++x)
+ OS << "\t" << (x - memrefs.begin()) << ": " << **x << "\n";
+
+ OS << " Pairwise dependence results:\n";
+ for (SmallVector<Instruction*, 8>::const_iterator x = memrefs.begin(),
+ end = memrefs.end(); x != end; ++x)
+ for (SmallVector<Instruction*, 8>::const_iterator y = x + 1;
+ y != end; ++y)
+ if (LDA->isDependencePair(*x, *y))
+ OS << "\t" << (x - memrefs.begin()) << "," << (y - memrefs.begin())
+ << ": " << (LDA->depends(*x, *y) ? "dependent" : "independent")
+ << "\n";
+}
+
+void LoopDependenceAnalysis::print(raw_ostream &OS, const Module*) const {
+ // TODO: doc why const_cast is safe
+ PrintLoopInfo(OS, const_cast<LoopDependenceAnalysis*>(this), this->L);
+}
diff --git a/src/LLVM/lib/Analysis/LoopInfo.cpp b/src/LLVM/lib/Analysis/LoopInfo.cpp
index 46219d1..85aacca 100644
--- a/src/LLVM/lib/Analysis/LoopInfo.cpp
+++ b/src/LLVM/lib/Analysis/LoopInfo.cpp
@@ -18,6 +18,7 @@
#include "llvm/Constants.h"
#include "llvm/Instructions.h"
#include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/LoopIterator.h"
#include "llvm/Assembly/Writer.h"
#include "llvm/Support/CFG.h"
#include "llvm/Support/CommandLine.h"
@@ -38,7 +39,9 @@
cl::desc("Verify loop info (time consuming)"));
char LoopInfo::ID = 0;
-INITIALIZE_PASS(LoopInfo, "loops", "Natural Loop Information", true, true);
+INITIALIZE_PASS_BEGIN(LoopInfo, "loops", "Natural Loop Information", true, true)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_PASS_END(LoopInfo, "loops", "Natural Loop Information", true, true)
//===----------------------------------------------------------------------===//
// Loop implementation
@@ -48,15 +51,18 @@
///
bool Loop::isLoopInvariant(Value *V) const {
if (Instruction *I = dyn_cast<Instruction>(V))
- return isLoopInvariant(I);
+ return !contains(I);
return true; // All non-instructions are loop invariant
}
-/// isLoopInvariant - Return true if the specified instruction is
-/// loop-invariant.
-///
-bool Loop::isLoopInvariant(Instruction *I) const {
- return !contains(I);
+/// hasLoopInvariantOperands - Return true if all the operands of the
+/// specified instruction are loop invariant.
+bool Loop::hasLoopInvariantOperands(Instruction *I) const {
+ for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
+ if (!isLoopInvariant(I->getOperand(i)))
+ return false;
+
+ return true;
}
/// makeLoopInvariant - If the given value is an instruciton inside of the
@@ -93,6 +99,9 @@
return false;
if (I->mayReadFromMemory())
return false;
+ // The landingpad instruction is immobile.
+ if (isa<LandingPadInst>(I))
+ return false;
// Determine the insertion point, unless one was given.
if (!InsertPt) {
BasicBlock *Preheader = getLoopPreheader();
@@ -105,6 +114,7 @@
for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
if (!makeLoopInvariant(I->getOperand(i), Changed, InsertPt))
return false;
+
// Hoist.
I->moveBefore(InsertPt);
Changed = true;
@@ -192,7 +202,7 @@
/// getSmallConstantTripCount - Returns the trip count of this loop as a
/// normal unsigned value, if possible. Returns 0 if the trip count is unknown
-/// of not constant. Will also return 0 if the trip count is very large
+/// or not constant. Will also return 0 if the trip count is very large
/// (>= 2^32)
unsigned Loop::getSmallConstantTripCount() const {
Value* TripCount = this->getTripCount();
@@ -377,6 +387,205 @@
}
//===----------------------------------------------------------------------===//
+// UnloopUpdater implementation
+//
+
+namespace {
+/// Find the new parent loop for all blocks within the "unloop" whose last
+/// backedges has just been removed.
+class UnloopUpdater {
+ Loop *Unloop;
+ LoopInfo *LI;
+
+ LoopBlocksDFS DFS;
+
+ // Map unloop's immediate subloops to their nearest reachable parents. Nested
+ // loops within these subloops will not change parents. However, an immediate
+ // subloop's new parent will be the nearest loop reachable from either its own
+ // exits *or* any of its nested loop's exits.
+ DenseMap<Loop*, Loop*> SubloopParents;
+
+ // Flag the presence of an irreducible backedge whose destination is a block
+ // directly contained by the original unloop.
+ bool FoundIB;
+
+public:
+ UnloopUpdater(Loop *UL, LoopInfo *LInfo) :
+ Unloop(UL), LI(LInfo), DFS(UL), FoundIB(false) {}
+
+ void updateBlockParents();
+
+ void removeBlocksFromAncestors();
+
+ void updateSubloopParents();
+
+protected:
+ Loop *getNearestLoop(BasicBlock *BB, Loop *BBLoop);
+};
+} // end anonymous namespace
+
+/// updateBlockParents - Update the parent loop for all blocks that are directly
+/// contained within the original "unloop".
+void UnloopUpdater::updateBlockParents() {
+ if (Unloop->getNumBlocks()) {
+ // Perform a post order CFG traversal of all blocks within this loop,
+ // propagating the nearest loop from sucessors to predecessors.
+ LoopBlocksTraversal Traversal(DFS, LI);
+ for (LoopBlocksTraversal::POTIterator POI = Traversal.begin(),
+ POE = Traversal.end(); POI != POE; ++POI) {
+
+ Loop *L = LI->getLoopFor(*POI);
+ Loop *NL = getNearestLoop(*POI, L);
+
+ if (NL != L) {
+ // For reducible loops, NL is now an ancestor of Unloop.
+ assert((NL != Unloop && (!NL || NL->contains(Unloop))) &&
+ "uninitialized successor");
+ LI->changeLoopFor(*POI, NL);
+ }
+ else {
+ // Or the current block is part of a subloop, in which case its parent
+ // is unchanged.
+ assert((FoundIB || Unloop->contains(L)) && "uninitialized successor");
+ }
+ }
+ }
+ // Each irreducible loop within the unloop induces a round of iteration using
+ // the DFS result cached by Traversal.
+ bool Changed = FoundIB;
+ for (unsigned NIters = 0; Changed; ++NIters) {
+ assert(NIters < Unloop->getNumBlocks() && "runaway iterative algorithm");
+
+ // Iterate over the postorder list of blocks, propagating the nearest loop
+ // from successors to predecessors as before.
+ Changed = false;
+ for (LoopBlocksDFS::POIterator POI = DFS.beginPostorder(),
+ POE = DFS.endPostorder(); POI != POE; ++POI) {
+
+ Loop *L = LI->getLoopFor(*POI);
+ Loop *NL = getNearestLoop(*POI, L);
+ if (NL != L) {
+ assert(NL != Unloop && (!NL || NL->contains(Unloop)) &&
+ "uninitialized successor");
+ LI->changeLoopFor(*POI, NL);
+ Changed = true;
+ }
+ }
+ }
+}
+
+/// removeBlocksFromAncestors - Remove unloop's blocks from all ancestors below
+/// their new parents.
+void UnloopUpdater::removeBlocksFromAncestors() {
+ // Remove unloop's blocks from all ancestors below their new parents.
+ for (Loop::block_iterator BI = Unloop->block_begin(),
+ BE = Unloop->block_end(); BI != BE; ++BI) {
+ Loop *NewParent = LI->getLoopFor(*BI);
+ // If this block is an immediate subloop, remove all blocks (including
+ // nested subloops) from ancestors below the new parent loop.
+ // Otherwise, if this block is in a nested subloop, skip it.
+ if (SubloopParents.count(NewParent))
+ NewParent = SubloopParents[NewParent];
+ else if (Unloop->contains(NewParent))
+ continue;
+
+ // Remove blocks from former Ancestors except Unloop itself which will be
+ // deleted.
+ for (Loop *OldParent = Unloop->getParentLoop(); OldParent != NewParent;
+ OldParent = OldParent->getParentLoop()) {
+ assert(OldParent && "new loop is not an ancestor of the original");
+ OldParent->removeBlockFromLoop(*BI);
+ }
+ }
+}
+
+/// updateSubloopParents - Update the parent loop for all subloops directly
+/// nested within unloop.
+void UnloopUpdater::updateSubloopParents() {
+ while (!Unloop->empty()) {
+ Loop *Subloop = *llvm::prior(Unloop->end());
+ Unloop->removeChildLoop(llvm::prior(Unloop->end()));
+
+ assert(SubloopParents.count(Subloop) && "DFS failed to visit subloop");
+ if (SubloopParents[Subloop])
+ SubloopParents[Subloop]->addChildLoop(Subloop);
+ else
+ LI->addTopLevelLoop(Subloop);
+ }
+}
+
+/// getNearestLoop - Return the nearest parent loop among this block's
+/// successors. If a successor is a subloop header, consider its parent to be
+/// the nearest parent of the subloop's exits.
+///
+/// For subloop blocks, simply update SubloopParents and return NULL.
+Loop *UnloopUpdater::getNearestLoop(BasicBlock *BB, Loop *BBLoop) {
+
+ // Initially for blocks directly contained by Unloop, NearLoop == Unloop and
+ // is considered uninitialized.
+ Loop *NearLoop = BBLoop;
+
+ Loop *Subloop = 0;
+ if (NearLoop != Unloop && Unloop->contains(NearLoop)) {
+ Subloop = NearLoop;
+ // Find the subloop ancestor that is directly contained within Unloop.
+ while (Subloop->getParentLoop() != Unloop) {
+ Subloop = Subloop->getParentLoop();
+ assert(Subloop && "subloop is not an ancestor of the original loop");
+ }
+ // Get the current nearest parent of the Subloop exits, initially Unloop.
+ if (!SubloopParents.count(Subloop))
+ SubloopParents[Subloop] = Unloop;
+ NearLoop = SubloopParents[Subloop];
+ }
+
+ succ_iterator I = succ_begin(BB), E = succ_end(BB);
+ if (I == E) {
+ assert(!Subloop && "subloop blocks must have a successor");
+ NearLoop = 0; // unloop blocks may now exit the function.
+ }
+ for (; I != E; ++I) {
+ if (*I == BB)
+ continue; // self loops are uninteresting
+
+ Loop *L = LI->getLoopFor(*I);
+ if (L == Unloop) {
+ // This successor has not been processed. This path must lead to an
+ // irreducible backedge.
+ assert((FoundIB || !DFS.hasPostorder(*I)) && "should have seen IB");
+ FoundIB = true;
+ }
+ if (L != Unloop && Unloop->contains(L)) {
+ // Successor is in a subloop.
+ if (Subloop)
+ continue; // Branching within subloops. Ignore it.
+
+ // BB branches from the original into a subloop header.
+ assert(L->getParentLoop() == Unloop && "cannot skip into nested loops");
+
+ // Get the current nearest parent of the Subloop's exits.
+ L = SubloopParents[L];
+ // L could be Unloop if the only exit was an irreducible backedge.
+ }
+ if (L == Unloop) {
+ continue;
+ }
+ // Handle critical edges from Unloop into a sibling loop.
+ if (L && !L->contains(Unloop)) {
+ L = L->getParentLoop();
+ }
+ // Remember the nearest parent loop among successors or subloop exits.
+ if (NearLoop == Unloop || !NearLoop || NearLoop->contains(L))
+ NearLoop = L;
+ }
+ if (Subloop) {
+ SubloopParents[Subloop] = NearLoop;
+ return BBLoop;
+ }
+ return NearLoop;
+}
+
+//===----------------------------------------------------------------------===//
// LoopInfo implementation
//
bool LoopInfo::runOnFunction(Function &) {
@@ -385,6 +594,68 @@
return false;
}
+/// updateUnloop - The last backedge has been removed from a loop--now the
+/// "unloop". Find a new parent for the blocks contained within unloop and
+/// update the loop tree. We don't necessarily have valid dominators at this
+/// point, but LoopInfo is still valid except for the removal of this loop.
+///
+/// Note that Unloop may now be an empty loop. Calling Loop::getHeader without
+/// checking first is illegal.
+void LoopInfo::updateUnloop(Loop *Unloop) {
+
+ // First handle the special case of no parent loop to simplify the algorithm.
+ if (!Unloop->getParentLoop()) {
+ // Since BBLoop had no parent, Unloop blocks are no longer in a loop.
+ for (Loop::block_iterator I = Unloop->block_begin(),
+ E = Unloop->block_end(); I != E; ++I) {
+
+ // Don't reparent blocks in subloops.
+ if (getLoopFor(*I) != Unloop)
+ continue;
+
+ // Blocks no longer have a parent but are still referenced by Unloop until
+ // the Unloop object is deleted.
+ LI.changeLoopFor(*I, 0);
+ }
+
+ // Remove the loop from the top-level LoopInfo object.
+ for (LoopInfo::iterator I = LI.begin();; ++I) {
+ assert(I != LI.end() && "Couldn't find loop");
+ if (*I == Unloop) {
+ LI.removeLoop(I);
+ break;
+ }
+ }
+
+ // Move all of the subloops to the top-level.
+ while (!Unloop->empty())
+ LI.addTopLevelLoop(Unloop->removeChildLoop(llvm::prior(Unloop->end())));
+
+ return;
+ }
+
+ // Update the parent loop for all blocks within the loop. Blocks within
+ // subloops will not change parents.
+ UnloopUpdater Updater(Unloop, this);
+ Updater.updateBlockParents();
+
+ // Remove blocks from former ancestor loops.
+ Updater.removeBlocksFromAncestors();
+
+ // Add direct subloops as children in their new parent loop.
+ Updater.updateSubloopParents();
+
+ // Remove unloop from its parent loop.
+ Loop *ParentLoop = Unloop->getParentLoop();
+ for (Loop::iterator I = ParentLoop->begin();; ++I) {
+ assert(I != ParentLoop->end() && "Couldn't find loop");
+ if (*I == Unloop) {
+ ParentLoop->removeChildLoop(I);
+ break;
+ }
+ }
+}
+
void LoopInfo::verifyAnalysis() const {
// LoopInfo is a FunctionPass, but verifying every loop in the function
// each time verifyAnalysis is called is very expensive. The
@@ -394,12 +665,21 @@
if (!VerifyLoopInfo) return;
+ DenseSet<const Loop*> Loops;
for (iterator I = begin(), E = end(); I != E; ++I) {
assert(!(*I)->getParentLoop() && "Top-level loop has a parent!");
- (*I)->verifyLoopNest();
+ (*I)->verifyLoopNest(&Loops);
}
- // TODO: check BBMap consistency.
+ // Verify that blocks are mapped to valid loops.
+ //
+ // FIXME: With an up-to-date DFS (see LoopIterator.h) and DominatorTree, we
+ // could also verify that the blocks are still in the correct loops.
+ for (DenseMap<BasicBlock*, Loop*>::const_iterator I = LI.BBMap.begin(),
+ E = LI.BBMap.end(); I != E; ++I) {
+ assert(Loops.count(I->second) && "orphaned loop");
+ assert(I->second->contains(I->first) && "orphaned block");
+ }
}
void LoopInfo::getAnalysisUsage(AnalysisUsage &AU) const {
@@ -411,3 +691,15 @@
LI.print(OS);
}
+//===----------------------------------------------------------------------===//
+// LoopBlocksDFS implementation
+//
+
+/// Traverse the loop blocks and store the DFS result.
+/// Useful for clients that just want the final DFS result and don't need to
+/// visit blocks during the initial traversal.
+void LoopBlocksDFS::perform(LoopInfo *LI) {
+ LoopBlocksTraversal Traversal(*this, LI);
+ for (LoopBlocksTraversal::POTIterator POI = Traversal.begin(),
+ POE = Traversal.end(); POI != POE; ++POI) ;
+}
diff --git a/src/LLVM/lib/Analysis/LoopPass.cpp b/src/LLVM/lib/Analysis/LoopPass.cpp
index 717e016..5ba1f40 100644
--- a/src/LLVM/lib/Analysis/LoopPass.cpp
+++ b/src/LLVM/lib/Analysis/LoopPass.cpp
@@ -14,8 +14,10 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/LoopPass.h"
+#include "llvm/DebugInfoProbe.h"
#include "llvm/Assembly/PrintModulePass.h"
#include "llvm/Support/Debug.h"
+#include "llvm/Support/ManagedStatic.h"
#include "llvm/Support/Timer.h"
using namespace llvm;
@@ -30,7 +32,6 @@
public:
static char ID;
- PrintLoopPass() : LoopPass(ID), Out(dbgs()) {}
PrintLoopPass(const std::string &B, raw_ostream &o)
: LoopPass(ID), Banner(B), Out(o) {}
@@ -53,78 +54,48 @@
}
//===----------------------------------------------------------------------===//
+// DebugInfoProbe
+
+static DebugInfoProbeInfo *TheDebugProbe;
+static void createDebugInfoProbe() {
+ if (TheDebugProbe) return;
+
+ // Constructed the first time this is called. This guarantees that the
+ // object will be constructed, if -enable-debug-info-probe is set,
+ // before static globals, thus it will be destroyed before them.
+ static ManagedStatic<DebugInfoProbeInfo> DIP;
+ TheDebugProbe = &*DIP;
+}
+
+//===----------------------------------------------------------------------===//
// LPPassManager
//
char LPPassManager::ID = 0;
-LPPassManager::LPPassManager(int Depth)
- : FunctionPass(ID), PMDataManager(Depth) {
+LPPassManager::LPPassManager()
+ : FunctionPass(ID), PMDataManager() {
skipThisLoop = false;
redoThisLoop = false;
LI = NULL;
CurrentLoop = NULL;
}
-/// Delete loop from the loop queue and loop hierarchy (LoopInfo).
+/// Delete loop from the loop queue and loop hierarchy (LoopInfo).
void LPPassManager::deleteLoopFromQueue(Loop *L) {
- if (Loop *ParentLoop = L->getParentLoop()) { // Not a top-level loop.
- // Reparent all of the blocks in this loop. Since BBLoop had a parent,
- // they are now all in it.
- for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
- I != E; ++I)
- if (LI->getLoopFor(*I) == L) // Don't change blocks in subloops.
- LI->changeLoopFor(*I, ParentLoop);
-
- // Remove the loop from its parent loop.
- for (Loop::iterator I = ParentLoop->begin(), E = ParentLoop->end();;
- ++I) {
- assert(I != E && "Couldn't find loop");
- if (*I == L) {
- ParentLoop->removeChildLoop(I);
- break;
- }
- }
-
- // Move all subloops into the parent loop.
- while (!L->empty())
- ParentLoop->addChildLoop(L->removeChildLoop(L->end()-1));
- } else {
- // Reparent all of the blocks in this loop. Since BBLoop had no parent,
- // they no longer in a loop at all.
-
- for (unsigned i = 0; i != L->getBlocks().size(); ++i) {
- // Don't change blocks in subloops.
- if (LI->getLoopFor(L->getBlocks()[i]) == L) {
- LI->removeBlock(L->getBlocks()[i]);
- --i;
- }
- }
-
- // Remove the loop from the top-level LoopInfo object.
- for (LoopInfo::iterator I = LI->begin(), E = LI->end();; ++I) {
- assert(I != E && "Couldn't find loop");
- if (*I == L) {
- LI->removeLoop(I);
- break;
- }
- }
-
- // Move all of the subloops to the top-level.
- while (!L->empty())
- LI->addTopLevelLoop(L->removeChildLoop(L->end()-1));
- }
-
- delete L;
+ LI->updateUnloop(L);
// If L is current loop then skip rest of the passes and let
// runOnFunction remove L from LQ. Otherwise, remove L from LQ now
// and continue applying other passes on CurrentLoop.
- if (CurrentLoop == L) {
+ if (CurrentLoop == L)
skipThisLoop = true;
+
+ delete L;
+
+ if (skipThisLoop)
return;
- }
for (std::deque<Loop *>::iterator I = LQ.begin(),
E = LQ.end(); I != E; ++I) {
@@ -151,10 +122,10 @@
void LPPassManager::insertLoopIntoQueue(Loop *L) {
// Insert L into loop queue
- if (L == CurrentLoop)
+ if (L == CurrentLoop)
redoLoop(L);
else if (!L->getParentLoop())
- // This is top level loop.
+ // This is top level loop.
LQ.push_front(L);
else {
// Insert L after the parent loop.
@@ -180,10 +151,10 @@
/// cloneBasicBlockSimpleAnalysis - Invoke cloneBasicBlockAnalysis hook for
/// all loop passes.
-void LPPassManager::cloneBasicBlockSimpleAnalysis(BasicBlock *From,
+void LPPassManager::cloneBasicBlockSimpleAnalysis(BasicBlock *From,
BasicBlock *To, Loop *L) {
- for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
- LoopPass *LP = (LoopPass *)getContainedPass(Index);
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ LoopPass *LP = getContainedPass(Index);
LP->cloneBasicBlockAnalysis(From, To, L);
}
}
@@ -191,14 +162,14 @@
/// deleteSimpleAnalysisValue - Invoke deleteAnalysisValue hook for all passes.
void LPPassManager::deleteSimpleAnalysisValue(Value *V, Loop *L) {
if (BasicBlock *BB = dyn_cast<BasicBlock>(V)) {
- for (BasicBlock::iterator BI = BB->begin(), BE = BB->end(); BI != BE;
+ for (BasicBlock::iterator BI = BB->begin(), BE = BB->end(); BI != BE;
++BI) {
Instruction &I = *BI;
deleteSimpleAnalysisValue(&I, L);
}
}
- for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
- LoopPass *LP = (LoopPass *)getContainedPass(Index);
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ LoopPass *LP = getContainedPass(Index);
LP->deleteAnalysisValue(V, L);
}
}
@@ -213,7 +184,7 @@
/// Pass Manager itself does not invalidate any analysis info.
void LPPassManager::getAnalysisUsage(AnalysisUsage &Info) const {
- // LPPassManager needs LoopInfo. In the long term LoopInfo class will
+ // LPPassManager needs LoopInfo. In the long term LoopInfo class will
// become part of LPPassManager.
Info.addRequired<LoopInfo>();
Info.setPreservesAll();
@@ -224,6 +195,7 @@
bool LPPassManager::runOnFunction(Function &F) {
LI = &getAnalysis<LoopInfo>();
bool Changed = false;
+ createDebugInfoProbe();
// Collect inherited analysis from Module level pass manager.
populateInheritedAnalysis(TPM->activeStack);
@@ -239,35 +211,37 @@
for (std::deque<Loop *>::const_iterator I = LQ.begin(), E = LQ.end();
I != E; ++I) {
Loop *L = *I;
- for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
- LoopPass *P = (LoopPass*)getContainedPass(Index);
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ LoopPass *P = getContainedPass(Index);
Changed |= P->doInitialization(L, *this);
}
}
// Walk Loops
while (!LQ.empty()) {
-
+
CurrentLoop = LQ.back();
skipThisLoop = false;
redoThisLoop = false;
// Run all passes on the current Loop.
- for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
- LoopPass *P = (LoopPass*)getContainedPass(Index);
-
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ LoopPass *P = getContainedPass(Index);
dumpPassInfo(P, EXECUTION_MSG, ON_LOOP_MSG,
CurrentLoop->getHeader()->getName());
dumpRequiredSet(P);
initializeAnalysisImpl(P);
-
+ if (TheDebugProbe)
+ TheDebugProbe->initialize(P, F);
{
PassManagerPrettyStackEntry X(P, *CurrentLoop->getHeader());
TimeRegion PassTimer(getPassTimer(P));
Changed |= P->runOnLoop(CurrentLoop, *this);
}
+ if (TheDebugProbe)
+ TheDebugProbe->finalize(P, F);
if (Changed)
dumpPassInfo(P, MODIFICATION_MSG, ON_LOOP_MSG,
@@ -301,26 +275,26 @@
// Do not run other passes on this loop.
break;
}
-
+
// If the loop was deleted, release all the loop passes. This frees up
// some memory, and avoids trouble with the pass manager trying to call
// verifyAnalysis on them.
if (skipThisLoop)
- for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
Pass *P = getContainedPass(Index);
freePass(P, "<deleted>", ON_LOOP_MSG);
}
// Pop the loop from queue after running all passes.
LQ.pop_back();
-
+
if (redoThisLoop)
LQ.push_back(CurrentLoop);
}
-
+
// Finalization
for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
- LoopPass *P = (LoopPass *)getContainedPass(Index);
+ LoopPass *P = getContainedPass(Index);
Changed |= P->doFinalization();
}
@@ -354,7 +328,7 @@
// LPPassManger as expected.
void LoopPass::preparePassManager(PMStack &PMS) {
- // Find LPPassManager
+ // Find LPPassManager
while (!PMS.empty() &&
PMS.top()->getPassManagerType() > PMT_LoopPassManager)
PMS.pop();
@@ -363,14 +337,14 @@
// by other passes that are managed by LPM then do not insert
// this pass in current LPM. Use new LPPassManager.
if (PMS.top()->getPassManagerType() == PMT_LoopPassManager &&
- !PMS.top()->preserveHigherLevelAnalysis(this))
+ !PMS.top()->preserveHigherLevelAnalysis(this))
PMS.pop();
}
/// Assign pass manager to manage this pass.
void LoopPass::assignPassManager(PMStack &PMS,
PassManagerType PreferredType) {
- // Find LPPassManager
+ // Find LPPassManager
while (!PMS.empty() &&
PMS.top()->getPassManagerType() > PMT_LoopPassManager)
PMS.pop();
@@ -379,12 +353,12 @@
if (PMS.top()->getPassManagerType() == PMT_LoopPassManager)
LPPM = (LPPassManager*)PMS.top();
else {
- // Create new Loop Pass Manager if it does not exist.
+ // Create new Loop Pass Manager if it does not exist.
assert (!PMS.empty() && "Unable to create Loop Pass Manager");
PMDataManager *PMD = PMS.top();
- // [1] Create new Call Graph Pass Manager
- LPPM = new LPPassManager(PMD->getDepth() + 1);
+ // [1] Create new Loop Pass Manager
+ LPPM = new LPPassManager();
LPPM->populateInheritedAnalysis(PMS);
// [2] Set up new manager's top level manager
diff --git a/src/LLVM/lib/Analysis/MemDepPrinter.cpp b/src/LLVM/lib/Analysis/MemDepPrinter.cpp
new file mode 100644
index 0000000..fde07ea
--- /dev/null
+++ b/src/LLVM/lib/Analysis/MemDepPrinter.cpp
@@ -0,0 +1,192 @@
+//===- MemDepPrinter.cpp - Printer for MemoryDependenceAnalysis -----------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/MemoryDependenceAnalysis.h"
+#include "llvm/LLVMContext.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Support/CallSite.h"
+#include "llvm/Support/InstIterator.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/SetVector.h"
+using namespace llvm;
+
+namespace {
+ struct MemDepPrinter : public FunctionPass {
+ const Function *F;
+
+ enum DepType {
+ Clobber = 0,
+ Def,
+ NonFuncLocal,
+ Unknown
+ };
+
+ static const char* DepTypeStr[];
+
+ typedef PointerIntPair<const Instruction *, 2, DepType> InstTypePair;
+ typedef std::pair<InstTypePair, const BasicBlock *> Dep;
+ typedef SmallSetVector<Dep, 4> DepSet;
+ typedef DenseMap<const Instruction *, DepSet> DepSetMap;
+ DepSetMap Deps;
+
+ static char ID; // Pass identifcation, replacement for typeid
+ MemDepPrinter() : FunctionPass(ID) {
+ initializeMemDepPrinterPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnFunction(Function &F);
+
+ void print(raw_ostream &OS, const Module * = 0) const;
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.addRequiredTransitive<AliasAnalysis>();
+ AU.addRequiredTransitive<MemoryDependenceAnalysis>();
+ AU.setPreservesAll();
+ }
+
+ virtual void releaseMemory() {
+ Deps.clear();
+ F = 0;
+ }
+
+ private:
+ static InstTypePair getInstTypePair(MemDepResult dep) {
+ if (dep.isClobber())
+ return InstTypePair(dep.getInst(), Clobber);
+ if (dep.isDef())
+ return InstTypePair(dep.getInst(), Def);
+ if (dep.isNonFuncLocal())
+ return InstTypePair(dep.getInst(), NonFuncLocal);
+ assert(dep.isUnknown() && "unexptected dependence type");
+ return InstTypePair(dep.getInst(), Unknown);
+ }
+ static InstTypePair getInstTypePair(const Instruction* inst, DepType type) {
+ return InstTypePair(inst, type);
+ }
+ };
+}
+
+char MemDepPrinter::ID = 0;
+INITIALIZE_PASS_BEGIN(MemDepPrinter, "print-memdeps",
+ "Print MemDeps of function", false, true)
+INITIALIZE_PASS_DEPENDENCY(MemoryDependenceAnalysis)
+INITIALIZE_PASS_END(MemDepPrinter, "print-memdeps",
+ "Print MemDeps of function", false, true)
+
+FunctionPass *llvm::createMemDepPrinter() {
+ return new MemDepPrinter();
+}
+
+const char* MemDepPrinter::DepTypeStr[]
+ = {"Clobber", "Def", "NonFuncLocal", "Unknown"};
+
+bool MemDepPrinter::runOnFunction(Function &F) {
+ this->F = &F;
+ AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
+ MemoryDependenceAnalysis &MDA = getAnalysis<MemoryDependenceAnalysis>();
+
+ // All this code uses non-const interfaces because MemDep is not
+ // const-friendly, though nothing is actually modified.
+ for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) {
+ Instruction *Inst = &*I;
+
+ if (!Inst->mayReadFromMemory() && !Inst->mayWriteToMemory())
+ continue;
+
+ MemDepResult Res = MDA.getDependency(Inst);
+ if (!Res.isNonLocal()) {
+ Deps[Inst].insert(std::make_pair(getInstTypePair(Res),
+ static_cast<BasicBlock *>(0)));
+ } else if (CallSite CS = cast<Value>(Inst)) {
+ const MemoryDependenceAnalysis::NonLocalDepInfo &NLDI =
+ MDA.getNonLocalCallDependency(CS);
+
+ DepSet &InstDeps = Deps[Inst];
+ for (MemoryDependenceAnalysis::NonLocalDepInfo::const_iterator
+ I = NLDI.begin(), E = NLDI.end(); I != E; ++I) {
+ const MemDepResult &Res = I->getResult();
+ InstDeps.insert(std::make_pair(getInstTypePair(Res), I->getBB()));
+ }
+ } else {
+ SmallVector<NonLocalDepResult, 4> NLDI;
+ if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
+ if (!LI->isUnordered()) {
+ // FIXME: Handle atomic/volatile loads.
+ Deps[Inst].insert(std::make_pair(getInstTypePair(0, Unknown),
+ static_cast<BasicBlock *>(0)));
+ continue;
+ }
+ AliasAnalysis::Location Loc = AA.getLocation(LI);
+ MDA.getNonLocalPointerDependency(Loc, true, LI->getParent(), NLDI);
+ } else if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
+ if (!LI->isUnordered()) {
+ // FIXME: Handle atomic/volatile stores.
+ Deps[Inst].insert(std::make_pair(getInstTypePair(0, Unknown),
+ static_cast<BasicBlock *>(0)));
+ continue;
+ }
+ AliasAnalysis::Location Loc = AA.getLocation(SI);
+ MDA.getNonLocalPointerDependency(Loc, false, SI->getParent(), NLDI);
+ } else if (VAArgInst *VI = dyn_cast<VAArgInst>(Inst)) {
+ AliasAnalysis::Location Loc = AA.getLocation(VI);
+ MDA.getNonLocalPointerDependency(Loc, false, VI->getParent(), NLDI);
+ } else {
+ llvm_unreachable("Unknown memory instruction!");
+ }
+
+ DepSet &InstDeps = Deps[Inst];
+ for (SmallVectorImpl<NonLocalDepResult>::const_iterator
+ I = NLDI.begin(), E = NLDI.end(); I != E; ++I) {
+ const MemDepResult &Res = I->getResult();
+ InstDeps.insert(std::make_pair(getInstTypePair(Res), I->getBB()));
+ }
+ }
+ }
+
+ return false;
+}
+
+void MemDepPrinter::print(raw_ostream &OS, const Module *M) const {
+ for (const_inst_iterator I = inst_begin(*F), E = inst_end(*F); I != E; ++I) {
+ const Instruction *Inst = &*I;
+
+ DepSetMap::const_iterator DI = Deps.find(Inst);
+ if (DI == Deps.end())
+ continue;
+
+ const DepSet &InstDeps = DI->second;
+
+ for (DepSet::const_iterator I = InstDeps.begin(), E = InstDeps.end();
+ I != E; ++I) {
+ const Instruction *DepInst = I->first.getPointer();
+ DepType type = I->first.getInt();
+ const BasicBlock *DepBB = I->second;
+
+ OS << " ";
+ OS << DepTypeStr[type];
+ if (DepBB) {
+ OS << " in block ";
+ WriteAsOperand(OS, DepBB, /*PrintType=*/false, M);
+ }
+ if (DepInst) {
+ OS << " from: ";
+ DepInst->print(OS);
+ }
+ OS << "\n";
+ }
+
+ Inst->print(OS);
+ OS << "\n\n";
+ }
+}
diff --git a/src/LLVM/lib/Analysis/MemoryBuiltins.cpp b/src/LLVM/lib/Analysis/MemoryBuiltins.cpp
new file mode 100644
index 0000000..8d451c4
--- /dev/null
+++ b/src/LLVM/lib/Analysis/MemoryBuiltins.cpp
@@ -0,0 +1,213 @@
+//===------ MemoryBuiltins.cpp - Identify calls to memory builtins --------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This family of functions identifies calls to builtin functions that allocate
+// or free memory.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/MemoryBuiltins.h"
+#include "llvm/Constants.h"
+#include "llvm/Instructions.h"
+#include "llvm/Module.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Target/TargetData.h"
+using namespace llvm;
+
+//===----------------------------------------------------------------------===//
+// malloc Call Utility Functions.
+//
+
+/// isMalloc - Returns true if the value is either a malloc call or a
+/// bitcast of the result of a malloc call.
+bool llvm::isMalloc(const Value *I) {
+ return extractMallocCall(I) || extractMallocCallFromBitCast(I);
+}
+
+static bool isMallocCall(const CallInst *CI) {
+ if (!CI)
+ return false;
+
+ Function *Callee = CI->getCalledFunction();
+ if (Callee == 0 || !Callee->isDeclaration())
+ return false;
+ if (Callee->getName() != "malloc" &&
+ Callee->getName() != "_Znwj" && // operator new(unsigned int)
+ Callee->getName() != "_Znwm" && // operator new(unsigned long)
+ Callee->getName() != "_Znaj" && // operator new[](unsigned int)
+ Callee->getName() != "_Znam") // operator new[](unsigned long)
+ return false;
+
+ // Check malloc prototype.
+ // FIXME: workaround for PR5130, this will be obsolete when a nobuiltin
+ // attribute will exist.
+ FunctionType *FTy = Callee->getFunctionType();
+ if (FTy->getNumParams() != 1)
+ return false;
+ return FTy->getParamType(0)->isIntegerTy(32) ||
+ FTy->getParamType(0)->isIntegerTy(64);
+}
+
+/// extractMallocCall - Returns the corresponding CallInst if the instruction
+/// is a malloc call. Since CallInst::CreateMalloc() only creates calls, we
+/// ignore InvokeInst here.
+const CallInst *llvm::extractMallocCall(const Value *I) {
+ const CallInst *CI = dyn_cast<CallInst>(I);
+ return (isMallocCall(CI)) ? CI : NULL;
+}
+
+CallInst *llvm::extractMallocCall(Value *I) {
+ CallInst *CI = dyn_cast<CallInst>(I);
+ return (isMallocCall(CI)) ? CI : NULL;
+}
+
+static bool isBitCastOfMallocCall(const BitCastInst *BCI) {
+ if (!BCI)
+ return false;
+
+ return isMallocCall(dyn_cast<CallInst>(BCI->getOperand(0)));
+}
+
+/// extractMallocCallFromBitCast - Returns the corresponding CallInst if the
+/// instruction is a bitcast of the result of a malloc call.
+CallInst *llvm::extractMallocCallFromBitCast(Value *I) {
+ BitCastInst *BCI = dyn_cast<BitCastInst>(I);
+ return (isBitCastOfMallocCall(BCI)) ? cast<CallInst>(BCI->getOperand(0))
+ : NULL;
+}
+
+const CallInst *llvm::extractMallocCallFromBitCast(const Value *I) {
+ const BitCastInst *BCI = dyn_cast<BitCastInst>(I);
+ return (isBitCastOfMallocCall(BCI)) ? cast<CallInst>(BCI->getOperand(0))
+ : NULL;
+}
+
+static Value *computeArraySize(const CallInst *CI, const TargetData *TD,
+ bool LookThroughSExt = false) {
+ if (!CI)
+ return NULL;
+
+ // The size of the malloc's result type must be known to determine array size.
+ Type *T = getMallocAllocatedType(CI);
+ if (!T || !T->isSized() || !TD)
+ return NULL;
+
+ unsigned ElementSize = TD->getTypeAllocSize(T);
+ if (StructType *ST = dyn_cast<StructType>(T))
+ ElementSize = TD->getStructLayout(ST)->getSizeInBytes();
+
+ // If malloc call's arg can be determined to be a multiple of ElementSize,
+ // return the multiple. Otherwise, return NULL.
+ Value *MallocArg = CI->getArgOperand(0);
+ Value *Multiple = NULL;
+ if (ComputeMultiple(MallocArg, ElementSize, Multiple,
+ LookThroughSExt))
+ return Multiple;
+
+ return NULL;
+}
+
+/// isArrayMalloc - Returns the corresponding CallInst if the instruction
+/// is a call to malloc whose array size can be determined and the array size
+/// is not constant 1. Otherwise, return NULL.
+const CallInst *llvm::isArrayMalloc(const Value *I, const TargetData *TD) {
+ const CallInst *CI = extractMallocCall(I);
+ Value *ArraySize = computeArraySize(CI, TD);
+
+ if (ArraySize &&
+ ArraySize != ConstantInt::get(CI->getArgOperand(0)->getType(), 1))
+ return CI;
+
+ // CI is a non-array malloc or we can't figure out that it is an array malloc.
+ return NULL;
+}
+
+/// getMallocType - Returns the PointerType resulting from the malloc call.
+/// The PointerType depends on the number of bitcast uses of the malloc call:
+/// 0: PointerType is the calls' return type.
+/// 1: PointerType is the bitcast's result type.
+/// >1: Unique PointerType cannot be determined, return NULL.
+PointerType *llvm::getMallocType(const CallInst *CI) {
+ assert(isMalloc(CI) && "getMallocType and not malloc call");
+
+ PointerType *MallocType = NULL;
+ unsigned NumOfBitCastUses = 0;
+
+ // Determine if CallInst has a bitcast use.
+ for (Value::const_use_iterator UI = CI->use_begin(), E = CI->use_end();
+ UI != E; )
+ if (const BitCastInst *BCI = dyn_cast<BitCastInst>(*UI++)) {
+ MallocType = cast<PointerType>(BCI->getDestTy());
+ NumOfBitCastUses++;
+ }
+
+ // Malloc call has 1 bitcast use, so type is the bitcast's destination type.
+ if (NumOfBitCastUses == 1)
+ return MallocType;
+
+ // Malloc call was not bitcast, so type is the malloc function's return type.
+ if (NumOfBitCastUses == 0)
+ return cast<PointerType>(CI->getType());
+
+ // Type could not be determined.
+ return NULL;
+}
+
+/// getMallocAllocatedType - Returns the Type allocated by malloc call.
+/// The Type depends on the number of bitcast uses of the malloc call:
+/// 0: PointerType is the malloc calls' return type.
+/// 1: PointerType is the bitcast's result type.
+/// >1: Unique PointerType cannot be determined, return NULL.
+Type *llvm::getMallocAllocatedType(const CallInst *CI) {
+ PointerType *PT = getMallocType(CI);
+ return PT ? PT->getElementType() : NULL;
+}
+
+/// getMallocArraySize - Returns the array size of a malloc call. If the
+/// argument passed to malloc is a multiple of the size of the malloced type,
+/// then return that multiple. For non-array mallocs, the multiple is
+/// constant 1. Otherwise, return NULL for mallocs whose array size cannot be
+/// determined.
+Value *llvm::getMallocArraySize(CallInst *CI, const TargetData *TD,
+ bool LookThroughSExt) {
+ assert(isMalloc(CI) && "getMallocArraySize and not malloc call");
+ return computeArraySize(CI, TD, LookThroughSExt);
+}
+
+//===----------------------------------------------------------------------===//
+// free Call Utility Functions.
+//
+
+/// isFreeCall - Returns non-null if the value is a call to the builtin free()
+const CallInst *llvm::isFreeCall(const Value *I) {
+ const CallInst *CI = dyn_cast<CallInst>(I);
+ if (!CI)
+ return 0;
+ Function *Callee = CI->getCalledFunction();
+ if (Callee == 0 || !Callee->isDeclaration())
+ return 0;
+
+ if (Callee->getName() != "free" &&
+ Callee->getName() != "_ZdlPv" && // operator delete(void*)
+ Callee->getName() != "_ZdaPv") // operator delete[](void*)
+ return 0;
+
+ // Check free prototype.
+ // FIXME: workaround for PR5130, this will be obsolete when a nobuiltin
+ // attribute will exist.
+ FunctionType *FTy = Callee->getFunctionType();
+ if (!FTy->getReturnType()->isVoidTy())
+ return 0;
+ if (FTy->getNumParams() != 1)
+ return 0;
+ if (FTy->getParamType(0) != Type::getInt8PtrTy(Callee->getContext()))
+ return 0;
+
+ return CI;
+}
diff --git a/src/LLVM/lib/Analysis/MemoryDependenceAnalysis.cpp b/src/LLVM/lib/Analysis/MemoryDependenceAnalysis.cpp
index 25d6f86..92967c0 100644
--- a/src/LLVM/lib/Analysis/MemoryDependenceAnalysis.cpp
+++ b/src/LLVM/lib/Analysis/MemoryDependenceAnalysis.cpp
@@ -16,17 +16,22 @@
#define DEBUG_TYPE "memdep"
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
+#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/Function.h"
+#include "llvm/LLVMContext.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/InstructionSimplify.h"
+#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Analysis/PHITransAddr.h"
+#include "llvm/Analysis/ValueTracking.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/PredIteratorCache.h"
#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetData.h"
using namespace llvm;
STATISTIC(NumCacheNonLocal, "Number of fully cached non-local responses");
@@ -42,14 +47,23 @@
STATISTIC(NumCacheCompleteNonLocalPtr,
"Number of block queries that were completely cached");
+// Limit for the number of instructions to scan in a block.
+// FIXME: Figure out what a sane value is for this.
+// (500 is relatively insane.)
+static const int BlockScanLimit = 500;
+
char MemoryDependenceAnalysis::ID = 0;
// Register this pass...
-INITIALIZE_PASS(MemoryDependenceAnalysis, "memdep",
- "Memory Dependence Analysis", false, true);
+INITIALIZE_PASS_BEGIN(MemoryDependenceAnalysis, "memdep",
+ "Memory Dependence Analysis", false, true)
+INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_END(MemoryDependenceAnalysis, "memdep",
+ "Memory Dependence Analysis", false, true)
MemoryDependenceAnalysis::MemoryDependenceAnalysis()
: FunctionPass(ID), PredCache(0) {
+ initializeMemoryDependenceAnalysisPass(*PassRegistry::getPassRegistry());
}
MemoryDependenceAnalysis::~MemoryDependenceAnalysis() {
}
@@ -76,6 +90,7 @@
bool MemoryDependenceAnalysis::runOnFunction(Function &) {
AA = &getAnalysis<AliasAnalysis>();
+ TD = getAnalysisIfAvailable<TargetData>();
if (PredCache == 0)
PredCache.reset(new PredIteratorCache());
return false;
@@ -91,158 +106,358 @@
InstIt = ReverseMap.find(Inst);
assert(InstIt != ReverseMap.end() && "Reverse map out of sync?");
bool Found = InstIt->second.erase(Val);
- assert(Found && "Invalid reverse map!"); Found=Found;
+ assert(Found && "Invalid reverse map!"); (void)Found;
if (InstIt->second.empty())
ReverseMap.erase(InstIt);
}
+/// GetLocation - If the given instruction references a specific memory
+/// location, fill in Loc with the details, otherwise set Loc.Ptr to null.
+/// Return a ModRefInfo value describing the general behavior of the
+/// instruction.
+static
+AliasAnalysis::ModRefResult GetLocation(const Instruction *Inst,
+ AliasAnalysis::Location &Loc,
+ AliasAnalysis *AA) {
+ if (const LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
+ if (LI->isUnordered()) {
+ Loc = AA->getLocation(LI);
+ return AliasAnalysis::Ref;
+ } else if (LI->getOrdering() == Monotonic) {
+ Loc = AA->getLocation(LI);
+ return AliasAnalysis::ModRef;
+ }
+ Loc = AliasAnalysis::Location();
+ return AliasAnalysis::ModRef;
+ }
+
+ if (const StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
+ if (SI->isUnordered()) {
+ Loc = AA->getLocation(SI);
+ return AliasAnalysis::Mod;
+ } else if (SI->getOrdering() == Monotonic) {
+ Loc = AA->getLocation(SI);
+ return AliasAnalysis::ModRef;
+ }
+ Loc = AliasAnalysis::Location();
+ return AliasAnalysis::ModRef;
+ }
+
+ if (const VAArgInst *V = dyn_cast<VAArgInst>(Inst)) {
+ Loc = AA->getLocation(V);
+ return AliasAnalysis::ModRef;
+ }
+
+ if (const CallInst *CI = isFreeCall(Inst)) {
+ // calls to free() deallocate the entire structure
+ Loc = AliasAnalysis::Location(CI->getArgOperand(0));
+ return AliasAnalysis::Mod;
+ }
+
+ if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst))
+ switch (II->getIntrinsicID()) {
+ case Intrinsic::lifetime_start:
+ case Intrinsic::lifetime_end:
+ case Intrinsic::invariant_start:
+ Loc = AliasAnalysis::Location(II->getArgOperand(1),
+ cast<ConstantInt>(II->getArgOperand(0))
+ ->getZExtValue(),
+ II->getMetadata(LLVMContext::MD_tbaa));
+ // These intrinsics don't really modify the memory, but returning Mod
+ // will allow them to be handled conservatively.
+ return AliasAnalysis::Mod;
+ case Intrinsic::invariant_end:
+ Loc = AliasAnalysis::Location(II->getArgOperand(2),
+ cast<ConstantInt>(II->getArgOperand(1))
+ ->getZExtValue(),
+ II->getMetadata(LLVMContext::MD_tbaa));
+ // These intrinsics don't really modify the memory, but returning Mod
+ // will allow them to be handled conservatively.
+ return AliasAnalysis::Mod;
+ default:
+ break;
+ }
+
+ // Otherwise, just do the coarse-grained thing that always works.
+ if (Inst->mayWriteToMemory())
+ return AliasAnalysis::ModRef;
+ if (Inst->mayReadFromMemory())
+ return AliasAnalysis::Ref;
+ return AliasAnalysis::NoModRef;
+}
/// getCallSiteDependencyFrom - Private helper for finding the local
/// dependencies of a call site.
MemDepResult MemoryDependenceAnalysis::
getCallSiteDependencyFrom(CallSite CS, bool isReadOnlyCall,
BasicBlock::iterator ScanIt, BasicBlock *BB) {
+ unsigned Limit = BlockScanLimit;
+
// Walk backwards through the block, looking for dependencies
while (ScanIt != BB->begin()) {
+ // Limit the amount of scanning we do so we don't end up with quadratic
+ // running time on extreme testcases.
+ --Limit;
+ if (!Limit)
+ return MemDepResult::getUnknown();
+
Instruction *Inst = --ScanIt;
// If this inst is a memory op, get the pointer it accessed
- Value *Pointer = 0;
- uint64_t PointerSize = 0;
- if (StoreInst *S = dyn_cast<StoreInst>(Inst)) {
- Pointer = S->getPointerOperand();
- PointerSize = AA->getTypeStoreSize(S->getOperand(0)->getType());
- } else if (CallSite InstCS = cast<Value>(Inst)) {
+ AliasAnalysis::Location Loc;
+ AliasAnalysis::ModRefResult MR = GetLocation(Inst, Loc, AA);
+ if (Loc.Ptr) {
+ // A simple instruction.
+ if (AA->getModRefInfo(CS, Loc) != AliasAnalysis::NoModRef)
+ return MemDepResult::getClobber(Inst);
+ continue;
+ }
+
+ if (CallSite InstCS = cast<Value>(Inst)) {
// Debug intrinsics don't cause dependences.
- if (ISA_DEBUG_INFO_INTRINSIC(Inst)) continue;
+ if (isa<DbgInfoIntrinsic>(Inst)) continue;
// If these two calls do not interfere, look past it.
switch (AA->getModRefInfo(CS, InstCS)) {
case AliasAnalysis::NoModRef:
// If the two calls are the same, return InstCS as a Def, so that
// CS can be found redundant and eliminated.
- if (isReadOnlyCall && InstCS.onlyReadsMemory() &&
+ if (isReadOnlyCall && !(MR & AliasAnalysis::Mod) &&
CS.getInstruction()->isIdenticalToWhenDefined(Inst))
return MemDepResult::getDef(Inst);
// Otherwise if the two calls don't interact (e.g. InstCS is readnone)
// keep scanning.
- continue;
+ break;
default:
return MemDepResult::getClobber(Inst);
}
- } else {
- // Non-memory instruction.
- continue;
}
-
- if (AA->getModRefInfo(CS, Pointer, PointerSize) != AliasAnalysis::NoModRef)
- return MemDepResult::getClobber(Inst);
}
- // No dependence found. If this is the entry block of the function, it is a
- // clobber, otherwise it is non-local.
+ // No dependence found. If this is the entry block of the function, it is
+ // unknown, otherwise it is non-local.
if (BB != &BB->getParent()->getEntryBlock())
return MemDepResult::getNonLocal();
- return MemDepResult::getClobber(ScanIt);
+ return MemDepResult::getNonFuncLocal();
+}
+
+/// isLoadLoadClobberIfExtendedToFullWidth - Return true if LI is a load that
+/// would fully overlap MemLoc if done as a wider legal integer load.
+///
+/// MemLocBase, MemLocOffset are lazily computed here the first time the
+/// base/offs of memloc is needed.
+static bool
+isLoadLoadClobberIfExtendedToFullWidth(const AliasAnalysis::Location &MemLoc,
+ const Value *&MemLocBase,
+ int64_t &MemLocOffs,
+ const LoadInst *LI,
+ const TargetData *TD) {
+ // If we have no target data, we can't do this.
+ if (TD == 0) return false;
+
+ // If we haven't already computed the base/offset of MemLoc, do so now.
+ if (MemLocBase == 0)
+ MemLocBase = GetPointerBaseWithConstantOffset(MemLoc.Ptr, MemLocOffs, *TD);
+
+ unsigned Size = MemoryDependenceAnalysis::
+ getLoadLoadClobberFullWidthSize(MemLocBase, MemLocOffs, MemLoc.Size,
+ LI, *TD);
+ return Size != 0;
+}
+
+/// getLoadLoadClobberFullWidthSize - This is a little bit of analysis that
+/// looks at a memory location for a load (specified by MemLocBase, Offs,
+/// and Size) and compares it against a load. If the specified load could
+/// be safely widened to a larger integer load that is 1) still efficient,
+/// 2) safe for the target, and 3) would provide the specified memory
+/// location value, then this function returns the size in bytes of the
+/// load width to use. If not, this returns zero.
+unsigned MemoryDependenceAnalysis::
+getLoadLoadClobberFullWidthSize(const Value *MemLocBase, int64_t MemLocOffs,
+ unsigned MemLocSize, const LoadInst *LI,
+ const TargetData &TD) {
+ // We can only extend simple integer loads.
+ if (!isa<IntegerType>(LI->getType()) || !LI->isSimple()) return 0;
+
+ // Get the base of this load.
+ int64_t LIOffs = 0;
+ const Value *LIBase =
+ GetPointerBaseWithConstantOffset(LI->getPointerOperand(), LIOffs, TD);
+
+ // If the two pointers are not based on the same pointer, we can't tell that
+ // they are related.
+ if (LIBase != MemLocBase) return 0;
+
+ // Okay, the two values are based on the same pointer, but returned as
+ // no-alias. This happens when we have things like two byte loads at "P+1"
+ // and "P+3". Check to see if increasing the size of the "LI" load up to its
+ // alignment (or the largest native integer type) will allow us to load all
+ // the bits required by MemLoc.
+
+ // If MemLoc is before LI, then no widening of LI will help us out.
+ if (MemLocOffs < LIOffs) return 0;
+
+ // Get the alignment of the load in bytes. We assume that it is safe to load
+ // any legal integer up to this size without a problem. For example, if we're
+ // looking at an i8 load on x86-32 that is known 1024 byte aligned, we can
+ // widen it up to an i32 load. If it is known 2-byte aligned, we can widen it
+ // to i16.
+ unsigned LoadAlign = LI->getAlignment();
+
+ int64_t MemLocEnd = MemLocOffs+MemLocSize;
+
+ // If no amount of rounding up will let MemLoc fit into LI, then bail out.
+ if (LIOffs+LoadAlign < MemLocEnd) return 0;
+
+ // This is the size of the load to try. Start with the next larger power of
+ // two.
+ unsigned NewLoadByteSize = LI->getType()->getPrimitiveSizeInBits()/8U;
+ NewLoadByteSize = NextPowerOf2(NewLoadByteSize);
+
+ while (1) {
+ // If this load size is bigger than our known alignment or would not fit
+ // into a native integer register, then we fail.
+ if (NewLoadByteSize > LoadAlign ||
+ !TD.fitsInLegalInteger(NewLoadByteSize*8))
+ return 0;
+
+ // If a load of this width would include all of MemLoc, then we succeed.
+ if (LIOffs+NewLoadByteSize >= MemLocEnd)
+ return NewLoadByteSize;
+
+ NewLoadByteSize <<= 1;
+ }
+
+ return 0;
}
/// getPointerDependencyFrom - Return the instruction on which a memory
-/// location depends. If isLoad is true, this routine ignore may-aliases with
-/// read-only operations.
+/// location depends. If isLoad is true, this routine ignores may-aliases with
+/// read-only operations. If isLoad is false, this routine ignores may-aliases
+/// with reads from read-only locations.
MemDepResult MemoryDependenceAnalysis::
-getPointerDependencyFrom(Value *MemPtr, uint64_t MemSize, bool isLoad,
+getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad,
BasicBlock::iterator ScanIt, BasicBlock *BB) {
- Value *InvariantTag = 0;
+ const Value *MemLocBase = 0;
+ int64_t MemLocOffset = 0;
+
+ unsigned Limit = BlockScanLimit;
// Walk backwards through the basic block, looking for dependencies.
while (ScanIt != BB->begin()) {
+ // Limit the amount of scanning we do so we don't end up with quadratic
+ // running time on extreme testcases.
+ --Limit;
+ if (!Limit)
+ return MemDepResult::getUnknown();
+
Instruction *Inst = --ScanIt;
- // If we're in an invariant region, no dependencies can be found before
- // we pass an invariant-begin marker.
- if (InvariantTag == Inst) {
- InvariantTag = 0;
- continue;
- }
-
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst)) {
- // Debug intrinsics don't cause dependences.
- if (ISA_DEBUG_INFO_INTRINSIC(Inst)) continue;
-
- // If we pass an invariant-end marker, then we've just entered an
- // invariant region and can start ignoring dependencies.
- if (II->getIntrinsicID() == Intrinsic::invariant_end) {
- // FIXME: This only considers queries directly on the invariant-tagged
- // pointer, not on query pointers that are indexed off of them. It'd
- // be nice to handle that at some point.
- AliasAnalysis::AliasResult R = AA->alias(II->getArgOperand(2), MemPtr);
- if (R == AliasAnalysis::MustAlias) {
- InvariantTag = II->getArgOperand(0);
- continue;
- }
+ // Debug intrinsics don't (and can't) cause dependences.
+ if (isa<DbgInfoIntrinsic>(II)) continue;
// If we reach a lifetime begin or end marker, then the query ends here
// because the value is undefined.
- } else if (II->getIntrinsicID() == Intrinsic::lifetime_start) {
+ if (II->getIntrinsicID() == Intrinsic::lifetime_start) {
// FIXME: This only considers queries directly on the invariant-tagged
// pointer, not on query pointers that are indexed off of them. It'd
- // be nice to handle that at some point.
- AliasAnalysis::AliasResult R = AA->alias(II->getArgOperand(1), MemPtr);
- if (R == AliasAnalysis::MustAlias)
+ // be nice to handle that at some point (the right approach is to use
+ // GetPointerBaseWithConstantOffset).
+ if (AA->isMustAlias(AliasAnalysis::Location(II->getArgOperand(1)),
+ MemLoc))
return MemDepResult::getDef(II);
+ continue;
}
}
- // If we're querying on a load and we're in an invariant region, we're done
- // at this point. Nothing a load depends on can live in an invariant region.
- if (isLoad && InvariantTag) continue;
-
// Values depend on loads if the pointers are must aliased. This means that
// a load depends on another must aliased load from the same value.
if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
- Value *Pointer = LI->getPointerOperand();
- uint64_t PointerSize = AA->getTypeStoreSize(LI->getType());
+ // Atomic loads have complications involved.
+ // FIXME: This is overly conservative.
+ if (!LI->isUnordered())
+ return MemDepResult::getClobber(LI);
+
+ AliasAnalysis::Location LoadLoc = AA->getLocation(LI);
// If we found a pointer, check if it could be the same as our pointer.
- AliasAnalysis::AliasResult R =
- AA->alias(Pointer, PointerSize, MemPtr, MemSize);
+ AliasAnalysis::AliasResult R = AA->alias(LoadLoc, MemLoc);
+
+ if (isLoad) {
+ if (R == AliasAnalysis::NoAlias) {
+ // If this is an over-aligned integer load (for example,
+ // "load i8* %P, align 4") see if it would obviously overlap with the
+ // queried location if widened to a larger load (e.g. if the queried
+ // location is 1 byte at P+1). If so, return it as a load/load
+ // clobber result, allowing the client to decide to widen the load if
+ // it wants to.
+ if (IntegerType *ITy = dyn_cast<IntegerType>(LI->getType()))
+ if (LI->getAlignment()*8 > ITy->getPrimitiveSizeInBits() &&
+ isLoadLoadClobberIfExtendedToFullWidth(MemLoc, MemLocBase,
+ MemLocOffset, LI, TD))
+ return MemDepResult::getClobber(Inst);
+
+ continue;
+ }
+
+ // Must aliased loads are defs of each other.
+ if (R == AliasAnalysis::MustAlias)
+ return MemDepResult::getDef(Inst);
+
+#if 0 // FIXME: Temporarily disabled. GVN is cleverly rewriting loads
+ // in terms of clobbering loads, but since it does this by looking
+ // at the clobbering load directly, it doesn't know about any
+ // phi translation that may have happened along the way.
+
+ // If we have a partial alias, then return this as a clobber for the
+ // client to handle.
+ if (R == AliasAnalysis::PartialAlias)
+ return MemDepResult::getClobber(Inst);
+#endif
+
+ // Random may-alias loads don't depend on each other without a
+ // dependence.
+ continue;
+ }
+
+ // Stores don't depend on other no-aliased accesses.
if (R == AliasAnalysis::NoAlias)
continue;
-
- // May-alias loads don't depend on each other without a dependence.
- if (isLoad && R == AliasAnalysis::MayAlias)
+
+ // Stores don't alias loads from read-only memory.
+ if (AA->pointsToConstantMemory(LoadLoc))
continue;
- // Stores depend on may and must aliased loads, loads depend on must-alias
- // loads.
+
+ // Stores depend on may/must aliased loads.
return MemDepResult::getDef(Inst);
}
if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
- // There can't be stores to the value we care about inside an
- // invariant region.
- if (InvariantTag) continue;
-
+ // Atomic stores have complications involved.
+ // FIXME: This is overly conservative.
+ if (!SI->isUnordered())
+ return MemDepResult::getClobber(SI);
+
// If alias analysis can tell that this store is guaranteed to not modify
// the query pointer, ignore it. Use getModRefInfo to handle cases where
// the query pointer points to constant memory etc.
- if (AA->getModRefInfo(SI, MemPtr, MemSize) == AliasAnalysis::NoModRef)
+ if (AA->getModRefInfo(SI, MemLoc) == AliasAnalysis::NoModRef)
continue;
// Ok, this store might clobber the query pointer. Check to see if it is
// a must alias: in this case, we want to return this as a def.
- Value *Pointer = SI->getPointerOperand();
- uint64_t PointerSize = AA->getTypeStoreSize(SI->getOperand(0)->getType());
+ AliasAnalysis::Location StoreLoc = AA->getLocation(SI);
// If we found a pointer, check if it could be the same as our pointer.
- AliasAnalysis::AliasResult R =
- AA->alias(Pointer, PointerSize, MemPtr, MemSize);
+ AliasAnalysis::AliasResult R = AA->alias(StoreLoc, MemLoc);
if (R == AliasAnalysis::NoAlias)
continue;
- if (R == AliasAnalysis::MayAlias)
- return MemDepResult::getClobber(Inst);
- return MemDepResult::getDef(Inst);
+ if (R == AliasAnalysis::MustAlias)
+ return MemDepResult::getDef(Inst);
+ return MemDepResult::getClobber(Inst);
}
// If this is an allocation, and if we know that the accessed pointer is to
@@ -253,24 +468,21 @@
// a subsequent bitcast of the malloc call result. There can be stores to
// the malloced memory between the malloc call and its bitcast uses, and we
// need to continue scanning until the malloc call.
- if (isa<AllocaInst>(Inst)) {
- Value *AccessPtr = MemPtr->getUnderlyingObject();
+ if (isa<AllocaInst>(Inst) ||
+ (isa<CallInst>(Inst) && extractMallocCall(Inst))) {
+ const Value *AccessPtr = GetUnderlyingObject(MemLoc.Ptr, TD);
- if (AccessPtr == Inst ||
- AA->alias(Inst, 1, AccessPtr, 1) == AliasAnalysis::MustAlias)
+ if (AccessPtr == Inst || AA->isMustAlias(Inst, AccessPtr))
return MemDepResult::getDef(Inst);
continue;
}
// See if this instruction (e.g. a call or vaarg) mod/ref's the pointer.
- switch (AA->getModRefInfo(Inst, MemPtr, MemSize)) {
+ switch (AA->getModRefInfo(Inst, MemLoc)) {
case AliasAnalysis::NoModRef:
// If the call has no effect on the queried pointer, just ignore it.
continue;
case AliasAnalysis::Mod:
- // If we're in an invariant region, we can ignore calls that ONLY
- // modify the pointer.
- if (InvariantTag) continue;
return MemDepResult::getClobber(Inst);
case AliasAnalysis::Ref:
// If the call is known to never store to the pointer, and if this is a
@@ -283,11 +495,11 @@
}
}
- // No dependence found. If this is the entry block of the function, it is a
- // clobber, otherwise it is non-local.
+ // No dependence found. If this is the entry block of the function, it is
+ // unknown, otherwise it is non-local.
if (BB != &BB->getParent()->getEntryBlock())
return MemDepResult::getNonLocal();
- return MemDepResult::getClobber(ScanIt);
+ return MemDepResult::getNonFuncLocal();
}
/// getDependency - Return the instruction on which a memory operation
@@ -313,72 +525,33 @@
BasicBlock *QueryParent = QueryInst->getParent();
- Value *MemPtr = 0;
- uint64_t MemSize = 0;
-
// Do the scan.
if (BasicBlock::iterator(QueryInst) == QueryParent->begin()) {
- // No dependence found. If this is the entry block of the function, it is a
- // clobber, otherwise it is non-local.
+ // No dependence found. If this is the entry block of the function, it is
+ // unknown, otherwise it is non-local.
if (QueryParent != &QueryParent->getParent()->getEntryBlock())
LocalCache = MemDepResult::getNonLocal();
else
- LocalCache = MemDepResult::getClobber(QueryInst);
- } else if (StoreInst *SI = dyn_cast<StoreInst>(QueryInst)) {
- // If this is a volatile store, don't mess around with it. Just return the
- // previous instruction as a clobber.
- if (SI->isVolatile())
- LocalCache = MemDepResult::getClobber(--BasicBlock::iterator(ScanPos));
- else {
- MemPtr = SI->getPointerOperand();
- MemSize = AA->getTypeStoreSize(SI->getOperand(0)->getType());
- }
- } else if (LoadInst *LI = dyn_cast<LoadInst>(QueryInst)) {
- // If this is a volatile load, don't mess around with it. Just return the
- // previous instruction as a clobber.
- if (LI->isVolatile())
- LocalCache = MemDepResult::getClobber(--BasicBlock::iterator(ScanPos));
- else {
- MemPtr = LI->getPointerOperand();
- MemSize = AA->getTypeStoreSize(LI->getType());
- }
- } else if (isa<CallInst>(QueryInst) || ISA_INVOKE_INST(QueryInst)) {
- int IntrinsicID = 0; // Intrinsic IDs start at 1.
- IntrinsicInst *II = dyn_cast<IntrinsicInst>(QueryInst);
- if (II)
- IntrinsicID = II->getIntrinsicID();
+ LocalCache = MemDepResult::getNonFuncLocal();
+ } else {
+ AliasAnalysis::Location MemLoc;
+ AliasAnalysis::ModRefResult MR = GetLocation(QueryInst, MemLoc, AA);
+ if (MemLoc.Ptr) {
+ // If we can do a pointer scan, make it happen.
+ bool isLoad = !(MR & AliasAnalysis::Mod);
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(QueryInst))
+ isLoad |= II->getIntrinsicID() == Intrinsic::lifetime_start;
- switch (IntrinsicID) {
- case Intrinsic::lifetime_start:
- case Intrinsic::lifetime_end:
- case Intrinsic::invariant_start:
- MemPtr = II->getArgOperand(1);
- MemSize = cast<ConstantInt>(II->getArgOperand(0))->getZExtValue();
- break;
- case Intrinsic::invariant_end:
- MemPtr = II->getArgOperand(2);
- MemSize = cast<ConstantInt>(II->getArgOperand(1))->getZExtValue();
- break;
- default:
+ LocalCache = getPointerDependencyFrom(MemLoc, isLoad, ScanPos,
+ QueryParent);
+ } else if (isa<CallInst>(QueryInst) || isa<InvokeInst>(QueryInst)) {
CallSite QueryCS(QueryInst);
bool isReadOnly = AA->onlyReadsMemory(QueryCS);
LocalCache = getCallSiteDependencyFrom(QueryCS, isReadOnly, ScanPos,
QueryParent);
- break;
- }
- } else {
- // Non-memory instruction.
- LocalCache = MemDepResult::getClobber(--BasicBlock::iterator(ScanPos));
- }
-
- // If we need to do a pointer scan, make it happen.
- if (MemPtr) {
- bool isLoad = !QueryInst->mayWriteToMemory();
- if (IntrinsicInst *II = dyn_cast<MemoryUseIntrinsic>(QueryInst)) {
- isLoad |= II->getIntrinsicID() == Intrinsic::lifetime_end;
- }
- LocalCache = getPointerDependencyFrom(MemPtr, MemSize, isLoad, ScanPos,
- QueryParent);
+ } else
+ // Non-memory instruction.
+ LocalCache = MemDepResult::getUnknown();
}
// Remember the result!
@@ -512,10 +685,10 @@
Dep = getCallSiteDependencyFrom(QueryCS, isReadonlyCall,ScanPos, DirtyBB);
} else if (DirtyBB != &DirtyBB->getParent()->getEntryBlock()) {
// No dependence found. If this is the entry block of the function, it is
- // a clobber, otherwise it is non-local.
+ // a clobber, otherwise it is unknown.
Dep = MemDepResult::getNonLocal();
} else {
- Dep = MemDepResult::getClobber(ScanPos);
+ Dep = MemDepResult::getNonFuncLocal();
}
// If we had a dirty entry for the block, update it. Otherwise, just add
@@ -552,31 +725,27 @@
/// own block.
///
void MemoryDependenceAnalysis::
-getNonLocalPointerDependency(Value *Pointer, bool isLoad, BasicBlock *FromBB,
+getNonLocalPointerDependency(const AliasAnalysis::Location &Loc, bool isLoad,
+ BasicBlock *FromBB,
SmallVectorImpl<NonLocalDepResult> &Result) {
- assert(Pointer->getType()->isPointerTy() &&
+ assert(Loc.Ptr->getType()->isPointerTy() &&
"Can't get pointer deps of a non-pointer!");
Result.clear();
- // We know that the pointer value is live into FromBB find the def/clobbers
- // from presecessors.
- const Type *EltTy = cast<PointerType>(Pointer->getType())->getElementType();
- uint64_t PointeeSize = AA->getTypeStoreSize(EltTy);
-
- PHITransAddr Address(Pointer, TD);
+ PHITransAddr Address(const_cast<Value *>(Loc.Ptr), TD);
// This is the set of blocks we've inspected, and the pointer we consider in
// each block. Because of critical edges, we currently bail out if querying
// a block with multiple different pointers. This can happen during PHI
// translation.
DenseMap<BasicBlock*, Value*> Visited;
- if (!getNonLocalPointerDepFromBB(Address, PointeeSize, isLoad, FromBB,
+ if (!getNonLocalPointerDepFromBB(Address, Loc, isLoad, FromBB,
Result, Visited, true))
return;
Result.clear();
Result.push_back(NonLocalDepResult(FromBB,
- MemDepResult::getClobber(FromBB->begin()),
- Pointer));
+ MemDepResult::getUnknown(),
+ const_cast<Value *>(Loc.Ptr)));
}
/// GetNonLocalInfoForBlock - Compute the memdep value for BB with
@@ -584,7 +753,7 @@
/// lookup (which may use dirty cache info if available). If we do a lookup,
/// add the result to the cache.
MemDepResult MemoryDependenceAnalysis::
-GetNonLocalInfoForBlock(Value *Pointer, uint64_t PointeeSize,
+GetNonLocalInfoForBlock(const AliasAnalysis::Location &Loc,
bool isLoad, BasicBlock *BB,
NonLocalDepInfo *Cache, unsigned NumSortedEntries) {
@@ -618,15 +787,14 @@
ScanPos = ExistingResult->getResult().getInst();
// Eliminating the dirty entry from 'Cache', so update the reverse info.
- ValueIsLoadPair CacheKey(Pointer, isLoad);
+ ValueIsLoadPair CacheKey(Loc.Ptr, isLoad);
RemoveFromReverseMap(ReverseNonLocalPtrDeps, ScanPos, CacheKey);
} else {
++NumUncacheNonLocalPtr;
}
// Scan the block for the dependency.
- MemDepResult Dep = getPointerDependencyFrom(Pointer, PointeeSize, isLoad,
- ScanPos, BB);
+ MemDepResult Dep = getPointerDependencyFrom(Loc, isLoad, ScanPos, BB);
// If we had a dirty entry for the block, update it. Otherwise, just add
// a new entry.
@@ -638,14 +806,14 @@
// If the block has a dependency (i.e. it isn't completely transparent to
// the value), remember the reverse association because we just added it
// to Cache!
- if (Dep.isNonLocal())
+ if (!Dep.isDef() && !Dep.isClobber())
return Dep;
// Keep the ReverseNonLocalPtrDeps map up to date so we can efficiently
// update MemDep when we remove instructions.
Instruction *Inst = Dep.getInst();
assert(Inst && "Didn't depend on anything?");
- ValueIsLoadPair CacheKey(Pointer, isLoad);
+ ValueIsLoadPair CacheKey(Loc.Ptr, isLoad);
ReverseNonLocalPtrDeps[Inst].insert(CacheKey);
return Dep;
}
@@ -699,7 +867,8 @@
/// not compute dependence information for some reason. This should be treated
/// as a clobber dependence on the first instruction in the predecessor block.
bool MemoryDependenceAnalysis::
-getNonLocalPointerDepFromBB(const PHITransAddr &Pointer, uint64_t PointeeSize,
+getNonLocalPointerDepFromBB(const PHITransAddr &Pointer,
+ const AliasAnalysis::Location &Loc,
bool isLoad, BasicBlock *StartBB,
SmallVectorImpl<NonLocalDepResult> &Result,
DenseMap<BasicBlock*, Value*> &Visited,
@@ -707,14 +876,68 @@
// Look up the cached info for Pointer.
