Google Git
Sign in
swiftshader / SwiftShader / refs/heads/llvm10-clean / . / third_party / llvm-10.0 / llvm / lib / CodeGen / SelectionDAG / LegalizeDAG.cpp
blob: 91404ee7728b85f66d3588c03b240ba5660cd341 [file] [log] [blame] [edit]
//===- LegalizeDAG.cpp - Implement SelectionDAG::Legalize -----------------===//
//
// 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 file implements the SelectionDAG::Legalize method.
//
//===----------------------------------------------------------------------===//
#include "llvm/ADT/APFloat.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/RuntimeLibcalls.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SelectionDAGNodes.h"
#include "llvm/CodeGen/TargetFrameLowering.h"
#include "llvm/CodeGen/TargetLowering.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Type.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MachineValueType.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <tuple>
#include <utility>
using namespace llvm;
#define DEBUG_TYPE "legalizedag"
namespace {
/// Keeps track of state when getting the sign of a floating-point value as an
/// integer.
struct FloatSignAsInt {
EVT FloatVT;
SDValue Chain;
SDValue FloatPtr;
SDValue IntPtr;
MachinePointerInfo IntPointerInfo;
MachinePointerInfo FloatPointerInfo;
SDValue IntValue;
APInt SignMask;
uint8_t SignBit;
};
//===----------------------------------------------------------------------===//
/// This takes an arbitrary SelectionDAG as input and
/// hacks on it until the target machine can handle it. This involves
/// eliminating value sizes the machine cannot handle (promoting small sizes to
/// large sizes or splitting up large values into small values) as well as
/// eliminating operations the machine cannot handle.
///
/// This code also does a small amount of optimization and recognition of idioms
/// as part of its processing. For example, if a target does not support a
/// 'setcc' instruction efficiently, but does support 'brcc' instruction, this
/// will attempt merge setcc and brc instructions into brcc's.
class SelectionDAGLegalize {
const TargetMachine &TM;
const TargetLowering &TLI;
SelectionDAG &DAG;
/// The set of nodes which have already been legalized. We hold a
/// reference to it in order to update as necessary on node deletion.
SmallPtrSetImpl<SDNode *> &LegalizedNodes;
/// A set of all the nodes updated during legalization.
SmallSetVector<SDNode *, 16> *UpdatedNodes;
EVT getSetCCResultType(EVT VT) const {
return TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT);
}
// Libcall insertion helpers.
public:
SelectionDAGLegalize(SelectionDAG &DAG,
SmallPtrSetImpl<SDNode *> &LegalizedNodes,
SmallSetVector<SDNode *, 16> *UpdatedNodes = nullptr)
: TM(DAG.getTarget()), TLI(DAG.getTargetLoweringInfo()), DAG(DAG),
LegalizedNodes(LegalizedNodes), UpdatedNodes(UpdatedNodes) {}
/// Legalizes the given operation.
void LegalizeOp(SDNode *Node);
private:
SDValue OptimizeFloatStore(StoreSDNode *ST);
void LegalizeLoadOps(SDNode *Node);
void LegalizeStoreOps(SDNode *Node);
/// Some targets cannot handle a variable
/// insertion index for the INSERT_VECTOR_ELT instruction. In this case, it
/// is necessary to spill the vector being inserted into to memory, perform
/// the insert there, and then read the result back.
SDValue PerformInsertVectorEltInMemory(SDValue Vec, SDValue Val, SDValue Idx,
const SDLoc &dl);
SDValue ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val, SDValue Idx,
const SDLoc &dl);
/// Return a vector shuffle operation which
/// performs the same shuffe in terms of order or result bytes, but on a type
/// whose vector element type is narrower than the original shuffle type.
/// e.g. <v4i32> <0, 1, 0, 1> -> v8i16 <0, 1, 2, 3, 0, 1, 2, 3>
SDValue ShuffleWithNarrowerEltType(EVT NVT, EVT VT, const SDLoc &dl,
SDValue N1, SDValue N2,
ArrayRef<int> Mask) const;
bool LegalizeSetCCCondCode(EVT VT, SDValue &LHS, SDValue &RHS, SDValue &CC,
bool &NeedInvert, const SDLoc &dl, SDValue &Chain,
bool IsSignaling = false);
SDValue ExpandLibCall(RTLIB::Libcall LC, SDNode *Node, bool isSigned);
void ExpandFPLibCall(SDNode *Node, RTLIB::Libcall Call_F32,
RTLIB::Libcall Call_F64, RTLIB::Libcall Call_F80,
RTLIB::Libcall Call_F128,
RTLIB::Libcall Call_PPCF128,
SmallVectorImpl<SDValue> &Results);
SDValue ExpandIntLibCall(SDNode *Node, bool isSigned,
RTLIB::Libcall Call_I8,
RTLIB::Libcall Call_I16,
RTLIB::Libcall Call_I32,
RTLIB::Libcall Call_I64,
RTLIB::Libcall Call_I128);
void ExpandArgFPLibCall(SDNode *Node,
RTLIB::Libcall Call_F32, RTLIB::Libcall Call_F64,
RTLIB::Libcall Call_F80, RTLIB::Libcall Call_F128,
RTLIB::Libcall Call_PPCF128,
SmallVectorImpl<SDValue> &Results);
void ExpandDivRemLibCall(SDNode *Node, SmallVectorImpl<SDValue> &Results);
void ExpandSinCosLibCall(SDNode *Node, SmallVectorImpl<SDValue> &Results);
SDValue EmitStackConvert(SDValue SrcOp, EVT SlotVT, EVT DestVT,
const SDLoc &dl);
SDValue EmitStackConvert(SDValue SrcOp, EVT SlotVT, EVT DestVT,
const SDLoc &dl, SDValue ChainIn);
SDValue ExpandBUILD_VECTOR(SDNode *Node);
SDValue ExpandSPLAT_VECTOR(SDNode *Node);
SDValue ExpandSCALAR_TO_VECTOR(SDNode *Node);
void ExpandDYNAMIC_STACKALLOC(SDNode *Node,
SmallVectorImpl<SDValue> &Results);
void getSignAsIntValue(FloatSignAsInt &State, const SDLoc &DL,
SDValue Value) const;
SDValue modifySignAsInt(const FloatSignAsInt &State, const SDLoc &DL,
SDValue NewIntValue) const;
SDValue ExpandFCOPYSIGN(SDNode *Node) const;
SDValue ExpandFABS(SDNode *Node) const;
SDValue ExpandLegalINT_TO_FP(SDNode *Node, SDValue &Chain);
void PromoteLegalINT_TO_FP(SDNode *N, const SDLoc &dl,
SmallVectorImpl<SDValue> &Results);
void PromoteLegalFP_TO_INT(SDNode *N, const SDLoc &dl,
SmallVectorImpl<SDValue> &Results);
SDValue ExpandBITREVERSE(SDValue Op, const SDLoc &dl);
SDValue ExpandBSWAP(SDValue Op, const SDLoc &dl);
SDValue ExpandExtractFromVectorThroughStack(SDValue Op);
SDValue ExpandInsertToVectorThroughStack(SDValue Op);
SDValue ExpandVectorBuildThroughStack(SDNode* Node);
SDValue ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP);
SDValue ExpandConstant(ConstantSDNode *CP);
// if ExpandNode returns false, LegalizeOp falls back to ConvertNodeToLibcall
bool ExpandNode(SDNode *Node);
void ConvertNodeToLibcall(SDNode *Node);
void PromoteNode(SDNode *Node);
public:
// Node replacement helpers
void ReplacedNode(SDNode *N) {
LegalizedNodes.erase(N);
if (UpdatedNodes)
UpdatedNodes->insert(N);
}
void ReplaceNode(SDNode *Old, SDNode *New) {
LLVM_DEBUG(dbgs() << " ... replacing: "; Old->dump(&DAG);
dbgs() << " with: "; New->dump(&DAG));
assert(Old->getNumValues() == New->getNumValues() &&
"Replacing one node with another that produces a different number "
"of values!");
DAG.ReplaceAllUsesWith(Old, New);
if (UpdatedNodes)
UpdatedNodes->insert(New);
ReplacedNode(Old);
}
void ReplaceNode(SDValue Old, SDValue New) {
LLVM_DEBUG(dbgs() << " ... replacing: "; Old->dump(&DAG);
dbgs() << " with: "; New->dump(&DAG));
DAG.ReplaceAllUsesWith(Old, New);
if (UpdatedNodes)
UpdatedNodes->insert(New.getNode());
ReplacedNode(Old.getNode());
}
void ReplaceNode(SDNode *Old, const SDValue *New) {
LLVM_DEBUG(dbgs() << " ... replacing: "; Old->dump(&DAG));
DAG.ReplaceAllUsesWith(Old, New);
for (unsigned i = 0, e = Old->getNumValues(); i != e; ++i) {
LLVM_DEBUG(dbgs() << (i == 0 ? " with: " : " and: ");
New[i]->dump(&DAG));
if (UpdatedNodes)
UpdatedNodes->insert(New[i].getNode());
}
ReplacedNode(Old);
}
void ReplaceNodeWithValue(SDValue Old, SDValue New) {
LLVM_DEBUG(dbgs() << " ... replacing: "; Old->dump(&DAG);
dbgs() << " with: "; New->dump(&DAG));
DAG.ReplaceAllUsesOfValueWith(Old, New);
if (UpdatedNodes)
UpdatedNodes->insert(New.getNode());
ReplacedNode(Old.getNode());
}
};
} // end anonymous namespace
/// Return a vector shuffle operation which
/// performs the same shuffle in terms of order or result bytes, but on a type
/// whose vector element type is narrower than the original shuffle type.
