blob: becc66a4ee90f5693254bf3d2f25a646ff123e87 [file] [log] [blame]
//===- HexagonHardwareLoops.cpp - Identify and generate hardware loops ----===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This pass identifies loops where we can generate the Hexagon hardware
// loop instruction. The hardware loop can perform loop branches with a
// zero-cycle overhead.
//
// The pattern that defines the induction variable can changed depending on
// prior optimizations. For example, the IndVarSimplify phase run by 'opt'
// normalizes induction variables, and the Loop Strength Reduction pass
// run by 'llc' may also make changes to the induction variable.
// The pattern detected by this phase is due to running Strength Reduction.
//
// Criteria for hardware loops:
// - Countable loops (w/ ind. var for a trip count)
// - Assumes loops are normalized by IndVarSimplify
// - Try inner-most loops first
// - No function calls in loops.
//
//===----------------------------------------------------------------------===//
#include "HexagonInstrInfo.h"
#include "HexagonSubtarget.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DebugLoc.h"
#include "llvm/InitializePasses.h"
#include "llvm/Pass.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <cassert>
#include <cstdint>
#include <cstdlib>
#include <iterator>
#include <map>
#include <set>
#include <string>
#include <utility>
#include <vector>
using namespace llvm;
#define DEBUG_TYPE "hwloops"
#ifndef NDEBUG
static cl::opt<int> HWLoopLimit("hexagon-max-hwloop", cl::Hidden, cl::init(-1));
// Option to create preheader only for a specific function.
static cl::opt<std::string> PHFn("hexagon-hwloop-phfn", cl::Hidden,
cl::init(""));
#endif
// Option to create a preheader if one doesn't exist.
static cl::opt<bool> HWCreatePreheader("hexagon-hwloop-preheader",
cl::Hidden, cl::init(true),
cl::desc("Add a preheader to a hardware loop if one doesn't exist"));
// Turn it off by default. If a preheader block is not created here, the
// software pipeliner may be unable to find a block suitable to serve as
// a preheader. In that case SWP will not run.
static cl::opt<bool> SpecPreheader("hwloop-spec-preheader", cl::Hidden,
cl::desc("Allow speculation of preheader "
"instructions"));
STATISTIC(NumHWLoops, "Number of loops converted to hardware loops");
namespace llvm {
FunctionPass *createHexagonHardwareLoops();
void initializeHexagonHardwareLoopsPass(PassRegistry&);
} // end namespace llvm
namespace {
class CountValue;
struct HexagonHardwareLoops : public MachineFunctionPass {
MachineLoopInfo *MLI;
MachineRegisterInfo *MRI;
MachineDominatorTree *MDT;
const HexagonInstrInfo *TII;
const HexagonRegisterInfo *TRI;
#ifndef NDEBUG
static int Counter;
#endif
public:
static char ID;
HexagonHardwareLoops() : MachineFunctionPass(ID) {}
bool runOnMachineFunction(MachineFunction &MF) override;
StringRef getPassName() const override { return "Hexagon Hardware Loops"; }
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.addRequired<MachineDominatorTree>();
AU.addRequired<MachineLoopInfo>();
MachineFunctionPass::getAnalysisUsage(AU);
}
private:
using LoopFeederMap = std::map<Register, MachineInstr *>;
/// Kinds of comparisons in the compare instructions.
struct Comparison {
enum Kind {
EQ = 0x01,
NE = 0x02,
L = 0x04,
G = 0x08,
U = 0x40,
LTs = L,
LEs = L | EQ,
GTs = G,
GEs = G | EQ,
LTu = L | U,
LEu = L | EQ | U,
GTu = G | U,
GEu = G | EQ | U
};
static Kind getSwappedComparison(Kind Cmp) {
assert ((!((Cmp & L) && (Cmp & G))) && "Malformed comparison operator");
if ((Cmp & L) || (Cmp & G))
return (Kind)(Cmp ^ (L|G));
return Cmp;
}
static Kind getNegatedComparison(Kind Cmp) {
if ((Cmp & L) || (Cmp & G))
return (Kind)((Cmp ^ (L | G)) ^ EQ);
if ((Cmp & NE) || (Cmp & EQ))
return (Kind)(Cmp ^ (EQ | NE));
return (Kind)0;
}
static bool isSigned(Kind Cmp) {
return (Cmp & (L | G) && !(Cmp & U));
}
static bool isUnsigned(Kind Cmp) {
return (Cmp & U);
}
};
/// Find the register that contains the loop controlling
/// induction variable.
/// If successful, it will return true and set the \p Reg, \p IVBump
/// and \p IVOp arguments. Otherwise it will return false.
/// The returned induction register is the register R that follows the
/// following induction pattern:
/// loop:
/// R = phi ..., [ R.next, LatchBlock ]
/// R.next = R + #bump
/// if (R.next < #N) goto loop
/// IVBump is the immediate value added to R, and IVOp is the instruction
/// "R.next = R + #bump".
bool findInductionRegister(MachineLoop *L, Register &Reg,
int64_t &IVBump, MachineInstr *&IVOp) const;
/// Return the comparison kind for the specified opcode.
Comparison::Kind getComparisonKind(unsigned CondOpc,
MachineOperand *InitialValue,
const MachineOperand *Endvalue,
int64_t IVBump) const;
/// Analyze the statements in a loop to determine if the loop
/// has a computable trip count and, if so, return a value that represents
/// the trip count expression.
CountValue *getLoopTripCount(MachineLoop *L,
SmallVectorImpl<MachineInstr *> &OldInsts);
/// Return the expression that represents the number of times
/// a loop iterates. The function takes the operands that represent the
/// loop start value, loop end value, and induction value. Based upon
/// these operands, the function attempts to compute the trip count.
/// If the trip count is not directly available (as an immediate value,
/// or a register), the function will attempt to insert computation of it
/// to the loop's preheader.
CountValue *computeCount(MachineLoop *Loop, const MachineOperand *Start,
const MachineOperand *End, Register IVReg,
int64_t IVBump, Comparison::Kind Cmp) const;
/// Return true if the instruction is not valid within a hardware
/// loop.
bool isInvalidLoopOperation(const MachineInstr *MI,
bool IsInnerHWLoop) const;
/// Return true if the loop contains an instruction that inhibits
/// using the hardware loop.
bool containsInvalidInstruction(MachineLoop *L, bool IsInnerHWLoop) const;
/// Given a loop, check if we can convert it to a hardware loop.
/// If so, then perform the conversion and return true.
bool convertToHardwareLoop(MachineLoop *L, bool &L0used, bool &L1used);
/// Return true if the instruction is now dead.
bool isDead(const MachineInstr *MI,
SmallVectorImpl<MachineInstr *> &DeadPhis) const;
/// Remove the instruction if it is now dead.
void removeIfDead(MachineInstr *MI);
/// Make sure that the "bump" instruction executes before the
/// compare. We need that for the IV fixup, so that the compare
/// instruction would not use a bumped value that has not yet been
/// defined. If the instructions are out of order, try to reorder them.
bool orderBumpCompare(MachineInstr *BumpI, MachineInstr *CmpI);
/// Return true if MO and MI pair is visited only once. If visited
/// more than once, this indicates there is recursion. In such a case,
/// return false.
bool isLoopFeeder(MachineLoop *L, MachineBasicBlock *A, MachineInstr *MI,
const MachineOperand *MO,
LoopFeederMap &LoopFeederPhi) const;
/// Return true if the Phi may generate a value that may underflow,
/// or may wrap.
bool phiMayWrapOrUnderflow(MachineInstr *Phi, const MachineOperand *EndVal,
MachineBasicBlock *MBB, MachineLoop *L,
LoopFeederMap &LoopFeederPhi) const;
/// Return true if the induction variable may underflow an unsigned
/// value in the first iteration.
bool loopCountMayWrapOrUnderFlow(const MachineOperand *InitVal,
const MachineOperand *EndVal,
MachineBasicBlock *MBB, MachineLoop *L,
LoopFeederMap &LoopFeederPhi) const;
/// Check if the given operand has a compile-time known constant
/// value. Return true if yes, and false otherwise. When returning true, set
/// Val to the corresponding constant value.
bool checkForImmediate(const MachineOperand &MO, int64_t &Val) const;
/// Check if the operand has a compile-time known constant value.
bool isImmediate(const MachineOperand &MO) const {
int64_t V;
return checkForImmediate(MO, V);
}
/// Return the immediate for the specified operand.
int64_t getImmediate(const MachineOperand &MO) const {
int64_t V;
if (!checkForImmediate(MO, V))
llvm_unreachable("Invalid operand");
return V;
}
/// Reset the given machine operand to now refer to a new immediate
/// value. Assumes that the operand was already referencing an immediate
/// value, either directly, or via a register.
void setImmediate(MachineOperand &MO, int64_t Val);
/// Fix the data flow of the induction variable.
/// The desired flow is: phi ---> bump -+-> comparison-in-latch.