ValueIsLoadPair CacheKey(Pointer.getAddr(), isLoad);
-
- std::pair<BBSkipFirstBlockPair, NonLocalDepInfo> *CacheInfo =
- &NonLocalPointerDeps[CacheKey];
- NonLocalDepInfo *Cache = &CacheInfo->second;
+
+ // Set up a temporary NLPI value. If the map doesn't yet have an entry for
+ // CacheKey, this value will be inserted as the associated value. Otherwise,
+ // it'll be ignored, and we'll have to check to see if the cached size and
+ // tbaa tag are consistent with the current query.
+ NonLocalPointerInfo InitialNLPI;
+ InitialNLPI.Size = Loc.Size;
+ InitialNLPI.TBAATag = Loc.TBAATag;
+
+ // Get the NLPI for CacheKey, inserting one into the map if it doesn't
+ // already have one.
+ std::pair<CachedNonLocalPointerInfo::iterator, bool> Pair =
+ NonLocalPointerDeps.insert(std::make_pair(CacheKey, InitialNLPI));
+ NonLocalPointerInfo *CacheInfo = &Pair.first->second;
+
+ // If we already have a cache entry for this CacheKey, we may need to do some
+ // work to reconcile the cache entry and the current query.
+ if (!Pair.second) {
+ if (CacheInfo->Size < Loc.Size) {
+ // The query's Size is greater than the cached one. Throw out the
+ // cached data and procede with the query at the greater size.
+ CacheInfo->Pair = BBSkipFirstBlockPair();
+ CacheInfo->Size = Loc.Size;
+ for (NonLocalDepInfo::iterator DI = CacheInfo->NonLocalDeps.begin(),
+ DE = CacheInfo->NonLocalDeps.end(); DI != DE; ++DI)
+ if (Instruction *Inst = DI->getResult().getInst())
+ RemoveFromReverseMap(ReverseNonLocalPtrDeps, Inst, CacheKey);
+ CacheInfo->NonLocalDeps.clear();
+ } else if (CacheInfo->Size > Loc.Size) {
+ // This query's Size is less than the cached one. Conservatively restart
+ // the query using the greater size.
+ return getNonLocalPointerDepFromBB(Pointer,
+ Loc.getWithNewSize(CacheInfo->Size),
+ isLoad, StartBB, Result, Visited,
+ SkipFirstBlock);
+ }
+
+ // If the query's TBAATag is inconsistent with the cached one,
+ // conservatively throw out the cached data and restart the query with
+ // no tag if needed.
+ if (CacheInfo->TBAATag != Loc.TBAATag) {
+ if (CacheInfo->TBAATag) {
+ CacheInfo->Pair = BBSkipFirstBlockPair();
+ CacheInfo->TBAATag = 0;
+ for (NonLocalDepInfo::iterator DI = CacheInfo->NonLocalDeps.begin(),
+ DE = CacheInfo->NonLocalDeps.end(); DI != DE; ++DI)
+ if (Instruction *Inst = DI->getResult().getInst())
+ RemoveFromReverseMap(ReverseNonLocalPtrDeps, Inst, CacheKey);
+ CacheInfo->NonLocalDeps.clear();
+ }
+ if (Loc.TBAATag)
+ return getNonLocalPointerDepFromBB(Pointer, Loc.getWithoutTBAATag(),
+ isLoad, StartBB, Result, Visited,
+ SkipFirstBlock);
+ }
+ }
+
+ NonLocalDepInfo *Cache = &CacheInfo->NonLocalDeps;
// If we have valid cached information for exactly the block we are
// investigating, just return it with no recomputation.
- if (CacheInfo->first == BBSkipFirstBlockPair(StartBB, SkipFirstBlock)) {
+ if (CacheInfo->Pair == BBSkipFirstBlockPair(StartBB, SkipFirstBlock)) {
// We have a fully cached result for this query then we can just return the
// cached results and populate the visited set. However, we have to verify
// that we don't already have conflicting results for these blocks. Check
@@ -750,13 +973,16 @@
// than its valid cache info. If empty, the result will be valid cache info,
// otherwise it isn't.
if (Cache->empty())
- CacheInfo->first = BBSkipFirstBlockPair(StartBB, SkipFirstBlock);
+ CacheInfo->Pair = BBSkipFirstBlockPair(StartBB, SkipFirstBlock);
else
- CacheInfo->first = BBSkipFirstBlockPair();
+ CacheInfo->Pair = BBSkipFirstBlockPair();
SmallVector<BasicBlock*, 32> Worklist;
Worklist.push_back(StartBB);
+ // PredList used inside loop.
+ SmallVector<std::pair<BasicBlock*, PHITransAddr>, 16> PredList;
+
// Keep track of the entries that we know are sorted. Previously cached
// entries will all be sorted. The entries we add we only sort on demand (we
// don't insert every element into its sorted position). We know that we
@@ -777,8 +1003,7 @@
// Get the dependency info for Pointer in BB. If we have cached
// information, we will use it, otherwise we compute it.
DEBUG(AssertSorted(*Cache, NumSortedEntries));
- MemDepResult Dep = GetNonLocalInfoForBlock(Pointer.getAddr(), PointeeSize,
- isLoad, BB, Cache,
+ MemDepResult Dep = GetNonLocalInfoForBlock(Loc, isLoad, BB, Cache,
NumSortedEntries);
// If we got a Def or Clobber, add this to the list of results.
@@ -794,22 +1019,29 @@
// the same Pointer.
if (!Pointer.NeedsPHITranslationFromBlock(BB)) {
SkipFirstBlock = false;
+ SmallVector<BasicBlock*, 16> NewBlocks;
for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI) {
// Verify that we haven't looked at this block yet.
std::pair<DenseMap<BasicBlock*,Value*>::iterator, bool>
InsertRes = Visited.insert(std::make_pair(*PI, Pointer.getAddr()));
if (InsertRes.second) {
// First time we've looked at *PI.
- Worklist.push_back(*PI);
+ NewBlocks.push_back(*PI);
continue;
}
// If we have seen this block before, but it was with a different
// pointer then we have a phi translation failure and we have to treat
// this as a clobber.
- if (InsertRes.first->second != Pointer.getAddr())
+ if (InsertRes.first->second != Pointer.getAddr()) {
+ // Make sure to clean up the Visited map before continuing on to
+ // PredTranslationFailure.
+ for (unsigned i = 0; i < NewBlocks.size(); i++)
+ Visited.erase(NewBlocks[i]);
goto PredTranslationFailure;
+ }
}
+ Worklist.append(NewBlocks.begin(), NewBlocks.end());
continue;
}
@@ -828,13 +1060,15 @@
NumSortedEntries = Cache->size();
}
Cache = 0;
-
+
+ PredList.clear();
for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI) {
BasicBlock *Pred = *PI;
-
+ PredList.push_back(std::make_pair(Pred, Pointer));
+
// Get the PHI translated pointer in this predecessor. This can fail if
// not translatable, in which case the getAddr() returns null.
- PHITransAddr PredPointer(Pointer);
+ PHITransAddr &PredPointer = PredList.back().second;
PredPointer.PHITranslateValue(BB, Pred, 0);
Value *PredPtrVal = PredPointer.getAddr();
@@ -848,6 +1082,9 @@
InsertRes = Visited.insert(std::make_pair(Pred, PredPtrVal));
if (!InsertRes.second) {
+ // We found the pred; take it off the list of preds to visit.
+ PredList.pop_back();
+
// If the predecessor was visited with PredPtr, then we already did
// the analysis and can ignore it.
if (InsertRes.first->second == PredPtrVal)
@@ -856,18 +1093,49 @@
// Otherwise, the block was previously analyzed with a different
// pointer. We can't represent the result of this case, so we just
// treat this as a phi translation failure.
+
+ // Make sure to clean up the Visited map before continuing on to
+ // PredTranslationFailure.
+ for (unsigned i = 0; i < PredList.size(); i++)
+ Visited.erase(PredList[i].first);
+
goto PredTranslationFailure;
}
-
+ }
+
+ // Actually process results here; this need to be a separate loop to avoid
+ // calling getNonLocalPointerDepFromBB for blocks we don't want to return
+ // any results for. (getNonLocalPointerDepFromBB will modify our
+ // datastructures in ways the code after the PredTranslationFailure label
+ // doesn't expect.)
+ for (unsigned i = 0; i < PredList.size(); i++) {
+ BasicBlock *Pred = PredList[i].first;
+ PHITransAddr &PredPointer = PredList[i].second;
+ Value *PredPtrVal = PredPointer.getAddr();
+
+ bool CanTranslate = true;
// If PHI translation was unable to find an available pointer in this
// predecessor, then we have to assume that the pointer is clobbered in
// that predecessor. We can still do PRE of the load, which would insert
// a computation of the pointer in this predecessor.
- if (PredPtrVal == 0) {
+ if (PredPtrVal == 0)
+ CanTranslate = false;
+
+ // FIXME: it is entirely possible that PHI translating will end up with
+ // the same value. Consider PHI translating something like:
+ // X = phi [x, bb1], [y, bb2]. PHI translating for bb1 doesn't *need*
+ // to recurse here, pedantically speaking.
+
+ // If getNonLocalPointerDepFromBB fails here, that means the cached
+ // result conflicted with the Visited list; we have to conservatively
+ // assume it is unknown, but this also does not block PRE of the load.
+ if (!CanTranslate ||
+ getNonLocalPointerDepFromBB(PredPointer,
+ Loc.getWithNewPtr(PredPtrVal),
+ isLoad, Pred,
+ Result, Visited)) {
// Add the entry to the Result list.
- NonLocalDepResult Entry(Pred,
- MemDepResult::getClobber(Pred->getTerminator()),
- PredPtrVal);
+ NonLocalDepResult Entry(Pred, MemDepResult::getUnknown(), PredPtrVal);
Result.push_back(Entry);
// Since we had a phi translation failure, the cache for CacheKey won't
@@ -875,41 +1143,34 @@
// queries. Mark this in NonLocalPointerDeps by setting the
// BBSkipFirstBlockPair pointer to null. This requires reuse of the
// cached value to do more work but not miss the phi trans failure.
- NonLocalPointerDeps[CacheKey].first = BBSkipFirstBlockPair();
+ NonLocalPointerInfo &NLPI = NonLocalPointerDeps[CacheKey];
+ NLPI.Pair = BBSkipFirstBlockPair();
continue;
}
-
- // FIXME: it is entirely possible that PHI translating will end up with
- // the same value. Consider PHI translating something like:
- // X = phi [x, bb1], [y, bb2]. PHI translating for bb1 doesn't *need*
- // to recurse here, pedantically speaking.
-
- // If we have a problem phi translating, fall through to the code below
- // to handle the failure condition.
- if (getNonLocalPointerDepFromBB(PredPointer, PointeeSize, isLoad, Pred,
- Result, Visited))
- goto PredTranslationFailure;
}
// Refresh the CacheInfo/Cache pointer so that it isn't invalidated.
CacheInfo = &NonLocalPointerDeps[CacheKey];
- Cache = &CacheInfo->second;
+ Cache = &CacheInfo->NonLocalDeps;
NumSortedEntries = Cache->size();
// Since we did phi translation, the "Cache" set won't contain all of the
// results for the query. This is ok (we can still use it to accelerate
// specific block queries) but we can't do the fastpath "return all
// results from the set" Clear out the indicator for this.
- CacheInfo->first = BBSkipFirstBlockPair();
+ CacheInfo->Pair = BBSkipFirstBlockPair();
SkipFirstBlock = false;
continue;
PredTranslationFailure:
+ // The following code is "failure"; we can't produce a sane translation
+ // for the given block. It assumes that we haven't modified any of
+ // our datastructures while processing the current block.
if (Cache == 0) {
// Refresh the CacheInfo/Cache pointer if it got invalidated.
CacheInfo = &NonLocalPointerDeps[CacheKey];
- Cache = &CacheInfo->second;
+ Cache = &CacheInfo->NonLocalDeps;
NumSortedEntries = Cache->size();
}
@@ -917,10 +1178,9 @@
// results for the query. This is ok (we can still use it to accelerate
// specific block queries) but we can't do the fastpath "return all
// results from the set". Clear out the indicator for this.
- CacheInfo->first = BBSkipFirstBlockPair();
+ CacheInfo->Pair = BBSkipFirstBlockPair();
- // If *nothing* works, mark the pointer as being clobbered by the first
- // instruction in this block.
+ // If *nothing* works, mark the pointer as unknown.
//
// If this is the magic first block, return this as a clobber of the whole
// incoming value. Since we can't phi translate to one of the predecessors,
@@ -935,8 +1195,7 @@
assert(I->getResult().isNonLocal() &&
"Should only be here with transparent block");
- I->setResult(MemDepResult::getClobber(BB->begin()));
- ReverseNonLocalPtrDeps[BB->begin()].insert(CacheKey);
+ I->setResult(MemDepResult::getUnknown());
Result.push_back(NonLocalDepResult(I->getBB(), I->getResult(),
Pointer.getAddr()));
break;
@@ -959,7 +1218,7 @@
// Remove all of the entries in the BB->val map. This involves removing
// instructions from the reverse map.
- NonLocalDepInfo &PInfo = It->second.second;
+ NonLocalDepInfo &PInfo = It->second.NonLocalDeps;
for (unsigned i = 0, e = PInfo.size(); i != e; ++i) {
Instruction *Target = PInfo[i].getResult().getInst();
@@ -1130,10 +1389,10 @@
assert(P.getPointer() != RemInst &&
"Already removed NonLocalPointerDeps info for RemInst");
- NonLocalDepInfo &NLPDI = NonLocalPointerDeps[P].second;
+ NonLocalDepInfo &NLPDI = NonLocalPointerDeps[P].NonLocalDeps;
// The cache is not valid for any specific block anymore.
- NonLocalPointerDeps[P].first = BBSkipFirstBlockPair();
+ NonLocalPointerDeps[P].Pair = BBSkipFirstBlockPair();
// Update any entries for RemInst to use the instruction after it.
for (NonLocalDepInfo::iterator DI = NLPDI.begin(), DE = NLPDI.end();
@@ -1179,7 +1438,7 @@
for (CachedNonLocalPointerInfo::const_iterator I =NonLocalPointerDeps.begin(),
E = NonLocalPointerDeps.end(); I != E; ++I) {
assert(I->first.getPointer() != D && "Inst occurs in NLPD map key");
- const NonLocalDepInfo &Val = I->second.second;
+ const NonLocalDepInfo &Val = I->second.NonLocalDeps;
for (NonLocalDepInfo::const_iterator II = Val.begin(), E = Val.end();
II != E; ++II)
assert(II->getResult().getInst() != D && "Inst occurs as NLPD value");
diff --git a/src/LLVM/lib/Analysis/ModuleDebugInfoPrinter.cpp b/src/LLVM/lib/Analysis/ModuleDebugInfoPrinter.cpp
new file mode 100644
index 0000000..e7e999c
--- /dev/null
+++ b/src/LLVM/lib/Analysis/ModuleDebugInfoPrinter.cpp
@@ -0,0 +1,87 @@
+//===-- ModuleDebugInfoPrinter.cpp - Prints module debug info metadata ----===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass decodes the debug info metadata in a module and prints in a
+// (sufficiently-prepared-) human-readable form.
+//
+// For example, run this pass from opt along with the -analyze option, and
+// it'll print to standard output.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/DebugInfo.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Pass.h"
+#include "llvm/Function.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/Statistic.h"
+using namespace llvm;
+
+namespace {
+ class ModuleDebugInfoPrinter : public ModulePass {
+ DebugInfoFinder Finder;
+ public:
+ static char ID; // Pass identification, replacement for typeid
+ ModuleDebugInfoPrinter() : ModulePass(ID) {
+ initializeModuleDebugInfoPrinterPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual bool runOnModule(Module &M);
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+ virtual void print(raw_ostream &O, const Module *M) const;
+ };
+}
+
+char ModuleDebugInfoPrinter::ID = 0;
+INITIALIZE_PASS(ModuleDebugInfoPrinter, "module-debuginfo",
+ "Decodes module-level debug info", false, true)
+
+ModulePass *llvm::createModuleDebugInfoPrinterPass() {
+ return new ModuleDebugInfoPrinter();
+}
+
+bool ModuleDebugInfoPrinter::runOnModule(Module &M) {
+ Finder.processModule(M);
+ return false;
+}
+
+void ModuleDebugInfoPrinter::print(raw_ostream &O, const Module *M) const {
+ for (DebugInfoFinder::iterator I = Finder.compile_unit_begin(),
+ E = Finder.compile_unit_end(); I != E; ++I) {
+ O << "Compile Unit: ";
+ DICompileUnit(*I).print(O);
+ O << '\n';
+ }
+
+ for (DebugInfoFinder::iterator I = Finder.subprogram_begin(),
+ E = Finder.subprogram_end(); I != E; ++I) {
+ O << "Subprogram: ";
+ DISubprogram(*I).print(O);
+ O << '\n';
+ }
+
+ for (DebugInfoFinder::iterator I = Finder.global_variable_begin(),
+ E = Finder.global_variable_end(); I != E; ++I) {
+ O << "GlobalVariable: ";
+ DIGlobalVariable(*I).print(O);
+ O << '\n';
+ }
+
+ for (DebugInfoFinder::iterator I = Finder.type_begin(),
+ E = Finder.type_end(); I != E; ++I) {
+ O << "Type: ";
+ DIType(*I).print(O);
+ O << '\n';
+ }
+}
diff --git a/src/LLVM/lib/Analysis/NoAliasAnalysis.cpp b/src/LLVM/lib/Analysis/NoAliasAnalysis.cpp
new file mode 100644
index 0000000..101c2d5
--- /dev/null
+++ b/src/LLVM/lib/Analysis/NoAliasAnalysis.cpp
@@ -0,0 +1,88 @@
+//===- NoAliasAnalysis.cpp - Minimal Alias Analysis Impl ------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the default implementation of the Alias Analysis interface
+// that simply returns "I don't know" for all queries.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Pass.h"
+#include "llvm/Target/TargetData.h"
+using namespace llvm;
+
+namespace {
+ /// NoAA - This class implements the -no-aa pass, which always returns "I
+ /// don't know" for alias queries. NoAA is unlike other alias analysis
+ /// implementations, in that it does not chain to a previous analysis. As
+ /// such it doesn't follow many of the rules that other alias analyses must.
+ ///
+ struct NoAA : public ImmutablePass, public AliasAnalysis {
+ static char ID; // Class identification, replacement for typeinfo
+ NoAA() : ImmutablePass(ID) {
+ initializeNoAAPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ }
+
+ virtual void initializePass() {
+ // Note: NoAA does not call InitializeAliasAnalysis because it's
+ // special and does not support chaining.
+ TD = getAnalysisIfAvailable<TargetData>();
+ }
+
+ virtual AliasResult alias(const Location &LocA, const Location &LocB) {
+ return MayAlias;
+ }
+
+ virtual ModRefBehavior getModRefBehavior(ImmutableCallSite CS) {
+ return UnknownModRefBehavior;
+ }
+ virtual ModRefBehavior getModRefBehavior(const Function *F) {
+ return UnknownModRefBehavior;
+ }
+
+ virtual bool pointsToConstantMemory(const Location &Loc,
+ bool OrLocal) {
+ return false;
+ }
+ virtual ModRefResult getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc) {
+ return ModRef;
+ }
+ virtual ModRefResult getModRefInfo(ImmutableCallSite CS1,
+ ImmutableCallSite CS2) {
+ return ModRef;
+ }
+
+ virtual void deleteValue(Value *V) {}
+ virtual void copyValue(Value *From, Value *To) {}
+ virtual void addEscapingUse(Use &U) {}
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ virtual void *getAdjustedAnalysisPointer(const void *ID) {
+ if (ID == &AliasAnalysis::ID)
+ return (AliasAnalysis*)this;
+ return this;
+ }
+ };
+} // End of anonymous namespace
+
+// Register this pass...
+char NoAA::ID = 0;
+INITIALIZE_AG_PASS(NoAA, AliasAnalysis, "no-aa",
+ "No Alias Analysis (always returns 'may' alias)",
+ true, true, true)
+
+ImmutablePass *llvm::createNoAAPass() { return new NoAA(); }
diff --git a/src/LLVM/lib/Analysis/PACKAGE.vcxproj b/src/LLVM/lib/Analysis/PACKAGE.vcxproj
new file mode 100644
index 0000000..ec92cdc
--- /dev/null
+++ b/src/LLVM/lib/Analysis/PACKAGE.vcxproj
@@ -0,0 +1,277 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup Label="ProjectConfigurations">
+ <ProjectConfiguration Include="Debug|Win32">
+ <Configuration>Debug</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="Release|Win32">
+ <Configuration>Release</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="MinSizeRel|Win32">
+ <Configuration>MinSizeRel</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ <ProjectConfiguration Include="RelWithDebInfo|Win32">
+ <Configuration>RelWithDebInfo</Configuration>
+ <Platform>Win32</Platform>
+ </ProjectConfiguration>
+ </ItemGroup>
+ <PropertyGroup Label="Globals">
+ <ProjectGUID>{1B050569-3318-48D9-8BB0-4DE9EF58B202}</ProjectGUID>
+ <Keyword>Win32Proj</Keyword>
+ <Platform>Win32</Platform>
+ <ProjectName>PACKAGE</ProjectName>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'" Label="Configuration">
+ <ConfigurationType></ConfigurationType>
+ <UseOfMfc>false</UseOfMfc>
+ <CharacterSet>MultiByte</CharacterSet>
+ </PropertyGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+ <ImportGroup Label="ExtensionSettings">
+ </ImportGroup>
+ <ImportGroup Label="PropertySheets">
+ <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" /> </ImportGroup>
+ <PropertyGroup Label="UserMacros" />
+ <PropertyGroup>
+ <_ProjectFileVersion>10.0.20506.1</_ProjectFileVersion>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ <IntDir Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
+ </PropertyGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+cd ..\..\..\LLVM
+if %errorlevel% neq 0 goto :cmEnd
+D:
+if %errorlevel% neq 0 goto :cmEnd
+"C:\Program Files (x86)\CMake 2.8\bin\cpack.exe" -C $(Configuration) --config ./CPackConfig.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+cd ..\..\..\LLVM
+if %errorlevel% neq 0 goto :cmEnd
+D:
+if %errorlevel% neq 0 goto :cmEnd
+"C:\Program Files (x86)\CMake 2.8\bin\cpack.exe" -C $(Configuration) --config ./CPackConfig.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+cd ..\..\..\LLVM
+if %errorlevel% neq 0 goto :cmEnd
+D:
+if %errorlevel% neq 0 goto :cmEnd
+"C:\Program Files (x86)\CMake 2.8\bin\cpack.exe" -C $(Configuration) --config ./CPackConfig.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">
+ <Midl>
+ <AdditionalIncludeDirectories>..\Analysis;..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+ <OutputDirectory>$(IntDir)</OutputDirectory>
+ <HeaderFileName>%(Filename).h</HeaderFileName>
+ <TypeLibraryName>%(Filename).tlb</TypeLibraryName>
+ <InterfaceIdentifierFileName>%(Filename)_i.c</InterfaceIdentifierFileName>
+ <ProxyFileName>%(Filename)_p.c</ProxyFileName>
+ </Midl>
+ <PostBuildEvent>
+ <Message></Message>
+ <Command>setlocal
+cd ..\..\..\LLVM
+if %errorlevel% neq 0 goto :cmEnd
+D:
+if %errorlevel% neq 0 goto :cmEnd
+"C:\Program Files (x86)\CMake 2.8\bin\cpack.exe" -C $(Configuration) --config ./CPackConfig.cmake
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ </PostBuildEvent>
+ </ItemDefinitionGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeFiles\PACKAGE_force.rule">
+ <Message Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles/PACKAGE_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles\PACKAGE_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='Release|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles/PACKAGE_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles\PACKAGE_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles/PACKAGE_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles\PACKAGE_force</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'"> </Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">setlocal
+cd .
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles/PACKAGE_force.rule;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles\PACKAGE_force</Outputs>
+ </CustomBuild>
+ </ItemGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeLists.txt">
+ <Message Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../.. -B../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../.. -B../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../.. -B../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='MinSizeRel|Win32'">CMakeFiles\generate.stamp</Outputs>
+ <Message Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">Building Custom Rule CMakeLists.txt</Message>
+ <Command Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">setlocal
+"C:\Program Files (x86)\CMake 2.8\bin\cmake.exe" -H../.. -B../.. --check-stamp-file CMakeFiles\generate.stamp
+if %errorlevel% neq 0 goto :cmEnd
+:cmEnd
+endlocal & call :cmErrorLevel %errorlevel% & goto :cmDone
+:cmErrorLevel
+exit /b %1
+:cmDone
+if %errorlevel% neq 0 goto :VCEnd</Command>
+ <AdditionalInputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeLists.txt;CMakeLists.txt;CMakeLists.txt;%(AdditionalInputs)</AdditionalInputs>
+ <Outputs Condition="'$(Configuration)|$(Platform)'=='RelWithDebInfo|Win32'">CMakeFiles\generate.stamp</Outputs>
+ </CustomBuild>
+ </ItemGroup>
+ <ItemGroup>
+ <ProjectReference Include="..\..\ALL_BUILD.vcxproj">
+ <Project>17AECBCF-B2AE-4524-9010-9A175A8F6BFE</Project>
+ </ProjectReference>
+ </ItemGroup>
+ <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+ <ImportGroup Label="ExtensionTargets">
+ </ImportGroup>
+</Project>
\ No newline at end of file
diff --git a/src/LLVM/lib/Analysis/PACKAGE.vcxproj.filters b/src/LLVM/lib/Analysis/PACKAGE.vcxproj.filters
new file mode 100644
index 0000000..a570359
--- /dev/null
+++ b/src/LLVM/lib/Analysis/PACKAGE.vcxproj.filters
@@ -0,0 +1,24 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+ <ItemGroup>
+ <CustomBuild Include="CMakeFiles\PACKAGE_force.rule">
+ <Filter>CMake Rules</Filter>
+ </CustomBuild>
+ <CustomBuild Include="CMakeLists.txt" />
+ </ItemGroup>
+ <ItemGroup>
+ <Filter Include="CMake Rules">
+ <UniqueIdentifier>{71794486-B3CB-4A48-93CC-DE95557E96E1}</UniqueIdentifier>
+ </Filter>
+ </ItemGroup>
+ <ItemGroup>
+ </ItemGroup>
+</Project>
diff --git a/src/LLVM/lib/Analysis/PHITransAddr.cpp b/src/LLVM/lib/Analysis/PHITransAddr.cpp
index cca6d65..7e22ddc 100644
--- a/src/LLVM/lib/Analysis/PHITransAddr.cpp
+++ b/src/LLVM/lib/Analysis/PHITransAddr.cpp
@@ -12,22 +12,28 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/PHITransAddr.h"
+#include "llvm/Constants.h"
+#include "llvm/Instructions.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
static bool CanPHITrans(Instruction *Inst) {
if (isa<PHINode>(Inst) ||
- isa<BitCastInst>(Inst) ||
isa<GetElementPtrInst>(Inst))
return true;
-
+
+ if (isa<CastInst>(Inst) &&
+ Inst->isSafeToSpeculativelyExecute())
+ return true;
+
if (Inst->getOpcode() == Instruction::Add &&
isa<ConstantInt>(Inst->getOperand(1)))
return true;
-
+
// cerr << "MEMDEP: Could not PHI translate: " << *Pointer;
// if (isa<BitCastInst>(PtrInst) || isa<GetElementPtrInst>(PtrInst))
// cerr << "OP:\t\t\t\t" << *PtrInst->getOperand(0);
@@ -50,7 +56,7 @@
// If this is a non-instruction value, there is nothing to do.
Instruction *I = dyn_cast<Instruction>(Expr);
if (I == 0) return true;
-
+
// If it's an instruction, it is either in Tmp or its operands recursively
// are.
SmallVectorImpl<Instruction*>::iterator Entry =
@@ -59,16 +65,17 @@
InstInputs.erase(Entry);
return true;
}
-
+
// If it isn't in the InstInputs list it is a subexpr incorporated into the
// address. Sanity check that it is phi translatable.
if (!CanPHITrans(I)) {
- errs() << "Non phi translatable instruction found in PHITransAddr, either "
- "something is missing from InstInputs or CanPHITrans is wrong:\n";
+ errs() << "Non phi translatable instruction found in PHITransAddr:\n";
errs() << *I << '\n';
+ llvm_unreachable("Either something is missing from InstInputs or "
+ "CanPHITrans is wrong.");
return false;
}
-
+
// Validate the operands of the instruction.
for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
if (!VerifySubExpr(I->getOperand(i), InstInputs))
@@ -82,19 +89,20 @@
/// returns false.
bool PHITransAddr::Verify() const {
if (Addr == 0) return true;
-
- SmallVector<Instruction*, 8> Tmp(InstInputs.begin(), InstInputs.end());
-
+
+ SmallVector<Instruction*, 8> Tmp(InstInputs.begin(), InstInputs.end());
+
if (!VerifySubExpr(Addr, Tmp))
return false;
-
+
if (!Tmp.empty()) {
- errs() << "PHITransAddr inconsistent, contains extra instructions:\n";
+ errs() << "PHITransAddr contains extra instructions:\n";
for (unsigned i = 0, e = InstInputs.size(); i != e; ++i)
errs() << " InstInput #" << i << " is " << *InstInputs[i] << "\n";
+ llvm_unreachable("This is unexpected.");
return false;
}
-
+
// a-ok.
return true;
}
@@ -111,11 +119,11 @@
}
-static void RemoveInstInputs(Value *V,
+static void RemoveInstInputs(Value *V,
SmallVectorImpl<Instruction*> &InstInputs) {
Instruction *I = dyn_cast<Instruction>(V);
if (I == 0) return;
-
+
// If the instruction is in the InstInputs list, remove it.
SmallVectorImpl<Instruction*>::iterator Entry =
std::find(InstInputs.begin(), InstInputs.end(), I);
@@ -123,9 +131,9 @@
InstInputs.erase(Entry);
return;
}
-
+
assert(!isa<PHINode>(I) && "Error, removing something that isn't an input");
-
+
// Otherwise, it must have instruction inputs itself. Zap them recursively.
for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
if (Instruction *Op = dyn_cast<Instruction>(I->getOperand(i)))
@@ -139,7 +147,7 @@
// If this is a non-instruction value, it can't require PHI translation.
Instruction *Inst = dyn_cast<Instruction>(V);
if (Inst == 0) return V;
-
+
// Determine whether 'Inst' is an input to our PHI translatable expression.
bool isInput = std::count(InstInputs.begin(), InstInputs.end(), Inst);
@@ -156,16 +164,16 @@
// In either case, the instruction itself isn't an input any longer.
InstInputs.erase(std::find(InstInputs.begin(), InstInputs.end(), Inst));
-
+
// If this is a PHI, go ahead and translate it.
if (PHINode *PN = dyn_cast<PHINode>(Inst))
return AddAsInput(PN->getIncomingValueForBlock(PredBB));
-
+
// If this is a non-phi value, and it is analyzable, we can incorporate it
// into the expression by making all instruction operands be inputs.
if (!CanPHITrans(Inst))
return 0;
-
+
// All instruction operands are now inputs (and of course, they may also be
// defined in this block, so they may need to be phi translated themselves.
for (unsigned i = 0, e = Inst->getNumOperands(); i != e; ++i)
@@ -176,31 +184,34 @@
// Ok, it must be an intermediate result (either because it started that way
// or because we just incorporated it into the expression). See if its
// operands need to be phi translated, and if so, reconstruct it.
-
- if (BitCastInst *BC = dyn_cast<BitCastInst>(Inst)) {
- Value *PHIIn = PHITranslateSubExpr(BC->getOperand(0), CurBB, PredBB, DT);
+
+ if (CastInst *Cast = dyn_cast<CastInst>(Inst)) {
+ if (!Cast->isSafeToSpeculativelyExecute()) return 0;
+ Value *PHIIn = PHITranslateSubExpr(Cast->getOperand(0), CurBB, PredBB, DT);
if (PHIIn == 0) return 0;
- if (PHIIn == BC->getOperand(0))
- return BC;
-
+ if (PHIIn == Cast->getOperand(0))
+ return Cast;
+
// Find an available version of this cast.
-
+
// Constants are trivial to find.
if (Constant *C = dyn_cast<Constant>(PHIIn))
- return AddAsInput(ConstantExpr::getBitCast(C, BC->getType()));
-
- // Otherwise we have to see if a bitcasted version of the incoming pointer
+ return AddAsInput(ConstantExpr::getCast(Cast->getOpcode(),
+ C, Cast->getType()));
+
+ // Otherwise we have to see if a casted version of the incoming pointer
// is available. If so, we can use it, otherwise we have to fail.
for (Value::use_iterator UI = PHIIn->use_begin(), E = PHIIn->use_end();
UI != E; ++UI) {
- if (BitCastInst *BCI = dyn_cast<BitCastInst>(*UI))
- if (BCI->getType() == BC->getType() &&
- (!DT || DT->dominates(BCI->getParent(), PredBB)))
- return BCI;
+ if (CastInst *CastI = dyn_cast<CastInst>(*UI))
+ if (CastI->getOpcode() == Cast->getOpcode() &&
+ CastI->getType() == Cast->getType() &&
+ (!DT || DT->dominates(CastI->getParent(), PredBB)))
+ return CastI;
}
return 0;
}
-
+
// Handle getelementptr with at least one PHI translatable operand.
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Inst)) {
SmallVector<Value*, 8> GEPOps;
@@ -208,22 +219,22 @@
for (unsigned i = 0, e = GEP->getNumOperands(); i != e; ++i) {
Value *GEPOp = PHITranslateSubExpr(GEP->getOperand(i), CurBB, PredBB, DT);
if (GEPOp == 0) return 0;
-
+
AnyChanged |= GEPOp != GEP->getOperand(i);
GEPOps.push_back(GEPOp);
}
-
+
if (!AnyChanged)
return GEP;
-
+
// Simplify the GEP to handle 'gep x, 0' -> x etc.
- if (Value *V = SimplifyGEPInst(&GEPOps[0], GEPOps.size(), TD)) {
+ if (Value *V = SimplifyGEPInst(GEPOps, TD, DT)) {
for (unsigned i = 0, e = GEPOps.size(); i != e; ++i)
RemoveInstInputs(GEPOps[i], InstInputs);
-
+
return AddAsInput(V);
}
-
+
// Scan to see if we have this GEP available.
Value *APHIOp = GEPOps[0];
for (Value::use_iterator UI = APHIOp->use_begin(), E = APHIOp->use_end();
@@ -245,7 +256,7 @@
}
return 0;
}
-
+
// Handle add with a constant RHS.
if (Inst->getOpcode() == Instruction::Add &&
isa<ConstantInt>(Inst->getOperand(1))) {
@@ -253,10 +264,10 @@
Constant *RHS = cast<ConstantInt>(Inst->getOperand(1));
bool isNSW = cast<BinaryOperator>(Inst)->hasNoSignedWrap();
bool isNUW = cast<BinaryOperator>(Inst)->hasNoUnsignedWrap();
-
+
Value *LHS = PHITranslateSubExpr(Inst->getOperand(0), CurBB, PredBB, DT);
if (LHS == 0) return 0;
-
+
// If the PHI translated LHS is an add of a constant, fold the immediates.
if (BinaryOperator *BOp = dyn_cast<BinaryOperator>(LHS))
if (BOp->getOpcode() == Instruction::Add)
@@ -264,16 +275,16 @@
LHS = BOp->getOperand(0);
RHS = ConstantExpr::getAdd(RHS, CI);
isNSW = isNUW = false;
-
+
// If the old 'LHS' was an input, add the new 'LHS' as an input.
if (std::count(InstInputs.begin(), InstInputs.end(), BOp)) {
RemoveInstInputs(BOp, InstInputs);
AddAsInput(LHS);
}
}
-
+
// See if the add simplifies away.
- if (Value *Res = SimplifyAddInst(LHS, RHS, isNSW, isNUW, TD)) {
+ if (Value *Res = SimplifyAddInst(LHS, RHS, isNSW, isNUW, TD, DT)) {
// If we simplified the operands, the LHS is no longer an input, but Res
// is.
RemoveInstInputs(LHS, InstInputs);
@@ -283,7 +294,7 @@
// If we didn't modify the add, just return it.
if (LHS == Inst->getOperand(0) && RHS == Inst->getOperand(1))
return Inst;
-
+
// Otherwise, see if we have this add available somewhere.
for (Value::use_iterator UI = LHS->use_begin(), E = LHS->use_end();
UI != E; ++UI) {
@@ -294,10 +305,10 @@
(!DT || DT->dominates(BO->getParent(), PredBB)))
return BO;
}
-
+
return 0;
}
-
+
// Otherwise, we failed.
return 0;
}
@@ -335,13 +346,13 @@
const DominatorTree &DT,
SmallVectorImpl<Instruction*> &NewInsts) {
unsigned NISize = NewInsts.size();
-
+
// Attempt to PHI translate with insertion.
Addr = InsertPHITranslatedSubExpr(Addr, CurBB, PredBB, DT, NewInsts);
-
+
// If successful, return the new value.
if (Addr) return Addr;
-
+
// If not, destroy any intermediate instructions inserted.
while (NewInsts.size() != NISize)
NewInsts.pop_back_val()->eraseFromParent();
@@ -367,20 +378,23 @@
// If we don't have an available version of this value, it must be an
// instruction.
Instruction *Inst = cast<Instruction>(InVal);
-
- // Handle bitcast of PHI translatable value.
- if (BitCastInst *BC = dyn_cast<BitCastInst>(Inst)) {
- Value *OpVal = InsertPHITranslatedSubExpr(BC->getOperand(0),
+
+ // Handle cast of PHI translatable value.
+ if (CastInst *Cast = dyn_cast<CastInst>(Inst)) {
+ if (!Cast->isSafeToSpeculativelyExecute()) return 0;
+ Value *OpVal = InsertPHITranslatedSubExpr(Cast->getOperand(0),
CurBB, PredBB, DT, NewInsts);
if (OpVal == 0) return 0;
-
- // Otherwise insert a bitcast at the end of PredBB.
- BitCastInst *New = new BitCastInst(OpVal, InVal->getType(),
- PredBB->getTerminator());
+
+ // Otherwise insert a cast at the end of PredBB.
+ CastInst *New = CastInst::Create(Cast->getOpcode(),
+ OpVal, InVal->getType(),
+ InVal->getName()+".phi.trans.insert",
+ PredBB->getTerminator());
NewInsts.push_back(New);
return New;
}
-
+
// Handle getelementptr with at least one PHI operand.
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Inst)) {
SmallVector<Value*, 8> GEPOps;
@@ -391,20 +405,21 @@
if (OpVal == 0) return 0;
GEPOps.push_back(OpVal);
}
-
- GetElementPtrInst *Result =
- GetElementPtrInst::Create(GEPOps[0], GEPOps.begin()+1, GEPOps.end(),
- PredBB->getTerminator());
+
+ GetElementPtrInst *Result =
+ GetElementPtrInst::Create(GEPOps[0], makeArrayRef(GEPOps).slice(1),
+ InVal->getName()+".phi.trans.insert",
+ PredBB->getTerminator());
Result->setIsInBounds(GEP->isInBounds());
NewInsts.push_back(Result);
return Result;
}
-
+
#if 0
// FIXME: This code works, but it is unclear that we actually want to insert
// a big chain of computation in order to make a value available in a block.
// This needs to be evaluated carefully to consider its cost trade offs.
-
+
// Handle add with a constant RHS.
if (Inst->getOpcode() == Instruction::Add &&
isa<ConstantInt>(Inst->getOperand(1))) {
@@ -412,7 +427,7 @@
Value *OpVal = InsertPHITranslatedSubExpr(Inst->getOperand(0),
CurBB, PredBB, DT, NewInsts);
if (OpVal == 0) return 0;
-
+
BinaryOperator *Res = BinaryOperator::CreateAdd(OpVal, Inst->getOperand(1),
InVal->getName()+".phi.trans.insert",
PredBB->getTerminator());
@@ -422,6 +437,6 @@
return Res;
}
#endif
-
+
return 0;
}
diff --git a/src/LLVM/lib/Analysis/PathNumbering.cpp b/src/LLVM/lib/Analysis/PathNumbering.cpp
new file mode 100644
index 0000000..0e3b6e6
--- /dev/null
+++ b/src/LLVM/lib/Analysis/PathNumbering.cpp
@@ -0,0 +1,522 @@
+//===- PathNumbering.cpp --------------------------------------*- C++ -*---===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Ball-Larus path numbers uniquely identify paths through a directed acyclic
+// graph (DAG) [Ball96]. For a CFG backedges are removed and replaced by phony
+// edges to obtain a DAG, and thus the unique path numbers [Ball96].
+//
+// The purpose of this analysis is to enumerate the edges in a CFG in order
+// to obtain paths from path numbers in a convenient manner. As described in
+// [Ball96] edges can be enumerated such that given a path number by following
+// the CFG and updating the path number, the path is obtained.
+//
+// [Ball96]
+// T. Ball and J. R. Larus. "Efficient Path Profiling."
+// International Symposium on Microarchitecture, pages 46-57, 1996.
+// http://portal.acm.org/citation.cfm?id=243857
+//
+//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "ball-larus-numbering"
+
+#include "llvm/Analysis/PathNumbering.h"
+#include "llvm/Constants.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/InstrTypes.h"
+#include "llvm/Instructions.h"
+#include "llvm/Module.h"
+#include "llvm/Pass.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Compiler.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/TypeBuilder.h"
+#include "llvm/Support/raw_ostream.h"
+
+#include <queue>
+#include <stack>
+#include <string>
+#include <utility>
+#include <sstream>
+
+using namespace llvm;
+
+// Are we enabling early termination
+static cl::opt<bool> ProcessEarlyTermination(
+ "path-profile-early-termination", cl::Hidden,
+ cl::desc("In path profiling, insert extra instrumentation to account for "
+ "unexpected function termination."));
+
+// Returns the basic block for the BallLarusNode
+BasicBlock* BallLarusNode::getBlock() {
+ return(_basicBlock);
+}
+
+// Returns the number of paths to the exit starting at the node.
+unsigned BallLarusNode::getNumberPaths() {
+ return(_numberPaths);
+}
+
+// Sets the number of paths to the exit starting at the node.
+void BallLarusNode::setNumberPaths(unsigned numberPaths) {
+ _numberPaths = numberPaths;
+}
+
+// Gets the NodeColor used in graph algorithms.
+BallLarusNode::NodeColor BallLarusNode::getColor() {
+ return(_color);
+}
+
+// Sets the NodeColor used in graph algorithms.
+void BallLarusNode::setColor(BallLarusNode::NodeColor color) {
+ _color = color;
+}
+
+// Returns an iterator over predecessor edges. Includes phony and
+// backedges.
+BLEdgeIterator BallLarusNode::predBegin() {
+ return(_predEdges.begin());
+}
+
+// Returns the end sentinel for the predecessor iterator.
+BLEdgeIterator BallLarusNode::predEnd() {
+ return(_predEdges.end());
+}
+
+// Returns the number of predecessor edges. Includes phony and
+// backedges.
+unsigned BallLarusNode::getNumberPredEdges() {
+ return(_predEdges.size());
+}
+
+// Returns an iterator over successor edges. Includes phony and
+// backedges.
+BLEdgeIterator BallLarusNode::succBegin() {
+ return(_succEdges.begin());
+}
+
+// Returns the end sentinel for the successor iterator.
+BLEdgeIterator BallLarusNode::succEnd() {
+ return(_succEdges.end());
+}
+
+// Returns the number of successor edges. Includes phony and
+// backedges.
+unsigned BallLarusNode::getNumberSuccEdges() {
+ return(_succEdges.size());
+}
+
+// Add an edge to the predecessor list.
+void BallLarusNode::addPredEdge(BallLarusEdge* edge) {
+ _predEdges.push_back(edge);
+}
+
+// Remove an edge from the predecessor list.
+void BallLarusNode::removePredEdge(BallLarusEdge* edge) {
+ removeEdge(_predEdges, edge);
+}
+
+// Add an edge to the successor list.
+void BallLarusNode::addSuccEdge(BallLarusEdge* edge) {
+ _succEdges.push_back(edge);
+}
+
+// Remove an edge from the successor list.
+void BallLarusNode::removeSuccEdge(BallLarusEdge* edge) {
+ removeEdge(_succEdges, edge);
+}
+
+// Returns the name of the BasicBlock being represented. If BasicBlock
+// is null then returns "<null>". If BasicBlock has no name, then
+// "<unnamed>" is returned. Intended for use with debug output.
+std::string BallLarusNode::getName() {
+ std::stringstream name;
+
+ if(getBlock() != NULL) {
+ if(getBlock()->hasName()) {
+ std::string tempName(getBlock()->getName());
+ name << tempName.c_str() << " (" << _uid << ")";
+ } else
+ name << "<unnamed> (" << _uid << ")";
+ } else
+ name << "<null> (" << _uid << ")";
+
+ return name.str();
+}
+
+// Removes an edge from an edgeVector. Used by removePredEdge and
+// removeSuccEdge.
+void BallLarusNode::removeEdge(BLEdgeVector& v, BallLarusEdge* e) {
+ // TODO: Avoid linear scan by using a set instead
+ for(BLEdgeIterator i = v.begin(),
+ end = v.end();
+ i != end;
+ ++i) {
+ if((*i) == e) {
+ v.erase(i);
+ break;
+ }
+ }
+}
+
+// Returns the source node of this edge.
+BallLarusNode* BallLarusEdge::getSource() const {
+ return(_source);
+}
+
+// Returns the target node of this edge.
+BallLarusNode* BallLarusEdge::getTarget() const {
+ return(_target);
+}
+
+// Sets the type of the edge.
+BallLarusEdge::EdgeType BallLarusEdge::getType() const {
+ return _edgeType;
+}
+
+// Gets the type of the edge.
+void BallLarusEdge::setType(EdgeType type) {
+ _edgeType = type;
+}
+
+// Returns the weight of this edge. Used to decode path numbers to sequences
+// of basic blocks.
+unsigned BallLarusEdge::getWeight() {
+ return(_weight);
+}
+
+// Sets the weight of the edge. Used during path numbering.
+void BallLarusEdge::setWeight(unsigned weight) {
+ _weight = weight;
+}
+
+// Gets the phony edge originating at the root.
+BallLarusEdge* BallLarusEdge::getPhonyRoot() {
+ return _phonyRoot;
+}
+
+// Sets the phony edge originating at the root.
+void BallLarusEdge::setPhonyRoot(BallLarusEdge* phonyRoot) {
+ _phonyRoot = phonyRoot;
+}
+
+// Gets the phony edge terminating at the exit.
+BallLarusEdge* BallLarusEdge::getPhonyExit() {
+ return _phonyExit;
+}
+
+// Sets the phony edge terminating at the exit.
+void BallLarusEdge::setPhonyExit(BallLarusEdge* phonyExit) {
+ _phonyExit = phonyExit;
+}
+
+// Gets the associated real edge if this is a phony edge.
+BallLarusEdge* BallLarusEdge::getRealEdge() {
+ return _realEdge;
+}
+
+// Sets the associated real edge if this is a phony edge.
+void BallLarusEdge::setRealEdge(BallLarusEdge* realEdge) {
+ _realEdge = realEdge;
+}
+
+// Returns the duplicate number of the edge.
+unsigned BallLarusEdge::getDuplicateNumber() {
+ return(_duplicateNumber);
+}
+
+// Initialization that requires virtual functions which are not fully
+// functional in the constructor.
+void BallLarusDag::init() {
+ BLBlockNodeMap inDag;
+ std::stack<BallLarusNode*> dfsStack;
+
+ _root = addNode(&(_function.getEntryBlock()));
+ _exit = addNode(NULL);
+
+ // start search from root
+ dfsStack.push(getRoot());
+
+ // dfs to add each bb into the dag
+ while(dfsStack.size())
+ buildNode(inDag, dfsStack);
+
+ // put in the final edge
+ addEdge(getExit(),getRoot(),0);
+}
+
+// Frees all memory associated with the DAG.
+BallLarusDag::~BallLarusDag() {
+ for(BLEdgeIterator edge = _edges.begin(), end = _edges.end(); edge != end;
+ ++edge)
+ delete (*edge);
+
+ for(BLNodeIterator node = _nodes.begin(), end = _nodes.end(); node != end;
+ ++node)
+ delete (*node);
+}
+
+// Calculate the path numbers by assigning edge increments as prescribed
+// in Ball-Larus path profiling.
+void BallLarusDag::calculatePathNumbers() {
+ BallLarusNode* node;
+ std::queue<BallLarusNode*> bfsQueue;
+ bfsQueue.push(getExit());
+
+ while(bfsQueue.size() > 0) {
+ node = bfsQueue.front();
+
+ DEBUG(dbgs() << "calculatePathNumbers on " << node->getName() << "\n");
+
+ bfsQueue.pop();
+ unsigned prevPathNumber = node->getNumberPaths();
+ calculatePathNumbersFrom(node);
+
+ // Check for DAG splitting
+ if( node->getNumberPaths() > 100000000 && node != getRoot() ) {
+ // Add new phony edge from the split-node to the DAG's exit
+ BallLarusEdge* exitEdge = addEdge(node, getExit(), 0);
+ exitEdge->setType(BallLarusEdge::SPLITEDGE_PHONY);
+
+ // Counters to handle the possibility of a multi-graph
+ BasicBlock* oldTarget = 0;
+ unsigned duplicateNumber = 0;
+
+ // Iterate through each successor edge, adding phony edges
+ for( BLEdgeIterator succ = node->succBegin(), end = node->succEnd();
+ succ != end; oldTarget = (*succ)->getTarget()->getBlock(), succ++ ) {
+
+ if( (*succ)->getType() == BallLarusEdge::NORMAL ) {
+ // is this edge a duplicate?
+ if( oldTarget != (*succ)->getTarget()->getBlock() )
+ duplicateNumber = 0;
+
+ // create the new phony edge: root -> succ
+ BallLarusEdge* rootEdge =
+ addEdge(getRoot(), (*succ)->getTarget(), duplicateNumber++);
+ rootEdge->setType(BallLarusEdge::SPLITEDGE_PHONY);
+ rootEdge->setRealEdge(*succ);
+
+ // split on this edge and reference it's exit/root phony edges
+ (*succ)->setType(BallLarusEdge::SPLITEDGE);
+ (*succ)->setPhonyRoot(rootEdge);
+ (*succ)->setPhonyExit(exitEdge);
+ (*succ)->setWeight(0);
+ }
+ }
+
+ calculatePathNumbersFrom(node);
+ }
+
+ DEBUG(dbgs() << "prev, new number paths " << prevPathNumber << ", "
+ << node->getNumberPaths() << ".\n");
+
+ if(prevPathNumber == 0 && node->getNumberPaths() != 0) {
+ DEBUG(dbgs() << "node ready : " << node->getName() << "\n");
+ for(BLEdgeIterator pred = node->predBegin(), end = node->predEnd();
+ pred != end; pred++) {
+ if( (*pred)->getType() == BallLarusEdge::BACKEDGE ||
+ (*pred)->getType() == BallLarusEdge::SPLITEDGE )
+ continue;
+
+ BallLarusNode* nextNode = (*pred)->getSource();
+ // not yet visited?
+ if(nextNode->getNumberPaths() == 0)
+ bfsQueue.push(nextNode);
+ }
+ }
+ }
+
+ DEBUG(dbgs() << "\tNumber of paths: " << getRoot()->getNumberPaths() << "\n");
+}
+
+// Returns the number of paths for the Dag.
+unsigned BallLarusDag::getNumberOfPaths() {
+ return(getRoot()->getNumberPaths());
+}
+
+// Returns the root (i.e. entry) node for the DAG.
+BallLarusNode* BallLarusDag::getRoot() {
+ return _root;
+}
+
+// Returns the exit node for the DAG.
+BallLarusNode* BallLarusDag::getExit() {
+ return _exit;
+}
+
+// Returns the function for the DAG.
+Function& BallLarusDag::getFunction() {
+ return(_function);
+}
+
+// Clears the node colors.
+void BallLarusDag::clearColors(BallLarusNode::NodeColor color) {
+ for (BLNodeIterator nodeIt = _nodes.begin(); nodeIt != _nodes.end(); nodeIt++)
+ (*nodeIt)->setColor(color);
+}
+
+// Processes one node and its imediate edges for building the DAG.
+void BallLarusDag::buildNode(BLBlockNodeMap& inDag, BLNodeStack& dfsStack) {
+ BallLarusNode* currentNode = dfsStack.top();
+ BasicBlock* currentBlock = currentNode->getBlock();
+
+ if(currentNode->getColor() != BallLarusNode::WHITE) {
+ // we have already visited this node
+ dfsStack.pop();
+ currentNode->setColor(BallLarusNode::BLACK);
+ } else {
+ // are there any external procedure calls?
+ if( ProcessEarlyTermination ) {
+ for( BasicBlock::iterator bbCurrent = currentNode->getBlock()->begin(),
+ bbEnd = currentNode->getBlock()->end(); bbCurrent != bbEnd;
+ bbCurrent++ ) {
+ Instruction& instr = *bbCurrent;
+ if( instr.getOpcode() == Instruction::Call ) {
+ BallLarusEdge* callEdge = addEdge(currentNode, getExit(), 0);
+ callEdge->setType(BallLarusEdge::CALLEDGE_PHONY);
+ break;
+ }
+ }
+ }
+
+ TerminatorInst* terminator = currentNode->getBlock()->getTerminator();
+ if(isa<ReturnInst>(terminator) || isa<UnreachableInst>(terminator)
+ || isa<ResumeInst>(terminator) || isa<UnwindInst>(terminator))
+ addEdge(currentNode, getExit(),0);
+
+ currentNode->setColor(BallLarusNode::GRAY);
+ inDag[currentBlock] = currentNode;
+
+ BasicBlock* oldSuccessor = 0;
+ unsigned duplicateNumber = 0;
+
+ // iterate through this node's successors
+ for(succ_iterator successor = succ_begin(currentBlock),
+ succEnd = succ_end(currentBlock); successor != succEnd;
+ oldSuccessor = *successor, ++successor ) {
+ BasicBlock* succBB = *successor;
+
+ // is this edge a duplicate?
+ if (oldSuccessor == succBB)
+ duplicateNumber++;
+ else
+ duplicateNumber = 0;
+
+ buildEdge(inDag, dfsStack, currentNode, succBB, duplicateNumber);
+ }
+ }
+}
+
+// Process an edge in the CFG for DAG building.
+void BallLarusDag::buildEdge(BLBlockNodeMap& inDag, std::stack<BallLarusNode*>&
+ dfsStack, BallLarusNode* currentNode,
+ BasicBlock* succBB, unsigned duplicateCount) {
+ BallLarusNode* succNode = inDag[succBB];
+
+ if(succNode && succNode->getColor() == BallLarusNode::BLACK) {
+ // visited node and forward edge
+ addEdge(currentNode, succNode, duplicateCount);
+ } else if(succNode && succNode->getColor() == BallLarusNode::GRAY) {
+ // visited node and back edge
+ DEBUG(dbgs() << "Backedge detected.\n");
+ addBackedge(currentNode, succNode, duplicateCount);
+ } else {
+ BallLarusNode* childNode;
+ // not visited node and forward edge
+ if(succNode) // an unvisited node that is child of a gray node
+ childNode = succNode;
+ else { // an unvisited node that is a child of a an unvisted node
+ childNode = addNode(succBB);
+ inDag[succBB] = childNode;
+ }
+ addEdge(currentNode, childNode, duplicateCount);
+ dfsStack.push(childNode);
+ }
+}
+
+// The weight on each edge is the increment required along any path that
+// contains that edge.
+void BallLarusDag::calculatePathNumbersFrom(BallLarusNode* node) {
+ if(node == getExit())
+ // The Exit node must be base case
+ node->setNumberPaths(1);
+ else {
+ unsigned sumPaths = 0;
+ BallLarusNode* succNode;
+
+ for(BLEdgeIterator succ = node->succBegin(), end = node->succEnd();
+ succ != end; succ++) {
+ if( (*succ)->getType() == BallLarusEdge::BACKEDGE ||
+ (*succ)->getType() == BallLarusEdge::SPLITEDGE )
+ continue;
+
+ (*succ)->setWeight(sumPaths);
+ succNode = (*succ)->getTarget();
+
+ if( !succNode->getNumberPaths() )
+ return;
+ sumPaths += succNode->getNumberPaths();
+ }
+
+ node->setNumberPaths(sumPaths);
+ }
+}
+
+// Allows subclasses to determine which type of Node is created.
+// Override this method to produce subclasses of BallLarusNode if
+// necessary. The destructor of BallLarusDag will call free on each
+// pointer created.
+BallLarusNode* BallLarusDag::createNode(BasicBlock* BB) {
+ return( new BallLarusNode(BB) );
+}
+
+// Allows subclasses to determine which type of Edge is created.
+// Override this method to produce subclasses of BallLarusEdge if
+// necessary. The destructor of BallLarusDag will call free on each
+// pointer created.
+BallLarusEdge* BallLarusDag::createEdge(BallLarusNode* source,
+ BallLarusNode* target,
+ unsigned duplicateCount) {
+ return( new BallLarusEdge(source, target, duplicateCount) );
+}
+
+// Proxy to node's constructor. Updates the DAG state.
+BallLarusNode* BallLarusDag::addNode(BasicBlock* BB) {
+ BallLarusNode* newNode = createNode(BB);
+ _nodes.push_back(newNode);
+ return( newNode );
+}
+
+// Proxy to edge's constructor. Updates the DAG state.
+BallLarusEdge* BallLarusDag::addEdge(BallLarusNode* source,
+ BallLarusNode* target,
+ unsigned duplicateCount) {
+ BallLarusEdge* newEdge = createEdge(source, target, duplicateCount);
+ _edges.push_back(newEdge);
+ source->addSuccEdge(newEdge);
+ target->addPredEdge(newEdge);
+ return(newEdge);
+}
+
+// Adds a backedge with its phony edges. Updates the DAG state.
+void BallLarusDag::addBackedge(BallLarusNode* source, BallLarusNode* target,
+ unsigned duplicateCount) {
+ BallLarusEdge* childEdge = addEdge(source, target, duplicateCount);
+ childEdge->setType(BallLarusEdge::BACKEDGE);
+
+ childEdge->setPhonyRoot(addEdge(getRoot(), target,0));
+ childEdge->setPhonyExit(addEdge(source, getExit(),0));
+
+ childEdge->getPhonyRoot()->setRealEdge(childEdge);
+ childEdge->getPhonyRoot()->setType(BallLarusEdge::BACKEDGE_PHONY);
+
+ childEdge->getPhonyExit()->setRealEdge(childEdge);
+ childEdge->getPhonyExit()->setType(BallLarusEdge::BACKEDGE_PHONY);
+ _backEdges.push_back(childEdge);
+}
diff --git a/src/LLVM/lib/Analysis/PathProfileInfo.cpp b/src/LLVM/lib/Analysis/PathProfileInfo.cpp
new file mode 100644
index 0000000..b361d3f
--- /dev/null
+++ b/src/LLVM/lib/Analysis/PathProfileInfo.cpp
@@ -0,0 +1,434 @@
+//===- PathProfileInfo.cpp ------------------------------------*- C++ -*---===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the interface used by optimizers to load path profiles,
+// and provides a loader pass which reads a path profile file.
+//
+//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "path-profile-info"
+
+#include "llvm/Module.h"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/ProfileInfoTypes.h"
+#include "llvm/Analysis/PathProfileInfo.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+
+#include <cstdio>
+
+using namespace llvm;
+
+// command line option for loading path profiles
+static cl::opt<std::string>
+PathProfileInfoFilename("path-profile-loader-file", cl::init("llvmprof.out"),
+ cl::value_desc("filename"),
+ cl::desc("Path profile file loaded by -path-profile-loader"), cl::Hidden);
+
+namespace {
+ class PathProfileLoaderPass : public ModulePass, public PathProfileInfo {
+ public:
+ PathProfileLoaderPass() : ModulePass(ID) { }
+ ~PathProfileLoaderPass();
+
+ // this pass doesn't change anything (only loads information)
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+
+ // the full name of the loader pass
+ virtual const char* getPassName() const {
+ return "Path Profiling Information Loader";
+ }
+
+ // required since this pass implements multiple inheritance
+ virtual void *getAdjustedAnalysisPointer(AnalysisID PI) {
+ if (PI == &PathProfileInfo::ID)
+ return (PathProfileInfo*)this;
+ return this;
+ }
+
+ // entry point to run the pass
+ bool runOnModule(Module &M);
+
+ // pass identification
+ static char ID;
+
+ private:
+ // make a reference table to refer to function by number
+ void buildFunctionRefs(Module &M);
+
+ // process argument info of a program from the input file
+ void handleArgumentInfo();
+
+ // process path number information from the input file
+ void handlePathInfo();
+
+ // array of references to the functions in the module
+ std::vector<Function*> _functions;
+
+ // path profile file handle
+ FILE* _file;
+
+ // path profile file name
+ std::string _filename;
+ };
+}
+
+// register PathLoader
+char PathProfileLoaderPass::ID = 0;
+
+INITIALIZE_ANALYSIS_GROUP(PathProfileInfo, "Path Profile Information",
+ NoPathProfileInfo)
+INITIALIZE_AG_PASS(PathProfileLoaderPass, PathProfileInfo,
+ "path-profile-loader",
+ "Load path profile information from file",
+ false, true, false)
+
+char &llvm::PathProfileLoaderPassID = PathProfileLoaderPass::ID;
+
+// link PathLoader as a pass, and make it available as an optimisation
+ModulePass *llvm::createPathProfileLoaderPass() {
+ return new PathProfileLoaderPass;
+}
+
+// ----------------------------------------------------------------------------
+// PathEdge implementation
+//
+ProfilePathEdge::ProfilePathEdge (BasicBlock* source, BasicBlock* target,
+ unsigned duplicateNumber)
+ : _source(source), _target(target), _duplicateNumber(duplicateNumber) {}
+
+// ----------------------------------------------------------------------------
+// Path implementation
+//
+
+ProfilePath::ProfilePath (unsigned int number, unsigned int count,
+ double countStdDev, PathProfileInfo* ppi)
+ : _number(number) , _count(count), _countStdDev(countStdDev), _ppi(ppi) {}
+
+double ProfilePath::getFrequency() const {
+ return 100 * double(_count) /
+ double(_ppi->_functionPathCounts[_ppi->_currentFunction]);
+}
+
+static BallLarusEdge* getNextEdge (BallLarusNode* node,
+ unsigned int pathNumber) {
+ BallLarusEdge* best = 0;
+
+ for( BLEdgeIterator next = node->succBegin(),
+ end = node->succEnd(); next != end; next++ ) {
+ if( (*next)->getType() != BallLarusEdge::BACKEDGE && // no backedges
+ (*next)->getType() != BallLarusEdge::SPLITEDGE && // no split edges
+ (*next)->getWeight() <= pathNumber && // weight must be <= pathNumber
+ (!best || (best->getWeight() < (*next)->getWeight())) ) // best one?
+ best = *next;
+ }
+
+ return best;
+}
+
+ProfilePathEdgeVector* ProfilePath::getPathEdges() const {
+ BallLarusNode* currentNode = _ppi->_currentDag->getRoot ();
+ unsigned int increment = _number;
+ ProfilePathEdgeVector* pev = new ProfilePathEdgeVector;
+
+ while (currentNode != _ppi->_currentDag->getExit()) {
+ BallLarusEdge* next = getNextEdge(currentNode, increment);
+
+ increment -= next->getWeight();
+
+ if( next->getType() != BallLarusEdge::BACKEDGE_PHONY &&
+ next->getType() != BallLarusEdge::SPLITEDGE_PHONY &&
+ next->getTarget() != _ppi->_currentDag->getExit() )
+ pev->push_back(ProfilePathEdge(
+ next->getSource()->getBlock(),
+ next->getTarget()->getBlock(),
+ next->getDuplicateNumber()));
+
+ if( next->getType() == BallLarusEdge::BACKEDGE_PHONY &&
+ next->getTarget() == _ppi->_currentDag->getExit() )
+ pev->push_back(ProfilePathEdge(
+ next->getRealEdge()->getSource()->getBlock(),
+ next->getRealEdge()->getTarget()->getBlock(),
+ next->getDuplicateNumber()));
+
+ if( next->getType() == BallLarusEdge::SPLITEDGE_PHONY &&
+ next->getSource() == _ppi->_currentDag->getRoot() )
+ pev->push_back(ProfilePathEdge(
+ next->getRealEdge()->getSource()->getBlock(),
+ next->getRealEdge()->getTarget()->getBlock(),
+ next->getDuplicateNumber()));
+
+ // set the new node
+ currentNode = next->getTarget();
+ }
+
+ return pev;
+}
+
+ProfilePathBlockVector* ProfilePath::getPathBlocks() const {
+ BallLarusNode* currentNode = _ppi->_currentDag->getRoot ();
+ unsigned int increment = _number;
+ ProfilePathBlockVector* pbv = new ProfilePathBlockVector;
+
+ while (currentNode != _ppi->_currentDag->getExit()) {
+ BallLarusEdge* next = getNextEdge(currentNode, increment);
+ increment -= next->getWeight();
+
+ // add block to the block list if it is a real edge
+ if( next->getType() == BallLarusEdge::NORMAL)
+ pbv->push_back (currentNode->getBlock());
+ // make the back edge the last edge since we are at the end
+ else if( next->getTarget() == _ppi->_currentDag->getExit() ) {
+ pbv->push_back (currentNode->getBlock());
+ pbv->push_back (next->getRealEdge()->getTarget()->getBlock());
+ }
+
+ // set the new node
+ currentNode = next->getTarget();
+ }
+
+ return pbv;
+}
+
+BasicBlock* ProfilePath::getFirstBlockInPath() const {
+ BallLarusNode* root = _ppi->_currentDag->getRoot();
+ BallLarusEdge* edge = getNextEdge(root, _number);
+
+ if( edge && (edge->getType() == BallLarusEdge::BACKEDGE_PHONY ||
+ edge->getType() == BallLarusEdge::SPLITEDGE_PHONY) )
+ return edge->getTarget()->getBlock();
+
+ return root->getBlock();
+}
+
+// ----------------------------------------------------------------------------
+// PathProfileInfo implementation
+//
+
+// Pass identification
+char llvm::PathProfileInfo::ID = 0;
+
+PathProfileInfo::PathProfileInfo () : _currentDag(0) , _currentFunction(0) {
+}
+
+PathProfileInfo::~PathProfileInfo() {
+ if (_currentDag)
+ delete _currentDag;
+}
+
+// set the function for which paths are currently begin processed
+void PathProfileInfo::setCurrentFunction(Function* F) {
+ // Make sure it exists
+ if (!F) return;
+
+ if (_currentDag)
+ delete _currentDag;
+
+ _currentFunction = F;
+ _currentDag = new BallLarusDag(*F);
+ _currentDag->init();
+ _currentDag->calculatePathNumbers();
+}
+
+// get the function for which paths are currently being processed
+Function* PathProfileInfo::getCurrentFunction() const {
+ return _currentFunction;
+}
+
+// get the entry block of the function
+BasicBlock* PathProfileInfo::getCurrentFunctionEntry() {
+ return _currentDag->getRoot()->getBlock();
+}
+
+// return the path based on its number
+ProfilePath* PathProfileInfo::getPath(unsigned int number) {
+ return _functionPaths[_currentFunction][number];
+}
+
+// return the number of paths which a function may potentially execute
+unsigned int PathProfileInfo::getPotentialPathCount() {
+ return _currentDag ? _currentDag->getNumberOfPaths() : 0;
+}
+
+// return an iterator for the beginning of a functions executed paths
+ProfilePathIterator PathProfileInfo::pathBegin() {
+ return _functionPaths[_currentFunction].begin();
+}
+
+// return an iterator for the end of a functions executed paths
+ProfilePathIterator PathProfileInfo::pathEnd() {
+ return _functionPaths[_currentFunction].end();
+}
+
+// returns the total number of paths run in the function
+unsigned int PathProfileInfo::pathsRun() {
+ return _currentFunction ? _functionPaths[_currentFunction].size() : 0;
+}
+
+// ----------------------------------------------------------------------------
+// PathLoader implementation
+//
+
+// remove all generated paths
+PathProfileLoaderPass::~PathProfileLoaderPass() {
+ for( FunctionPathIterator funcNext = _functionPaths.begin(),
+ funcEnd = _functionPaths.end(); funcNext != funcEnd; funcNext++)
+ for( ProfilePathIterator pathNext = funcNext->second.begin(),
+ pathEnd = funcNext->second.end(); pathNext != pathEnd; pathNext++)
+ delete pathNext->second;
+}
+
+// entry point of the pass; this loads and parses a file
+bool PathProfileLoaderPass::runOnModule(Module &M) {
+ // get the filename and setup the module's function references
+ _filename = PathProfileInfoFilename;
+ buildFunctionRefs (M);
+
+ if (!(_file = fopen(_filename.c_str(), "rb"))) {
+ errs () << "error: input '" << _filename << "' file does not exist.\n";
+ return false;
+ }
+
+ ProfilingType profType;
+
+ while( fread(&profType, sizeof(ProfilingType), 1, _file) ) {
+ switch (profType) {
+ case ArgumentInfo:
+ handleArgumentInfo ();
+ break;
+ case PathInfo:
+ handlePathInfo ();
+ break;
+ default:
+ errs () << "error: bad path profiling file syntax, " << profType << "\n";
+ fclose (_file);
+ return false;
+ }
+ }
+
+ fclose (_file);
+
+ return true;
+}
+
+// create a reference table for functions defined in the path profile file
+void PathProfileLoaderPass::buildFunctionRefs (Module &M) {
+ _functions.push_back(0); // make the 0 index a null pointer
+
+ for (Module::iterator F = M.begin(), E = M.end(); F != E; F++) {
+ if (F->isDeclaration())
+ continue;
+ _functions.push_back(F);
+ }
+}
+
+// handle command like argument infor in the output file
+void PathProfileLoaderPass::handleArgumentInfo() {
+ // get the argument list's length
+ unsigned savedArgsLength;
+ if( fread(&savedArgsLength, sizeof(unsigned), 1, _file) != 1 ) {
+ errs() << "warning: argument info header/data mismatch\n";
+ return;
+ }
+
+ // allocate a buffer, and get the arguments
+ char* args = new char[savedArgsLength+1];
+ if( fread(args, 1, savedArgsLength, _file) != savedArgsLength )
+ errs() << "warning: argument info header/data mismatch\n";
+
+ args[savedArgsLength] = '\0';
+ argList = std::string(args);
+ delete [] args; // cleanup dynamic string
+
+ // byte alignment
+ if (savedArgsLength & 3)
+ fseek(_file, 4-(savedArgsLength&3), SEEK_CUR);
+}
+
+// Handle path profile information in the output file
+void PathProfileLoaderPass::handlePathInfo () {
+ // get the number of functions in this profile
+ unsigned functionCount;
+ if( fread(&functionCount, sizeof(functionCount), 1, _file) != 1 ) {
+ errs() << "warning: path info header/data mismatch\n";
+ return;
+ }
+
+ // gather path information for each function
+ for (unsigned i = 0; i < functionCount; i++) {
+ PathProfileHeader pathHeader;
+ if( fread(&pathHeader, sizeof(pathHeader), 1, _file) != 1 ) {
+ errs() << "warning: bad header for path function info\n";
+ break;
+ }
+
+ Function* f = _functions[pathHeader.fnNumber];
+
+ // dynamically allocate a table to store path numbers
+ PathProfileTableEntry* pathTable =
+ new PathProfileTableEntry[pathHeader.numEntries];
+
+ if( fread(pathTable, sizeof(PathProfileTableEntry),
+ pathHeader.numEntries, _file) != pathHeader.numEntries) {
+ delete [] pathTable;
+ errs() << "warning: path function info header/data mismatch\n";
+ return;
+ }
+
+ // Build a new path for the current function
+ unsigned int totalPaths = 0;
+ for (unsigned int j = 0; j < pathHeader.numEntries; j++) {
+ totalPaths += pathTable[j].pathCounter;
+ _functionPaths[f][pathTable[j].pathNumber]
+ = new ProfilePath(pathTable[j].pathNumber, pathTable[j].pathCounter,
+ 0, this);
+ }
+
+ _functionPathCounts[f] = totalPaths;
+
+ delete [] pathTable;
+ }
+}
+
+//===----------------------------------------------------------------------===//
+// NoProfile PathProfileInfo implementation
+//
+
+namespace {
+ struct NoPathProfileInfo : public ImmutablePass, public PathProfileInfo {
+ static char ID; // Class identification, replacement for typeinfo
+ NoPathProfileInfo() : ImmutablePass(ID) {
+ initializeNoPathProfileInfoPass(*PassRegistry::getPassRegistry());
+ }
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ virtual void *getAdjustedAnalysisPointer(AnalysisID PI) {
+ if (PI == &PathProfileInfo::ID)
+ return (PathProfileInfo*)this;
+ return this;
+ }
+
+ virtual const char *getPassName() const {
+ return "NoPathProfileInfo";
+ }
+ };
+} // End of anonymous namespace
+
+char NoPathProfileInfo::ID = 0;
+// Register this pass...