/// e.g. <v4i32> <0, 1, 0, 1> -> v8i16 <0, 1, 2, 3, 0, 1, 2, 3>
SDValue SelectionDAGLegalize::ShuffleWithNarrowerEltType(
EVT NVT, EVT VT, const SDLoc &dl, SDValue N1, SDValue N2,
ArrayRef<int> Mask) const {
unsigned NumMaskElts = VT.getVectorNumElements();
unsigned NumDestElts = NVT.getVectorNumElements();
unsigned NumEltsGrowth = NumDestElts / NumMaskElts;
assert(NumEltsGrowth && "Cannot promote to vector type with fewer elts!");
if (NumEltsGrowth == 1)
return DAG.getVectorShuffle(NVT, dl, N1, N2, Mask);
SmallVector<int, 8> NewMask;
for (unsigned i = 0; i != NumMaskElts; ++i) {
int Idx = Mask[i];
for (unsigned j = 0; j != NumEltsGrowth; ++j) {
if (Idx < 0)
NewMask.push_back(-1);
else
NewMask.push_back(Idx * NumEltsGrowth + j);
}
}
assert(NewMask.size() == NumDestElts && "Non-integer NumEltsGrowth?");
assert(TLI.isShuffleMaskLegal(NewMask, NVT) && "Shuffle not legal?");
return DAG.getVectorShuffle(NVT, dl, N1, N2, NewMask);
}
/// Expands the ConstantFP node to an integer constant or
/// a load from the constant pool.
SDValue
SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) {
bool Extend = false;
SDLoc dl(CFP);
// If a FP immediate is precise when represented as a float and if the
// target can do an extending load from float to double, we put it into
// the constant pool as a float, even if it's is statically typed as a
// double. This shrinks FP constants and canonicalizes them for targets where
// an FP extending load is the same cost as a normal load (such as on the x87
// fp stack or PPC FP unit).
EVT VT = CFP->getValueType(0);
ConstantFP *LLVMC = const_cast<ConstantFP*>(CFP->getConstantFPValue());
if (!UseCP) {
assert((VT == MVT::f64 || VT == MVT::f32) && "Invalid type expansion");
return DAG.getConstant(LLVMC->getValueAPF().bitcastToAPInt(), dl,
(VT == MVT::f64) ? MVT::i64 : MVT::i32);
}
APFloat APF = CFP->getValueAPF();
EVT OrigVT = VT;
EVT SVT = VT;
// We don't want to shrink SNaNs. Converting the SNaN back to its real type
// can cause it to be changed into a QNaN on some platforms (e.g. on SystemZ).
if (!APF.isSignaling()) {
while (SVT != MVT::f32 && SVT != MVT::f16) {
SVT = (MVT::SimpleValueType)(SVT.getSimpleVT().SimpleTy - 1);
if (ConstantFPSDNode::isValueValidForType(SVT, APF) &&
// Only do this if the target has a native EXTLOAD instruction from
// smaller type.
TLI.isLoadExtLegal(ISD::EXTLOAD, OrigVT, SVT) &&
TLI.ShouldShrinkFPConstant(OrigVT)) {
Type *SType = SVT.getTypeForEVT(*DAG.getContext());
LLVMC = cast<ConstantFP>(ConstantExpr::getFPTrunc(LLVMC, SType));
VT = SVT;
Extend = true;
}
}
}
SDValue CPIdx =
DAG.getConstantPool(LLVMC, TLI.getPointerTy(DAG.getDataLayout()));
unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
if (Extend) {
SDValue Result = DAG.getExtLoad(
ISD::EXTLOAD, dl, OrigVT, DAG.getEntryNode(), CPIdx,
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()), VT,
Alignment);
return Result;
}
SDValue Result = DAG.getLoad(
OrigVT, dl, DAG.getEntryNode(), CPIdx,
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()), Alignment);
return Result;
}
/// Expands the Constant node to a load from the constant pool.
SDValue SelectionDAGLegalize::ExpandConstant(ConstantSDNode *CP) {
SDLoc dl(CP);
EVT VT = CP->getValueType(0);
SDValue CPIdx = DAG.getConstantPool(CP->getConstantIntValue(),
TLI.getPointerTy(DAG.getDataLayout()));
unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
SDValue Result = DAG.getLoad(
VT, dl, DAG.getEntryNode(), CPIdx,
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()), Alignment);
return Result;
}
/// Some target cannot handle a variable insertion index for the
/// INSERT_VECTOR_ELT instruction. In this case, it
/// is necessary to spill the vector being inserted into to memory, perform
/// the insert there, and then read the result back.
SDValue SelectionDAGLegalize::PerformInsertVectorEltInMemory(SDValue Vec,
SDValue Val,
SDValue Idx,
const SDLoc &dl) {
SDValue Tmp1 = Vec;
SDValue Tmp2 = Val;
SDValue Tmp3 = Idx;
// If the target doesn't support this, we have to spill the input vector
// to a temporary stack slot, update the element, then reload it. This is
// badness. We could also load the value into a vector register (either
// with a "move to register" or "extload into register" instruction, then
// permute it into place, if the idx is a constant and if the idx is
// supported by the target.
EVT VT = Tmp1.getValueType();
EVT EltVT = VT.getVectorElementType();
SDValue StackPtr = DAG.CreateStackTemporary(VT);
int SPFI = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
// Store the vector.
SDValue Ch = DAG.getStore(
DAG.getEntryNode(), dl, Tmp1, StackPtr,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SPFI));
SDValue StackPtr2 = TLI.getVectorElementPointer(DAG, StackPtr, VT, Tmp3);
// Store the scalar value.
Ch = DAG.getTruncStore(Ch, dl, Tmp2, StackPtr2, MachinePointerInfo(), EltVT);
// Load the updated vector.
return DAG.getLoad(VT, dl, Ch, StackPtr, MachinePointerInfo::getFixedStack(
DAG.getMachineFunction(), SPFI));
}
SDValue SelectionDAGLegalize::ExpandINSERT_VECTOR_ELT(SDValue Vec, SDValue Val,
SDValue Idx,
const SDLoc &dl) {
if (ConstantSDNode *InsertPos = dyn_cast<ConstantSDNode>(Idx)) {
// SCALAR_TO_VECTOR requires that the type of the value being inserted
// match the element type of the vector being created, except for
// integers in which case the inserted value can be over width.