/// |
/// +-> back to phi
/// where "bump" is the increment of the induction variable:
/// iv = iv + #const.
/// Due to some prior code transformations, the actual flow may look
/// like this:
/// phi -+-> bump ---> back to phi
/// |
/// +-> comparison-in-latch (against upper_bound-bump),
/// i.e. the comparison that controls the loop execution may be using
/// the value of the induction variable from before the increment.
///
/// Return true if the loop's flow is the desired one (i.e. it's
/// either been fixed, or no fixing was necessary).
/// Otherwise, return false. This can happen if the induction variable
/// couldn't be identified, or if the value in the latch's comparison
/// cannot be adjusted to reflect the post-bump value.
bool fixupInductionVariable(MachineLoop *L);
/// Given a loop, if it does not have a preheader, create one.
/// Return the block that is the preheader.
MachineBasicBlock *createPreheaderForLoop(MachineLoop *L);
};
char HexagonHardwareLoops::ID = 0;
#ifndef NDEBUG
int HexagonHardwareLoops::Counter = 0;
#endif
/// Abstraction for a trip count of a loop. A smaller version
/// of the MachineOperand class without the concerns of changing the
/// operand representation.
class CountValue {
public:
enum CountValueType {
CV_Register,
CV_Immediate
};
private:
CountValueType Kind;
union Values {
Values() : R{Register(), 0} {}
Values(const Values&) = default;
struct {
Register Reg;
unsigned Sub;
} R;
unsigned ImmVal;
} Contents;
public:
explicit CountValue(CountValueType t, Register v, unsigned u = 0) {
Kind = t;
if (Kind == CV_Register) {
Contents.R.Reg = v;
Contents.R.Sub = u;
} else {
Contents.ImmVal = v;
}
}
bool isReg() const { return Kind == CV_Register; }
bool isImm() const { return Kind == CV_Immediate; }
Register getReg() const {
assert(isReg() && "Wrong CountValue accessor");
return Contents.R.Reg;
}
unsigned getSubReg() const {
assert(isReg() && "Wrong CountValue accessor");
return Contents.R.Sub;
}
unsigned getImm() const {
assert(isImm() && "Wrong CountValue accessor");
return Contents.ImmVal;
}
void print(raw_ostream &OS, const TargetRegisterInfo *TRI = nullptr) const {
if (isReg()) { OS << printReg(Contents.R.Reg, TRI, Contents.R.Sub); }
if (isImm()) { OS << Contents.ImmVal; }
}
};
} // end anonymous namespace
INITIALIZE_PASS_BEGIN(HexagonHardwareLoops, "hwloops",
"Hexagon Hardware Loops", false, false)
INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
INITIALIZE_PASS_END(HexagonHardwareLoops, "hwloops",
"Hexagon Hardware Loops", false, false)
FunctionPass *llvm::createHexagonHardwareLoops() {
return new HexagonHardwareLoops();
}
bool HexagonHardwareLoops::runOnMachineFunction(MachineFunction &MF) {
LLVM_DEBUG(dbgs() << "********* Hexagon Hardware Loops *********\n");
if (skipFunction(MF.getFunction()))
return false;
bool Changed = false;
MLI = &getAnalysis<MachineLoopInfo>();
MRI = &MF.getRegInfo();
MDT = &getAnalysis<MachineDominatorTree>();
const HexagonSubtarget &HST = MF.getSubtarget<HexagonSubtarget>();
TII = HST.getInstrInfo();
TRI = HST.getRegisterInfo();
for (auto &L : *MLI)
if (L->isOutermost()) {
bool L0Used = false;
bool L1Used = false;
Changed |= convertToHardwareLoop(L, L0Used, L1Used);
}
return Changed;
}
bool HexagonHardwareLoops::findInductionRegister(MachineLoop *L,
Register &Reg,
int64_t &IVBump,
MachineInstr *&IVOp
) const {
MachineBasicBlock *Header = L->getHeader();
MachineBasicBlock *Preheader = MLI->findLoopPreheader(L, SpecPreheader);
MachineBasicBlock *Latch = L->getLoopLatch();
MachineBasicBlock *ExitingBlock = L->findLoopControlBlock();
if (!Header || !Preheader || !Latch || !ExitingBlock)
return false;
// This pair represents an induction register together with an immediate
// value that will be added to it in each loop iteration.
using RegisterBump = std::pair<Register, int64_t>;
// Mapping: R.next -> (R, bump), where R, R.next and bump are derived
// from an induction operation
// R.next = R + bump
// where bump is an immediate value.
using InductionMap = std::map<Register, RegisterBump>;
InductionMap IndMap;
using instr_iterator = MachineBasicBlock::instr_iterator;
for (instr_iterator I = Header->instr_begin(), E = Header->instr_end();
I != E && I->isPHI(); ++I) {
MachineInstr *Phi = &*I;
// Have a PHI instruction. Get the operand that corresponds to the
// latch block, and see if is a result of an addition of form "reg+imm",
// where the "reg" is defined by the PHI node we are looking at.
for (unsigned i = 1, n = Phi->getNumOperands(); i < n; i += 2) {
if (Phi->getOperand(i+1).getMBB() != Latch)
continue;
Register PhiOpReg = Phi->getOperand(i).getReg();
MachineInstr *DI = MRI->getVRegDef(PhiOpReg);
if (DI->getDesc().isAdd()) {
// If the register operand to the add is the PHI we're looking at, this
// meets the induction pattern.
Register IndReg = DI->getOperand(1).getReg();
MachineOperand &Opnd2 = DI->getOperand(2);
int64_t V;
if (MRI->getVRegDef(IndReg) == Phi && checkForImmediate(Opnd2, V)) {
Register UpdReg = DI->getOperand(0).getReg();
IndMap.insert(std::make_pair(UpdReg, std::make_pair(IndReg, V)));
}
}
} // for (i)
} // for (instr)
SmallVector<MachineOperand,2> Cond;
MachineBasicBlock *TB = nullptr, *FB = nullptr;
bool NotAnalyzed = TII->analyzeBranch(*ExitingBlock, TB, FB, Cond, false);
if (NotAnalyzed)
return false;
Register PredR;
unsigned PredPos, PredRegFlags;
if (!TII->getPredReg(Cond, PredR, PredPos, PredRegFlags))
return false;
MachineInstr *PredI = MRI->getVRegDef(PredR);
if (!PredI->isCompare())
return false;
Register CmpReg1, CmpReg2;
int64_t CmpImm = 0, CmpMask = 0;
bool CmpAnalyzed =
TII->analyzeCompare(*PredI, CmpReg1, CmpReg2, CmpMask, CmpImm);
// Fail if the compare was not analyzed, or it's not comparing a register
// with an immediate value. Not checking the mask here, since we handle
// the individual compare opcodes (including A4_cmpb*) later on.
if (!CmpAnalyzed)
return false;
// Exactly one of the input registers to the comparison should be among
// the induction registers.
InductionMap::iterator IndMapEnd = IndMap.end();
InductionMap::iterator F = IndMapEnd;
if (CmpReg1 != 0) {
InductionMap::iterator F1 = IndMap.find(CmpReg1);
if (F1 != IndMapEnd)
F = F1;
}
if (CmpReg2 != 0) {
InductionMap::iterator F2 = IndMap.find(CmpReg2);
if (F2 != IndMapEnd) {
if (F != IndMapEnd)
return false;
F = F2;
}
}
if (F == IndMapEnd)
return false;
Reg = F->second.first;
IVBump = F->second.second;
IVOp = MRI->getVRegDef(F->first);
return true;
}
// Return the comparison kind for the specified opcode.
HexagonHardwareLoops::Comparison::Kind
HexagonHardwareLoops::getComparisonKind(unsigned CondOpc,
MachineOperand *InitialValue,
const MachineOperand *EndValue,
int64_t IVBump) const {
Comparison::Kind Cmp = (Comparison::Kind)0;
switch (CondOpc) {
case Hexagon::C2_cmpeq:
case Hexagon::C2_cmpeqi:
case Hexagon::C2_cmpeqp:
Cmp = Comparison::EQ;
break;
case Hexagon::C4_cmpneq:
case Hexagon::C4_cmpneqi:
Cmp = Comparison::NE;
break;
case Hexagon::C2_cmplt:
Cmp = Comparison::LTs;
break;
case Hexagon::C2_cmpltu:
Cmp = Comparison::LTu;
break;
case Hexagon::C4_cmplte:
case Hexagon::C4_cmpltei:
Cmp = Comparison::LEs;
break;
case Hexagon::C4_cmplteu:
case Hexagon::C4_cmplteui:
Cmp = Comparison::LEu;
break;
case Hexagon::C2_cmpgt:
case Hexagon::C2_cmpgti:
case Hexagon::C2_cmpgtp:
Cmp = Comparison::GTs;
break;
case Hexagon::C2_cmpgtu:
case Hexagon::C2_cmpgtui:
case Hexagon::C2_cmpgtup:
Cmp = Comparison::GTu;
break;
case Hexagon::C2_cmpgei:
Cmp = Comparison::GEs;
break;
case Hexagon::C2_cmpgeui:
Cmp = Comparison::GEs;
break;
default:
return (Comparison::Kind)0;
}
return Cmp;
}
/// Analyze the statements in a loop to determine if the loop has
/// a computable trip count and, if so, return a value that represents
/// the trip count expression.