+INITIALIZE_AG_PASS(NoPathProfileInfo, PathProfileInfo, "no-path-profile",
+ "No Path Profile Information", false, true, true)
+
+ImmutablePass *llvm::createNoPathProfileInfoPass() { return new NoPathProfileInfo(); }
diff --git a/src/LLVM/lib/Analysis/PathProfileVerifier.cpp b/src/LLVM/lib/Analysis/PathProfileVerifier.cpp
new file mode 100644
index 0000000..0ae734e
--- /dev/null
+++ b/src/LLVM/lib/Analysis/PathProfileVerifier.cpp
@@ -0,0 +1,207 @@
+//===- PathProfileVerifier.cpp --------------------------------*- C++ -*---===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This verifier derives an edge profile file from current path profile
+// information
+//
+//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "path-profile-verifier"
+
+#include "llvm/Module.h"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/ProfileInfoTypes.h"
+#include "llvm/Analysis/PathProfileInfo.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/raw_ostream.h"
+
+#include <stdio.h>
+
+using namespace llvm;
+
+namespace {
+ class PathProfileVerifier : public ModulePass {
+ private:
+ bool runOnModule(Module &M);
+
+ public:
+ static char ID; // Pass identification, replacement for typeid
+ PathProfileVerifier() : ModulePass(ID) {
+ initializePathProfileVerifierPass(*PassRegistry::getPassRegistry());
+ }
+
+
+ virtual const char *getPassName() const {
+ return "Path Profiler Verifier";
+ }
+
+ // The verifier requires the path profile and edge profile.
+ virtual void getAnalysisUsage(AnalysisUsage& AU) const;
+ };
+}
+
+static cl::opt<std::string>
+EdgeProfileFilename("path-profile-verifier-file",
+ cl::init("edgefrompath.llvmprof.out"),
+ cl::value_desc("filename"),
+ cl::desc("Edge profile file generated by -path-profile-verifier"),
+ cl::Hidden);
+
+char PathProfileVerifier::ID = 0;
+INITIALIZE_PASS(PathProfileVerifier, "path-profile-verifier",
+ "Compare the path profile derived edge profile against the "
+ "edge profile.", true, true)
+
+ModulePass *llvm::createPathProfileVerifierPass() {
+ return new PathProfileVerifier();
+}
+
+// The verifier requires the path profile and edge profile.
+void PathProfileVerifier::getAnalysisUsage(AnalysisUsage& AU) const {
+ AU.addRequired<PathProfileInfo>();
+ AU.addPreserved<PathProfileInfo>();
+}
+
+typedef std::map<unsigned, unsigned> DuplicateToIndexMap;
+typedef std::map<BasicBlock*,DuplicateToIndexMap> BlockToDuplicateMap;
+typedef std::map<BasicBlock*,BlockToDuplicateMap> NestedBlockToIndexMap;
+
+// the verifier iterates through each path to gather the total
+// number of edge frequencies
+bool PathProfileVerifier::runOnModule (Module &M) {
+ PathProfileInfo& pathProfileInfo = getAnalysis<PathProfileInfo>();
+
+ // setup a data structure to map path edges which index an
+ // array of edge counters
+ NestedBlockToIndexMap arrayMap;
+ unsigned i = 0;
+ for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
+ if (F->isDeclaration()) continue;
+
+ arrayMap[0][F->begin()][0] = i++;
+
+ for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
+ TerminatorInst *TI = BB->getTerminator();
+
+ unsigned duplicate = 0;
+ BasicBlock* prev = 0;
+ for (unsigned s = 0, e = TI->getNumSuccessors(); s != e;
+ prev = TI->getSuccessor(s), ++s) {
+ if (prev == TI->getSuccessor(s))
+ duplicate++;
+ else duplicate = 0;
+
+ arrayMap[BB][TI->getSuccessor(s)][duplicate] = i++;
+ }
+ }
+ }
+
+ std::vector<unsigned> edgeArray(i);
+
+ // iterate through each path and increment the edge counters as needed
+ for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
+ if (F->isDeclaration()) continue;
+
+ pathProfileInfo.setCurrentFunction(F);
+
+ DEBUG(dbgs() << "function '" << F->getName() << "' ran "
+ << pathProfileInfo.pathsRun()
+ << "/" << pathProfileInfo.getPotentialPathCount()
+ << " potential paths\n");
+
+ for( ProfilePathIterator nextPath = pathProfileInfo.pathBegin(),
+ endPath = pathProfileInfo.pathEnd();
+ nextPath != endPath; nextPath++ ) {
+ ProfilePath* currentPath = nextPath->second;
+
+ ProfilePathEdgeVector* pev = currentPath->getPathEdges();
+ DEBUG(dbgs () << "path #" << currentPath->getNumber() << ": "
+ << currentPath->getCount() << "\n");
+ // setup the entry edge (normally path profiling doesn't care about this)
+ if (currentPath->getFirstBlockInPath() == &F->getEntryBlock())
+ edgeArray[arrayMap[0][currentPath->getFirstBlockInPath()][0]]
+ += currentPath->getCount();
+
+ for( ProfilePathEdgeIterator nextEdge = pev->begin(),
+ endEdge = pev->end(); nextEdge != endEdge; nextEdge++ ) {
+ if (nextEdge != pev->begin())
+ DEBUG(dbgs() << " :: ");
+
+ BasicBlock* source = nextEdge->getSource();
+ BasicBlock* target = nextEdge->getTarget();
+ unsigned duplicateNumber = nextEdge->getDuplicateNumber();
+ DEBUG(dbgs () << source->getNameStr() << " --{" << duplicateNumber
+ << "}--> " << target->getNameStr());
+
+ // Ensure all the referenced edges exist
+ // TODO: make this a separate function
+ if( !arrayMap.count(source) ) {
+ errs() << " error [" << F->getNameStr() << "()]: source '"
+ << source->getNameStr()
+ << "' does not exist in the array map.\n";
+ } else if( !arrayMap[source].count(target) ) {
+ errs() << " error [" << F->getNameStr() << "()]: target '"
+ << target->getNameStr()
+ << "' does not exist in the array map.\n";
+ } else if( !arrayMap[source][target].count(duplicateNumber) ) {
+ errs() << " error [" << F->getNameStr() << "()]: edge "
+ << source->getNameStr() << " -> " << target->getNameStr()
+ << " duplicate number " << duplicateNumber
+ << " does not exist in the array map.\n";
+ } else {
+ edgeArray[arrayMap[source][target][duplicateNumber]]
+ += currentPath->getCount();
+ }
+ }
+
+ DEBUG(errs() << "\n");
+
+ delete pev;
+ }
+ }
+
+ std::string errorInfo;
+ std::string filename = EdgeProfileFilename;
+
+ // Open a handle to the file
+ FILE* edgeFile = fopen(filename.c_str(),"wb");
+
+ if (!edgeFile) {
+ errs() << "error: unable to open file '" << filename << "' for output.\n";
+ return false;
+ }
+
+ errs() << "Generating edge profile '" << filename << "' ...\n";
+
+ // write argument info
+ unsigned type = ArgumentInfo;
+ unsigned num = pathProfileInfo.argList.size();
+ int zeros = 0;
+
+ fwrite(&type,sizeof(unsigned),1,edgeFile);
+ fwrite(&num,sizeof(unsigned),1,edgeFile);
+ fwrite(pathProfileInfo.argList.c_str(),1,num,edgeFile);
+ if (num&3)
+ fwrite(&zeros, 1, 4-(num&3), edgeFile);
+
+ type = EdgeInfo;
+ num = edgeArray.size();
+ fwrite(&type,sizeof(unsigned),1,edgeFile);
+ fwrite(&num,sizeof(unsigned),1,edgeFile);
+
+ // write each edge to the file
+ for( std::vector<unsigned>::iterator s = edgeArray.begin(),
+ e = edgeArray.end(); s != e; s++)
+ fwrite(&*s, sizeof (unsigned), 1, edgeFile);
+
+ fclose (edgeFile);
+
+ return true;
+}
diff --git a/src/LLVM/lib/Analysis/PostDominators.cpp b/src/LLVM/lib/Analysis/PostDominators.cpp
new file mode 100644
index 0000000..6ed2729
--- /dev/null
+++ b/src/LLVM/lib/Analysis/PostDominators.cpp
@@ -0,0 +1,51 @@
+//===- PostDominators.cpp - Post-Dominator Calculation --------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the post-dominator construction algorithms.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "postdomtree"
+
+#include "llvm/Analysis/PostDominators.h"
+#include "llvm/Instructions.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/ADT/SetOperations.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Analysis/DominatorInternals.h"
+using namespace llvm;
+
+//===----------------------------------------------------------------------===//
+// PostDominatorTree Implementation
+//===----------------------------------------------------------------------===//
+
+char PostDominatorTree::ID = 0;
+INITIALIZE_PASS(PostDominatorTree, "postdomtree",
+ "Post-Dominator Tree Construction", true, true)
+
+bool PostDominatorTree::runOnFunction(Function &F) {
+ DT->recalculate(F);
+ return false;
+}
+
+PostDominatorTree::~PostDominatorTree() {
+ delete DT;
+}
+
+void PostDominatorTree::print(raw_ostream &OS, const Module *) const {
+ DT->print(OS);
+}
+
+
+FunctionPass* llvm::createPostDomTree() {
+ return new PostDominatorTree();
+}
+
diff --git a/src/LLVM/lib/Analysis/ProfileEstimatorPass.cpp b/src/LLVM/lib/Analysis/ProfileEstimatorPass.cpp
new file mode 100644
index 0000000..b594e2b
--- /dev/null
+++ b/src/LLVM/lib/Analysis/ProfileEstimatorPass.cpp
@@ -0,0 +1,426 @@
+//===- ProfileEstimatorPass.cpp - LLVM Pass to estimate profile info ------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a concrete implementation of profiling information that
+// estimates the profiling information in a very crude and unimaginative way.
+//
+//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "profile-estimator"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/ProfileInfo.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/Format.h"
+using namespace llvm;
+
+static cl::opt<double>
+LoopWeight(
+ "profile-estimator-loop-weight", cl::init(10),
+ cl::value_desc("loop-weight"),
+ cl::desc("Number of loop executions used for profile-estimator")
+);
+
+namespace {
+ class ProfileEstimatorPass : public FunctionPass, public ProfileInfo {
+ double ExecCount;
+ LoopInfo *LI;
+ std::set<BasicBlock*> BBToVisit;
+ std::map<Loop*,double> LoopExitWeights;
+ std::map<Edge,double> MinimalWeight;
+ public:
+ static char ID; // Class identification, replacement for typeinfo
+ explicit ProfileEstimatorPass(const double execcount = 0)
+ : FunctionPass(ID), ExecCount(execcount) {
+ initializeProfileEstimatorPassPass(*PassRegistry::getPassRegistry());
+ if (execcount == 0) ExecCount = LoopWeight;
+ }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ AU.addRequired<LoopInfo>();
+ }
+
+ virtual const char *getPassName() const {
+ return "Profiling information estimator";
+ }
+
+ /// run - Estimate the profile information from the specified file.
+ virtual bool runOnFunction(Function &F);
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ virtual void *getAdjustedAnalysisPointer(AnalysisID PI) {
+ if (PI == &ProfileInfo::ID)
+ return (ProfileInfo*)this;
+ return this;
+ }
+
+ virtual void recurseBasicBlock(BasicBlock *BB);
+
+ void inline printEdgeWeight(Edge);
+ };
+} // End of anonymous namespace
+
+char ProfileEstimatorPass::ID = 0;
+INITIALIZE_AG_PASS_BEGIN(ProfileEstimatorPass, ProfileInfo, "profile-estimator",
+ "Estimate profiling information", false, true, false)
+INITIALIZE_PASS_DEPENDENCY(LoopInfo)
+INITIALIZE_AG_PASS_END(ProfileEstimatorPass, ProfileInfo, "profile-estimator",
+ "Estimate profiling information", false, true, false)
+
+namespace llvm {
+ char &ProfileEstimatorPassID = ProfileEstimatorPass::ID;
+
+ FunctionPass *createProfileEstimatorPass() {
+ return new ProfileEstimatorPass();
+ }
+
+ /// createProfileEstimatorPass - This function returns a Pass that estimates
+ /// profiling information using the given loop execution count.
+ Pass *createProfileEstimatorPass(const unsigned execcount) {
+ return new ProfileEstimatorPass(execcount);
+ }
+}
+
+static double ignoreMissing(double w) {
+ if (w == ProfileInfo::MissingValue) return 0;
+ return w;
+}
+
+static void inline printEdgeError(ProfileInfo::Edge e, const char *M) {
+ DEBUG(dbgs() << "-- Edge " << e << " is not calculated, " << M << "\n");
+}
+
+void inline ProfileEstimatorPass::printEdgeWeight(Edge E) {
+ DEBUG(dbgs() << "-- Weight of Edge " << E << ":"
+ << format("%20.20g", getEdgeWeight(E)) << "\n");
+}
+
+// recurseBasicBlock() - This calculates the ProfileInfo estimation for a
+// single block and then recurses into the successors.
+// The algorithm preserves the flow condition, meaning that the sum of the
+// weight of the incoming edges must be equal the block weight which must in
+// turn be equal to the sume of the weights of the outgoing edges.
+// Since the flow of an block is deterimined from the current state of the
+// flow, once an edge has a flow assigned this flow is never changed again,
+// otherwise it would be possible to violate the flow condition in another
+// block.
+void ProfileEstimatorPass::recurseBasicBlock(BasicBlock *BB) {
+
+ // Break the recursion if this BasicBlock was already visited.
+ if (BBToVisit.find(BB) == BBToVisit.end()) return;
+
+ // Read the LoopInfo for this block.
+ bool BBisHeader = LI->isLoopHeader(BB);
+ Loop* BBLoop = LI->getLoopFor(BB);
+
+ // To get the block weight, read all incoming edges.
+ double BBWeight = 0;
+ std::set<BasicBlock*> ProcessedPreds;
+ for ( pred_iterator bbi = pred_begin(BB), bbe = pred_end(BB);
+ bbi != bbe; ++bbi ) {
+ // If this block was not considered already, add weight.
+ Edge edge = getEdge(*bbi,BB);
+ double w = getEdgeWeight(edge);
+ if (ProcessedPreds.insert(*bbi).second) {
+ BBWeight += ignoreMissing(w);
+ }
+ // If this block is a loop header and the predecessor is contained in this
+ // loop, thus the edge is a backedge, continue and do not check if the
+ // value is valid.
+ if (BBisHeader && BBLoop->contains(*bbi)) {
+ printEdgeError(edge, "but is backedge, continuing");
+ continue;
+ }
+ // If the edges value is missing (and this is no loop header, and this is
+ // no backedge) return, this block is currently non estimatable.
+ if (w == MissingValue) {
+ printEdgeError(edge, "returning");
+ return;
+ }
+ }
+ if (getExecutionCount(BB) != MissingValue) {
+ BBWeight = getExecutionCount(BB);
+ }
+
+ // Fetch all necessary information for current block.
+ SmallVector<Edge, 8> ExitEdges;
+ SmallVector<Edge, 8> Edges;
+ if (BBLoop) {
+ BBLoop->getExitEdges(ExitEdges);
+ }
+
+ // If this is a loop header, consider the following:
+ // Exactly the flow that is entering this block, must exit this block too. So
+ // do the following:
+ // *) get all the exit edges, read the flow that is already leaving this
+ // loop, remember the edges that do not have any flow on them right now.
+ // (The edges that have already flow on them are most likely exiting edges of
+ // other loops, do not touch those flows because the previously caclulated
+ // loopheaders would not be exact anymore.)
+ // *) In case there is not a single exiting edge left, create one at the loop
+ // latch to prevent the flow from building up in the loop.
+ // *) Take the flow that is not leaving the loop already and distribute it on
+ // the remaining exiting edges.
+ // (This ensures that all flow that enters the loop also leaves it.)
+ // *) Increase the flow into the loop by increasing the weight of this block.
+ // There is at least one incoming backedge that will bring us this flow later
+ // on. (So that the flow condition in this node is valid again.)
+ if (BBisHeader) {
+ double incoming = BBWeight;
+ // Subtract the flow leaving the loop.
+ std::set<Edge> ProcessedExits;
+ for (SmallVector<Edge, 8>::iterator ei = ExitEdges.begin(),
+ ee = ExitEdges.end(); ei != ee; ++ei) {
+ if (ProcessedExits.insert(*ei).second) {
+ double w = getEdgeWeight(*ei);
+ if (w == MissingValue) {
+ Edges.push_back(*ei);
+ // Check if there is a necessary minimal weight, if yes, subtract it
+ // from weight.
+ if (MinimalWeight.find(*ei) != MinimalWeight.end()) {
+ incoming -= MinimalWeight[*ei];
+ DEBUG(dbgs() << "Reserving " << format("%.20g",MinimalWeight[*ei]) << " at " << (*ei) << "\n");
+ }
+ } else {
+ incoming -= w;
+ }
+ }
+ }
+ // If no exit edges, create one:
+ if (Edges.size() == 0) {
+ BasicBlock *Latch = BBLoop->getLoopLatch();
+ if (Latch) {
+ Edge edge = getEdge(Latch,0);
+ EdgeInformation[BB->getParent()][edge] = BBWeight;
+ printEdgeWeight(edge);
+ edge = getEdge(Latch, BB);
+ EdgeInformation[BB->getParent()][edge] = BBWeight * ExecCount;
+ printEdgeWeight(edge);
+ }
+ }
+
+ // Distribute remaining weight to the exting edges. To prevent fractions
+ // from building up and provoking precision problems the weight which is to
+ // be distributed is split and the rounded, the last edge gets a somewhat
+ // bigger value, but we are close enough for an estimation.
+ double fraction = floor(incoming/Edges.size());
+ for (SmallVector<Edge, 8>::iterator ei = Edges.begin(), ee = Edges.end();
+ ei != ee; ++ei) {
+ double w = 0;
+ if (ei != (ee-1)) {
+ w = fraction;
+ incoming -= fraction;
+ } else {
+ w = incoming;
+ }
+ EdgeInformation[BB->getParent()][*ei] += w;
+ // Read necessary minimal weight.
+ if (MinimalWeight.find(*ei) != MinimalWeight.end()) {
+ EdgeInformation[BB->getParent()][*ei] += MinimalWeight[*ei];
+ DEBUG(dbgs() << "Additionally " << format("%.20g",MinimalWeight[*ei]) << " at " << (*ei) << "\n");
+ }
+ printEdgeWeight(*ei);
+
+ // Add minimal weight to paths to all exit edges, this is used to ensure
+ // that enough flow is reaching this edges.
+ Path p;
+ const BasicBlock *Dest = GetPath(BB, (*ei).first, p, GetPathToDest);
+ while (Dest != BB) {
+ const BasicBlock *Parent = p.find(Dest)->second;
+ Edge e = getEdge(Parent, Dest);
+ if (MinimalWeight.find(e) == MinimalWeight.end()) {
+ MinimalWeight[e] = 0;
+ }
+ MinimalWeight[e] += w;
+ DEBUG(dbgs() << "Minimal Weight for " << e << ": " << format("%.20g",MinimalWeight[e]) << "\n");
+ Dest = Parent;
+ }
+ }
+ // Increase flow into the loop.
+ BBWeight *= (ExecCount+1);
+ }
+
+ BlockInformation[BB->getParent()][BB] = BBWeight;
+ // Up until now we considered only the loop exiting edges, now we have a
+ // definite block weight and must distribute this onto the outgoing edges.
+ // Since there may be already flow attached to some of the edges, read this
+ // flow first and remember the edges that have still now flow attached.
+ Edges.clear();
+ std::set<BasicBlock*> ProcessedSuccs;
+
+ succ_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
+ // Also check for (BB,0) edges that may already contain some flow. (But only
+ // in case there are no successors.)
+ if (bbi == bbe) {
+ Edge edge = getEdge(BB,0);
+ EdgeInformation[BB->getParent()][edge] = BBWeight;
+ printEdgeWeight(edge);
+ }
+ for ( ; bbi != bbe; ++bbi ) {
+ if (ProcessedSuccs.insert(*bbi).second) {
+ Edge edge = getEdge(BB,*bbi);
+ double w = getEdgeWeight(edge);
+ if (w != MissingValue) {
+ BBWeight -= getEdgeWeight(edge);
+ } else {
+ Edges.push_back(edge);
+ // If minimal weight is necessary, reserve weight by subtracting weight
+ // from block weight, this is readded later on.
+ if (MinimalWeight.find(edge) != MinimalWeight.end()) {
+ BBWeight -= MinimalWeight[edge];
+ DEBUG(dbgs() << "Reserving " << format("%.20g",MinimalWeight[edge]) << " at " << edge << "\n");
+ }
+ }
+ }
+ }
+
+ double fraction = floor(BBWeight/Edges.size());
+ // Finally we know what flow is still not leaving the block, distribute this
+ // flow onto the empty edges.
+ for (SmallVector<Edge, 8>::iterator ei = Edges.begin(), ee = Edges.end();
+ ei != ee; ++ei) {
+ if (ei != (ee-1)) {
+ EdgeInformation[BB->getParent()][*ei] += fraction;
+ BBWeight -= fraction;
+ } else {
+ EdgeInformation[BB->getParent()][*ei] += BBWeight;
+ }
+ // Readd minial necessary weight.
+ if (MinimalWeight.find(*ei) != MinimalWeight.end()) {
+ EdgeInformation[BB->getParent()][*ei] += MinimalWeight[*ei];
+ DEBUG(dbgs() << "Additionally " << format("%.20g",MinimalWeight[*ei]) << " at " << (*ei) << "\n");
+ }
+ printEdgeWeight(*ei);
+ }
+
+ // This block is visited, mark this before the recursion.
+ BBToVisit.erase(BB);
+
+ // Recurse into successors.
+ for (succ_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
+ bbi != bbe; ++bbi) {
+ recurseBasicBlock(*bbi);
+ }
+}
+
+bool ProfileEstimatorPass::runOnFunction(Function &F) {
+ if (F.isDeclaration()) return false;
+
+ // Fetch LoopInfo and clear ProfileInfo for this function.
+ LI = &getAnalysis<LoopInfo>();
+ FunctionInformation.erase(&F);
+ BlockInformation[&F].clear();
+ EdgeInformation[&F].clear();
+ BBToVisit.clear();
+
+ // Mark all blocks as to visit.
+ for (Function::iterator bi = F.begin(), be = F.end(); bi != be; ++bi)
+ BBToVisit.insert(bi);
+
+ // Clear Minimal Edges.
+ MinimalWeight.clear();
+
+ DEBUG(dbgs() << "Working on function " << F.getNameStr() << "\n");
+
+ // Since the entry block is the first one and has no predecessors, the edge
+ // (0,entry) is inserted with the starting weight of 1.
+ BasicBlock *entry = &F.getEntryBlock();
+ BlockInformation[&F][entry] = pow(2.0, 32.0);
+ Edge edge = getEdge(0,entry);
+ EdgeInformation[&F][edge] = BlockInformation[&F][entry];
+ printEdgeWeight(edge);
+
+ // Since recurseBasicBlock() maybe returns with a block which was not fully
+ // estimated, use recurseBasicBlock() until everything is calculated.
+ bool cleanup = false;
+ recurseBasicBlock(entry);
+ while (BBToVisit.size() > 0 && !cleanup) {
+ // Remember number of open blocks, this is later used to check if progress
+ // was made.
+ unsigned size = BBToVisit.size();
+
+ // Try to calculate all blocks in turn.
+ for (std::set<BasicBlock*>::iterator bi = BBToVisit.begin(),
+ be = BBToVisit.end(); bi != be; ++bi) {
+ recurseBasicBlock(*bi);
+ // If at least one block was finished, break because iterator may be
+ // invalid.
+ if (BBToVisit.size() < size) break;
+ }
+
+ // If there was not a single block resolved, make some assumptions.
+ if (BBToVisit.size() == size) {
+ bool found = false;
+ for (std::set<BasicBlock*>::iterator BBI = BBToVisit.begin(), BBE = BBToVisit.end();
+ (BBI != BBE) && (!found); ++BBI) {
+ BasicBlock *BB = *BBI;
+ // Try each predecessor if it can be assumend.
+ for (pred_iterator bbi = pred_begin(BB), bbe = pred_end(BB);
+ (bbi != bbe) && (!found); ++bbi) {
+ Edge e = getEdge(*bbi,BB);
+ double w = getEdgeWeight(e);
+ // Check that edge from predecessor is still free.
+ if (w == MissingValue) {
+ // Check if there is a circle from this block to predecessor.
+ Path P;
+ const BasicBlock *Dest = GetPath(BB, *bbi, P, GetPathToDest);
+ if (Dest != *bbi) {
+ // If there is no circle, just set edge weight to 0
+ EdgeInformation[&F][e] = 0;
+ DEBUG(dbgs() << "Assuming edge weight: ");
+ printEdgeWeight(e);
+ found = true;
+ }
+ }
+ }
+ }
+ if (!found) {
+ cleanup = true;
+ DEBUG(dbgs() << "No assumption possible in Fuction "<<F.getName()<<", setting all to zero\n");
+ }
+ }
+ }
+ // In case there was no safe way to assume edges, set as a last measure,
+ // set _everything_ to zero.
+ if (cleanup) {
+ FunctionInformation[&F] = 0;
+ BlockInformation[&F].clear();
+ EdgeInformation[&F].clear();
+ for (Function::const_iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) {
+ const BasicBlock *BB = &(*FI);
+ BlockInformation[&F][BB] = 0;
+ const_pred_iterator predi = pred_begin(BB), prede = pred_end(BB);
+ if (predi == prede) {
+ Edge e = getEdge(0,BB);
+ setEdgeWeight(e,0);
+ }
+ for (;predi != prede; ++predi) {
+ Edge e = getEdge(*predi,BB);
+ setEdgeWeight(e,0);
+ }
+ succ_const_iterator succi = succ_begin(BB), succe = succ_end(BB);
+ if (succi == succe) {
+ Edge e = getEdge(BB,0);
+ setEdgeWeight(e,0);
+ }
+ for (;succi != succe; ++succi) {
+ Edge e = getEdge(*succi,BB);
+ setEdgeWeight(e,0);
+ }
+ }
+ }
+
+ return false;
+}
diff --git a/src/LLVM/lib/Analysis/ProfileInfo.cpp b/src/LLVM/lib/Analysis/ProfileInfo.cpp
index fc7f286..173de2c 100644
--- a/src/LLVM/lib/Analysis/ProfileInfo.cpp
+++ b/src/LLVM/lib/Analysis/ProfileInfo.cpp
@@ -24,8 +24,12 @@
#include <limits>
using namespace llvm;
+namespace llvm {
+ template<> char ProfileInfoT<Function,BasicBlock>::ID = 0;
+}
+
// Register the ProfileInfo interface, providing a nice name to refer to.
-static RegisterAnalysisGroup<ProfileInfo> Z("Profile Information");
+INITIALIZE_ANALYSIS_GROUP(ProfileInfo, "Profile Information", NoProfileInfo)
namespace llvm {
@@ -44,9 +48,6 @@
}
template<>
-char ProfileInfoT<Function,BasicBlock>::ID = 0;
-
-template<>
char ProfileInfoT<MachineFunction, MachineBasicBlock>::ID = 0;
template<>
@@ -308,9 +309,9 @@
removeEdge(oldedge);
}
-/// Replaces all occurences of RmBB in the ProfilingInfo with DestBB.
+/// Replaces all occurrences of RmBB in the ProfilingInfo with DestBB.
/// This checks all edges of the function the blocks reside in and replaces the
-/// occurences of RmBB with DestBB.
+/// occurrences of RmBB with DestBB.
template<>
void ProfileInfoT<Function,BasicBlock>::
replaceAllUses(const BasicBlock *RmBB, const BasicBlock *DestBB) {
@@ -811,7 +812,7 @@
}
if (iw < 0) continue;
- // Check the recieving end of the path if it can handle the flow.
+ // Check the receiving end of the path if it can handle the flow.
double ow = getExecutionCount(Dest);
Processed.clear();
for (succ_const_iterator NBB = succ_begin(BB), End = succ_end(BB);
@@ -888,7 +889,7 @@
FI = Unvisited.begin(), FE = Unvisited.end();
while(FI != FE && !FoundPath) {
const BasicBlock *BB = *FI; ++FI;
- const BasicBlock *Dest;
+ const BasicBlock *Dest = 0;
Path P;
bool BackEdgeFound = false;
for (const_pred_iterator NBB = pred_begin(BB), End = pred_end(BB);
@@ -1076,7 +1077,9 @@
namespace {
struct NoProfileInfo : public ImmutablePass, public ProfileInfo {
static char ID; // Class identification, replacement for typeinfo
- NoProfileInfo() : ImmutablePass(ID) {}
+ NoProfileInfo() : ImmutablePass(ID) {
+ initializeNoProfileInfoPass(*PassRegistry::getPassRegistry());
+ }
/// getAdjustedAnalysisPointer - This method is used when a pass implements
/// an analysis interface through multiple inheritance. If needed, it
@@ -1097,6 +1100,6 @@
char NoProfileInfo::ID = 0;
// Register this pass...
INITIALIZE_AG_PASS(NoProfileInfo, ProfileInfo, "no-profile",
- "No Profile Information", false, true, true);
+ "No Profile Information", false, true, true)
ImmutablePass *llvm::createNoProfileInfoPass() { return new NoProfileInfo(); }
diff --git a/src/LLVM/lib/Analysis/ProfileInfoLoader.cpp b/src/LLVM/lib/Analysis/ProfileInfoLoader.cpp
new file mode 100644
index 0000000..eaa38da
--- /dev/null
+++ b/src/LLVM/lib/Analysis/ProfileInfoLoader.cpp
@@ -0,0 +1,157 @@
+//===- ProfileInfoLoad.cpp - Load profile information from disk -----------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// The ProfileInfoLoader class is used to load and represent profiling
+// information read in from the dump file.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/ProfileInfoLoader.h"
+#include "llvm/Analysis/ProfileInfoTypes.h"
+#include "llvm/Module.h"
+#include "llvm/InstrTypes.h"
+#include "llvm/Support/raw_ostream.h"
+#include <cstdio>
+#include <cstdlib>
+using namespace llvm;
+
+// ByteSwap - Byteswap 'Var' if 'Really' is true.
+//
+static inline unsigned ByteSwap(unsigned Var, bool Really) {
+ if (!Really) return Var;
+ return ((Var & (255U<< 0U)) << 24U) |
+ ((Var & (255U<< 8U)) << 8U) |
+ ((Var & (255U<<16U)) >> 8U) |
+ ((Var & (255U<<24U)) >> 24U);
+}
+
+static unsigned AddCounts(unsigned A, unsigned B) {
+ // If either value is undefined, use the other.
+ if (A == ProfileInfoLoader::Uncounted) return B;
+ if (B == ProfileInfoLoader::Uncounted) return A;
+ return A + B;
+}
+
+static void ReadProfilingBlock(const char *ToolName, FILE *F,
+ bool ShouldByteSwap,
+ std::vector<unsigned> &Data) {
+ // Read the number of entries...
+ unsigned NumEntries;
+ if (fread(&NumEntries, sizeof(unsigned), 1, F) != 1) {
+ errs() << ToolName << ": data packet truncated!\n";
+ perror(0);
+ exit(1);
+ }
+ NumEntries = ByteSwap(NumEntries, ShouldByteSwap);
+
+ // Read the counts...
+ std::vector<unsigned> TempSpace(NumEntries);
+
+ // Read in the block of data...
+ if (fread(&TempSpace[0], sizeof(unsigned)*NumEntries, 1, F) != 1) {
+ errs() << ToolName << ": data packet truncated!\n";
+ perror(0);
+ exit(1);
+ }
+
+ // Make sure we have enough space... The space is initialised to -1 to
+ // facitiltate the loading of missing values for OptimalEdgeProfiling.
+ if (Data.size() < NumEntries)
+ Data.resize(NumEntries, ProfileInfoLoader::Uncounted);
+
+ // Accumulate the data we just read into the data.
+ if (!ShouldByteSwap) {
+ for (unsigned i = 0; i != NumEntries; ++i) {
+ Data[i] = AddCounts(TempSpace[i], Data[i]);
+ }
+ } else {
+ for (unsigned i = 0; i != NumEntries; ++i) {
+ Data[i] = AddCounts(ByteSwap(TempSpace[i], true), Data[i]);
+ }
+ }
+}
+
+const unsigned ProfileInfoLoader::Uncounted = ~0U;
+
+// ProfileInfoLoader ctor - Read the specified profiling data file, exiting the
+// program if the file is invalid or broken.
+//
+ProfileInfoLoader::ProfileInfoLoader(const char *ToolName,
+ const std::string &Filename,
+ Module &TheModule) :
+ Filename(Filename),
+ M(TheModule), Warned(false) {
+ FILE *F = fopen(Filename.c_str(), "rb");
+ if (F == 0) {
+ errs() << ToolName << ": Error opening '" << Filename << "': ";
+ perror(0);
+ exit(1);
+ }
+
+ // Keep reading packets until we run out of them.
+ unsigned PacketType;
+ while (fread(&PacketType, sizeof(unsigned), 1, F) == 1) {
+ // If the low eight bits of the packet are zero, we must be dealing with an
+ // endianness mismatch. Byteswap all words read from the profiling
+ // information.
+ bool ShouldByteSwap = (char)PacketType == 0;
+ PacketType = ByteSwap(PacketType, ShouldByteSwap);
+
+ switch (PacketType) {
+ case ArgumentInfo: {
+ unsigned ArgLength;
+ if (fread(&ArgLength, sizeof(unsigned), 1, F) != 1) {
+ errs() << ToolName << ": arguments packet truncated!\n";
+ perror(0);
+ exit(1);
+ }
+ ArgLength = ByteSwap(ArgLength, ShouldByteSwap);
+
+ // Read in the arguments...
+ std::vector<char> Chars(ArgLength+4);
+
+ if (ArgLength)
+ if (fread(&Chars[0], (ArgLength+3) & ~3, 1, F) != 1) {
+ errs() << ToolName << ": arguments packet truncated!\n";
+ perror(0);
+ exit(1);
+ }
+ CommandLines.push_back(std::string(&Chars[0], &Chars[ArgLength]));
+ break;
+ }
+
+ case FunctionInfo:
+ ReadProfilingBlock(ToolName, F, ShouldByteSwap, FunctionCounts);
+ break;
+
+ case BlockInfo:
+ ReadProfilingBlock(ToolName, F, ShouldByteSwap, BlockCounts);
+ break;
+
+ case EdgeInfo:
+ ReadProfilingBlock(ToolName, F, ShouldByteSwap, EdgeCounts);
+ break;
+
+ case OptEdgeInfo:
+ ReadProfilingBlock(ToolName, F, ShouldByteSwap, OptimalEdgeCounts);
+ break;
+
+ case BBTraceInfo:
+ ReadProfilingBlock(ToolName, F, ShouldByteSwap, BBTrace);
+ break;
+
+ default:
+ errs() << ToolName << ": Unknown packet type #" << PacketType << "!\n";
+ exit(1);
+ }
+ }
+
+ fclose(F);
+}
+
diff --git a/src/LLVM/lib/Analysis/ProfileInfoLoaderPass.cpp b/src/LLVM/lib/Analysis/ProfileInfoLoaderPass.cpp
new file mode 100644
index 0000000..098079b
--- /dev/null
+++ b/src/LLVM/lib/Analysis/ProfileInfoLoaderPass.cpp
@@ -0,0 +1,267 @@
+//===- ProfileInfoLoaderPass.cpp - LLVM Pass to load profile info ---------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a concrete implementation of profiling information that
+// loads the information from a profile dump file.
+//
+//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "profile-loader"
+#include "llvm/BasicBlock.h"
+#include "llvm/InstrTypes.h"
+#include "llvm/Module.h"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/ProfileInfo.h"
+#include "llvm/Analysis/ProfileInfoLoader.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/Format.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/SmallSet.h"
+#include <set>
+using namespace llvm;
+
+STATISTIC(NumEdgesRead, "The # of edges read.");
+
+static cl::opt<std::string>
+ProfileInfoFilename("profile-info-file", cl::init("llvmprof.out"),
+ cl::value_desc("filename"),
+ cl::desc("Profile file loaded by -profile-loader"));
+
+namespace {
+ class LoaderPass : public ModulePass, public ProfileInfo {
+ std::string Filename;
+ std::set<Edge> SpanningTree;
+ std::set<const BasicBlock*> BBisUnvisited;
+ unsigned ReadCount;
+ public:
+ static char ID; // Class identification, replacement for typeinfo
+ explicit LoaderPass(const std::string &filename = "")
+ : ModulePass(ID), Filename(filename) {
+ initializeLoaderPassPass(*PassRegistry::getPassRegistry());
+ if (filename.empty()) Filename = ProfileInfoFilename;
+ }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+
+ virtual const char *getPassName() const {
+ return "Profiling information loader";
+ }
+
+ // recurseBasicBlock() - Calculates the edge weights for as much basic
+ // blocks as possbile.
+ virtual void recurseBasicBlock(const BasicBlock *BB);
+ virtual void readEdgeOrRemember(Edge, Edge&, unsigned &, double &);
+ virtual void readEdge(ProfileInfo::Edge, std::vector<unsigned>&);
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ virtual void *getAdjustedAnalysisPointer(AnalysisID PI) {
+ if (PI == &ProfileInfo::ID)
+ return (ProfileInfo*)this;
+ return this;
+ }
+
+ /// run - Load the profile information from the specified file.
+ virtual bool runOnModule(Module &M);
+ };
+} // End of anonymous namespace
+
+char LoaderPass::ID = 0;
+INITIALIZE_AG_PASS(LoaderPass, ProfileInfo, "profile-loader",
+ "Load profile information from llvmprof.out", false, true, false)
+
+char &llvm::ProfileLoaderPassID = LoaderPass::ID;
+
+ModulePass *llvm::createProfileLoaderPass() { return new LoaderPass(); }
+
+/// createProfileLoaderPass - This function returns a Pass that loads the
+/// profiling information for the module from the specified filename, making it
+/// available to the optimizers.
+Pass *llvm::createProfileLoaderPass(const std::string &Filename) {
+ return new LoaderPass(Filename);
+}
+
+void LoaderPass::readEdgeOrRemember(Edge edge, Edge &tocalc,
+ unsigned &uncalc, double &count) {
+ double w;
+ if ((w = getEdgeWeight(edge)) == MissingValue) {
+ tocalc = edge;
+ uncalc++;
+ } else {
+ count+=w;
+ }
+}
+
+// recurseBasicBlock - Visits all neighbours of a block and then tries to
+// calculate the missing edge values.
+void LoaderPass::recurseBasicBlock(const BasicBlock *BB) {
+
+ // break recursion if already visited
+ if (BBisUnvisited.find(BB) == BBisUnvisited.end()) return;
+ BBisUnvisited.erase(BB);
+ if (!BB) return;
+
+ for (succ_const_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
+ bbi != bbe; ++bbi) {
+ recurseBasicBlock(*bbi);
+ }
+ for (const_pred_iterator bbi = pred_begin(BB), bbe = pred_end(BB);
+ bbi != bbe; ++bbi) {
+ recurseBasicBlock(*bbi);
+ }
+
+ Edge tocalc;
+ if (CalculateMissingEdge(BB, tocalc)) {
+ SpanningTree.erase(tocalc);
+ }
+}
+
+void LoaderPass::readEdge(ProfileInfo::Edge e,
+ std::vector<unsigned> &ECs) {
+ if (ReadCount < ECs.size()) {
+ double weight = ECs[ReadCount++];
+ if (weight != ProfileInfoLoader::Uncounted) {
+ // Here the data realm changes from the unsigned of the file to the
+ // double of the ProfileInfo. This conversion is save because we know
+ // that everything thats representable in unsinged is also representable
+ // in double.
+ EdgeInformation[getFunction(e)][e] += (double)weight;
+
+ DEBUG(dbgs() << "--Read Edge Counter for " << e
+ << " (# "<< (ReadCount-1) << "): "
+ << (unsigned)getEdgeWeight(e) << "\n");
+ } else {
+ // This happens only if reading optimal profiling information, not when
+ // reading regular profiling information.
+ SpanningTree.insert(e);
+ }
+ }
+}
+
+bool LoaderPass::runOnModule(Module &M) {
+ ProfileInfoLoader PIL("profile-loader", Filename, M);
+
+ EdgeInformation.clear();
+ std::vector<unsigned> Counters = PIL.getRawEdgeCounts();
+ if (Counters.size() > 0) {
+ ReadCount = 0;
+ for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
+ if (F->isDeclaration()) continue;
+ DEBUG(dbgs()<<"Working on "<<F->getNameStr()<<"\n");
+ readEdge(getEdge(0,&F->getEntryBlock()), Counters);
+ for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
+ TerminatorInst *TI = BB->getTerminator();
+ for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
+ readEdge(getEdge(BB,TI->getSuccessor(s)), Counters);
+ }
+ }
+ }
+ if (ReadCount != Counters.size()) {
+ errs() << "WARNING: profile information is inconsistent with "
+ << "the current program!\n";
+ }
+ NumEdgesRead = ReadCount;
+ }
+
+ Counters = PIL.getRawOptimalEdgeCounts();
+ if (Counters.size() > 0) {
+ ReadCount = 0;
+ for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
+ if (F->isDeclaration()) continue;
+ DEBUG(dbgs()<<"Working on "<<F->getNameStr()<<"\n");
+ readEdge(getEdge(0,&F->getEntryBlock()), Counters);
+ for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
+ TerminatorInst *TI = BB->getTerminator();
+ if (TI->getNumSuccessors() == 0) {
+ readEdge(getEdge(BB,0), Counters);
+ }
+ for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
+ readEdge(getEdge(BB,TI->getSuccessor(s)), Counters);
+ }
+ }
+ while (SpanningTree.size() > 0) {
+
+ unsigned size = SpanningTree.size();
+
+ BBisUnvisited.clear();
+ for (std::set<Edge>::iterator ei = SpanningTree.begin(),
+ ee = SpanningTree.end(); ei != ee; ++ei) {
+ BBisUnvisited.insert(ei->first);
+ BBisUnvisited.insert(ei->second);
+ }
+ while (BBisUnvisited.size() > 0) {
+ recurseBasicBlock(*BBisUnvisited.begin());
+ }
+
+ if (SpanningTree.size() == size) {
+ DEBUG(dbgs()<<"{");
+ for (std::set<Edge>::iterator ei = SpanningTree.begin(),
+ ee = SpanningTree.end(); ei != ee; ++ei) {
+ DEBUG(dbgs()<< *ei <<",");
+ }
+ assert(0 && "No edge calculated!");
+ }
+
+ }
+ }
+ if (ReadCount != Counters.size()) {
+ errs() << "WARNING: profile information is inconsistent with "
+ << "the current program!\n";
+ }
+ NumEdgesRead = ReadCount;
+ }
+
+ BlockInformation.clear();
+ Counters = PIL.getRawBlockCounts();
+ if (Counters.size() > 0) {
+ ReadCount = 0;
+ for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
+ if (F->isDeclaration()) continue;
+ for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
+ if (ReadCount < Counters.size())
+ // Here the data realm changes from the unsigned of the file to the
+ // double of the ProfileInfo. This conversion is save because we know
+ // that everything thats representable in unsinged is also
+ // representable in double.
+ BlockInformation[F][BB] = (double)Counters[ReadCount++];
+ }
+ if (ReadCount != Counters.size()) {
+ errs() << "WARNING: profile information is inconsistent with "
+ << "the current program!\n";
+ }
+ }
+
+ FunctionInformation.clear();
+ Counters = PIL.getRawFunctionCounts();
+ if (Counters.size() > 0) {
+ ReadCount = 0;
+ for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
+ if (F->isDeclaration()) continue;
+ if (ReadCount < Counters.size())
+ // Here the data realm changes from the unsigned of the file to the
+ // double of the ProfileInfo. This conversion is save because we know
+ // that everything thats representable in unsinged is also
+ // representable in double.
+ FunctionInformation[F] = (double)Counters[ReadCount++];
+ }
+ if (ReadCount != Counters.size()) {
+ errs() << "WARNING: profile information is inconsistent with "
+ << "the current program!\n";
+ }
+ }
+
+ return false;
+}
diff --git a/src/LLVM/lib/Analysis/ProfileVerifierPass.cpp b/src/LLVM/lib/Analysis/ProfileVerifierPass.cpp
new file mode 100644
index 0000000..a017518
--- /dev/null
+++ b/src/LLVM/lib/Analysis/ProfileVerifierPass.cpp
@@ -0,0 +1,382 @@
+//===- ProfileVerifierPass.cpp - LLVM Pass to estimate profile info -------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a pass that checks profiling information for
+// plausibility.
+//
+//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "profile-verifier"
+#include "llvm/Instructions.h"
+#include "llvm/Module.h"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/ProfileInfo.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/CallSite.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/InstIterator.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/Debug.h"
+#include <set>
+using namespace llvm;
+
+static cl::opt<bool,false>
+ProfileVerifierDisableAssertions("profile-verifier-noassert",
+ cl::desc("Disable assertions"));
+
+namespace llvm {
+ template<class FType, class BType>
+ class ProfileVerifierPassT : public FunctionPass {
+
+ struct DetailedBlockInfo {
+ const BType *BB;
+ double BBWeight;
+ double inWeight;
+ int inCount;
+ double outWeight;
+ int outCount;
+ };
+
+ ProfileInfoT<FType, BType> *PI;
+ std::set<const BType*> BBisVisited;
+ std::set<const FType*> FisVisited;
+ bool DisableAssertions;
+
+ // When debugging is enabled, the verifier prints a whole slew of debug
+ // information, otherwise its just the assert. These are all the helper
+ // functions.
+ bool PrintedDebugTree;
+ std::set<const BType*> BBisPrinted;
+ void debugEntry(DetailedBlockInfo*);
+ void printDebugInfo(const BType *BB);
+
+ public:
+ static char ID; // Class identification, replacement for typeinfo
+
+ explicit ProfileVerifierPassT () : FunctionPass(ID) {
+ initializeProfileVerifierPassPass(*PassRegistry::getPassRegistry());
+ DisableAssertions = ProfileVerifierDisableAssertions;
+ }
+ explicit ProfileVerifierPassT (bool da) : FunctionPass(ID),
+ DisableAssertions(da) {
+ initializeProfileVerifierPassPass(*PassRegistry::getPassRegistry());
+ }
+
+ void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ AU.addRequired<ProfileInfoT<FType, BType> >();
+ }
+
+ const char *getPassName() const {
+ return "Profiling information verifier";
+ }
+
+ /// run - Verify the profile information.
+ bool runOnFunction(FType &F);
+ void recurseBasicBlock(const BType*);
+
+ bool exitReachable(const FType*);
+ double ReadOrAssert(typename ProfileInfoT<FType, BType>::Edge);
+ void CheckValue(bool, const char*, DetailedBlockInfo*);
+ };
+
+ typedef ProfileVerifierPassT<Function, BasicBlock> ProfileVerifierPass;
+
+ template<class FType, class BType>
+ void ProfileVerifierPassT<FType, BType>::printDebugInfo(const BType *BB) {
+
+ if (BBisPrinted.find(BB) != BBisPrinted.end()) return;
+
+ double BBWeight = PI->getExecutionCount(BB);
+ if (BBWeight == ProfileInfoT<FType, BType>::MissingValue) { BBWeight = 0; }
+ double inWeight = 0;
+ int inCount = 0;
+ std::set<const BType*> ProcessedPreds;
+ for (const_pred_iterator bbi = pred_begin(BB), bbe = pred_end(BB);
+ bbi != bbe; ++bbi ) {
+ if (ProcessedPreds.insert(*bbi).second) {
+ typename ProfileInfoT<FType, BType>::Edge E = PI->getEdge(*bbi,BB);
+ double EdgeWeight = PI->getEdgeWeight(E);
+ if (EdgeWeight == ProfileInfoT<FType, BType>::MissingValue) { EdgeWeight = 0; }
+ dbgs() << "calculated in-edge " << E << ": "
+ << format("%20.20g",EdgeWeight) << "\n";
+ inWeight += EdgeWeight;
+ inCount++;
+ }
+ }
+ double outWeight = 0;
+ int outCount = 0;
+ std::set<const BType*> ProcessedSuccs;
+ for ( succ_const_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
+ bbi != bbe; ++bbi ) {
+ if (ProcessedSuccs.insert(*bbi).second) {
+ typename ProfileInfoT<FType, BType>::Edge E = PI->getEdge(BB,*bbi);
+ double EdgeWeight = PI->getEdgeWeight(E);
+ if (EdgeWeight == ProfileInfoT<FType, BType>::MissingValue) { EdgeWeight = 0; }
+ dbgs() << "calculated out-edge " << E << ": "
+ << format("%20.20g",EdgeWeight) << "\n";
+ outWeight += EdgeWeight;
+ outCount++;
+ }
+ }
+ dbgs() << "Block " << BB->getNameStr() << " in "
+ << BB->getParent()->getNameStr() << ":"
+ << "BBWeight=" << format("%20.20g",BBWeight) << ","
+ << "inWeight=" << format("%20.20g",inWeight) << ","
+ << "inCount=" << inCount << ","
+ << "outWeight=" << format("%20.20g",outWeight) << ","
+ << "outCount" << outCount << "\n";
+
+ // mark as visited and recurse into subnodes
+ BBisPrinted.insert(BB);
+ for ( succ_const_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
+ bbi != bbe; ++bbi ) {
+ printDebugInfo(*bbi);
+ }
+ }
+
+ template<class FType, class BType>
+ void ProfileVerifierPassT<FType, BType>::debugEntry (DetailedBlockInfo *DI) {
+ dbgs() << "TROUBLE: Block " << DI->BB->getNameStr() << " in "
+ << DI->BB->getParent()->getNameStr() << ":"
+ << "BBWeight=" << format("%20.20g",DI->BBWeight) << ","
+ << "inWeight=" << format("%20.20g",DI->inWeight) << ","
+ << "inCount=" << DI->inCount << ","
+ << "outWeight=" << format("%20.20g",DI->outWeight) << ","
+ << "outCount=" << DI->outCount << "\n";
+ if (!PrintedDebugTree) {
+ PrintedDebugTree = true;
+ printDebugInfo(&(DI->BB->getParent()->getEntryBlock()));
+ }
+ }
+
+ // This compares A and B for equality.
+ static bool Equals(double A, double B) {
+ return A == B;
+ }
+
+ // This checks if the function "exit" is reachable from an given function
+ // via calls, this is necessary to check if a profile is valid despite the
+ // counts not fitting exactly.
+ template<class FType, class BType>
+ bool ProfileVerifierPassT<FType, BType>::exitReachable(const FType *F) {
+ if (!F) return false;
+
+ if (FisVisited.count(F)) return false;
+
+ FType *Exit = F->getParent()->getFunction("exit");
+ if (Exit == F) {
+ return true;
+ }
+
+ FisVisited.insert(F);
+ bool exits = false;
+ for (const_inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) {
+ if (const CallInst *CI = dyn_cast<CallInst>(&*I)) {
+ FType *F = CI->getCalledFunction();
+ if (F) {
+ exits |= exitReachable(F);
+ } else {
+ // This is a call to a pointer, all bets are off...
+ exits = true;
+ }
+ if (exits) break;
+ }
+ }
+ return exits;
+ }
+
+ #define ASSERTMESSAGE(M) \
+ { dbgs() << "ASSERT:" << (M) << "\n"; \
+ if (!DisableAssertions) assert(0 && (M)); }
+
+ template<class FType, class BType>
+ double ProfileVerifierPassT<FType, BType>::ReadOrAssert(typename ProfileInfoT<FType, BType>::Edge E) {
+ double EdgeWeight = PI->getEdgeWeight(E);
+ if (EdgeWeight == ProfileInfoT<FType, BType>::MissingValue) {
+ dbgs() << "Edge " << E << " in Function "
+ << ProfileInfoT<FType, BType>::getFunction(E)->getNameStr() << ": ";
+ ASSERTMESSAGE("Edge has missing value");
+ return 0;
+ } else {
+ if (EdgeWeight < 0) {
+ dbgs() << "Edge " << E << " in Function "
+ << ProfileInfoT<FType, BType>::getFunction(E)->getNameStr() << ": ";
+ ASSERTMESSAGE("Edge has negative value");
+ }
+ return EdgeWeight;
+ }
+ }
+
+ template<class FType, class BType>
+ void ProfileVerifierPassT<FType, BType>::CheckValue(bool Error,
+ const char *Message,
+ DetailedBlockInfo *DI) {
+ if (Error) {
+ DEBUG(debugEntry(DI));
+ dbgs() << "Block " << DI->BB->getNameStr() << " in Function "
+ << DI->BB->getParent()->getNameStr() << ": ";
+ ASSERTMESSAGE(Message);
+ }
+ return;
+ }
+
+ // This calculates the Information for a block and then recurses into the
+ // successors.
+ template<class FType, class BType>
+ void ProfileVerifierPassT<FType, BType>::recurseBasicBlock(const BType *BB) {
+
+ // Break the recursion by remembering all visited blocks.
+ if (BBisVisited.find(BB) != BBisVisited.end()) return;
+
+ // Use a data structure to store all the information, this can then be handed
+ // to debug printers.
+ DetailedBlockInfo DI;
+ DI.BB = BB;
+ DI.outCount = DI.inCount = 0;
+ DI.inWeight = DI.outWeight = 0;
+
+ // Read predecessors.
+ std::set<const BType*> ProcessedPreds;
+ const_pred_iterator bpi = pred_begin(BB), bpe = pred_end(BB);
+ // If there are none, check for (0,BB) edge.
+ if (bpi == bpe) {
+ DI.inWeight += ReadOrAssert(PI->getEdge(0,BB));
+ DI.inCount++;
+ }
+ for (;bpi != bpe; ++bpi) {
+ if (ProcessedPreds.insert(*bpi).second) {
+ DI.inWeight += ReadOrAssert(PI->getEdge(*bpi,BB));
+ DI.inCount++;
+ }
+ }
+
+ // Read successors.
+ std::set<const BType*> ProcessedSuccs;
+ succ_const_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
+ // If there is an (0,BB) edge, consider it too. (This is done not only when
+ // there are no successors, but every time; not every function contains
+ // return blocks with no successors (think loop latch as return block)).
+ double w = PI->getEdgeWeight(PI->getEdge(BB,0));
+ if (w != ProfileInfoT<FType, BType>::MissingValue) {
+ DI.outWeight += w;
+ DI.outCount++;
+ }
+ for (;bbi != bbe; ++bbi) {
+ if (ProcessedSuccs.insert(*bbi).second) {
+ DI.outWeight += ReadOrAssert(PI->getEdge(BB,*bbi));
+ DI.outCount++;
+ }
+ }
+
+ // Read block weight.
+ DI.BBWeight = PI->getExecutionCount(BB);
+ CheckValue(DI.BBWeight == ProfileInfoT<FType, BType>::MissingValue,
+ "BasicBlock has missing value", &DI);
+ CheckValue(DI.BBWeight < 0,
+ "BasicBlock has negative value", &DI);
+
+ // Check if this block is a setjmp target.
+ bool isSetJmpTarget = false;
+ if (DI.outWeight > DI.inWeight) {
+ for (typename BType::const_iterator i = BB->begin(), ie = BB->end();
+ i != ie; ++i) {
+ if (const CallInst *CI = dyn_cast<CallInst>(&*i)) {
+ FType *F = CI->getCalledFunction();
+ if (F && (F->getName() == "_setjmp")) {
+ isSetJmpTarget = true; break;
+ }
+ }
+ }
+ }
+ // Check if this block is eventually reaching exit.
+ bool isExitReachable = false;
+ if (DI.inWeight > DI.outWeight) {
+ for (typename BType::const_iterator i = BB->begin(), ie = BB->end();
+ i != ie; ++i) {
+ if (const CallInst *CI = dyn_cast<CallInst>(&*i)) {
+ FType *F = CI->getCalledFunction();
+ if (F) {
+ FisVisited.clear();
+ isExitReachable |= exitReachable(F);
+ } else {
+ // This is a call to a pointer, all bets are off...
+ isExitReachable = true;
+ }
+ if (isExitReachable) break;
+ }
+ }
+ }
+
+ if (DI.inCount > 0 && DI.outCount == 0) {
+ // If this is a block with no successors.
+ if (!isSetJmpTarget) {
+ CheckValue(!Equals(DI.inWeight,DI.BBWeight),
+ "inWeight and BBWeight do not match", &DI);
+ }
+ } else if (DI.inCount == 0 && DI.outCount > 0) {
+ // If this is a block with no predecessors.
+ if (!isExitReachable)
+ CheckValue(!Equals(DI.BBWeight,DI.outWeight),
+ "BBWeight and outWeight do not match", &DI);
+ } else {
+ // If this block has successors and predecessors.
+ if (DI.inWeight > DI.outWeight && !isExitReachable)
+ CheckValue(!Equals(DI.inWeight,DI.outWeight),
+ "inWeight and outWeight do not match", &DI);
+ if (DI.inWeight < DI.outWeight && !isSetJmpTarget)
+ CheckValue(!Equals(DI.inWeight,DI.outWeight),
+ "inWeight and outWeight do not match", &DI);
+ }
+
+
+ // Mark this block as visited, rescurse into successors.
+ BBisVisited.insert(BB);
+ for ( succ_const_iterator bbi = succ_begin(BB), bbe = succ_end(BB);
+ bbi != bbe; ++bbi ) {
+ recurseBasicBlock(*bbi);
+ }
+ }
+
+ template<class FType, class BType>
+ bool ProfileVerifierPassT<FType, BType>::runOnFunction(FType &F) {
+ PI = getAnalysisIfAvailable<ProfileInfoT<FType, BType> >();
+ if (!PI)
+ ASSERTMESSAGE("No ProfileInfo available");
+
+ // Prepare global variables.
+ PrintedDebugTree = false;
+ BBisVisited.clear();
+
+ // Fetch entry block and recurse into it.
+ const BType *entry = &F.getEntryBlock();
+ recurseBasicBlock(entry);
+
+ if (PI->getExecutionCount(&F) != PI->getExecutionCount(entry))
+ ASSERTMESSAGE("Function count and entry block count do not match");
+
+ return false;
+ }
+
+ template<class FType, class BType>
+ char ProfileVerifierPassT<FType, BType>::ID = 0;
+}
+
+INITIALIZE_PASS_BEGIN(ProfileVerifierPass, "profile-verifier",
+ "Verify profiling information", false, true)
+INITIALIZE_AG_DEPENDENCY(ProfileInfo)
+INITIALIZE_PASS_END(ProfileVerifierPass, "profile-verifier",
+ "Verify profiling information", false, true)
+
+namespace llvm {
+ FunctionPass *createProfileVerifierPass() {
+ return new ProfileVerifierPass(ProfileVerifierDisableAssertions);
+ }
+}
+
diff --git a/src/LLVM/lib/Analysis/RegionInfo.cpp b/src/LLVM/lib/Analysis/RegionInfo.cpp
new file mode 100644
index 0000000..52753cb
--- /dev/null
+++ b/src/LLVM/lib/Analysis/RegionInfo.cpp
@@ -0,0 +1,851 @@
+//===- RegionInfo.cpp - SESE region detection analysis --------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+// Detects single entry single exit regions in the control flow graph.
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/RegionInfo.h"
+#include "llvm/Analysis/RegionIterator.h"
+
+#include "llvm/ADT/PostOrderIterator.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Assembly/Writer.h"
+
+#define DEBUG_TYPE "region"
+#include "llvm/Support/Debug.h"
+
+#include <set>
+#include <algorithm>
+
+using namespace llvm;
+
+// Always verify if expensive checking is enabled.
+#ifdef XDEBUG
+static bool VerifyRegionInfo = true;
+#else
+static bool VerifyRegionInfo = false;
+#endif
+
+static cl::opt<bool,true>
+VerifyRegionInfoX("verify-region-info", cl::location(VerifyRegionInfo),
+ cl::desc("Verify region info (time consuming)"));
+
+STATISTIC(numRegions, "The # of regions");
+STATISTIC(numSimpleRegions, "The # of simple regions");
+
+static cl::opt<enum Region::PrintStyle> printStyle("print-region-style",
+ cl::Hidden,
+ cl::desc("style of printing regions"),
+ cl::values(
+ clEnumValN(Region::PrintNone, "none", "print no details"),
+ clEnumValN(Region::PrintBB, "bb",
+ "print regions in detail with block_iterator"),
+ clEnumValN(Region::PrintRN, "rn",
+ "print regions in detail with element_iterator"),
+ clEnumValEnd));
+//===----------------------------------------------------------------------===//
+/// Region Implementation
+Region::Region(BasicBlock *Entry, BasicBlock *Exit, RegionInfo* RInfo,
+ DominatorTree *dt, Region *Parent)
+ : RegionNode(Parent, Entry, 1), RI(RInfo), DT(dt), exit(Exit) {}
+
+Region::~Region() {
+ // Free the cached nodes.
+ for (BBNodeMapT::iterator it = BBNodeMap.begin(),
+ ie = BBNodeMap.end(); it != ie; ++it)
+ delete it->second;
+
+ // Only clean the cache for this Region. Caches of child Regions will be
+ // cleaned when the child Regions are deleted.
+ BBNodeMap.clear();
+
+ for (iterator I = begin(), E = end(); I != E; ++I)
+ delete *I;
+}
+
+void Region::replaceEntry(BasicBlock *BB) {
+ entry.setPointer(BB);
+}
+
+void Region::replaceExit(BasicBlock *BB) {
+ assert(exit && "No exit to replace!");
+ exit = BB;
+}
+
+bool Region::contains(const BasicBlock *B) const {
+ BasicBlock *BB = const_cast<BasicBlock*>(B);
+
+ assert(DT->getNode(BB) && "BB not part of the dominance tree");
+
+ BasicBlock *entry = getEntry(), *exit = getExit();
+
+ // Toplevel region.
+ if (!exit)
+ return true;
+
+ return (DT->dominates(entry, BB)
+ && !(DT->dominates(exit, BB) && DT->dominates(entry, exit)));
+}
+
+bool Region::contains(const Loop *L) const {
+ // BBs that are not part of any loop are element of the Loop
+ // described by the NULL pointer. This loop is not part of any region,
+ // except if the region describes the whole function.
+ if (L == 0)
+ return getExit() == 0;
+
+ if (!contains(L->getHeader()))
+ return false;
+
+ SmallVector<BasicBlock *, 8> ExitingBlocks;
+ L->getExitingBlocks(ExitingBlocks);
+
+ for (SmallVectorImpl<BasicBlock*>::iterator BI = ExitingBlocks.begin(),
+ BE = ExitingBlocks.end(); BI != BE; ++BI)
+ if (!contains(*BI))
+ return false;
+
+ return true;
+}
+
+Loop *Region::outermostLoopInRegion(Loop *L) const {
+ if (!contains(L))
+ return 0;
+
+ while (L && contains(L->getParentLoop())) {
+ L = L->getParentLoop();
+ }
+
+ return L;
+}
+
+Loop *Region::outermostLoopInRegion(LoopInfo *LI, BasicBlock* BB) const {
+ assert(LI && BB && "LI and BB cannot be null!");
+ Loop *L = LI->getLoopFor(BB);
+ return outermostLoopInRegion(L);
+}
+
+BasicBlock *Region::getEnteringBlock() const {
+ BasicBlock *entry = getEntry();
+ BasicBlock *Pred;
+ BasicBlock *enteringBlock = 0;
+
+ for (pred_iterator PI = pred_begin(entry), PE = pred_end(entry); PI != PE;
+ ++PI) {
+ Pred = *PI;
+ if (DT->getNode(Pred) && !contains(Pred)) {
+ if (enteringBlock)
+ return 0;
+
+ enteringBlock = Pred;
+ }
+ }
+
+ return enteringBlock;
+}
+
+BasicBlock *Region::getExitingBlock() const {
+ BasicBlock *exit = getExit();
+ BasicBlock *Pred;
+ BasicBlock *exitingBlock = 0;
+
+ if (!exit)
+ return 0;
+
+ for (pred_iterator PI = pred_begin(exit), PE = pred_end(exit); PI != PE;
+ ++PI) {
+ Pred = *PI;
+ if (contains(Pred)) {
+ if (exitingBlock)
+ return 0;
+
+ exitingBlock = Pred;
+ }
+ }
+
+ return exitingBlock;
+}
+
+bool Region::isSimple() const {
+ return !isTopLevelRegion() && getEnteringBlock() && getExitingBlock();
+}
+
+std::string Region::getNameStr() const {
+ std::string exitName;
+ std::string entryName;
+
+ if (getEntry()->getName().empty()) {
+ raw_string_ostream OS(entryName);
+
+ WriteAsOperand(OS, getEntry(), false);
+ entryName = OS.str();
+ } else
+ entryName = getEntry()->getNameStr();
+
+ if (getExit()) {
+ if (getExit()->getName().empty()) {
+ raw_string_ostream OS(exitName);
+
+ WriteAsOperand(OS, getExit(), false);
+ exitName = OS.str();
+ } else
+ exitName = getExit()->getNameStr();
+ } else
+ exitName = "<Function Return>";
+
+ return entryName + " => " + exitName;
+}
+
+void Region::verifyBBInRegion(BasicBlock *BB) const {
+ if (!contains(BB))
+ llvm_unreachable("Broken region found!");
+
+ BasicBlock *entry = getEntry(), *exit = getExit();
+
+ for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB); SI != SE; ++SI)
+ if (!contains(*SI) && exit != *SI)
+ llvm_unreachable("Broken region found!");
+
+ if (entry != BB)
+ for (pred_iterator SI = pred_begin(BB), SE = pred_end(BB); SI != SE; ++SI)
+ if (!contains(*SI))
+ llvm_unreachable("Broken region found!");
+}
+
+void Region::verifyWalk(BasicBlock *BB, std::set<BasicBlock*> *visited) const {
+ BasicBlock *exit = getExit();
+
+ visited->insert(BB);
+
+ verifyBBInRegion(BB);
+
+ for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB); SI != SE; ++SI)
+ if (*SI != exit && visited->find(*SI) == visited->end())
+ verifyWalk(*SI, visited);
+}
+
+void Region::verifyRegion() const {
+ // Only do verification when user wants to, otherwise this expensive
+ // check will be invoked by PassManager.
+ if (!VerifyRegionInfo) return;
+
+ std::set<BasicBlock*> visited;
+ verifyWalk(getEntry(), &visited);
+}
+
+void Region::verifyRegionNest() const {
+ for (Region::const_iterator RI = begin(), RE = end(); RI != RE; ++RI)
+ (*RI)->verifyRegionNest();
+
+ verifyRegion();
+}
+
+Region::block_iterator Region::block_begin() {
+ return GraphTraits<FlatIt<Region*> >::nodes_begin(this);
+}
+
+Region::block_iterator Region::block_end() {
+ return GraphTraits<FlatIt<Region*> >::nodes_end(this);
+}
+
+Region::const_block_iterator Region::block_begin() const {
+ return GraphTraits<FlatIt<const Region*> >::nodes_begin(this);
+}
+
+Region::const_block_iterator Region::block_end() const {
+ return GraphTraits<FlatIt<const Region*> >::nodes_end(this);
+}
+
+Region::element_iterator Region::element_begin() {
+ return GraphTraits<Region*>::nodes_begin(this);
+}
+
+Region::element_iterator Region::element_end() {
+ return GraphTraits<Region*>::nodes_end(this);
+}
+
+Region::const_element_iterator Region::element_begin() const {
+ return GraphTraits<const Region*>::nodes_begin(this);
+}
+
+Region::const_element_iterator Region::element_end() const {
+ return GraphTraits<const Region*>::nodes_end(this);
+}
+
+Region* Region::getSubRegionNode(BasicBlock *BB) const {
+ Region *R = RI->getRegionFor(BB);
+
+ if (!R || R == this)
+ return 0;
+
+ // If we pass the BB out of this region, that means our code is broken.
+ assert(contains(R) && "BB not in current region!");
+
+ while (contains(R->getParent()) && R->getParent() != this)
+ R = R->getParent();
+
+ if (R->getEntry() != BB)
+ return 0;
+
+ return R;
+}
+
+RegionNode* Region::getBBNode(BasicBlock *BB) const {
+ assert(contains(BB) && "Can get BB node out of this region!");
+
+ BBNodeMapT::const_iterator at = BBNodeMap.find(BB);
+
+ if (at != BBNodeMap.end())
+ return at->second;
+
+ RegionNode *NewNode = new RegionNode(const_cast<Region*>(this), BB);
+ BBNodeMap.insert(std::make_pair(BB, NewNode));
+ return NewNode;
+}
+
+RegionNode* Region::getNode(BasicBlock *BB) const {
+ assert(contains(BB) && "Can get BB node out of this region!");
+ if (Region* Child = getSubRegionNode(BB))
+ return Child->getNode();
+
+ return getBBNode(BB);
+}
+
+void Region::transferChildrenTo(Region *To) {
+ for (iterator I = begin(), E = end(); I != E; ++I) {
+ (*I)->parent = To;
+ To->children.push_back(*I);
+ }
+ children.clear();
+}
+
+void Region::addSubRegion(Region *SubRegion, bool moveChildren) {
+ assert(SubRegion->parent == 0 && "SubRegion already has a parent!");
+ assert(std::find(begin(), end(), SubRegion) == children.end()
+ && "Subregion already exists!");
+
+ SubRegion->parent = this;
+ children.push_back(SubRegion);
+
+ if (!moveChildren)
+ return;
+
+ assert(SubRegion->children.size() == 0
+ && "SubRegions that contain children are not supported");
+
+ for (element_iterator I = element_begin(), E = element_end(); I != E; ++I)
+ if (!(*I)->isSubRegion()) {
+ BasicBlock *BB = (*I)->getNodeAs<BasicBlock>();
+
+ if (SubRegion->contains(BB))
+ RI->setRegionFor(BB, SubRegion);
+ }
+
+ std::vector<Region*> Keep;
+ for (iterator I = begin(), E = end(); I != E; ++I)
+ if (SubRegion->contains(*I) && *I != SubRegion) {
+ SubRegion->children.push_back(*I);
+ (*I)->parent = SubRegion;
+ } else
+ Keep.push_back(*I);
+
+ children.clear();
+ children.insert(children.begin(), Keep.begin(), Keep.end());
+}
+
+
+Region *Region::removeSubRegion(Region *Child) {
+ assert(Child->parent == this && "Child is not a child of this region!");
+ Child->parent = 0;
+ RegionSet::iterator I = std::find(children.begin(), children.end(), Child);
+ assert(I != children.end() && "Region does not exit. Unable to remove.");
+ children.erase(children.begin()+(I-begin()));
+ return Child;
+}
+
+unsigned Region::getDepth() const {
+ unsigned Depth = 0;
+
+ for (Region *R = parent; R != 0; R = R->parent)
+ ++Depth;
+
+ return Depth;
+}
+
+Region *Region::getExpandedRegion() const {
+ unsigned NumSuccessors = exit->getTerminator()->getNumSuccessors();
+
+ if (NumSuccessors == 0)
+ return NULL;
+
+ for (pred_iterator PI = pred_begin(getExit()), PE = pred_end(getExit());
+ PI != PE; ++PI)
+ if (!DT->dominates(getEntry(), *PI))
+ return NULL;
+
+ Region *R = RI->getRegionFor(exit);
+
+ if (R->getEntry() != exit) {
+ if (exit->getTerminator()->getNumSuccessors() == 1)
+ return new Region(getEntry(), *succ_begin(exit), RI, DT);
+ else
+ return NULL;
+ }
+
+ while (R->getParent() && R->getParent()->getEntry() == exit)
+ R = R->getParent();
+
+ if (!DT->dominates(getEntry(), R->getExit()))
+ for (pred_iterator PI = pred_begin(getExit()), PE = pred_end(getExit());
+ PI != PE; ++PI)
+ if (!DT->dominates(R->getExit(), *PI))
+ return NULL;
+
+ return new Region(getEntry(), R->getExit(), RI, DT);
+}
+
+void Region::print(raw_ostream &OS, bool print_tree, unsigned level,
+ enum PrintStyle Style) const {
+ if (print_tree)
+ OS.indent(level*2) << "[" << level << "] " << getNameStr();
+ else
+ OS.indent(level*2) << getNameStr();
+
+ OS << "\n";
+
+
+ if (Style != PrintNone) {
+ OS.indent(level*2) << "{\n";
+ OS.indent(level*2 + 2);
+
+ if (Style == PrintBB) {
+ for (const_block_iterator I = block_begin(), E = block_end(); I!=E; ++I)
+ OS << **I << ", "; // TODO: remove the last ","
+ } else if (Style == PrintRN) {
+ for (const_element_iterator I = element_begin(), E = element_end(); I!=E; ++I)
+ OS << **I << ", "; // TODO: remove the last ",
+ }
+
+ OS << "\n";
+ }
+
+ if (print_tree)
+ for (const_iterator RI = begin(), RE = end(); RI != RE; ++RI)
+ (*RI)->print(OS, print_tree, level+1, Style);
+
+ if (Style != PrintNone)
+ OS.indent(level*2) << "} \n";
+}
+
+void Region::dump() const {
+ print(dbgs(), true, getDepth(), printStyle.getValue());
+}
+
+void Region::clearNodeCache() {
+ // Free the cached nodes.