EVT EltVT = Vec.getValueType().getVectorElementType();
if (Val.getValueType() == EltVT ||
(EltVT.isInteger() && Val.getValueType().bitsGE(EltVT))) {
SDValue ScVec = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl,
Vec.getValueType(), Val);
unsigned NumElts = Vec.getValueType().getVectorNumElements();
// We generate a shuffle of InVec and ScVec, so the shuffle mask
// should be 0,1,2,3,4,5... with the appropriate element replaced with
// elt 0 of the RHS.
SmallVector<int, 8> ShufOps;
for (unsigned i = 0; i != NumElts; ++i)
ShufOps.push_back(i != InsertPos->getZExtValue() ? i : NumElts);
return DAG.getVectorShuffle(Vec.getValueType(), dl, Vec, ScVec, ShufOps);
}
}
return PerformInsertVectorEltInMemory(Vec, Val, Idx, dl);
}
SDValue SelectionDAGLegalize::OptimizeFloatStore(StoreSDNode* ST) {
if (!ISD::isNormalStore(ST))
return SDValue();
LLVM_DEBUG(dbgs() << "Optimizing float store operations\n");
// Turn 'store float 1.0, Ptr' -> 'store int 0x12345678, Ptr'
// FIXME: We shouldn't do this for TargetConstantFP's.
// FIXME: move this to the DAG Combiner! Note that we can't regress due
// to phase ordering between legalized code and the dag combiner. This
// probably means that we need to integrate dag combiner and legalizer
// together.
// We generally can't do this one for long doubles.
SDValue Chain = ST->getChain();
SDValue Ptr = ST->getBasePtr();
unsigned Alignment = ST->getAlignment();
MachineMemOperand::Flags MMOFlags = ST->getMemOperand()->getFlags();
AAMDNodes AAInfo = ST->getAAInfo();
SDLoc dl(ST);
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(ST->getValue())) {
if (CFP->getValueType(0) == MVT::f32 &&
TLI.isTypeLegal(MVT::i32)) {
SDValue Con = DAG.getConstant(CFP->getValueAPF().
bitcastToAPInt().zextOrTrunc(32),
SDLoc(CFP), MVT::i32);
return DAG.getStore(Chain, dl, Con, Ptr, ST->getPointerInfo(), Alignment,
MMOFlags, AAInfo);
}
if (CFP->getValueType(0) == MVT::f64) {
// If this target supports 64-bit registers, do a single 64-bit store.
if (TLI.isTypeLegal(MVT::i64)) {
SDValue Con = DAG.getConstant(CFP->getValueAPF().bitcastToAPInt().
zextOrTrunc(64), SDLoc(CFP), MVT::i64);
return DAG.getStore(Chain, dl, Con, Ptr, ST->getPointerInfo(),
Alignment, MMOFlags, AAInfo);
}
if (TLI.isTypeLegal(MVT::i32) && !ST->isVolatile()) {
// Otherwise, if the target supports 32-bit registers, use 2 32-bit
// stores. If the target supports neither 32- nor 64-bits, this
// xform is certainly not worth it.
const APInt &IntVal = CFP->getValueAPF().bitcastToAPInt();
SDValue Lo = DAG.getConstant(IntVal.trunc(32), dl, MVT::i32);
SDValue Hi = DAG.getConstant(IntVal.lshr(32).trunc(32), dl, MVT::i32);
if (DAG.getDataLayout().isBigEndian())
std::swap(Lo, Hi);
Lo = DAG.getStore(Chain, dl, Lo, Ptr, ST->getPointerInfo(), Alignment,
MMOFlags, AAInfo);
Ptr = DAG.getMemBasePlusOffset(Ptr, 4, dl);
Hi = DAG.getStore(Chain, dl, Hi, Ptr,
ST->getPointerInfo().getWithOffset(4),
MinAlign(Alignment, 4U), MMOFlags, AAInfo);
return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo, Hi);
}
}
}
return SDValue(nullptr, 0);
}
void SelectionDAGLegalize::LegalizeStoreOps(SDNode *Node) {
StoreSDNode *ST = cast<StoreSDNode>(Node);
SDValue Chain = ST->getChain();
SDValue Ptr = ST->getBasePtr();
SDLoc dl(Node);
unsigned Alignment = ST->getAlignment();
MachineMemOperand::Flags MMOFlags = ST->getMemOperand()->getFlags();
AAMDNodes AAInfo = ST->getAAInfo();
if (!ST->isTruncatingStore()) {
LLVM_DEBUG(dbgs() << "Legalizing store operation\n");
if (SDNode *OptStore = OptimizeFloatStore(ST).getNode()) {
ReplaceNode(ST, OptStore);
return;
}
SDValue Value = ST->getValue();
MVT VT = Value.getSimpleValueType();
switch (TLI.getOperationAction(ISD::STORE, VT)) {
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Legal: {
// If this is an unaligned store and the target doesn't support it,
// expand it.
EVT MemVT = ST->getMemoryVT();
const DataLayout &DL = DAG.getDataLayout();
if (!TLI.allowsMemoryAccessForAlignment(*DAG.getContext(), DL, MemVT,
*ST->getMemOperand())) {
LLVM_DEBUG(dbgs() << "Expanding unsupported unaligned store\n");
SDValue Result = TLI.expandUnalignedStore(ST, DAG);
ReplaceNode(SDValue(ST, 0), Result);
} else
LLVM_DEBUG(dbgs() << "Legal store\n");
break;
}
case TargetLowering::Custom: {
LLVM_DEBUG(dbgs() << "Trying custom lowering\n");
SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
if (Res && Res != SDValue(Node, 0))
ReplaceNode(SDValue(Node, 0), Res);
return;
}
case TargetLowering::Promote: {
MVT NVT = TLI.getTypeToPromoteTo(ISD::STORE, VT);
assert(NVT.getSizeInBits() == VT.getSizeInBits() &&
"Can only promote stores to same size type");
Value = DAG.getNode(ISD::BITCAST, dl, NVT, Value);
SDValue Result =
DAG.getStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
Alignment, MMOFlags, AAInfo);
ReplaceNode(SDValue(Node, 0), Result);
break;
}
}
return;
}
LLVM_DEBUG(dbgs() << "Legalizing truncating store operations\n");
SDValue Value = ST->getValue();
EVT StVT = ST->getMemoryVT();
unsigned StWidth = StVT.getSizeInBits();
auto &DL = DAG.getDataLayout();
if (StWidth != StVT.getStoreSizeInBits()) {
// Promote to a byte-sized store with upper bits zero if not
// storing an integral number of bytes. For example, promote
// TRUNCSTORE:i1 X -> TRUNCSTORE:i8 (and X, 1)
EVT NVT = EVT::getIntegerVT(*DAG.getContext(),
StVT.getStoreSizeInBits());
Value = DAG.getZeroExtendInReg(Value, dl, StVT);
SDValue Result =
DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(), NVT,
Alignment, MMOFlags, AAInfo);
ReplaceNode(SDValue(Node, 0), Result);
} else if (StWidth & (StWidth - 1)) {
// If not storing a power-of-2 number of bits, expand as two stores.
assert(!StVT.isVector() && "Unsupported truncstore!");
unsigned LogStWidth = Log2_32(StWidth);
assert(LogStWidth < 32);
unsigned RoundWidth = 1 << LogStWidth;
assert(RoundWidth < StWidth);
unsigned ExtraWidth = StWidth - RoundWidth;
assert(ExtraWidth < RoundWidth);
assert(!(RoundWidth % 8) && !(ExtraWidth % 8) &&
"Store size not an integral number of bytes!");
EVT RoundVT = EVT::getIntegerVT(*DAG.getContext(), RoundWidth);
EVT ExtraVT = EVT::getIntegerVT(*DAG.getContext(), ExtraWidth);
SDValue Lo, Hi;
unsigned IncrementSize;
if (DL.isLittleEndian()) {
// TRUNCSTORE:i24 X -> TRUNCSTORE:i16 X, TRUNCSTORE@+2:i8 (srl X, 16)
// Store the bottom RoundWidth bits.