///
/// This function iterates over the phi nodes in the loop to check for
/// induction variable patterns that are used in the calculation for
/// the number of time the loop is executed.
CountValue *HexagonHardwareLoops::getLoopTripCount(MachineLoop *L,
SmallVectorImpl<MachineInstr *> &OldInsts) {
MachineBasicBlock *TopMBB = L->getTopBlock();
MachineBasicBlock::pred_iterator PI = TopMBB->pred_begin();
assert(PI != TopMBB->pred_end() &&
"Loop must have more than one incoming edge!");
MachineBasicBlock *Backedge = *PI++;
if (PI == TopMBB->pred_end()) // dead loop?
return nullptr;
MachineBasicBlock *Incoming = *PI++;
if (PI != TopMBB->pred_end()) // multiple backedges?
return nullptr;
// Make sure there is one incoming and one backedge and determine which
// is which.
if (L->contains(Incoming)) {
if (L->contains(Backedge))
return nullptr;
std::swap(Incoming, Backedge);
} else if (!L->contains(Backedge))
return nullptr;
// Look for the cmp instruction to determine if we can get a useful trip
// count. The trip count can be either a register or an immediate. The
// location of the value depends upon the type (reg or imm).
MachineBasicBlock *ExitingBlock = L->findLoopControlBlock();
if (!ExitingBlock)
return nullptr;
Register IVReg = 0;
int64_t IVBump = 0;
MachineInstr *IVOp;
bool FoundIV = findInductionRegister(L, IVReg, IVBump, IVOp);
if (!FoundIV)
return nullptr;
MachineBasicBlock *Preheader = MLI->findLoopPreheader(L, SpecPreheader);
MachineOperand *InitialValue = nullptr;
MachineInstr *IV_Phi = MRI->getVRegDef(IVReg);
MachineBasicBlock *Latch = L->getLoopLatch();
for (unsigned i = 1, n = IV_Phi->getNumOperands(); i < n; i += 2) {
MachineBasicBlock *MBB = IV_Phi->getOperand(i+1).getMBB();
if (MBB == Preheader)
InitialValue = &IV_Phi->getOperand(i);
else if (MBB == Latch)
IVReg = IV_Phi->getOperand(i).getReg(); // Want IV reg after bump.
}
if (!InitialValue)
return nullptr;
SmallVector<MachineOperand,2> Cond;
MachineBasicBlock *TB = nullptr, *FB = nullptr;
bool NotAnalyzed = TII->analyzeBranch(*ExitingBlock, TB, FB, Cond, false);
if (NotAnalyzed)
return nullptr;
MachineBasicBlock *Header = L->getHeader();
// TB must be non-null. If FB is also non-null, one of them must be
// the header. Otherwise, branch to TB could be exiting the loop, and
// the fall through can go to the header.
assert (TB && "Exit block without a branch?");
if (ExitingBlock != Latch && (TB == Latch || FB == Latch)) {
MachineBasicBlock *LTB = nullptr, *LFB = nullptr;
SmallVector<MachineOperand,2> LCond;
bool NotAnalyzed = TII->analyzeBranch(*Latch, LTB, LFB, LCond, false);
if (NotAnalyzed)
return nullptr;
if (TB == Latch)
TB = (LTB == Header) ? LTB : LFB;
else
FB = (LTB == Header) ? LTB: LFB;
}
assert ((!FB || TB == Header || FB == Header) && "Branches not to header?");
if (!TB || (FB && TB != Header && FB != Header))
return nullptr;
// Branches of form "if (!P) ..." cause HexagonInstrInfo::analyzeBranch
// to put imm(0), followed by P in the vector Cond.
// If TB is not the header, it means that the "not-taken" path must lead
// to the header.
bool Negated = TII->predOpcodeHasNot(Cond) ^ (TB != Header);
Register PredReg;
unsigned PredPos, PredRegFlags;
if (!TII->getPredReg(Cond, PredReg, PredPos, PredRegFlags))
return nullptr;
MachineInstr *CondI = MRI->getVRegDef(PredReg);
unsigned CondOpc = CondI->getOpcode();
Register CmpReg1, CmpReg2;
int64_t Mask = 0, ImmValue = 0;
bool AnalyzedCmp =
TII->analyzeCompare(*CondI, CmpReg1, CmpReg2, Mask, ImmValue);
if (!AnalyzedCmp)
return nullptr;
// The comparison operator type determines how we compute the loop
// trip count.
OldInsts.push_back(CondI);
OldInsts.push_back(IVOp);
// Sadly, the following code gets information based on the position
// of the operands in the compare instruction. This has to be done
// this way, because the comparisons check for a specific relationship
// between the operands (e.g. is-less-than), rather than to find out
// what relationship the operands are in (as on PPC).
Comparison::Kind Cmp;
bool isSwapped = false;
const MachineOperand &Op1 = CondI->getOperand(1);
const MachineOperand &Op2 = CondI->getOperand(2);
const MachineOperand *EndValue = nullptr;
if (Op1.isReg()) {
if (Op2.isImm() || Op1.getReg() == IVReg)
EndValue = &Op2;
else {
EndValue = &Op1;
isSwapped = true;
}
}
if (!EndValue)
return nullptr;
Cmp = getComparisonKind(CondOpc, InitialValue, EndValue, IVBump);
if (!Cmp)
return nullptr;
if (Negated)
Cmp = Comparison::getNegatedComparison(Cmp);
if (isSwapped)
Cmp = Comparison::getSwappedComparison(Cmp);
if (InitialValue->isReg()) {
Register R = InitialValue->getReg();
MachineBasicBlock *DefBB = MRI->getVRegDef(R)->getParent();
if (!MDT->properlyDominates(DefBB, Header)) {
int64_t V;
if (!checkForImmediate(*InitialValue, V))
return nullptr;
}
OldInsts.push_back(MRI->getVRegDef(R));
}
if (EndValue->isReg()) {
Register R = EndValue->getReg();
MachineBasicBlock *DefBB = MRI->getVRegDef(R)->getParent();
if (!MDT->properlyDominates(DefBB, Header)) {
int64_t V;
if (!checkForImmediate(*EndValue, V))
return nullptr;
}
OldInsts.push_back(MRI->getVRegDef(R));
}
return computeCount(L, InitialValue, EndValue, IVReg, IVBump, Cmp);
}
/// Helper function that returns the expression that represents the
/// number of times a loop iterates. The function takes the operands that
/// represent the loop start value, loop end value, and induction value.
/// Based upon these operands, the function attempts to compute the trip count.
CountValue *HexagonHardwareLoops::computeCount(MachineLoop *Loop,
const MachineOperand *Start,
const MachineOperand *End,
Register IVReg,
int64_t IVBump,
Comparison::Kind Cmp) const {
// Cannot handle comparison EQ, i.e. while (A == B).
if (Cmp == Comparison::EQ)
return nullptr;
// Check if either the start or end values are an assignment of an immediate.
// If so, use the immediate value rather than the register.
if (Start->isReg()) {
const MachineInstr *StartValInstr = MRI->getVRegDef(Start->getReg());
if (StartValInstr && (StartValInstr->getOpcode() == Hexagon::A2_tfrsi ||
StartValInstr->getOpcode() == Hexagon::A2_tfrpi))
Start = &StartValInstr->getOperand(1);
}
if (End->isReg()) {
const MachineInstr *EndValInstr = MRI->getVRegDef(End->getReg());
if (EndValInstr && (EndValInstr->getOpcode() == Hexagon::A2_tfrsi ||
EndValInstr->getOpcode() == Hexagon::A2_tfrpi))
End = &EndValInstr->getOperand(1);
}
if (!Start->isReg() && !Start->isImm())
return nullptr;
if (!End->isReg() && !End->isImm())
return nullptr;
bool CmpLess = Cmp & Comparison::L;
bool CmpGreater = Cmp & Comparison::G;
bool CmpHasEqual = Cmp & Comparison::EQ;
// Avoid certain wrap-arounds. This doesn't detect all wrap-arounds.
if (CmpLess && IVBump < 0)
// Loop going while iv is "less" with the iv value going down. Must wrap.
return nullptr;
if (CmpGreater && IVBump > 0)
// Loop going while iv is "greater" with the iv value going up. Must wrap.
return nullptr;
// Phis that may feed into the loop.