+ for (BBNodeMapT::iterator I = BBNodeMap.begin(),
+ IE = BBNodeMap.end(); I != IE; ++I)
+ delete I->second;
+
+ BBNodeMap.clear();
+ for (Region::iterator RI = begin(), RE = end(); RI != RE; ++RI)
+ (*RI)->clearNodeCache();
+}
+
+//===----------------------------------------------------------------------===//
+// RegionInfo implementation
+//
+
+bool RegionInfo::isCommonDomFrontier(BasicBlock *BB, BasicBlock *entry,
+ BasicBlock *exit) const {
+ for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE; ++PI) {
+ BasicBlock *P = *PI;
+ if (DT->dominates(entry, P) && !DT->dominates(exit, P))
+ return false;
+ }
+ return true;
+}
+
+bool RegionInfo::isRegion(BasicBlock *entry, BasicBlock *exit) const {
+ assert(entry && exit && "entry and exit must not be null!");
+ typedef DominanceFrontier::DomSetType DST;
+
+ DST *entrySuccs = &DF->find(entry)->second;
+
+ // Exit is the header of a loop that contains the entry. In this case,
+ // the dominance frontier must only contain the exit.
+ if (!DT->dominates(entry, exit)) {
+ for (DST::iterator SI = entrySuccs->begin(), SE = entrySuccs->end();
+ SI != SE; ++SI)
+ if (*SI != exit && *SI != entry)
+ return false;
+
+ return true;
+ }
+
+ DST *exitSuccs = &DF->find(exit)->second;
+
+ // Do not allow edges leaving the region.
+ for (DST::iterator SI = entrySuccs->begin(), SE = entrySuccs->end();
+ SI != SE; ++SI) {
+ if (*SI == exit || *SI == entry)
+ continue;
+ if (exitSuccs->find(*SI) == exitSuccs->end())
+ return false;
+ if (!isCommonDomFrontier(*SI, entry, exit))
+ return false;
+ }
+
+ // Do not allow edges pointing into the region.
+ for (DST::iterator SI = exitSuccs->begin(), SE = exitSuccs->end();
+ SI != SE; ++SI)
+ if (DT->properlyDominates(entry, *SI) && *SI != exit)
+ return false;
+
+
+ return true;
+}
+
+void RegionInfo::insertShortCut(BasicBlock *entry, BasicBlock *exit,
+ BBtoBBMap *ShortCut) const {
+ assert(entry && exit && "entry and exit must not be null!");
+
+ BBtoBBMap::iterator e = ShortCut->find(exit);
+
+ if (e == ShortCut->end())
+ // No further region at exit available.
+ (*ShortCut)[entry] = exit;
+ else {
+ // We found a region e that starts at exit. Therefore (entry, e->second)
+ // is also a region, that is larger than (entry, exit). Insert the
+ // larger one.
+ BasicBlock *BB = e->second;
+ (*ShortCut)[entry] = BB;
+ }
+}
+
+DomTreeNode* RegionInfo::getNextPostDom(DomTreeNode* N,
+ BBtoBBMap *ShortCut) const {
+ BBtoBBMap::iterator e = ShortCut->find(N->getBlock());
+
+ if (e == ShortCut->end())
+ return N->getIDom();
+
+ return PDT->getNode(e->second)->getIDom();
+}
+
+bool RegionInfo::isTrivialRegion(BasicBlock *entry, BasicBlock *exit) const {
+ assert(entry && exit && "entry and exit must not be null!");
+
+ unsigned num_successors = succ_end(entry) - succ_begin(entry);
+
+ if (num_successors <= 1 && exit == *(succ_begin(entry)))
+ return true;
+
+ return false;
+}
+
+void RegionInfo::updateStatistics(Region *R) {
+ ++numRegions;
+
+ // TODO: Slow. Should only be enabled if -stats is used.
+ if (R->isSimple()) ++numSimpleRegions;
+}
+
+Region *RegionInfo::createRegion(BasicBlock *entry, BasicBlock *exit) {
+ assert(entry && exit && "entry and exit must not be null!");
+
+ if (isTrivialRegion(entry, exit))
+ return 0;
+
+ Region *region = new Region(entry, exit, this, DT);
+ BBtoRegion.insert(std::make_pair(entry, region));
+
+ #ifdef XDEBUG
+ region->verifyRegion();
+ #else
+ DEBUG(region->verifyRegion());
+ #endif
+
+ updateStatistics(region);
+ return region;
+}
+
+void RegionInfo::findRegionsWithEntry(BasicBlock *entry, BBtoBBMap *ShortCut) {
+ assert(entry);
+
+ DomTreeNode *N = PDT->getNode(entry);
+
+ if (!N)
+ return;
+
+ Region *lastRegion= 0;
+ BasicBlock *lastExit = entry;
+
+ // As only a BasicBlock that postdominates entry can finish a region, walk the
+ // post dominance tree upwards.
+ while ((N = getNextPostDom(N, ShortCut))) {
+ BasicBlock *exit = N->getBlock();
+
+ if (!exit)
+ break;
+
+ if (isRegion(entry, exit)) {
+ Region *newRegion = createRegion(entry, exit);
+
+ if (lastRegion)
+ newRegion->addSubRegion(lastRegion);
+
+ lastRegion = newRegion;
+ lastExit = exit;
+ }
+
+ // This can never be a region, so stop the search.
+ if (!DT->dominates(entry, exit))
+ break;
+ }
+
+ // Tried to create regions from entry to lastExit. Next time take a
+ // shortcut from entry to lastExit.
+ if (lastExit != entry)
+ insertShortCut(entry, lastExit, ShortCut);
+}
+
+void RegionInfo::scanForRegions(Function &F, BBtoBBMap *ShortCut) {
+ BasicBlock *entry = &(F.getEntryBlock());
+ DomTreeNode *N = DT->getNode(entry);
+
+ // Iterate over the dominance tree in post order to start with the small
+ // regions from the bottom of the dominance tree. If the small regions are
+ // detected first, detection of bigger regions is faster, as we can jump
+ // over the small regions.
+ for (po_iterator<DomTreeNode*> FI = po_begin(N), FE = po_end(N); FI != FE;
+ ++FI) {
+ findRegionsWithEntry(FI->getBlock(), ShortCut);
+ }
+}
+
+Region *RegionInfo::getTopMostParent(Region *region) {
+ while (region->parent)
+ region = region->getParent();
+
+ return region;
+}
+
+void RegionInfo::buildRegionsTree(DomTreeNode *N, Region *region) {
+ BasicBlock *BB = N->getBlock();
+
+ // Passed region exit
+ while (BB == region->getExit())
+ region = region->getParent();
+
+ BBtoRegionMap::iterator it = BBtoRegion.find(BB);
+
+ // This basic block is a start block of a region. It is already in the
+ // BBtoRegion relation. Only the child basic blocks have to be updated.
+ if (it != BBtoRegion.end()) {
+ Region *newRegion = it->second;;
+ region->addSubRegion(getTopMostParent(newRegion));
+ region = newRegion;
+ } else {
+ BBtoRegion[BB] = region;
+ }
+
+ for (DomTreeNode::iterator CI = N->begin(), CE = N->end(); CI != CE; ++CI)
+ buildRegionsTree(*CI, region);
+}
+
+void RegionInfo::releaseMemory() {
+ BBtoRegion.clear();
+ if (TopLevelRegion)
+ delete TopLevelRegion;
+ TopLevelRegion = 0;
+}
+
+RegionInfo::RegionInfo() : FunctionPass(ID) {
+ initializeRegionInfoPass(*PassRegistry::getPassRegistry());
+ TopLevelRegion = 0;
+}
+
+RegionInfo::~RegionInfo() {
+ releaseMemory();
+}
+
+void RegionInfo::Calculate(Function &F) {
+ // ShortCut a function where for every BB the exit of the largest region
+ // starting with BB is stored. These regions can be threated as single BBS.
+ // This improves performance on linear CFGs.
+ BBtoBBMap ShortCut;
+
+ scanForRegions(F, &ShortCut);
+ BasicBlock *BB = &F.getEntryBlock();
+ buildRegionsTree(DT->getNode(BB), TopLevelRegion);
+}
+
+bool RegionInfo::runOnFunction(Function &F) {
+ releaseMemory();
+
+ DT = &getAnalysis<DominatorTree>();
+ PDT = &getAnalysis<PostDominatorTree>();
+ DF = &getAnalysis<DominanceFrontier>();
+
+ TopLevelRegion = new Region(&F.getEntryBlock(), 0, this, DT, 0);
+ updateStatistics(TopLevelRegion);
+
+ Calculate(F);
+
+ return false;
+}
+
+void RegionInfo::getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ AU.addRequiredTransitive<DominatorTree>();
+ AU.addRequired<PostDominatorTree>();
+ AU.addRequired<DominanceFrontier>();
+}
+
+void RegionInfo::print(raw_ostream &OS, const Module *) const {
+ OS << "Region tree:\n";
+ TopLevelRegion->print(OS, true, 0, printStyle.getValue());
+ OS << "End region tree\n";
+}
+
+void RegionInfo::verifyAnalysis() const {
+ // Only do verification when user wants to, otherwise this expensive check
+ // will be invoked by PMDataManager::verifyPreservedAnalysis when
+ // a regionpass (marked PreservedAll) finish.
+ if (!VerifyRegionInfo) return;
+
+ TopLevelRegion->verifyRegionNest();
+}
+
+// Region pass manager support.
+Region *RegionInfo::getRegionFor(BasicBlock *BB) const {
+ BBtoRegionMap::const_iterator I=
+ BBtoRegion.find(BB);
+ return I != BBtoRegion.end() ? I->second : 0;
+}
+
+void RegionInfo::setRegionFor(BasicBlock *BB, Region *R) {
+ BBtoRegion[BB] = R;
+}
+
+Region *RegionInfo::operator[](BasicBlock *BB) const {
+ return getRegionFor(BB);
+}
+
+BasicBlock *RegionInfo::getMaxRegionExit(BasicBlock *BB) const {
+ BasicBlock *Exit = NULL;
+
+ while (true) {
+ // Get largest region that starts at BB.
+ Region *R = getRegionFor(BB);
+ while (R && R->getParent() && R->getParent()->getEntry() == BB)
+ R = R->getParent();
+
+ // Get the single exit of BB.
+ if (R && R->getEntry() == BB)
+ Exit = R->getExit();
+ else if (++succ_begin(BB) == succ_end(BB))
+ Exit = *succ_begin(BB);
+ else // No single exit exists.
+ return Exit;
+
+ // Get largest region that starts at Exit.
+ Region *ExitR = getRegionFor(Exit);
+ while (ExitR && ExitR->getParent()
+ && ExitR->getParent()->getEntry() == Exit)
+ ExitR = ExitR->getParent();
+
+ for (pred_iterator PI = pred_begin(Exit), PE = pred_end(Exit); PI != PE;
+ ++PI)
+ if (!R->contains(*PI) && !ExitR->contains(*PI))
+ break;
+
+ // This stops infinite cycles.
+ if (DT->dominates(Exit, BB))
+ break;
+
+ BB = Exit;
+ }
+
+ return Exit;
+}
+
+Region*
+RegionInfo::getCommonRegion(Region *A, Region *B) const {
+ assert (A && B && "One of the Regions is NULL");
+
+ if (A->contains(B)) return A;
+
+ while (!B->contains(A))
+ B = B->getParent();
+
+ return B;
+}
+
+Region*
+RegionInfo::getCommonRegion(SmallVectorImpl<Region*> &Regions) const {
+ Region* ret = Regions.back();
+ Regions.pop_back();
+
+ for (SmallVectorImpl<Region*>::const_iterator I = Regions.begin(),
+ E = Regions.end(); I != E; ++I)
+ ret = getCommonRegion(ret, *I);
+
+ return ret;
+}
+
+Region*
+RegionInfo::getCommonRegion(SmallVectorImpl<BasicBlock*> &BBs) const {
+ Region* ret = getRegionFor(BBs.back());
+ BBs.pop_back();
+
+ for (SmallVectorImpl<BasicBlock*>::const_iterator I = BBs.begin(),
+ E = BBs.end(); I != E; ++I)
+ ret = getCommonRegion(ret, getRegionFor(*I));
+
+ return ret;
+}
+
+void RegionInfo::splitBlock(BasicBlock* NewBB, BasicBlock *OldBB)
+{
+ Region *R = getRegionFor(OldBB);
+
+ setRegionFor(NewBB, R);
+
+ while (R->getEntry() == OldBB && !R->isTopLevelRegion()) {
+ R->replaceEntry(NewBB);
+ R = R->getParent();
+ }
+
+ setRegionFor(OldBB, R);
+}
+
+char RegionInfo::ID = 0;
+INITIALIZE_PASS_BEGIN(RegionInfo, "regions",
+ "Detect single entry single exit regions", true, true)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_PASS_DEPENDENCY(PostDominatorTree)
+INITIALIZE_PASS_DEPENDENCY(DominanceFrontier)
+INITIALIZE_PASS_END(RegionInfo, "regions",
+ "Detect single entry single exit regions", true, true)
+
+// Create methods available outside of this file, to use them
+// "include/llvm/LinkAllPasses.h". Otherwise the pass would be deleted by
+// the link time optimization.
+
+namespace llvm {
+ FunctionPass *createRegionInfoPass() {
+ return new RegionInfo();
+ }
+}
+
diff --git a/src/LLVM/lib/Analysis/RegionPass.cpp b/src/LLVM/lib/Analysis/RegionPass.cpp
new file mode 100644
index 0000000..3a3529b
--- /dev/null
+++ b/src/LLVM/lib/Analysis/RegionPass.cpp
@@ -0,0 +1,275 @@
+//===- RegionPass.cpp - Region Pass and Region Pass Manager ---------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements RegionPass and RGPassManager. All region optimization
+// and transformation passes are derived from RegionPass. RGPassManager is
+// responsible for managing RegionPasses.
+// most of these codes are COPY from LoopPass.cpp
+//
+//===----------------------------------------------------------------------===//
+#include "llvm/Analysis/RegionPass.h"
+#include "llvm/Analysis/RegionIterator.h"
+#include "llvm/Support/Timer.h"
+
+#define DEBUG_TYPE "regionpassmgr"
+#include "llvm/Support/Debug.h"
+using namespace llvm;
+
+//===----------------------------------------------------------------------===//
+// RGPassManager
+//
+
+char RGPassManager::ID = 0;
+
+RGPassManager::RGPassManager()
+ : FunctionPass(ID), PMDataManager() {
+ skipThisRegion = false;
+ redoThisRegion = false;
+ RI = NULL;
+ CurrentRegion = NULL;
+}
+
+// Recurse through all subregions and all regions into RQ.
+static void addRegionIntoQueue(Region *R, std::deque<Region *> &RQ) {
+ RQ.push_back(R);
+ for (Region::iterator I = R->begin(), E = R->end(); I != E; ++I)
+ addRegionIntoQueue(*I, RQ);
+}
+
+/// Pass Manager itself does not invalidate any analysis info.
+void RGPassManager::getAnalysisUsage(AnalysisUsage &Info) const {
+ Info.addRequired<RegionInfo>();
+ Info.setPreservesAll();
+}
+
+/// run - Execute all of the passes scheduled for execution. Keep track of
+/// whether any of the passes modifies the function, and if so, return true.
+bool RGPassManager::runOnFunction(Function &F) {
+ RI = &getAnalysis<RegionInfo>();
+ bool Changed = false;
+
+ // Collect inherited analysis from Module level pass manager.
+ populateInheritedAnalysis(TPM->activeStack);
+
+ addRegionIntoQueue(RI->getTopLevelRegion(), RQ);
+
+ if (RQ.empty()) // No regions, skip calling finalizers
+ return false;
+
+ // Initialization
+ for (std::deque<Region *>::const_iterator I = RQ.begin(), E = RQ.end();
+ I != E; ++I) {
+ Region *R = *I;
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ RegionPass *RP = (RegionPass *)getContainedPass(Index);
+ Changed |= RP->doInitialization(R, *this);
+ }
+ }
+
+ // Walk Regions
+ while (!RQ.empty()) {
+
+ CurrentRegion = RQ.back();
+ skipThisRegion = false;
+ redoThisRegion = false;
+
+ // Run all passes on the current Region.
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ RegionPass *P = (RegionPass*)getContainedPass(Index);
+
+ dumpPassInfo(P, EXECUTION_MSG, ON_REGION_MSG,
+ CurrentRegion->getNameStr());
+ dumpRequiredSet(P);
+
+ initializeAnalysisImpl(P);
+
+ {
+ PassManagerPrettyStackEntry X(P, *CurrentRegion->getEntry());
+
+ TimeRegion PassTimer(getPassTimer(P));
+ Changed |= P->runOnRegion(CurrentRegion, *this);
+ }
+
+ if (Changed)
+ dumpPassInfo(P, MODIFICATION_MSG, ON_REGION_MSG,
+ skipThisRegion ? "<deleted>" :
+ CurrentRegion->getNameStr());
+ dumpPreservedSet(P);
+
+ if (!skipThisRegion) {
+ // Manually check that this region is still healthy. This is done
+ // instead of relying on RegionInfo::verifyRegion since RegionInfo
+ // is a function pass and it's really expensive to verify every
+ // Region in the function every time. That level of checking can be
+ // enabled with the -verify-region-info option.
+ {
+ TimeRegion PassTimer(getPassTimer(P));
+ CurrentRegion->verifyRegion();
+ }
+
+ // Then call the regular verifyAnalysis functions.
+ verifyPreservedAnalysis(P);
+ }
+
+ removeNotPreservedAnalysis(P);
+ recordAvailableAnalysis(P);
+ removeDeadPasses(P,
+ skipThisRegion ? "<deleted>" :
+ CurrentRegion->getNameStr(),
+ ON_REGION_MSG);
+
+ if (skipThisRegion)
+ // Do not run other passes on this region.
+ break;
+ }
+
+ // If the region was deleted, release all the region passes. This frees up
+ // some memory, and avoids trouble with the pass manager trying to call
+ // verifyAnalysis on them.
+ if (skipThisRegion)
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ Pass *P = getContainedPass(Index);
+ freePass(P, "<deleted>", ON_REGION_MSG);
+ }
+
+ // Pop the region from queue after running all passes.
+ RQ.pop_back();
+
+ if (redoThisRegion)
+ RQ.push_back(CurrentRegion);
+
+ // Free all region nodes created in region passes.
+ RI->clearNodeCache();
+ }
+
+ // Finalization
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ RegionPass *P = (RegionPass*)getContainedPass(Index);
+ Changed |= P->doFinalization();
+ }
+
+ // Print the region tree after all pass.
+ DEBUG(
+ dbgs() << "\nRegion tree of function " << F.getName()
+ << " after all region Pass:\n";
+ RI->dump();
+ dbgs() << "\n";
+ );
+
+ return Changed;
+}
+
+/// Print passes managed by this manager
+void RGPassManager::dumpPassStructure(unsigned Offset) {
+ errs().indent(Offset*2) << "Region Pass Manager\n";
+ for (unsigned Index = 0; Index < getNumContainedPasses(); ++Index) {
+ Pass *P = getContainedPass(Index);
+ P->dumpPassStructure(Offset + 1);
+ dumpLastUses(P, Offset+1);
+ }
+}
+
+namespace {
+//===----------------------------------------------------------------------===//
+// PrintRegionPass
+class PrintRegionPass : public RegionPass {
+private:
+ std::string Banner;
+ raw_ostream &Out; // raw_ostream to print on.
+
+public:
+ static char ID;
+ PrintRegionPass() : RegionPass(ID), Out(dbgs()) {}
+ PrintRegionPass(const std::string &B, raw_ostream &o)
+ : RegionPass(ID), Banner(B), Out(o) {}
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ }
+
+ virtual bool runOnRegion(Region *R, RGPassManager &RGM) {
+ Out << Banner;
+ for (Region::block_iterator I = R->block_begin(), E = R->block_end();
+ I != E; ++I)
+ (*I)->getEntry()->print(Out);
+
+ return false;
+ }
+};
+
+char PrintRegionPass::ID = 0;
+} //end anonymous namespace
+
+//===----------------------------------------------------------------------===//
+// RegionPass
+
+// Check if this pass is suitable for the current RGPassManager, if
+// available. This pass P is not suitable for a RGPassManager if P
+// is not preserving higher level analysis info used by other
+// RGPassManager passes. In such case, pop RGPassManager from the
+// stack. This will force assignPassManager() to create new
+// LPPassManger as expected.
+void RegionPass::preparePassManager(PMStack &PMS) {
+
+ // Find RGPassManager
+ while (!PMS.empty() &&
+ PMS.top()->getPassManagerType() > PMT_RegionPassManager)
+ PMS.pop();
+
+
+ // If this pass is destroying high level information that is used
+ // by other passes that are managed by LPM then do not insert
+ // this pass in current LPM. Use new RGPassManager.
+ if (PMS.top()->getPassManagerType() == PMT_RegionPassManager &&
+ !PMS.top()->preserveHigherLevelAnalysis(this))
+ PMS.pop();
+}
+
+/// Assign pass manager to manage this pass.
+void RegionPass::assignPassManager(PMStack &PMS,
+ PassManagerType PreferredType) {
+ // Find RGPassManager
+ while (!PMS.empty() &&
+ PMS.top()->getPassManagerType() > PMT_RegionPassManager)
+ PMS.pop();
+
+ RGPassManager *RGPM;
+
+ // Create new Region Pass Manager if it does not exist.
+ if (PMS.top()->getPassManagerType() == PMT_RegionPassManager)
+ RGPM = (RGPassManager*)PMS.top();
+ else {
+
+ assert (!PMS.empty() && "Unable to create Region Pass Manager");
+ PMDataManager *PMD = PMS.top();
+
+ // [1] Create new Region Pass Manager
+ RGPM = new RGPassManager();
+ RGPM->populateInheritedAnalysis(PMS);
+
+ // [2] Set up new manager's top level manager
+ PMTopLevelManager *TPM = PMD->getTopLevelManager();
+ TPM->addIndirectPassManager(RGPM);
+
+ // [3] Assign manager to manage this new manager. This may create
+ // and push new managers into PMS
+ TPM->schedulePass(RGPM);
+
+ // [4] Push new manager into PMS
+ PMS.push(RGPM);
+ }
+
+ RGPM->add(this);
+}
+
+/// Get the printer pass
+Pass *RegionPass::createPrinterPass(raw_ostream &O,
+ const std::string &Banner) const {
+ return new PrintRegionPass(Banner, O);
+}
diff --git a/src/LLVM/lib/Analysis/RegionPrinter.cpp b/src/LLVM/lib/Analysis/RegionPrinter.cpp
new file mode 100644
index 0000000..a1730b0
--- /dev/null
+++ b/src/LLVM/lib/Analysis/RegionPrinter.cpp
@@ -0,0 +1,220 @@
+//===- RegionPrinter.cpp - Print regions tree pass ------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+// Print out the region tree of a function using dotty/graphviz.
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/RegionInfo.h"
+#include "llvm/Analysis/RegionIterator.h"
+#include "llvm/Analysis/RegionPrinter.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/DOTGraphTraitsPass.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/PostOrderIterator.h"
+#include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+//===----------------------------------------------------------------------===//
+/// onlySimpleRegion - Show only the simple regions in the RegionViewer.
+static cl::opt<bool>
+onlySimpleRegions("only-simple-regions",
+ cl::desc("Show only simple regions in the graphviz viewer"),
+ cl::Hidden,
+ cl::init(false));
+
+namespace llvm {
+template<>
+struct DOTGraphTraits<RegionNode*> : public DefaultDOTGraphTraits {
+
+ DOTGraphTraits (bool isSimple=false)
+ : DefaultDOTGraphTraits(isSimple) {}
+
+ std::string getNodeLabel(RegionNode *Node, RegionNode *Graph) {
+
+ if (!Node->isSubRegion()) {
+ BasicBlock *BB = Node->getNodeAs<BasicBlock>();
+
+ if (isSimple())
+ return DOTGraphTraits<const Function*>
+ ::getSimpleNodeLabel(BB, BB->getParent());
+ else
+ return DOTGraphTraits<const Function*>
+ ::getCompleteNodeLabel(BB, BB->getParent());
+ }
+
+ return "Not implemented";
+ }
+};
+
+template<>
+struct DOTGraphTraits<RegionInfo*> : public DOTGraphTraits<RegionNode*> {
+
+ DOTGraphTraits (bool isSimple=false)
+ : DOTGraphTraits<RegionNode*>(isSimple) {}
+
+ static std::string getGraphName(RegionInfo *DT) {
+ return "Region Graph";
+ }
+
+ std::string getNodeLabel(RegionNode *Node, RegionInfo *G) {
+ return DOTGraphTraits<RegionNode*>::getNodeLabel(Node,
+ G->getTopLevelRegion());
+ }
+
+ std::string getEdgeAttributes(RegionNode *srcNode,
+ GraphTraits<RegionInfo*>::ChildIteratorType CI, RegionInfo *RI) {
+
+ RegionNode *destNode = *CI;
+
+ if (srcNode->isSubRegion() || destNode->isSubRegion())
+ return "";
+
+ // In case of a backedge, do not use it to define the layout of the nodes.
+ BasicBlock *srcBB = srcNode->getNodeAs<BasicBlock>();
+ BasicBlock *destBB = destNode->getNodeAs<BasicBlock>();
+
+ Region *R = RI->getRegionFor(destBB);
+
+ while (R && R->getParent())
+ if (R->getParent()->getEntry() == destBB)
+ R = R->getParent();
+ else
+ break;
+
+ if (R->getEntry() == destBB && R->contains(srcBB))
+ return "constraint=false";
+
+ return "";
+ }
+
+ // Print the cluster of the subregions. This groups the single basic blocks
+ // and adds a different background color for each group.
+ static void printRegionCluster(const Region *R, GraphWriter<RegionInfo*> &GW,
+ unsigned depth = 0) {
+ raw_ostream &O = GW.getOStream();
+ O.indent(2 * depth) << "subgraph cluster_" << static_cast<const void*>(R)
+ << " {\n";
+ O.indent(2 * (depth + 1)) << "label = \"\";\n";
+
+ if (!onlySimpleRegions || R->isSimple()) {
+ O.indent(2 * (depth + 1)) << "style = filled;\n";
+ O.indent(2 * (depth + 1)) << "color = "
+ << ((R->getDepth() * 2 % 12) + 1) << "\n";
+
+ } else {
+ O.indent(2 * (depth + 1)) << "style = solid;\n";
+ O.indent(2 * (depth + 1)) << "color = "
+ << ((R->getDepth() * 2 % 12) + 2) << "\n";
+ }
+
+ for (Region::const_iterator RI = R->begin(), RE = R->end(); RI != RE; ++RI)
+ printRegionCluster(*RI, GW, depth + 1);
+
+ RegionInfo *RI = R->getRegionInfo();
+
+ for (Region::const_block_iterator BI = R->block_begin(),
+ BE = R->block_end(); BI != BE; ++BI) {
+ BasicBlock *BB = (*BI)->getNodeAs<BasicBlock>();
+ if (RI->getRegionFor(BB) == R)
+ O.indent(2 * (depth + 1)) << "Node"
+ << static_cast<const void*>(RI->getTopLevelRegion()->getBBNode(BB))
+ << ";\n";
+ }
+
+ O.indent(2 * depth) << "}\n";
+ }
+
+ static void addCustomGraphFeatures(const RegionInfo* RI,
+ GraphWriter<RegionInfo*> &GW) {
+ raw_ostream &O = GW.getOStream();
+ O << "\tcolorscheme = \"paired12\"\n";
+ printRegionCluster(RI->getTopLevelRegion(), GW, 4);
+ }
+};
+} //end namespace llvm
+
+namespace {
+
+struct RegionViewer
+ : public DOTGraphTraitsViewer<RegionInfo, false> {
+ static char ID;
+ RegionViewer() : DOTGraphTraitsViewer<RegionInfo, false>("reg", ID){
+ initializeRegionViewerPass(*PassRegistry::getPassRegistry());
+ }
+};
+char RegionViewer::ID = 0;
+
+struct RegionOnlyViewer
+ : public DOTGraphTraitsViewer<RegionInfo, true> {
+ static char ID;
+ RegionOnlyViewer() : DOTGraphTraitsViewer<RegionInfo, true>("regonly", ID) {
+ initializeRegionOnlyViewerPass(*PassRegistry::getPassRegistry());
+ }
+};
+char RegionOnlyViewer::ID = 0;
+
+struct RegionPrinter
+ : public DOTGraphTraitsPrinter<RegionInfo, false> {
+ static char ID;
+ RegionPrinter() :
+ DOTGraphTraitsPrinter<RegionInfo, false>("reg", ID) {
+ initializeRegionPrinterPass(*PassRegistry::getPassRegistry());
+ }
+};
+char RegionPrinter::ID = 0;
+} //end anonymous namespace
+
+INITIALIZE_PASS(RegionPrinter, "dot-regions",
+ "Print regions of function to 'dot' file", true, true)
+
+INITIALIZE_PASS(RegionViewer, "view-regions", "View regions of function",
+ true, true)
+
+INITIALIZE_PASS(RegionOnlyViewer, "view-regions-only",
+ "View regions of function (with no function bodies)",
+ true, true)
+
+namespace {
+
+struct RegionOnlyPrinter
+ : public DOTGraphTraitsPrinter<RegionInfo, true> {
+ static char ID;
+ RegionOnlyPrinter() :
+ DOTGraphTraitsPrinter<RegionInfo, true>("reg", ID) {
+ initializeRegionOnlyPrinterPass(*PassRegistry::getPassRegistry());
+ }
+};
+
+}
+
+char RegionOnlyPrinter::ID = 0;
+INITIALIZE_PASS(RegionOnlyPrinter, "dot-regions-only",
+ "Print regions of function to 'dot' file "
+ "(with no function bodies)",
+ true, true)
+
+FunctionPass* llvm::createRegionViewerPass() {
+ return new RegionViewer();
+}
+
+FunctionPass* llvm::createRegionOnlyViewerPass() {
+ return new RegionOnlyViewer();
+}
+
+FunctionPass* llvm::createRegionPrinterPass() {
+ return new RegionPrinter();
+}
+
+FunctionPass* llvm::createRegionOnlyPrinterPass() {
+ return new RegionOnlyPrinter();
+}
+
diff --git a/src/LLVM/lib/Analysis/ScalarEvolution.cpp b/src/LLVM/lib/Analysis/ScalarEvolution.cpp
index b3ddfc0..e0ac56c 100644
--- a/src/LLVM/lib/Analysis/ScalarEvolution.cpp
+++ b/src/LLVM/lib/Analysis/ScalarEvolution.cpp
@@ -69,6 +69,7 @@
#include "llvm/Operator.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Assembly/Writer.h"
@@ -103,8 +104,12 @@
"derived loop"),
cl::init(100));
-INITIALIZE_PASS(ScalarEvolution, "scalar-evolution",
- "Scalar Evolution Analysis", false, true);
+INITIALIZE_PASS_BEGIN(ScalarEvolution, "scalar-evolution",
+ "Scalar Evolution Analysis", false, true)
+INITIALIZE_PASS_DEPENDENCY(LoopInfo)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_PASS_END(ScalarEvolution, "scalar-evolution",
+ "Scalar Evolution Analysis", false, true)
char ScalarEvolution::ID = 0;
//===----------------------------------------------------------------------===//
@@ -115,13 +120,142 @@
// Implementation of the SCEV class.
//
-SCEV::~SCEV() {}
-
void SCEV::dump() const {
print(dbgs());
dbgs() << '\n';
}
+void SCEV::print(raw_ostream &OS) const {
+ switch (getSCEVType()) {
+ case scConstant:
+ WriteAsOperand(OS, cast<SCEVConstant>(this)->getValue(), false);
+ return;
+ case scTruncate: {
+ const SCEVTruncateExpr *Trunc = cast<SCEVTruncateExpr>(this);
+ const SCEV *Op = Trunc->getOperand();
+ OS << "(trunc " << *Op->getType() << " " << *Op << " to "
+ << *Trunc->getType() << ")";
+ return;
+ }
+ case scZeroExtend: {
+ const SCEVZeroExtendExpr *ZExt = cast<SCEVZeroExtendExpr>(this);
+ const SCEV *Op = ZExt->getOperand();
+ OS << "(zext " << *Op->getType() << " " << *Op << " to "
+ << *ZExt->getType() << ")";
+ return;
+ }
+ case scSignExtend: {
+ const SCEVSignExtendExpr *SExt = cast<SCEVSignExtendExpr>(this);
+ const SCEV *Op = SExt->getOperand();
+ OS << "(sext " << *Op->getType() << " " << *Op << " to "
+ << *SExt->getType() << ")";
+ return;
+ }
+ case scAddRecExpr: {
+ const SCEVAddRecExpr *AR = cast<SCEVAddRecExpr>(this);
+ OS << "{" << *AR->getOperand(0);
+ for (unsigned i = 1, e = AR->getNumOperands(); i != e; ++i)
+ OS << ",+," << *AR->getOperand(i);
+ OS << "}<";
+ if (AR->getNoWrapFlags(FlagNUW))
+ OS << "nuw><";
+ if (AR->getNoWrapFlags(FlagNSW))
+ OS << "nsw><";
+ if (AR->getNoWrapFlags(FlagNW) &&
+ !AR->getNoWrapFlags((NoWrapFlags)(FlagNUW | FlagNSW)))
+ OS << "nw><";
+ WriteAsOperand(OS, AR->getLoop()->getHeader(), /*PrintType=*/false);
+ OS << ">";
+ return;
+ }
+ case scAddExpr:
+ case scMulExpr:
+ case scUMaxExpr:
+ case scSMaxExpr: {
+ const SCEVNAryExpr *NAry = cast<SCEVNAryExpr>(this);
+ const char *OpStr = 0;
+ switch (NAry->getSCEVType()) {
+ case scAddExpr: OpStr = " + "; break;
+ case scMulExpr: OpStr = " * "; break;
+ case scUMaxExpr: OpStr = " umax "; break;
+ case scSMaxExpr: OpStr = " smax "; break;
+ }
+ OS << "(";
+ for (SCEVNAryExpr::op_iterator I = NAry->op_begin(), E = NAry->op_end();
+ I != E; ++I) {
+ OS << **I;
+ if (llvm::next(I) != E)
+ OS << OpStr;
+ }
+ OS << ")";
+ return;
+ }
+ case scUDivExpr: {
+ const SCEVUDivExpr *UDiv = cast<SCEVUDivExpr>(this);
+ OS << "(" << *UDiv->getLHS() << " /u " << *UDiv->getRHS() << ")";
+ return;
+ }
+ case scUnknown: {
+ const SCEVUnknown *U = cast<SCEVUnknown>(this);
+ Type *AllocTy;
+ if (U->isSizeOf(AllocTy)) {
+ OS << "sizeof(" << *AllocTy << ")";
+ return;
+ }
+ if (U->isAlignOf(AllocTy)) {
+ OS << "alignof(" << *AllocTy << ")";
+ return;
+ }
+
+ Type *CTy;
+ Constant *FieldNo;
+ if (U->isOffsetOf(CTy, FieldNo)) {
+ OS << "offsetof(" << *CTy << ", ";
+ WriteAsOperand(OS, FieldNo, false);
+ OS << ")";
+ return;
+ }
+
+ // Otherwise just print it normally.
+ WriteAsOperand(OS, U->getValue(), false);
+ return;
+ }
+ case scCouldNotCompute:
+ OS << "***COULDNOTCOMPUTE***";
+ return;
+ default: break;
+ }
+ llvm_unreachable("Unknown SCEV kind!");
+}
+
+Type *SCEV::getType() const {
+ switch (getSCEVType()) {
+ case scConstant:
+ return cast<SCEVConstant>(this)->getType();
+ case scTruncate:
+ case scZeroExtend:
+ case scSignExtend:
+ return cast<SCEVCastExpr>(this)->getType();
+ case scAddRecExpr:
+ case scMulExpr:
+ case scUMaxExpr:
+ case scSMaxExpr:
+ return cast<SCEVNAryExpr>(this)->getType();
+ case scAddExpr:
+ return cast<SCEVAddExpr>(this)->getType();
+ case scUDivExpr:
+ return cast<SCEVUDivExpr>(this)->getType();
+ case scUnknown:
+ return cast<SCEVUnknown>(this)->getType();
+ case scCouldNotCompute:
+ llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
+ return 0;
+ default: break;
+ }
+ llvm_unreachable("Unknown SCEV kind!");
+ return 0;
+}
+
bool SCEV::isZero() const {
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(this))
return SC->getValue()->isZero();
@@ -143,30 +277,6 @@
SCEVCouldNotCompute::SCEVCouldNotCompute() :
SCEV(FoldingSetNodeIDRef(), scCouldNotCompute) {}
-bool SCEVCouldNotCompute::isLoopInvariant(const Loop *L) const {
- llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
- return false;
-}
-
-const Type *SCEVCouldNotCompute::getType() const {
- llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
- return 0;
-}
-
-bool SCEVCouldNotCompute::hasComputableLoopEvolution(const Loop *L) const {
- llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
- return false;
-}
-
-bool SCEVCouldNotCompute::hasOperand(const SCEV *) const {
- llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
- return false;
-}
-
-void SCEVCouldNotCompute::print(raw_ostream &OS) const {
- OS << "***COULDNOTCOMPUTE***";
-}
-
bool SCEVCouldNotCompute::classof(const SCEV *S) {
return S->getSCEVType() == scCouldNotCompute;
}
@@ -187,159 +297,42 @@
}
const SCEV *
-ScalarEvolution::getConstant(const Type *Ty, uint64_t V, bool isSigned) {
- const IntegerType *ITy = cast<IntegerType>(getEffectiveSCEVType(Ty));
+ScalarEvolution::getConstant(Type *Ty, uint64_t V, bool isSigned) {
+ IntegerType *ITy = cast<IntegerType>(getEffectiveSCEVType(Ty));
return getConstant(ConstantInt::get(ITy, V, isSigned));
}
-const Type *SCEVConstant::getType() const { return V->getType(); }
-
-void SCEVConstant::print(raw_ostream &OS) const {
- WriteAsOperand(OS, V, false);
-}
-
SCEVCastExpr::SCEVCastExpr(const FoldingSetNodeIDRef ID,
- unsigned SCEVTy, const SCEV *op, const Type *ty)
+ unsigned SCEVTy, const SCEV *op, Type *ty)
: SCEV(ID, SCEVTy), Op(op), Ty(ty) {}
-bool SCEVCastExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
- return Op->dominates(BB, DT);
-}
-
-bool SCEVCastExpr::properlyDominates(BasicBlock *BB, DominatorTree *DT) const {
- return Op->properlyDominates(BB, DT);
-}
-
SCEVTruncateExpr::SCEVTruncateExpr(const FoldingSetNodeIDRef ID,
- const SCEV *op, const Type *ty)
+ const SCEV *op, Type *ty)
: SCEVCastExpr(ID, scTruncate, op, ty) {
assert((Op->getType()->isIntegerTy() || Op->getType()->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot truncate non-integer value!");
}
-void SCEVTruncateExpr::print(raw_ostream &OS) const {
- OS << "(trunc " << *Op->getType() << " " << *Op << " to " << *Ty << ")";
-}
-
SCEVZeroExtendExpr::SCEVZeroExtendExpr(const FoldingSetNodeIDRef ID,
- const SCEV *op, const Type *ty)
+ const SCEV *op, Type *ty)
: SCEVCastExpr(ID, scZeroExtend, op, ty) {
assert((Op->getType()->isIntegerTy() || Op->getType()->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot zero extend non-integer value!");
}
-void SCEVZeroExtendExpr::print(raw_ostream &OS) const {
- OS << "(zext " << *Op->getType() << " " << *Op << " to " << *Ty << ")";
-}
-
SCEVSignExtendExpr::SCEVSignExtendExpr(const FoldingSetNodeIDRef ID,
- const SCEV *op, const Type *ty)
+ const SCEV *op, Type *ty)
: SCEVCastExpr(ID, scSignExtend, op, ty) {
assert((Op->getType()->isIntegerTy() || Op->getType()->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot sign extend non-integer value!");
}
-void SCEVSignExtendExpr::print(raw_ostream &OS) const {
- OS << "(sext " << *Op->getType() << " " << *Op << " to " << *Ty << ")";
-}
-
-void SCEVCommutativeExpr::print(raw_ostream &OS) const {
- const char *OpStr = getOperationStr();
- OS << "(";
- for (op_iterator I = op_begin(), E = op_end(); I != E; ++I) {
- OS << **I;
- if (llvm::next(I) != E)
- OS << OpStr;
- }
- OS << ")";
-}
-
-bool SCEVNAryExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
- for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
- if (!getOperand(i)->dominates(BB, DT))
- return false;
- }
- return true;
-}
-
-bool SCEVNAryExpr::properlyDominates(BasicBlock *BB, DominatorTree *DT) const {
- for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
- if (!getOperand(i)->properlyDominates(BB, DT))
- return false;
- }
- return true;
-}
-
-bool SCEVUDivExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
- return LHS->dominates(BB, DT) && RHS->dominates(BB, DT);
-}
-
-bool SCEVUDivExpr::properlyDominates(BasicBlock *BB, DominatorTree *DT) const {
- return LHS->properlyDominates(BB, DT) && RHS->properlyDominates(BB, DT);
-}
-
-void SCEVUDivExpr::print(raw_ostream &OS) const {
- OS << "(" << *LHS << " /u " << *RHS << ")";
-}
-
-const Type *SCEVUDivExpr::getType() const {
- // In most cases the types of LHS and RHS will be the same, but in some
- // crazy cases one or the other may be a pointer. ScalarEvolution doesn't
- // depend on the type for correctness, but handling types carefully can
- // avoid extra casts in the SCEVExpander. The LHS is more likely to be
- // a pointer type than the RHS, so use the RHS' type here.
- return RHS->getType();
-}
-
-bool SCEVAddRecExpr::isLoopInvariant(const Loop *QueryLoop) const {
- // Add recurrences are never invariant in the function-body (null loop).
- if (!QueryLoop)
- return false;
-
- // This recurrence is variant w.r.t. QueryLoop if QueryLoop contains L.
- if (QueryLoop->contains(L))
- return false;
-
- // This recurrence is variant w.r.t. QueryLoop if any of its operands
- // are variant.
- for (unsigned i = 0, e = getNumOperands(); i != e; ++i)
- if (!getOperand(i)->isLoopInvariant(QueryLoop))
- return false;
-
- // Otherwise it's loop-invariant.
- return true;
-}
-
-bool
-SCEVAddRecExpr::dominates(BasicBlock *BB, DominatorTree *DT) const {
- return DT->dominates(L->getHeader(), BB) &&
- SCEVNAryExpr::dominates(BB, DT);
-}
-
-bool
-SCEVAddRecExpr::properlyDominates(BasicBlock *BB, DominatorTree *DT) const {
- // This uses a "dominates" query instead of "properly dominates" query because
- // the instruction which produces the addrec's value is a PHI, and a PHI
- // effectively properly dominates its entire containing block.
- return DT->dominates(L->getHeader(), BB) &&
- SCEVNAryExpr::properlyDominates(BB, DT);
-}
-
-void SCEVAddRecExpr::print(raw_ostream &OS) const {
- OS << "{" << *Operands[0];
- for (unsigned i = 1, e = NumOperands; i != e; ++i)
- OS << ",+," << *Operands[i];
- OS << "}<";
- WriteAsOperand(OS, L->getHeader(), /*PrintType=*/false);
- OS << ">";
-}
-
void SCEVUnknown::deleted() {
- // Clear this SCEVUnknown from ValuesAtScopes.
- SE->ValuesAtScopes.erase(this);
+ // Clear this SCEVUnknown from various maps.
+ SE->forgetMemoizedResults(this);
// Remove this SCEVUnknown from the uniquing map.
SE->UniqueSCEVs.RemoveNode(this);
@@ -349,8 +342,8 @@
}
void SCEVUnknown::allUsesReplacedWith(Value *New) {
- // Clear this SCEVUnknown from ValuesAtScopes.
- SE->ValuesAtScopes.erase(this);
+ // Clear this SCEVUnknown from various maps.
+ SE->forgetMemoizedResults(this);
// Remove this SCEVUnknown from the uniquing map.
SE->UniqueSCEVs.RemoveNode(this);
@@ -361,33 +354,7 @@
setValPtr(New);
}
-bool SCEVUnknown::isLoopInvariant(const Loop *L) const {
- // All non-instruction values are loop invariant. All instructions are loop
- // invariant if they are not contained in the specified loop.
- // Instructions are never considered invariant in the function body
- // (null loop) because they are defined within the "loop".
- if (Instruction *I = dyn_cast<Instruction>(getValue()))
- return L && !L->contains(I);
- return true;
-}
-
-bool SCEVUnknown::dominates(BasicBlock *BB, DominatorTree *DT) const {
- if (Instruction *I = dyn_cast<Instruction>(getValue()))
- return DT->dominates(I->getParent(), BB);
- return true;
-}
-
-bool SCEVUnknown::properlyDominates(BasicBlock *BB, DominatorTree *DT) const {
- if (Instruction *I = dyn_cast<Instruction>(getValue()))
- return DT->properlyDominates(I->getParent(), BB);
- return true;
-}
-
-const Type *SCEVUnknown::getType() const {
- return getValue()->getType();
-}
-
-bool SCEVUnknown::isSizeOf(const Type *&AllocTy) const {
+bool SCEVUnknown::isSizeOf(Type *&AllocTy) const {
if (ConstantExpr *VCE = dyn_cast<ConstantExpr>(getValue()))
if (VCE->getOpcode() == Instruction::PtrToInt)
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(VCE->getOperand(0)))
@@ -404,15 +371,15 @@
return false;
}
-bool SCEVUnknown::isAlignOf(const Type *&AllocTy) const {
+bool SCEVUnknown::isAlignOf(Type *&AllocTy) const {
if (ConstantExpr *VCE = dyn_cast<ConstantExpr>(getValue()))
if (VCE->getOpcode() == Instruction::PtrToInt)
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(VCE->getOperand(0)))
if (CE->getOpcode() == Instruction::GetElementPtr &&
CE->getOperand(0)->isNullValue()) {
- const Type *Ty =
+ Type *Ty =
cast<PointerType>(CE->getOperand(0)->getType())->getElementType();
- if (const StructType *STy = dyn_cast<StructType>(Ty))
+ if (StructType *STy = dyn_cast<StructType>(Ty))
if (!STy->isPacked() &&
CE->getNumOperands() == 3 &&
CE->getOperand(1)->isNullValue()) {
@@ -429,7 +396,7 @@
return false;
}
-bool SCEVUnknown::isOffsetOf(const Type *&CTy, Constant *&FieldNo) const {
+bool SCEVUnknown::isOffsetOf(Type *&CTy, Constant *&FieldNo) const {
if (ConstantExpr *VCE = dyn_cast<ConstantExpr>(getValue()))
if (VCE->getOpcode() == Instruction::PtrToInt)
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(VCE->getOperand(0)))
@@ -437,7 +404,7 @@
CE->getNumOperands() == 3 &&
CE->getOperand(0)->isNullValue() &&
CE->getOperand(1)->isNullValue()) {
- const Type *Ty =
+ Type *Ty =
cast<PointerType>(CE->getOperand(0)->getType())->getElementType();
// Ignore vector types here so that ScalarEvolutionExpander doesn't
// emit getelementptrs that index into vectors.
@@ -451,197 +418,180 @@
return false;
}
-void SCEVUnknown::print(raw_ostream &OS) const {
- const Type *AllocTy;
- if (isSizeOf(AllocTy)) {
- OS << "sizeof(" << *AllocTy << ")";
- return;
- }
- if (isAlignOf(AllocTy)) {
- OS << "alignof(" << *AllocTy << ")";
- return;
- }
-
- const Type *CTy;
- Constant *FieldNo;
- if (isOffsetOf(CTy, FieldNo)) {
- OS << "offsetof(" << *CTy << ", ";
- WriteAsOperand(OS, FieldNo, false);
- OS << ")";
- return;
- }
-
- // Otherwise just print it normally.
- WriteAsOperand(OS, getValue(), false);
-}
-
//===----------------------------------------------------------------------===//
// SCEV Utilities
//===----------------------------------------------------------------------===//
-static bool CompareTypes(const Type *A, const Type *B) {
- if (A->getTypeID() != B->getTypeID())
- return A->getTypeID() < B->getTypeID();
- if (const IntegerType *AI = dyn_cast<IntegerType>(A)) {
- const IntegerType *BI = cast<IntegerType>(B);
- return AI->getBitWidth() < BI->getBitWidth();
- }
- if (const PointerType *AI = dyn_cast<PointerType>(A)) {
- const PointerType *BI = cast<PointerType>(B);
- return CompareTypes(AI->getElementType(), BI->getElementType());
- }
- if (const ArrayType *AI = dyn_cast<ArrayType>(A)) {
- const ArrayType *BI = cast<ArrayType>(B);
- if (AI->getNumElements() != BI->getNumElements())
- return AI->getNumElements() < BI->getNumElements();
- return CompareTypes(AI->getElementType(), BI->getElementType());
- }
- if (const VectorType *AI = dyn_cast<VectorType>(A)) {
- const VectorType *BI = cast<VectorType>(B);
- if (AI->getNumElements() != BI->getNumElements())
- return AI->getNumElements() < BI->getNumElements();
- return CompareTypes(AI->getElementType(), BI->getElementType());
- }
- if (const StructType *AI = dyn_cast<StructType>(A)) {
- const StructType *BI = cast<StructType>(B);
- if (AI->getNumElements() != BI->getNumElements())
- return AI->getNumElements() < BI->getNumElements();
- for (unsigned i = 0, e = AI->getNumElements(); i != e; ++i)
- if (CompareTypes(AI->getElementType(i), BI->getElementType(i)) ||
- CompareTypes(BI->getElementType(i), AI->getElementType(i)))
- return CompareTypes(AI->getElementType(i), BI->getElementType(i));
- }
- return false;
-}
-
namespace {
/// SCEVComplexityCompare - Return true if the complexity of the LHS is less
/// than the complexity of the RHS. This comparator is used to canonicalize
/// expressions.
class SCEVComplexityCompare {
- const LoopInfo *LI;
+ const LoopInfo *const LI;
public:
explicit SCEVComplexityCompare(const LoopInfo *li) : LI(li) {}
+ // Return true or false if LHS is less than, or at least RHS, respectively.
bool operator()(const SCEV *LHS, const SCEV *RHS) const {
+ return compare(LHS, RHS) < 0;
+ }
+
+ // Return negative, zero, or positive, if LHS is less than, equal to, or
+ // greater than RHS, respectively. A three-way result allows recursive
+ // comparisons to be more efficient.
+ int compare(const SCEV *LHS, const SCEV *RHS) const {
// Fast-path: SCEVs are uniqued so we can do a quick equality check.
if (LHS == RHS)
- return false;
+ return 0;
// Primarily, sort the SCEVs by their getSCEVType().
unsigned LType = LHS->getSCEVType(), RType = RHS->getSCEVType();
if (LType != RType)
- return LType < RType;
+ return (int)LType - (int)RType;
// Aside from the getSCEVType() ordering, the particular ordering
// isn't very important except that it's beneficial to be consistent,
// so that (a + b) and (b + a) don't end up as different expressions.
-
- // Sort SCEVUnknown values with some loose heuristics. TODO: This is
- // not as complete as it could be.
- if (const SCEVUnknown *LU = dyn_cast<SCEVUnknown>(LHS)) {
+ switch (LType) {
+ case scUnknown: {
+ const SCEVUnknown *LU = cast<SCEVUnknown>(LHS);
const SCEVUnknown *RU = cast<SCEVUnknown>(RHS);
+ // Sort SCEVUnknown values with some loose heuristics. TODO: This is
+ // not as complete as it could be.
+ const Value *LV = LU->getValue(), *RV = RU->getValue();
+
// Order pointer values after integer values. This helps SCEVExpander
// form GEPs.
- bool LIsPointer = LU->getType()->isPointerTy(),
- RIsPointer = RU->getType()->isPointerTy();
+ bool LIsPointer = LV->getType()->isPointerTy(),
+ RIsPointer = RV->getType()->isPointerTy();
if (LIsPointer != RIsPointer)
- return RIsPointer;
+ return (int)LIsPointer - (int)RIsPointer;
// Compare getValueID values.
- unsigned LID = LU->getValue()->getValueID(),
- RID = RU->getValue()->getValueID();
+ unsigned LID = LV->getValueID(),
+ RID = RV->getValueID();
if (LID != RID)
- return LID < RID;
+ return (int)LID - (int)RID;
// Sort arguments by their position.
- if (const Argument *LA = dyn_cast<Argument>(LU->getValue())) {
- const Argument *RA = cast<Argument>(RU->getValue());
- return LA->getArgNo() < RA->getArgNo();
+ if (const Argument *LA = dyn_cast<Argument>(LV)) {
+ const Argument *RA = cast<Argument>(RV);
+ unsigned LArgNo = LA->getArgNo(), RArgNo = RA->getArgNo();
+ return (int)LArgNo - (int)RArgNo;
}
- // For instructions, compare their loop depth, and their opcode.
- // This is pretty loose.
- if (const Instruction *LV = dyn_cast<Instruction>(LU->getValue())) {
- const Instruction *RV = cast<Instruction>(RU->getValue());
+ // For instructions, compare their loop depth, and their operand
+ // count. This is pretty loose.
+ if (const Instruction *LInst = dyn_cast<Instruction>(LV)) {
+ const Instruction *RInst = cast<Instruction>(RV);
// Compare loop depths.
- unsigned LDepth = LI->getLoopDepth(LV->getParent()),
- RDepth = LI->getLoopDepth(RV->getParent());
- if (LDepth != RDepth)
- return LDepth < RDepth;
+ const BasicBlock *LParent = LInst->getParent(),
+ *RParent = RInst->getParent();
+ if (LParent != RParent) {
+ unsigned LDepth = LI->getLoopDepth(LParent),
+ RDepth = LI->getLoopDepth(RParent);
+ if (LDepth != RDepth)
+ return (int)LDepth - (int)RDepth;
+ }
// Compare the number of operands.
- unsigned LNumOps = LV->getNumOperands(),
- RNumOps = RV->getNumOperands();
- if (LNumOps != RNumOps)
- return LNumOps < RNumOps;
+ unsigned LNumOps = LInst->getNumOperands(),
+ RNumOps = RInst->getNumOperands();
+ return (int)LNumOps - (int)RNumOps;
}
- return false;
+ return 0;
}
- // Compare constant values.
- if (const SCEVConstant *LC = dyn_cast<SCEVConstant>(LHS)) {
+ case scConstant: {
+ const SCEVConstant *LC = cast<SCEVConstant>(LHS);
const SCEVConstant *RC = cast<SCEVConstant>(RHS);
- const ConstantInt *LCC = LC->getValue();
- const ConstantInt *RCC = RC->getValue();
- unsigned LBitWidth = LCC->getBitWidth(), RBitWidth = RCC->getBitWidth();
+
+ // Compare constant values.
+ const APInt &LA = LC->getValue()->getValue();
+ const APInt &RA = RC->getValue()->getValue();
+ unsigned LBitWidth = LA.getBitWidth(), RBitWidth = RA.getBitWidth();
if (LBitWidth != RBitWidth)
- return LBitWidth < RBitWidth;
- return LCC->getValue().ult(RCC->getValue());
+ return (int)LBitWidth - (int)RBitWidth;
+ return LA.ult(RA) ? -1 : 1;
}
- // Compare addrec loop depths.
- if (const SCEVAddRecExpr *LA = dyn_cast<SCEVAddRecExpr>(LHS)) {
+ case scAddRecExpr: {
+ const SCEVAddRecExpr *LA = cast<SCEVAddRecExpr>(LHS);
const SCEVAddRecExpr *RA = cast<SCEVAddRecExpr>(RHS);
- unsigned LDepth = LA->getLoop()->getLoopDepth(),
- RDepth = RA->getLoop()->getLoopDepth();
- if (LDepth != RDepth)
- return LDepth < RDepth;
+
+ // Compare addrec loop depths.
+ const Loop *LLoop = LA->getLoop(), *RLoop = RA->getLoop();
+ if (LLoop != RLoop) {
+ unsigned LDepth = LLoop->getLoopDepth(),
+ RDepth = RLoop->getLoopDepth();
+ if (LDepth != RDepth)
+ return (int)LDepth - (int)RDepth;
+ }
+
+ // Addrec complexity grows with operand count.
+ unsigned LNumOps = LA->getNumOperands(), RNumOps = RA->getNumOperands();
+ if (LNumOps != RNumOps)
+ return (int)LNumOps - (int)RNumOps;
+
+ // Lexicographically compare.
+ for (unsigned i = 0; i != LNumOps; ++i) {
+ long X = compare(LA->getOperand(i), RA->getOperand(i));
+ if (X != 0)
+ return X;
+ }
+
+ return 0;
}
- // Lexicographically compare n-ary expressions.
- if (const SCEVNAryExpr *LC = dyn_cast<SCEVNAryExpr>(LHS)) {
+ case scAddExpr:
+ case scMulExpr:
+ case scSMaxExpr:
+ case scUMaxExpr: {
+ const SCEVNAryExpr *LC = cast<SCEVNAryExpr>(LHS);
const SCEVNAryExpr *RC = cast<SCEVNAryExpr>(RHS);
+
+ // Lexicographically compare n-ary expressions.
unsigned LNumOps = LC->getNumOperands(), RNumOps = RC->getNumOperands();
for (unsigned i = 0; i != LNumOps; ++i) {
if (i >= RNumOps)
- return false;
- const SCEV *LOp = LC->getOperand(i), *ROp = RC->getOperand(i);
- if (operator()(LOp, ROp))
- return true;
- if (operator()(ROp, LOp))
- return false;
+ return 1;
+ long X = compare(LC->getOperand(i), RC->getOperand(i));
+ if (X != 0)
+ return X;
}
- return LNumOps < RNumOps;
+ return (int)LNumOps - (int)RNumOps;
}
- // Lexicographically compare udiv expressions.
- if (const SCEVUDivExpr *LC = dyn_cast<SCEVUDivExpr>(LHS)) {
+ case scUDivExpr: {
+ const SCEVUDivExpr *LC = cast<SCEVUDivExpr>(LHS);
const SCEVUDivExpr *RC = cast<SCEVUDivExpr>(RHS);
- const SCEV *LL = LC->getLHS(), *LR = LC->getRHS(),
- *RL = RC->getLHS(), *RR = RC->getRHS();
- if (operator()(LL, RL))
- return true;
- if (operator()(RL, LL))
- return false;
- if (operator()(LR, RR))
- return true;
- if (operator()(RR, LR))
- return false;
- return false;
+
+ // Lexicographically compare udiv expressions.
+ long X = compare(LC->getLHS(), RC->getLHS());
+ if (X != 0)
+ return X;
+ return compare(LC->getRHS(), RC->getRHS());
}
- // Compare cast expressions by operand.
- if (const SCEVCastExpr *LC = dyn_cast<SCEVCastExpr>(LHS)) {
+ case scTruncate:
+ case scZeroExtend:
+ case scSignExtend: {
+ const SCEVCastExpr *LC = cast<SCEVCastExpr>(LHS);
const SCEVCastExpr *RC = cast<SCEVCastExpr>(RHS);
- return operator()(LC->getOperand(), RC->getOperand());
+
+ // Compare cast expressions by operand.
+ return compare(LC->getOperand(), RC->getOperand());
+ }
+
+ default:
+ break;
}
llvm_unreachable("Unknown SCEV kind!");
- return false;
+ return 0;
}
};
}
@@ -662,8 +612,9 @@
if (Ops.size() == 2) {
// This is the common case, which also happens to be trivially simple.
// Special case it.
- if (SCEVComplexityCompare(LI)(Ops[1], Ops[0]))
- std::swap(Ops[0], Ops[1]);
+ const SCEV *&LHS = Ops[0], *&RHS = Ops[1];
+ if (SCEVComplexityCompare(LI)(RHS, LHS))
+ std::swap(LHS, RHS);
return;
}
@@ -701,7 +652,7 @@
/// Assume, K > 0.
static const SCEV *BinomialCoefficient(const SCEV *It, unsigned K,
ScalarEvolution &SE,
- const Type* ResultTy) {
+ Type *ResultTy) {
// Handle the simplest case efficiently.
if (K == 1)
return SE.getTruncateOrZeroExtend(It, ResultTy);
@@ -791,7 +742,7 @@
MultiplyFactor = MultiplyFactor.trunc(W);
// Calculate the product, at width T+W
- const IntegerType *CalculationTy = IntegerType::get(SE.getContext(),
+ IntegerType *CalculationTy = IntegerType::get(SE.getContext(),
CalculationBits);
const SCEV *Dividend = SE.getTruncateOrZeroExtend(It, CalculationTy);
for (unsigned i = 1; i != K; ++i) {
@@ -839,7 +790,7 @@
//===----------------------------------------------------------------------===//
const SCEV *ScalarEvolution::getTruncateExpr(const SCEV *Op,
- const Type *Ty) {
+ Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) > getTypeSizeInBits(Ty) &&
"This is not a truncating conversion!");
assert(isSCEVable(Ty) &&
@@ -871,12 +822,42 @@
if (const SCEVZeroExtendExpr *SZ = dyn_cast<SCEVZeroExtendExpr>(Op))
return getTruncateOrZeroExtend(SZ->getOperand(), Ty);
+ // trunc(x1+x2+...+xN) --> trunc(x1)+trunc(x2)+...+trunc(xN) if we can
+ // eliminate all the truncates.
+ if (const SCEVAddExpr *SA = dyn_cast<SCEVAddExpr>(Op)) {
+ SmallVector<const SCEV *, 4> Operands;
+ bool hasTrunc = false;
+ for (unsigned i = 0, e = SA->getNumOperands(); i != e && !hasTrunc; ++i) {
+ const SCEV *S = getTruncateExpr(SA->getOperand(i), Ty);
+ hasTrunc = isa<SCEVTruncateExpr>(S);
+ Operands.push_back(S);
+ }
+ if (!hasTrunc)
+ return getAddExpr(Operands);
+ UniqueSCEVs.FindNodeOrInsertPos(ID, IP); // Mutates IP, returns NULL.
+ }
+
+ // trunc(x1*x2*...*xN) --> trunc(x1)*trunc(x2)*...*trunc(xN) if we can
+ // eliminate all the truncates.
+ if (const SCEVMulExpr *SM = dyn_cast<SCEVMulExpr>(Op)) {
+ SmallVector<const SCEV *, 4> Operands;
+ bool hasTrunc = false;
+ for (unsigned i = 0, e = SM->getNumOperands(); i != e && !hasTrunc; ++i) {
+ const SCEV *S = getTruncateExpr(SM->getOperand(i), Ty);
+ hasTrunc = isa<SCEVTruncateExpr>(S);
+ Operands.push_back(S);
+ }
+ if (!hasTrunc)
+ return getMulExpr(Operands);
+ UniqueSCEVs.FindNodeOrInsertPos(ID, IP); // Mutates IP, returns NULL.
+ }
+
// If the input value is a chrec scev, truncate the chrec's operands.
if (const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(Op)) {
SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i)
Operands.push_back(getTruncateExpr(AddRec->getOperand(i), Ty));
- return getAddRecExpr(Operands, AddRec->getLoop());
+ return getAddRecExpr(Operands, AddRec->getLoop(), SCEV::FlagAnyWrap);
}
// As a special case, fold trunc(undef) to undef. We don't want to
@@ -896,7 +877,7 @@
}
const SCEV *ScalarEvolution::getZeroExtendExpr(const SCEV *Op,
- const Type *Ty) {
+ Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
assert(isSCEVable(Ty) &&
@@ -922,6 +903,19 @@
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
+ // zext(trunc(x)) --> zext(x) or x or trunc(x)
+ if (const SCEVTruncateExpr *ST = dyn_cast<SCEVTruncateExpr>(Op)) {
+ // It's possible the bits taken off by the truncate were all zero bits. If
+ // so, we should be able to simplify this further.
+ const SCEV *X = ST->getOperand();
+ ConstantRange CR = getUnsignedRange(X);
+ unsigned TruncBits = getTypeSizeInBits(ST->getType());
+ unsigned NewBits = getTypeSizeInBits(Ty);
+ if (CR.truncate(TruncBits).zeroExtend(NewBits).contains(
+ CR.zextOrTrunc(NewBits)))
+ return getTruncateOrZeroExtend(X, Ty);
+ }
+
// If the input value is a chrec scev, and we can prove that the value
// did not overflow the old, smaller, value, we can zero extend all of the
// operands (often constants). This allows analysis of something like
@@ -935,10 +929,10 @@
// If we have special knowledge that this addrec won't overflow,
// we don't need to do any further analysis.
- if (AR->hasNoUnsignedWrap())
+ if (AR->getNoWrapFlags(SCEV::FlagNUW))
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getZeroExtendExpr(Step, Ty),
- L);
+ L, AR->getNoWrapFlags());
// Check whether the backedge-taken count is SCEVCouldNotCompute.
// Note that this serves two purposes: It filters out loops that are
@@ -960,7 +954,7 @@
const SCEV *RecastedMaxBECount =
getTruncateOrZeroExtend(CastedMaxBECount, MaxBECount->getType());
if (MaxBECount == RecastedMaxBECount) {
- const Type *WideTy = IntegerType::get(getContext(), BitWidth * 2);
+ Type *WideTy = IntegerType::get(getContext(), BitWidth * 2);
// Check whether Start+Step*MaxBECount has no unsigned overflow.
const SCEV *ZMul = getMulExpr(CastedMaxBECount, Step);
const SCEV *Add = getAddExpr(Start, ZMul);
@@ -968,12 +962,14 @@
getAddExpr(getZeroExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getZeroExtendExpr(Step, WideTy)));
- if (getZeroExtendExpr(Add, WideTy) == OperandExtendedAdd)
+ if (getZeroExtendExpr(Add, WideTy) == OperandExtendedAdd) {
+ // Cache knowledge of AR NUW, which is propagated to this AddRec.
+ const_cast<SCEVAddRecExpr *>(AR)->setNoWrapFlags(SCEV::FlagNUW);
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getZeroExtendExpr(Step, Ty),
- L);
-
+ L, AR->getNoWrapFlags());
+ }
// Similar to above, only this time treat the step value as signed.
// This covers loops that count down.
const SCEV *SMul = getMulExpr(CastedMaxBECount, Step);
@@ -982,11 +978,15 @@
getAddExpr(getZeroExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getSignExtendExpr(Step, WideTy)));
- if (getZeroExtendExpr(Add, WideTy) == OperandExtendedAdd)
+ if (getZeroExtendExpr(Add, WideTy) == OperandExtendedAdd) {
+ // Cache knowledge of AR NW, which is propagated to this AddRec.
+ // Negative step causes unsigned wrap, but it still can't self-wrap.
+ const_cast<SCEVAddRecExpr *>(AR)->setNoWrapFlags(SCEV::FlagNW);
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
- L);
+ L, AR->getNoWrapFlags());
+ }
}
// If the backedge is guarded by a comparison with the pre-inc value
@@ -999,22 +999,29 @@
if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_ULT, AR, N) ||
(isLoopEntryGuardedByCond(L, ICmpInst::ICMP_ULT, Start, N) &&
isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_ULT,
- AR->getPostIncExpr(*this), N)))
+ AR->getPostIncExpr(*this), N))) {
+ // Cache knowledge of AR NUW, which is propagated to this AddRec.
+ const_cast<SCEVAddRecExpr *>(AR)->setNoWrapFlags(SCEV::FlagNUW);
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getZeroExtendExpr(Step, Ty),
- L);
+ L, AR->getNoWrapFlags());
+ }
} else if (isKnownNegative(Step)) {
const SCEV *N = getConstant(APInt::getMaxValue(BitWidth) -
getSignedRange(Step).getSignedMin());
if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT, AR, N) ||
(isLoopEntryGuardedByCond(L, ICmpInst::ICMP_UGT, Start, N) &&
isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT,
- AR->getPostIncExpr(*this), N)))
+ AR->getPostIncExpr(*this), N))) {
+ // Cache knowledge of AR NW, which is propagated to this AddRec.
+ // Negative step causes unsigned wrap, but it still can't self-wrap.
+ const_cast<SCEVAddRecExpr *>(AR)->setNoWrapFlags(SCEV::FlagNW);
// Return the expression with the addrec on the outside.
return getAddRecExpr(getZeroExtendExpr(Start, Ty),
getSignExtendExpr(Step, Ty),
- L);
+ L, AR->getNoWrapFlags());
+ }
}
}
}
@@ -1028,8 +1035,107 @@
return S;
}
+// Get the limit of a recurrence such that incrementing by Step cannot cause
+// signed overflow as long as the value of the recurrence within the loop does
+// not exceed this limit before incrementing.
+static const SCEV *getOverflowLimitForStep(const SCEV *Step,
+ ICmpInst::Predicate *Pred,
+ ScalarEvolution *SE) {
+ unsigned BitWidth = SE->getTypeSizeInBits(Step->getType());
+ if (SE->isKnownPositive(Step)) {
+ *Pred = ICmpInst::ICMP_SLT;
+ return SE->getConstant(APInt::getSignedMinValue(BitWidth) -
+ SE->getSignedRange(Step).getSignedMax());
+ }
+ if (SE->isKnownNegative(Step)) {
+ *Pred = ICmpInst::ICMP_SGT;
+ return SE->getConstant(APInt::getSignedMaxValue(BitWidth) -
+ SE->getSignedRange(Step).getSignedMin());
+ }
+ return 0;
+}
+
+// The recurrence AR has been shown to have no signed wrap. Typically, if we can
+// prove NSW for AR, then we can just as easily prove NSW for its preincrement
+// or postincrement sibling. This allows normalizing a sign extended AddRec as
+// such: {sext(Step + Start),+,Step} => {(Step + sext(Start),+,Step} As a
+// result, the expression "Step + sext(PreIncAR)" is congruent with
+// "sext(PostIncAR)"
+static const SCEV *getPreStartForSignExtend(const SCEVAddRecExpr *AR,
+ Type *Ty,
+ ScalarEvolution *SE) {
+ const Loop *L = AR->getLoop();
+ const SCEV *Start = AR->getStart();
+ const SCEV *Step = AR->getStepRecurrence(*SE);
+
+ // Check for a simple looking step prior to loop entry.
+ const SCEVAddExpr *SA = dyn_cast<SCEVAddExpr>(Start);
+ if (!SA)
+ return 0;
+
+ // Create an AddExpr for "PreStart" after subtracting Step. Full SCEV
+ // subtraction is expensive. For this purpose, perform a quick and dirty
+ // difference, by checking for Step in the operand list.
+ SmallVector<const SCEV *, 4> DiffOps;
+ for (SCEVAddExpr::op_iterator I = SA->op_begin(), E = SA->op_end();
+ I != E; ++I) {
+ if (*I != Step)
+ DiffOps.push_back(*I);
+ }
+ if (DiffOps.size() == SA->getNumOperands())
+ return 0;
+
+ // This is a postinc AR. Check for overflow on the preinc recurrence using the
+ // same three conditions that getSignExtendedExpr checks.
+
+ // 1. NSW flags on the step increment.