Lo = DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
RoundVT, Alignment, MMOFlags, AAInfo);
// Store the remaining ExtraWidth bits.
IncrementSize = RoundWidth / 8;
Ptr = DAG.getMemBasePlusOffset(Ptr, IncrementSize, dl);
Hi = DAG.getNode(
ISD::SRL, dl, Value.getValueType(), Value,
DAG.getConstant(RoundWidth, dl,
TLI.getShiftAmountTy(Value.getValueType(), DL)));
Hi = DAG.getTruncStore(
Chain, dl, Hi, Ptr,
ST->getPointerInfo().getWithOffset(IncrementSize), ExtraVT,
MinAlign(Alignment, IncrementSize), MMOFlags, AAInfo);
} else {
// Big endian - avoid unaligned stores.
// TRUNCSTORE:i24 X -> TRUNCSTORE:i16 (srl X, 8), TRUNCSTORE@+2:i8 X
// Store the top RoundWidth bits.
Hi = DAG.getNode(
ISD::SRL, dl, Value.getValueType(), Value,
DAG.getConstant(ExtraWidth, dl,
TLI.getShiftAmountTy(Value.getValueType(), DL)));
Hi = DAG.getTruncStore(Chain, dl, Hi, Ptr, ST->getPointerInfo(),
RoundVT, Alignment, MMOFlags, AAInfo);
// Store the remaining ExtraWidth bits.
IncrementSize = RoundWidth / 8;
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
DAG.getConstant(IncrementSize, dl,
Ptr.getValueType()));
Lo = DAG.getTruncStore(
Chain, dl, Value, Ptr,
ST->getPointerInfo().getWithOffset(IncrementSize), ExtraVT,
MinAlign(Alignment, IncrementSize), MMOFlags, AAInfo);
}
// The order of the stores doesn't matter.
SDValue Result = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo, Hi);
ReplaceNode(SDValue(Node, 0), Result);
} else {
switch (TLI.getTruncStoreAction(ST->getValue().getValueType(), StVT)) {
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Legal: {
EVT MemVT = ST->getMemoryVT();
// If this is an unaligned store and the target doesn't support it,
// expand it.
if (!TLI.allowsMemoryAccessForAlignment(*DAG.getContext(), DL, MemVT,
*ST->getMemOperand())) {
SDValue Result = TLI.expandUnalignedStore(ST, DAG);
ReplaceNode(SDValue(ST, 0), Result);
}
break;
}
case TargetLowering::Custom: {
SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG);
if (Res && Res != SDValue(Node, 0))
ReplaceNode(SDValue(Node, 0), Res);
return;
}
case TargetLowering::Expand:
assert(!StVT.isVector() &&
"Vector Stores are handled in LegalizeVectorOps");
SDValue Result;
// TRUNCSTORE:i16 i32 -> STORE i16
if (TLI.isTypeLegal(StVT)) {
Value = DAG.getNode(ISD::TRUNCATE, dl, StVT, Value);
Result = DAG.getStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
Alignment, MMOFlags, AAInfo);
} else {
// The in-memory type isn't legal. Truncate to the type it would promote
// to, and then do a truncstore.
Value = DAG.getNode(ISD::TRUNCATE, dl,
TLI.getTypeToTransformTo(*DAG.getContext(), StVT),
Value);
Result = DAG.getTruncStore(Chain, dl, Value, Ptr, ST->getPointerInfo(),
StVT, Alignment, MMOFlags, AAInfo);
}
ReplaceNode(SDValue(Node, 0), Result);
break;
}
}
}
void SelectionDAGLegalize::LegalizeLoadOps(SDNode *Node) {
LoadSDNode *LD = cast<LoadSDNode>(Node);
SDValue Chain = LD->getChain(); // The chain.
SDValue Ptr = LD->getBasePtr(); // The base pointer.
SDValue Value; // The value returned by the load op.
SDLoc dl(Node);
ISD::LoadExtType ExtType = LD->getExtensionType();
if (ExtType == ISD::NON_EXTLOAD) {
LLVM_DEBUG(dbgs() << "Legalizing non-extending load operation\n");
MVT VT = Node->getSimpleValueType(0);
SDValue RVal = SDValue(Node, 0);
SDValue RChain = SDValue(Node, 1);
switch (TLI.getOperationAction(Node->getOpcode(), VT)) {
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Legal: {
EVT MemVT = LD->getMemoryVT();
const DataLayout &DL = DAG.getDataLayout();
// If this is an unaligned load and the target doesn't support it,
// expand it.
if (!TLI.allowsMemoryAccessForAlignment(*DAG.getContext(), DL, MemVT,
*LD->getMemOperand())) {
std::tie(RVal, RChain) = TLI.expandUnalignedLoad(LD, DAG);
}
break;
}
case TargetLowering::Custom:
if (SDValue Res = TLI.LowerOperation(RVal, DAG)) {
RVal = Res;
RChain = Res.getValue(1);
}
break;
case TargetLowering::Promote: {
MVT NVT = TLI.getTypeToPromoteTo(Node->getOpcode(), VT);
assert(NVT.getSizeInBits() == VT.getSizeInBits() &&
"Can only promote loads to same size type");
SDValue Res = DAG.getLoad(NVT, dl, Chain, Ptr, LD->getMemOperand());
RVal = DAG.getNode(ISD::BITCAST, dl, VT, Res);
RChain = Res.getValue(1);
break;
}
}
if (RChain.getNode() != Node) {
assert(RVal.getNode() != Node && "Load must be completely replaced");
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), RVal);
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), RChain);
if (UpdatedNodes) {
UpdatedNodes->insert(RVal.getNode());
UpdatedNodes->insert(RChain.getNode());
}
ReplacedNode(Node);
}
return;
}
LLVM_DEBUG(dbgs() << "Legalizing extending load operation\n");
EVT SrcVT = LD->getMemoryVT();
unsigned SrcWidth = SrcVT.getSizeInBits();
unsigned Alignment = LD->getAlignment();
MachineMemOperand::Flags MMOFlags = LD->getMemOperand()->getFlags();
AAMDNodes AAInfo = LD->getAAInfo();
if (SrcWidth != SrcVT.getStoreSizeInBits() &&
// Some targets pretend to have an i1 loading operation, and actually
// load an i8. This trick is correct for ZEXTLOAD because the top 7
// bits are guaranteed to be zero; it helps the optimizers understand
// that these bits are zero. It is also useful for EXTLOAD, since it
// tells the optimizers that those bits are undefined. It would be
// nice to have an effective generic way of getting these benefits...
// Until such a way is found, don't insist on promoting i1 here.
(SrcVT != MVT::i1 ||
TLI.getLoadExtAction(ExtType, Node->getValueType(0), MVT::i1) ==
TargetLowering::Promote)) {
// Promote to a byte-sized load if not loading an integral number of
// bytes. For example, promote EXTLOAD:i20 -> EXTLOAD:i24.
unsigned NewWidth = SrcVT.getStoreSizeInBits();
EVT NVT = EVT::getIntegerVT(*DAG.getContext(), NewWidth);
SDValue Ch;
// The extra bits are guaranteed to be zero, since we stored them that
// way. A zext load from NVT thus automatically gives zext from SrcVT.
ISD::LoadExtType NewExtType =
ExtType == ISD::ZEXTLOAD ? ISD::ZEXTLOAD : ISD::EXTLOAD;
SDValue Result =
DAG.getExtLoad(NewExtType, dl, Node->getValueType(0), Chain, Ptr,
LD->getPointerInfo(), NVT, Alignment, MMOFlags, AAInfo);
Ch = Result.getValue(1); // The chain.
if (ExtType == ISD::SEXTLOAD)
// Having the top bits zero doesn't help when sign extending.
Result = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl,
Result.getValueType(),
Result, DAG.getValueType(SrcVT));
else if (ExtType == ISD::ZEXTLOAD || NVT == Result.getValueType())
// All the top bits are guaranteed to be zero - inform the optimizers.