LoopFeederMap LoopFeederPhi;
// Check if the initial value may be zero and can be decremented in the first
// iteration. If the value is zero, the endloop instruction will not decrement
// the loop counter, so we shouldn't generate a hardware loop in this case.
if (loopCountMayWrapOrUnderFlow(Start, End, Loop->getLoopPreheader(), Loop,
LoopFeederPhi))
return nullptr;
if (Start->isImm() && End->isImm()) {
// Both, start and end are immediates.
int64_t StartV = Start->getImm();
int64_t EndV = End->getImm();
int64_t Dist = EndV - StartV;
if (Dist == 0)
return nullptr;
bool Exact = (Dist % IVBump) == 0;
if (Cmp == Comparison::NE) {
if (!Exact)
return nullptr;
if ((Dist < 0) ^ (IVBump < 0))
return nullptr;
}
// For comparisons that include the final value (i.e. include equality
// with the final value), we need to increase the distance by 1.
if (CmpHasEqual)
Dist = Dist > 0 ? Dist+1 : Dist-1;
// For the loop to iterate, CmpLess should imply Dist > 0. Similarly,
// CmpGreater should imply Dist < 0. These conditions could actually
// fail, for example, in unreachable code (which may still appear to be
// reachable in the CFG).
if ((CmpLess && Dist < 0) || (CmpGreater && Dist > 0))
return nullptr;
// "Normalized" distance, i.e. with the bump set to +-1.
int64_t Dist1 = (IVBump > 0) ? (Dist + (IVBump - 1)) / IVBump
: (-Dist + (-IVBump - 1)) / (-IVBump);
assert (Dist1 > 0 && "Fishy thing. Both operands have the same sign.");
uint64_t Count = Dist1;
if (Count > 0xFFFFFFFFULL)
return nullptr;
return new CountValue(CountValue::CV_Immediate, Count);
}
// A general case: Start and End are some values, but the actual
// iteration count may not be available. If it is not, insert
// a computation of it into the preheader.
// If the induction variable bump is not a power of 2, quit.
// Othwerise we'd need a general integer division.
if (!isPowerOf2_64(std::abs(IVBump)))
return nullptr;
MachineBasicBlock *PH = MLI->findLoopPreheader(Loop, SpecPreheader);
assert (PH && "Should have a preheader by now");
MachineBasicBlock::iterator InsertPos = PH->getFirstTerminator();
DebugLoc DL;
if (InsertPos != PH->end())
DL = InsertPos->getDebugLoc();
// If Start is an immediate and End is a register, the trip count
// will be "reg - imm". Hexagon's "subtract immediate" instruction
// is actually "reg + -imm".
// If the loop IV is going downwards, i.e. if the bump is negative,
// then the iteration count (computed as End-Start) will need to be
// negated. To avoid the negation, just swap Start and End.
if (IVBump < 0) {
std::swap(Start, End);
IVBump = -IVBump;
}
// Cmp may now have a wrong direction, e.g. LEs may now be GEs.
// Signedness, and "including equality" are preserved.
bool RegToImm = Start->isReg() && End->isImm(); // for (reg..imm)
bool RegToReg = Start->isReg() && End->isReg(); // for (reg..reg)
int64_t StartV = 0, EndV = 0;
if (Start->isImm())
StartV = Start->getImm();
if (End->isImm())
EndV = End->getImm();
int64_t AdjV = 0;
// To compute the iteration count, we would need this computation:
// Count = (End - Start + (IVBump-1)) / IVBump
// or, when CmpHasEqual:
// Count = (End - Start + (IVBump-1)+1) / IVBump
// The "IVBump-1" part is the adjustment (AdjV). We can avoid
// generating an instruction specifically to add it if we can adjust
// the immediate values for Start or End.
if (CmpHasEqual) {
// Need to add 1 to the total iteration count.
if (Start->isImm())
StartV--;
else if (End->isImm())
EndV++;
else
AdjV += 1;
}
if (Cmp != Comparison::NE) {
if (Start->isImm())
StartV -= (IVBump-1);
else if (End->isImm())
EndV += (IVBump-1);
else
AdjV += (IVBump-1);
}
Register R = 0;
unsigned SR = 0;
if (Start->isReg()) {
R = Start->getReg();
SR = Start->getSubReg();
} else {
R = End->getReg();
SR = End->getSubReg();
}
const TargetRegisterClass *RC = MRI->getRegClass(R);
// Hardware loops cannot handle 64-bit registers. If it's a double
// register, it has to have a subregister.
if (!SR && RC == &Hexagon::DoubleRegsRegClass)
return nullptr;
const TargetRegisterClass *IntRC = &Hexagon::IntRegsRegClass;
// Compute DistR (register with the distance between Start and End).
Register DistR;
unsigned DistSR;
// Avoid special case, where the start value is an imm(0).
if (Start->isImm() && StartV == 0) {
DistR = End->getReg();
DistSR = End->getSubReg();
} else {
const MCInstrDesc &SubD = RegToReg ? TII->get(Hexagon::A2_sub) :
(RegToImm ? TII->get(Hexagon::A2_subri) :
TII->get(Hexagon::A2_addi));
if (RegToReg || RegToImm) {
Register SubR = MRI->createVirtualRegister(IntRC);
MachineInstrBuilder SubIB =
BuildMI(*PH, InsertPos, DL, SubD, SubR);
if (RegToReg)
SubIB.addReg(End->getReg(), 0, End->getSubReg())
.addReg(Start->getReg(), 0, Start->getSubReg());
else
SubIB.addImm(EndV)
.addReg(Start->getReg(), 0, Start->getSubReg());
DistR = SubR;
} else {
// If the loop has been unrolled, we should use the original loop count
// instead of recalculating the value. This will avoid additional
// 'Add' instruction.
const MachineInstr *EndValInstr = MRI->getVRegDef(End->getReg());
if (EndValInstr->getOpcode() == Hexagon::A2_addi &&
EndValInstr->getOperand(1).getSubReg() == 0 &&
EndValInstr->getOperand(2).getImm() == StartV) {
DistR = EndValInstr->getOperand(1).getReg();
} else {
Register SubR = MRI->createVirtualRegister(IntRC);
MachineInstrBuilder SubIB =
BuildMI(*PH, InsertPos, DL, SubD, SubR);
SubIB.addReg(End->getReg(), 0, End->getSubReg())
.addImm(-StartV);
DistR = SubR;
}
}
DistSR = 0;
}
// From DistR, compute AdjR (register with the adjusted distance).
Register AdjR;
unsigned AdjSR;
if (AdjV == 0) {
AdjR = DistR;
AdjSR = DistSR;
} else {
// Generate CountR = ADD DistR, AdjVal
Register AddR = MRI->createVirtualRegister(IntRC);
MCInstrDesc const &AddD = TII->get(Hexagon::A2_addi);
BuildMI(*PH, InsertPos, DL, AddD, AddR)
.addReg(DistR, 0, DistSR)
.addImm(AdjV);
AdjR = AddR;
AdjSR = 0;
}
// From AdjR, compute CountR (register with the final count).
Register CountR;
unsigned CountSR;
if (IVBump == 1) {
CountR = AdjR;
CountSR = AdjSR;
} else {
// The IV bump is a power of two. Log_2(IV bump) is the shift amount.
unsigned Shift = Log2_32(IVBump);
// Generate NormR = LSR DistR, Shift.
Register LsrR = MRI->createVirtualRegister(IntRC);
const MCInstrDesc &LsrD = TII->get(Hexagon::S2_lsr_i_r);
BuildMI(*PH, InsertPos, DL, LsrD, LsrR)
.addReg(AdjR, 0, AdjSR)
.addImm(Shift);
CountR = LsrR;
CountSR = 0;
}
return new CountValue(CountValue::CV_Register, CountR, CountSR);
}
/// Return true if the operation is invalid within hardware loop.
bool HexagonHardwareLoops::isInvalidLoopOperation(const MachineInstr *MI,
bool IsInnerHWLoop) const {
// Call is not allowed because the callee may use a hardware loop except for
// the case when the call never returns.
if (MI->getDesc().isCall())
return !TII->doesNotReturn(*MI);
// Check if the instruction defines a hardware loop register.
using namespace Hexagon;
static const Register Regs01[] = { LC0, SA0, LC1, SA1 };
static const Register Regs1[] = { LC1, SA1 };
auto CheckRegs = IsInnerHWLoop ? ArrayRef(Regs01, std::size(Regs01))
: ArrayRef(Regs1, std::size(Regs1));
for (Register R : CheckRegs)
if (MI->modifiesRegister(R, TRI))
return true;
return false;
}
/// Return true if the loop contains an instruction that inhibits
/// the use of the hardware loop instruction.
bool HexagonHardwareLoops::containsInvalidInstruction(MachineLoop *L,
bool IsInnerHWLoop) const {
LLVM_DEBUG(dbgs() << "\nhw_loop head, "
<< printMBBReference(**L->block_begin()));
for (MachineBasicBlock *MBB : L->getBlocks()) {
for (const MachineInstr &MI : *MBB) {
if (isInvalidLoopOperation(&MI, IsInnerHWLoop)) {
LLVM_DEBUG(dbgs() << "\nCannot convert to hw_loop due to:";
MI.dump(););
return true;
}
}
}
return false;
}
/// Returns true if the instruction is dead. This was essentially
/// copied from DeadMachineInstructionElim::isDead, but with special cases
/// for inline asm, physical registers and instructions with side effects
/// removed.