+ const SCEV *PreStart = SE->getAddExpr(DiffOps, SA->getNoWrapFlags());
+ const SCEVAddRecExpr *PreAR = dyn_cast<SCEVAddRecExpr>(
+ SE->getAddRecExpr(PreStart, Step, L, SCEV::FlagAnyWrap));
+
+ if (PreAR && PreAR->getNoWrapFlags(SCEV::FlagNSW))
+ return PreStart;
+
+ // 2. Direct overflow check on the step operation's expression.
+ unsigned BitWidth = SE->getTypeSizeInBits(AR->getType());
+ Type *WideTy = IntegerType::get(SE->getContext(), BitWidth * 2);
+ const SCEV *OperandExtendedStart =
+ SE->getAddExpr(SE->getSignExtendExpr(PreStart, WideTy),
+ SE->getSignExtendExpr(Step, WideTy));
+ if (SE->getSignExtendExpr(Start, WideTy) == OperandExtendedStart) {
+ // Cache knowledge of PreAR NSW.
+ if (PreAR)
+ const_cast<SCEVAddRecExpr *>(PreAR)->setNoWrapFlags(SCEV::FlagNSW);
+ // FIXME: this optimization needs a unit test
+ DEBUG(dbgs() << "SCEV: untested prestart overflow check\n");
+ return PreStart;
+ }
+
+ // 3. Loop precondition.
+ ICmpInst::Predicate Pred;
+ const SCEV *OverflowLimit = getOverflowLimitForStep(Step, &Pred, SE);
+
+ if (OverflowLimit &&
+ SE->isLoopEntryGuardedByCond(L, Pred, PreStart, OverflowLimit)) {
+ return PreStart;
+ }
+ return 0;
+}
+
+// Get the normalized sign-extended expression for this AddRec's Start.
+static const SCEV *getSignExtendAddRecStart(const SCEVAddRecExpr *AR,
+ Type *Ty,
+ ScalarEvolution *SE) {
+ const SCEV *PreStart = getPreStartForSignExtend(AR, Ty, SE);
+ if (!PreStart)
+ return SE->getSignExtendExpr(AR->getStart(), Ty);
+
+ return SE->getAddExpr(SE->getSignExtendExpr(AR->getStepRecurrence(*SE), Ty),
+ SE->getSignExtendExpr(PreStart, Ty));
+}
+
const SCEV *ScalarEvolution::getSignExtendExpr(const SCEV *Op,
- const Type *Ty) {
+ Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
assert(isSCEVable(Ty) &&
@@ -1046,6 +1152,10 @@
if (const SCEVSignExtendExpr *SS = dyn_cast<SCEVSignExtendExpr>(Op))
return getSignExtendExpr(SS->getOperand(), Ty);
+ // sext(zext(x)) --> zext(x)
+ if (const SCEVZeroExtendExpr *SZ = dyn_cast<SCEVZeroExtendExpr>(Op))
+ return getZeroExtendExpr(SZ->getOperand(), Ty);
+
// Before doing any expensive analysis, check to see if we've already
// computed a SCEV for this Op and Ty.
FoldingSetNodeID ID;
@@ -1055,6 +1165,23 @@
void *IP = 0;
if (const SCEV *S = UniqueSCEVs.FindNodeOrInsertPos(ID, IP)) return S;
+ // If the input value is provably positive, build a zext instead.
+ if (isKnownNonNegative(Op))
+ return getZeroExtendExpr(Op, Ty);
+
+ // sext(trunc(x)) --> sext(x) or x or trunc(x)
+ if (const SCEVTruncateExpr *ST = dyn_cast<SCEVTruncateExpr>(Op)) {
+ // It's possible the bits taken off by the truncate were all sign bits. If
+ // so, we should be able to simplify this further.
+ const SCEV *X = ST->getOperand();
+ ConstantRange CR = getSignedRange(X);
+ unsigned TruncBits = getTypeSizeInBits(ST->getType());
+ unsigned NewBits = getTypeSizeInBits(Ty);
+ if (CR.truncate(TruncBits).signExtend(NewBits).contains(
+ CR.sextOrTrunc(NewBits)))
+ return getTruncateOrSignExtend(X, Ty);
+ }
+
// If the input value is a chrec scev, and we can prove that the value
// did not overflow the old, smaller, value, we can sign extend all of the
// operands (often constants). This allows analysis of something like
@@ -1068,10 +1195,10 @@
// If we have special knowledge that this addrec won't overflow,
// we don't need to do any further analysis.
- if (AR->hasNoSignedWrap())
- return getAddRecExpr(getSignExtendExpr(Start, Ty),
+ if (AR->getNoWrapFlags(SCEV::FlagNSW))
+ return getAddRecExpr(getSignExtendAddRecStart(AR, Ty, this),
getSignExtendExpr(Step, Ty),
- L);
+ L, SCEV::FlagNSW);
// Check whether the backedge-taken count is SCEVCouldNotCompute.
// Note that this serves two purposes: It filters out loops that are
@@ -1093,7 +1220,7 @@
const SCEV *RecastedMaxBECount =
getTruncateOrZeroExtend(CastedMaxBECount, MaxBECount->getType());
if (MaxBECount == RecastedMaxBECount) {
- const Type *WideTy = IntegerType::get(getContext(), BitWidth * 2);
+ Type *WideTy = IntegerType::get(getContext(), BitWidth * 2);
// Check whether Start+Step*MaxBECount has no signed overflow.
const SCEV *SMul = getMulExpr(CastedMaxBECount, Step);
const SCEV *Add = getAddExpr(Start, SMul);
@@ -1101,12 +1228,14 @@
getAddExpr(getSignExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getSignExtendExpr(Step, WideTy)));
- if (getSignExtendExpr(Add, WideTy) == OperandExtendedAdd)
+ if (getSignExtendExpr(Add, WideTy) == OperandExtendedAdd) {
+ // Cache knowledge of AR NSW, which is propagated to this AddRec.
+ const_cast<SCEVAddRecExpr *>(AR)->setNoWrapFlags(SCEV::FlagNSW);
// Return the expression with the addrec on the outside.
- return getAddRecExpr(getSignExtendExpr(Start, Ty),
+ return getAddRecExpr(getSignExtendAddRecStart(AR, Ty, this),
getSignExtendExpr(Step, Ty),
- L);
-
+ L, AR->getNoWrapFlags());
+ }
// Similar to above, only this time treat the step value as unsigned.
// This covers loops that count up with an unsigned step.
const SCEV *UMul = getMulExpr(CastedMaxBECount, Step);
@@ -1115,39 +1244,32 @@
getAddExpr(getSignExtendExpr(Start, WideTy),
getMulExpr(getZeroExtendExpr(CastedMaxBECount, WideTy),
getZeroExtendExpr(Step, WideTy)));
- if (getSignExtendExpr(Add, WideTy) == OperandExtendedAdd)
+ if (getSignExtendExpr(Add, WideTy) == OperandExtendedAdd) {
+ // Cache knowledge of AR NSW, which is propagated to this AddRec.
+ const_cast<SCEVAddRecExpr *>(AR)->setNoWrapFlags(SCEV::FlagNSW);
// Return the expression with the addrec on the outside.
- return getAddRecExpr(getSignExtendExpr(Start, Ty),
+ return getAddRecExpr(getSignExtendAddRecStart(AR, Ty, this),
getZeroExtendExpr(Step, Ty),
- L);
+ L, AR->getNoWrapFlags());
+ }
}
// If the backedge is guarded by a comparison with the pre-inc value
// the addrec is safe. Also, if the entry is guarded by a comparison
// with the start value and the backedge is guarded by a comparison
// with the post-inc value, the addrec is safe.
- if (isKnownPositive(Step)) {
- const SCEV *N = getConstant(APInt::getSignedMinValue(BitWidth) -
- getSignedRange(Step).getSignedMax());
- if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SLT, AR, N) ||
- (isLoopEntryGuardedByCond(L, ICmpInst::ICMP_SLT, Start, N) &&
- isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SLT,
- AR->getPostIncExpr(*this), N)))
- // Return the expression with the addrec on the outside.
- return getAddRecExpr(getSignExtendExpr(Start, Ty),
- getSignExtendExpr(Step, Ty),
- L);
- } else if (isKnownNegative(Step)) {
- const SCEV *N = getConstant(APInt::getSignedMaxValue(BitWidth) -
- getSignedRange(Step).getSignedMin());
- if (isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SGT, AR, N) ||
- (isLoopEntryGuardedByCond(L, ICmpInst::ICMP_SGT, Start, N) &&
- isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_SGT,
- AR->getPostIncExpr(*this), N)))
- // Return the expression with the addrec on the outside.
- return getAddRecExpr(getSignExtendExpr(Start, Ty),
- getSignExtendExpr(Step, Ty),
- L);
+ ICmpInst::Predicate Pred;
+ const SCEV *OverflowLimit = getOverflowLimitForStep(Step, &Pred, this);
+ if (OverflowLimit &&
+ (isLoopBackedgeGuardedByCond(L, Pred, AR, OverflowLimit) ||
+ (isLoopEntryGuardedByCond(L, Pred, Start, OverflowLimit) &&
+ isLoopBackedgeGuardedByCond(L, Pred, AR->getPostIncExpr(*this),
+ OverflowLimit)))) {
+ // Cache knowledge of AR NSW, then propagate NSW to the wide AddRec.
+ const_cast<SCEVAddRecExpr *>(AR)->setNoWrapFlags(SCEV::FlagNSW);
+ return getAddRecExpr(getSignExtendAddRecStart(AR, Ty, this),
+ getSignExtendExpr(Step, Ty),
+ L, AR->getNoWrapFlags());
}
}
}
@@ -1165,7 +1287,7 @@
/// unspecified bits out to the given type.
///
const SCEV *ScalarEvolution::getAnyExtendExpr(const SCEV *Op,
- const Type *Ty) {
+ Type *Ty) {
assert(getTypeSizeInBits(Op->getType()) < getTypeSizeInBits(Ty) &&
"This is not an extending conversion!");
assert(isSCEVable(Ty) &&
@@ -1201,7 +1323,7 @@
for (SCEVAddRecExpr::op_iterator I = AR->op_begin(), E = AR->op_end();
I != E; ++I)
Ops.push_back(getAnyExtendExpr(*I, Ty));
- return getAddRecExpr(Ops, AR->getLoop());
+ return getAddRecExpr(Ops, AR->getLoop(), SCEV::FlagNW);
}
// As a special case, fold anyext(undef) to undef. We don't want to
@@ -1322,25 +1444,31 @@
/// getAddExpr - Get a canonical add expression, or something simpler if
/// possible.
const SCEV *ScalarEvolution::getAddExpr(SmallVectorImpl<const SCEV *> &Ops,
- bool HasNUW, bool HasNSW) {
+ SCEV::NoWrapFlags Flags) {
+ assert(!(Flags & ~(SCEV::FlagNUW | SCEV::FlagNSW)) &&
+ "only nuw or nsw allowed");
assert(!Ops.empty() && "Cannot get empty add!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
- const Type *ETy = getEffectiveSCEVType(Ops[0]->getType());
+ Type *ETy = getEffectiveSCEVType(Ops[0]->getType());
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
assert(getEffectiveSCEVType(Ops[i]->getType()) == ETy &&
"SCEVAddExpr operand types don't match!");
#endif
- // If HasNSW is true and all the operands are non-negative, infer HasNUW.
- if (!HasNUW && HasNSW) {
+ // If FlagNSW is true and all the operands are non-negative, infer FlagNUW.
+ // And vice-versa.
+ int SignOrUnsignMask = SCEV::FlagNUW | SCEV::FlagNSW;
+ SCEV::NoWrapFlags SignOrUnsignWrap = maskFlags(Flags, SignOrUnsignMask);
+ if (SignOrUnsignWrap && (SignOrUnsignWrap != SignOrUnsignMask)) {
bool All = true;
- for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (!isKnownNonNegative(Ops[i])) {
+ for (SmallVectorImpl<const SCEV *>::const_iterator I = Ops.begin(),
+ E = Ops.end(); I != E; ++I)
+ if (!isKnownNonNegative(*I)) {
All = false;
break;
}
- if (All) HasNUW = true;
+ if (All) Flags = setFlags(Flags, (SCEV::NoWrapFlags)SignOrUnsignMask);
}
// Sort by complexity, this groups all similar expression types together.
@@ -1369,22 +1497,29 @@
if (Ops.size() == 1) return Ops[0];
}
- // Okay, check to see if the same value occurs in the operand list twice. If
- // so, merge them together into an multiply expression. Since we sorted the
- // list, these values are required to be adjacent.
- const Type *Ty = Ops[0]->getType();
- for (unsigned i = 0, e = Ops.size()-1; i != e; ++i)
+ // Okay, check to see if the same value occurs in the operand list more than
+ // once. If so, merge them together into an multiply expression. Since we
+ // sorted the list, these values are required to be adjacent.
+ Type *Ty = Ops[0]->getType();
+ bool FoundMatch = false;
+ for (unsigned i = 0, e = Ops.size(); i != e-1; ++i)
if (Ops[i] == Ops[i+1]) { // X + Y + Y --> X + Y*2
- // Found a match, merge the two values into a multiply, and add any
- // remaining values to the result.
- const SCEV *Two = getConstant(Ty, 2);
- const SCEV *Mul = getMulExpr(Ops[i], Two);
- if (Ops.size() == 2)
+ // Scan ahead to count how many equal operands there are.
+ unsigned Count = 2;
+ while (i+Count != e && Ops[i+Count] == Ops[i])
+ ++Count;
+ // Merge the values into a multiply.
+ const SCEV *Scale = getConstant(Ty, Count);
+ const SCEV *Mul = getMulExpr(Scale, Ops[i]);
+ if (Ops.size() == Count)
return Mul;
- Ops.erase(Ops.begin()+i, Ops.begin()+i+2);
- Ops.push_back(Mul);
- return getAddExpr(Ops, HasNUW, HasNSW);
+ Ops[i] = Mul;
+ Ops.erase(Ops.begin()+i+1, Ops.begin()+i+Count);
+ --i; e -= Count - 1;
+ FoundMatch = true;
}
+ if (FoundMatch)
+ return getAddExpr(Ops, Flags);
// Check for truncates. If all the operands are truncated from the same
// type, see if factoring out the truncate would permit the result to be
@@ -1392,8 +1527,8 @@
// if the contents of the resulting outer trunc fold to something simple.
for (; Idx < Ops.size() && isa<SCEVTruncateExpr>(Ops[Idx]); ++Idx) {
const SCEVTruncateExpr *Trunc = cast<SCEVTruncateExpr>(Ops[Idx]);
- const Type *DstType = Trunc->getType();
- const Type *SrcType = Trunc->getOperand()->getType();
+ Type *DstType = Trunc->getType();
+ Type *SrcType = Trunc->getOperand()->getType();
SmallVector<const SCEV *, 8> LargeOps;
bool Ok = true;
// Check all the operands to see if they can be represented in the
@@ -1434,7 +1569,7 @@
}
if (Ok) {
// Evaluate the expression in the larger type.
- const SCEV *Fold = getAddExpr(LargeOps, HasNUW, HasNSW);
+ const SCEV *Fold = getAddExpr(LargeOps, Flags);
// If it folds to something simple, use it. Otherwise, don't.
if (isa<SCEVConstant>(Fold) || isa<SCEVUnknown>(Fold))
return getTruncateExpr(Fold, DstType);
@@ -1481,7 +1616,7 @@
// re-generate the operands list. Group the operands by constant scale,
// to avoid multiplying by the same constant scale multiple times.
std::map<APInt, SmallVector<const SCEV *, 4>, APIntCompare> MulOpLists;
- for (SmallVector<const SCEV *, 8>::iterator I = NewOps.begin(),
+ for (SmallVector<const SCEV *, 8>::const_iterator I = NewOps.begin(),
E = NewOps.end(); I != E; ++I)
MulOpLists[M.find(*I)->second].push_back(*I);
// Re-generate the operands list.
@@ -1508,20 +1643,23 @@
const SCEVMulExpr *Mul = cast<SCEVMulExpr>(Ops[Idx]);
for (unsigned MulOp = 0, e = Mul->getNumOperands(); MulOp != e; ++MulOp) {
const SCEV *MulOpSCEV = Mul->getOperand(MulOp);
+ if (isa<SCEVConstant>(MulOpSCEV))
+ continue;
for (unsigned AddOp = 0, e = Ops.size(); AddOp != e; ++AddOp)
- if (MulOpSCEV == Ops[AddOp] && !isa<SCEVConstant>(Ops[AddOp])) {
+ if (MulOpSCEV == Ops[AddOp]) {
// Fold W + X + (X * Y * Z) --> W + (X * ((Y*Z)+1))
const SCEV *InnerMul = Mul->getOperand(MulOp == 0);
if (Mul->getNumOperands() != 2) {
// If the multiply has more than two operands, we must get the
// Y*Z term.
- SmallVector<const SCEV *, 4> MulOps(Mul->op_begin(), Mul->op_end());
- MulOps.erase(MulOps.begin()+MulOp);
+ SmallVector<const SCEV *, 4> MulOps(Mul->op_begin(),
+ Mul->op_begin()+MulOp);
+ MulOps.append(Mul->op_begin()+MulOp+1, Mul->op_end());
InnerMul = getMulExpr(MulOps);
}
const SCEV *One = getConstant(Ty, 1);
- const SCEV *AddOne = getAddExpr(InnerMul, One);
- const SCEV *OuterMul = getMulExpr(AddOne, Ops[AddOp]);
+ const SCEV *AddOne = getAddExpr(One, InnerMul);
+ const SCEV *OuterMul = getMulExpr(AddOne, MulOpSCEV);
if (Ops.size() == 2) return OuterMul;
if (AddOp < Idx) {
Ops.erase(Ops.begin()+AddOp);
@@ -1548,15 +1686,15 @@
const SCEV *InnerMul1 = Mul->getOperand(MulOp == 0);
if (Mul->getNumOperands() != 2) {
SmallVector<const SCEV *, 4> MulOps(Mul->op_begin(),
- Mul->op_end());
- MulOps.erase(MulOps.begin()+MulOp);
+ Mul->op_begin()+MulOp);
+ MulOps.append(Mul->op_begin()+MulOp+1, Mul->op_end());
InnerMul1 = getMulExpr(MulOps);
}
const SCEV *InnerMul2 = OtherMul->getOperand(OMulOp == 0);
if (OtherMul->getNumOperands() != 2) {
SmallVector<const SCEV *, 4> MulOps(OtherMul->op_begin(),
- OtherMul->op_end());
- MulOps.erase(MulOps.begin()+OMulOp);
+ OtherMul->op_begin()+OMulOp);
+ MulOps.append(OtherMul->op_begin()+OMulOp+1, OtherMul->op_end());
InnerMul2 = getMulExpr(MulOps);
}
const SCEV *InnerMulSum = getAddExpr(InnerMul1,InnerMul2);
@@ -1585,7 +1723,7 @@
const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
const Loop *AddRecLoop = AddRec->getLoop();
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (Ops[i]->isLoopInvariant(AddRecLoop)) {
+ if (isLoopInvariant(Ops[i], AddRecLoop)) {
LIOps.push_back(Ops[i]);
Ops.erase(Ops.begin()+i);
--i; --e;
@@ -1602,14 +1740,14 @@
// Build the new addrec. Propagate the NUW and NSW flags if both the
// outer add and the inner addrec are guaranteed to have no overflow.
- const SCEV *NewRec = getAddRecExpr(AddRecOps, AddRecLoop,
- HasNUW && AddRec->hasNoUnsignedWrap(),
- HasNSW && AddRec->hasNoSignedWrap());
+ // Always propagate NW.
+ Flags = AddRec->getNoWrapFlags(setFlags(Flags, SCEV::FlagNW));
+ const SCEV *NewRec = getAddRecExpr(AddRecOps, AddRecLoop, Flags);
// If all of the other operands were loop invariant, we are done.
if (Ops.size() == 1) return NewRec;
- // Otherwise, add the folded AddRec by the non-liv parts.
+ // Otherwise, add the folded AddRec by the non-invariant parts.
for (unsigned i = 0;; ++i)
if (Ops[i] == AddRec) {
Ops[i] = NewRec;
@@ -1622,30 +1760,32 @@
// there are multiple AddRec's with the same loop induction variable being
// added together. If so, we can fold them.
for (unsigned OtherIdx = Idx+1;
- OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);++OtherIdx)
- if (OtherIdx != Idx) {
- const SCEVAddRecExpr *OtherAddRec = cast<SCEVAddRecExpr>(Ops[OtherIdx]);
- if (AddRecLoop == OtherAddRec->getLoop()) {
- // Other + {A,+,B} + {C,+,D} --> Other + {A+C,+,B+D}
- SmallVector<const SCEV *, 4> NewOps(AddRec->op_begin(),
- AddRec->op_end());
- for (unsigned i = 0, e = OtherAddRec->getNumOperands(); i != e; ++i) {
- if (i >= NewOps.size()) {
- NewOps.append(OtherAddRec->op_begin()+i,
- OtherAddRec->op_end());
- break;
+ OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);
+ ++OtherIdx)
+ if (AddRecLoop == cast<SCEVAddRecExpr>(Ops[OtherIdx])->getLoop()) {
+ // Other + {A,+,B}<L> + {C,+,D}<L> --> Other + {A+C,+,B+D}<L>
+ SmallVector<const SCEV *, 4> AddRecOps(AddRec->op_begin(),
+ AddRec->op_end());
+ for (; OtherIdx != Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);
+ ++OtherIdx)
+ if (const SCEVAddRecExpr *OtherAddRec =
+ dyn_cast<SCEVAddRecExpr>(Ops[OtherIdx]))
+ if (OtherAddRec->getLoop() == AddRecLoop) {
+ for (unsigned i = 0, e = OtherAddRec->getNumOperands();
+ i != e; ++i) {
+ if (i >= AddRecOps.size()) {
+ AddRecOps.append(OtherAddRec->op_begin()+i,
+ OtherAddRec->op_end());
+ break;
+ }
+ AddRecOps[i] = getAddExpr(AddRecOps[i],
+ OtherAddRec->getOperand(i));
+ }
+ Ops.erase(Ops.begin() + OtherIdx); --OtherIdx;
}
- NewOps[i] = getAddExpr(NewOps[i], OtherAddRec->getOperand(i));
- }
- const SCEV *NewAddRec = getAddRecExpr(NewOps, AddRecLoop);
-
- if (Ops.size() == 2) return NewAddRec;
-
- Ops.erase(Ops.begin()+Idx);
- Ops.erase(Ops.begin()+OtherIdx-1);
- Ops.push_back(NewAddRec);
- return getAddExpr(Ops);
- }
+ // Step size has changed, so we cannot guarantee no self-wraparound.
+ Ops[Idx] = getAddRecExpr(AddRecOps, AddRecLoop, SCEV::FlagAnyWrap);
+ return getAddExpr(Ops);
}
// Otherwise couldn't fold anything into this recurrence. Move onto the
@@ -1656,7 +1796,6 @@
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scAddExpr);
- ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
@@ -1669,33 +1808,70 @@
O, Ops.size());
UniqueSCEVs.InsertNode(S, IP);
}
- if (HasNUW) S->setHasNoUnsignedWrap(true);
- if (HasNSW) S->setHasNoSignedWrap(true);
+ S->setNoWrapFlags(Flags);
return S;
}
+static uint64_t umul_ov(uint64_t i, uint64_t j, bool &Overflow) {
+ uint64_t k = i*j;
+ if (j > 1 && k / j != i) Overflow = true;
+ return k;
+}
+
+/// Compute the result of "n choose k", the binomial coefficient. If an
+/// intermediate computation overflows, Overflow will be set and the return will
+/// be garbage. Overflow is not cleared on absense of overflow.
+static uint64_t Choose(uint64_t n, uint64_t k, bool &Overflow) {
+ // We use the multiplicative formula:
+ // n(n-1)(n-2)...(n-(k-1)) / k(k-1)(k-2)...1 .
+ // At each iteration, we take the n-th term of the numeral and divide by the
+ // (k-n)th term of the denominator. This division will always produce an
+ // integral result, and helps reduce the chance of overflow in the
+ // intermediate computations. However, we can still overflow even when the
+ // final result would fit.
+
+ if (n == 0 || n == k) return 1;
+ if (k > n) return 0;
+
+ if (k > n/2)
+ k = n-k;
+
+ uint64_t r = 1;
+ for (uint64_t i = 1; i <= k; ++i) {
+ r = umul_ov(r, n-(i-1), Overflow);
+ r /= i;
+ }
+ return r;
+}
+
/// getMulExpr - Get a canonical multiply expression, or something simpler if
/// possible.
const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops,
- bool HasNUW, bool HasNSW) {
+ SCEV::NoWrapFlags Flags) {
+ assert(Flags == maskFlags(Flags, SCEV::FlagNUW | SCEV::FlagNSW) &&
+ "only nuw or nsw allowed");
assert(!Ops.empty() && "Cannot get empty mul!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
+ Type *ETy = getEffectiveSCEVType(Ops[0]->getType());
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
- assert(getEffectiveSCEVType(Ops[i]->getType()) ==
- getEffectiveSCEVType(Ops[0]->getType()) &&
+ assert(getEffectiveSCEVType(Ops[i]->getType()) == ETy &&
"SCEVMulExpr operand types don't match!");
#endif
- // If HasNSW is true and all the operands are non-negative, infer HasNUW.
- if (!HasNUW && HasNSW) {
+ // If FlagNSW is true and all the operands are non-negative, infer FlagNUW.
+ // And vice-versa.
+ int SignOrUnsignMask = SCEV::FlagNUW | SCEV::FlagNSW;
+ SCEV::NoWrapFlags SignOrUnsignWrap = maskFlags(Flags, SignOrUnsignMask);
+ if (SignOrUnsignWrap && (SignOrUnsignWrap != SignOrUnsignMask)) {
bool All = true;
- for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (!isKnownNonNegative(Ops[i])) {
+ for (SmallVectorImpl<const SCEV *>::const_iterator I = Ops.begin(),
+ E = Ops.end(); I != E; ++I)
+ if (!isKnownNonNegative(*I)) {
All = false;
break;
}
- if (All) HasNUW = true;
+ if (All) Flags = setFlags(Flags, (SCEV::NoWrapFlags)SignOrUnsignMask);
}
// Sort by complexity, this groups all similar expression types together.
@@ -1735,12 +1911,12 @@
} else if (Ops[0]->isAllOnesValue()) {
// If we have a mul by -1 of an add, try distributing the -1 among the
// add operands.
- if (Ops.size() == 2)
+ if (Ops.size() == 2) {
if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Ops[1])) {
SmallVector<const SCEV *, 4> NewOps;
bool AnyFolded = false;
- for (SCEVAddRecExpr::op_iterator I = Add->op_begin(), E = Add->op_end();
- I != E; ++I) {
+ for (SCEVAddRecExpr::op_iterator I = Add->op_begin(),
+ E = Add->op_end(); I != E; ++I) {
const SCEV *Mul = getMulExpr(Ops[0], *I);
if (!isa<SCEVMulExpr>(Mul)) AnyFolded = true;
NewOps.push_back(Mul);
@@ -1748,6 +1924,18 @@
if (AnyFolded)
return getAddExpr(NewOps);
}
+ else if (const SCEVAddRecExpr *
+ AddRec = dyn_cast<SCEVAddRecExpr>(Ops[1])) {
+ // Negation preserves a recurrence's no self-wrap property.
+ SmallVector<const SCEV *, 4> Operands;
+ for (SCEVAddRecExpr::op_iterator I = AddRec->op_begin(),
+ E = AddRec->op_end(); I != E; ++I) {
+ Operands.push_back(getMulExpr(Ops[0], *I));
+ }
+ return getAddRecExpr(Operands, AddRec->getLoop(),
+ AddRec->getNoWrapFlags(SCEV::FlagNW));
+ }
+ }
}
if (Ops.size() == 1)
@@ -1788,8 +1976,9 @@
// they are loop invariant w.r.t. the recurrence.
SmallVector<const SCEV *, 8> LIOps;
const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
+ const Loop *AddRecLoop = AddRec->getLoop();
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
- if (Ops[i]->isLoopInvariant(AddRec->getLoop())) {
+ if (isLoopInvariant(Ops[i], AddRecLoop)) {
LIOps.push_back(Ops[i]);
Ops.erase(Ops.begin()+i);
--i; --e;
@@ -1806,14 +1995,16 @@
// Build the new addrec. Propagate the NUW and NSW flags if both the
// outer mul and the inner addrec are guaranteed to have no overflow.
- const SCEV *NewRec = getAddRecExpr(NewOps, AddRec->getLoop(),
- HasNUW && AddRec->hasNoUnsignedWrap(),
- HasNSW && AddRec->hasNoSignedWrap());
+ //
+ // No self-wrap cannot be guaranteed after changing the step size, but
+ // will be inferred if either NUW or NSW is true.
+ Flags = AddRec->getNoWrapFlags(clearFlags(Flags, SCEV::FlagNW));
+ const SCEV *NewRec = getAddRecExpr(NewOps, AddRecLoop, Flags);
// If all of the other operands were loop invariant, we are done.
if (Ops.size() == 1) return NewRec;
- // Otherwise, multiply the folded AddRec by the non-liv parts.
+ // Otherwise, multiply the folded AddRec by the non-invariant parts.
for (unsigned i = 0;; ++i)
if (Ops[i] == AddRec) {
Ops[i] = NewRec;
@@ -1826,29 +2017,66 @@
// there are multiple AddRec's with the same loop induction variable being
// multiplied together. If so, we can fold them.
for (unsigned OtherIdx = Idx+1;
- OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);++OtherIdx)
- if (OtherIdx != Idx) {
- const SCEVAddRecExpr *OtherAddRec = cast<SCEVAddRecExpr>(Ops[OtherIdx]);
- if (AddRec->getLoop() == OtherAddRec->getLoop()) {
- // F * G --> {A,+,B} * {C,+,D} --> {A*C,+,F*D + G*B + B*D}
- const SCEVAddRecExpr *F = AddRec, *G = OtherAddRec;
- const SCEV *NewStart = getMulExpr(F->getStart(),
- G->getStart());
- const SCEV *B = F->getStepRecurrence(*this);
- const SCEV *D = G->getStepRecurrence(*this);
- const SCEV *NewStep = getAddExpr(getMulExpr(F, D),
- getMulExpr(G, B),
- getMulExpr(B, D));
- const SCEV *NewAddRec = getAddRecExpr(NewStart, NewStep,
- F->getLoop());
- if (Ops.size() == 2) return NewAddRec;
-
- Ops.erase(Ops.begin()+Idx);
- Ops.erase(Ops.begin()+OtherIdx-1);
- Ops.push_back(NewAddRec);
+ OtherIdx < Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);
+ ++OtherIdx) {
+ if (AddRecLoop == cast<SCEVAddRecExpr>(Ops[OtherIdx])->getLoop()) {
+ // {A1,+,A2,+,...,+,An}<L> * {B1,+,B2,+,...,+,Bn}<L>
+ // = {x=1 in [ sum y=x..2x [ sum z=max(y-x, y-n)..min(x,n) [
+ // choose(x, 2x)*choose(2x-y, x-z)*A_{y-z}*B_z
+ // ]]],+,...up to x=2n}.
+ // Note that the arguments to choose() are always integers with values
+ // known at compile time, never SCEV objects.
+ //
+ // The implementation avoids pointless extra computations when the two
+ // addrec's are of different length (mathematically, it's equivalent to
+ // an infinite stream of zeros on the right).
+ bool OpsModified = false;
+ for (; OtherIdx != Ops.size() && isa<SCEVAddRecExpr>(Ops[OtherIdx]);
+ ++OtherIdx)
+ if (const SCEVAddRecExpr *OtherAddRec =
+ dyn_cast<SCEVAddRecExpr>(Ops[OtherIdx]))
+ if (OtherAddRec->getLoop() == AddRecLoop) {
+ bool Overflow = false;
+ Type *Ty = AddRec->getType();
+ bool LargerThan64Bits = getTypeSizeInBits(Ty) > 64;
+ SmallVector<const SCEV*, 7> AddRecOps;
+ for (int x = 0, xe = AddRec->getNumOperands() +
+ OtherAddRec->getNumOperands() - 1;
+ x != xe && !Overflow; ++x) {
+ const SCEV *Term = getConstant(Ty, 0);
+ for (int y = x, ye = 2*x+1; y != ye && !Overflow; ++y) {
+ uint64_t Coeff1 = Choose(x, 2*x - y, Overflow);
+ for (int z = std::max(y-x, y-(int)AddRec->getNumOperands()+1),
+ ze = std::min(x+1, (int)OtherAddRec->getNumOperands());
+ z < ze && !Overflow; ++z) {
+ uint64_t Coeff2 = Choose(2*x - y, x-z, Overflow);
+ uint64_t Coeff;
+ if (LargerThan64Bits)
+ Coeff = umul_ov(Coeff1, Coeff2, Overflow);
+ else
+ Coeff = Coeff1*Coeff2;
+ const SCEV *CoeffTerm = getConstant(Ty, Coeff);
+ const SCEV *Term1 = AddRec->getOperand(y-z);
+ const SCEV *Term2 = OtherAddRec->getOperand(z);
+ Term = getAddExpr(Term, getMulExpr(CoeffTerm, Term1,Term2));
+ }
+ }
+ AddRecOps.push_back(Term);
+ }
+ if (!Overflow) {
+ const SCEV *NewAddRec = getAddRecExpr(AddRecOps,
+ AddRec->getLoop(),
+ SCEV::FlagAnyWrap);
+ if (Ops.size() == 2) return NewAddRec;
+ Ops[Idx] = AddRec = cast<SCEVAddRecExpr>(NewAddRec);
+ Ops.erase(Ops.begin() + OtherIdx); --OtherIdx;
+ OpsModified = true;
+ }
+ }
+ if (OpsModified)
return getMulExpr(Ops);
- }
}
+ }
// Otherwise couldn't fold anything into this recurrence. Move onto the
// next one.
@@ -1858,7 +2086,6 @@
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scMulExpr);
- ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
@@ -1871,8 +2098,7 @@
O, Ops.size());
UniqueSCEVs.InsertNode(S, IP);
}
- if (HasNUW) S->setHasNoUnsignedWrap(true);
- if (HasNSW) S->setHasNoSignedWrap(true);
+ S->setNoWrapFlags(Flags);
return S;
}
@@ -1894,30 +2120,48 @@
// Determine if the division can be folded into the operands of
// its operands.
// TODO: Generalize this to non-constants by using known-bits information.
- const Type *Ty = LHS->getType();
+ Type *Ty = LHS->getType();
unsigned LZ = RHSC->getValue()->getValue().countLeadingZeros();
unsigned MaxShiftAmt = getTypeSizeInBits(Ty) - LZ - 1;
// For non-power-of-two values, effectively round the value up to the
// nearest power of two.
if (!RHSC->getValue()->getValue().isPowerOf2())
++MaxShiftAmt;
- const IntegerType *ExtTy =
+ IntegerType *ExtTy =
IntegerType::get(getContext(), getTypeSizeInBits(Ty) + MaxShiftAmt);
- // {X,+,N}/C --> {X/C,+,N/C} if safe and N/C can be folded.
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(LHS))
if (const SCEVConstant *Step =
- dyn_cast<SCEVConstant>(AR->getStepRecurrence(*this)))
- if (!Step->getValue()->getValue()
- .urem(RHSC->getValue()->getValue()) &&
+ dyn_cast<SCEVConstant>(AR->getStepRecurrence(*this))) {
+ // {X,+,N}/C --> {X/C,+,N/C} if safe and N/C can be folded.
+ const APInt &StepInt = Step->getValue()->getValue();
+ const APInt &DivInt = RHSC->getValue()->getValue();
+ if (!StepInt.urem(DivInt) &&
getZeroExtendExpr(AR, ExtTy) ==
getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
getZeroExtendExpr(Step, ExtTy),
- AR->getLoop())) {
+ AR->getLoop(), SCEV::FlagAnyWrap)) {
SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = AR->getNumOperands(); i != e; ++i)
Operands.push_back(getUDivExpr(AR->getOperand(i), RHS));
- return getAddRecExpr(Operands, AR->getLoop());
+ return getAddRecExpr(Operands, AR->getLoop(),
+ SCEV::FlagNW);
}
+ /// Get a canonical UDivExpr for a recurrence.
+ /// {X,+,N}/C => {Y,+,N}/C where Y=X-(X%N). Safe when C%N=0.
+ // We can currently only fold X%N if X is constant.
+ const SCEVConstant *StartC = dyn_cast<SCEVConstant>(AR->getStart());
+ if (StartC && !DivInt.urem(StepInt) &&
+ getZeroExtendExpr(AR, ExtTy) ==
+ getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
+ getZeroExtendExpr(Step, ExtTy),
+ AR->getLoop(), SCEV::FlagAnyWrap)) {
+ const APInt &StartInt = StartC->getValue()->getValue();
+ const APInt &StartRem = StartInt.urem(StepInt);
+ if (StartRem != 0)
+ LHS = getAddRecExpr(getConstant(StartInt - StartRem), Step,
+ AR->getLoop(), SCEV::FlagNW);
+ }
+ }
// (A*B)/C --> A*(B/C) if safe and B/C can be folded.
if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(LHS)) {
SmallVector<const SCEV *, 4> Operands;
@@ -1937,7 +2181,7 @@
}
}
// (A+B)/C --> (A/C + B/C) if safe and A/C and B/C can be folded.
- if (const SCEVAddRecExpr *A = dyn_cast<SCEVAddRecExpr>(LHS)) {
+ if (const SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(LHS)) {
SmallVector<const SCEV *, 4> Operands;
for (unsigned i = 0, e = A->getNumOperands(); i != e; ++i)
Operands.push_back(getZeroExtendExpr(A->getOperand(i), ExtTy));
@@ -1980,38 +2224,40 @@
/// getAddRecExpr - Get an add recurrence expression for the specified loop.
/// Simplify the expression as much as possible.
-const SCEV *ScalarEvolution::getAddRecExpr(const SCEV *Start,
- const SCEV *Step, const Loop *L,
- bool HasNUW, bool HasNSW) {
+const SCEV *ScalarEvolution::getAddRecExpr(const SCEV *Start, const SCEV *Step,
+ const Loop *L,
+ SCEV::NoWrapFlags Flags) {
SmallVector<const SCEV *, 4> Operands;
Operands.push_back(Start);
if (const SCEVAddRecExpr *StepChrec = dyn_cast<SCEVAddRecExpr>(Step))
if (StepChrec->getLoop() == L) {
Operands.append(StepChrec->op_begin(), StepChrec->op_end());
- return getAddRecExpr(Operands, L);
+ return getAddRecExpr(Operands, L, maskFlags(Flags, SCEV::FlagNW));
}
Operands.push_back(Step);
- return getAddRecExpr(Operands, L, HasNUW, HasNSW);
+ return getAddRecExpr(Operands, L, Flags);
}
/// getAddRecExpr - Get an add recurrence expression for the specified loop.
/// Simplify the expression as much as possible.
const SCEV *
ScalarEvolution::getAddRecExpr(SmallVectorImpl<const SCEV *> &Operands,
- const Loop *L,
- bool HasNUW, bool HasNSW) {
+ const Loop *L, SCEV::NoWrapFlags Flags) {
if (Operands.size() == 1) return Operands[0];
#ifndef NDEBUG
+ Type *ETy = getEffectiveSCEVType(Operands[0]->getType());
for (unsigned i = 1, e = Operands.size(); i != e; ++i)
- assert(getEffectiveSCEVType(Operands[i]->getType()) ==
- getEffectiveSCEVType(Operands[0]->getType()) &&
+ assert(getEffectiveSCEVType(Operands[i]->getType()) == ETy &&
"SCEVAddRecExpr operand types don't match!");
+ for (unsigned i = 0, e = Operands.size(); i != e; ++i)
+ assert(isLoopInvariant(Operands[i], L) &&
+ "SCEVAddRecExpr operand is not loop-invariant!");
#endif
if (Operands.back()->isZero()) {
Operands.pop_back();
- return getAddRecExpr(Operands, L, HasNUW, HasNSW); // {X,+,0} --> X
+ return getAddRecExpr(Operands, L, SCEV::FlagAnyWrap); // {X,+,0} --> X
}
// It's tempting to want to call getMaxBackedgeTakenCount count here and
@@ -2020,23 +2266,27 @@
// meaningful BE count at this point (and if we don't, we'd be stuck
// with a SCEVCouldNotCompute as the cached BE count).
- // If HasNSW is true and all the operands are non-negative, infer HasNUW.
- if (!HasNUW && HasNSW) {
+ // If FlagNSW is true and all the operands are non-negative, infer FlagNUW.
+ // And vice-versa.
+ int SignOrUnsignMask = SCEV::FlagNUW | SCEV::FlagNSW;
+ SCEV::NoWrapFlags SignOrUnsignWrap = maskFlags(Flags, SignOrUnsignMask);
+ if (SignOrUnsignWrap && (SignOrUnsignWrap != SignOrUnsignMask)) {
bool All = true;
- for (unsigned i = 0, e = Operands.size(); i != e; ++i)
- if (!isKnownNonNegative(Operands[i])) {
+ for (SmallVectorImpl<const SCEV *>::const_iterator I = Operands.begin(),
+ E = Operands.end(); I != E; ++I)
+ if (!isKnownNonNegative(*I)) {
All = false;
break;
}
- if (All) HasNUW = true;
+ if (All) Flags = setFlags(Flags, (SCEV::NoWrapFlags)SignOrUnsignMask);
}
// Canonicalize nested AddRecs in by nesting them in order of loop depth.
if (const SCEVAddRecExpr *NestedAR = dyn_cast<SCEVAddRecExpr>(Operands[0])) {
const Loop *NestedLoop = NestedAR->getLoop();
- if (L->contains(NestedLoop->getHeader()) ?
+ if (L->contains(NestedLoop) ?
(L->getLoopDepth() < NestedLoop->getLoopDepth()) :
- (!NestedLoop->contains(L->getHeader()) &&
+ (!NestedLoop->contains(L) &&
DT->dominates(L->getHeader(), NestedLoop->getHeader()))) {
SmallVector<const SCEV *, 4> NestedOperands(NestedAR->op_begin(),
NestedAR->op_end());
@@ -2046,21 +2296,34 @@
// requirement.
bool AllInvariant = true;
for (unsigned i = 0, e = Operands.size(); i != e; ++i)
- if (!Operands[i]->isLoopInvariant(L)) {
+ if (!isLoopInvariant(Operands[i], L)) {
AllInvariant = false;
break;
}
if (AllInvariant) {
- NestedOperands[0] = getAddRecExpr(Operands, L);
+ // Create a recurrence for the outer loop with the same step size.
+ //
+ // The outer recurrence keeps its NW flag but only keeps NUW/NSW if the
+ // inner recurrence has the same property.
+ SCEV::NoWrapFlags OuterFlags =
+ maskFlags(Flags, SCEV::FlagNW | NestedAR->getNoWrapFlags());
+
+ NestedOperands[0] = getAddRecExpr(Operands, L, OuterFlags);
AllInvariant = true;
for (unsigned i = 0, e = NestedOperands.size(); i != e; ++i)
- if (!NestedOperands[i]->isLoopInvariant(NestedLoop)) {
+ if (!isLoopInvariant(NestedOperands[i], NestedLoop)) {
AllInvariant = false;
break;
}
- if (AllInvariant)
+ if (AllInvariant) {
// Ok, both add recurrences are valid after the transformation.
- return getAddRecExpr(NestedOperands, NestedLoop, HasNUW, HasNSW);
+ //
+ // The inner recurrence keeps its NW flag but only keeps NUW/NSW if
+ // the outer recurrence has the same property.
+ SCEV::NoWrapFlags InnerFlags =
+ maskFlags(NestedAR->getNoWrapFlags(), SCEV::FlagNW | Flags);
+ return getAddRecExpr(NestedOperands, NestedLoop, InnerFlags);
+ }
}
// Reset Operands to its original state.
Operands[0] = NestedAR;
@@ -2071,7 +2334,6 @@
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scAddRecExpr);
- ID.AddInteger(Operands.size());
for (unsigned i = 0, e = Operands.size(); i != e; ++i)
ID.AddPointer(Operands[i]);
ID.AddPointer(L);
@@ -2085,8 +2347,7 @@
O, Operands.size(), L);
UniqueSCEVs.InsertNode(S, IP);
}
- if (HasNUW) S->setHasNoUnsignedWrap(true);
- if (HasNSW) S->setHasNoSignedWrap(true);
+ S->setNoWrapFlags(Flags);
return S;
}
@@ -2103,9 +2364,9 @@
assert(!Ops.empty() && "Cannot get empty smax!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
+ Type *ETy = getEffectiveSCEVType(Ops[0]->getType());
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
- assert(getEffectiveSCEVType(Ops[i]->getType()) ==
- getEffectiveSCEVType(Ops[0]->getType()) &&
+ assert(getEffectiveSCEVType(Ops[i]->getType()) == ETy &&
"SCEVSMaxExpr operand types don't match!");
#endif
@@ -2182,7 +2443,6 @@
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scSMaxExpr);
- ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
@@ -2208,9 +2468,9 @@
assert(!Ops.empty() && "Cannot get empty umax!");
if (Ops.size() == 1) return Ops[0];
#ifndef NDEBUG
+ Type *ETy = getEffectiveSCEVType(Ops[0]->getType());
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
- assert(getEffectiveSCEVType(Ops[i]->getType()) ==
- getEffectiveSCEVType(Ops[0]->getType()) &&
+ assert(getEffectiveSCEVType(Ops[i]->getType()) == ETy &&
"SCEVUMaxExpr operand types don't match!");
#endif
@@ -2287,7 +2547,6 @@
// already have one, otherwise create a new one.
FoldingSetNodeID ID;
ID.AddInteger(scUMaxExpr);
- ID.AddInteger(Ops.size());
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
ID.AddPointer(Ops[i]);
void *IP = 0;
@@ -2312,7 +2571,7 @@
return getNotSCEV(getUMaxExpr(getNotSCEV(LHS), getNotSCEV(RHS)));
}
-const SCEV *ScalarEvolution::getSizeOfExpr(const Type *AllocTy) {
+const SCEV *ScalarEvolution::getSizeOfExpr(Type *AllocTy) {
// If we have TargetData, we can bypass creating a target-independent
// constant expression and then folding it back into a ConstantInt.
// This is just a compile-time optimization.
@@ -2324,20 +2583,20 @@
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
if (Constant *Folded = ConstantFoldConstantExpression(CE, TD))
C = Folded;
- const Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(AllocTy));
+ Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(AllocTy));
return getTruncateOrZeroExtend(getSCEV(C), Ty);
}
-const SCEV *ScalarEvolution::getAlignOfExpr(const Type *AllocTy) {
+const SCEV *ScalarEvolution::getAlignOfExpr(Type *AllocTy) {
Constant *C = ConstantExpr::getAlignOf(AllocTy);
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
if (Constant *Folded = ConstantFoldConstantExpression(CE, TD))
C = Folded;
- const Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(AllocTy));
+ Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(AllocTy));
return getTruncateOrZeroExtend(getSCEV(C), Ty);
}
-const SCEV *ScalarEvolution::getOffsetOfExpr(const StructType *STy,
+const SCEV *ScalarEvolution::getOffsetOfExpr(StructType *STy,
unsigned FieldNo) {
// If we have TargetData, we can bypass creating a target-independent
// constant expression and then folding it back into a ConstantInt.
@@ -2350,17 +2609,17 @@
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
if (Constant *Folded = ConstantFoldConstantExpression(CE, TD))
C = Folded;
- const Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(STy));
+ Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(STy));
return getTruncateOrZeroExtend(getSCEV(C), Ty);
}
-const SCEV *ScalarEvolution::getOffsetOfExpr(const Type *CTy,
+const SCEV *ScalarEvolution::getOffsetOfExpr(Type *CTy,
Constant *FieldNo) {
Constant *C = ConstantExpr::getOffsetOf(CTy, FieldNo);
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
if (Constant *Folded = ConstantFoldConstantExpression(CE, TD))
C = Folded;
- const Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(CTy));
+ Type *Ty = getEffectiveSCEVType(PointerType::getUnqual(CTy));
return getTruncateOrZeroExtend(getSCEV(C), Ty);
}
@@ -2394,14 +2653,14 @@
/// the SCEV framework. This primarily includes integer types, and it
/// can optionally include pointer types if the ScalarEvolution class
/// has access to target-specific information.
-bool ScalarEvolution::isSCEVable(const Type *Ty) const {
+bool ScalarEvolution::isSCEVable(Type *Ty) const {
// Integers and pointers are always SCEVable.
return Ty->isIntegerTy() || Ty->isPointerTy();
}
/// getTypeSizeInBits - Return the size in bits of the specified type,
/// for which isSCEVable must return true.
-uint64_t ScalarEvolution::getTypeSizeInBits(const Type *Ty) const {
+uint64_t ScalarEvolution::getTypeSizeInBits(Type *Ty) const {
assert(isSCEVable(Ty) && "Type is not SCEVable!");
// If we have a TargetData, use it!
@@ -2422,7 +2681,7 @@
/// the given type and which represents how SCEV will treat the given
/// type, for which isSCEVable must return true. For pointer types,
/// this is the pointer-sized integer type.
-const Type *ScalarEvolution::getEffectiveSCEVType(const Type *Ty) const {
+Type *ScalarEvolution::getEffectiveSCEVType(Type *Ty) const {
assert(isSCEVable(Ty) && "Type is not SCEVable!");
if (Ty->isIntegerTy())
@@ -2445,10 +2704,15 @@
const SCEV *ScalarEvolution::getSCEV(Value *V) {
assert(isSCEVable(V->getType()) && "Value is not SCEVable!");
- std::map<SCEVCallbackVH, const SCEV *>::iterator I = Scalars.find(V);
- if (I != Scalars.end()) return I->second;
+ ValueExprMapType::const_iterator I = ValueExprMap.find(V);
+ if (I != ValueExprMap.end()) return I->second;
const SCEV *S = createSCEV(V);
- Scalars.insert(std::make_pair(SCEVCallbackVH(V, this), S));
+
+ // The process of creating a SCEV for V may have caused other SCEVs
+ // to have been created, so it's necessary to insert the new entry
+ // from scratch, rather than trying to remember the insert position
+ // above.
+ ValueExprMap.insert(std::make_pair(SCEVCallbackVH(V, this), S));
return S;
}
@@ -2459,7 +2723,7 @@
return getConstant(
cast<ConstantInt>(ConstantExpr::getNeg(VC->getValue())));
- const Type *Ty = V->getType();
+ Type *Ty = V->getType();
Ty = getEffectiveSCEVType(Ty);
return getMulExpr(V,
getConstant(cast<ConstantInt>(Constant::getAllOnesValue(Ty))));
@@ -2471,32 +2735,32 @@
return getConstant(
cast<ConstantInt>(ConstantExpr::getNot(VC->getValue())));
- const Type *Ty = V->getType();
+ Type *Ty = V->getType();
Ty = getEffectiveSCEVType(Ty);
const SCEV *AllOnes =
getConstant(cast<ConstantInt>(Constant::getAllOnesValue(Ty)));
return getMinusSCEV(AllOnes, V);
}
-/// getMinusSCEV - Return a SCEV corresponding to LHS - RHS.
-///
-const SCEV *ScalarEvolution::getMinusSCEV(const SCEV *LHS,
- const SCEV *RHS) {
+/// getMinusSCEV - Return LHS-RHS. Minus is represented in SCEV as A+B*-1.
+const SCEV *ScalarEvolution::getMinusSCEV(const SCEV *LHS, const SCEV *RHS,
+ SCEV::NoWrapFlags Flags) {
+ assert(!maskFlags(Flags, SCEV::FlagNUW) && "subtraction does not have NUW");
+
// Fast path: X - X --> 0.
if (LHS == RHS)
return getConstant(LHS->getType(), 0);
// X - Y --> X + -Y
- return getAddExpr(LHS, getNegativeSCEV(RHS));
+ return getAddExpr(LHS, getNegativeSCEV(RHS), Flags);
}
/// getTruncateOrZeroExtend - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. If the type must be extended, it is zero
/// extended.
const SCEV *
-ScalarEvolution::getTruncateOrZeroExtend(const SCEV *V,
- const Type *Ty) {
- const Type *SrcTy = V->getType();
+ScalarEvolution::getTruncateOrZeroExtend(const SCEV *V, Type *Ty) {
+ Type *SrcTy = V->getType();
assert((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot truncate or zero extend with non-integer arguments!");
@@ -2512,8 +2776,8 @@
/// extended.
const SCEV *
ScalarEvolution::getTruncateOrSignExtend(const SCEV *V,
- const Type *Ty) {
- const Type *SrcTy = V->getType();
+ Type *Ty) {
+ Type *SrcTy = V->getType();
assert((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot truncate or zero extend with non-integer arguments!");
@@ -2528,8 +2792,8 @@
/// input value to the specified type. If the type must be extended, it is zero
/// extended. The conversion must not be narrowing.
const SCEV *
-ScalarEvolution::getNoopOrZeroExtend(const SCEV *V, const Type *Ty) {
- const Type *SrcTy = V->getType();
+ScalarEvolution::getNoopOrZeroExtend(const SCEV *V, Type *Ty) {
+ Type *SrcTy = V->getType();
assert((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot noop or zero extend with non-integer arguments!");
@@ -2544,8 +2808,8 @@
/// input value to the specified type. If the type must be extended, it is sign
/// extended. The conversion must not be narrowing.
const SCEV *
-ScalarEvolution::getNoopOrSignExtend(const SCEV *V, const Type *Ty) {
- const Type *SrcTy = V->getType();
+ScalarEvolution::getNoopOrSignExtend(const SCEV *V, Type *Ty) {
+ Type *SrcTy = V->getType();
assert((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot noop or sign extend with non-integer arguments!");
@@ -2561,8 +2825,8 @@
/// it is extended with unspecified bits. The conversion must not be
/// narrowing.
const SCEV *
-ScalarEvolution::getNoopOrAnyExtend(const SCEV *V, const Type *Ty) {
- const Type *SrcTy = V->getType();
+ScalarEvolution::getNoopOrAnyExtend(const SCEV *V, Type *Ty) {
+ Type *SrcTy = V->getType();
assert((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot noop or any extend with non-integer arguments!");
@@ -2576,8 +2840,8 @@
/// getTruncateOrNoop - Return a SCEV corresponding to a conversion of the
/// input value to the specified type. The conversion must not be widening.
const SCEV *
-ScalarEvolution::getTruncateOrNoop(const SCEV *V, const Type *Ty) {
- const Type *SrcTy = V->getType();
+ScalarEvolution::getTruncateOrNoop(const SCEV *V, Type *Ty) {
+ Type *SrcTy = V->getType();
assert((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
(Ty->isIntegerTy() || Ty->isPointerTy()) &&
"Cannot truncate or noop with non-integer arguments!");
@@ -2620,6 +2884,36 @@
return getUMinExpr(PromotedLHS, PromotedRHS);
}
+/// getPointerBase - Transitively follow the chain of pointer-type operands
+/// until reaching a SCEV that does not have a single pointer operand. This
+/// returns a SCEVUnknown pointer for well-formed pointer-type expressions,
+/// but corner cases do exist.
+const SCEV *ScalarEvolution::getPointerBase(const SCEV *V) {
+ // A pointer operand may evaluate to a nonpointer expression, such as null.
+ if (!V->getType()->isPointerTy())
+ return V;
+
+ if (const SCEVCastExpr *Cast = dyn_cast<SCEVCastExpr>(V)) {
+ return getPointerBase(Cast->getOperand());
+ }
+ else if (const SCEVNAryExpr *NAry = dyn_cast<SCEVNAryExpr>(V)) {
+ const SCEV *PtrOp = 0;
+ for (SCEVNAryExpr::op_iterator I = NAry->op_begin(), E = NAry->op_end();
+ I != E; ++I) {
+ if ((*I)->getType()->isPointerTy()) {
+ // Cannot find the base of an expression with multiple pointer operands.
+ if (PtrOp)
+ return V;
+ PtrOp = *I;
+ }
+ }
+ if (!PtrOp)
+ return V;
+ return getPointerBase(PtrOp);
+ }
+ return V;
+}
+
/// PushDefUseChildren - Push users of the given Instruction
/// onto the given Worklist.
static void
@@ -2633,7 +2927,7 @@
/// ForgetSymbolicValue - This looks up computed SCEV values for all
/// instructions that depend on the given instruction and removes them from
-/// the Scalars map if they reference SymName. This is used during PHI
+/// the ValueExprMapType map if they reference SymName. This is used during PHI
/// resolution.
void
ScalarEvolution::ForgetSymbolicName(Instruction *PN, const SCEV *SymName) {
@@ -2646,12 +2940,14 @@
Instruction *I = Worklist.pop_back_val();
if (!Visited.insert(I)) continue;
- std::map<SCEVCallbackVH, const SCEV *>::iterator It =
- Scalars.find(static_cast<Value *>(I));
- if (It != Scalars.end()) {
+ ValueExprMapType::iterator It =
+ ValueExprMap.find(static_cast<Value *>(I));
+ if (It != ValueExprMap.end()) {
+ const SCEV *Old = It->second;
+
// Short-circuit the def-use traversal if the symbolic name
// ceases to appear in expressions.
- if (It->second != SymName && !It->second->hasOperand(SymName))
+ if (Old != SymName && !hasOperand(Old, SymName))
continue;
// SCEVUnknown for a PHI either means that it has an unrecognized
@@ -2662,10 +2958,10 @@
// updates on its own when it gets to that point. In the third, we do
// want to forget the SCEVUnknown.
if (!isa<PHINode>(I) ||
- !isa<SCEVUnknown>(It->second) ||
- (I != PN && It->second == SymName)) {
- ValuesAtScopes.erase(It->second);
- Scalars.erase(It);
+ !isa<SCEVUnknown>(Old) ||
+ (I != PN && Old == SymName)) {
+ forgetMemoizedResults(Old);
+ ValueExprMap.erase(It);
}
}
@@ -2702,9 +2998,9 @@
if (BEValueV && StartValueV) {
// While we are analyzing this PHI node, handle its value symbolically.
const SCEV *SymbolicName = getUnknown(PN);
- assert(Scalars.find(PN) == Scalars.end() &&
+ assert(ValueExprMap.find(PN) == ValueExprMap.end() &&
"PHI node already processed?");
- Scalars.insert(std::make_pair(SCEVCallbackVH(PN, this), SymbolicName));
+ ValueExprMap.insert(std::make_pair(SCEVCallbackVH(PN, this), SymbolicName));
// Using this symbolic name for the PHI, analyze the value coming around
// the back-edge.
@@ -2736,36 +3032,43 @@
// This is not a valid addrec if the step amount is varying each
// loop iteration, but is not itself an addrec in this loop.
- if (Accum->isLoopInvariant(L) ||
+ if (isLoopInvariant(Accum, L) ||
(isa<SCEVAddRecExpr>(Accum) &&
cast<SCEVAddRecExpr>(Accum)->getLoop() == L)) {
- bool HasNUW = false;
- bool HasNSW = false;
+ SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap;
// If the increment doesn't overflow, then neither the addrec nor
// the post-increment will overflow.
if (const AddOperator *OBO = dyn_cast<AddOperator>(BEValueV)) {
if (OBO->hasNoUnsignedWrap())
- HasNUW = true;
+ Flags = setFlags(Flags, SCEV::FlagNUW);
if (OBO->hasNoSignedWrap())
- HasNSW = true;
+ Flags = setFlags(Flags, SCEV::FlagNSW);
+ } else if (const GEPOperator *GEP =
+ dyn_cast<GEPOperator>(BEValueV)) {
+ // If the increment is an inbounds GEP, then we know the address
+ // space cannot be wrapped around. We cannot make any guarantee
+ // about signed or unsigned overflow because pointers are
+ // unsigned but we may have a negative index from the base
+ // pointer.
+ if (GEP->isInBounds())
+ Flags = setFlags(Flags, SCEV::FlagNW);
}
const SCEV *StartVal = getSCEV(StartValueV);
- const SCEV *PHISCEV =
- getAddRecExpr(StartVal, Accum, L, HasNUW, HasNSW);
+ const SCEV *PHISCEV = getAddRecExpr(StartVal, Accum, L, Flags);
// Since the no-wrap flags are on the increment, they apply to the
// post-incremented value as well.
- if (Accum->isLoopInvariant(L))
+ if (isLoopInvariant(Accum, L))
(void)getAddRecExpr(getAddExpr(StartVal, Accum),
- Accum, L, HasNUW, HasNSW);
+ Accum, L, Flags);
// Okay, for the entire analysis of this edge we assumed the PHI
// to be symbolic. We now need to go back and purge all of the
// entries for the scalars that use the symbolic expression.
ForgetSymbolicName(PN, SymbolicName);
- Scalars[SCEVCallbackVH(PN, this)] = PHISCEV;
+ ValueExprMap[SCEVCallbackVH(PN, this)] = PHISCEV;
return PHISCEV;
}
}
@@ -2783,14 +3086,17 @@
// initial step of the addrec evolution.
if (StartVal == getMinusSCEV(AddRec->getOperand(0),
AddRec->getOperand(1))) {
+ // FIXME: For constant StartVal, we should be able to infer
+ // no-wrap flags.
const SCEV *PHISCEV =
- getAddRecExpr(StartVal, AddRec->getOperand(1), L);
+ getAddRecExpr(StartVal, AddRec->getOperand(1), L,
+ SCEV::FlagAnyWrap);
// Okay, for the entire analysis of this edge we assumed the PHI
// to be symbolic. We now need to go back and purge all of the
// entries for the scalars that use the symbolic expression.
ForgetSymbolicName(PN, SymbolicName);
- Scalars[SCEVCallbackVH(PN, this)] = PHISCEV;
+ ValueExprMap[SCEVCallbackVH(PN, this)] = PHISCEV;
return PHISCEV;
}
}
@@ -2802,17 +3108,9 @@
// PHI's incoming blocks are in a different loop, in which case doing so
// risks breaking LCSSA form. Instcombine would normally zap these, but
// it doesn't have DominatorTree information, so it may miss cases.
- if (Value *V = PN->hasConstantValue(DT)) {
- bool AllSameLoop = true;
- Loop *PNLoop = LI->getLoopFor(PN->getParent());
- for (size_t i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
- if (LI->getLoopFor(PN->getIncomingBlock(i)) != PNLoop) {
- AllSameLoop = false;
- break;
- }
- if (AllSameLoop)
+ if (Value *V = SimplifyInstruction(PN, TD, DT))
+ if (LI->replacementPreservesLCSSAForm(PN, V))
return getSCEV(V);
- }
// If it's not a loop phi, we can't handle it yet.
return getUnknown(PN);
@@ -2827,8 +3125,9 @@
// Add expression, because the Instruction may be guarded by control flow
// and the no-overflow bits may not be valid for the expression in any
// context.
+ bool isInBounds = GEP->isInBounds();
- const Type *IntPtrTy = getEffectiveSCEVType(GEP->getType());
+ Type *IntPtrTy = getEffectiveSCEVType(GEP->getType());
Value *Base = GEP->getOperand(0);
// Don't attempt to analyze GEPs over unsized objects.
if (!cast<PointerType>(Base->getType())->getElementType()->isSized())
@@ -2840,7 +3139,7 @@
I != E; ++I) {
Value *Index = *I;
// Compute the (potentially symbolic) offset in bytes for this index.
- if (const StructType *STy = dyn_cast<StructType>(*GTI++)) {
+ if (StructType *STy = dyn_cast<StructType>(*GTI++)) {
// For a struct, add the member offset.
unsigned FieldNo = cast<ConstantInt>(Index)->getZExtValue();
const SCEV *FieldOffset = getOffsetOfExpr(STy, FieldNo);
@@ -2855,7 +3154,9 @@
IndexS = getTruncateOrSignExtend(IndexS, IntPtrTy);
// Multiply the index by the element size to compute the element offset.
- const SCEV *LocalOffset = getMulExpr(IndexS, ElementSize);
+ const SCEV *LocalOffset = getMulExpr(IndexS, ElementSize,
+ isInBounds ? SCEV::FlagNSW :
+ SCEV::FlagAnyWrap);
// Add the element offset to the running total offset.
TotalOffset = getAddExpr(TotalOffset, LocalOffset);
@@ -2866,7 +3167,8 @@
const SCEV *BaseS = getSCEV(Base);
// Add the total offset from all the GEP indices to the base.
- return getAddExpr(BaseS, TotalOffset);
+ return getAddExpr(BaseS, TotalOffset,
+ isInBounds ? SCEV::FlagNSW : SCEV::FlagAnyWrap);
}
/// GetMinTrailingZeros - Determine the minimum number of zero bits that S is
@@ -2954,9 +3256,13 @@
///
ConstantRange
ScalarEvolution::getUnsignedRange(const SCEV *S) {
+ // See if we've computed this range already.
+ DenseMap<const SCEV *, ConstantRange>::iterator I = UnsignedRanges.find(S);
+ if (I != UnsignedRanges.end())
+ return I->second;
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S))
- return ConstantRange(C->getValue()->getValue());
+ return setUnsignedRange(C, ConstantRange(C->getValue()->getValue()));
unsigned BitWidth = getTypeSizeInBits(S->getType());
ConstantRange ConservativeResult(BitWidth, /*isFullSet=*/true);
@@ -2973,55 +3279,58 @@
ConstantRange X = getUnsignedRange(Add->getOperand(0));
for (unsigned i = 1, e = Add->getNumOperands(); i != e; ++i)
X = X.add(getUnsignedRange(Add->getOperand(i)));
- return ConservativeResult.intersectWith(X);
+ return setUnsignedRange(Add, ConservativeResult.intersectWith(X));
}
if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(S)) {
ConstantRange X = getUnsignedRange(Mul->getOperand(0));
for (unsigned i = 1, e = Mul->getNumOperands(); i != e; ++i)
X = X.multiply(getUnsignedRange(Mul->getOperand(i)));
- return ConservativeResult.intersectWith(X);
+ return setUnsignedRange(Mul, ConservativeResult.intersectWith(X));
}
if (const SCEVSMaxExpr *SMax = dyn_cast<SCEVSMaxExpr>(S)) {
ConstantRange X = getUnsignedRange(SMax->getOperand(0));
for (unsigned i = 1, e = SMax->getNumOperands(); i != e; ++i)
X = X.smax(getUnsignedRange(SMax->getOperand(i)));
- return ConservativeResult.intersectWith(X);
+ return setUnsignedRange(SMax, ConservativeResult.intersectWith(X));
}
if (const SCEVUMaxExpr *UMax = dyn_cast<SCEVUMaxExpr>(S)) {
ConstantRange X = getUnsignedRange(UMax->getOperand(0));
for (unsigned i = 1, e = UMax->getNumOperands(); i != e; ++i)
X = X.umax(getUnsignedRange(UMax->getOperand(i)));
- return ConservativeResult.intersectWith(X);
+ return setUnsignedRange(UMax, ConservativeResult.intersectWith(X));
}
if (const SCEVUDivExpr *UDiv = dyn_cast<SCEVUDivExpr>(S)) {
ConstantRange X = getUnsignedRange(UDiv->getLHS());
ConstantRange Y = getUnsignedRange(UDiv->getRHS());
- return ConservativeResult.intersectWith(X.udiv(Y));
+ return setUnsignedRange(UDiv, ConservativeResult.intersectWith(X.udiv(Y)));
}
if (const SCEVZeroExtendExpr *ZExt = dyn_cast<SCEVZeroExtendExpr>(S)) {
ConstantRange X = getUnsignedRange(ZExt->getOperand());
- return ConservativeResult.intersectWith(X.zeroExtend(BitWidth));
+ return setUnsignedRange(ZExt,
+ ConservativeResult.intersectWith(X.zeroExtend(BitWidth)));
}
if (const SCEVSignExtendExpr *SExt = dyn_cast<SCEVSignExtendExpr>(S)) {
ConstantRange X = getUnsignedRange(SExt->getOperand());
- return ConservativeResult.intersectWith(X.signExtend(BitWidth));
+ return setUnsignedRange(SExt,
+ ConservativeResult.intersectWith(X.signExtend(BitWidth)));
}
if (const SCEVTruncateExpr *Trunc = dyn_cast<SCEVTruncateExpr>(S)) {
ConstantRange X = getUnsignedRange(Trunc->getOperand());
- return ConservativeResult.intersectWith(X.truncate(BitWidth));
+ return setUnsignedRange(Trunc,
+ ConservativeResult.intersectWith(X.truncate(BitWidth)));
}
if (const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(S)) {
// If there's no unsigned wrap, the value will never be less than its
// initial value.
- if (AddRec->hasNoUnsignedWrap())
+ if (AddRec->getNoWrapFlags(SCEV::FlagNUW))
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(AddRec->getStart()))
if (!C->getValue()->isZero())
ConservativeResult =
@@ -3030,7 +3339,7 @@
// TODO: non-affine addrec
if (AddRec->isAffine()) {
- const Type *Ty = AddRec->getType();
+ Type *Ty = AddRec->getType();
const SCEV *MaxBECount = getMaxBackedgeTakenCount(AddRec->getLoop());
if (!isa<SCEVCouldNotCompute>(MaxBECount) &&
getTypeSizeInBits(MaxBECount->getType()) <= BitWidth) {
@@ -3055,19 +3364,20 @@
ConstantRange ExtEndRange = EndRange.zextOrTrunc(BitWidth*2+1);
if (ExtStartRange.add(ExtMaxBECountRange.multiply(ExtStepRange)) !=
ExtEndRange)
- return ConservativeResult;
+ return setUnsignedRange(AddRec, ConservativeResult);
APInt Min = APIntOps::umin(StartRange.getUnsignedMin(),
EndRange.getUnsignedMin());
APInt Max = APIntOps::umax(StartRange.getUnsignedMax(),
EndRange.getUnsignedMax());
if (Min.isMinValue() && Max.isMaxValue())
- return ConservativeResult;
- return ConservativeResult.intersectWith(ConstantRange(Min, Max+1));
+ return setUnsignedRange(AddRec, ConservativeResult);
+ return setUnsignedRange(AddRec,
+ ConservativeResult.intersectWith(ConstantRange(Min, Max+1)));
}
}
- return ConservativeResult;
+ return setUnsignedRange(AddRec, ConservativeResult);
}
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
@@ -3076,20 +3386,25 @@
APInt Zeros(BitWidth, 0), Ones(BitWidth, 0);
ComputeMaskedBits(U->getValue(), Mask, Zeros, Ones, TD);
if (Ones == ~Zeros + 1)
- return ConservativeResult;
- return ConservativeResult.intersectWith(ConstantRange(Ones, ~Zeros + 1));
+ return setUnsignedRange(U, ConservativeResult);
+ return setUnsignedRange(U,
+ ConservativeResult.intersectWith(ConstantRange(Ones, ~Zeros + 1)));
}
- return ConservativeResult;
+ return setUnsignedRange(S, ConservativeResult);
}
/// getSignedRange - Determine the signed range for a particular SCEV.
///
ConstantRange
ScalarEvolution::getSignedRange(const SCEV *S) {
+ // See if we've computed this range already.