Result = DAG.getNode(ISD::AssertZext, dl,
Result.getValueType(), Result,
DAG.getValueType(SrcVT));
Value = Result;
Chain = Ch;
} else if (SrcWidth & (SrcWidth - 1)) {
// If not loading a power-of-2 number of bits, expand as two loads.
assert(!SrcVT.isVector() && "Unsupported extload!");
unsigned LogSrcWidth = Log2_32(SrcWidth);
assert(LogSrcWidth < 32);
unsigned RoundWidth = 1 << LogSrcWidth;
assert(RoundWidth < SrcWidth);
unsigned ExtraWidth = SrcWidth - RoundWidth;
assert(ExtraWidth < RoundWidth);
assert(!(RoundWidth % 8) && !(ExtraWidth % 8) &&
"Load size not an integral number of bytes!");
EVT RoundVT = EVT::getIntegerVT(*DAG.getContext(), RoundWidth);
EVT ExtraVT = EVT::getIntegerVT(*DAG.getContext(), ExtraWidth);
SDValue Lo, Hi, Ch;
unsigned IncrementSize;
auto &DL = DAG.getDataLayout();
if (DL.isLittleEndian()) {
// EXTLOAD:i24 -> ZEXTLOAD:i16 | (shl EXTLOAD@+2:i8, 16)
// Load the bottom RoundWidth bits.
Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, Node->getValueType(0), Chain, Ptr,
LD->getPointerInfo(), RoundVT, Alignment, MMOFlags,
AAInfo);
// Load the remaining ExtraWidth bits.
IncrementSize = RoundWidth / 8;
Ptr = DAG.getMemBasePlusOffset(Ptr, IncrementSize, dl);
Hi = DAG.getExtLoad(ExtType, dl, Node->getValueType(0), Chain, Ptr,
LD->getPointerInfo().getWithOffset(IncrementSize),
ExtraVT, MinAlign(Alignment, IncrementSize), MMOFlags,
AAInfo);
// Build a factor node to remember that this load is independent of
// the other one.
Ch = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo.getValue(1),
Hi.getValue(1));
// Move the top bits to the right place.
Hi = DAG.getNode(
ISD::SHL, dl, Hi.getValueType(), Hi,
DAG.getConstant(RoundWidth, dl,
TLI.getShiftAmountTy(Hi.getValueType(), DL)));
// Join the hi and lo parts.
Value = DAG.getNode(ISD::OR, dl, Node->getValueType(0), Lo, Hi);
} else {
// Big endian - avoid unaligned loads.
// EXTLOAD:i24 -> (shl EXTLOAD:i16, 8) | ZEXTLOAD@+2:i8
// Load the top RoundWidth bits.
Hi = DAG.getExtLoad(ExtType, dl, Node->getValueType(0), Chain, Ptr,
LD->getPointerInfo(), RoundVT, Alignment, MMOFlags,
AAInfo);
// Load the remaining ExtraWidth bits.
IncrementSize = RoundWidth / 8;
Ptr = DAG.getMemBasePlusOffset(Ptr, IncrementSize, dl);
Lo = DAG.getExtLoad(ISD::ZEXTLOAD, dl, Node->getValueType(0), Chain, Ptr,
LD->getPointerInfo().getWithOffset(IncrementSize),
ExtraVT, MinAlign(Alignment, IncrementSize), MMOFlags,
AAInfo);
// Build a factor node to remember that this load is independent of
// the other one.
Ch = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo.getValue(1),
Hi.getValue(1));
// Move the top bits to the right place.
Hi = DAG.getNode(
ISD::SHL, dl, Hi.getValueType(), Hi,
DAG.getConstant(ExtraWidth, dl,
TLI.getShiftAmountTy(Hi.getValueType(), DL)));
// Join the hi and lo parts.
Value = DAG.getNode(ISD::OR, dl, Node->getValueType(0), Lo, Hi);
}
Chain = Ch;
} else {
bool isCustom = false;
switch (TLI.getLoadExtAction(ExtType, Node->getValueType(0),
SrcVT.getSimpleVT())) {
default: llvm_unreachable("This action is not supported yet!");
case TargetLowering::Custom:
isCustom = true;
LLVM_FALLTHROUGH;
case TargetLowering::Legal:
Value = SDValue(Node, 0);
Chain = SDValue(Node, 1);
if (isCustom) {
if (SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG)) {
Value = Res;
Chain = Res.getValue(1);
}
} else {
// If this is an unaligned load and the target doesn't support it,
// expand it.
EVT MemVT = LD->getMemoryVT();
const DataLayout &DL = DAG.getDataLayout();
if (!TLI.allowsMemoryAccess(*DAG.getContext(), DL, MemVT,
*LD->getMemOperand())) {
std::tie(Value, Chain) = TLI.expandUnalignedLoad(LD, DAG);
}
}
break;
case TargetLowering::Expand: {
EVT DestVT = Node->getValueType(0);
if (!TLI.isLoadExtLegal(ISD::EXTLOAD, DestVT, SrcVT)) {
// If the source type is not legal, see if there is a legal extload to
// an intermediate type that we can then extend further.
EVT LoadVT = TLI.getRegisterType(SrcVT.getSimpleVT());
if (TLI.isTypeLegal(SrcVT) || // Same as SrcVT == LoadVT?
TLI.isLoadExtLegal(ExtType, LoadVT, SrcVT)) {
// If we are loading a legal type, this is a non-extload followed by a
// full extend.
ISD::LoadExtType MidExtType =
(LoadVT == SrcVT) ? ISD::NON_EXTLOAD : ExtType;
SDValue Load = DAG.getExtLoad(MidExtType, dl, LoadVT, Chain, Ptr,
SrcVT, LD->getMemOperand());
unsigned ExtendOp =
ISD::getExtForLoadExtType(SrcVT.isFloatingPoint(), ExtType);
Value = DAG.getNode(ExtendOp, dl, Node->getValueType(0), Load);
Chain = Load.getValue(1);
break;
}
// Handle the special case of fp16 extloads. EXTLOAD doesn't have the
// normal undefined upper bits behavior to allow using an in-reg extend
// with the illegal FP type, so load as an integer and do the
// from-integer conversion.
if (SrcVT.getScalarType() == MVT::f16) {
EVT ISrcVT = SrcVT.changeTypeToInteger();
EVT IDestVT = DestVT.changeTypeToInteger();
EVT ILoadVT = TLI.getRegisterType(IDestVT.getSimpleVT());
SDValue Result = DAG.getExtLoad(ISD::ZEXTLOAD, dl, ILoadVT, Chain,
Ptr, ISrcVT, LD->getMemOperand());
Value = DAG.getNode(ISD::FP16_TO_FP, dl, DestVT, Result);
Chain = Result.getValue(1);
break;
}
}
assert(!SrcVT.isVector() &&
"Vector Loads are handled in LegalizeVectorOps");
// FIXME: This does not work for vectors on most targets. Sign-
// and zero-extend operations are currently folded into extending
// loads, whether they are legal or not, and then we end up here
// without any support for legalizing them.
assert(ExtType != ISD::EXTLOAD &&
"EXTLOAD should always be supported!");
// Turn the unsupported load into an EXTLOAD followed by an
// explicit zero/sign extend inreg.