bool HexagonHardwareLoops::isDead(const MachineInstr *MI,
SmallVectorImpl<MachineInstr *> &DeadPhis) const {
// Examine each operand.
for (const MachineOperand &MO : MI->operands()) {
if (!MO.isReg() || !MO.isDef())
continue;
Register Reg = MO.getReg();
if (MRI->use_nodbg_empty(Reg))
continue;
using use_nodbg_iterator = MachineRegisterInfo::use_nodbg_iterator;
// This instruction has users, but if the only user is the phi node for the
// parent block, and the only use of that phi node is this instruction, then
// this instruction is dead: both it (and the phi node) can be removed.
use_nodbg_iterator I = MRI->use_nodbg_begin(Reg);
use_nodbg_iterator End = MRI->use_nodbg_end();
if (std::next(I) != End || !I->getParent()->isPHI())
return false;
MachineInstr *OnePhi = I->getParent();
for (unsigned j = 0, f = OnePhi->getNumOperands(); j != f; ++j) {
const MachineOperand &OPO = OnePhi->getOperand(j);
if (!OPO.isReg() || !OPO.isDef())
continue;
Register OPReg = OPO.getReg();
use_nodbg_iterator nextJ;
for (use_nodbg_iterator J = MRI->use_nodbg_begin(OPReg);
J != End; J = nextJ) {
nextJ = std::next(J);
MachineOperand &Use = *J;
MachineInstr *UseMI = Use.getParent();
// If the phi node has a user that is not MI, bail.
if (MI != UseMI)
return false;
}
}
DeadPhis.push_back(OnePhi);
}
// If there are no defs with uses, the instruction is dead.
return true;
}
void HexagonHardwareLoops::removeIfDead(MachineInstr *MI) {
// This procedure was essentially copied from DeadMachineInstructionElim.
SmallVector<MachineInstr*, 1> DeadPhis;
if (isDead(MI, DeadPhis)) {
LLVM_DEBUG(dbgs() << "HW looping will remove: " << *MI);
// It is possible that some DBG_VALUE instructions refer to this
// instruction. Examine each def operand for such references;
// if found, mark the DBG_VALUE as undef (but don't delete it).
for (const MachineOperand &MO : MI->operands()) {
if (!MO.isReg() || !MO.isDef())
continue;
Register Reg = MO.getReg();
// We use make_early_inc_range here because setReg below invalidates the
// iterator.
for (MachineOperand &MO :
llvm::make_early_inc_range(MRI->use_operands(Reg))) {
MachineInstr *UseMI = MO.getParent();
if (UseMI == MI)
continue;
if (MO.isDebug())
MO.setReg(0U);
}
}
MI->eraseFromParent();
for (unsigned i = 0; i < DeadPhis.size(); ++i)
DeadPhis[i]->eraseFromParent();
}
}
/// Check if the loop is a candidate for converting to a hardware
/// loop. If so, then perform the transformation.
///
/// This function works on innermost loops first. A loop can be converted
/// if it is a counting loop; either a register value or an immediate.
///
/// The code makes several assumptions about the representation of the loop
/// in llvm.
bool HexagonHardwareLoops::convertToHardwareLoop(MachineLoop *L,
bool &RecL0used,
bool &RecL1used) {
// This is just to confirm basic correctness.
assert(L->getHeader() && "Loop without a header?");
bool Changed = false;
bool L0Used = false;
bool L1Used = false;
// Process nested loops first.
for (MachineLoop *I : *L) {
Changed |= convertToHardwareLoop(I, RecL0used, RecL1used);
L0Used |= RecL0used;
L1Used |= RecL1used;
}
// If a nested loop has been converted, then we can't convert this loop.
if (Changed && L0Used && L1Used)
return Changed;
unsigned LOOP_i;
unsigned LOOP_r;
unsigned ENDLOOP;
// Flag used to track loopN instruction:
// 1 - Hardware loop is being generated for the inner most loop.
// 0 - Hardware loop is being generated for the outer loop.
unsigned IsInnerHWLoop = 1;
if (L0Used) {
LOOP_i = Hexagon::J2_loop1i;
LOOP_r = Hexagon::J2_loop1r;
ENDLOOP = Hexagon::ENDLOOP1;
IsInnerHWLoop = 0;
} else {
LOOP_i = Hexagon::J2_loop0i;
LOOP_r = Hexagon::J2_loop0r;
ENDLOOP = Hexagon::ENDLOOP0;
}
#ifndef NDEBUG
// Stop trying after reaching the limit (if any).
int Limit = HWLoopLimit;
if (Limit >= 0) {
if (Counter >= HWLoopLimit)
return false;
Counter++;
}
#endif
// Does the loop contain any invalid instructions?
if (containsInvalidInstruction(L, IsInnerHWLoop))
return false;
MachineBasicBlock *LastMBB = L->findLoopControlBlock();
// Don't generate hw loop if the loop has more than one exit.
if (!LastMBB)
return false;
MachineBasicBlock::iterator LastI = LastMBB->getFirstTerminator();
if (LastI == LastMBB->end())
return false;
// Is the induction variable bump feeding the latch condition?
if (!fixupInductionVariable(L))
return false;
// Ensure the loop has a preheader: the loop instruction will be
// placed there.
MachineBasicBlock *Preheader = MLI->findLoopPreheader(L, SpecPreheader);
if (!Preheader) {
Preheader = createPreheaderForLoop(L);
if (!Preheader)
return false;
}
MachineBasicBlock::iterator InsertPos = Preheader->getFirstTerminator();
SmallVector<MachineInstr*, 2> OldInsts;
// Are we able to determine the trip count for the loop?
CountValue *TripCount = getLoopTripCount(L, OldInsts);
if (!TripCount)
return false;
// Is the trip count available in the preheader?
if (TripCount->isReg()) {
// There will be a use of the register inserted into the preheader,
// so make sure that the register is actually defined at that point.
MachineInstr *TCDef = MRI->getVRegDef(TripCount->getReg());
MachineBasicBlock *BBDef = TCDef->getParent();
if (!MDT->dominates(BBDef, Preheader))
return false;
}
// Determine the loop start.
MachineBasicBlock *TopBlock = L->getTopBlock();
MachineBasicBlock *ExitingBlock = L->findLoopControlBlock();
MachineBasicBlock *LoopStart = nullptr;
if (ExitingBlock != L->getLoopLatch()) {
MachineBasicBlock *TB = nullptr, *FB = nullptr;
SmallVector<MachineOperand, 2> Cond;
if (TII->analyzeBranch(*ExitingBlock, TB, FB, Cond, false))
return false;
if (L->contains(TB))
LoopStart = TB;
else if (L->contains(FB))
LoopStart = FB;
else
return false;
}
else
LoopStart = TopBlock;
// Convert the loop to a hardware loop.
LLVM_DEBUG(dbgs() << "Change to hardware loop at "; L->dump());
DebugLoc DL;
if (InsertPos != Preheader->end())
DL = InsertPos->getDebugLoc();
if (TripCount->isReg()) {
// Create a copy of the loop count register.
Register CountReg = MRI->createVirtualRegister(&Hexagon::IntRegsRegClass);
BuildMI(*Preheader, InsertPos, DL, TII->get(TargetOpcode::COPY), CountReg)
.addReg(TripCount->getReg(), 0, TripCount->getSubReg());
// Add the Loop instruction to the beginning of the loop.
BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_r)).addMBB(LoopStart)
.addReg(CountReg);
} else {
assert(TripCount->isImm() && "Expecting immediate value for trip count");
// Add the Loop immediate instruction to the beginning of the loop,
// if the immediate fits in the instructions. Otherwise, we need to
// create a new virtual register.
int64_t CountImm = TripCount->getImm();
if (!TII->isValidOffset(LOOP_i, CountImm, TRI)) {
Register CountReg = MRI->createVirtualRegister(&Hexagon::IntRegsRegClass);
BuildMI(*Preheader, InsertPos, DL, TII->get(Hexagon::A2_tfrsi), CountReg)
.addImm(CountImm);
BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_r))
.addMBB(LoopStart).addReg(CountReg);
} else
BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_i))
.addMBB(LoopStart).addImm(CountImm);
}
// Make sure the loop start always has a reference in the CFG.
LoopStart->setMachineBlockAddressTaken();
// Replace the loop branch with an endloop instruction.