+ DenseMap<const SCEV *, ConstantRange>::iterator I = SignedRanges.find(S);
+ if (I != SignedRanges.end())
+ return I->second;
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(S))
- return ConstantRange(C->getValue()->getValue());
+ return setSignedRange(C, ConstantRange(C->getValue()->getValue()));
unsigned BitWidth = getTypeSizeInBits(S->getType());
ConstantRange ConservativeResult(BitWidth, /*isFullSet=*/true);
@@ -3106,55 +3421,58 @@
ConstantRange X = getSignedRange(Add->getOperand(0));
for (unsigned i = 1, e = Add->getNumOperands(); i != e; ++i)
X = X.add(getSignedRange(Add->getOperand(i)));
- return ConservativeResult.intersectWith(X);
+ return setSignedRange(Add, ConservativeResult.intersectWith(X));
}
if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(S)) {
ConstantRange X = getSignedRange(Mul->getOperand(0));
for (unsigned i = 1, e = Mul->getNumOperands(); i != e; ++i)
X = X.multiply(getSignedRange(Mul->getOperand(i)));
- return ConservativeResult.intersectWith(X);
+ return setSignedRange(Mul, ConservativeResult.intersectWith(X));
}
if (const SCEVSMaxExpr *SMax = dyn_cast<SCEVSMaxExpr>(S)) {
ConstantRange X = getSignedRange(SMax->getOperand(0));
for (unsigned i = 1, e = SMax->getNumOperands(); i != e; ++i)
X = X.smax(getSignedRange(SMax->getOperand(i)));
- return ConservativeResult.intersectWith(X);
+ return setSignedRange(SMax, ConservativeResult.intersectWith(X));
}
if (const SCEVUMaxExpr *UMax = dyn_cast<SCEVUMaxExpr>(S)) {
ConstantRange X = getSignedRange(UMax->getOperand(0));
for (unsigned i = 1, e = UMax->getNumOperands(); i != e; ++i)
X = X.umax(getSignedRange(UMax->getOperand(i)));
- return ConservativeResult.intersectWith(X);
+ return setSignedRange(UMax, ConservativeResult.intersectWith(X));
}
if (const SCEVUDivExpr *UDiv = dyn_cast<SCEVUDivExpr>(S)) {
ConstantRange X = getSignedRange(UDiv->getLHS());
ConstantRange Y = getSignedRange(UDiv->getRHS());
- return ConservativeResult.intersectWith(X.udiv(Y));
+ return setSignedRange(UDiv, ConservativeResult.intersectWith(X.udiv(Y)));
}
if (const SCEVZeroExtendExpr *ZExt = dyn_cast<SCEVZeroExtendExpr>(S)) {
ConstantRange X = getSignedRange(ZExt->getOperand());
- return ConservativeResult.intersectWith(X.zeroExtend(BitWidth));
+ return setSignedRange(ZExt,
+ ConservativeResult.intersectWith(X.zeroExtend(BitWidth)));
}
if (const SCEVSignExtendExpr *SExt = dyn_cast<SCEVSignExtendExpr>(S)) {
ConstantRange X = getSignedRange(SExt->getOperand());
- return ConservativeResult.intersectWith(X.signExtend(BitWidth));
+ return setSignedRange(SExt,
+ ConservativeResult.intersectWith(X.signExtend(BitWidth)));
}
if (const SCEVTruncateExpr *Trunc = dyn_cast<SCEVTruncateExpr>(S)) {
ConstantRange X = getSignedRange(Trunc->getOperand());
- return ConservativeResult.intersectWith(X.truncate(BitWidth));
+ return setSignedRange(Trunc,
+ ConservativeResult.intersectWith(X.truncate(BitWidth)));
}
if (const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(S)) {
// If there's no signed wrap, and all the operands have the same sign or
// zero, the value won't ever change sign.
- if (AddRec->hasNoSignedWrap()) {
+ if (AddRec->getNoWrapFlags(SCEV::FlagNSW)) {
bool AllNonNeg = true;
bool AllNonPos = true;
for (unsigned i = 0, e = AddRec->getNumOperands(); i != e; ++i) {
@@ -3173,7 +3491,7 @@
// TODO: non-affine addrec
if (AddRec->isAffine()) {
- const Type *Ty = AddRec->getType();
+ Type *Ty = AddRec->getType();
const SCEV *MaxBECount = getMaxBackedgeTakenCount(AddRec->getLoop());
if (!isa<SCEVCouldNotCompute>(MaxBECount) &&
getTypeSizeInBits(MaxBECount->getType()) <= BitWidth) {
@@ -3198,34 +3516,35 @@
ConstantRange ExtEndRange = EndRange.sextOrTrunc(BitWidth*2+1);
if (ExtStartRange.add(ExtMaxBECountRange.multiply(ExtStepRange)) !=
ExtEndRange)
- return ConservativeResult;
+ return setSignedRange(AddRec, ConservativeResult);
APInt Min = APIntOps::smin(StartRange.getSignedMin(),
EndRange.getSignedMin());
APInt Max = APIntOps::smax(StartRange.getSignedMax(),
EndRange.getSignedMax());
if (Min.isMinSignedValue() && Max.isMaxSignedValue())
- return ConservativeResult;
- return ConservativeResult.intersectWith(ConstantRange(Min, Max+1));
+ return setSignedRange(AddRec, ConservativeResult);
+ return setSignedRange(AddRec,
+ ConservativeResult.intersectWith(ConstantRange(Min, Max+1)));
}
}
- return ConservativeResult;
+ return setSignedRange(AddRec, ConservativeResult);
}
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
// For a SCEVUnknown, ask ValueTracking.
if (!U->getValue()->getType()->isIntegerTy() && !TD)
- return ConservativeResult;
+ return setSignedRange(U, ConservativeResult);
unsigned NS = ComputeNumSignBits(U->getValue(), TD);
if (NS == 1)
- return ConservativeResult;
- return ConservativeResult.intersectWith(
+ return setSignedRange(U, ConservativeResult);
+ return setSignedRange(U, ConservativeResult.intersectWith(
ConstantRange(APInt::getSignedMinValue(BitWidth).ashr(NS - 1),
- APInt::getSignedMaxValue(BitWidth).ashr(NS - 1)+1));
+ APInt::getSignedMaxValue(BitWidth).ashr(NS - 1)+1)));
}
- return ConservativeResult;
+ return setSignedRange(S, ConservativeResult);
}
/// createSCEV - We know that there is no SCEV for the specified value.
@@ -3258,12 +3577,48 @@
Operator *U = cast<Operator>(V);
switch (Opcode) {
- case Instruction::Add:
- return getAddExpr(getSCEV(U->getOperand(0)),
- getSCEV(U->getOperand(1)));
- case Instruction::Mul:
- return getMulExpr(getSCEV(U->getOperand(0)),
- getSCEV(U->getOperand(1)));
+ case Instruction::Add: {
+ // The simple thing to do would be to just call getSCEV on both operands
+ // and call getAddExpr with the result. However if we're looking at a
+ // bunch of things all added together, this can be quite inefficient,
+ // because it leads to N-1 getAddExpr calls for N ultimate operands.
+ // Instead, gather up all the operands and make a single getAddExpr call.
+ // LLVM IR canonical form means we need only traverse the left operands.
+ SmallVector<const SCEV *, 4> AddOps;
+ AddOps.push_back(getSCEV(U->getOperand(1)));
+ for (Value *Op = U->getOperand(0); ; Op = U->getOperand(0)) {
+ unsigned Opcode = Op->getValueID() - Value::InstructionVal;
+ if (Opcode != Instruction::Add && Opcode != Instruction::Sub)
+ break;
+ U = cast<Operator>(Op);
+ const SCEV *Op1 = getSCEV(U->getOperand(1));
+ if (Opcode == Instruction::Sub)
+ AddOps.push_back(getNegativeSCEV(Op1));
+ else
+ AddOps.push_back(Op1);
+ }
+ AddOps.push_back(getSCEV(U->getOperand(0)));
+ SCEV::NoWrapFlags Flags = SCEV::FlagAnyWrap;
+ OverflowingBinaryOperator *OBO = cast<OverflowingBinaryOperator>(V);
+ if (OBO->hasNoSignedWrap())
+ setFlags(Flags, SCEV::FlagNSW);
+ if (OBO->hasNoUnsignedWrap())
+ setFlags(Flags, SCEV::FlagNUW);
+ return getAddExpr(AddOps, Flags);
+ }
+ case Instruction::Mul: {
+ // See the Add code above.
+ SmallVector<const SCEV *, 4> MulOps;
+ MulOps.push_back(getSCEV(U->getOperand(1)));
+ for (Value *Op = U->getOperand(0);
+ Op->getValueID() == Instruction::Mul + Value::InstructionVal;
+ Op = U->getOperand(0)) {
+ U = cast<Operator>(Op);
+ MulOps.push_back(getSCEV(U->getOperand(1)));
+ }
+ MulOps.push_back(getSCEV(U->getOperand(0)));
+ return getMulExpr(MulOps);
+ }
case Instruction::UDiv:
return getUDivExpr(getSCEV(U->getOperand(0)),
getSCEV(U->getOperand(1)));
@@ -3318,10 +3673,8 @@
// transfer the no-wrap flags, since an or won't introduce a wrap.
if (const SCEVAddRecExpr *NewAR = dyn_cast<SCEVAddRecExpr>(S)) {
const SCEVAddRecExpr *OldAR = cast<SCEVAddRecExpr>(LHS);
- if (OldAR->hasNoUnsignedWrap())
- const_cast<SCEVAddRecExpr *>(NewAR)->setHasNoUnsignedWrap(true);
- if (OldAR->hasNoSignedWrap())
- const_cast<SCEVAddRecExpr *>(NewAR)->setHasNoSignedWrap(true);
+ const_cast<SCEVAddRecExpr *>(NewAR)->setNoWrapFlags(
+ OldAR->getNoWrapFlags());
}
return S;
}
@@ -3349,9 +3702,9 @@
LCI->getValue() == CI->getValue())
if (const SCEVZeroExtendExpr *Z =
dyn_cast<SCEVZeroExtendExpr>(getSCEV(U->getOperand(0)))) {
- const Type *UTy = U->getType();
+ Type *UTy = U->getType();
const SCEV *Z0 = Z->getOperand();
- const Type *Z0Ty = Z0->getType();
+ Type *Z0Ty = Z0->getType();
unsigned Z0TySize = getTypeSizeInBits(Z0Ty);
// If C is a low-bits mask, the zero extend is serving to
@@ -3363,8 +3716,8 @@
// If C is a single bit, it may be in the sign-bit position
// before the zero-extend. In this case, represent the xor
// using an add, which is equivalent, and re-apply the zext.
- APInt Trunc = APInt(CI->getValue()).trunc(Z0TySize);
- if (APInt(Trunc).zext(getTypeSizeInBits(UTy)) == CI->getValue() &&
+ APInt Trunc = CI->getValue().trunc(Z0TySize);
+ if (Trunc.zext(getTypeSizeInBits(UTy)) == CI->getValue() &&
Trunc.isSignBit())
return getZeroExtendExpr(getAddExpr(Z0, getConstant(Trunc)),
UTy);
@@ -3525,7 +3878,7 @@
const SCEV *LDiff = getMinusSCEV(LA, LS);
const SCEV *RDiff = getMinusSCEV(RA, One);
if (LDiff == RDiff)
- return getAddExpr(getUMaxExpr(LS, One), LDiff);
+ return getAddExpr(getUMaxExpr(One, LS), LDiff);
}
break;
case ICmpInst::ICMP_EQ:
@@ -3540,7 +3893,7 @@
const SCEV *LDiff = getMinusSCEV(LA, One);
const SCEV *RDiff = getMinusSCEV(RA, LS);
if (LDiff == RDiff)
- return getAddExpr(getUMaxExpr(LS, One), LDiff);
+ return getAddExpr(getUMaxExpr(One, LS), LDiff);
}
break;
default:
@@ -3561,6 +3914,70 @@
// Iteration Count Computation Code
//
+/// getSmallConstantTripCount - Returns the maximum trip count of this loop as a
+/// normal unsigned value, if possible. Returns 0 if the trip count is unknown
+/// or not constant. Will also return 0 if the maximum trip count is very large
+/// (>= 2^32)
+unsigned ScalarEvolution::getSmallConstantTripCount(Loop *L,
+ BasicBlock *ExitBlock) {
+ const SCEVConstant *ExitCount =
+ dyn_cast<SCEVConstant>(getExitCount(L, ExitBlock));
+ if (!ExitCount)
+ return 0;
+
+ ConstantInt *ExitConst = ExitCount->getValue();
+
+ // Guard against huge trip counts.
+ if (ExitConst->getValue().getActiveBits() > 32)
+ return 0;
+
+ // In case of integer overflow, this returns 0, which is correct.
+ return ((unsigned)ExitConst->getZExtValue()) + 1;
+}
+
+/// getSmallConstantTripMultiple - Returns the largest constant divisor of the
+/// trip count of this loop as a normal unsigned value, if possible. This
+/// means that the actual trip count is always a multiple of the returned
+/// value (don't forget the trip count could very well be zero as well!).
+///
+/// Returns 1 if the trip count is unknown or not guaranteed to be the
+/// multiple of a constant (which is also the case if the trip count is simply
+/// constant, use getSmallConstantTripCount for that case), Will also return 1
+/// if the trip count is very large (>= 2^32).
+unsigned ScalarEvolution::getSmallConstantTripMultiple(Loop *L,
+ BasicBlock *ExitBlock) {
+ const SCEV *ExitCount = getExitCount(L, ExitBlock);
+ if (ExitCount == getCouldNotCompute())
+ return 1;
+
+ // Get the trip count from the BE count by adding 1.
+ const SCEV *TCMul = getAddExpr(ExitCount,
+ getConstant(ExitCount->getType(), 1));
+ // FIXME: SCEV distributes multiplication as V1*C1 + V2*C1. We could attempt
+ // to factor simple cases.
+ if (const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(TCMul))
+ TCMul = Mul->getOperand(0);
+
+ const SCEVConstant *MulC = dyn_cast<SCEVConstant>(TCMul);
+ if (!MulC)
+ return 1;
+
+ ConstantInt *Result = MulC->getValue();
+
+ // Guard against huge trip counts.
+ if (!Result || Result->getValue().getActiveBits() > 32)
+ return 1;
+
+ return (unsigned)Result->getZExtValue();
+}
+
+// getExitCount - Get the expression for the number of loop iterations for which
+// this loop is guaranteed not to exit via ExitintBlock. Otherwise return
+// SCEVCouldNotCompute.
+const SCEV *ScalarEvolution::getExitCount(Loop *L, BasicBlock *ExitingBlock) {
+ return getBackedgeTakenInfo(L).getExact(ExitingBlock, this);
+}
+
/// getBackedgeTakenCount - If the specified loop has a predictable
/// backedge-taken count, return it, otherwise return a SCEVCouldNotCompute
/// object. The backedge-taken count is the number of times the loop header
@@ -3573,14 +3990,14 @@
/// hasLoopInvariantBackedgeTakenCount).
///
const SCEV *ScalarEvolution::getBackedgeTakenCount(const Loop *L) {
- return getBackedgeTakenInfo(L).Exact;
+ return getBackedgeTakenInfo(L).getExact(this);
}
/// getMaxBackedgeTakenCount - Similar to getBackedgeTakenCount, except
/// return the least SCEV value that is known never to be less than the
/// actual backedge taken count.
const SCEV *ScalarEvolution::getMaxBackedgeTakenCount(const Loop *L) {
- return getBackedgeTakenInfo(L).Max;
+ return getBackedgeTakenInfo(L).getMax(this);
}
/// PushLoopPHIs - Push PHI nodes in the header of the given loop
@@ -3597,68 +4014,76 @@
const ScalarEvolution::BackedgeTakenInfo &
ScalarEvolution::getBackedgeTakenInfo(const Loop *L) {
- // Initially insert a CouldNotCompute for this loop. If the insertion
+ // Initially insert an invalid entry for this loop. If the insertion
// succeeds, proceed to actually compute a backedge-taken count and
// update the value. The temporary CouldNotCompute value tells SCEV
// code elsewhere that it shouldn't attempt to request a new
// backedge-taken count, which could result in infinite recursion.
- std::pair<std::map<const Loop *, BackedgeTakenInfo>::iterator, bool> Pair =
- BackedgeTakenCounts.insert(std::make_pair(L, getCouldNotCompute()));
- if (Pair.second) {
- BackedgeTakenInfo BECount = ComputeBackedgeTakenCount(L);
- if (BECount.Exact != getCouldNotCompute()) {
- assert(BECount.Exact->isLoopInvariant(L) &&
- BECount.Max->isLoopInvariant(L) &&
- "Computed backedge-taken count isn't loop invariant for loop!");
- ++NumTripCountsComputed;
+ std::pair<DenseMap<const Loop *, BackedgeTakenInfo>::iterator, bool> Pair =
+ BackedgeTakenCounts.insert(std::make_pair(L, BackedgeTakenInfo()));
+ if (!Pair.second)
+ return Pair.first->second;
- // Update the value in the map.
- Pair.first->second = BECount;
- } else {
- if (BECount.Max != getCouldNotCompute())
- // Update the value in the map.
- Pair.first->second = BECount;
- if (isa<PHINode>(L->getHeader()->begin()))
- // Only count loops that have phi nodes as not being computable.
- ++NumTripCountsNotComputed;
- }
+ // ComputeBackedgeTakenCount may allocate memory for its result. Inserting it
+ // into the BackedgeTakenCounts map transfers ownership. Otherwise, the result
+ // must be cleared in this scope.
+ BackedgeTakenInfo Result = ComputeBackedgeTakenCount(L);
- // Now that we know more about the trip count for this loop, forget any
- // existing SCEV values for PHI nodes in this loop since they are only
- // conservative estimates made without the benefit of trip count
- // information. This is similar to the code in forgetLoop, except that
- // it handles SCEVUnknown PHI nodes specially.
- if (BECount.hasAnyInfo()) {
- SmallVector<Instruction *, 16> Worklist;
- PushLoopPHIs(L, Worklist);
+ if (Result.getExact(this) != getCouldNotCompute()) {
+ assert(isLoopInvariant(Result.getExact(this), L) &&
+ isLoopInvariant(Result.getMax(this), L) &&
+ "Computed backedge-taken count isn't loop invariant for loop!");
+ ++NumTripCountsComputed;
+ }
+ else if (Result.getMax(this) == getCouldNotCompute() &&
+ isa<PHINode>(L->getHeader()->begin())) {
+ // Only count loops that have phi nodes as not being computable.
+ ++NumTripCountsNotComputed;
+ }
- SmallPtrSet<Instruction *, 8> Visited;
- while (!Worklist.empty()) {
- Instruction *I = Worklist.pop_back_val();
- if (!Visited.insert(I)) continue;
+ // Now that we know more about the trip count for this loop, forget any
+ // existing SCEV values for PHI nodes in this loop since they are only
+ // conservative estimates made without the benefit of trip count
+ // information. This is similar to the code in forgetLoop, except that
+ // it handles SCEVUnknown PHI nodes specially.
+ if (Result.hasAnyInfo()) {
+ SmallVector<Instruction *, 16> Worklist;
+ PushLoopPHIs(L, Worklist);
- std::map<SCEVCallbackVH, const SCEV *>::iterator It =
- Scalars.find(static_cast<Value *>(I));
- if (It != Scalars.end()) {
- // SCEVUnknown for a PHI either means that it has an unrecognized
- // structure, or it's a PHI that's in the progress of being computed
- // by createNodeForPHI. In the former case, additional loop trip
- // count information isn't going to change anything. In the later
- // case, createNodeForPHI will perform the necessary updates on its
- // own when it gets to that point.
- if (!isa<PHINode>(I) || !isa<SCEVUnknown>(It->second)) {
- ValuesAtScopes.erase(It->second);
- Scalars.erase(It);
- }
- if (PHINode *PN = dyn_cast<PHINode>(I))
- ConstantEvolutionLoopExitValue.erase(PN);
+ SmallPtrSet<Instruction *, 8> Visited;
+ while (!Worklist.empty()) {
+ Instruction *I = Worklist.pop_back_val();
+ if (!Visited.insert(I)) continue;
+
+ ValueExprMapType::iterator It =
+ ValueExprMap.find(static_cast<Value *>(I));
+ if (It != ValueExprMap.end()) {
+ const SCEV *Old = It->second;
+
+ // SCEVUnknown for a PHI either means that it has an unrecognized
+ // structure, or it's a PHI that's in the progress of being computed
+ // by createNodeForPHI. In the former case, additional loop trip
+ // count information isn't going to change anything. In the later
+ // case, createNodeForPHI will perform the necessary updates on its
+ // own when it gets to that point.
+ if (!isa<PHINode>(I) || !isa<SCEVUnknown>(Old)) {
+ forgetMemoizedResults(Old);
+ ValueExprMap.erase(It);
}
-
- PushDefUseChildren(I, Worklist);
+ if (PHINode *PN = dyn_cast<PHINode>(I))
+ ConstantEvolutionLoopExitValue.erase(PN);
}
+
+ PushDefUseChildren(I, Worklist);
}
}
- return Pair.first->second;
+
+ // Re-lookup the insert position, since the call to
+ // ComputeBackedgeTakenCount above could result in a
+ // recusive call to getBackedgeTakenInfo (on a different
+ // loop), which would invalidate the iterator computed
+ // earlier.
+ return BackedgeTakenCounts.find(L)->second = Result;
}
/// forgetLoop - This method should be called by the client when it has
@@ -3666,7 +4091,12 @@
/// compute a trip count, or if the loop is deleted.
void ScalarEvolution::forgetLoop(const Loop *L) {
// Drop any stored trip count value.
- BackedgeTakenCounts.erase(L);
+ DenseMap<const Loop*, BackedgeTakenInfo>::iterator BTCPos =
+ BackedgeTakenCounts.find(L);
+ if (BTCPos != BackedgeTakenCounts.end()) {
+ BTCPos->second.clear();
+ BackedgeTakenCounts.erase(BTCPos);
+ }
// Drop information about expressions based on loop-header PHIs.
SmallVector<Instruction *, 16> Worklist;
@@ -3677,17 +4107,21 @@
Instruction *I = Worklist.pop_back_val();
if (!Visited.insert(I)) continue;
- std::map<SCEVCallbackVH, const SCEV *>::iterator It =
- Scalars.find(static_cast<Value *>(I));
- if (It != Scalars.end()) {
- ValuesAtScopes.erase(It->second);
- Scalars.erase(It);
+ ValueExprMapType::iterator It = ValueExprMap.find(static_cast<Value *>(I));
+ if (It != ValueExprMap.end()) {
+ forgetMemoizedResults(It->second);
+ ValueExprMap.erase(It);
if (PHINode *PN = dyn_cast<PHINode>(I))
ConstantEvolutionLoopExitValue.erase(PN);
}
PushDefUseChildren(I, Worklist);
}
+
+ // Forget all contained loops too, to avoid dangling entries in the
+ // ValuesAtScopes map.
+ for (Loop::iterator I = L->begin(), E = L->end(); I != E; ++I)
+ forgetLoop(*I);
}
/// forgetValue - This method should be called by the client when it has
@@ -3706,11 +4140,10 @@
I = Worklist.pop_back_val();
if (!Visited.insert(I)) continue;
- std::map<SCEVCallbackVH, const SCEV *>::iterator It =
- Scalars.find(static_cast<Value *>(I));
- if (It != Scalars.end()) {
- ValuesAtScopes.erase(It->second);
- Scalars.erase(It);
+ ValueExprMapType::iterator It = ValueExprMap.find(static_cast<Value *>(I));
+ if (It != ValueExprMap.end()) {
+ forgetMemoizedResults(It->second);
+ ValueExprMap.erase(It);
if (PHINode *PN = dyn_cast<PHINode>(I))
ConstantEvolutionLoopExitValue.erase(PN);
}
@@ -3719,6 +4152,85 @@
}
}
+/// getExact - Get the exact loop backedge taken count considering all loop
+/// exits. If all exits are computable, this is the minimum computed count.
+const SCEV *
+ScalarEvolution::BackedgeTakenInfo::getExact(ScalarEvolution *SE) const {
+ // If any exits were not computable, the loop is not computable.
+ if (!ExitNotTaken.isCompleteList()) return SE->getCouldNotCompute();
+
+ // We need at least one computable exit.
+ if (!ExitNotTaken.ExitingBlock) return SE->getCouldNotCompute();
+ assert(ExitNotTaken.ExactNotTaken && "uninitialized not-taken info");
+
+ const SCEV *BECount = 0;
+ for (const ExitNotTakenInfo *ENT = &ExitNotTaken;
+ ENT != 0; ENT = ENT->getNextExit()) {
+
+ assert(ENT->ExactNotTaken != SE->getCouldNotCompute() && "bad exit SCEV");
+
+ if (!BECount)
+ BECount = ENT->ExactNotTaken;
+ else
+ BECount = SE->getUMinFromMismatchedTypes(BECount, ENT->ExactNotTaken);
+ }
+ assert(BECount && "Invalid not taken count for loop exit");
+ return BECount;
+}
+
+/// getExact - Get the exact not taken count for this loop exit.
+const SCEV *
+ScalarEvolution::BackedgeTakenInfo::getExact(BasicBlock *ExitingBlock,
+ ScalarEvolution *SE) const {
+ for (const ExitNotTakenInfo *ENT = &ExitNotTaken;
+ ENT != 0; ENT = ENT->getNextExit()) {
+
+ if (ENT->ExitingBlock == ExitingBlock)
+ return ENT->ExactNotTaken;
+ }
+ return SE->getCouldNotCompute();
+}
+
+/// getMax - Get the max backedge taken count for the loop.
+const SCEV *
+ScalarEvolution::BackedgeTakenInfo::getMax(ScalarEvolution *SE) const {
+ return Max ? Max : SE->getCouldNotCompute();
+}
+
+/// Allocate memory for BackedgeTakenInfo and copy the not-taken count of each
+/// computable exit into a persistent ExitNotTakenInfo array.
+ScalarEvolution::BackedgeTakenInfo::BackedgeTakenInfo(
+ SmallVectorImpl< std::pair<BasicBlock *, const SCEV *> > &ExitCounts,
+ bool Complete, const SCEV *MaxCount) : Max(MaxCount) {
+
+ if (!Complete)
+ ExitNotTaken.setIncomplete();
+
+ unsigned NumExits = ExitCounts.size();
+ if (NumExits == 0) return;
+
+ ExitNotTaken.ExitingBlock = ExitCounts[0].first;
+ ExitNotTaken.ExactNotTaken = ExitCounts[0].second;
+ if (NumExits == 1) return;
+
+ // Handle the rare case of multiple computable exits.
+ ExitNotTakenInfo *ENT = new ExitNotTakenInfo[NumExits-1];
+
+ ExitNotTakenInfo *PrevENT = &ExitNotTaken;
+ for (unsigned i = 1; i < NumExits; ++i, PrevENT = ENT, ++ENT) {
+ PrevENT->setNextExit(ENT);
+ ENT->ExitingBlock = ExitCounts[i].first;
+ ENT->ExactNotTaken = ExitCounts[i].second;
+ }
+}
+
+/// clear - Invalidate this result and free the ExitNotTakenInfo array.
+void ScalarEvolution::BackedgeTakenInfo::clear() {
+ ExitNotTaken.ExitingBlock = 0;
+ ExitNotTaken.ExactNotTaken = 0;
+ delete[] ExitNotTaken.getNextExit();
+}
+
/// ComputeBackedgeTakenCount - Compute the number of times the backedge
/// of the specified loop will execute.
ScalarEvolution::BackedgeTakenInfo
@@ -3727,38 +4239,31 @@
L->getExitingBlocks(ExitingBlocks);
// Examine all exits and pick the most conservative values.
- const SCEV *BECount = getCouldNotCompute();
const SCEV *MaxBECount = getCouldNotCompute();
- bool CouldNotComputeBECount = false;
+ bool CouldComputeBECount = true;
+ SmallVector<std::pair<BasicBlock *, const SCEV *>, 4> ExitCounts;
for (unsigned i = 0, e = ExitingBlocks.size(); i != e; ++i) {
- BackedgeTakenInfo NewBTI =
- ComputeBackedgeTakenCountFromExit(L, ExitingBlocks[i]);
-
- if (NewBTI.Exact == getCouldNotCompute()) {
+ ExitLimit EL = ComputeExitLimit(L, ExitingBlocks[i]);
+ if (EL.Exact == getCouldNotCompute())
// We couldn't compute an exact value for this exit, so
// we won't be able to compute an exact value for the loop.
- CouldNotComputeBECount = true;
- BECount = getCouldNotCompute();
- } else if (!CouldNotComputeBECount) {
- if (BECount == getCouldNotCompute())
- BECount = NewBTI.Exact;
- else
- BECount = getUMinFromMismatchedTypes(BECount, NewBTI.Exact);
- }
+ CouldComputeBECount = false;
+ else
+ ExitCounts.push_back(std::make_pair(ExitingBlocks[i], EL.Exact));
+
if (MaxBECount == getCouldNotCompute())
- MaxBECount = NewBTI.Max;
- else if (NewBTI.Max != getCouldNotCompute())
- MaxBECount = getUMinFromMismatchedTypes(MaxBECount, NewBTI.Max);
+ MaxBECount = EL.Max;
+ else if (EL.Max != getCouldNotCompute())
+ MaxBECount = getUMinFromMismatchedTypes(MaxBECount, EL.Max);
}
- return BackedgeTakenInfo(BECount, MaxBECount);
+ return BackedgeTakenInfo(ExitCounts, CouldComputeBECount, MaxBECount);
}
-/// ComputeBackedgeTakenCountFromExit - Compute the number of times the backedge
-/// of the specified loop will execute if it exits via the specified block.
-ScalarEvolution::BackedgeTakenInfo
-ScalarEvolution::ComputeBackedgeTakenCountFromExit(const Loop *L,
- BasicBlock *ExitingBlock) {
+/// ComputeExitLimit - Compute the number of times the backedge of the specified
+/// loop will execute if it exits via the specified block.
+ScalarEvolution::ExitLimit
+ScalarEvolution::ComputeExitLimit(const Loop *L, BasicBlock *ExitingBlock) {
// Okay, we've chosen an exiting block. See what condition causes us to
// exit at this block.
@@ -3816,95 +4321,91 @@
}
// Proceed to the next level to examine the exit condition expression.
- return ComputeBackedgeTakenCountFromExitCond(L, ExitBr->getCondition(),
- ExitBr->getSuccessor(0),
- ExitBr->getSuccessor(1));
+ return ComputeExitLimitFromCond(L, ExitBr->getCondition(),
+ ExitBr->getSuccessor(0),
+ ExitBr->getSuccessor(1));
}
-/// ComputeBackedgeTakenCountFromExitCond - Compute the number of times the
+/// ComputeExitLimitFromCond - Compute the number of times the
/// backedge of the specified loop will execute if its exit condition
/// were a conditional branch of ExitCond, TBB, and FBB.
-ScalarEvolution::BackedgeTakenInfo
-ScalarEvolution::ComputeBackedgeTakenCountFromExitCond(const Loop *L,
- Value *ExitCond,
- BasicBlock *TBB,
- BasicBlock *FBB) {
+ScalarEvolution::ExitLimit
+ScalarEvolution::ComputeExitLimitFromCond(const Loop *L,
+ Value *ExitCond,
+ BasicBlock *TBB,
+ BasicBlock *FBB) {
// Check if the controlling expression for this loop is an And or Or.
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(ExitCond)) {
if (BO->getOpcode() == Instruction::And) {
// Recurse on the operands of the and.
- BackedgeTakenInfo BTI0 =
- ComputeBackedgeTakenCountFromExitCond(L, BO->getOperand(0), TBB, FBB);
- BackedgeTakenInfo BTI1 =
- ComputeBackedgeTakenCountFromExitCond(L, BO->getOperand(1), TBB, FBB);
+ ExitLimit EL0 = ComputeExitLimitFromCond(L, BO->getOperand(0), TBB, FBB);
+ ExitLimit EL1 = ComputeExitLimitFromCond(L, BO->getOperand(1), TBB, FBB);
const SCEV *BECount = getCouldNotCompute();
const SCEV *MaxBECount = getCouldNotCompute();
if (L->contains(TBB)) {
// Both conditions must be true for the loop to continue executing.
// Choose the less conservative count.
- if (BTI0.Exact == getCouldNotCompute() ||
- BTI1.Exact == getCouldNotCompute())
+ if (EL0.Exact == getCouldNotCompute() ||
+ EL1.Exact == getCouldNotCompute())
BECount = getCouldNotCompute();
else
- BECount = getUMinFromMismatchedTypes(BTI0.Exact, BTI1.Exact);
- if (BTI0.Max == getCouldNotCompute())
- MaxBECount = BTI1.Max;
- else if (BTI1.Max == getCouldNotCompute())
- MaxBECount = BTI0.Max;
+ BECount = getUMinFromMismatchedTypes(EL0.Exact, EL1.Exact);
+ if (EL0.Max == getCouldNotCompute())
+ MaxBECount = EL1.Max;
+ else if (EL1.Max == getCouldNotCompute())
+ MaxBECount = EL0.Max;
else
- MaxBECount = getUMinFromMismatchedTypes(BTI0.Max, BTI1.Max);
+ MaxBECount = getUMinFromMismatchedTypes(EL0.Max, EL1.Max);
} else {
// Both conditions must be true at the same time for the loop to exit.
// For now, be conservative.
assert(L->contains(FBB) && "Loop block has no successor in loop!");
- if (BTI0.Max == BTI1.Max)
- MaxBECount = BTI0.Max;
- if (BTI0.Exact == BTI1.Exact)
- BECount = BTI0.Exact;
+ if (EL0.Max == EL1.Max)
+ MaxBECount = EL0.Max;
+ if (EL0.Exact == EL1.Exact)
+ BECount = EL0.Exact;
}
- return BackedgeTakenInfo(BECount, MaxBECount);
+ return ExitLimit(BECount, MaxBECount);
}
if (BO->getOpcode() == Instruction::Or) {
// Recurse on the operands of the or.
- BackedgeTakenInfo BTI0 =
- ComputeBackedgeTakenCountFromExitCond(L, BO->getOperand(0), TBB, FBB);
- BackedgeTakenInfo BTI1 =
- ComputeBackedgeTakenCountFromExitCond(L, BO->getOperand(1), TBB, FBB);
+ ExitLimit EL0 = ComputeExitLimitFromCond(L, BO->getOperand(0), TBB, FBB);
+ ExitLimit EL1 = ComputeExitLimitFromCond(L, BO->getOperand(1), TBB, FBB);
const SCEV *BECount = getCouldNotCompute();
const SCEV *MaxBECount = getCouldNotCompute();
if (L->contains(FBB)) {
// Both conditions must be false for the loop to continue executing.
// Choose the less conservative count.
- if (BTI0.Exact == getCouldNotCompute() ||
- BTI1.Exact == getCouldNotCompute())
+ if (EL0.Exact == getCouldNotCompute() ||
+ EL1.Exact == getCouldNotCompute())
BECount = getCouldNotCompute();
else
- BECount = getUMinFromMismatchedTypes(BTI0.Exact, BTI1.Exact);
- if (BTI0.Max == getCouldNotCompute())
- MaxBECount = BTI1.Max;
- else if (BTI1.Max == getCouldNotCompute())
- MaxBECount = BTI0.Max;
+ BECount = getUMinFromMismatchedTypes(EL0.Exact, EL1.Exact);
+ if (EL0.Max == getCouldNotCompute())
+ MaxBECount = EL1.Max;
+ else if (EL1.Max == getCouldNotCompute())
+ MaxBECount = EL0.Max;
else
- MaxBECount = getUMinFromMismatchedTypes(BTI0.Max, BTI1.Max);
+ MaxBECount = getUMinFromMismatchedTypes(EL0.Max, EL1.Max);
} else {
// Both conditions must be false at the same time for the loop to exit.
// For now, be conservative.
assert(L->contains(TBB) && "Loop block has no successor in loop!");
- if (BTI0.Max == BTI1.Max)
- MaxBECount = BTI0.Max;
- if (BTI0.Exact == BTI1.Exact)
- BECount = BTI0.Exact;
+ if (EL0.Max == EL1.Max)
+ MaxBECount = EL0.Max;
+ if (EL0.Exact == EL1.Exact)
+ BECount = EL0.Exact;
}
- return BackedgeTakenInfo(BECount, MaxBECount);
+ return ExitLimit(BECount, MaxBECount);
}
}
// With an icmp, it may be feasible to compute an exact backedge-taken count.
// Proceed to the next level to examine the icmp.
if (ICmpInst *ExitCondICmp = dyn_cast<ICmpInst>(ExitCond))
- return ComputeBackedgeTakenCountFromExitCondICmp(L, ExitCondICmp, TBB, FBB);
+ return ComputeExitLimitFromICmp(L, ExitCondICmp, TBB, FBB);
// Check for a constant condition. These are normally stripped out by
// SimplifyCFG, but ScalarEvolution may be used by a pass which wishes to
@@ -3920,17 +4421,17 @@
}
// If it's not an integer or pointer comparison then compute it the hard way.
- return ComputeBackedgeTakenCountExhaustively(L, ExitCond, !L->contains(TBB));
+ return ComputeExitCountExhaustively(L, ExitCond, !L->contains(TBB));
}
-/// ComputeBackedgeTakenCountFromExitCondICmp - Compute the number of times the
+/// ComputeExitLimitFromICmp - Compute the number of times the
/// backedge of the specified loop will execute if its exit condition
/// were a conditional branch of the ICmpInst ExitCond, TBB, and FBB.
-ScalarEvolution::BackedgeTakenInfo
-ScalarEvolution::ComputeBackedgeTakenCountFromExitCondICmp(const Loop *L,
- ICmpInst *ExitCond,
- BasicBlock *TBB,
- BasicBlock *FBB) {
+ScalarEvolution::ExitLimit
+ScalarEvolution::ComputeExitLimitFromICmp(const Loop *L,
+ ICmpInst *ExitCond,
+ BasicBlock *TBB,
+ BasicBlock *FBB) {
// If the condition was exit on true, convert the condition to exit on false
ICmpInst::Predicate Cond;
@@ -3942,8 +4443,8 @@
// Handle common loops like: for (X = "string"; *X; ++X)
if (LoadInst *LI = dyn_cast<LoadInst>(ExitCond->getOperand(0)))
if (Constant *RHS = dyn_cast<Constant>(ExitCond->getOperand(1))) {
- BackedgeTakenInfo ItCnt =
- ComputeLoadConstantCompareBackedgeTakenCount(LI, RHS, L, Cond);
+ ExitLimit ItCnt =
+ ComputeLoadConstantCompareExitLimit(LI, RHS, L, Cond);
if (ItCnt.hasAnyInfo())
return ItCnt;
}
@@ -3957,7 +4458,7 @@
// At this point, we would like to compute how many iterations of the
// loop the predicate will return true for these inputs.
- if (LHS->isLoopInvariant(L) && !RHS->isLoopInvariant(L)) {
+ if (isLoopInvariant(LHS, L) && !isLoopInvariant(RHS, L)) {
// If there is a loop-invariant, force it into the RHS.
std::swap(LHS, RHS);
Cond = ICmpInst::getSwappedPredicate(Cond);
@@ -3982,36 +4483,36 @@
switch (Cond) {
case ICmpInst::ICMP_NE: { // while (X != Y)
// Convert to: while (X-Y != 0)
- BackedgeTakenInfo BTI = HowFarToZero(getMinusSCEV(LHS, RHS), L);
- if (BTI.hasAnyInfo()) return BTI;
+ ExitLimit EL = HowFarToZero(getMinusSCEV(LHS, RHS), L);
+ if (EL.hasAnyInfo()) return EL;
break;
}
case ICmpInst::ICMP_EQ: { // while (X == Y)
// Convert to: while (X-Y == 0)
- BackedgeTakenInfo BTI = HowFarToNonZero(getMinusSCEV(LHS, RHS), L);
- if (BTI.hasAnyInfo()) return BTI;
+ ExitLimit EL = HowFarToNonZero(getMinusSCEV(LHS, RHS), L);
+ if (EL.hasAnyInfo()) return EL;
break;
}
case ICmpInst::ICMP_SLT: {
- BackedgeTakenInfo BTI = HowManyLessThans(LHS, RHS, L, true);
- if (BTI.hasAnyInfo()) return BTI;
+ ExitLimit EL = HowManyLessThans(LHS, RHS, L, true);
+ if (EL.hasAnyInfo()) return EL;
break;
}
case ICmpInst::ICMP_SGT: {
- BackedgeTakenInfo BTI = HowManyLessThans(getNotSCEV(LHS),
+ ExitLimit EL = HowManyLessThans(getNotSCEV(LHS),
getNotSCEV(RHS), L, true);
- if (BTI.hasAnyInfo()) return BTI;
+ if (EL.hasAnyInfo()) return EL;
break;
}
case ICmpInst::ICMP_ULT: {
- BackedgeTakenInfo BTI = HowManyLessThans(LHS, RHS, L, false);
- if (BTI.hasAnyInfo()) return BTI;
+ ExitLimit EL = HowManyLessThans(LHS, RHS, L, false);
+ if (EL.hasAnyInfo()) return EL;
break;
}
case ICmpInst::ICMP_UGT: {
- BackedgeTakenInfo BTI = HowManyLessThans(getNotSCEV(LHS),
+ ExitLimit EL = HowManyLessThans(getNotSCEV(LHS),
getNotSCEV(RHS), L, false);
- if (BTI.hasAnyInfo()) return BTI;
+ if (EL.hasAnyInfo()) return EL;
break;
}
default:
@@ -4025,8 +4526,7 @@
#endif
break;
}
- return
- ComputeBackedgeTakenCountExhaustively(L, ExitCond, !L->contains(TBB));
+ return ComputeExitCountExhaustively(L, ExitCond, !L->contains(TBB));
}
static ConstantInt *
@@ -4056,10 +4556,10 @@
if (Idx >= CA->getNumOperands()) return 0; // Bogus program
Init = cast<Constant>(CA->getOperand(Idx));
} else if (isa<ConstantAggregateZero>(Init)) {
- if (const StructType *STy = dyn_cast<StructType>(Init->getType())) {
+ if (StructType *STy = dyn_cast<StructType>(Init->getType())) {
assert(Idx < STy->getNumElements() && "Bad struct index!");
Init = Constant::getNullValue(STy->getElementType(Idx));
- } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Init->getType())) {
+ } else if (ArrayType *ATy = dyn_cast<ArrayType>(Init->getType())) {
if (Idx >= ATy->getNumElements()) return 0; // Bogus program
Init = Constant::getNullValue(ATy->getElementType());
} else {
@@ -4073,15 +4573,16 @@
return Init;
}
-/// ComputeLoadConstantCompareBackedgeTakenCount - Given an exit condition of
+/// ComputeLoadConstantCompareExitLimit - Given an exit condition of
/// 'icmp op load X, cst', try to see if we can compute the backedge
/// execution count.
-ScalarEvolution::BackedgeTakenInfo
-ScalarEvolution::ComputeLoadConstantCompareBackedgeTakenCount(
- LoadInst *LI,
- Constant *RHS,
- const Loop *L,
- ICmpInst::Predicate predicate) {
+ScalarEvolution::ExitLimit
+ScalarEvolution::ComputeLoadConstantCompareExitLimit(
+ LoadInst *LI,
+ Constant *RHS,
+ const Loop *L,
+ ICmpInst::Predicate predicate) {
+
if (LI->isVolatile()) return getCouldNotCompute();
// Check to see if the loaded pointer is a getelementptr of a global.
@@ -4119,7 +4620,7 @@
// We can only recognize very limited forms of loop index expressions, in
// particular, only affine AddRec's like {C1,+,C2}.
const SCEVAddRecExpr *IdxExpr = dyn_cast<SCEVAddRecExpr>(Idx);
- if (!IdxExpr || !IdxExpr->isAffine() || IdxExpr->isLoopInvariant(L) ||
+ if (!IdxExpr || !IdxExpr->isAffine() || isLoopInvariant(IdxExpr, L) ||
!isa<SCEVConstant>(IdxExpr->getOperand(0)) ||
!isa<SCEVConstant>(IdxExpr->getOperand(1)))
return getCouldNotCompute();
@@ -4166,69 +4667,117 @@
return false;
}
+/// Determine whether this instruction can constant evolve within this loop
+/// assuming its operands can all constant evolve.
+static bool canConstantEvolve(Instruction *I, const Loop *L) {
+ // An instruction outside of the loop can't be derived from a loop PHI.
+ if (!L->contains(I)) return false;
+
+ if (isa<PHINode>(I)) {
+ if (L->getHeader() == I->getParent())
+ return true;
+ else
+ // We don't currently keep track of the control flow needed to evaluate
+ // PHIs, so we cannot handle PHIs inside of loops.
+ return false;
+ }
+
+ // If we won't be able to constant fold this expression even if the operands
+ // are constants, bail early.
+ return CanConstantFold(I);
+}
+
+/// getConstantEvolvingPHIOperands - Implement getConstantEvolvingPHI by
+/// recursing through each instruction operand until reaching a loop header phi.
+static PHINode *
+getConstantEvolvingPHIOperands(Instruction *UseInst, const Loop *L,
+ DenseMap<Instruction *, PHINode *> &PHIMap) {
+
+ // Otherwise, we can evaluate this instruction if all of its operands are
+ // constant or derived from a PHI node themselves.
+ PHINode *PHI = 0;
+ for (Instruction::op_iterator OpI = UseInst->op_begin(),
+ OpE = UseInst->op_end(); OpI != OpE; ++OpI) {
+
+ if (isa<Constant>(*OpI)) continue;
+
+ Instruction *OpInst = dyn_cast<Instruction>(*OpI);
+ if (!OpInst || !canConstantEvolve(OpInst, L)) return 0;
+
+ PHINode *P = dyn_cast<PHINode>(OpInst);
+ if (!P)
+ // If this operand is already visited, reuse the prior result.
+ // We may have P != PHI if this is the deepest point at which the
+ // inconsistent paths meet.
+ P = PHIMap.lookup(OpInst);
+ if (!P) {
+ // Recurse and memoize the results, whether a phi is found or not.
+ // This recursive call invalidates pointers into PHIMap.
+ P = getConstantEvolvingPHIOperands(OpInst, L, PHIMap);
+ PHIMap[OpInst] = P;
+ }
+ if (P == 0) return 0; // Not evolving from PHI
+ if (PHI && PHI != P) return 0; // Evolving from multiple different PHIs.
+ PHI = P;
+ }
+ // This is a expression evolving from a constant PHI!
+ return PHI;
+}
+
/// getConstantEvolvingPHI - Given an LLVM value and a loop, return a PHI node
/// in the loop that V is derived from. We allow arbitrary operations along the
/// way, but the operands of an operation must either be constants or a value
/// derived from a constant PHI. If this expression does not fit with these
/// constraints, return null.
static PHINode *getConstantEvolvingPHI(Value *V, const Loop *L) {
- // If this is not an instruction, or if this is an instruction outside of the
- // loop, it can't be derived from a loop PHI.
Instruction *I = dyn_cast<Instruction>(V);
- if (I == 0 || !L->contains(I)) return 0;
+ if (I == 0 || !canConstantEvolve(I, L)) return 0;
if (PHINode *PN = dyn_cast<PHINode>(I)) {
- if (L->getHeader() == I->getParent())
- return PN;
- else
- // We don't currently keep track of the control flow needed to evaluate
- // PHIs, so we cannot handle PHIs inside of loops.
- return 0;
+ return PN;
}
- // If we won't be able to constant fold this expression even if the operands
- // are constants, return early.
- if (!CanConstantFold(I)) return 0;
-
- // Otherwise, we can evaluate this instruction if all of its operands are
- // constant or derived from a PHI node themselves.
- PHINode *PHI = 0;
- for (unsigned Op = 0, e = I->getNumOperands(); Op != e; ++Op)
- if (!isa<Constant>(I->getOperand(Op))) {
- PHINode *P = getConstantEvolvingPHI(I->getOperand(Op), L);
- if (P == 0) return 0; // Not evolving from PHI
- if (PHI == 0)
- PHI = P;
- else if (PHI != P)
- return 0; // Evolving from multiple different PHIs.
- }
-
- // This is a expression evolving from a constant PHI!
- return PHI;
+ // Record non-constant instructions contained by the loop.
+ DenseMap<Instruction *, PHINode *> PHIMap;
+ return getConstantEvolvingPHIOperands(I, L, PHIMap);
}
/// EvaluateExpression - Given an expression that passes the
/// getConstantEvolvingPHI predicate, evaluate its value assuming the PHI node
/// in the loop has the value PHIVal. If we can't fold this expression for some
/// reason, return null.
-static Constant *EvaluateExpression(Value *V, Constant *PHIVal,
+static Constant *EvaluateExpression(Value *V, const Loop *L,
+ DenseMap<Instruction *, Constant *> &Vals,
const TargetData *TD) {
- if (isa<PHINode>(V)) return PHIVal;
+ // Convenient constant check, but redundant for recursive calls.
if (Constant *C = dyn_cast<Constant>(V)) return C;
+
Instruction *I = cast<Instruction>(V);
+ if (Constant *C = Vals.lookup(I)) return C;
+
+ assert(!isa<PHINode>(I) && "loop header phis should be mapped to constant");
+ assert(canConstantEvolve(I, L) && "cannot evaluate expression in this loop");
+ (void)L;
std::vector<Constant*> Operands(I->getNumOperands());
for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
- Operands[i] = EvaluateExpression(I->getOperand(i), PHIVal, TD);
- if (Operands[i] == 0) return 0;
+ Instruction *Operand = dyn_cast<Instruction>(I->getOperand(i));
+ if (!Operand) {
+ Operands[i] = dyn_cast<Constant>(I->getOperand(i));
+ if (!Operands[i]) return 0;
+ continue;
+ }
+ Constant *C = EvaluateExpression(Operand, L, Vals, TD);
+ Vals[Operand] = C;
+ if (!C) return 0;
+ Operands[i] = C;
}
if (const CmpInst *CI = dyn_cast<CmpInst>(I))
return ConstantFoldCompareInstOperands(CI->getPredicate(), Operands[0],
Operands[1], TD);
- return ConstantFoldInstOperands(I->getOpcode(), I->getType(),
- &Operands[0], Operands.size(), TD);
+ return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Operands, TD);
}
/// getConstantEvolutionLoopExitValue - If we know that the specified Phi is
@@ -4239,7 +4788,7 @@
ScalarEvolution::getConstantEvolutionLoopExitValue(PHINode *PN,
const APInt &BEs,
const Loop *L) {
- std::map<PHINode*, Constant*>::iterator I =
+ DenseMap<PHINode*, Constant*>::const_iterator I =
ConstantEvolutionLoopExitValue.find(PN);
if (I != ConstantEvolutionLoopExitValue.end())
return I->second;
@@ -4249,6 +4798,9 @@
Constant *&RetVal = ConstantEvolutionLoopExitValue[PN];
+ // FIXME: Nick's fix for PR11034 will seed constants for multiple header phis.
+ DenseMap<Instruction *, Constant *> CurrentIterVals;
+
// Since the loop is canonicalized, the PHI node must have two entries. One
// entry must be a constant (coming in from outside of the loop), and the
// second must be derived from the same PHI.
@@ -4257,6 +4809,7 @@
dyn_cast<Constant>(PN->getIncomingValue(!SecondIsBackedge));
if (StartCST == 0)
return RetVal = 0; // Must be a constant.
+ CurrentIterVals[PN] = StartCST;
Value *BEValue = PN->getIncomingValue(SecondIsBackedge);
if (getConstantEvolvingPHI(BEValue, L) != PN &&
@@ -4269,29 +4822,31 @@
unsigned NumIterations = BEs.getZExtValue(); // must be in range
unsigned IterationNum = 0;
- for (Constant *PHIVal = StartCST; ; ++IterationNum) {
+ for (; ; ++IterationNum) {
if (IterationNum == NumIterations)
- return RetVal = PHIVal; // Got exit value!
+ return RetVal = CurrentIterVals[PN]; // Got exit value!
// Compute the value of the PHI node for the next iteration.
- Constant *NextPHI = EvaluateExpression(BEValue, PHIVal, TD);
- if (NextPHI == PHIVal)
+ // EvaluateExpression adds non-phi values to the CurrentIterVals map.
+ Constant *NextPHI = EvaluateExpression(BEValue, L, CurrentIterVals, TD);
+ if (NextPHI == CurrentIterVals[PN])
return RetVal = NextPHI; // Stopped evolving!
if (NextPHI == 0)
return 0; // Couldn't evaluate!
- PHIVal = NextPHI;
+ DenseMap<Instruction *, Constant *> NextIterVals;
+ NextIterVals[PN] = NextPHI;
+ CurrentIterVals.swap(NextIterVals);
}
}
-/// ComputeBackedgeTakenCountExhaustively - If the loop is known to execute a
+/// ComputeExitCountExhaustively - If the loop is known to execute a
/// constant number of times (the condition evolves only from constants),
/// try to evaluate a few iterations of the loop until we get the exit
/// condition gets a value of ExitWhen (true or false). If we cannot
/// evaluate the trip count of the loop, return getCouldNotCompute().
-const SCEV *
-ScalarEvolution::ComputeBackedgeTakenCountExhaustively(const Loop *L,
- Value *Cond,
- bool ExitWhen) {
+const SCEV * ScalarEvolution::ComputeExitCountExhaustively(const Loop *L,
+ Value *Cond,
+ bool ExitWhen) {
PHINode *PN = getConstantEvolvingPHI(Cond, L);
if (PN == 0) return getCouldNotCompute();
@@ -4318,8 +4873,10 @@
unsigned MaxIterations = MaxBruteForceIterations; // Limit analysis.
for (Constant *PHIVal = StartCST;
IterationNum != MaxIterations; ++IterationNum) {
+ DenseMap<Instruction *, Constant *> PHIValMap;
+ PHIValMap[PN] = PHIVal;
ConstantInt *CondVal =
- dyn_cast_or_null<ConstantInt>(EvaluateExpression(Cond, PHIVal, TD));
+ dyn_cast_or_null<ConstantInt>(EvaluateExpression(Cond, L, PHIValMap, TD));
// Couldn't symbolically evaluate.
if (!CondVal) return getCouldNotCompute();
@@ -4330,7 +4887,7 @@
}
// Compute the value of the PHI node for the next iteration.
- Constant *NextPHI = EvaluateExpression(BEValue, PHIVal, TD);
+ Constant *NextPHI = EvaluateExpression(BEValue, L, PHIValMap, TD);
if (NextPHI == 0 || NextPHI == PHIVal)
return getCouldNotCompute();// Couldn't evaluate or not making progress...
PHIVal = NextPHI;
@@ -4438,7 +4995,7 @@
Operands[0], Operands[1], TD);
else
C = ConstantFoldInstOperands(I->getOpcode(), I->getType(),
- &Operands[0], Operands.size(), TD);
+ Operands, TD);
if (!C) return V;
return getSCEV(C);
}
@@ -4507,7 +5064,15 @@
for (++i; i != e; ++i)
NewOps.push_back(getSCEVAtScope(AddRec->getOperand(i), L));
- AddRec = cast<SCEVAddRecExpr>(getAddRecExpr(NewOps, AddRec->getLoop()));
+ const SCEV *FoldedRec =
+ getAddRecExpr(NewOps, AddRec->getLoop(),
+ AddRec->getNoWrapFlags(SCEV::FlagNW));
+ AddRec = dyn_cast<SCEVAddRecExpr>(FoldedRec);
+ // The addrec may be folded to a nonrecurrence, for example, if the
+ // induction variable is multiplied by zero after constant folding. Go
+ // ahead and return the folded value.
+ if (!AddRec)
+ return FoldedRec;
break;
}
@@ -4593,7 +5158,7 @@
// bit width during computations.
APInt AD = A.lshr(Mult2).zext(BW + 1); // AD = A / D
APInt Mod(BW + 1, 0);
- Mod.set(BW - Mult2); // Mod = N / D
+ Mod.setBit(BW - Mult2); // Mod = N / D
APInt I = AD.multiplicativeInverse(Mod);
// 4. Compute the minimum unsigned root of the equation:
@@ -4652,7 +5217,7 @@
// Compute the two solutions for the quadratic formula.
// The divisions must be performed as signed divisions.
APInt NegB(-B);
- APInt TwoA( A << 1 );
+ APInt TwoA(A << 1);
if (TwoA.isMinValue()) {
const SCEV *CNC = SE.getCouldNotCompute();
return std::make_pair(CNC, CNC);
@@ -4667,12 +5232,17 @@
return std::make_pair(SE.getConstant(Solution1),
SE.getConstant(Solution2));
- } // end APIntOps namespace
+ } // end APIntOps namespace
}
/// HowFarToZero - Return the number of times a backedge comparing the specified
/// value to zero will execute. If not computable, return CouldNotCompute.
-ScalarEvolution::BackedgeTakenInfo
+///
+/// This is only used for loops with a "x != y" exit test. The exit condition is
+/// now expressed as a single expression, V = x-y. So the exit test is
+/// effectively V != 0. We know and take advantage of the fact that this
+/// expression only being used in a comparison by zero context.
+ScalarEvolution::ExitLimit
ScalarEvolution::HowFarToZero(const SCEV *V, const Loop *L) {
// If the value is a constant
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(V)) {
@@ -4685,55 +5255,23 @@
if (!AddRec || AddRec->getLoop() != L)
return getCouldNotCompute();
- if (AddRec->isAffine()) {
- // If this is an affine expression, the execution count of this branch is
- // the minimum unsigned root of the following equation:
- //
- // Start + Step*N = 0 (mod 2^BW)
- //
- // equivalent to:
- //
- // Step*N = -Start (mod 2^BW)
- //
- // where BW is the common bit width of Start and Step.
-
- // Get the initial value for the loop.
- const SCEV *Start = getSCEVAtScope(AddRec->getStart(),
- L->getParentLoop());
- const SCEV *Step = getSCEVAtScope(AddRec->getOperand(1),
- L->getParentLoop());
-
- if (const SCEVConstant *StepC = dyn_cast<SCEVConstant>(Step)) {
- // For now we handle only constant steps.
-
- // First, handle unitary steps.
- if (StepC->getValue()->equalsInt(1)) // 1*N = -Start (mod 2^BW), so:
- return getNegativeSCEV(Start); // N = -Start (as unsigned)
- if (StepC->getValue()->isAllOnesValue()) // -1*N = -Start (mod 2^BW), so:
- return Start; // N = Start (as unsigned)
-
- // Then, try to solve the above equation provided that Start is constant.
- if (const SCEVConstant *StartC = dyn_cast<SCEVConstant>(Start))
- return SolveLinEquationWithOverflow(StepC->getValue()->getValue(),
- -StartC->getValue()->getValue(),
- *this);
- }
- } else if (AddRec->isQuadratic() && AddRec->getType()->isIntegerTy()) {
- // If this is a quadratic (3-term) AddRec {L,+,M,+,N}, find the roots of
- // the quadratic equation to solve it.
- std::pair<const SCEV *,const SCEV *> Roots = SolveQuadraticEquation(AddRec,
- *this);
+ // If this is a quadratic (3-term) AddRec {L,+,M,+,N}, find the roots of
+ // the quadratic equation to solve it.
+ if (AddRec->isQuadratic() && AddRec->getType()->isIntegerTy()) {
+ std::pair<const SCEV *,const SCEV *> Roots =
+ SolveQuadraticEquation(AddRec, *this);
const SCEVConstant *R1 = dyn_cast<SCEVConstant>(Roots.first);
const SCEVConstant *R2 = dyn_cast<SCEVConstant>(Roots.second);
- if (R1) {
+ if (R1 && R2) {
#if 0
dbgs() << "HFTZ: " << *V << " - sol#1: " << *R1
<< " sol#2: " << *R2 << "\n";
#endif
// Pick the smallest positive root value.
if (ConstantInt *CB =
- dyn_cast<ConstantInt>(ConstantExpr::getICmp(ICmpInst::ICMP_ULT,
- R1->getValue(), R2->getValue()))) {
+ dyn_cast<ConstantInt>(ConstantExpr::getICmp(CmpInst::ICMP_ULT,
+ R1->getValue(),
+ R2->getValue()))) {
if (CB->getZExtValue() == false)
std::swap(R1, R2); // R1 is the minimum root now.
@@ -4745,15 +5283,89 @@
return R1; // We found a quadratic root!
}
}
+ return getCouldNotCompute();
}
+ // Otherwise we can only handle this if it is affine.
+ if (!AddRec->isAffine())
+ return getCouldNotCompute();
+
+ // If this is an affine expression, the execution count of this branch is
+ // the minimum unsigned root of the following equation:
+ //
+ // Start + Step*N = 0 (mod 2^BW)
+ //
+ // equivalent to:
+ //
+ // Step*N = -Start (mod 2^BW)
+ //
+ // where BW is the common bit width of Start and Step.
+
+ // Get the initial value for the loop.
+ const SCEV *Start = getSCEVAtScope(AddRec->getStart(), L->getParentLoop());
+ const SCEV *Step = getSCEVAtScope(AddRec->getOperand(1), L->getParentLoop());
+
+ // For now we handle only constant steps.
+ //
+ // TODO: Handle a nonconstant Step given AddRec<NUW>. If the
+ // AddRec is NUW, then (in an unsigned sense) it cannot be counting up to wrap
+ // to 0, it must be counting down to equal 0. Consequently, N = Start / -Step.
+ // We have not yet seen any such cases.
+ const SCEVConstant *StepC = dyn_cast<SCEVConstant>(Step);
+ if (StepC == 0)
+ return getCouldNotCompute();
+
+ // For positive steps (counting up until unsigned overflow):
+ // N = -Start/Step (as unsigned)
+ // For negative steps (counting down to zero):
+ // N = Start/-Step
+ // First compute the unsigned distance from zero in the direction of Step.
+ bool CountDown = StepC->getValue()->getValue().isNegative();
+ const SCEV *Distance = CountDown ? Start : getNegativeSCEV(Start);
+
+ // Handle unitary steps, which cannot wraparound.
+ // 1*N = -Start; -1*N = Start (mod 2^BW), so:
+ // N = Distance (as unsigned)
+ if (StepC->getValue()->equalsInt(1) || StepC->getValue()->isAllOnesValue()) {
+ ConstantRange CR = getUnsignedRange(Start);
+ const SCEV *MaxBECount;
+ if (!CountDown && CR.getUnsignedMin().isMinValue())
+ // When counting up, the worst starting value is 1, not 0.
+ MaxBECount = CR.getUnsignedMax().isMinValue()
+ ? getConstant(APInt::getMinValue(CR.getBitWidth()))
+ : getConstant(APInt::getMaxValue(CR.getBitWidth()));
+ else
+ MaxBECount = getConstant(CountDown ? CR.getUnsignedMax()
+ : -CR.getUnsignedMin());
+ return ExitLimit(Distance, MaxBECount);
+ }
+
+ // If the recurrence is known not to wraparound, unsigned divide computes the
+ // back edge count. We know that the value will either become zero (and thus
+ // the loop terminates), that the loop will terminate through some other exit
+ // condition first, or that the loop has undefined behavior. This means
+ // we can't "miss" the exit value, even with nonunit stride.
+ //
+ // FIXME: Prove that loops always exhibits *acceptable* undefined
+ // behavior. Loops must exhibit defined behavior until a wrapped value is
+ // actually used. So the trip count computed by udiv could be smaller than the
+ // number of well-defined iterations.
+ if (AddRec->getNoWrapFlags(SCEV::FlagNW))
+ // FIXME: We really want an "isexact" bit for udiv.
+ return getUDivExpr(Distance, CountDown ? getNegativeSCEV(Step) : Step);
+
+ // Then, try to solve the above equation provided that Start is constant.
+ if (const SCEVConstant *StartC = dyn_cast<SCEVConstant>(Start))
+ return SolveLinEquationWithOverflow(StepC->getValue()->getValue(),
+ -StartC->getValue()->getValue(),
+ *this);
return getCouldNotCompute();
}
/// HowFarToNonZero - Return the number of times a backedge checking the
/// specified value for nonzero will execute. If not computable, return
/// CouldNotCompute
-ScalarEvolution::BackedgeTakenInfo
+ScalarEvolution::ExitLimit
ScalarEvolution::HowFarToNonZero(const SCEV *V, const Loop *L) {
// Loops that look like: while (X == 0) are very strange indeed. We don't
// handle them yet except for the trivial case. This could be expanded in the
@@ -4846,7 +5458,7 @@
// as both operands could be addrecs loop-invariant in each other's loop.
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(RHS)) {
const Loop *L = AR->getLoop();
- if (LHS->isLoopInvariant(L) && LHS->properlyDominates(L->getHeader(), DT)) {
+ if (isLoopInvariant(LHS, L) && properlyDominates(LHS, L->getHeader())) {
std::swap(LHS, RHS);
Pred = ICmpInst::getSwappedPredicate(Pred);
Changed = true;
@@ -5007,12 +5619,12 @@
case ICmpInst::ICMP_SLE:
if (!getSignedRange(RHS).getSignedMax().isMaxSignedValue()) {
RHS = getAddExpr(getConstant(RHS->getType(), 1, true), RHS,
- /*HasNUW=*/false, /*HasNSW=*/true);
+ SCEV::FlagNSW);
Pred = ICmpInst::ICMP_SLT;
Changed = true;
} else if (!getSignedRange(LHS).getSignedMin().isMinSignedValue()) {
LHS = getAddExpr(getConstant(RHS->getType(), (uint64_t)-1, true), LHS,
- /*HasNUW=*/false, /*HasNSW=*/true);
+ SCEV::FlagNSW);
Pred = ICmpInst::ICMP_SLT;
Changed = true;
}
@@ -5020,12 +5632,12 @@
case ICmpInst::ICMP_SGE:
if (!getSignedRange(RHS).getSignedMin().isMinSignedValue()) {
RHS = getAddExpr(getConstant(RHS->getType(), (uint64_t)-1, true), RHS,
- /*HasNUW=*/false, /*HasNSW=*/true);
+ SCEV::FlagNSW);
Pred = ICmpInst::ICMP_SGT;
Changed = true;
} else if (!getSignedRange(LHS).getSignedMax().isMaxSignedValue()) {
LHS = getAddExpr(getConstant(RHS->getType(), 1, true), LHS,
- /*HasNUW=*/false, /*HasNSW=*/true);
+ SCEV::FlagNSW);
Pred = ICmpInst::ICMP_SGT;
Changed = true;
}
@@ -5033,12 +5645,12 @@
case ICmpInst::ICMP_ULE:
if (!getUnsignedRange(RHS).getUnsignedMax().isMaxValue()) {
RHS = getAddExpr(getConstant(RHS->getType(), 1, true), RHS,
- /*HasNUW=*/true, /*HasNSW=*/false);
+ SCEV::FlagNUW);
Pred = ICmpInst::ICMP_ULT;
Changed = true;
} else if (!getUnsignedRange(LHS).getUnsignedMin().isMinValue()) {
LHS = getAddExpr(getConstant(RHS->getType(), (uint64_t)-1, true), LHS,
- /*HasNUW=*/true, /*HasNSW=*/false);
+ SCEV::FlagNUW);
Pred = ICmpInst::ICMP_ULT;
Changed = true;
}
@@ -5046,12 +5658,12 @@
case ICmpInst::ICMP_UGE:
if (!getUnsignedRange(RHS).getUnsignedMin().isMinValue()) {
RHS = getAddExpr(getConstant(RHS->getType(), (uint64_t)-1, true), RHS,
- /*HasNUW=*/true, /*HasNSW=*/false);
+ SCEV::FlagNUW);
Pred = ICmpInst::ICMP_UGT;
Changed = true;
} else if (!getUnsignedRange(LHS).getUnsignedMax().isMaxValue()) {
LHS = getAddExpr(getConstant(RHS->getType(), 1, true), LHS,
- /*HasNUW=*/true, /*HasNSW=*/false);
+ SCEV::FlagNUW);
Pred = ICmpInst::ICMP_UGT;
Changed = true;
}
@@ -5066,13 +5678,13 @@
trivially_true:
// Return 0 == 0.
- LHS = RHS = getConstant(Type::getInt1Ty(getContext()), 0);
+ LHS = RHS = getConstant(ConstantInt::getFalse(getContext()));
Pred = ICmpInst::ICMP_EQ;
return true;
trivially_false:
// Return 0 != 0.
- LHS = RHS = getConstant(Type::getInt1Ty(getContext()), 0);
+ LHS = RHS = getConstant(ConstantInt::getFalse(getContext()));
Pred = ICmpInst::ICMP_NE;
return true;
}
@@ -5432,7 +6044,14 @@
assert(!isKnownNegative(Step) &&
"This code doesn't handle negative strides yet!");
- const Type *Ty = Start->getType();
+ Type *Ty = Start->getType();
+
+ // When Start == End, we have an exact BECount == 0. Short-circuit this case
+ // here because SCEV may not be able to determine that the unsigned division
+ // after rounding is zero.
+ if (Start == End)
+ return getConstant(Ty, 0);
+
const SCEV *NegOne = getConstant(Ty, (uint64_t)-1);
const SCEV *Diff = getMinusSCEV(End, Start);
const SCEV *RoundUp = getAddExpr(Step, NegOne);
@@ -5444,7 +6063,7 @@
if (!NoWrap) {
// Check Add for unsigned overflow.
// TODO: More sophisticated things could be done here.
- const Type *WideTy = IntegerType::get(getContext(),
+ Type *WideTy = IntegerType::get(getContext(),
getTypeSizeInBits(Ty) + 1);
const SCEV *EDiff = getZeroExtendExpr(Diff, WideTy);
const SCEV *ERoundUp = getZeroExtendExpr(RoundUp, WideTy);
@@ -5459,19 +6078,19 @@
/// HowManyLessThans - Return the number of times a backedge containing the
/// specified less-than comparison will execute. If not computable, return
/// CouldNotCompute.
-ScalarEvolution::BackedgeTakenInfo
+ScalarEvolution::ExitLimit
ScalarEvolution::HowManyLessThans(const SCEV *LHS, const SCEV *RHS,
const Loop *L, bool isSigned) {
// Only handle: "ADDREC < LoopInvariant".
- if (!RHS->isLoopInvariant(L)) return getCouldNotCompute();
+ if (!isLoopInvariant(RHS, L)) return getCouldNotCompute();
const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(LHS);
if (!AddRec || AddRec->getLoop() != L)
return getCouldNotCompute();
// Check to see if we have a flag which makes analysis easy.
- bool NoWrap = isSigned ? AddRec->hasNoSignedWrap() :
- AddRec->hasNoUnsignedWrap();
+ bool NoWrap = isSigned ? AddRec->getNoWrapFlags(SCEV::FlagNSW) :
+ AddRec->getNoWrapFlags(SCEV::FlagNUW);
if (AddRec->isAffine()) {
unsigned BitWidth = getTypeSizeInBits(AddRec->getType());
@@ -5555,9 +6174,18 @@
// The maximum backedge count is similar, except using the minimum start
// value and the maximum end value.
- const SCEV *MaxBECount = getBECount(MinStart, MaxEnd, Step, NoWrap);
+ // If we already have an exact constant BECount, use it instead.
+ const SCEV *MaxBECount = isa<SCEVConstant>(BECount) ? BECount
+ : getBECount(MinStart, MaxEnd, Step, NoWrap);
- return BackedgeTakenInfo(BECount, MaxBECount);
+ // If the stride is nonconstant, and NoWrap == true, then
+ // getBECount(MinStart, MaxEnd) may not compute. This would result in an
+ // exact BECount and invalid MaxBECount, which should be avoided to catch
+ // more optimization opportunities.
+ if (isa<SCEVCouldNotCompute>(MaxBECount))
+ MaxBECount = BECount;
+
+ return ExitLimit(BECount, MaxBECount);
}
return getCouldNotCompute();
@@ -5578,7 +6206,8 @@
if (!SC->getValue()->isZero()) {
SmallVector<const SCEV *, 4> Operands(op_begin(), op_end());
Operands[0] = SE.getConstant(SC->getType(), 0);
- const SCEV *Shifted = SE.getAddRecExpr(Operands, getLoop());
+ const SCEV *Shifted = SE.getAddRecExpr(Operands, getLoop(),
+ getNoWrapFlags(FlagNW));
if (const SCEVAddRecExpr *ShiftedAddRec =
dyn_cast<SCEVAddRecExpr>(Shifted))
return ShiftedAddRec->getNumIterationsInRange(
@@ -5639,7 +6268,9 @@
// Range.getUpper() is crossed.
SmallVector<const SCEV *, 4> NewOps(op_begin(), op_end());
NewOps[0] = SE.getNegativeSCEV(SE.getConstant(Range.getUpper()));
- const SCEV *NewAddRec = SE.getAddRecExpr(NewOps, getLoop());
+ const SCEV *NewAddRec = SE.getAddRecExpr(NewOps, getLoop(),
+ // getNoWrapFlags(FlagNW)
+ FlagAnyWrap);
// Next, solve the constructed addrec
std::pair<const SCEV *,const SCEV *> Roots =
@@ -5696,7 +6327,7 @@
assert(SE && "SCEVCallbackVH called with a null ScalarEvolution!");
if (PHINode *PN = dyn_cast<PHINode>(getValPtr()))
SE->ConstantEvolutionLoopExitValue.erase(PN);
- SE->Scalars.erase(getValPtr());
+ SE->ValueExprMap.erase(getValPtr());
// this now dangles!
}
@@ -5722,7 +6353,7 @@
continue;
if (PHINode *PN = dyn_cast<PHINode>(U))
SE->ConstantEvolutionLoopExitValue.erase(PN);
- SE->Scalars.erase(U);
+ SE->ValueExprMap.erase(U);
for (Value::use_iterator UI = U->use_begin(), UE = U->use_end();
UI != UE; ++UI)
Worklist.push_back(*UI);
@@ -5730,7 +6361,7 @@
// Delete the Old value.
if (PHINode *PN = dyn_cast<PHINode>(Old))
SE->ConstantEvolutionLoopExitValue.erase(PN);
- SE->Scalars.erase(Old);
+ SE->ValueExprMap.erase(Old);
// this now dangles!