SDValue Result = DAG.getExtLoad(ISD::EXTLOAD, dl,
Node->getValueType(0),
Chain, Ptr, SrcVT,
LD->getMemOperand());
SDValue ValRes;
if (ExtType == ISD::SEXTLOAD)
ValRes = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl,
Result.getValueType(),
Result, DAG.getValueType(SrcVT));
else
ValRes = DAG.getZeroExtendInReg(Result, dl, SrcVT.getScalarType());
Value = ValRes;
Chain = Result.getValue(1);
break;
}
}
}
// Since loads produce two values, make sure to remember that we legalized
// both of them.
if (Chain.getNode() != Node) {
assert(Value.getNode() != Node && "Load must be completely replaced");
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 0), Value);
DAG.ReplaceAllUsesOfValueWith(SDValue(Node, 1), Chain);
if (UpdatedNodes) {
UpdatedNodes->insert(Value.getNode());
UpdatedNodes->insert(Chain.getNode());
}
ReplacedNode(Node);
}
}
/// Return a legal replacement for the given operation, with all legal operands.
void SelectionDAGLegalize::LegalizeOp(SDNode *Node) {
LLVM_DEBUG(dbgs() << "\nLegalizing: "; Node->dump(&DAG));
// Allow illegal target nodes and illegal registers.
if (Node->getOpcode() == ISD::TargetConstant ||
Node->getOpcode() == ISD::Register)
return;
#ifndef NDEBUG
for (unsigned i = 0, e = Node->getNumValues(); i != e; ++i)
assert(TLI.getTypeAction(*DAG.getContext(), Node->getValueType(i)) ==
TargetLowering::TypeLegal &&
"Unexpected illegal type!");
for (const SDValue &Op : Node->op_values())
assert((TLI.getTypeAction(*DAG.getContext(), Op.getValueType()) ==
TargetLowering::TypeLegal ||
Op.getOpcode() == ISD::TargetConstant ||
Op.getOpcode() == ISD::Register) &&
"Unexpected illegal type!");
#endif
// Figure out the correct action; the way to query this varies by opcode
TargetLowering::LegalizeAction Action = TargetLowering::Legal;
bool SimpleFinishLegalizing = true;
switch (Node->getOpcode()) {
case ISD::INTRINSIC_W_CHAIN:
case ISD::INTRINSIC_WO_CHAIN:
case ISD::INTRINSIC_VOID:
case ISD::STACKSAVE:
Action = TLI.getOperationAction(Node->getOpcode(), MVT::Other);
break;
case ISD::GET_DYNAMIC_AREA_OFFSET:
Action = TLI.getOperationAction(Node->getOpcode(),
Node->getValueType(0));
break;
case ISD::VAARG:
Action = TLI.getOperationAction(Node->getOpcode(),
Node->getValueType(0));
if (Action != TargetLowering::Promote)
Action = TLI.getOperationAction(Node->getOpcode(), MVT::Other);
break;
case ISD::FP_TO_FP16:
case ISD::SINT_TO_FP:
case ISD::UINT_TO_FP:
case ISD::EXTRACT_VECTOR_ELT:
case ISD::LROUND:
case ISD::LLROUND:
case ISD::LRINT:
case ISD::LLRINT:
Action = TLI.getOperationAction(Node->getOpcode(),
Node->getOperand(0).getValueType());
break;
case ISD::STRICT_SINT_TO_FP:
case ISD::STRICT_UINT_TO_FP:
case ISD::STRICT_LRINT:
case ISD::STRICT_LLRINT:
case ISD::STRICT_LROUND:
case ISD::STRICT_LLROUND:
// These pseudo-ops are the same as the other STRICT_ ops except
// they are registered with setOperationAction() using the input type
// instead of the output type.
Action = TLI.getOperationAction(Node->getOpcode(),
Node->getOperand(1).getValueType());
break;
case ISD::SIGN_EXTEND_INREG: {
EVT InnerType = cast<VTSDNode>(Node->getOperand(1))->getVT();
Action = TLI.getOperationAction(Node->getOpcode(), InnerType);
break;
}
case ISD::ATOMIC_STORE:
Action = TLI.getOperationAction(Node->getOpcode(),
Node->getOperand(2).getValueType());
break;
case ISD::SELECT_CC:
case ISD::STRICT_FSETCC:
case ISD::STRICT_FSETCCS:
case ISD::SETCC:
case ISD::BR_CC: {
unsigned CCOperand = Node->getOpcode() == ISD::SELECT_CC ? 4 :
Node->getOpcode() == ISD::STRICT_FSETCC ? 3 :
Node->getOpcode() == ISD::STRICT_FSETCCS ? 3 :
Node->getOpcode() == ISD::SETCC ? 2 : 1;
unsigned CompareOperand = Node->getOpcode() == ISD::BR_CC ? 2 :
Node->getOpcode() == ISD::STRICT_FSETCC ? 1 :
Node->getOpcode() == ISD::STRICT_FSETCCS ? 1 : 0;
MVT OpVT = Node->getOperand(CompareOperand).getSimpleValueType();
ISD::CondCode CCCode =
cast<CondCodeSDNode>(Node->getOperand(CCOperand))->get();
Action = TLI.getCondCodeAction(CCCode, OpVT);
if (Action == TargetLowering::Legal) {
if (Node->getOpcode() == ISD::SELECT_CC)
Action = TLI.getOperationAction(Node->getOpcode(),
Node->getValueType(0));
else
Action = TLI.getOperationAction(Node->getOpcode(), OpVT);
}
break;
}
case ISD::LOAD:
case ISD::STORE:
// FIXME: Model these properly. LOAD and STORE are complicated, and
// STORE expects the unlegalized operand in some cases.
SimpleFinishLegalizing = false;
break;
case ISD::CALLSEQ_START:
case ISD::CALLSEQ_END:
// FIXME: This shouldn't be necessary. These nodes have special properties
// dealing with the recursive nature of legalization. Removing this
// special case should be done as part of making LegalizeDAG non-recursive.
SimpleFinishLegalizing = false;
break;
case ISD::EXTRACT_ELEMENT:
case ISD::FLT_ROUNDS_:
case ISD::MERGE_VALUES:
case ISD::EH_RETURN:
case ISD::FRAME_TO_ARGS_OFFSET:
case ISD::EH_DWARF_CFA:
case ISD::EH_SJLJ_SETJMP:
case ISD::EH_SJLJ_LONGJMP:
case ISD::EH_SJLJ_SETUP_DISPATCH:
// These operations lie about being legal: when they claim to be legal,
// they should actually be expanded.
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
if (Action == TargetLowering::Legal)
Action = TargetLowering::Expand;
break;
case ISD::INIT_TRAMPOLINE:
case ISD::ADJUST_TRAMPOLINE:
case ISD::FRAMEADDR:
case ISD::RETURNADDR:
case ISD::ADDROFRETURNADDR:
case ISD::SPONENTRY:
// These operations lie about being legal: when they claim to be legal,
// they should actually be custom-lowered.
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
if (Action == TargetLowering::Legal)
Action = TargetLowering::Custom;
break;
case ISD::READCYCLECOUNTER:
// READCYCLECOUNTER returns an i64, even if type legalization might have
// expanded that to several smaller types.
Action = TLI.getOperationAction(Node->getOpcode(), MVT::i64);
break;
case ISD::READ_REGISTER:
case ISD::WRITE_REGISTER:
// Named register is legal in the DAG, but blocked by register name
// selection if not implemented by target (to chose the correct register)
// They'll be converted to Copy(To/From)Reg.
Action = TargetLowering::Legal;
break;
case ISD::DEBUGTRAP:
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
if (Action == TargetLowering::Expand) {
// replace ISD::DEBUGTRAP with ISD::TRAP
SDValue NewVal;
NewVal = DAG.getNode(ISD::TRAP, SDLoc(Node), Node->getVTList(),
Node->getOperand(0));
ReplaceNode(Node, NewVal.getNode());
LegalizeOp(NewVal.getNode());
return;
}
break;
case ISD::SADDSAT:
case ISD::UADDSAT:
case ISD::SSUBSAT:
case ISD::USUBSAT: {
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
break;
}
case ISD::SMULFIX:
case ISD::SMULFIXSAT:
case ISD::UMULFIX:
case ISD::UMULFIXSAT:
case ISD::SDIVFIX:
case ISD::UDIVFIX: {
unsigned Scale = Node->getConstantOperandVal(2);
Action = TLI.getFixedPointOperationAction(Node->getOpcode(),
Node->getValueType(0), Scale);
break;
}
case ISD::MSCATTER:
Action = TLI.getOperationAction(Node->getOpcode(),
cast<MaskedScatterSDNode>(Node)->getValue().getValueType());
break;
case ISD::MSTORE:
Action = TLI.getOperationAction(Node->getOpcode(),
cast<MaskedStoreSDNode>(Node)->getValue().getValueType());
break;
case ISD::VECREDUCE_FADD:
case ISD::VECREDUCE_FMUL:
case ISD::VECREDUCE_ADD:
case ISD::VECREDUCE_MUL:
case ISD::VECREDUCE_AND:
case ISD::VECREDUCE_OR:
case ISD::VECREDUCE_XOR:
case ISD::VECREDUCE_SMAX:
case ISD::VECREDUCE_SMIN:
case ISD::VECREDUCE_UMAX:
case ISD::VECREDUCE_UMIN:
case ISD::VECREDUCE_FMAX:
case ISD::VECREDUCE_FMIN:
Action = TLI.getOperationAction(
Node->getOpcode(), Node->getOperand(0).getValueType());
break;
default:
if (Node->getOpcode() >= ISD::BUILTIN_OP_END) {
Action = TargetLowering::Legal;
} else {
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
}
break;
}
if (SimpleFinishLegalizing) {
SDNode *NewNode = Node;
switch (Node->getOpcode()) {
default: break;
case ISD::SHL:
case ISD::SRL:
case ISD::SRA:
case ISD::ROTL:
case ISD::ROTR: {
// Legalizing shifts/rotates requires adjusting the shift amount
// to the appropriate width.