DebugLoc LastIDL = LastI->getDebugLoc();
BuildMI(*LastMBB, LastI, LastIDL, TII->get(ENDLOOP)).addMBB(LoopStart);
// The loop ends with either:
// - a conditional branch followed by an unconditional branch, or
// - a conditional branch to the loop start.
if (LastI->getOpcode() == Hexagon::J2_jumpt ||
LastI->getOpcode() == Hexagon::J2_jumpf) {
// Delete one and change/add an uncond. branch to out of the loop.
MachineBasicBlock *BranchTarget = LastI->getOperand(1).getMBB();
LastI = LastMBB->erase(LastI);
if (!L->contains(BranchTarget)) {
if (LastI != LastMBB->end())
LastI = LastMBB->erase(LastI);
SmallVector<MachineOperand, 0> Cond;
TII->insertBranch(*LastMBB, BranchTarget, nullptr, Cond, LastIDL);
}
} else {
// Conditional branch to loop start; just delete it.
LastMBB->erase(LastI);
}
delete TripCount;
// The induction operation and the comparison may now be
// unneeded. If these are unneeded, then remove them.
for (unsigned i = 0; i < OldInsts.size(); ++i)
removeIfDead(OldInsts[i]);
++NumHWLoops;
// Set RecL1used and RecL0used only after hardware loop has been
// successfully generated. Doing it earlier can cause wrong loop instruction
// to be used.
if (L0Used) // Loop0 was already used. So, the correct loop must be loop1.
RecL1used = true;
else
RecL0used = true;
return true;
}
bool HexagonHardwareLoops::orderBumpCompare(MachineInstr *BumpI,
MachineInstr *CmpI) {
assert (BumpI != CmpI && "Bump and compare in the same instruction?");
MachineBasicBlock *BB = BumpI->getParent();
if (CmpI->getParent() != BB)
return false;
using instr_iterator = MachineBasicBlock::instr_iterator;
// Check if things are in order to begin with.
for (instr_iterator I(BumpI), E = BB->instr_end(); I != E; ++I)
if (&*I == CmpI)
return true;
// Out of order.
Register PredR = CmpI->getOperand(0).getReg();
bool FoundBump = false;
instr_iterator CmpIt = CmpI->getIterator(), NextIt = std::next(CmpIt);
for (instr_iterator I = NextIt, E = BB->instr_end(); I != E; ++I) {
MachineInstr *In = &*I;
for (unsigned i = 0, n = In->getNumOperands(); i < n; ++i) {
MachineOperand &MO = In->getOperand(i);
if (MO.isReg() && MO.isUse()) {
if (MO.getReg() == PredR) // Found an intervening use of PredR.
return false;
}
}
if (In == BumpI) {
BB->splice(++BumpI->getIterator(), BB, CmpI->getIterator());
FoundBump = true;
break;
}
}
assert (FoundBump && "Cannot determine instruction order");
return FoundBump;
}
/// This function is required to break recursion. Visiting phis in a loop may
/// result in recursion during compilation. We break the recursion by making
/// sure that we visit a MachineOperand and its definition in a
/// MachineInstruction only once. If we attempt to visit more than once, then
/// there is recursion, and will return false.
bool HexagonHardwareLoops::isLoopFeeder(MachineLoop *L, MachineBasicBlock *A,
MachineInstr *MI,
const MachineOperand *MO,
LoopFeederMap &LoopFeederPhi) const {
if (LoopFeederPhi.find(MO->getReg()) == LoopFeederPhi.end()) {
LLVM_DEBUG(dbgs() << "\nhw_loop head, "
<< printMBBReference(**L->block_begin()));
// Ignore all BBs that form Loop.
if (llvm::is_contained(L->getBlocks(), A))
return false;
MachineInstr *Def = MRI->getVRegDef(MO->getReg());
LoopFeederPhi.insert(std::make_pair(MO->getReg(), Def));
return true;
} else
// Already visited node.
return false;
}
/// Return true if a Phi may generate a value that can underflow.
/// This function calls loopCountMayWrapOrUnderFlow for each Phi operand.
bool HexagonHardwareLoops::phiMayWrapOrUnderflow(
MachineInstr *Phi, const MachineOperand *EndVal, MachineBasicBlock *MBB,
MachineLoop *L, LoopFeederMap &LoopFeederPhi) const {
assert(Phi->isPHI() && "Expecting a Phi.");
// Walk through each Phi, and its used operands. Make sure that
// if there is recursion in Phi, we won't generate hardware loops.
for (int i = 1, n = Phi->getNumOperands(); i < n; i += 2)
if (isLoopFeeder(L, MBB, Phi, &(Phi->getOperand(i)), LoopFeederPhi))
if (loopCountMayWrapOrUnderFlow(&(Phi->getOperand(i)), EndVal,
Phi->getParent(), L, LoopFeederPhi))
return true;
return false;
}
/// Return true if the induction variable can underflow in the first iteration.
/// An example, is an initial unsigned value that is 0 and is decrement in the
/// first itertion of a do-while loop. In this case, we cannot generate a
/// hardware loop because the endloop instruction does not decrement the loop
/// counter if it is <= 1. We only need to perform this analysis if the
/// initial value is a register.
///
/// This function assumes the initial value may underfow unless proven
/// otherwise. If the type is signed, then we don't care because signed
/// underflow is undefined. We attempt to prove the initial value is not
/// zero by perfoming a crude analysis of the loop counter. This function
/// checks if the initial value is used in any comparison prior to the loop
/// and, if so, assumes the comparison is a range check. This is inexact,
/// but will catch the simple cases.
bool HexagonHardwareLoops::loopCountMayWrapOrUnderFlow(
const MachineOperand *InitVal, const MachineOperand *EndVal,
MachineBasicBlock *MBB, MachineLoop *L,
LoopFeederMap &LoopFeederPhi) const {
// Only check register values since they are unknown.
if (!InitVal->isReg())
return false;
if (!EndVal->isImm())
return false;
// A register value that is assigned an immediate is a known value, and it
// won't underflow in the first iteration.
int64_t Imm;
if (checkForImmediate(*InitVal, Imm))
return (EndVal->getImm() == Imm);
Register Reg = InitVal->getReg();
// We don't know the value of a physical register.
if (!Reg.isVirtual())
return true;
MachineInstr *Def = MRI->getVRegDef(Reg);
if (!Def)
return true;
// If the initial value is a Phi or copy and the operands may not underflow,
// then the definition cannot be underflow either.
if (Def->isPHI() && !phiMayWrapOrUnderflow(Def, EndVal, Def->getParent(),
L, LoopFeederPhi))
return false;
if (Def->isCopy() && !loopCountMayWrapOrUnderFlow(&(Def->getOperand(1)),
EndVal, Def->getParent(),
L, LoopFeederPhi))
return false;
// Iterate over the uses of the initial value. If the initial value is used
// in a compare, then we assume this is a range check that ensures the loop
// doesn't underflow. This is not an exact test and should be improved.
for (MachineRegisterInfo::use_instr_nodbg_iterator I = MRI->use_instr_nodbg_begin(Reg),
E = MRI->use_instr_nodbg_end(); I != E; ++I) {
MachineInstr *MI = &*I;
Register CmpReg1, CmpReg2;
int64_t CmpMask = 0, CmpValue = 0;
if (!TII->analyzeCompare(*MI, CmpReg1, CmpReg2, CmpMask, CmpValue))
continue;
MachineBasicBlock *TBB = nullptr, *FBB = nullptr;
SmallVector<MachineOperand, 2> Cond;
if (TII->analyzeBranch(*MI->getParent(), TBB, FBB, Cond, false))
continue;
Comparison::Kind Cmp =
getComparisonKind(MI->getOpcode(), nullptr, nullptr, 0);
if (Cmp == 0)
continue;
if (TII->predOpcodeHasNot(Cond) ^ (TBB != MBB))
Cmp = Comparison::getNegatedComparison(Cmp);
if (CmpReg2 != 0 && CmpReg2 == Reg)
Cmp = Comparison::getSwappedComparison(Cmp);
// Signed underflow is undefined.
if (Comparison::isSigned(Cmp))
return false;
// Check if there is a comparison of the initial value. If the initial value
// is greater than or not equal to another value, then assume this is a
// range check.
if ((Cmp & Comparison::G) || Cmp == Comparison::NE)
return false;
}
// OK - this is a hack that needs to be improved. We really need to analyze
// the instructions performed on the initial value. This works on the simplest
// cases only.
if (!Def->isCopy() && !Def->isPHI())
return false;
return true;
}
bool HexagonHardwareLoops::checkForImmediate(const MachineOperand &MO,
int64_t &Val) const {
if (MO.isImm()) {
Val = MO.getImm();
return true;
}
if (!MO.isReg())
return false;
// MO is a register. Check whether it is defined as an immediate value,
// and if so, get the value of it in TV. That value will then need to be
// processed to handle potential subregisters in MO.