}
@@ -5743,6 +6374,7 @@
ScalarEvolution::ScalarEvolution()
: FunctionPass(ID), FirstUnknown(0) {
+ initializeScalarEvolutionPass(*PassRegistry::getPassRegistry());
}
bool ScalarEvolution::runOnFunction(Function &F) {
@@ -5760,10 +6392,23 @@
U->~SCEVUnknown();
FirstUnknown = 0;
- Scalars.clear();
+ ValueExprMap.clear();
+
+ // Free any extra memory created for ExitNotTakenInfo in the unlikely event
+ // that a loop had multiple computable exits.
+ for (DenseMap<const Loop*, BackedgeTakenInfo>::iterator I =
+ BackedgeTakenCounts.begin(), E = BackedgeTakenCounts.end();
+ I != E; ++I) {
+ I->second.clear();
+ }
+
BackedgeTakenCounts.clear();
ConstantEvolutionLoopExitValue.clear();
ValuesAtScopes.clear();
+ LoopDispositions.clear();
+ BlockDispositions.clear();
+ UnsignedRanges.clear();
+ SignedRanges.clear();
UniqueSCEVs.clear();
SCEVAllocator.Reset();
}
@@ -5843,7 +6488,7 @@
if (L) {
OS << "\t\t" "Exits: ";
const SCEV *ExitValue = SE.getSCEVAtScope(SV, L->getParentLoop());
- if (!ExitValue->isLoopInvariant(L)) {
+ if (!SE.isLoopInvariant(ExitValue, L)) {
OS << "<<Unknown>>";
} else {
OS << *ExitValue;
@@ -5860,3 +6505,240 @@
PrintLoopInfo(OS, &SE, *I);
}
+ScalarEvolution::LoopDisposition
+ScalarEvolution::getLoopDisposition(const SCEV *S, const Loop *L) {
+ std::map<const Loop *, LoopDisposition> &Values = LoopDispositions[S];
+ std::pair<std::map<const Loop *, LoopDisposition>::iterator, bool> Pair =
+ Values.insert(std::make_pair(L, LoopVariant));
+ if (!Pair.second)
+ return Pair.first->second;
+
+ LoopDisposition D = computeLoopDisposition(S, L);
+ return LoopDispositions[S][L] = D;
+}
+
+ScalarEvolution::LoopDisposition
+ScalarEvolution::computeLoopDisposition(const SCEV *S, const Loop *L) {
+ switch (S->getSCEVType()) {
+ case scConstant:
+ return LoopInvariant;
+ case scTruncate:
+ case scZeroExtend:
+ case scSignExtend:
+ return getLoopDisposition(cast<SCEVCastExpr>(S)->getOperand(), L);
+ case scAddRecExpr: {
+ const SCEVAddRecExpr *AR = cast<SCEVAddRecExpr>(S);
+
+ // If L is the addrec's loop, it's computable.
+ if (AR->getLoop() == L)
+ return LoopComputable;
+
+ // Add recurrences are never invariant in the function-body (null loop).
+ if (!L)
+ return LoopVariant;
+
+ // This recurrence is variant w.r.t. L if L contains AR's loop.
+ if (L->contains(AR->getLoop()))
+ return LoopVariant;
+
+ // This recurrence is invariant w.r.t. L if AR's loop contains L.
+ if (AR->getLoop()->contains(L))
+ return LoopInvariant;
+
+ // This recurrence is variant w.r.t. L if any of its operands
+ // are variant.
+ for (SCEVAddRecExpr::op_iterator I = AR->op_begin(), E = AR->op_end();
+ I != E; ++I)
+ if (!isLoopInvariant(*I, L))
+ return LoopVariant;
+
+ // Otherwise it's loop-invariant.
+ return LoopInvariant;
+ }
+ case scAddExpr:
+ case scMulExpr:
+ case scUMaxExpr:
+ case scSMaxExpr: {
+ const SCEVNAryExpr *NAry = cast<SCEVNAryExpr>(S);
+ bool HasVarying = false;
+ for (SCEVNAryExpr::op_iterator I = NAry->op_begin(), E = NAry->op_end();
+ I != E; ++I) {
+ LoopDisposition D = getLoopDisposition(*I, L);
+ if (D == LoopVariant)
+ return LoopVariant;
+ if (D == LoopComputable)
+ HasVarying = true;
+ }
+ return HasVarying ? LoopComputable : LoopInvariant;
+ }
+ case scUDivExpr: {
+ const SCEVUDivExpr *UDiv = cast<SCEVUDivExpr>(S);
+ LoopDisposition LD = getLoopDisposition(UDiv->getLHS(), L);
+ if (LD == LoopVariant)
+ return LoopVariant;
+ LoopDisposition RD = getLoopDisposition(UDiv->getRHS(), L);
+ if (RD == LoopVariant)
+ return LoopVariant;
+ return (LD == LoopInvariant && RD == LoopInvariant) ?
+ LoopInvariant : LoopComputable;
+ }
+ case scUnknown:
+ // All non-instruction values are loop invariant. All instructions are loop
+ // invariant if they are not contained in the specified loop.
+ // Instructions are never considered invariant in the function body
+ // (null loop) because they are defined within the "loop".
+ if (Instruction *I = dyn_cast<Instruction>(cast<SCEVUnknown>(S)->getValue()))
+ return (L && !L->contains(I)) ? LoopInvariant : LoopVariant;
+ return LoopInvariant;
+ case scCouldNotCompute:
+ llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
+ return LoopVariant;
+ default: break;
+ }
+ llvm_unreachable("Unknown SCEV kind!");
+ return LoopVariant;
+}
+
+bool ScalarEvolution::isLoopInvariant(const SCEV *S, const Loop *L) {
+ return getLoopDisposition(S, L) == LoopInvariant;
+}
+
+bool ScalarEvolution::hasComputableLoopEvolution(const SCEV *S, const Loop *L) {
+ return getLoopDisposition(S, L) == LoopComputable;
+}
+
+ScalarEvolution::BlockDisposition
+ScalarEvolution::getBlockDisposition(const SCEV *S, const BasicBlock *BB) {
+ std::map<const BasicBlock *, BlockDisposition> &Values = BlockDispositions[S];
+ std::pair<std::map<const BasicBlock *, BlockDisposition>::iterator, bool>
+ Pair = Values.insert(std::make_pair(BB, DoesNotDominateBlock));
+ if (!Pair.second)
+ return Pair.first->second;
+
+ BlockDisposition D = computeBlockDisposition(S, BB);
+ return BlockDispositions[S][BB] = D;
+}
+
+ScalarEvolution::BlockDisposition
+ScalarEvolution::computeBlockDisposition(const SCEV *S, const BasicBlock *BB) {
+ switch (S->getSCEVType()) {
+ case scConstant:
+ return ProperlyDominatesBlock;
+ case scTruncate:
+ case scZeroExtend:
+ case scSignExtend:
+ return getBlockDisposition(cast<SCEVCastExpr>(S)->getOperand(), BB);
+ case scAddRecExpr: {
+ // This uses a "dominates" query instead of "properly dominates" query
+ // to test for proper dominance too, because the instruction which
+ // produces the addrec's value is a PHI, and a PHI effectively properly
+ // dominates its entire containing block.
+ const SCEVAddRecExpr *AR = cast<SCEVAddRecExpr>(S);
+ if (!DT->dominates(AR->getLoop()->getHeader(), BB))
+ return DoesNotDominateBlock;
+ }
+ // FALL THROUGH into SCEVNAryExpr handling.
+ case scAddExpr:
+ case scMulExpr:
+ case scUMaxExpr:
+ case scSMaxExpr: {
+ const SCEVNAryExpr *NAry = cast<SCEVNAryExpr>(S);
+ bool Proper = true;
+ for (SCEVNAryExpr::op_iterator I = NAry->op_begin(), E = NAry->op_end();
+ I != E; ++I) {
+ BlockDisposition D = getBlockDisposition(*I, BB);
+ if (D == DoesNotDominateBlock)
+ return DoesNotDominateBlock;
+ if (D == DominatesBlock)
+ Proper = false;
+ }
+ return Proper ? ProperlyDominatesBlock : DominatesBlock;
+ }
+ case scUDivExpr: {
+ const SCEVUDivExpr *UDiv = cast<SCEVUDivExpr>(S);
+ const SCEV *LHS = UDiv->getLHS(), *RHS = UDiv->getRHS();
+ BlockDisposition LD = getBlockDisposition(LHS, BB);
+ if (LD == DoesNotDominateBlock)
+ return DoesNotDominateBlock;
+ BlockDisposition RD = getBlockDisposition(RHS, BB);
+ if (RD == DoesNotDominateBlock)
+ return DoesNotDominateBlock;
+ return (LD == ProperlyDominatesBlock && RD == ProperlyDominatesBlock) ?
+ ProperlyDominatesBlock : DominatesBlock;
+ }
+ case scUnknown:
+ if (Instruction *I =
+ dyn_cast<Instruction>(cast<SCEVUnknown>(S)->getValue())) {
+ if (I->getParent() == BB)
+ return DominatesBlock;
+ if (DT->properlyDominates(I->getParent(), BB))
+ return ProperlyDominatesBlock;
+ return DoesNotDominateBlock;
+ }
+ return ProperlyDominatesBlock;
+ case scCouldNotCompute:
+ llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
+ return DoesNotDominateBlock;
+ default: break;
+ }
+ llvm_unreachable("Unknown SCEV kind!");
+ return DoesNotDominateBlock;
+}
+
+bool ScalarEvolution::dominates(const SCEV *S, const BasicBlock *BB) {
+ return getBlockDisposition(S, BB) >= DominatesBlock;
+}
+
+bool ScalarEvolution::properlyDominates(const SCEV *S, const BasicBlock *BB) {
+ return getBlockDisposition(S, BB) == ProperlyDominatesBlock;
+}
+
+bool ScalarEvolution::hasOperand(const SCEV *S, const SCEV *Op) const {
+ switch (S->getSCEVType()) {
+ case scConstant:
+ return false;
+ case scTruncate:
+ case scZeroExtend:
+ case scSignExtend: {
+ const SCEVCastExpr *Cast = cast<SCEVCastExpr>(S);
+ const SCEV *CastOp = Cast->getOperand();
+ return Op == CastOp || hasOperand(CastOp, Op);
+ }
+ case scAddRecExpr:
+ case scAddExpr:
+ case scMulExpr:
+ case scUMaxExpr:
+ case scSMaxExpr: {
+ const SCEVNAryExpr *NAry = cast<SCEVNAryExpr>(S);
+ for (SCEVNAryExpr::op_iterator I = NAry->op_begin(), E = NAry->op_end();
+ I != E; ++I) {
+ const SCEV *NAryOp = *I;
+ if (NAryOp == Op || hasOperand(NAryOp, Op))
+ return true;
+ }
+ return false;
+ }
+ case scUDivExpr: {
+ const SCEVUDivExpr *UDiv = cast<SCEVUDivExpr>(S);
+ const SCEV *LHS = UDiv->getLHS(), *RHS = UDiv->getRHS();
+ return LHS == Op || hasOperand(LHS, Op) ||
+ RHS == Op || hasOperand(RHS, Op);
+ }
+ case scUnknown:
+ return false;
+ case scCouldNotCompute:
+ llvm_unreachable("Attempt to use a SCEVCouldNotCompute object!");
+ return false;
+ default: break;
+ }
+ llvm_unreachable("Unknown SCEV kind!");
+ return false;
+}
+
+void ScalarEvolution::forgetMemoizedResults(const SCEV *S) {
+ ValuesAtScopes.erase(S);
+ LoopDispositions.erase(S);
+ BlockDispositions.erase(S);
+ UnsignedRanges.erase(S);
+ SignedRanges.erase(S);
+}
diff --git a/src/LLVM/lib/Analysis/ScalarEvolutionAliasAnalysis.cpp b/src/LLVM/lib/Analysis/ScalarEvolutionAliasAnalysis.cpp
new file mode 100644
index 0000000..e9edb3e
--- /dev/null
+++ b/src/LLVM/lib/Analysis/ScalarEvolutionAliasAnalysis.cpp
@@ -0,0 +1,173 @@
+//===- ScalarEvolutionAliasAnalysis.cpp - SCEV-based Alias Analysis -------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the ScalarEvolutionAliasAnalysis pass, which implements a
+// simple alias analysis implemented in terms of ScalarEvolution queries.
+//
+// This differs from traditional loop dependence analysis in that it tests
+// for dependencies within a single iteration of a loop, rather than
+// dependencies between different iterations.
+//
+// ScalarEvolution has a more complete understanding of pointer arithmetic
+// than BasicAliasAnalysis' collection of ad-hoc analyses.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/ScalarEvolutionExpressions.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Pass.h"
+using namespace llvm;
+
+namespace {
+ /// ScalarEvolutionAliasAnalysis - This is a simple alias analysis
+ /// implementation that uses ScalarEvolution to answer queries.
+ class ScalarEvolutionAliasAnalysis : public FunctionPass,
+ public AliasAnalysis {
+ ScalarEvolution *SE;
+
+ public:
+ static char ID; // Class identification, replacement for typeinfo
+ ScalarEvolutionAliasAnalysis() : FunctionPass(ID), SE(0) {
+ initializeScalarEvolutionAliasAnalysisPass(
+ *PassRegistry::getPassRegistry());
+ }
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ virtual void *getAdjustedAnalysisPointer(AnalysisID PI) {
+ if (PI == &AliasAnalysis::ID)
+ return (AliasAnalysis*)this;
+ return this;
+ }
+
+ private:
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+ virtual bool runOnFunction(Function &F);
+ virtual AliasResult alias(const Location &LocA, const Location &LocB);
+
+ Value *GetBaseValue(const SCEV *S);
+ };
+} // End of anonymous namespace
+
+// Register this pass...
+char ScalarEvolutionAliasAnalysis::ID = 0;
+INITIALIZE_AG_PASS_BEGIN(ScalarEvolutionAliasAnalysis, AliasAnalysis, "scev-aa",
+ "ScalarEvolution-based Alias Analysis", false, true, false)
+INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
+INITIALIZE_AG_PASS_END(ScalarEvolutionAliasAnalysis, AliasAnalysis, "scev-aa",
+ "ScalarEvolution-based Alias Analysis", false, true, false)
+
+FunctionPass *llvm::createScalarEvolutionAliasAnalysisPass() {
+ return new ScalarEvolutionAliasAnalysis();
+}
+
+void
+ScalarEvolutionAliasAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.addRequiredTransitive<ScalarEvolution>();
+ AU.setPreservesAll();
+ AliasAnalysis::getAnalysisUsage(AU);
+}
+
+bool
+ScalarEvolutionAliasAnalysis::runOnFunction(Function &F) {
+ InitializeAliasAnalysis(this);
+ SE = &getAnalysis<ScalarEvolution>();
+ return false;
+}
+
+/// GetBaseValue - Given an expression, try to find a
+/// base value. Return null is none was found.
+Value *
+ScalarEvolutionAliasAnalysis::GetBaseValue(const SCEV *S) {
+ if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S)) {
+ // In an addrec, assume that the base will be in the start, rather
+ // than the step.
+ return GetBaseValue(AR->getStart());
+ } else if (const SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(S)) {
+ // If there's a pointer operand, it'll be sorted at the end of the list.
+ const SCEV *Last = A->getOperand(A->getNumOperands()-1);
+ if (Last->getType()->isPointerTy())
+ return GetBaseValue(Last);
+ } else if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
+ // This is a leaf node.
+ return U->getValue();
+ }
+ // No Identified object found.
+ return 0;
+}
+
+AliasAnalysis::AliasResult
+ScalarEvolutionAliasAnalysis::alias(const Location &LocA,
+ const Location &LocB) {
+ // If either of the memory references is empty, it doesn't matter what the
+ // pointer values are. This allows the code below to ignore this special
+ // case.
+ if (LocA.Size == 0 || LocB.Size == 0)
+ return NoAlias;
+
+ // This is ScalarEvolutionAliasAnalysis. Get the SCEVs!
+ const SCEV *AS = SE->getSCEV(const_cast<Value *>(LocA.Ptr));
+ const SCEV *BS = SE->getSCEV(const_cast<Value *>(LocB.Ptr));
+
+ // If they evaluate to the same expression, it's a MustAlias.
+ if (AS == BS) return MustAlias;
+
+ // If something is known about the difference between the two addresses,
+ // see if it's enough to prove a NoAlias.
+ if (SE->getEffectiveSCEVType(AS->getType()) ==
+ SE->getEffectiveSCEVType(BS->getType())) {
+ unsigned BitWidth = SE->getTypeSizeInBits(AS->getType());
+ APInt ASizeInt(BitWidth, LocA.Size);
+ APInt BSizeInt(BitWidth, LocB.Size);
+
+ // Compute the difference between the two pointers.
+ const SCEV *BA = SE->getMinusSCEV(BS, AS);
+
+ // Test whether the difference is known to be great enough that memory of
+ // the given sizes don't overlap. This assumes that ASizeInt and BSizeInt
+ // are non-zero, which is special-cased above.
+ if (ASizeInt.ule(SE->getUnsignedRange(BA).getUnsignedMin()) &&
+ (-BSizeInt).uge(SE->getUnsignedRange(BA).getUnsignedMax()))
+ return NoAlias;
+
+ // Folding the subtraction while preserving range information can be tricky
+ // (because of INT_MIN, etc.); if the prior test failed, swap AS and BS
+ // and try again to see if things fold better that way.
+
+ // Compute the difference between the two pointers.
+ const SCEV *AB = SE->getMinusSCEV(AS, BS);
+
+ // Test whether the difference is known to be great enough that memory of
+ // the given sizes don't overlap. This assumes that ASizeInt and BSizeInt
+ // are non-zero, which is special-cased above.
+ if (BSizeInt.ule(SE->getUnsignedRange(AB).getUnsignedMin()) &&
+ (-ASizeInt).uge(SE->getUnsignedRange(AB).getUnsignedMax()))
+ return NoAlias;
+ }
+
+ // If ScalarEvolution can find an underlying object, form a new query.
+ // The correctness of this depends on ScalarEvolution not recognizing
+ // inttoptr and ptrtoint operators.
+ Value *AO = GetBaseValue(AS);
+ Value *BO = GetBaseValue(BS);
+ if ((AO && AO != LocA.Ptr) || (BO && BO != LocB.Ptr))
+ if (alias(Location(AO ? AO : LocA.Ptr,
+ AO ? +UnknownSize : LocA.Size,
+ AO ? 0 : LocA.TBAATag),
+ Location(BO ? BO : LocB.Ptr,
+ BO ? +UnknownSize : LocB.Size,
+ BO ? 0 : LocB.TBAATag)) == NoAlias)
+ return NoAlias;
+
+ // Forward the query to the next analysis.
+ return AliasAnalysis::alias(LocA, LocB);
+}
diff --git a/src/LLVM/lib/Analysis/ScalarEvolutionExpander.cpp b/src/LLVM/lib/Analysis/ScalarEvolutionExpander.cpp
index 8f31297..47f0f32 100644
--- a/src/LLVM/lib/Analysis/ScalarEvolutionExpander.cpp
+++ b/src/LLVM/lib/Analysis/ScalarEvolutionExpander.cpp
@@ -17,15 +17,17 @@
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/LLVMContext.h"
+#include "llvm/Support/Debug.h"
#include "llvm/Target/TargetData.h"
#include "llvm/ADT/STLExtras.h"
+
using namespace llvm;
/// ReuseOrCreateCast - Arrange for there to be a cast of V to Ty at IP,
/// reusing an existing cast if a suitable one exists, moving an existing
/// cast if a suitable one exists but isn't in the right place, or
/// creating a new one.
-Value *SCEVExpander::ReuseOrCreateCast(Value *V, const Type *Ty,
+Value *SCEVExpander::ReuseOrCreateCast(Value *V, Type *Ty,
Instruction::CastOps Op,
BasicBlock::iterator IP) {
// Check to see if there is already a cast!
@@ -40,7 +42,7 @@
// Create a new cast, and leave the old cast in place in case
// it is being used as an insert point. Clear its operand
// so that it doesn't hold anything live.
- Instruction *NewCI = CastInst::Create(Op, V, Ty, IP);
+ Instruction *NewCI = CastInst::Create(Op, V, Ty, "", IP);
NewCI->takeName(CI);
CI->replaceAllUsesWith(NewCI);
CI->setOperand(0, UndefValue::get(V->getType()));
@@ -53,7 +55,7 @@
}
// Create a new cast.
- Instruction *I = CastInst::Create(Op, V, Ty, IP);
+ Instruction *I = CastInst::Create(Op, V, Ty, V->getName(), IP);
rememberInstruction(I);
return I;
}
@@ -61,7 +63,7 @@
/// InsertNoopCastOfTo - Insert a cast of V to the specified type,
/// which must be possible with a noop cast, doing what we can to share
/// the casts.
-Value *SCEVExpander::InsertNoopCastOfTo(Value *V, const Type *Ty) {
+Value *SCEVExpander::InsertNoopCastOfTo(Value *V, Type *Ty) {
Instruction::CastOps Op = CastInst::getCastOpcode(V, false, Ty, false);
assert((Op == Instruction::BitCast ||
Op == Instruction::PtrToInt ||
@@ -101,7 +103,9 @@
BasicBlock::iterator IP = A->getParent()->getEntryBlock().begin();
while ((isa<BitCastInst>(IP) &&
isa<Argument>(cast<BitCastInst>(IP)->getOperand(0)) &&
- cast<BitCastInst>(IP)->getOperand(0) != A))
+ cast<BitCastInst>(IP)->getOperand(0) != A) ||
+ isa<DbgInfoIntrinsic>(IP) ||
+ isa<LandingPadInst>(IP))
++IP;
return ReuseOrCreateCast(A, Ty, Op, IP);
}
@@ -109,7 +113,11 @@
// Cast the instruction immediately after the instruction.
Instruction *I = cast<Instruction>(V);
BasicBlock::iterator IP = I; ++IP;
- while (isa<PHINode>(IP)) ++IP;
+ if (InvokeInst *II = dyn_cast<InvokeInst>(I))
+ IP = II->getNormalDest()->begin();
+ while (isa<PHINode>(IP) || isa<DbgInfoIntrinsic>(IP) ||
+ isa<LandingPadInst>(IP))
+ ++IP;
return ReuseOrCreateCast(I, Ty, Op, IP);
}
@@ -132,7 +140,7 @@
for (; ScanLimit; --IP, --ScanLimit) {
// Don't count dbg.value against the ScanLimit, to avoid perturbing the
// generated code.
- if (ISA_DEBUG_INFO_INTRINSIC(IP))
+ if (isa<DbgInfoIntrinsic>(IP))
ScanLimit++;
if (IP->getOpcode() == (unsigned)Opcode && IP->getOperand(0) == LHS &&
IP->getOperand(1) == RHS)
@@ -156,7 +164,8 @@
}
// If we haven't found this binop, insert it.
- Value *BO = Builder.CreateBinOp(Opcode, LHS, RHS);
+ Instruction *BO = cast<Instruction>(Builder.CreateBinOp(Opcode, LHS, RHS));
+ BO->setDebugLoc(SaveInsertPt->getDebugLoc());
rememberInstruction(BO);
// Restore the original insert point.
@@ -259,7 +268,8 @@
const SCEV *Start = A->getStart();
if (!FactorOutConstant(Start, Remainder, Factor, SE, TD))
return false;
- S = SE.getAddRecExpr(Start, Step, A->getLoop());
+ // FIXME: can use A->getNoWrapFlags(FlagNW)
+ S = SE.getAddRecExpr(Start, Step, A->getLoop(), SCEV::FlagAnyWrap);
return true;
}
@@ -271,7 +281,7 @@
/// the list.
///
static void SimplifyAddOperands(SmallVectorImpl<const SCEV *> &Ops,
- const Type *Ty,
+ Type *Ty,
ScalarEvolution &SE) {
unsigned NumAddRecs = 0;
for (unsigned i = Ops.size(); i > 0 && isa<SCEVAddRecExpr>(Ops[i-1]); --i)
@@ -300,7 +310,7 @@
/// into GEP indices.
///
static void SplitAddRecs(SmallVectorImpl<const SCEV *> &Ops,
- const Type *Ty,
+ Type *Ty,
ScalarEvolution &SE) {
// Find the addrecs.
SmallVector<const SCEV *, 8> AddRecs;
@@ -311,7 +321,9 @@
const SCEV *Zero = SE.getConstant(Ty, 0);
AddRecs.push_back(SE.getAddRecExpr(Zero,
A->getStepRecurrence(SE),
- A->getLoop()));
+ A->getLoop(),
+ // FIXME: A->getNoWrapFlags(FlagNW)
+ SCEV::FlagAnyWrap));
if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Start)) {
Ops[i] = Zero;
Ops.append(Add->op_begin(), Add->op_end());
@@ -357,10 +369,10 @@
///
Value *SCEVExpander::expandAddToGEP(const SCEV *const *op_begin,
const SCEV *const *op_end,
- const PointerType *PTy,
- const Type *Ty,
+ PointerType *PTy,
+ Type *Ty,
Value *V) {
- const Type *ElTy = PTy->getElementType();
+ Type *ElTy = PTy->getElementType();
SmallVector<Value *, 4> GepIndices;
SmallVector<const SCEV *, 8> Ops(op_begin, op_end);
bool AnyNonZeroIndices = false;
@@ -415,7 +427,7 @@
GepIndices.push_back(Scaled);
// Collect struct field index operands.
- while (const StructType *STy = dyn_cast<StructType>(ElTy)) {
+ while (StructType *STy = dyn_cast<StructType>(ElTy)) {
bool FoundFieldNo = false;
// An empty struct has no fields.
if (STy->getNumElements() == 0) break;
@@ -443,7 +455,7 @@
// appropriate struct type.
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(Ops[i])) {
- const Type *CTy;
+ Type *CTy;
Constant *FieldNo;
if (U->isOffsetOf(CTy, FieldNo) && CTy == STy) {
GepIndices.push_back(FieldNo);
@@ -466,7 +478,7 @@
}
}
- if (const ArrayType *ATy = dyn_cast<ArrayType>(ElTy))
+ if (ArrayType *ATy = dyn_cast<ArrayType>(ElTy))
ElTy = ATy->getElementType();
else
break;
@@ -486,7 +498,7 @@
// Fold a GEP with constant operands.
if (Constant *CLHS = dyn_cast<Constant>(V))
if (Constant *CRHS = dyn_cast<Constant>(Idx))
- return ConstantExpr::getGetElementPtr(CLHS, &CRHS, 1);
+ return ConstantExpr::getGetElementPtr(CLHS, CRHS);
// Do a quick scan to see if we have this GEP nearby. If so, reuse it.
unsigned ScanLimit = 6;
@@ -498,7 +510,7 @@
for (; ScanLimit; --IP, --ScanLimit) {
// Don't count dbg.value against the ScanLimit, to avoid perturbing the
// generated code.
- if (ISA_DEBUG_INFO_INTRINSIC(IP))
+ if (isa<DbgInfoIntrinsic>(IP))
ScanLimit++;
if (IP->getOpcode() == Instruction::GetElementPtr &&
IP->getOperand(0) == V && IP->getOperand(1) == Idx)
@@ -522,7 +534,7 @@
}
// Emit a GEP.
- Value *GEP = Builder.CreateGEP(V, Idx);
+ Value *GEP = Builder.CreateGEP(V, Idx, "uglygep");
rememberInstruction(GEP);
// Restore the original insert point.
@@ -564,8 +576,8 @@
if (V->getType() != PTy)
Casted = InsertNoopCastOfTo(Casted, PTy);
Value *GEP = Builder.CreateGEP(Casted,
- GepIndices.begin(),
- GepIndices.end());
+ GepIndices,
+ "scevgep");
Ops.push_back(SE.getUnknown(GEP));
rememberInstruction(GEP);
@@ -604,15 +616,22 @@
return A; // Arbitrarily break the tie.
}
-/// GetRelevantLoop - Get the most relevant loop associated with the given
+/// getRelevantLoop - Get the most relevant loop associated with the given
/// expression, according to PickMostRelevantLoop.
-static const Loop *GetRelevantLoop(const SCEV *S, LoopInfo &LI,
- DominatorTree &DT) {
+const Loop *SCEVExpander::getRelevantLoop(const SCEV *S) {
+ // Test whether we've already computed the most relevant loop for this SCEV.
+ std::pair<DenseMap<const SCEV *, const Loop *>::iterator, bool> Pair =
+ RelevantLoops.insert(std::make_pair(S, static_cast<const Loop *>(0)));
+ if (!Pair.second)
+ return Pair.first->second;
+
if (isa<SCEVConstant>(S))
+ // A constant has no relevant loops.
return 0;
if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
if (const Instruction *I = dyn_cast<Instruction>(U->getValue()))
- return LI.getLoopFor(I->getParent());
+ return Pair.first->second = SE.LI->getLoopFor(I->getParent());
+ // A non-instruction has no relevant loops.
return 0;
}
if (const SCEVNAryExpr *N = dyn_cast<SCEVNAryExpr>(S)) {
@@ -621,16 +640,22 @@
L = AR->getLoop();
for (SCEVNAryExpr::op_iterator I = N->op_begin(), E = N->op_end();
I != E; ++I)
- L = PickMostRelevantLoop(L, GetRelevantLoop(*I, LI, DT), DT);
- return L;
+ L = PickMostRelevantLoop(L, getRelevantLoop(*I), *SE.DT);
+ return RelevantLoops[N] = L;
}
- if (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(S))
- return GetRelevantLoop(C->getOperand(), LI, DT);
- if (const SCEVUDivExpr *D = dyn_cast<SCEVUDivExpr>(S))
- return PickMostRelevantLoop(GetRelevantLoop(D->getLHS(), LI, DT),
- GetRelevantLoop(D->getRHS(), LI, DT),
- DT);
+ if (const SCEVCastExpr *C = dyn_cast<SCEVCastExpr>(S)) {
+ const Loop *Result = getRelevantLoop(C->getOperand());
+ return RelevantLoops[C] = Result;
+ }
+ if (const SCEVUDivExpr *D = dyn_cast<SCEVUDivExpr>(S)) {
+ const Loop *Result =
+ PickMostRelevantLoop(getRelevantLoop(D->getLHS()),
+ getRelevantLoop(D->getRHS()),
+ *SE.DT);
+ return RelevantLoops[D] = Result;
+ }
llvm_unreachable("Unexpected SCEV type!");
+ return 0;
}
namespace {
@@ -669,7 +694,7 @@
}
Value *SCEVExpander::visitAddExpr(const SCEVAddExpr *S) {
- const Type *Ty = SE.getEffectiveSCEVType(S->getType());
+ Type *Ty = SE.getEffectiveSCEVType(S->getType());
// Collect all the add operands in a loop, along with their associated loops.
// Iterate in reverse so that constants are emitted last, all else equal, and
@@ -678,8 +703,7 @@
SmallVector<std::pair<const Loop *, const SCEV *>, 8> OpsAndLoops;
for (std::reverse_iterator<SCEVAddExpr::op_iterator> I(S->op_end()),
E(S->op_begin()); I != E; ++I)
- OpsAndLoops.push_back(std::make_pair(GetRelevantLoop(*I, *SE.LI, *SE.DT),
- *I));
+ OpsAndLoops.push_back(std::make_pair(getRelevantLoop(*I), *I));
// Sort by loop. Use a stable sort so that constants follow non-constants and
// pointer operands precede non-pointer operands.
@@ -696,7 +720,7 @@
// This is the first operand. Just expand it.
Sum = expand(Op);
++I;
- } else if (const PointerType *PTy = dyn_cast<PointerType>(Sum->getType())) {
+ } else if (PointerType *PTy = dyn_cast<PointerType>(Sum->getType())) {
// The running sum expression is a pointer. Try to form a getelementptr
// at this level with that as the base.
SmallVector<const SCEV *, 4> NewOps;
@@ -710,7 +734,7 @@
NewOps.push_back(X);
}
Sum = expandAddToGEP(NewOps.begin(), NewOps.end(), PTy, Ty, Sum);
- } else if (const PointerType *PTy = dyn_cast<PointerType>(Op->getType())) {
+ } else if (PointerType *PTy = dyn_cast<PointerType>(Op->getType())) {
// The running sum is an integer, and there's a pointer at this level.
// Try to form a getelementptr. If the running sum is instructions,
// use a SCEVUnknown to avoid re-analyzing them.
@@ -741,15 +765,14 @@
}
Value *SCEVExpander::visitMulExpr(const SCEVMulExpr *S) {
- const Type *Ty = SE.getEffectiveSCEVType(S->getType());
+ Type *Ty = SE.getEffectiveSCEVType(S->getType());
// Collect all the mul operands in a loop, along with their associated loops.
// Iterate in reverse so that constants are emitted last, all else equal.
SmallVector<std::pair<const Loop *, const SCEV *>, 8> OpsAndLoops;
for (std::reverse_iterator<SCEVMulExpr::op_iterator> I(S->op_end()),
E(S->op_begin()); I != E; ++I)
- OpsAndLoops.push_back(std::make_pair(GetRelevantLoop(*I, *SE.LI, *SE.DT),
- *I));
+ OpsAndLoops.push_back(std::make_pair(getRelevantLoop(*I), *I));
// Sort by loop. Use a stable sort so that constants follow non-constants.
std::stable_sort(OpsAndLoops.begin(), OpsAndLoops.end(), LoopCompare(*SE.DT));
@@ -784,7 +807,7 @@
}
Value *SCEVExpander::visitUDivExpr(const SCEVUDivExpr *S) {
- const Type *Ty = SE.getEffectiveSCEVType(S->getType());
+ Type *Ty = SE.getEffectiveSCEVType(S->getType());
Value *LHS = expandCodeFor(S->getLHS(), Ty);
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(S->getRHS())) {
@@ -808,7 +831,9 @@
Rest = SE.getAddExpr(Rest,
SE.getAddRecExpr(SE.getConstant(A->getType(), 0),
A->getStepRecurrence(SE),
- A->getLoop()));
+ A->getLoop(),
+ // FIXME: A->getNoWrapFlags(FlagNW)
+ SCEV::FlagAnyWrap));
}
if (const SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(Base)) {
Base = A->getOperand(A->getNumOperands()-1);
@@ -819,76 +844,141 @@
}
}
+/// Determine if this is a well-behaved chain of instructions leading back to
+/// the PHI. If so, it may be reused by expanded expressions.
+bool SCEVExpander::isNormalAddRecExprPHI(PHINode *PN, Instruction *IncV,
+ const Loop *L) {
+ if (IncV->getNumOperands() == 0 || isa<PHINode>(IncV) ||
+ (isa<CastInst>(IncV) && !isa<BitCastInst>(IncV)))
+ return false;
+ // If any of the operands don't dominate the insert position, bail.
+ // Addrec operands are always loop-invariant, so this can only happen
+ // if there are instructions which haven't been hoisted.
+ if (L == IVIncInsertLoop) {
+ for (User::op_iterator OI = IncV->op_begin()+1,
+ OE = IncV->op_end(); OI != OE; ++OI)
+ if (Instruction *OInst = dyn_cast<Instruction>(OI))
+ if (!SE.DT->dominates(OInst, IVIncInsertPos))
+ return false;
+ }
+ // Advance to the next instruction.
+ IncV = dyn_cast<Instruction>(IncV->getOperand(0));
+ if (!IncV)
+ return false;
+
+ if (IncV->mayHaveSideEffects())
+ return false;
+
+ if (IncV != PN)
+ return true;
+
+ return isNormalAddRecExprPHI(PN, IncV, L);
+}
+
+/// Determine if this cyclic phi is in a form that would have been generated by
+/// LSR. We don't care if the phi was actually expanded in this pass, as long
+/// as it is in a low-cost form, for example, no implied multiplication. This
+/// should match any patterns generated by getAddRecExprPHILiterally and
+/// expandAddtoGEP.
+bool SCEVExpander::isExpandedAddRecExprPHI(PHINode *PN, Instruction *IncV,
+ const Loop *L) {
+ switch (IncV->getOpcode()) {
+ // Check for a simple Add/Sub or GEP of a loop invariant step.
+ case Instruction::Add:
+ case Instruction::Sub:
+ return IncV->getOperand(0) == PN
+ && L->isLoopInvariant(IncV->getOperand(1));
+ case Instruction::BitCast:
+ IncV = dyn_cast<GetElementPtrInst>(IncV->getOperand(0));
+ if (!IncV)
+ return false;
+ // fall-thru to GEP handling
+ case Instruction::GetElementPtr: {
+ // This must be a pointer addition of constants (pretty) or some number of
+ // address-size elements (ugly).
+ for (Instruction::op_iterator I = IncV->op_begin()+1, E = IncV->op_end();
+ I != E; ++I) {
+ if (isa<Constant>(*I))
+ continue;
+ // ugly geps have 2 operands.
+ // i1* is used by the expander to represent an address-size element.
+ if (IncV->getNumOperands() != 2)
+ return false;
+ unsigned AS = cast<PointerType>(IncV->getType())->getAddressSpace();
+ if (IncV->getType() != Type::getInt1PtrTy(SE.getContext(), AS)
+ && IncV->getType() != Type::getInt8PtrTy(SE.getContext(), AS))
+ return false;
+ // Ensure the operands dominate the insertion point. I don't know of a
+ // case when this would not be true, so this is somewhat untested.
+ if (L == IVIncInsertLoop) {
+ for (User::op_iterator OI = IncV->op_begin()+1,
+ OE = IncV->op_end(); OI != OE; ++OI)
+ if (Instruction *OInst = dyn_cast<Instruction>(OI))
+ if (!SE.DT->dominates(OInst, IVIncInsertPos))
+ return false;
+ }
+ break;
+ }
+ IncV = dyn_cast<Instruction>(IncV->getOperand(0));
+ if (IncV && IncV->getOpcode() == Instruction::BitCast)
+ IncV = dyn_cast<Instruction>(IncV->getOperand(0));
+ return IncV == PN;
+ }
+ default:
+ return false;
+ }
+}
+
/// getAddRecExprPHILiterally - Helper for expandAddRecExprLiterally. Expand
/// the base addrec, which is the addrec without any non-loop-dominating
/// values, and return the PHI.
PHINode *
SCEVExpander::getAddRecExprPHILiterally(const SCEVAddRecExpr *Normalized,
const Loop *L,
- const Type *ExpandTy,
- const Type *IntTy) {
+ Type *ExpandTy,
+ Type *IntTy) {
+ assert((!IVIncInsertLoop||IVIncInsertPos) && "Uninitialized insert position");
+
// Reuse a previously-inserted PHI, if present.
- for (BasicBlock::iterator I = L->getHeader()->begin();
- PHINode *PN = dyn_cast<PHINode>(I); ++I)
- if (SE.isSCEVable(PN->getType()) &&
- (SE.getEffectiveSCEVType(PN->getType()) ==
- SE.getEffectiveSCEVType(Normalized->getType())) &&
- SE.getSCEV(PN) == Normalized)
- if (BasicBlock *LatchBlock = L->getLoopLatch()) {
- Instruction *IncV =
- cast<Instruction>(PN->getIncomingValueForBlock(LatchBlock));
+ BasicBlock *LatchBlock = L->getLoopLatch();
+ if (LatchBlock) {
+ for (BasicBlock::iterator I = L->getHeader()->begin();
+ PHINode *PN = dyn_cast<PHINode>(I); ++I) {
+ if (!SE.isSCEVable(PN->getType()) ||
+ (SE.getEffectiveSCEVType(PN->getType()) !=
+ SE.getEffectiveSCEVType(Normalized->getType())) ||
+ SE.getSCEV(PN) != Normalized)
+ continue;
- // Determine if this is a well-behaved chain of instructions leading
- // back to the PHI. It probably will be, if we're scanning an inner
- // loop already visited by LSR for example, but it wouldn't have
- // to be.
- do {
- if (IncV->getNumOperands() == 0 || isa<PHINode>(IncV)) {
- IncV = 0;
- break;
- }
- // If any of the operands don't dominate the insert position, bail.
- // Addrec operands are always loop-invariant, so this can only happen
- // if there are instructions which haven't been hoisted.
- for (User::op_iterator OI = IncV->op_begin()+1,
- OE = IncV->op_end(); OI != OE; ++OI)
- if (Instruction *OInst = dyn_cast<Instruction>(OI))
- if (!SE.DT->dominates(OInst, IVIncInsertPos)) {
- IncV = 0;
- break;
- }
- if (!IncV)
- break;
- // Advance to the next instruction.
- IncV = dyn_cast<Instruction>(IncV->getOperand(0));
- if (!IncV)
- break;
- if (IncV->mayHaveSideEffects()) {
- IncV = 0;
- break;
- }
- } while (IncV != PN);
+ Instruction *IncV =
+ cast<Instruction>(PN->getIncomingValueForBlock(LatchBlock));
- if (IncV) {
- // Ok, the add recurrence looks usable.
- // Remember this PHI, even in post-inc mode.
- InsertedValues.insert(PN);
- // Remember the increment.
- IncV = cast<Instruction>(PN->getIncomingValueForBlock(LatchBlock));
- rememberInstruction(IncV);
- if (L == IVIncInsertLoop)
- do {
- if (SE.DT->dominates(IncV, IVIncInsertPos))
- break;
- // Make sure the increment is where we want it. But don't move it
- // down past a potential existing post-inc user.
- IncV->moveBefore(IVIncInsertPos);
- IVIncInsertPos = IncV;
- IncV = cast<Instruction>(IncV->getOperand(0));
- } while (IncV != PN);
- return PN;
- }
+ if (LSRMode) {
+ if (!isExpandedAddRecExprPHI(PN, IncV, L))
+ continue;
}
+ else {
+ if (!isNormalAddRecExprPHI(PN, IncV, L))
+ continue;
+ }
+ // Ok, the add recurrence looks usable.
+ // Remember this PHI, even in post-inc mode.
+ InsertedValues.insert(PN);
+ // Remember the increment.
+ rememberInstruction(IncV);
+ if (L == IVIncInsertLoop)
+ do {
+ if (SE.DT->dominates(IncV, IVIncInsertPos))
+ break;
+ // Make sure the increment is where we want it. But don't move it
+ // down past a potential existing post-inc user.
+ IncV->moveBefore(IVIncInsertPos);
+ IVIncInsertPos = IncV;
+ IncV = cast<Instruction>(IncV->getOperand(0));
+ } while (IncV != PN);
+ return PN;
+ }
+ }
// Save the original insertion point so we can restore it when we're done.
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
@@ -898,6 +988,11 @@
Value *StartV = expandCodeFor(Normalized->getStart(), ExpandTy,
L->getHeader()->begin());
+ // StartV must be hoisted into L's preheader to dominate the new phi.
+ assert(!isa<Instruction>(StartV) ||
+ SE.DT->properlyDominates(cast<Instruction>(StartV)->getParent(),
+ L->getHeader()));
+
// Expand code for the step value. Insert instructions right before the
// terminator corresponding to the back-edge. Do this before creating the PHI
// so that PHI reuse code doesn't see an incomplete PHI. If the stride is
@@ -911,14 +1006,15 @@
Value *StepV = expandCodeFor(Step, IntTy, L->getHeader()->begin());
// Create the PHI.
- Builder.SetInsertPoint(L->getHeader(), L->getHeader()->begin());
- PHINode *PN = Builder.CreatePHI(ExpandTy);
+ BasicBlock *Header = L->getHeader();
+ Builder.SetInsertPoint(Header, Header->begin());
+ pred_iterator HPB = pred_begin(Header), HPE = pred_end(Header);
+ PHINode *PN = Builder.CreatePHI(ExpandTy, std::distance(HPB, HPE),
+ Twine(IVName) + ".iv");
rememberInstruction(PN);
// Create the step instructions and populate the PHI.
- BasicBlock *Header = L->getHeader();
- for (pred_iterator HPI = pred_begin(Header), HPE = pred_end(Header);
- HPI != HPE; ++HPI) {
+ for (pred_iterator HPI = HPB; HPI != HPE; ++HPI) {
BasicBlock *Pred = *HPI;
// Add a start value.
@@ -932,11 +1028,11 @@
// at IVIncInsertPos.
Instruction *InsertPos = L == IVIncInsertLoop ?
IVIncInsertPos : Pred->getTerminator();
- Builder.SetInsertPoint(InsertPos->getParent(), InsertPos);
+ Builder.SetInsertPoint(InsertPos);
Value *IncV;
// If the PHI is a pointer, use a GEP, otherwise use an add or sub.
if (isPointer) {
- const PointerType *GEPPtrTy = cast<PointerType>(ExpandTy);
+ PointerType *GEPPtrTy = cast<PointerType>(ExpandTy);
// If the step isn't constant, don't use an implicitly scaled GEP, because
// that would require a multiply inside the loop.
if (!isa<ConstantInt>(StepV))
@@ -950,8 +1046,8 @@
}
} else {
IncV = isNegative ?
- Builder.CreateSub(PN, StepV) :
- Builder.CreateAdd(PN, StepV);
+ Builder.CreateSub(PN, StepV, Twine(IVName) + ".iv.next") :
+ Builder.CreateAdd(PN, StepV, Twine(IVName) + ".iv.next");
rememberInstruction(IncV);
}
PN->addIncoming(IncV, Pred);
@@ -968,8 +1064,8 @@
}
Value *SCEVExpander::expandAddRecExprLiterally(const SCEVAddRecExpr *S) {
- const Type *STy = S->getType();
- const Type *IntTy = SE.getEffectiveSCEVType(STy);
+ Type *STy = S->getType();
+ Type *IntTy = SE.getEffectiveSCEVType(STy);
const Loop *L = S->getLoop();
// Determine a normalized form of this expression, which is the expression
@@ -986,29 +1082,33 @@
// Strip off any non-loop-dominating component from the addrec start.
const SCEV *Start = Normalized->getStart();
const SCEV *PostLoopOffset = 0;
- if (!Start->properlyDominates(L->getHeader(), SE.DT)) {
+ if (!SE.properlyDominates(Start, L->getHeader())) {
PostLoopOffset = Start;
Start = SE.getConstant(Normalized->getType(), 0);
- Normalized =
- cast<SCEVAddRecExpr>(SE.getAddRecExpr(Start,
- Normalized->getStepRecurrence(SE),
- Normalized->getLoop()));
+ Normalized = cast<SCEVAddRecExpr>(
+ SE.getAddRecExpr(Start, Normalized->getStepRecurrence(SE),
+ Normalized->getLoop(),
+ // FIXME: Normalized->getNoWrapFlags(FlagNW)
+ SCEV::FlagAnyWrap));
}
// Strip off any non-loop-dominating component from the addrec step.
const SCEV *Step = Normalized->getStepRecurrence(SE);
const SCEV *PostLoopScale = 0;
- if (!Step->dominates(L->getHeader(), SE.DT)) {
+ if (!SE.dominates(Step, L->getHeader())) {
PostLoopScale = Step;
Step = SE.getConstant(Normalized->getType(), 1);
Normalized =
cast<SCEVAddRecExpr>(SE.getAddRecExpr(Start, Step,
- Normalized->getLoop()));
+ Normalized->getLoop(),
+ // FIXME: Normalized
+ // ->getNoWrapFlags(FlagNW)
+ SCEV::FlagAnyWrap));
}
// Expand the core addrec. If we need post-loop scaling, force it to
// expand to an integer type to avoid the need for additional casting.
- const Type *ExpandTy = PostLoopScale ? IntTy : STy;
+ Type *ExpandTy = PostLoopScale ? IntTy : STy;
PHINode *PN = getAddRecExprPHILiterally(Normalized, L, ExpandTy, IntTy);
// Accommodate post-inc mode, if necessary.
@@ -1020,6 +1120,14 @@
BasicBlock *LatchBlock = L->getLoopLatch();
assert(LatchBlock && "PostInc mode requires a unique loop latch!");
Result = PN->getIncomingValueForBlock(LatchBlock);
+
+ // For an expansion to use the postinc form, the client must call
+ // expandCodeFor with an InsertPoint that is either outside the PostIncLoop
+ // or dominated by IVIncInsertPos.
+ assert((!isa<Instruction>(Result) ||
+ SE.DT->dominates(cast<Instruction>(Result),
+ Builder.GetInsertPoint())) &&
+ "postinc expansion does not dominate use");
}
// Re-apply any non-loop-dominating scale.
@@ -1032,7 +1140,7 @@
// Re-apply any non-loop-dominating offset.
if (PostLoopOffset) {
- if (const PointerType *PTy = dyn_cast<PointerType>(ExpandTy)) {
+ if (PointerType *PTy = dyn_cast<PointerType>(ExpandTy)) {
const SCEV *const OffsetArray[1] = { PostLoopOffset };
Result = expandAddToGEP(OffsetArray, OffsetArray+1, PTy, IntTy, Result);
} else {
@@ -1049,7 +1157,7 @@
Value *SCEVExpander::visitAddRecExpr(const SCEVAddRecExpr *S) {
if (!CanonicalMode) return expandAddRecExprLiterally(S);
- const Type *Ty = SE.getEffectiveSCEVType(S->getType());
+ Type *Ty = SE.getEffectiveSCEVType(S->getType());
const Loop *L = S->getLoop();
// First check for an existing canonical IV in a suitable type.
@@ -1066,12 +1174,15 @@
SmallVector<const SCEV *, 4> NewOps(S->getNumOperands());
for (unsigned i = 0, e = S->getNumOperands(); i != e; ++i)
NewOps[i] = SE.getAnyExtendExpr(S->op_begin()[i], CanonicalIV->getType());
- Value *V = expand(SE.getAddRecExpr(NewOps, S->getLoop()));
+ Value *V = expand(SE.getAddRecExpr(NewOps, S->getLoop(),
+ // FIXME: S->getNoWrapFlags(FlagNW)
+ SCEV::FlagAnyWrap));
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
BasicBlock::iterator SaveInsertPt = Builder.GetInsertPoint();
BasicBlock::iterator NewInsertPt =
llvm::next(BasicBlock::iterator(cast<Instruction>(V)));
- while (isa<PHINode>(NewInsertPt) || ISA_DEBUG_INFO_INTRINSIC(NewInsertPt))
+ while (isa<PHINode>(NewInsertPt) || isa<DbgInfoIntrinsic>(NewInsertPt) ||
+ isa<LandingPadInst>(NewInsertPt))
++NewInsertPt;
V = expandCodeFor(SE.getTruncateExpr(SE.getUnknown(V), Ty), 0,
NewInsertPt);
@@ -1083,7 +1194,8 @@
if (!S->getStart()->isZero()) {
SmallVector<const SCEV *, 4> NewOps(S->op_begin(), S->op_end());
NewOps[0] = SE.getConstant(Ty, 0);
- const SCEV *Rest = SE.getAddRecExpr(NewOps, L);
+ // FIXME: can use S->getNoWrapFlags()
+ const SCEV *Rest = SE.getAddRecExpr(NewOps, L, SCEV::FlagAnyWrap);
// Turn things like ptrtoint+arithmetic+inttoptr into GEP. See the
// comments on expandAddToGEP for details.
@@ -1092,7 +1204,7 @@
// Dig into the expression to find the pointer base for a GEP.
ExposePointerBase(Base, RestArray[0], SE);
// If we found a pointer, expand the AddRec with a GEP.
- if (const PointerType *PTy = dyn_cast<PointerType>(Base->getType())) {
+ if (PointerType *PTy = dyn_cast<PointerType>(Base->getType())) {
// Make sure the Base isn't something exotic, such as a multiplied
// or divided pointer value. In those cases, the result type isn't
// actually a pointer type.
@@ -1113,18 +1225,21 @@
// Create and insert the PHI node for the induction variable in the
// specified loop.
BasicBlock *Header = L->getHeader();
- CanonicalIV = PHINode::Create(Ty, Header->begin());
+ pred_iterator HPB = pred_begin(Header), HPE = pred_end(Header);
+ CanonicalIV = PHINode::Create(Ty, std::distance(HPB, HPE), "indvar",
+ Header->begin());
rememberInstruction(CanonicalIV);
Constant *One = ConstantInt::get(Ty, 1);
- for (pred_iterator HPI = pred_begin(Header), HPE = pred_end(Header);
- HPI != HPE; ++HPI) {
+ for (pred_iterator HPI = HPB; HPI != HPE; ++HPI) {
BasicBlock *HP = *HPI;
if (L->contains(HP)) {
// Insert a unit add instruction right before the terminator
// corresponding to the back-edge.
Instruction *Add = BinaryOperator::CreateAdd(CanonicalIV, One,
+ "indvar.next",
HP->getTerminator());
+ Add->setDebugLoc(HP->getTerminator()->getDebugLoc());
rememberInstruction(Add);
CanonicalIV->addIncoming(Add, HP);
} else {
@@ -1173,7 +1288,7 @@
}
Value *SCEVExpander::visitTruncateExpr(const SCEVTruncateExpr *S) {
- const Type *Ty = SE.getEffectiveSCEVType(S->getType());
+ Type *Ty = SE.getEffectiveSCEVType(S->getType());
Value *V = expandCodeFor(S->getOperand(),
SE.getEffectiveSCEVType(S->getOperand()->getType()));
Value *I = Builder.CreateTrunc(V, Ty);
@@ -1182,7 +1297,7 @@
}
Value *SCEVExpander::visitZeroExtendExpr(const SCEVZeroExtendExpr *S) {
- const Type *Ty = SE.getEffectiveSCEVType(S->getType());
+ Type *Ty = SE.getEffectiveSCEVType(S->getType());
Value *V = expandCodeFor(S->getOperand(),
SE.getEffectiveSCEVType(S->getOperand()->getType()));
Value *I = Builder.CreateZExt(V, Ty);
@@ -1191,7 +1306,7 @@
}
Value *SCEVExpander::visitSignExtendExpr(const SCEVSignExtendExpr *S) {
- const Type *Ty = SE.getEffectiveSCEVType(S->getType());
+ Type *Ty = SE.getEffectiveSCEVType(S->getType());
Value *V = expandCodeFor(S->getOperand(),
SE.getEffectiveSCEVType(S->getOperand()->getType()));
Value *I = Builder.CreateSExt(V, Ty);
@@ -1201,7 +1316,7 @@
Value *SCEVExpander::visitSMaxExpr(const SCEVSMaxExpr *S) {
Value *LHS = expand(S->getOperand(S->getNumOperands()-1));
- const Type *Ty = LHS->getType();
+ Type *Ty = LHS->getType();
for (int i = S->getNumOperands()-2; i >= 0; --i) {
// In the case of mixed integer and pointer types, do the
// rest of the comparisons as integer.
@@ -1212,7 +1327,7 @@
Value *RHS = expandCodeFor(S->getOperand(i), Ty);
Value *ICmp = Builder.CreateICmpSGT(LHS, RHS);
rememberInstruction(ICmp);
- Value *Sel = Builder.CreateSelect(ICmp, LHS, RHS);
+ Value *Sel = Builder.CreateSelect(ICmp, LHS, RHS, "smax");
rememberInstruction(Sel);
LHS = Sel;
}
@@ -1225,7 +1340,7 @@
Value *SCEVExpander::visitUMaxExpr(const SCEVUMaxExpr *S) {
Value *LHS = expand(S->getOperand(S->getNumOperands()-1));
- const Type *Ty = LHS->getType();
+ Type *Ty = LHS->getType();
for (int i = S->getNumOperands()-2; i >= 0; --i) {
// In the case of mixed integer and pointer types, do the
// rest of the comparisons as integer.
@@ -1236,7 +1351,7 @@
Value *RHS = expandCodeFor(S->getOperand(i), Ty);
Value *ICmp = Builder.CreateICmpUGT(LHS, RHS);
rememberInstruction(ICmp);
- Value *Sel = Builder.CreateSelect(ICmp, LHS, RHS);
+ Value *Sel = Builder.CreateSelect(ICmp, LHS, RHS, "umax");
rememberInstruction(Sel);
LHS = Sel;
}
@@ -1247,16 +1362,16 @@
return LHS;
}
-Value *SCEVExpander::expandCodeFor(const SCEV *SH, const Type *Ty,
+Value *SCEVExpander::expandCodeFor(const SCEV *SH, Type *Ty,
Instruction *I) {
BasicBlock::iterator IP = I;
- while (isInsertedInstruction(IP) || ISA_DEBUG_INFO_INTRINSIC(IP))
+ while (isInsertedInstruction(IP) || isa<DbgInfoIntrinsic>(IP))
++IP;
Builder.SetInsertPoint(IP->getParent(), IP);
return expandCodeFor(SH, Ty);
}
-Value *SCEVExpander::expandCodeFor(const SCEV *SH, const Type *Ty) {
+Value *SCEVExpander::expandCodeFor(const SCEV *SH, Type *Ty) {
// Expand the code for this SCEV.
Value *V = expand(SH);
if (Ty) {
@@ -1273,7 +1388,7 @@
Instruction *InsertPt = Builder.GetInsertPoint();
for (Loop *L = SE.LI->getLoopFor(Builder.GetInsertBlock()); ;
L = L->getParentLoop())
- if (S->isLoopInvariant(L)) {
+ if (SE.isLoopInvariant(S, L)) {
if (!L) break;
if (BasicBlock *Preheader = L->getLoopPreheader())
InsertPt = Preheader->getTerminator();
@@ -1281,9 +1396,9 @@
// If the SCEV is computable at this level, insert it into the header
// after the PHIs (and after any other instructions that we've inserted
// there) so that it is guaranteed to dominate any user inside the loop.
- if (L && S->hasComputableLoopEvolution(L) && !PostIncLoops.count(L))
- InsertPt = L->getHeader()->getFirstNonPHI();
- while (isInsertedInstruction(InsertPt) || ISA_DEBUG_INFO_INTRINSIC(InsertPt))
+ if (L && SE.hasComputableLoopEvolution(S, L) && !PostIncLoops.count(L))
+ InsertPt = L->getHeader()->getFirstInsertionPt();
+ while (isInsertedInstruction(InsertPt) || isa<DbgInfoIntrinsic>(InsertPt))
InsertPt = llvm::next(BasicBlock::iterator(InsertPt));
break;
}
@@ -1303,8 +1418,12 @@
Value *V = visit(S);
// Remember the expanded value for this SCEV at this location.
- if (PostIncLoops.empty())
- InsertedExpressions[std::make_pair(S, InsertPt)] = V;
+ //
+ // This is independent of PostIncLoops. The mapped value simply materializes
+ // the expression at this insertion point. If the mapped value happened to be
+ // a postinc expansion, it could be reused by a non postinc user, but only if
+ // its insertion point was already at the head of the loop.
+ InsertedExpressions[std::make_pair(S, InsertPt)] = V;
restoreInsertPoint(SaveInsertBB, SaveInsertPt);
return V;
@@ -1317,12 +1436,12 @@
InsertedValues.insert(I);
// If we just claimed an existing instruction and that instruction had
- // been the insert point, adjust the insert point forward so that
+ // been the insert point, adjust the insert point forward so that
// subsequently inserted code will be dominated.
if (Builder.GetInsertPoint() == I) {
BasicBlock::iterator It = cast<Instruction>(I);
do { ++It; } while (isInsertedInstruction(It) ||
- ISA_DEBUG_INFO_INTRINSIC(It));
+ isa<DbgInfoIntrinsic>(It));
Builder.SetInsertPoint(Builder.GetInsertBlock(), It);
}
}
@@ -1330,7 +1449,7 @@
void SCEVExpander::restoreInsertPoint(BasicBlock *BB, BasicBlock::iterator I) {
// If we acquired more instructions since the old insert point was saved,
// advance past them.
- while (isInsertedInstruction(I) || ISA_DEBUG_INFO_INTRINSIC(I)) ++I;
+ while (isInsertedInstruction(I) || isa<DbgInfoIntrinsic>(I)) ++I;
Builder.SetInsertPoint(BB, I);
}
@@ -1341,12 +1460,13 @@
/// starts at zero and steps by one on each iteration.
PHINode *
SCEVExpander::getOrInsertCanonicalInductionVariable(const Loop *L,
- const Type *Ty) {
+ Type *Ty) {
assert(Ty->isIntegerTy() && "Can only insert integer induction variables!");
// Build a SCEV for {0,+,1}<L>.
+ // Conservatively use FlagAnyWrap for now.
const SCEV *H = SE.getAddRecExpr(SE.getConstant(Ty, 0),
- SE.getConstant(Ty, 1), L);
+ SE.getConstant(Ty, 1), L, SCEV::FlagAnyWrap);
// Emit code for it.
BasicBlock *SaveInsertBB = Builder.GetInsertBlock();
@@ -1357,3 +1477,102 @@
return V;
}
+
+/// hoistStep - Attempt to hoist an IV increment above a potential use.
+///
+/// To successfully hoist, two criteria must be met:
+/// - IncV operands dominate InsertPos and
+/// - InsertPos dominates IncV
+///
+/// Meeting the second condition means that we don't need to check all of IncV's
+/// existing uses (it's moving up in the domtree).
+///
+/// This does not yet recursively hoist the operands, although that would
+/// not be difficult.
+///
+/// This does not require a SCEVExpander instance and could be replaced by a
+/// general code-insertion helper.
+bool SCEVExpander::hoistStep(Instruction *IncV, Instruction *InsertPos,
+ const DominatorTree *DT) {
+ if (DT->dominates(IncV, InsertPos))
+ return true;
+
+ if (!DT->dominates(InsertPos->getParent(), IncV->getParent()))
+ return false;
+
+ if (IncV->mayHaveSideEffects())
+ return false;
+
+ // Attempt to hoist IncV
+ for (User::op_iterator OI = IncV->op_begin(), OE = IncV->op_end();
+ OI != OE; ++OI) {
+ Instruction *OInst = dyn_cast<Instruction>(OI);
+ if (OInst && !DT->dominates(OInst, InsertPos))
+ return false;
+ }
+ IncV->moveBefore(InsertPos);
+ return true;
+}
+
+/// replaceCongruentIVs - Check for congruent phis in this loop header and
+/// replace them with their most canonical representative. Return the number of
+/// phis eliminated.
+///
+/// This does not depend on any SCEVExpander state but should be used in
+/// the same context that SCEVExpander is used.
+unsigned SCEVExpander::replaceCongruentIVs(Loop *L, const DominatorTree *DT,
+ SmallVectorImpl<WeakVH> &DeadInsts) {
+ unsigned NumElim = 0;
+ DenseMap<const SCEV *, PHINode *> ExprToIVMap;
+ for (BasicBlock::iterator I = L->getHeader()->begin(); isa<PHINode>(I); ++I) {
+ PHINode *Phi = cast<PHINode>(I);
+ if (!SE.isSCEVable(Phi->getType()))
+ continue;
+
+ PHINode *&OrigPhiRef = ExprToIVMap[SE.getSCEV(Phi)];
+ if (!OrigPhiRef) {
+ OrigPhiRef = Phi;
+ continue;
+ }
+
+ // If one phi derives from the other via GEPs, types may differ.
+ // We could consider adding a bitcast here to handle it.
+ if (OrigPhiRef->getType() != Phi->getType())
+ continue;
+
+ if (BasicBlock *LatchBlock = L->getLoopLatch()) {
+ Instruction *OrigInc =
+ cast<Instruction>(OrigPhiRef->getIncomingValueForBlock(LatchBlock));
+ Instruction *IsomorphicInc =
+ cast<Instruction>(Phi->getIncomingValueForBlock(LatchBlock));
+
+ // If this phi is more canonical, swap it with the original.
+ if (!isExpandedAddRecExprPHI(OrigPhiRef, OrigInc, L)
+ && isExpandedAddRecExprPHI(Phi, IsomorphicInc, L)) {
+ std::swap(OrigPhiRef, Phi);
+ std::swap(OrigInc, IsomorphicInc);
+ }
+ // Replacing the congruent phi is sufficient because acyclic redundancy
+ // elimination, CSE/GVN, should handle the rest. However, once SCEV proves
+ // that a phi is congruent, it's often the head of an IV user cycle that
+ // is isomorphic with the original phi. So it's worth eagerly cleaning up
+ // the common case of a single IV increment.
+ if (OrigInc != IsomorphicInc &&
+ OrigInc->getType() == IsomorphicInc->getType() &&
+ SE.getSCEV(OrigInc) == SE.getSCEV(IsomorphicInc) &&
+ hoistStep(OrigInc, IsomorphicInc, DT)) {
+ DEBUG_WITH_TYPE(DebugType, dbgs()
+ << "INDVARS: Eliminated congruent iv.inc: "
+ << *IsomorphicInc << '\n');
+ IsomorphicInc->replaceAllUsesWith(OrigInc);
+ DeadInsts.push_back(IsomorphicInc);
+ }
+ }
+ DEBUG_WITH_TYPE(DebugType, dbgs()
+ << "INDVARS: Eliminated congruent iv: " << *Phi << '\n');
+ ++NumElim;
+ Phi->replaceAllUsesWith(OrigPhiRef);
+ DeadInsts.push_back(Phi);
+ }
+ return NumElim;
+}
diff --git a/src/LLVM/lib/Analysis/ScalarEvolutionNormalization.cpp b/src/LLVM/lib/Analysis/ScalarEvolutionNormalization.cpp
index 498387a..c66ecd6 100644
--- a/src/LLVM/lib/Analysis/ScalarEvolutionNormalization.cpp
+++ b/src/LLVM/lib/Analysis/ScalarEvolutionNormalization.cpp
@@ -60,20 +60,40 @@
return true;
}
-const SCEV *llvm::TransformForPostIncUse(TransformKind Kind,
- const SCEV *S,
- Instruction *User,
- Value *OperandValToReplace,
- PostIncLoopSet &Loops,
- ScalarEvolution &SE,
- DominatorTree &DT) {
- if (isa<SCEVConstant>(S) || isa<SCEVUnknown>(S))
- return S;
+namespace {
+
+/// Hold the state used during post-inc expression transformation, including a
+/// map of transformed expressions.
+class PostIncTransform {
+ TransformKind Kind;
+ PostIncLoopSet &Loops;
+ ScalarEvolution &SE;
+ DominatorTree &DT;
+
+ DenseMap<const SCEV*, const SCEV*> Transformed;
+
+public:
+ PostIncTransform(TransformKind kind, PostIncLoopSet &loops,
+ ScalarEvolution &se, DominatorTree &dt):
+ Kind(kind), Loops(loops), SE(se), DT(dt) {}
+
+ const SCEV *TransformSubExpr(const SCEV *S, Instruction *User,
+ Value *OperandValToReplace);
+
+protected:
+ const SCEV *TransformImpl(const SCEV *S, Instruction *User,
+ Value *OperandValToReplace);
+};
+
+} // namespace
+
+/// Implement post-inc transformation for all valid expression types.
+const SCEV *PostIncTransform::
+TransformImpl(const SCEV *S, Instruction *User, Value *OperandValToReplace) {
if (const SCEVCastExpr *X = dyn_cast<SCEVCastExpr>(S)) {
const SCEV *O = X->getOperand();
- const SCEV *N = TransformForPostIncUse(Kind, O, User, OperandValToReplace,
- Loops, SE, DT);
+ const SCEV *N = TransformSubExpr(O, User, OperandValToReplace);
if (O != N)
switch (S->getSCEVType()) {
case scZeroExtend: return SE.getZeroExtendExpr(N, S->getType());
@@ -93,39 +113,39 @@
// Transform each operand.
for (SCEVNAryExpr::op_iterator I = AR->op_begin(), E = AR->op_end();
I != E; ++I) {
- const SCEV *O = *I;
- const SCEV *N = TransformForPostIncUse(Kind, O, LUser, 0, Loops, SE, DT);
- Operands.push_back(N);
+ Operands.push_back(TransformSubExpr(*I, LUser, 0));
}
- const SCEV *Result = SE.getAddRecExpr(Operands, L);
+ // Conservatively use AnyWrap until/unless we need FlagNW.
+ const SCEV *Result = SE.getAddRecExpr(Operands, L, SCEV::FlagAnyWrap);
switch (Kind) {
default: llvm_unreachable("Unexpected transform name!");
case NormalizeAutodetect:
if (IVUseShouldUsePostIncValue(User, OperandValToReplace, L, &DT)) {
const SCEV *TransformedStep =
- TransformForPostIncUse(Kind, AR->getStepRecurrence(SE),
- User, OperandValToReplace, Loops, SE, DT);
+ TransformSubExpr(AR->getStepRecurrence(SE),
+ User, OperandValToReplace);
Result = SE.getMinusSCEV(Result, TransformedStep);
Loops.insert(L);
}
-#ifdef XDEBUG
- assert(S == TransformForPostIncUse(Denormalize, Result,
- User, OperandValToReplace,
- Loops, SE, DT) &&
+#if 0
+ // This assert is conceptually correct, but ScalarEvolution currently
+ // sometimes fails to canonicalize two equal SCEVs to exactly the same
+ // form. It's possibly a pessimization when this happens, but it isn't a
+ // correctness problem, so disable this assert for now.
+ assert(S == TransformSubExpr(Result, User, OperandValToReplace) &&
"SCEV normalization is not invertible!");
#endif
break;
case Normalize:
if (Loops.count(L)) {
const SCEV *TransformedStep =
- TransformForPostIncUse(Kind, AR->getStepRecurrence(SE),
- User, OperandValToReplace, Loops, SE, DT);
+ TransformSubExpr(AR->getStepRecurrence(SE),
+ User, OperandValToReplace);
Result = SE.getMinusSCEV(Result, TransformedStep);
}
-#ifdef XDEBUG
- assert(S == TransformForPostIncUse(Denormalize, Result,
- User, OperandValToReplace,
- Loops, SE, DT) &&
+#if 0
+ // See the comment on the assert above.
+ assert(S == TransformSubExpr(Result, User, OperandValToReplace) &&
"SCEV normalization is not invertible!");
#endif
break;
@@ -144,8 +164,7 @@
for (SCEVNAryExpr::op_iterator I = X->op_begin(), E = X->op_end();
I != E; ++I) {
const SCEV *O = *I;
- const SCEV *N = TransformForPostIncUse(Kind, O, User, OperandValToReplace,
- Loops, SE, DT);
+ const SCEV *N = TransformSubExpr(O, User, OperandValToReplace);
Changed |= N != O;
Operands.push_back(N);
}
@@ -164,10 +183,8 @@
if (const SCEVUDivExpr *X = dyn_cast<SCEVUDivExpr>(S)) {
const SCEV *LO = X->getLHS();
const SCEV *RO = X->getRHS();
- const SCEV *LN = TransformForPostIncUse(Kind, LO, User, OperandValToReplace,
- Loops, SE, DT);
- const SCEV *RN = TransformForPostIncUse(Kind, RO, User, OperandValToReplace,
- Loops, SE, DT);
+ const SCEV *LN = TransformSubExpr(LO, User, OperandValToReplace);
+ const SCEV *RN = TransformSubExpr(RO, User, OperandValToReplace);
if (LO != LN || RO != RN)
return SE.getUDivExpr(LN, RN);
return S;
@@ -176,3 +193,33 @@
llvm_unreachable("Unexpected SCEV kind!");
return 0;
}
+
+/// Manage recursive transformation across an expression DAG. Revisiting
+/// expressions would lead to exponential recursion.
+const SCEV *PostIncTransform::
+TransformSubExpr(const SCEV *S, Instruction *User, Value *OperandValToReplace) {
+
+ if (isa<SCEVConstant>(S) || isa<SCEVUnknown>(S))
+ return S;
+
+ const SCEV *Result = Transformed.lookup(S);
+ if (Result)
+ return Result;
+
+ Result = TransformImpl(S, User, OperandValToReplace);
+ Transformed[S] = Result;
+ return Result;
+}
+
+/// Top level driver for transforming an expression DAG into its requested
+/// post-inc form (either "Normalized" or "Denormalized".
+const SCEV *llvm::TransformForPostIncUse(TransformKind Kind,
+ const SCEV *S,
+ Instruction *User,
+ Value *OperandValToReplace,
+ PostIncLoopSet &Loops,
+ ScalarEvolution &SE,
+ DominatorTree &DT) {
+ PostIncTransform Transform(Kind, Loops, SE, DT);
+ return Transform.TransformSubExpr(S, User, OperandValToReplace);
+}
diff --git a/src/LLVM/lib/Analysis/SparsePropagation.cpp b/src/LLVM/lib/Analysis/SparsePropagation.cpp
new file mode 100644
index 0000000..d8c207b
--- /dev/null
+++ b/src/LLVM/lib/Analysis/SparsePropagation.cpp
@@ -0,0 +1,347 @@
+//===- SparsePropagation.cpp - Sparse Conditional Property Propagation ----===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements an abstract sparse conditional propagation algorithm,
+// modeled after SCCP, but with a customizable lattice function.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "sparseprop"
+#include "llvm/Analysis/SparsePropagation.h"
+#include "llvm/Constants.h"
+#include "llvm/Function.h"
+#include "llvm/Instructions.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+//===----------------------------------------------------------------------===//
+// AbstractLatticeFunction Implementation
+//===----------------------------------------------------------------------===//
+
+AbstractLatticeFunction::~AbstractLatticeFunction() {}
+
+/// PrintValue - Render the specified lattice value to the specified stream.