SDValue Op0 = Node->getOperand(0);
SDValue Op1 = Node->getOperand(1);
if (!Op1.getValueType().isVector()) {
SDValue SAO = DAG.getShiftAmountOperand(Op0.getValueType(), Op1);
// The getShiftAmountOperand() may create a new operand node or
// return the existing one. If new operand is created we need
// to update the parent node.
// Do not try to legalize SAO here! It will be automatically legalized
// in the next round.
if (SAO != Op1)
NewNode = DAG.UpdateNodeOperands(Node, Op0, SAO);
}
}
break;
case ISD::FSHL:
case ISD::FSHR:
case ISD::SRL_PARTS:
case ISD::SRA_PARTS:
case ISD::SHL_PARTS: {
// Legalizing shifts/rotates requires adjusting the shift amount
// to the appropriate width.
SDValue Op0 = Node->getOperand(0);
SDValue Op1 = Node->getOperand(1);
SDValue Op2 = Node->getOperand(2);
if (!Op2.getValueType().isVector()) {
SDValue SAO = DAG.getShiftAmountOperand(Op0.getValueType(), Op2);
// The getShiftAmountOperand() may create a new operand node or
// return the existing one. If new operand is created we need
// to update the parent node.
if (SAO != Op2)
NewNode = DAG.UpdateNodeOperands(Node, Op0, Op1, SAO);
}
break;
}
}
if (NewNode != Node) {
ReplaceNode(Node, NewNode);
Node = NewNode;
}
switch (Action) {
case TargetLowering::Legal:
LLVM_DEBUG(dbgs() << "Legal node: nothing to do\n");
return;
case TargetLowering::Custom:
LLVM_DEBUG(dbgs() << "Trying custom legalization\n");
// FIXME: The handling for custom lowering with multiple results is
// a complete mess.
if (SDValue Res = TLI.LowerOperation(SDValue(Node, 0), DAG)) {
if (!(Res.getNode() != Node || Res.getResNo() != 0))
return;
if (Node->getNumValues() == 1) {
LLVM_DEBUG(dbgs() << "Successfully custom legalized node\n");
// We can just directly replace this node with the lowered value.
ReplaceNode(SDValue(Node, 0), Res);
return;
}
SmallVector<SDValue, 8> ResultVals;
for (unsigned i = 0, e = Node->getNumValues(); i != e; ++i)
ResultVals.push_back(Res.getValue(i));
LLVM_DEBUG(dbgs() << "Successfully custom legalized node\n");
ReplaceNode(Node, ResultVals.data());
return;
}
LLVM_DEBUG(dbgs() << "Could not custom legalize node\n");
LLVM_FALLTHROUGH;
case TargetLowering::Expand:
if (ExpandNode(Node))
return;
LLVM_FALLTHROUGH;
case TargetLowering::LibCall:
ConvertNodeToLibcall(Node);
return;
case TargetLowering::Promote:
PromoteNode(Node);
return;
}
}
switch (Node->getOpcode()) {
default:
#ifndef NDEBUG
dbgs() << "NODE: ";
Node->dump( &DAG);
dbgs() << "\n";
#endif
llvm_unreachable("Do not know how to legalize this operator!");
case ISD::CALLSEQ_START:
case ISD::CALLSEQ_END:
break;
case ISD::LOAD:
return LegalizeLoadOps(Node);
case ISD::STORE:
return LegalizeStoreOps(Node);
}
}
SDValue SelectionDAGLegalize::ExpandExtractFromVectorThroughStack(SDValue Op) {
SDValue Vec = Op.getOperand(0);
SDValue Idx = Op.getOperand(1);
SDLoc dl(Op);
// Before we generate a new store to a temporary stack slot, see if there is
// already one that we can use. There often is because when we scalarize
// vector operations (using SelectionDAG::UnrollVectorOp for example) a whole
// series of EXTRACT_VECTOR_ELT nodes are generated, one for each element in
// the vector. If all are expanded here, we don't want one store per vector
// element.
// Caches for hasPredecessorHelper
SmallPtrSet<const SDNode *, 32> Visited;
SmallVector<const SDNode *, 16> Worklist;
Visited.insert(Op.getNode());
Worklist.push_back(Idx.getNode());
SDValue StackPtr, Ch;
for (SDNode::use_iterator UI = Vec.getNode()->use_begin(),
UE = Vec.getNode()->use_end(); UI != UE; ++UI) {
SDNode *User = *UI;
if (StoreSDNode *ST = dyn_cast<StoreSDNode>(User)) {
if (ST->isIndexed() || ST->isTruncatingStore() ||
ST->getValue() != Vec)
continue;
// Make sure that nothing else could have stored into the destination of
// this store.
if (!ST->getChain().reachesChainWithoutSideEffects(DAG.getEntryNode()))
continue;
// If the index is dependent on the store we will introduce a cycle when
// creating the load (the load uses the index, and by replacing the chain
// we will make the index dependent on the load). Also, the store might be
// dependent on the extractelement and introduce a cycle when creating
// the load.
if (SDNode::hasPredecessorHelper(ST, Visited, Worklist) ||
ST->hasPredecessor(Op.getNode()))
continue;
StackPtr = ST->getBasePtr();
Ch = SDValue(ST, 0);
break;
}
}
EVT VecVT = Vec.getValueType();
if (!Ch.getNode()) {
// Store the value to a temporary stack slot, then LOAD the returned part.
StackPtr = DAG.CreateStackTemporary(VecVT);
Ch = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr,
MachinePointerInfo());
}
StackPtr = TLI.getVectorElementPointer(DAG, StackPtr, VecVT, Idx);
SDValue NewLoad;
if (Op.getValueType().isVector())
NewLoad =
DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr, MachinePointerInfo());
else
NewLoad = DAG.getExtLoad(ISD::EXTLOAD, dl, Op.getValueType(), Ch, StackPtr,
MachinePointerInfo(),
VecVT.getVectorElementType());
// Replace the chain going out of the store, by the one out of the load.
DAG.ReplaceAllUsesOfValueWith(Ch, SDValue(NewLoad.getNode(), 1));
// We introduced a cycle though, so update the loads operands, making sure
// to use the original store's chain as an incoming chain.
SmallVector<SDValue, 6> NewLoadOperands(NewLoad->op_begin(),
NewLoad->op_end());
NewLoadOperands[0] = Ch;
NewLoad =
SDValue(DAG.UpdateNodeOperands(NewLoad.getNode(), NewLoadOperands), 0);
return NewLoad;
}
SDValue SelectionDAGLegalize::ExpandInsertToVectorThroughStack(SDValue Op) {
assert(Op.getValueType().isVector() && "Non-vector insert subvector!");
SDValue Vec = Op.getOperand(0);
SDValue Part = Op.getOperand(1);
SDValue Idx = Op.getOperand(2);
SDLoc dl(Op);
// Store the value to a temporary stack slot, then LOAD the returned part.