int64_t TV;
Register R = MO.getReg();
if (!R.isVirtual())
return false;
MachineInstr *DI = MRI->getVRegDef(R);
unsigned DOpc = DI->getOpcode();
switch (DOpc) {
case TargetOpcode::COPY:
case Hexagon::A2_tfrsi:
case Hexagon::A2_tfrpi:
case Hexagon::CONST32:
case Hexagon::CONST64:
// Call recursively to avoid an extra check whether operand(1) is
// indeed an immediate (it could be a global address, for example),
// plus we can handle COPY at the same time.
if (!checkForImmediate(DI->getOperand(1), TV))
return false;
break;
case Hexagon::A2_combineii:
case Hexagon::A4_combineir:
case Hexagon::A4_combineii:
case Hexagon::A4_combineri:
case Hexagon::A2_combinew: {
const MachineOperand &S1 = DI->getOperand(1);
const MachineOperand &S2 = DI->getOperand(2);
int64_t V1, V2;
if (!checkForImmediate(S1, V1) || !checkForImmediate(S2, V2))
return false;
TV = V2 | (static_cast<uint64_t>(V1) << 32);
break;
}
case TargetOpcode::REG_SEQUENCE: {
const MachineOperand &S1 = DI->getOperand(1);
const MachineOperand &S3 = DI->getOperand(3);
int64_t V1, V3;
if (!checkForImmediate(S1, V1) || !checkForImmediate(S3, V3))
return false;
unsigned Sub2 = DI->getOperand(2).getImm();
unsigned Sub4 = DI->getOperand(4).getImm();
if (Sub2 == Hexagon::isub_lo && Sub4 == Hexagon::isub_hi)
TV = V1 | (V3 << 32);
else if (Sub2 == Hexagon::isub_hi && Sub4 == Hexagon::isub_lo)
TV = V3 | (V1 << 32);
else
llvm_unreachable("Unexpected form of REG_SEQUENCE");
break;
}
default:
return false;
}
// By now, we should have successfully obtained the immediate value defining
// the register referenced in MO. Handle a potential use of a subregister.
switch (MO.getSubReg()) {
case Hexagon::isub_lo:
Val = TV & 0xFFFFFFFFULL;
break;
case Hexagon::isub_hi:
Val = (TV >> 32) & 0xFFFFFFFFULL;
break;
default:
Val = TV;
break;
}
return true;
}
void HexagonHardwareLoops::setImmediate(MachineOperand &MO, int64_t Val) {
if (MO.isImm()) {
MO.setImm(Val);
return;
}
assert(MO.isReg());
Register R = MO.getReg();
MachineInstr *DI = MRI->getVRegDef(R);
const TargetRegisterClass *RC = MRI->getRegClass(R);
Register NewR = MRI->createVirtualRegister(RC);
MachineBasicBlock &B = *DI->getParent();
DebugLoc DL = DI->getDebugLoc();
BuildMI(B, DI, DL, TII->get(DI->getOpcode()), NewR).addImm(Val);
MO.setReg(NewR);
}
bool HexagonHardwareLoops::fixupInductionVariable(MachineLoop *L) {
MachineBasicBlock *Header = L->getHeader();
MachineBasicBlock *Latch = L->getLoopLatch();
MachineBasicBlock *ExitingBlock = L->findLoopControlBlock();
if (!(Header && Latch && ExitingBlock))
return false;
// These data structures follow the same concept as the corresponding
// ones in findInductionRegister (where some comments are).
using RegisterBump = std::pair<Register, int64_t>;
using RegisterInduction = std::pair<Register, RegisterBump>;
using RegisterInductionSet = std::set<RegisterInduction>;
// Register candidates for induction variables, with their associated bumps.
RegisterInductionSet IndRegs;
// Look for induction patterns:
// %1 = PHI ..., [ latch, %2 ]
// %2 = ADD %1, imm
using instr_iterator = MachineBasicBlock::instr_iterator;
for (instr_iterator I = Header->instr_begin(), E = Header->instr_end();
I != E && I->isPHI(); ++I) {
MachineInstr *Phi = &*I;
// Have a PHI instruction.
for (unsigned i = 1, n = Phi->getNumOperands(); i < n; i += 2) {
if (Phi->getOperand(i+1).getMBB() != Latch)
continue;
Register PhiReg = Phi->getOperand(i).getReg();
MachineInstr *DI = MRI->getVRegDef(PhiReg);
if (DI->getDesc().isAdd()) {
// If the register operand to the add/sub is the PHI we are looking
// at, this meets the induction pattern.
Register IndReg = DI->getOperand(1).getReg();
MachineOperand &Opnd2 = DI->getOperand(2);
int64_t V;
if (MRI->getVRegDef(IndReg) == Phi && checkForImmediate(Opnd2, V)) {
Register UpdReg = DI->getOperand(0).getReg();
IndRegs.insert(std::make_pair(UpdReg, std::make_pair(IndReg, V)));
}
}
} // for (i)
} // for (instr)
if (IndRegs.empty())
return false;
MachineBasicBlock *TB = nullptr, *FB = nullptr;
SmallVector<MachineOperand,2> Cond;
// analyzeBranch returns true if it fails to analyze branch.
bool NotAnalyzed = TII->analyzeBranch(*ExitingBlock, TB, FB, Cond, false);
if (NotAnalyzed || Cond.empty())
return false;
if (ExitingBlock != Latch && (TB == Latch || FB == Latch)) {
MachineBasicBlock *LTB = nullptr, *LFB = nullptr;
SmallVector<MachineOperand,2> LCond;
bool NotAnalyzed = TII->analyzeBranch(*Latch, LTB, LFB, LCond, false);
if (NotAnalyzed)
return false;
// Since latch is not the exiting block, the latch branch should be an
// unconditional branch to the loop header.
if (TB == Latch)
TB = (LTB == Header) ? LTB : LFB;
else
FB = (LTB == Header) ? LTB : LFB;
}
if (TB != Header) {
if (FB != Header) {
// The latch/exit block does not go back to the header.
return false;
}
// FB is the header (i.e., uncond. jump to branch header)
// In this case, the LoopBody -> TB should not be a back edge otherwise
// it could result in an infinite loop after conversion to hw_loop.
// This case can happen when the Latch has two jumps like this:
// Jmp_c OuterLoopHeader <-- TB
// Jmp InnerLoopHeader <-- FB
if (MDT->dominates(TB, FB))
return false;
}
// Expecting a predicate register as a condition. It won't be a hardware
// predicate register at this point yet, just a vreg.
// HexagonInstrInfo::analyzeBranch for negated branches inserts imm(0)
// into Cond, followed by the predicate register. For non-negated branches
// it's just the register.
unsigned CSz = Cond.size();
if (CSz != 1 && CSz != 2)
return false;
if (!Cond[CSz-1].isReg())
return false;
Register P = Cond[CSz - 1].getReg();
MachineInstr *PredDef = MRI->getVRegDef(P);
if (!PredDef->isCompare())
return false;
SmallSet<Register,2> CmpRegs;
MachineOperand *CmpImmOp = nullptr;
// Go over all operands to the compare and look for immediate and register
// operands. Assume that if the compare has a single register use and a
// single immediate operand, then the register is being compared with the
// immediate value.
for (unsigned i = 0, n = PredDef->getNumOperands(); i < n; ++i) {
MachineOperand &MO = PredDef->getOperand(i);
if (MO.isReg()) {
// Skip all implicit references. In one case there was:
// %140 = FCMPUGT32_rr %138, %139, implicit %usr
if (MO.isImplicit())
continue;
if (MO.isUse()) {
if (!isImmediate(MO)) {
CmpRegs.insert(MO.getReg());
continue;
}
// Consider the register to be the "immediate" operand.
if (CmpImmOp)
return false;
CmpImmOp = &MO;
}
} else if (MO.isImm()) {
if (CmpImmOp) // A second immediate argument? Confusing. Bail out.
return false;
CmpImmOp = &MO;
}
}
if (CmpRegs.empty())
return false;
// Check if the compared register follows the order we want. Fix if needed.
for (RegisterInductionSet::iterator I = IndRegs.begin(), E = IndRegs.end();
I != E; ++I) {
// This is a success. If the register used in the comparison is one that
// we have identified as a bumped (updated) induction register, there is
// nothing to do.
if (CmpRegs.count(I->first))
return true;
// Otherwise, if the register being compared comes out of a PHI node,
// and has been recognized as following the induction pattern, and is
// compared against an immediate, we can fix it.
const RegisterBump &RB = I->second;
if (CmpRegs.count(RB.first)) {
if (!CmpImmOp) {
// If both operands to the compare instruction are registers, see if
// it can be changed to use induction register as one of the operands.