+void AbstractLatticeFunction::PrintValue(LatticeVal V, raw_ostream &OS) {
+ if (V == UndefVal)
+ OS << "undefined";
+ else if (V == OverdefinedVal)
+ OS << "overdefined";
+ else if (V == UntrackedVal)
+ OS << "untracked";
+ else
+ OS << "unknown lattice value";
+}
+
+//===----------------------------------------------------------------------===//
+// SparseSolver Implementation
+//===----------------------------------------------------------------------===//
+
+/// getOrInitValueState - Return the LatticeVal object that corresponds to the
+/// value, initializing the value's state if it hasn't been entered into the
+/// map yet. This function is necessary because not all values should start
+/// out in the underdefined state... Arguments should be overdefined, and
+/// constants should be marked as constants.
+///
+SparseSolver::LatticeVal SparseSolver::getOrInitValueState(Value *V) {
+ DenseMap<Value*, LatticeVal>::iterator I = ValueState.find(V);
+ if (I != ValueState.end()) return I->second; // Common case, in the map
+
+ LatticeVal LV;
+ if (LatticeFunc->IsUntrackedValue(V))
+ return LatticeFunc->getUntrackedVal();
+ else if (Constant *C = dyn_cast<Constant>(V))
+ LV = LatticeFunc->ComputeConstant(C);
+ else if (Argument *A = dyn_cast<Argument>(V))
+ LV = LatticeFunc->ComputeArgument(A);
+ else if (!isa<Instruction>(V))
+ // All other non-instructions are overdefined.
+ LV = LatticeFunc->getOverdefinedVal();
+ else
+ // All instructions are underdefined by default.
+ LV = LatticeFunc->getUndefVal();
+
+ // If this value is untracked, don't add it to the map.
+ if (LV == LatticeFunc->getUntrackedVal())
+ return LV;
+ return ValueState[V] = LV;
+}
+
+/// UpdateState - When the state for some instruction is potentially updated,
+/// this function notices and adds I to the worklist if needed.
+void SparseSolver::UpdateState(Instruction &Inst, LatticeVal V) {
+ DenseMap<Value*, LatticeVal>::iterator I = ValueState.find(&Inst);
+ if (I != ValueState.end() && I->second == V)
+ return; // No change.
+
+ // An update. Visit uses of I.
+ ValueState[&Inst] = V;
+ InstWorkList.push_back(&Inst);
+}
+
+/// MarkBlockExecutable - This method can be used by clients to mark all of
+/// the blocks that are known to be intrinsically live in the processed unit.
+void SparseSolver::MarkBlockExecutable(BasicBlock *BB) {
+ DEBUG(dbgs() << "Marking Block Executable: " << BB->getName() << "\n");
+ BBExecutable.insert(BB); // Basic block is executable!
+ BBWorkList.push_back(BB); // Add the block to the work list!
+}
+
+/// markEdgeExecutable - Mark a basic block as executable, adding it to the BB
+/// work list if it is not already executable...
+void SparseSolver::markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest) {
+ if (!KnownFeasibleEdges.insert(Edge(Source, Dest)).second)
+ return; // This edge is already known to be executable!
+
+ DEBUG(dbgs() << "Marking Edge Executable: " << Source->getName()
+ << " -> " << Dest->getName() << "\n");
+
+ if (BBExecutable.count(Dest)) {
+ // The destination is already executable, but we just made an edge
+ // feasible that wasn't before. Revisit the PHI nodes in the block
+ // because they have potentially new operands.
+ for (BasicBlock::iterator I = Dest->begin(); isa<PHINode>(I); ++I)
+ visitPHINode(*cast<PHINode>(I));
+
+ } else {
+ MarkBlockExecutable(Dest);
+ }
+}
+
+
+/// getFeasibleSuccessors - Return a vector of booleans to indicate which
+/// successors are reachable from a given terminator instruction.
+void SparseSolver::getFeasibleSuccessors(TerminatorInst &TI,
+ SmallVectorImpl<bool> &Succs,
+ bool AggressiveUndef) {
+ Succs.resize(TI.getNumSuccessors());
+ if (TI.getNumSuccessors() == 0) return;
+
+ if (BranchInst *BI = dyn_cast<BranchInst>(&TI)) {
+ if (BI->isUnconditional()) {
+ Succs[0] = true;
+ return;
+ }
+
+ LatticeVal BCValue;
+ if (AggressiveUndef)
+ BCValue = getOrInitValueState(BI->getCondition());
+ else
+ BCValue = getLatticeState(BI->getCondition());
+
+ if (BCValue == LatticeFunc->getOverdefinedVal() ||
+ BCValue == LatticeFunc->getUntrackedVal()) {
+ // Overdefined condition variables can branch either way.
+ Succs[0] = Succs[1] = true;
+ return;
+ }
+
+ // If undefined, neither is feasible yet.
+ if (BCValue == LatticeFunc->getUndefVal())
+ return;
+
+ Constant *C = LatticeFunc->GetConstant(BCValue, BI->getCondition(), *this);
+ if (C == 0 || !isa<ConstantInt>(C)) {
+ // Non-constant values can go either way.
+ Succs[0] = Succs[1] = true;
+ return;
+ }
+
+ // Constant condition variables mean the branch can only go a single way
+ Succs[C->isNullValue()] = true;
+ return;
+ }
+
+ if (isa<InvokeInst>(TI)) {
+ // Invoke instructions successors are always executable.
+ // TODO: Could ask the lattice function if the value can throw.
+ Succs[0] = Succs[1] = true;
+ return;
+ }
+
+ if (isa<IndirectBrInst>(TI)) {
+ Succs.assign(Succs.size(), true);
+ return;
+ }
+
+ SwitchInst &SI = cast<SwitchInst>(TI);
+ LatticeVal SCValue;
+ if (AggressiveUndef)
+ SCValue = getOrInitValueState(SI.getCondition());
+ else
+ SCValue = getLatticeState(SI.getCondition());
+
+ if (SCValue == LatticeFunc->getOverdefinedVal() ||
+ SCValue == LatticeFunc->getUntrackedVal()) {
+ // All destinations are executable!
+ Succs.assign(TI.getNumSuccessors(), true);
+ return;
+ }
+
+ // If undefined, neither is feasible yet.
+ if (SCValue == LatticeFunc->getUndefVal())
+ return;
+
+ Constant *C = LatticeFunc->GetConstant(SCValue, SI.getCondition(), *this);
+ if (C == 0 || !isa<ConstantInt>(C)) {
+ // All destinations are executable!
+ Succs.assign(TI.getNumSuccessors(), true);
+ return;
+ }
+
+ Succs[SI.findCaseValue(cast<ConstantInt>(C))] = true;
+}
+
+
+/// isEdgeFeasible - Return true if the control flow edge from the 'From'
+/// basic block to the 'To' basic block is currently feasible...
+bool SparseSolver::isEdgeFeasible(BasicBlock *From, BasicBlock *To,
+ bool AggressiveUndef) {
+ SmallVector<bool, 16> SuccFeasible;
+ TerminatorInst *TI = From->getTerminator();
+ getFeasibleSuccessors(*TI, SuccFeasible, AggressiveUndef);
+
+ for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
+ if (TI->getSuccessor(i) == To && SuccFeasible[i])
+ return true;
+
+ return false;
+}
+
+void SparseSolver::visitTerminatorInst(TerminatorInst &TI) {
+ SmallVector<bool, 16> SuccFeasible;
+ getFeasibleSuccessors(TI, SuccFeasible, true);
+
+ BasicBlock *BB = TI.getParent();
+
+ // Mark all feasible successors executable...
+ for (unsigned i = 0, e = SuccFeasible.size(); i != e; ++i)
+ if (SuccFeasible[i])
+ markEdgeExecutable(BB, TI.getSuccessor(i));
+}
+
+void SparseSolver::visitPHINode(PHINode &PN) {
+ // The lattice function may store more information on a PHINode than could be
+ // computed from its incoming values. For example, SSI form stores its sigma
+ // functions as PHINodes with a single incoming value.
+ if (LatticeFunc->IsSpecialCasedPHI(&PN)) {
+ LatticeVal IV = LatticeFunc->ComputeInstructionState(PN, *this);
+ if (IV != LatticeFunc->getUntrackedVal())
+ UpdateState(PN, IV);
+ return;
+ }
+
+ LatticeVal PNIV = getOrInitValueState(&PN);
+ LatticeVal Overdefined = LatticeFunc->getOverdefinedVal();
+
+ // If this value is already overdefined (common) just return.
+ if (PNIV == Overdefined || PNIV == LatticeFunc->getUntrackedVal())
+ return; // Quick exit
+
+ // Super-extra-high-degree PHI nodes are unlikely to ever be interesting,
+ // and slow us down a lot. Just mark them overdefined.
+ if (PN.getNumIncomingValues() > 64) {
+ UpdateState(PN, Overdefined);
+ return;
+ }
+
+ // Look at all of the executable operands of the PHI node. If any of them
+ // are overdefined, the PHI becomes overdefined as well. Otherwise, ask the
+ // transfer function to give us the merge of the incoming values.
+ for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) {
+ // If the edge is not yet known to be feasible, it doesn't impact the PHI.
+ if (!isEdgeFeasible(PN.getIncomingBlock(i), PN.getParent(), true))
+ continue;
+
+ // Merge in this value.
+ LatticeVal OpVal = getOrInitValueState(PN.getIncomingValue(i));
+ if (OpVal != PNIV)
+ PNIV = LatticeFunc->MergeValues(PNIV, OpVal);
+
+ if (PNIV == Overdefined)
+ break; // Rest of input values don't matter.
+ }
+
+ // Update the PHI with the compute value, which is the merge of the inputs.
+ UpdateState(PN, PNIV);
+}
+
+
+void SparseSolver::visitInst(Instruction &I) {
+ // PHIs are handled by the propagation logic, they are never passed into the
+ // transfer functions.
+ if (PHINode *PN = dyn_cast<PHINode>(&I))
+ return visitPHINode(*PN);
+
+ // Otherwise, ask the transfer function what the result is. If this is
+ // something that we care about, remember it.
+ LatticeVal IV = LatticeFunc->ComputeInstructionState(I, *this);
+ if (IV != LatticeFunc->getUntrackedVal())
+ UpdateState(I, IV);
+
+ if (TerminatorInst *TI = dyn_cast<TerminatorInst>(&I))
+ visitTerminatorInst(*TI);
+}
+
+void SparseSolver::Solve(Function &F) {
+ MarkBlockExecutable(&F.getEntryBlock());
+
+ // Process the work lists until they are empty!
+ while (!BBWorkList.empty() || !InstWorkList.empty()) {
+ // Process the instruction work list.
+ while (!InstWorkList.empty()) {
+ Instruction *I = InstWorkList.back();
+ InstWorkList.pop_back();
+
+ DEBUG(dbgs() << "\nPopped off I-WL: " << *I << "\n");
+
+ // "I" got into the work list because it made a transition. See if any
+ // users are both live and in need of updating.
+ for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
+ UI != E; ++UI) {
+ Instruction *U = cast<Instruction>(*UI);
+ if (BBExecutable.count(U->getParent())) // Inst is executable?
+ visitInst(*U);
+ }
+ }
+
+ // Process the basic block work list.
+ while (!BBWorkList.empty()) {
+ BasicBlock *BB = BBWorkList.back();
+ BBWorkList.pop_back();
+
+ DEBUG(dbgs() << "\nPopped off BBWL: " << *BB);
+
+ // Notify all instructions in this basic block that they are newly
+ // executable.
+ for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
+ visitInst(*I);
+ }
+ }
+}
+
+void SparseSolver::Print(Function &F, raw_ostream &OS) const {
+ OS << "\nFUNCTION: " << F.getNameStr() << "\n";
+ for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
+ if (!BBExecutable.count(BB))
+ OS << "INFEASIBLE: ";
+ OS << "\t";
+ if (BB->hasName())
+ OS << BB->getNameStr() << ":\n";
+ else
+ OS << "; anon bb\n";
+ for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
+ LatticeFunc->PrintValue(getLatticeState(I), OS);
+ OS << *I << "\n";
+ }
+
+ OS << "\n";
+ }
+}
+
diff --git a/src/LLVM/lib/Analysis/Trace.cpp b/src/LLVM/lib/Analysis/Trace.cpp
new file mode 100644
index 0000000..68a39cd
--- /dev/null
+++ b/src/LLVM/lib/Analysis/Trace.cpp
@@ -0,0 +1,51 @@
+//===- Trace.cpp - Implementation of Trace class --------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class represents a single trace of LLVM basic blocks. A trace is a
+// single entry, multiple exit, region of code that is often hot. Trace-based
+// optimizations treat traces almost like they are a large, strange, basic
+// block: because the trace path is assumed to be hot, optimizations for the
+// fall-through path are made at the expense of the non-fall-through paths.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/Trace.h"
+#include "llvm/Function.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+Function *Trace::getFunction() const {
+ return getEntryBasicBlock()->getParent();
+}
+
+Module *Trace::getModule() const {
+ return getFunction()->getParent();
+}
+
+/// print - Write trace to output stream.
+///
+void Trace::print(raw_ostream &O) const {
+ Function *F = getFunction();
+ O << "; Trace from function " << F->getNameStr() << ", blocks:\n";
+ for (const_iterator i = begin(), e = end(); i != e; ++i) {
+ O << "; ";
+ WriteAsOperand(O, *i, true, getModule());
+ O << "\n";
+ }
+ O << "; Trace parent function: \n" << *F;
+}
+
+/// dump - Debugger convenience method; writes trace to standard error
+/// output stream.
+///
+void Trace::dump() const {
+ print(dbgs());
+}
diff --git a/src/LLVM/lib/Analysis/TypeBasedAliasAnalysis.cpp b/src/LLVM/lib/Analysis/TypeBasedAliasAnalysis.cpp
new file mode 100644
index 0000000..0faf139
--- /dev/null
+++ b/src/LLVM/lib/Analysis/TypeBasedAliasAnalysis.cpp
@@ -0,0 +1,300 @@
+//===- TypeBasedAliasAnalysis.cpp - Type-Based Alias Analysis -------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the TypeBasedAliasAnalysis pass, which implements
+// metadata-based TBAA.
+//
+// In LLVM IR, memory does not have types, so LLVM's own type system is not
+// suitable for doing TBAA. Instead, metadata is added to the IR to describe
+// a type system of a higher level language. This can be used to implement
+// typical C/C++ TBAA, but it can also be used to implement custom alias
+// analysis behavior for other languages.
+//
+// The current metadata format is very simple. TBAA MDNodes have up to
+// three fields, e.g.:
+// !0 = metadata !{ metadata !"an example type tree" }
+// !1 = metadata !{ metadata !"int", metadata !0 }
+// !2 = metadata !{ metadata !"float", metadata !0 }
+// !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
+//
+// The first field is an identity field. It can be any value, usually
+// an MDString, which uniquely identifies the type. The most important
+// name in the tree is the name of the root node. Two trees with
+// different root node names are entirely disjoint, even if they
+// have leaves with common names.
+//
+// The second field identifies the type's parent node in the tree, or
+// is null or omitted for a root node. A type is considered to alias
+// all of its descendants and all of its ancestors in the tree. Also,
+// a type is considered to alias all types in other trees, so that
+// bitcode produced from multiple front-ends is handled conservatively.
+//
+// If the third field is present, it's an integer which if equal to 1
+// indicates that the type is "constant" (meaning pointsToConstantMemory
+// should return true; see
+// http://llvm.org/docs/AliasAnalysis.html#OtherItfs).
+//
+// TODO: The current metadata format doesn't support struct
+// fields. For example:
+// struct X {
+// double d;
+// int i;
+// };
+// void foo(struct X *x, struct X *y, double *p) {
+// *x = *y;
+// *p = 0.0;
+// }
+// Struct X has a double member, so the store to *x can alias the store to *p.
+// Currently it's not possible to precisely describe all the things struct X
+// aliases, so struct assignments must use conservative TBAA nodes. There's
+// no scheme for attaching metadata to @llvm.memcpy yet either.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Constants.h"
+#include "llvm/LLVMContext.h"
+#include "llvm/Module.h"
+#include "llvm/Metadata.h"
+#include "llvm/Pass.h"
+#include "llvm/Support/CommandLine.h"
+using namespace llvm;
+
+// A handy option for disabling TBAA functionality. The same effect can also be
+// achieved by stripping the !tbaa tags from IR, but this option is sometimes
+// more convenient.
+static cl::opt<bool> EnableTBAA("enable-tbaa", cl::init(true));
+
+namespace {
+ /// TBAANode - This is a simple wrapper around an MDNode which provides a
+ /// higher-level interface by hiding the details of how alias analysis
+ /// information is encoded in its operands.
+ class TBAANode {
+ const MDNode *Node;
+
+ public:
+ TBAANode() : Node(0) {}
+ explicit TBAANode(const MDNode *N) : Node(N) {}
+
+ /// getNode - Get the MDNode for this TBAANode.
+ const MDNode *getNode() const { return Node; }
+
+ /// getParent - Get this TBAANode's Alias tree parent.
+ TBAANode getParent() const {
+ if (Node->getNumOperands() < 2)
+ return TBAANode();
+ MDNode *P = dyn_cast_or_null<MDNode>(Node->getOperand(1));
+ if (!P)
+ return TBAANode();
+ // Ok, this node has a valid parent. Return it.
+ return TBAANode(P);
+ }
+
+ /// TypeIsImmutable - Test if this TBAANode represents a type for objects
+ /// which are not modified (by any means) in the context where this
+ /// AliasAnalysis is relevant.
+ bool TypeIsImmutable() const {
+ if (Node->getNumOperands() < 3)
+ return false;
+ ConstantInt *CI = dyn_cast<ConstantInt>(Node->getOperand(2));
+ if (!CI)
+ return false;
+ return CI->getValue()[0];
+ }
+ };
+}
+
+namespace {
+ /// TypeBasedAliasAnalysis - This is a simple alias analysis
+ /// implementation that uses TypeBased to answer queries.
+ class TypeBasedAliasAnalysis : public ImmutablePass,
+ public AliasAnalysis {
+ public:
+ static char ID; // Class identification, replacement for typeinfo
+ TypeBasedAliasAnalysis() : ImmutablePass(ID) {
+ initializeTypeBasedAliasAnalysisPass(*PassRegistry::getPassRegistry());
+ }
+
+ virtual void initializePass() {
+ InitializeAliasAnalysis(this);
+ }
+
+ /// getAdjustedAnalysisPointer - This method is used when a pass implements
+ /// an analysis interface through multiple inheritance. If needed, it
+ /// should override this to adjust the this pointer as needed for the
+ /// specified pass info.
+ virtual void *getAdjustedAnalysisPointer(const void *PI) {
+ if (PI == &AliasAnalysis::ID)
+ return (AliasAnalysis*)this;
+ return this;
+ }
+
+ bool Aliases(const MDNode *A, const MDNode *B) const;
+
+ private:
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+ virtual AliasResult alias(const Location &LocA, const Location &LocB);
+ virtual bool pointsToConstantMemory(const Location &Loc, bool OrLocal);
+ virtual ModRefBehavior getModRefBehavior(ImmutableCallSite CS);
+ virtual ModRefBehavior getModRefBehavior(const Function *F);
+ virtual ModRefResult getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc);
+ virtual ModRefResult getModRefInfo(ImmutableCallSite CS1,
+ ImmutableCallSite CS2);
+ };
+} // End of anonymous namespace
+
+// Register this pass...
+char TypeBasedAliasAnalysis::ID = 0;
+INITIALIZE_AG_PASS(TypeBasedAliasAnalysis, AliasAnalysis, "tbaa",
+ "Type-Based Alias Analysis", false, true, false)
+
+ImmutablePass *llvm::createTypeBasedAliasAnalysisPass() {
+ return new TypeBasedAliasAnalysis();
+}
+
+void
+TypeBasedAliasAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesAll();
+ AliasAnalysis::getAnalysisUsage(AU);
+}
+
+/// Aliases - Test whether the type represented by A may alias the
+/// type represented by B.
+bool
+TypeBasedAliasAnalysis::Aliases(const MDNode *A,
+ const MDNode *B) const {
+ // Keep track of the root node for A and B.
+ TBAANode RootA, RootB;
+
+ // Climb the tree from A to see if we reach B.
+ for (TBAANode T(A); ; ) {
+ if (T.getNode() == B)
+ // B is an ancestor of A.
+ return true;
+
+ RootA = T;
+ T = T.getParent();
+ if (!T.getNode())
+ break;
+ }
+
+ // Climb the tree from B to see if we reach A.
+ for (TBAANode T(B); ; ) {
+ if (T.getNode() == A)
+ // A is an ancestor of B.
+ return true;
+
+ RootB = T;
+ T = T.getParent();
+ if (!T.getNode())
+ break;
+ }
+
+ // Neither node is an ancestor of the other.
+
+ // If they have different roots, they're part of different potentially
+ // unrelated type systems, so we must be conservative.
+ if (RootA.getNode() != RootB.getNode())
+ return true;
+
+ // If they have the same root, then we've proved there's no alias.
+ return false;
+}
+
+AliasAnalysis::AliasResult
+TypeBasedAliasAnalysis::alias(const Location &LocA,
+ const Location &LocB) {
+ if (!EnableTBAA)
+ return AliasAnalysis::alias(LocA, LocB);
+
+ // Get the attached MDNodes. If either value lacks a tbaa MDNode, we must
+ // be conservative.
+ const MDNode *AM = LocA.TBAATag;
+ if (!AM) return AliasAnalysis::alias(LocA, LocB);
+ const MDNode *BM = LocB.TBAATag;
+ if (!BM) return AliasAnalysis::alias(LocA, LocB);
+
+ // If they may alias, chain to the next AliasAnalysis.
+ if (Aliases(AM, BM))
+ return AliasAnalysis::alias(LocA, LocB);
+
+ // Otherwise return a definitive result.
+ return NoAlias;
+}
+
+bool TypeBasedAliasAnalysis::pointsToConstantMemory(const Location &Loc,
+ bool OrLocal) {
+ if (!EnableTBAA)
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+
+ const MDNode *M = Loc.TBAATag;
+ if (!M) return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+
+ // If this is an "immutable" type, we can assume the pointer is pointing
+ // to constant memory.
+ if (TBAANode(M).TypeIsImmutable())
+ return true;
+
+ return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+}
+
+AliasAnalysis::ModRefBehavior
+TypeBasedAliasAnalysis::getModRefBehavior(ImmutableCallSite CS) {
+ if (!EnableTBAA)
+ return AliasAnalysis::getModRefBehavior(CS);
+
+ ModRefBehavior Min = UnknownModRefBehavior;
+
+ // If this is an "immutable" type, we can assume the call doesn't write
+ // to memory.
+ if (const MDNode *M = CS.getInstruction()->getMetadata(LLVMContext::MD_tbaa))
+ if (TBAANode(M).TypeIsImmutable())
+ Min = OnlyReadsMemory;
+
+ return ModRefBehavior(AliasAnalysis::getModRefBehavior(CS) & Min);
+}
+
+AliasAnalysis::ModRefBehavior
+TypeBasedAliasAnalysis::getModRefBehavior(const Function *F) {
+ // Functions don't have metadata. Just chain to the next implementation.
+ return AliasAnalysis::getModRefBehavior(F);
+}
+
+AliasAnalysis::ModRefResult
+TypeBasedAliasAnalysis::getModRefInfo(ImmutableCallSite CS,
+ const Location &Loc) {
+ if (!EnableTBAA)
+ return AliasAnalysis::getModRefInfo(CS, Loc);
+
+ if (const MDNode *L = Loc.TBAATag)
+ if (const MDNode *M =
+ CS.getInstruction()->getMetadata(LLVMContext::MD_tbaa))
+ if (!Aliases(L, M))
+ return NoModRef;
+
+ return AliasAnalysis::getModRefInfo(CS, Loc);
+}
+
+AliasAnalysis::ModRefResult
+TypeBasedAliasAnalysis::getModRefInfo(ImmutableCallSite CS1,
+ ImmutableCallSite CS2) {
+ if (!EnableTBAA)
+ return AliasAnalysis::getModRefInfo(CS1, CS2);
+
+ if (const MDNode *M1 =
+ CS1.getInstruction()->getMetadata(LLVMContext::MD_tbaa))
+ if (const MDNode *M2 =
+ CS2.getInstruction()->getMetadata(LLVMContext::MD_tbaa))
+ if (!Aliases(M1, M2))
+ return NoModRef;
+
+ return AliasAnalysis::getModRefInfo(CS1, CS2);
+}
diff --git a/src/LLVM/lib/Analysis/ValueTracking.cpp b/src/LLVM/lib/Analysis/ValueTracking.cpp
index f636694..4d94f61 100644
--- a/src/LLVM/lib/Analysis/ValueTracking.cpp
+++ b/src/LLVM/lib/Analysis/ValueTracking.cpp
@@ -13,6 +13,7 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Constants.h"
#include "llvm/Instructions.h"
#include "llvm/GlobalVariable.h"
@@ -23,9 +24,22 @@
#include "llvm/Target/TargetData.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
+#include "llvm/Support/PatternMatch.h"
#include "llvm/ADT/SmallPtrSet.h"
#include <cstring>
using namespace llvm;
+using namespace llvm::PatternMatch;
+
+const unsigned MaxDepth = 6;
+
+/// getBitWidth - Returns the bitwidth of the given scalar or pointer type (if
+/// unknown returns 0). For vector types, returns the element type's bitwidth.
+static unsigned getBitWidth(Type *Ty, const TargetData *TD) {
+ if (unsigned BitWidth = Ty->getScalarSizeInBits())
+ return BitWidth;
+ assert(isa<PointerType>(Ty) && "Expected a pointer type!");
+ return TD ? TD->getPointerSizeInBits() : 0;
+}
/// ComputeMaskedBits - Determine which of the bits specified in Mask are
/// known to be either zero or one and return them in the KnownZero/KnownOne
@@ -46,7 +60,6 @@
void llvm::ComputeMaskedBits(Value *V, const APInt &Mask,
APInt &KnownZero, APInt &KnownOne,
const TargetData *TD, unsigned Depth) {
- const unsigned MaxDepth = 6;
assert(V && "No Value?");
assert(Depth <= MaxDepth && "Limit Search Depth");
unsigned BitWidth = Mask.getBitWidth();
@@ -69,14 +82,14 @@
// Null and aggregate-zero are all-zeros.
if (isa<ConstantPointerNull>(V) ||
isa<ConstantAggregateZero>(V)) {
- KnownOne.clear();
+ KnownOne.clearAllBits();
KnownZero = Mask;
return;
}
// Handle a constant vector by taking the intersection of the known bits of
// each element.
if (ConstantVector *CV = dyn_cast<ConstantVector>(V)) {
- KnownZero.set(); KnownOne.set();
+ KnownZero.setAllBits(); KnownOne.setAllBits();
for (unsigned i = 0, e = CV->getNumOperands(); i != e; ++i) {
APInt KnownZero2(BitWidth, 0), KnownOne2(BitWidth, 0);
ComputeMaskedBits(CV->getOperand(i), Mask, KnownZero2, KnownOne2,
@@ -90,7 +103,7 @@
if (GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
unsigned Align = GV->getAlignment();
if (Align == 0 && TD && GV->getType()->getElementType()->isSized()) {
- const Type *ObjectType = GV->getType()->getElementType();
+ Type *ObjectType = GV->getType()->getElementType();
// If the object is defined in the current Module, we'll be giving
// it the preferred alignment. Otherwise, we have to assume that it
// may only have the minimum ABI alignment.
@@ -103,23 +116,33 @@
KnownZero = Mask & APInt::getLowBitsSet(BitWidth,
CountTrailingZeros_32(Align));
else
- KnownZero.clear();
- KnownOne.clear();
+ KnownZero.clearAllBits();
+ KnownOne.clearAllBits();
return;
}
// A weak GlobalAlias is totally unknown. A non-weak GlobalAlias has
// the bits of its aliasee.
if (GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
if (GA->mayBeOverridden()) {
- KnownZero.clear(); KnownOne.clear();
+ KnownZero.clearAllBits(); KnownOne.clearAllBits();
} else {
ComputeMaskedBits(GA->getAliasee(), Mask, KnownZero, KnownOne,
TD, Depth+1);
}
return;
}
+
+ if (Argument *A = dyn_cast<Argument>(V)) {
+ // Get alignment information off byval arguments if specified in the IR.
+ if (A->hasByValAttr())
+ if (unsigned Align = A->getParamAlignment())
+ KnownZero = Mask & APInt::getLowBitsSet(BitWidth,
+ CountTrailingZeros_32(Align));
+ return;
+ }
- KnownZero.clear(); KnownOne.clear(); // Start out not knowing anything.
+ // Start out not knowing anything.
+ KnownZero.clearAllBits(); KnownOne.clearAllBits();
if (Depth == MaxDepth || Mask == 0)
return; // Limit search depth.
@@ -185,7 +208,7 @@
// Also compute a conserative estimate for high known-0 bits.
// More trickiness is possible, but this is sufficient for the
// interesting case of alignment computation.
- KnownOne.clear();
+ KnownOne.clearAllBits();
unsigned TrailZ = KnownZero.countTrailingOnes() +
KnownZero2.countTrailingOnes();
unsigned LeadZ = std::max(KnownZero.countLeadingOnes() +
@@ -208,8 +231,8 @@
AllOnes, KnownZero2, KnownOne2, TD, Depth+1);
unsigned LeadZ = KnownZero2.countLeadingOnes();
- KnownOne2.clear();
- KnownZero2.clear();
+ KnownOne2.clearAllBits();
+ KnownZero2.clearAllBits();
ComputeMaskedBits(I->getOperand(1),
AllOnes, KnownZero2, KnownOne2, TD, Depth+1);
unsigned RHSUnknownLeadingOnes = KnownOne2.countLeadingZeros();
@@ -245,7 +268,7 @@
// FALL THROUGH and handle them the same as zext/trunc.
case Instruction::ZExt:
case Instruction::Trunc: {
- const Type *SrcTy = I->getOperand(0)->getType();
+ Type *SrcTy = I->getOperand(0)->getType();
unsigned SrcBitWidth;
// Note that we handle pointer operands here because of inttoptr/ptrtoint
@@ -255,21 +278,20 @@
else
SrcBitWidth = SrcTy->getScalarSizeInBits();
- APInt MaskIn(Mask);
- MaskIn.zextOrTrunc(SrcBitWidth);
- KnownZero.zextOrTrunc(SrcBitWidth);
- KnownOne.zextOrTrunc(SrcBitWidth);
+ APInt MaskIn = Mask.zextOrTrunc(SrcBitWidth);
+ KnownZero = KnownZero.zextOrTrunc(SrcBitWidth);
+ KnownOne = KnownOne.zextOrTrunc(SrcBitWidth);
ComputeMaskedBits(I->getOperand(0), MaskIn, KnownZero, KnownOne, TD,
Depth+1);
- KnownZero.zextOrTrunc(BitWidth);
- KnownOne.zextOrTrunc(BitWidth);
+ KnownZero = KnownZero.zextOrTrunc(BitWidth);
+ KnownOne = KnownOne.zextOrTrunc(BitWidth);
// Any top bits are known to be zero.
if (BitWidth > SrcBitWidth)
KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - SrcBitWidth);
return;
}
case Instruction::BitCast: {
- const Type *SrcTy = I->getOperand(0)->getType();
+ Type *SrcTy = I->getOperand(0)->getType();
if ((SrcTy->isIntegerTy() || SrcTy->isPointerTy()) &&
// TODO: For now, not handling conversions like:
// (bitcast i64 %x to <2 x i32>)
@@ -284,15 +306,14 @@
// Compute the bits in the result that are not present in the input.
unsigned SrcBitWidth = I->getOperand(0)->getType()->getScalarSizeInBits();
- APInt MaskIn(Mask);
- MaskIn.trunc(SrcBitWidth);
- KnownZero.trunc(SrcBitWidth);
- KnownOne.trunc(SrcBitWidth);
+ APInt MaskIn = Mask.trunc(SrcBitWidth);
+ KnownZero = KnownZero.trunc(SrcBitWidth);
+ KnownOne = KnownOne.trunc(SrcBitWidth);
ComputeMaskedBits(I->getOperand(0), MaskIn, KnownZero, KnownOne, TD,
Depth+1);
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
- KnownZero.zext(BitWidth);
- KnownOne.zext(BitWidth);
+ KnownZero = KnownZero.zext(BitWidth);
+ KnownOne = KnownOne.zext(BitWidth);
// If the sign bit of the input is known set or clear, then we know the
// top bits of the result.
@@ -338,7 +359,7 @@
// (ashr X, C1) & C2 == 0 iff (-1 >> C1) & C2 == 0
if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
// Compute the new bits that are at the top now.
- uint64_t ShiftAmt = SA->getLimitedValue(BitWidth);
+ uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
// Signed shift right.
APInt Mask2(Mask.shl(ShiftAmt));
@@ -418,6 +439,29 @@
KnownZero |= LHSKnownZero & Mask;
KnownOne |= LHSKnownOne & Mask;
}
+
+ // Are we still trying to solve for the sign bit?
+ if (Mask.isNegative() && !KnownZero.isNegative() && !KnownOne.isNegative()){
+ OverflowingBinaryOperator *OBO = cast<OverflowingBinaryOperator>(I);
+ if (OBO->hasNoSignedWrap()) {
+ if (I->getOpcode() == Instruction::Add) {
+ // Adding two positive numbers can't wrap into negative
+ if (LHSKnownZero.isNegative() && KnownZero2.isNegative())
+ KnownZero |= APInt::getSignBit(BitWidth);
+ // and adding two negative numbers can't wrap into positive.
+ else if (LHSKnownOne.isNegative() && KnownOne2.isNegative())
+ KnownOne |= APInt::getSignBit(BitWidth);
+ } else {
+ // Subtracting a negative number from a positive one can't wrap
+ if (LHSKnownZero.isNegative() && KnownOne2.isNegative())
+ KnownZero |= APInt::getSignBit(BitWidth);
+ // neither can subtracting a positive number from a negative one.
+ else if (LHSKnownOne.isNegative() && KnownZero2.isNegative())
+ KnownOne |= APInt::getSignBit(BitWidth);
+ }
+ }
+ }
+
return;
}
case Instruction::SRem:
@@ -449,6 +493,19 @@
assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
}
}
+
+ // The sign bit is the LHS's sign bit, except when the result of the
+ // remainder is zero.
+ if (Mask.isNegative() && KnownZero.isNonNegative()) {
+ APInt Mask2 = APInt::getSignBit(BitWidth);
+ APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
+ ComputeMaskedBits(I->getOperand(0), Mask2, LHSKnownZero, LHSKnownOne, TD,
+ Depth+1);
+ // If it's known zero, our sign bit is also zero.
+ if (LHSKnownZero.isNegative())
+ KnownZero |= LHSKnownZero;
+ }
+
break;
case Instruction::URem: {
if (ConstantInt *Rem = dyn_cast<ConstantInt>(I->getOperand(1))) {
@@ -474,7 +531,7 @@
unsigned Leaders = std::max(KnownZero.countLeadingOnes(),
KnownZero2.countLeadingOnes());
- KnownOne.clear();
+ KnownOne.clearAllBits();
KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & Mask;
break;
}
@@ -502,7 +559,7 @@
gep_type_iterator GTI = gep_type_begin(I);
for (unsigned i = 1, e = I->getNumOperands(); i != e; ++i, ++GTI) {
Value *Index = I->getOperand(i);
- if (const StructType *STy = dyn_cast<StructType>(*GTI)) {
+ if (StructType *STy = dyn_cast<StructType>(*GTI)) {
// Handle struct member offset arithmetic.
if (!TD) return;
const StructLayout *SL = TD->getStructLayout(STy);
@@ -512,7 +569,7 @@
CountTrailingZeros_64(Offset));
} else {
// Handle array index arithmetic.
- const Type *IndexedTy = GTI.getIndexedType();
+ Type *IndexedTy = GTI.getIndexedType();
if (!IndexedTy->isSized()) return;
unsigned GEPOpiBits = Index->getType()->getScalarSizeInBits();
uint64_t TypeSize = TD ? TD->getTypeAllocSize(IndexedTy) : 1;
@@ -579,9 +636,17 @@
}
}
+ // Unreachable blocks may have zero-operand PHI nodes.
+ if (P->getNumIncomingValues() == 0)
+ return;
+
// Otherwise take the unions of the known bit sets of the operands,
// taking conservative care to avoid excessive recursion.
if (Depth < MaxDepth - 1 && !KnownZero && !KnownOne) {
+ // Skip if every incoming value references to ourself.
+ if (P->hasConstantValue() == P)
+ break;
+
KnownZero = APInt::getAllOnesValue(BitWidth);
KnownOne = APInt::getAllOnesValue(BitWidth);
for (unsigned i = 0, e = P->getNumIncomingValues(); i != e; ++i) {
@@ -615,12 +680,192 @@
KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - LowBits);
break;
}
+ case Intrinsic::x86_sse42_crc32_64_8:
+ case Intrinsic::x86_sse42_crc32_64_64:
+ KnownZero = APInt::getHighBitsSet(64, 32);
+ break;
}
}
break;
}
}
+/// ComputeSignBit - Determine whether the sign bit is known to be zero or
+/// one. Convenience wrapper around ComputeMaskedBits.
+void llvm::ComputeSignBit(Value *V, bool &KnownZero, bool &KnownOne,
+ const TargetData *TD, unsigned Depth) {
+ unsigned BitWidth = getBitWidth(V->getType(), TD);
+ if (!BitWidth) {
+ KnownZero = false;
+ KnownOne = false;
+ return;
+ }
+ APInt ZeroBits(BitWidth, 0);
+ APInt OneBits(BitWidth, 0);
+ ComputeMaskedBits(V, APInt::getSignBit(BitWidth), ZeroBits, OneBits, TD,
+ Depth);
+ KnownOne = OneBits[BitWidth - 1];
+ KnownZero = ZeroBits[BitWidth - 1];
+}
+
+/// isPowerOfTwo - Return true if the given value is known to have exactly one
+/// bit set when defined. For vectors return true if every element is known to
+/// be a power of two when defined. Supports values with integer or pointer
+/// types and vectors of integers.
+bool llvm::isPowerOfTwo(Value *V, const TargetData *TD, unsigned Depth) {
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(V))
+ return CI->getValue().isPowerOf2();
+ // TODO: Handle vector constants.
+
+ // 1 << X is clearly a power of two if the one is not shifted off the end. If
+ // it is shifted off the end then the result is undefined.
+ if (match(V, m_Shl(m_One(), m_Value())))
+ return true;
+
+ // (signbit) >>l X is clearly a power of two if the one is not shifted off the
+ // bottom. If it is shifted off the bottom then the result is undefined.
+ if (match(V, m_LShr(m_SignBit(), m_Value())))
+ return true;
+
+ // The remaining tests are all recursive, so bail out if we hit the limit.
+ if (Depth++ == MaxDepth)
+ return false;
+
+ if (ZExtInst *ZI = dyn_cast<ZExtInst>(V))
+ return isPowerOfTwo(ZI->getOperand(0), TD, Depth);
+
+ if (SelectInst *SI = dyn_cast<SelectInst>(V))
+ return isPowerOfTwo(SI->getTrueValue(), TD, Depth) &&
+ isPowerOfTwo(SI->getFalseValue(), TD, Depth);
+
+ // An exact divide or right shift can only shift off zero bits, so the result
+ // is a power of two only if the first operand is a power of two and not
+ // copying a sign bit (sdiv int_min, 2).
+ if (match(V, m_LShr(m_Value(), m_Value())) ||
+ match(V, m_UDiv(m_Value(), m_Value()))) {
+ PossiblyExactOperator *PEO = cast<PossiblyExactOperator>(V);
+ if (PEO->isExact())
+ return isPowerOfTwo(PEO->getOperand(0), TD, Depth);
+ }
+
+ return false;
+}
+
+/// isKnownNonZero - Return true if the given value is known to be non-zero
+/// when defined. For vectors return true if every element is known to be
+/// non-zero when defined. Supports values with integer or pointer type and
+/// vectors of integers.
+bool llvm::isKnownNonZero(Value *V, const TargetData *TD, unsigned Depth) {
+ if (Constant *C = dyn_cast<Constant>(V)) {
+ if (C->isNullValue())
+ return false;
+ if (isa<ConstantInt>(C))
+ // Must be non-zero due to null test above.
+ return true;
+ // TODO: Handle vectors
+ return false;
+ }
+
+ // The remaining tests are all recursive, so bail out if we hit the limit.
+ if (Depth++ == MaxDepth)
+ return false;
+
+ unsigned BitWidth = getBitWidth(V->getType(), TD);
+
+ // X | Y != 0 if X != 0 or Y != 0.
+ Value *X = 0, *Y = 0;
+ if (match(V, m_Or(m_Value(X), m_Value(Y))))
+ return isKnownNonZero(X, TD, Depth) || isKnownNonZero(Y, TD, Depth);
+
+ // ext X != 0 if X != 0.
+ if (isa<SExtInst>(V) || isa<ZExtInst>(V))
+ return isKnownNonZero(cast<Instruction>(V)->getOperand(0), TD, Depth);
+
+ // shl X, Y != 0 if X is odd. Note that the value of the shift is undefined
+ // if the lowest bit is shifted off the end.
+ if (BitWidth && match(V, m_Shl(m_Value(X), m_Value(Y)))) {
+ // shl nuw can't remove any non-zero bits.
+ BinaryOperator *BO = cast<BinaryOperator>(V);
+ if (BO->hasNoUnsignedWrap())
+ return isKnownNonZero(X, TD, Depth);
+
+ APInt KnownZero(BitWidth, 0);
+ APInt KnownOne(BitWidth, 0);
+ ComputeMaskedBits(X, APInt(BitWidth, 1), KnownZero, KnownOne, TD, Depth);
+ if (KnownOne[0])
+ return true;
+ }
+ // shr X, Y != 0 if X is negative. Note that the value of the shift is not
+ // defined if the sign bit is shifted off the end.
+ else if (match(V, m_Shr(m_Value(X), m_Value(Y)))) {
+ // shr exact can only shift out zero bits.
+ BinaryOperator *BO = cast<BinaryOperator>(V);
+ if (BO->isExact())
+ return isKnownNonZero(X, TD, Depth);
+
+ bool XKnownNonNegative, XKnownNegative;
+ ComputeSignBit(X, XKnownNonNegative, XKnownNegative, TD, Depth);
+ if (XKnownNegative)
+ return true;
+ }
+ // div exact can only produce a zero if the dividend is zero.
+ else if (match(V, m_IDiv(m_Value(X), m_Value()))) {
+ BinaryOperator *BO = cast<BinaryOperator>(V);
+ if (BO->isExact())
+ return isKnownNonZero(X, TD, Depth);
+ }
+ // X + Y.
+ else if (match(V, m_Add(m_Value(X), m_Value(Y)))) {
+ bool XKnownNonNegative, XKnownNegative;
+ bool YKnownNonNegative, YKnownNegative;
+ ComputeSignBit(X, XKnownNonNegative, XKnownNegative, TD, Depth);
+ ComputeSignBit(Y, YKnownNonNegative, YKnownNegative, TD, Depth);
+
+ // If X and Y are both non-negative (as signed values) then their sum is not
+ // zero unless both X and Y are zero.
+ if (XKnownNonNegative && YKnownNonNegative)
+ if (isKnownNonZero(X, TD, Depth) || isKnownNonZero(Y, TD, Depth))
+ return true;
+
+ // If X and Y are both negative (as signed values) then their sum is not
+ // zero unless both X and Y equal INT_MIN.
+ if (BitWidth && XKnownNegative && YKnownNegative) {
+ APInt KnownZero(BitWidth, 0);
+ APInt KnownOne(BitWidth, 0);
+ APInt Mask = APInt::getSignedMaxValue(BitWidth);
+ // The sign bit of X is set. If some other bit is set then X is not equal
+ // to INT_MIN.
+ ComputeMaskedBits(X, Mask, KnownZero, KnownOne, TD, Depth);
+ if ((KnownOne & Mask) != 0)
+ return true;
+ // The sign bit of Y is set. If some other bit is set then Y is not equal
+ // to INT_MIN.
+ ComputeMaskedBits(Y, Mask, KnownZero, KnownOne, TD, Depth);
+ if ((KnownOne & Mask) != 0)
+ return true;
+ }
+
+ // The sum of a non-negative number and a power of two is not zero.
+ if (XKnownNonNegative && isPowerOfTwo(Y, TD, Depth))
+ return true;
+ if (YKnownNonNegative && isPowerOfTwo(X, TD, Depth))
+ return true;
+ }
+ // (C ? X : Y) != 0 if X != 0 and Y != 0.
+ else if (SelectInst *SI = dyn_cast<SelectInst>(V)) {
+ if (isKnownNonZero(SI->getTrueValue(), TD, Depth) &&
+ isKnownNonZero(SI->getFalseValue(), TD, Depth))
+ return true;
+ }
+
+ if (!BitWidth) return false;
+ APInt KnownZero(BitWidth, 0);
+ APInt KnownOne(BitWidth, 0);
+ ComputeMaskedBits(V, APInt::getAllOnesValue(BitWidth), KnownZero, KnownOne,
+ TD, Depth);
+ return KnownOne != 0;
+}
+
/// MaskedValueIsZero - Return true if 'V & Mask' is known to be zero. We use
/// this predicate to simplify operations downstream. Mask is known to be zero
/// for bits that V cannot have.
@@ -653,7 +898,7 @@
assert((TD || V->getType()->isIntOrIntVectorTy()) &&
"ComputeNumSignBits requires a TargetData object to operate "
"on non-integer values!");
- const Type *Ty = V->getType();
+ Type *Ty = V->getType();
unsigned TyBits = TD ? TD->getTypeSizeInBits(V->getType()->getScalarType()) :
Ty->getScalarSizeInBits();
unsigned Tmp, Tmp2;
@@ -679,6 +924,13 @@
Tmp += C->getZExtValue();
if (Tmp > TyBits) Tmp = TyBits;
}
+ // vector ashr X, <C, C, C, C> -> adds C sign bits
+ if (ConstantVector *C = dyn_cast<ConstantVector>(U->getOperand(1))) {
+ if (ConstantInt *CI = dyn_cast_or_null<ConstantInt>(C->getSplatValue())) {
+ Tmp += CI->getZExtValue();
+ if (Tmp > TyBits) Tmp = TyBits;
+ }
+ }
return Tmp;
case Instruction::Shl:
if (ConstantInt *C = dyn_cast<ConstantInt>(U->getOperand(1))) {
@@ -826,7 +1078,7 @@
assert(Depth <= MaxDepth && "Limit Search Depth");
assert(V->getType()->isIntegerTy() && "Not integer or pointer type!");
- const Type *T = V->getType();
+ Type *T = V->getType();
ConstantInt *CI = dyn_cast<ConstantInt>(V);
@@ -875,24 +1127,26 @@
// Turn Op0 << Op1 into Op0 * 2^Op1
APInt Op1Int = Op1CI->getValue();
uint64_t BitToSet = Op1Int.getLimitedValue(Op1Int.getBitWidth() - 1);
- Op1 = ConstantInt::get(V->getContext(),
- APInt(Op1Int.getBitWidth(), 0).set(BitToSet));
+ APInt API(Op1Int.getBitWidth(), 0);
+ API.setBit(BitToSet);
+ Op1 = ConstantInt::get(V->getContext(), API);
}
Value *Mul0 = NULL;
- Value *Mul1 = NULL;
- bool M0 = ComputeMultiple(Op0, Base, Mul0,
- LookThroughSExt, Depth+1);
- bool M1 = ComputeMultiple(Op1, Base, Mul1,
- LookThroughSExt, Depth+1);
-
- if (M0) {
- if (isa<Constant>(Op1) && isa<Constant>(Mul0)) {
- // V == Base * (Mul0 * Op1), so return (Mul0 * Op1)
- Multiple = ConstantExpr::getMul(cast<Constant>(Mul0),
- cast<Constant>(Op1));
- return true;
- }
+ if (ComputeMultiple(Op0, Base, Mul0, LookThroughSExt, Depth+1)) {
+ if (Constant *Op1C = dyn_cast<Constant>(Op1))
+ if (Constant *MulC = dyn_cast<Constant>(Mul0)) {
+ if (Op1C->getType()->getPrimitiveSizeInBits() <
+ MulC->getType()->getPrimitiveSizeInBits())
+ Op1C = ConstantExpr::getZExt(Op1C, MulC->getType());
+ if (Op1C->getType()->getPrimitiveSizeInBits() >
+ MulC->getType()->getPrimitiveSizeInBits())
+ MulC = ConstantExpr::getZExt(MulC, Op1C->getType());
+
+ // V == Base * (Mul0 * Op1), so return (Mul0 * Op1)
+ Multiple = ConstantExpr::getMul(MulC, Op1C);
+ return true;
+ }
if (ConstantInt *Mul0CI = dyn_cast<ConstantInt>(Mul0))
if (Mul0CI->getValue() == 1) {
@@ -902,13 +1156,21 @@
}
}
- if (M1) {
- if (isa<Constant>(Op0) && isa<Constant>(Mul1)) {
- // V == Base * (Mul1 * Op0), so return (Mul1 * Op0)
- Multiple = ConstantExpr::getMul(cast<Constant>(Mul1),
- cast<Constant>(Op0));
- return true;
- }
+ Value *Mul1 = NULL;
+ if (ComputeMultiple(Op1, Base, Mul1, LookThroughSExt, Depth+1)) {
+ if (Constant *Op0C = dyn_cast<Constant>(Op0))
+ if (Constant *MulC = dyn_cast<Constant>(Mul1)) {
+ if (Op0C->getType()->getPrimitiveSizeInBits() <
+ MulC->getType()->getPrimitiveSizeInBits())
+ Op0C = ConstantExpr::getZExt(Op0C, MulC->getType());
+ if (Op0C->getType()->getPrimitiveSizeInBits() >
+ MulC->getType()->getPrimitiveSizeInBits())
+ MulC = ConstantExpr::getZExt(MulC, Op0C->getType());
+
+ // V == Base * (Mul1 * Op0), so return (Mul1 * Op0)
+ Multiple = ConstantExpr::getMul(MulC, Op0C);
+ return true;
+ }
if (ConstantInt *Mul1CI = dyn_cast<ConstantInt>(Mul1))
if (Mul1CI->getValue() == 1) {
@@ -973,192 +1235,77 @@
return false;
}
+/// isBytewiseValue - If the specified value can be set by repeating the same
+/// byte in memory, return the i8 value that it is represented with. This is
+/// true for all i8 values obviously, but is also true for i32 0, i32 -1,
+/// i16 0xF0F0, double 0.0 etc. If the value can't be handled with a repeated
+/// byte store (e.g. i16 0x1234), return null.
+Value *llvm::isBytewiseValue(Value *V) {
+ // All byte-wide stores are splatable, even of arbitrary variables.
+ if (V->getType()->isIntegerTy(8)) return V;
-/// GetLinearExpression - Analyze the specified value as a linear expression:
-/// "A*V + B", where A and B are constant integers. Return the scale and offset
-/// values as APInts and return V as a Value*. The incoming Value is known to
-/// have IntegerType. Note that this looks through extends, so the high bits
-/// may not be represented in the result.
-static Value *GetLinearExpression(Value *V, APInt &Scale, APInt &Offset,
- const TargetData *TD, unsigned Depth) {
- assert(V->getType()->isIntegerTy() && "Not an integer value");
-
- // Limit our recursion depth.
- if (Depth == 6) {
- Scale = 1;
- Offset = 0;
- return V;
+ // Handle 'null' ConstantArrayZero etc.
+ if (Constant *C = dyn_cast<Constant>(V))
+ if (C->isNullValue())
+ return Constant::getNullValue(Type::getInt8Ty(V->getContext()));
+
+ // Constant float and double values can be handled as integer values if the
+ // corresponding integer value is "byteable". An important case is 0.0.
+ if (ConstantFP *CFP = dyn_cast<ConstantFP>(V)) {
+ if (CFP->getType()->isFloatTy())
+ V = ConstantExpr::getBitCast(CFP, Type::getInt32Ty(V->getContext()));
+ if (CFP->getType()->isDoubleTy())
+ V = ConstantExpr::getBitCast(CFP, Type::getInt64Ty(V->getContext()));
+ // Don't handle long double formats, which have strange constraints.
}
- if (BinaryOperator *BOp = dyn_cast<BinaryOperator>(V)) {
- if (ConstantInt *RHSC = dyn_cast<ConstantInt>(BOp->getOperand(1))) {
- switch (BOp->getOpcode()) {
- default: break;
- case Instruction::Or:
- // X|C == X+C if all the bits in C are unset in X. Otherwise we can't
- // analyze it.
- if (!MaskedValueIsZero(BOp->getOperand(0), RHSC->getValue(), TD))
- break;
- // FALL THROUGH.
- case Instruction::Add:
- V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, TD, Depth+1);
- Offset += RHSC->getValue();
- return V;
- case Instruction::Mul:
- V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, TD, Depth+1);
- Offset *= RHSC->getValue();
- Scale *= RHSC->getValue();
- return V;
- case Instruction::Shl:
- V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, TD, Depth+1);
- Offset <<= RHSC->getValue().getLimitedValue();
- Scale <<= RHSC->getValue().getLimitedValue();
- return V;
- }
- }
- }
-
- // Since clients don't care about the high bits of the value, just scales and
- // offsets, we can look through extensions.
- if (isa<SExtInst>(V) || isa<ZExtInst>(V)) {
- Value *CastOp = cast<CastInst>(V)->getOperand(0);
- unsigned OldWidth = Scale.getBitWidth();
- unsigned SmallWidth = CastOp->getType()->getPrimitiveSizeInBits();
- Scale.trunc(SmallWidth);
- Offset.trunc(SmallWidth);
- Value *Result = GetLinearExpression(CastOp, Scale, Offset, TD, Depth+1);
- Scale.zext(OldWidth);
- Offset.zext(OldWidth);
- return Result;
- }
-
- Scale = 1;
- Offset = 0;
- return V;
-}
-
-/// DecomposeGEPExpression - If V is a symbolic pointer expression, decompose it
-/// into a base pointer with a constant offset and a number of scaled symbolic
-/// offsets.
-///
-/// The scaled symbolic offsets (represented by pairs of a Value* and a scale in
-/// the VarIndices vector) are Value*'s that are known to be scaled by the
-/// specified amount, but which may have other unrepresented high bits. As such,
-/// the gep cannot necessarily be reconstructed from its decomposed form.
-///
-/// When TargetData is around, this function is capable of analyzing everything
-/// that Value::getUnderlyingObject() can look through. When not, it just looks
-/// through pointer casts.
-///
-const Value *llvm::DecomposeGEPExpression(const Value *V, int64_t &BaseOffs,
- SmallVectorImpl<std::pair<const Value*, int64_t> > &VarIndices,
- const TargetData *TD) {
- // Limit recursion depth to limit compile time in crazy cases.
- unsigned MaxLookup = 6;
-
- BaseOffs = 0;
- do {
- // See if this is a bitcast or GEP.
- const Operator *Op = dyn_cast<Operator>(V);
- if (Op == 0) {
- // The only non-operator case we can handle are GlobalAliases.
- if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
- if (!GA->mayBeOverridden()) {
- V = GA->getAliasee();
- continue;
- }
- }
- return V;
- }
-
- if (Op->getOpcode() == Instruction::BitCast) {
- V = Op->getOperand(0);
- continue;
- }
-
- const GEPOperator *GEPOp = dyn_cast<GEPOperator>(Op);
- if (GEPOp == 0)
- return V;
-
- // Don't attempt to analyze GEPs over unsized objects.
- if (!cast<PointerType>(GEPOp->getOperand(0)->getType())
- ->getElementType()->isSized())
- return V;
-
- // If we are lacking TargetData information, we can't compute the offets of
- // elements computed by GEPs. However, we can handle bitcast equivalent
- // GEPs.
- if (!TD) {
- if (!GEPOp->hasAllZeroIndices())
- return V;
- V = GEPOp->getOperand(0);
- continue;
- }
-
- // Walk the indices of the GEP, accumulating them into BaseOff/VarIndices.
- gep_type_iterator GTI = gep_type_begin(GEPOp);
- for (User::const_op_iterator I = GEPOp->op_begin()+1,
- E = GEPOp->op_end(); I != E; ++I) {
- Value *Index = *I;
- // Compute the (potentially symbolic) offset in bytes for this index.
- if (const StructType *STy = dyn_cast<StructType>(*GTI++)) {
- // For a struct, add the member offset.
- unsigned FieldNo = cast<ConstantInt>(Index)->getZExtValue();
- if (FieldNo == 0) continue;
+ // We can handle constant integers that are power of two in size and a
+ // multiple of 8 bits.
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
+ unsigned Width = CI->getBitWidth();
+ if (isPowerOf2_32(Width) && Width > 8) {
+ // We can handle this value if the recursive binary decomposition is the
+ // same at all levels.
+ APInt Val = CI->getValue();
+ APInt Val2;
+ while (Val.getBitWidth() != 8) {
+ unsigned NextWidth = Val.getBitWidth()/2;
+ Val2 = Val.lshr(NextWidth);
+ Val2 = Val2.trunc(Val.getBitWidth()/2);
+ Val = Val.trunc(Val.getBitWidth()/2);
- BaseOffs += TD->getStructLayout(STy)->getElementOffset(FieldNo);
- continue;
+ // If the top/bottom halves aren't the same, reject it.
+ if (Val != Val2)
+ return 0;
}
-
- // For an array/pointer, add the element offset, explicitly scaled.
- if (ConstantInt *CIdx = dyn_cast<ConstantInt>(Index)) {
- if (CIdx->isZero()) continue;
- BaseOffs += TD->getTypeAllocSize(*GTI)*CIdx->getSExtValue();
- continue;
- }
-
- uint64_t Scale = TD->getTypeAllocSize(*GTI);
-
- // Use GetLinearExpression to decompose the index into a C1*V+C2 form.
- unsigned Width = cast<IntegerType>(Index->getType())->getBitWidth();
- APInt IndexScale(Width, 0), IndexOffset(Width, 0);
- Index = GetLinearExpression(Index, IndexScale, IndexOffset, TD, 0);
-
- // The GEP index scale ("Scale") scales C1*V+C2, yielding (C1*V+C2)*Scale.
- // This gives us an aggregate computation of (C1*Scale)*V + C2*Scale.
- BaseOffs += IndexOffset.getZExtValue()*Scale;
- Scale *= IndexScale.getZExtValue();
-
-
- // If we already had an occurrance of this index variable, merge this
- // scale into it. For example, we want to handle:
- // A[x][x] -> x*16 + x*4 -> x*20
- // This also ensures that 'x' only appears in the index list once.
- for (unsigned i = 0, e = VarIndices.size(); i != e; ++i) {
- if (VarIndices[i].first == Index) {
- Scale += VarIndices[i].second;
- VarIndices.erase(VarIndices.begin()+i);
- break;
- }
- }
-
- // Make sure that we have a scale that makes sense for this target's
- // pointer size.
- if (unsigned ShiftBits = 64-TD->getPointerSizeInBits()) {
- Scale <<= ShiftBits;
- Scale >>= ShiftBits;
- }
-
- if (Scale)
- VarIndices.push_back(std::make_pair(Index, Scale));
+ return ConstantInt::get(V->getContext(), Val);
}
-
- // Analyze the base pointer next.
- V = GEPOp->getOperand(0);
- } while (--MaxLookup);
+ }
- // If the chain of expressions is too deep, just return early.
- return V;
+ // A ConstantArray is splatable if all its members are equal and also
+ // splatable.
+ if (ConstantArray *CA = dyn_cast<ConstantArray>(V)) {
+ if (CA->getNumOperands() == 0)
+ return 0;
+
+ Value *Val = isBytewiseValue(CA->getOperand(0));
+ if (!Val)
+ return 0;
+
+ for (unsigned I = 1, E = CA->getNumOperands(); I != E; ++I)
+ if (CA->getOperand(I-1) != CA->getOperand(I))
+ return 0;
+
+ return Val;
+ }
+
+ // Conceptually, we could handle things like:
+ // %a = zext i8 %X to i16
+ // %b = shl i16 %a, 8
+ // %c = or i16 %a, %b
+ // but until there is an example that actually needs this, it doesn't seem
+ // worth worrying about.
+ return 0;
}
@@ -1168,11 +1315,11 @@
// indices from Idxs that should be left out when inserting into the resulting
// struct. To is the result struct built so far, new insertvalue instructions
// build on that.
-static Value *BuildSubAggregate(Value *From, Value* To, const Type *IndexedType,
+static Value *BuildSubAggregate(Value *From, Value* To, Type *IndexedType,
SmallVector<unsigned, 10> &Idxs,
unsigned IdxSkip,
Instruction *InsertBefore) {
- const llvm::StructType *STy = llvm::dyn_cast<llvm::StructType>(IndexedType);
+ llvm::StructType *STy = llvm::dyn_cast<llvm::StructType>(IndexedType);
if (STy) {
// Save the original To argument so we can modify it
Value *OrigTo = To;
@@ -1195,7 +1342,7 @@
break;
}
}
- // If we succesfully found a value for each of our subaggregates
+ // If we successfully found a value for each of our subaggregates
if (To)
return To;
}
@@ -1205,14 +1352,14 @@
// we might be able to find the complete struct somewhere.
// Find the value that is at that particular spot
- Value *V = FindInsertedValue(From, Idxs.begin(), Idxs.end());
+ Value *V = FindInsertedValue(From, Idxs);
if (!V)
return NULL;
// Insert the value in the new (sub) aggregrate
- return llvm::InsertValueInst::Create(To, V, Idxs.begin() + IdxSkip,
- Idxs.end(), InsertBefore);
+ return llvm::InsertValueInst::Create(To, V, makeArrayRef(Idxs).slice(IdxSkip),
+ "tmp", InsertBefore);
}
// This helper takes a nested struct and extracts a part of it (which is again a
@@ -1227,15 +1374,13 @@
// insertvalue instruction somewhere).
//
// All inserted insertvalue instructions are inserted before InsertBefore
-static Value *BuildSubAggregate(Value *From, const unsigned *idx_begin,
- const unsigned *idx_end,
+static Value *BuildSubAggregate(Value *From, ArrayRef<unsigned> idx_range,
Instruction *InsertBefore) {
assert(InsertBefore && "Must have someplace to insert!");
- const Type *IndexedType = ExtractValueInst::getIndexedType(From->getType(),
- idx_begin,
- idx_end);
+ Type *IndexedType = ExtractValueInst::getIndexedType(From->getType(),
+ idx_range);
Value *To = UndefValue::get(IndexedType);
- SmallVector<unsigned, 10> Idxs(idx_begin, idx_end);
+ SmallVector<unsigned, 10> Idxs(idx_range.begin(), idx_range.end());
unsigned IdxSkip = Idxs.size();
return BuildSubAggregate(From, To, IndexedType, Idxs, IdxSkip, InsertBefore);
@@ -1247,39 +1392,37 @@
///
/// If InsertBefore is not null, this function will duplicate (modified)
/// insertvalues when a part of a nested struct is extracted.
-Value *llvm::FindInsertedValue(Value *V, const unsigned *idx_begin,
- const unsigned *idx_end, Instruction *InsertBefore) {
+Value *llvm::FindInsertedValue(Value *V, ArrayRef<unsigned> idx_range,
+ Instruction *InsertBefore) {
// Nothing to index? Just return V then (this is useful at the end of our
// recursion)
- if (idx_begin == idx_end)
+ if (idx_range.empty())
return V;
// We have indices, so V should have an indexable type
assert((V->getType()->isStructTy() || V->getType()->isArrayTy())
&& "Not looking at a struct or array?");
- assert(ExtractValueInst::getIndexedType(V->getType(), idx_begin, idx_end)
+ assert(ExtractValueInst::getIndexedType(V->getType(), idx_range)
&& "Invalid indices for type?");
- const CompositeType *PTy = cast<CompositeType>(V->getType());
+ CompositeType *PTy = cast<CompositeType>(V->getType());
if (isa<UndefValue>(V))
return UndefValue::get(ExtractValueInst::getIndexedType(PTy,
- idx_begin,
- idx_end));
+ idx_range));
else if (isa<ConstantAggregateZero>(V))
return Constant::getNullValue(ExtractValueInst::getIndexedType(PTy,
- idx_begin,
- idx_end));
+ idx_range));
else if (Constant *C = dyn_cast<Constant>(V)) {
if (isa<ConstantArray>(C) || isa<ConstantStruct>(C))
// Recursively process this constant
- return FindInsertedValue(C->getOperand(*idx_begin), idx_begin + 1,
- idx_end, InsertBefore);
+ return FindInsertedValue(C->getOperand(idx_range[0]), idx_range.slice(1),
+ InsertBefore);
} else if (InsertValueInst *I = dyn_cast<InsertValueInst>(V)) {
// Loop the indices for the insertvalue instruction in parallel with the
// requested indices
- const unsigned *req_idx = idx_begin;
+ const unsigned *req_idx = idx_range.begin();
for (const unsigned *i = I->idx_begin(), *e = I->idx_end();
i != e; ++i, ++req_idx) {
- if (req_idx == idx_end) {
+ if (req_idx == idx_range.end()) {
if (InsertBefore)
// The requested index identifies a part of a nested aggregate. Handle
// this specially. For example,
@@ -1291,7 +1434,8 @@
// %C = insertvalue {i32, i32 } %A, i32 11, 1
// which allows the unused 0,0 element from the nested struct to be
// removed.
- return BuildSubAggregate(V, idx_begin, req_idx, InsertBefore);
+ return BuildSubAggregate(V, makeArrayRef(idx_range.begin(), req_idx),
+ InsertBefore);
else
// We can't handle this without inserting insertvalues
return 0;
@@ -1301,13 +1445,14 @@
// See if the (aggregrate) value inserted into has the value we are
// looking for, then.
if (*req_idx != *i)
- return FindInsertedValue(I->getAggregateOperand(), idx_begin, idx_end,
+ return FindInsertedValue(I->getAggregateOperand(), idx_range,
InsertBefore);
}
// If we end up here, the indices of the insertvalue match with those
// requested (though possibly only partially). Now we recursively look at
// the inserted value, passing any remaining indices.
- return FindInsertedValue(I->getInsertedValueOperand(), req_idx, idx_end,
+ return FindInsertedValue(I->getInsertedValueOperand(),
+ makeArrayRef(req_idx, idx_range.end()),
InsertBefore);
} else if (ExtractValueInst *I = dyn_cast<ExtractValueInst>(V)) {
// If we're extracting a value from an aggregrate that was extracted from
@@ -1315,30 +1460,67 @@
// However, we will need to chain I's indices with the requested indices.
// Calculate the number of indices required
- unsigned size = I->getNumIndices() + (idx_end - idx_begin);
+ unsigned size = I->getNumIndices() + idx_range.size();
// Allocate some space to put the new indices in
SmallVector<unsigned, 5> Idxs;
Idxs.reserve(size);
// Add indices from the extract value instruction
- for (const unsigned *i = I->idx_begin(), *e = I->idx_end();
- i != e; ++i)
- Idxs.push_back(*i);
+ Idxs.append(I->idx_begin(), I->idx_end());
// Add requested indices
- for (const unsigned *i = idx_begin, *e = idx_end; i != e; ++i)
- Idxs.push_back(*i);
+ Idxs.append(idx_range.begin(), idx_range.end());
assert(Idxs.size() == size
&& "Number of indices added not correct?");
- return FindInsertedValue(I->getAggregateOperand(), Idxs.begin(), Idxs.end(),
- InsertBefore);
+ return FindInsertedValue(I->getAggregateOperand(), Idxs, InsertBefore);
}
// Otherwise, we don't know (such as, extracting from a function return value
// or load instruction)
return 0;
}
+/// GetPointerBaseWithConstantOffset - Analyze the specified pointer to see if
+/// it can be expressed as a base pointer plus a constant offset. Return the
+/// base and offset to the caller.
+Value *llvm::GetPointerBaseWithConstantOffset(Value *Ptr, int64_t &Offset,
+ const TargetData &TD) {
+ Operator *PtrOp = dyn_cast<Operator>(Ptr);
+ if (PtrOp == 0) return Ptr;
+
+ // Just look through bitcasts.
+ if (PtrOp->getOpcode() == Instruction::BitCast)
+ return GetPointerBaseWithConstantOffset(PtrOp->getOperand(0), Offset, TD);
+
+ // If this is a GEP with constant indices, we can look through it.
+ GEPOperator *GEP = dyn_cast<GEPOperator>(PtrOp);
+ if (GEP == 0 || !GEP->hasAllConstantIndices()) return Ptr;
+
+ gep_type_iterator GTI = gep_type_begin(GEP);
+ for (User::op_iterator I = GEP->idx_begin(), E = GEP->idx_end(); I != E;
+ ++I, ++GTI) {
+ ConstantInt *OpC = cast<ConstantInt>(*I);
+ if (OpC->isZero()) continue;
+
+ // Handle a struct and array indices which add their offset to the pointer.
+ if (StructType *STy = dyn_cast<StructType>(*GTI)) {
+ Offset += TD.getStructLayout(STy)->getElementOffset(OpC->getZExtValue());
+ } else {
+ uint64_t Size = TD.getTypeAllocSize(GTI.getIndexedType());
+ Offset += OpC->getSExtValue()*Size;
+ }
+ }
+
+ // Re-sign extend from the pointer size if needed to get overflow edge cases
+ // right.
+ unsigned PtrSize = TD.getPointerSizeInBits();
+ if (PtrSize < 64)
+ Offset = (Offset << (64-PtrSize)) >> (64-PtrSize);
+
+ return GetPointerBaseWithConstantOffset(GEP->getPointerOperand(), Offset, TD);
+}
+
+
/// GetConstantStringInfo - This function computes the length of a
/// null-terminated C string pointed to by V. If successful, it returns true
/// and returns the string in Str. If unsuccessful, it returns false.
@@ -1372,8 +1554,8 @@
return false;
// Make sure the index-ee is a pointer to array of i8.
- const PointerType *PT = cast<PointerType>(GEP->getOperand(0)->getType());
- const ArrayType *AT = dyn_cast<ArrayType>(PT->getElementType());
+ PointerType *PT = cast<PointerType>(GEP->getOperand(0)->getType());
+ ArrayType *AT = dyn_cast<ArrayType>(PT->getElementType());
if (AT == 0 || !AT->getElementType()->isIntegerTy(8))
return false;
@@ -1566,3 +1748,48 @@
// an empty string as a length.
return Len == ~0ULL ? 1 : Len;
}
+
+Value *
+llvm::GetUnderlyingObject(Value *V, const TargetData *TD, unsigned MaxLookup) {
+ if (!V->getType()->isPointerTy())
+ return V;
+ for (unsigned Count = 0; MaxLookup == 0 || Count < MaxLookup; ++Count) {
+ if (GEPOperator *GEP = dyn_cast<GEPOperator>(V)) {
+ V = GEP->getPointerOperand();
+ } else if (Operator::getOpcode(V) == Instruction::BitCast) {
+ V = cast<Operator>(V)->getOperand(0);
+ } else if (GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
+ if (GA->mayBeOverridden())
+ return V;
+ V = GA->getAliasee();
+ } else {
+ // See if InstructionSimplify knows any relevant tricks.
+ if (Instruction *I = dyn_cast<Instruction>(V))
+ // TODO: Acquire a DominatorTree and use it.
+ if (Value *Simplified = SimplifyInstruction(I, TD, 0)) {
+ V = Simplified;
+ continue;
+ }
+
+ return V;
+ }
+ assert(V->getType()->isPointerTy() && "Unexpected operand type!");
+ }
+ return V;
+}
+
+/// onlyUsedByLifetimeMarkers - Return true if the only users of this pointer
+/// are lifetime markers.
+///
+bool llvm::onlyUsedByLifetimeMarkers(const Value *V) {
+ for (Value::const_use_iterator UI = V->use_begin(), UE = V->use_end();
+ UI != UE; ++UI) {
+ const IntrinsicInst *II = dyn_cast<IntrinsicInst>(*UI);
+ if (!II) return false;
+
+ if (II->getIntrinsicID() != Intrinsic::lifetime_start &&
+ II->getIntrinsicID() != Intrinsic::lifetime_end)
+ return false;
+ }
+ return true;
+}