MachineInstr *IndI = nullptr;
MachineInstr *nonIndI = nullptr;
MachineOperand *IndMO = nullptr;
MachineOperand *nonIndMO = nullptr;
for (unsigned i = 1, n = PredDef->getNumOperands(); i < n; ++i) {
MachineOperand &MO = PredDef->getOperand(i);
if (MO.isReg() && MO.getReg() == RB.first) {
LLVM_DEBUG(dbgs() << "\n DefMI(" << i
<< ") = " << *(MRI->getVRegDef(I->first)));
if (IndI)
return false;
IndI = MRI->getVRegDef(I->first);
IndMO = &MO;
} else if (MO.isReg()) {
LLVM_DEBUG(dbgs() << "\n DefMI(" << i
<< ") = " << *(MRI->getVRegDef(MO.getReg())));
if (nonIndI)
return false;
nonIndI = MRI->getVRegDef(MO.getReg());
nonIndMO = &MO;
}
}
if (IndI && nonIndI &&
nonIndI->getOpcode() == Hexagon::A2_addi &&
nonIndI->getOperand(2).isImm() &&
nonIndI->getOperand(2).getImm() == - RB.second) {
bool Order = orderBumpCompare(IndI, PredDef);
if (Order) {
IndMO->setReg(I->first);
nonIndMO->setReg(nonIndI->getOperand(1).getReg());
return true;
}
}
return false;
}
// It is not valid to do this transformation on an unsigned comparison
// because it may underflow.
Comparison::Kind Cmp =
getComparisonKind(PredDef->getOpcode(), nullptr, nullptr, 0);
if (!Cmp || Comparison::isUnsigned(Cmp))
return false;
// If the register is being compared against an immediate, try changing
// the compare instruction to use induction register and adjust the
// immediate operand.
int64_t CmpImm = getImmediate(*CmpImmOp);
int64_t V = RB.second;
// Handle Overflow (64-bit).
if (((V > 0) && (CmpImm > INT64_MAX - V)) ||
((V < 0) && (CmpImm < INT64_MIN - V)))
return false;
CmpImm += V;
// Most comparisons of register against an immediate value allow
// the immediate to be constant-extended. There are some exceptions
// though. Make sure the new combination will work.
if (CmpImmOp->isImm() && !TII->isExtendable(*PredDef) &&
!TII->isValidOffset(PredDef->getOpcode(), CmpImm, TRI, false))
return false;
// Make sure that the compare happens after the bump. Otherwise,
// after the fixup, the compare would use a yet-undefined register.
MachineInstr *BumpI = MRI->getVRegDef(I->first);
bool Order = orderBumpCompare(BumpI, PredDef);
if (!Order)
return false;
// Finally, fix the compare instruction.
setImmediate(*CmpImmOp, CmpImm);
for (unsigned i = 0, n = PredDef->getNumOperands(); i < n; ++i) {
MachineOperand &MO = PredDef->getOperand(i);
if (MO.isReg() && MO.getReg() == RB.first) {
MO.setReg(I->first);
return true;
}
}
}
}
return false;
}
/// createPreheaderForLoop - Create a preheader for a given loop.
MachineBasicBlock *HexagonHardwareLoops::createPreheaderForLoop(
MachineLoop *L) {
if (MachineBasicBlock *TmpPH = MLI->findLoopPreheader(L, SpecPreheader))
return TmpPH;
if (!HWCreatePreheader)
return nullptr;
MachineBasicBlock *Header = L->getHeader();
MachineBasicBlock *Latch = L->getLoopLatch();
MachineBasicBlock *ExitingBlock = L->findLoopControlBlock();
MachineFunction *MF = Header->getParent();
DebugLoc DL;
#ifndef NDEBUG
if ((!PHFn.empty()) && (PHFn != MF->getName()))
return nullptr;
#endif
if (!Latch || !ExitingBlock || Header->hasAddressTaken())
return nullptr;
using instr_iterator = MachineBasicBlock::instr_iterator;
// Verify that all existing predecessors have analyzable branches
// (or no branches at all).
using MBBVector = std::vector<MachineBasicBlock *>;
MBBVector Preds(Header->pred_begin(), Header->pred_end());
SmallVector<MachineOperand,2> Tmp1;
MachineBasicBlock *TB = nullptr, *FB = nullptr;
if (TII->analyzeBranch(*ExitingBlock, TB, FB, Tmp1, false))
return nullptr;
for (MachineBasicBlock *PB : Preds) {
bool NotAnalyzed = TII->analyzeBranch(*PB, TB, FB, Tmp1, false);
if (NotAnalyzed)
return nullptr;
}
MachineBasicBlock *NewPH = MF->CreateMachineBasicBlock();
MF->insert(Header->getIterator(), NewPH);
if (Header->pred_size() > 2) {
// Ensure that the header has only two predecessors: the preheader and
// the loop latch. Any additional predecessors of the header should
// join at the newly created preheader. Inspect all PHI nodes from the
// header and create appropriate corresponding PHI nodes in the preheader.
for (instr_iterator I = Header->instr_begin(), E = Header->instr_end();
I != E && I->isPHI(); ++I) {
MachineInstr *PN = &*I;
const MCInstrDesc &PD = TII->get(TargetOpcode::PHI);
MachineInstr *NewPN = MF->CreateMachineInstr(PD, DL);
NewPH->insert(NewPH->end(), NewPN);
Register PR = PN->getOperand(0).getReg();
const TargetRegisterClass *RC = MRI->getRegClass(PR);
Register NewPR = MRI->createVirtualRegister(RC);
NewPN->addOperand(MachineOperand::CreateReg(NewPR, true));
// Copy all non-latch operands of a header's PHI node to the newly
// created PHI node in the preheader.
for (unsigned i = 1, n = PN->getNumOperands(); i < n; i += 2) {
Register PredR = PN->getOperand(i).getReg();
unsigned PredRSub = PN->getOperand(i).getSubReg();
MachineBasicBlock *PredB = PN->getOperand(i+1).getMBB();
if (PredB == Latch)
continue;
MachineOperand MO = MachineOperand::CreateReg(PredR, false);
MO.setSubReg(PredRSub);
NewPN->addOperand(MO);
NewPN->addOperand(MachineOperand::CreateMBB(PredB));
}
// Remove copied operands from the old PHI node and add the value
// coming from the preheader's PHI.
for (int i = PN->getNumOperands()-2; i > 0; i -= 2) {
MachineBasicBlock *PredB = PN->getOperand(i+1).getMBB();
if (PredB != Latch) {
PN->removeOperand(i+1);
PN->removeOperand(i);
}
}
PN->addOperand(MachineOperand::CreateReg(NewPR, false));
PN->addOperand(MachineOperand::CreateMBB(NewPH));
}
} else {
assert(Header->pred_size() == 2);
// The header has only two predecessors, but the non-latch predecessor
// is not a preheader (e.g. it has other successors, etc.)
// In such a case we don't need any extra PHI nodes in the new preheader,
// all we need is to adjust existing PHIs in the header to now refer to
// the new preheader.
for (instr_iterator I = Header->instr_begin(), E = Header->instr_end();
I != E && I->isPHI(); ++I) {
MachineInstr *PN = &*I;
for (unsigned i = 1, n = PN->getNumOperands(); i < n; i += 2) {
MachineOperand &MO = PN->getOperand(i+1);
if (MO.getMBB() != Latch)
MO.setMBB(NewPH);
}
}
}
// "Reroute" the CFG edges to link in the new preheader.
// If any of the predecessors falls through to the header, insert a branch
// to the new preheader in that place.
SmallVector<MachineOperand,1> Tmp2;
SmallVector<MachineOperand,1> EmptyCond;
TB = FB = nullptr;
for (MachineBasicBlock *PB : Preds) {
if (PB != Latch) {
Tmp2.clear();
bool NotAnalyzed = TII->analyzeBranch(*PB, TB, FB, Tmp2, false);
(void)NotAnalyzed; // suppress compiler warning
assert (!NotAnalyzed && "Should be analyzable!");
if (TB != Header && (Tmp2.empty() || FB != Header))
TII->insertBranch(*PB, NewPH, nullptr, EmptyCond, DL);
PB->ReplaceUsesOfBlockWith(Header, NewPH);
}
}
// It can happen that the latch block will fall through into the header.
// Insert an unconditional branch to the header.
TB = FB = nullptr;
bool LatchNotAnalyzed = TII->analyzeBranch(*Latch, TB, FB, Tmp2, false);
(void)LatchNotAnalyzed; // suppress compiler warning
assert (!LatchNotAnalyzed && "Should be analyzable!");
if (!TB && !FB)
TII->insertBranch(*Latch, Header, nullptr, EmptyCond, DL);
// Finally, the branch from the preheader to the header.
TII->insertBranch(*NewPH, Header, nullptr, EmptyCond, DL);
NewPH->addSuccessor(Header);
MachineLoop *ParentLoop = L->getParentLoop();
if (ParentLoop)
ParentLoop->addBasicBlockToLoop(NewPH, MLI->getBase());
// Update the dominator information with the new preheader.
if (MDT) {
if (MachineDomTreeNode *HN = MDT->getNode(Header)) {
if (MachineDomTreeNode *DHN = HN->getIDom()) {
MDT->addNewBlock(NewPH, DHN->getBlock());
MDT->changeImmediateDominator(Header, NewPH);
}
}
}
return NewPH;
}