EVT VecVT = Vec.getValueType();
SDValue StackPtr = DAG.CreateStackTemporary(VecVT);
int FI = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
MachinePointerInfo PtrInfo =
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI);
// First store the whole vector.
SDValue Ch = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr, PtrInfo);
// Then store the inserted part.
SDValue SubStackPtr = TLI.getVectorElementPointer(DAG, StackPtr, VecVT, Idx);
// Store the subvector.
Ch = DAG.getStore(Ch, dl, Part, SubStackPtr, MachinePointerInfo());
// Finally, load the updated vector.
return DAG.getLoad(Op.getValueType(), dl, Ch, StackPtr, PtrInfo);
}
SDValue SelectionDAGLegalize::ExpandVectorBuildThroughStack(SDNode* Node) {
// We can't handle this case efficiently. Allocate a sufficiently
// aligned object on the stack, store each element into it, then load
// the result as a vector.
// Create the stack frame object.
EVT VT = Node->getValueType(0);
EVT EltVT = VT.getVectorElementType();
SDLoc dl(Node);
SDValue FIPtr = DAG.CreateStackTemporary(VT);
int FI = cast<FrameIndexSDNode>(FIPtr.getNode())->getIndex();
MachinePointerInfo PtrInfo =
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI);
// Emit a store of each element to the stack slot.
SmallVector<SDValue, 8> Stores;
unsigned TypeByteSize = EltVT.getSizeInBits() / 8;
assert(TypeByteSize > 0 && "Vector element type too small for stack store!");
// Store (in the right endianness) the elements to memory.
for (unsigned i = 0, e = Node->getNumOperands(); i != e; ++i) {
// Ignore undef elements.
if (Node->getOperand(i).isUndef()) continue;
unsigned Offset = TypeByteSize*i;
SDValue Idx = DAG.getConstant(Offset, dl, FIPtr.getValueType());
Idx = DAG.getMemBasePlusOffset(FIPtr, Idx, dl);
// If the destination vector element type is narrower than the source
// element type, only store the bits necessary.
if (EltVT.bitsLT(Node->getOperand(i).getValueType().getScalarType())) {
Stores.push_back(DAG.getTruncStore(DAG.getEntryNode(), dl,
Node->getOperand(i), Idx,
PtrInfo.getWithOffset(Offset), EltVT));
} else
Stores.push_back(DAG.getStore(DAG.getEntryNode(), dl, Node->getOperand(i),
Idx, PtrInfo.getWithOffset(Offset)));
}
SDValue StoreChain;
if (!Stores.empty()) // Not all undef elements?
StoreChain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Stores);
else
StoreChain = DAG.getEntryNode();
// Result is a load from the stack slot.
return DAG.getLoad(VT, dl, StoreChain, FIPtr, PtrInfo);
}
/// Bitcast a floating-point value to an integer value. Only bitcast the part
/// containing the sign bit if the target has no integer value capable of
/// holding all bits of the floating-point value.
void SelectionDAGLegalize::getSignAsIntValue(FloatSignAsInt &State,
const SDLoc &DL,
SDValue Value) const {
EVT FloatVT = Value.getValueType();
unsigned NumBits = FloatVT.getSizeInBits();
State.FloatVT = FloatVT;
EVT IVT = EVT::getIntegerVT(*DAG.getContext(), NumBits);
// Convert to an integer of the same size.
if (TLI.isTypeLegal(IVT)) {
State.IntValue = DAG.getNode(ISD::BITCAST, DL, IVT, Value);
State.SignMask = APInt::getSignMask(NumBits);
State.SignBit = NumBits - 1;
return;
}
auto &DataLayout = DAG.getDataLayout();
// Store the float to memory, then load the sign part out as an integer.
MVT LoadTy = TLI.getRegisterType(*DAG.getContext(), MVT::i8);
// First create a temporary that is aligned for both the load and store.
SDValue StackPtr = DAG.CreateStackTemporary(FloatVT, LoadTy);
int FI = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
// Then store the float to it.
State.FloatPtr = StackPtr;
MachineFunction &MF = DAG.getMachineFunction();
State.FloatPointerInfo = MachinePointerInfo::getFixedStack(MF, FI);
State.Chain = DAG.getStore(DAG.getEntryNode(), DL, Value, State.FloatPtr,
State.FloatPointerInfo);
SDValue IntPtr;
if (DataLayout.isBigEndian()) {
assert(FloatVT.isByteSized() && "Unsupported floating point type!");
// Load out a legal integer with the same sign bit as the float.
IntPtr = StackPtr;
State.IntPointerInfo = State.FloatPointerInfo;
} else {
// Advance the pointer so that the loaded byte will contain the sign bit.
unsigned ByteOffset = (FloatVT.getSizeInBits() / 8) - 1;
IntPtr = DAG.getMemBasePlusOffset(StackPtr, ByteOffset, DL);
State.IntPointerInfo = MachinePointerInfo::getFixedStack(MF, FI,
ByteOffset);
}
State.IntPtr = IntPtr;
State.IntValue = DAG.getExtLoad(ISD::EXTLOAD, DL, LoadTy, State.Chain, IntPtr,
State.IntPointerInfo, MVT::i8);
State.SignMask = APInt::getOneBitSet(LoadTy.getSizeInBits(), 7);
State.SignBit = 7;
}
/// Replace the integer value produced by getSignAsIntValue() with a new value
/// and cast the result back to a floating-point type.
SDValue SelectionDAGLegalize::modifySignAsInt(const FloatSignAsInt &State,
const SDLoc &DL,
SDValue NewIntValue) const {
if (!State.Chain)
return DAG.getNode(ISD::BITCAST, DL, State.FloatVT, NewIntValue);
// Override the part containing the sign bit in the value stored on the stack.
SDValue Chain = DAG.getTruncStore(State.Chain, DL, NewIntValue, State.IntPtr,
State.IntPointerInfo, MVT::i8);
return DAG.getLoad(State.FloatVT, DL, Chain, State.FloatPtr,
State.FloatPointerInfo);
}
SDValue SelectionDAGLegalize::ExpandFCOPYSIGN(SDNode *Node) const {
SDLoc DL(Node);
SDValue Mag = Node->getOperand(0);
SDValue Sign = Node->getOperand(1);
// Get sign bit into an integer value.
FloatSignAsInt SignAsInt;
getSignAsIntValue(SignAsInt, DL, Sign);
EVT IntVT = SignAsInt.IntValue.getValueType();
SDValue SignMask = DAG.getConstant(SignAsInt.SignMask, DL, IntVT);
SDValue SignBit = DAG.getNode(ISD::AND, DL, IntVT, SignAsInt.IntValue,
SignMask);
// If FABS is legal transform FCOPYSIGN(x, y) => sign(x) ? -FABS(x) : FABS(X)
EVT FloatVT = Mag.getValueType();
if (TLI.isOperationLegalOrCustom(ISD::FABS, FloatVT) &&
TLI.isOperationLegalOrCustom(ISD::FNEG, FloatVT)) {
SDValue AbsValue = DAG.getNode(ISD::FABS, DL, FloatVT, Mag);
SDValue NegValue = DAG.getNode(ISD::FNEG, DL, FloatVT, AbsValue);
SDValue Cond = DAG.getSetCC(DL, getSetCCResultType(IntVT), SignBit,
DAG.getConstant(0, DL, IntVT), ISD::SETNE);
return DAG.getSelect(DL, FloatVT, Cond, NegValue, AbsValue);
}
// Transform Mag value to integer, and clear the sign bit.
FloatSignAsInt MagAsInt;
getSignAsIntValue(MagAsInt, DL, Mag);
EVT MagVT = MagAsInt.IntValue.getValueType();
SDValue ClearSignMask = DAG.getConstant(~MagAsInt.SignMask, DL, MagVT);
SDValue ClearedSign = DAG.getNode(ISD::AND, DL, MagVT, MagAsInt.IntValue,
ClearSignMask);
// Get the signbit at the right position for MagAsInt.
int ShiftAmount = SignAsInt.SignBit - MagAsInt.SignBit;
EVT ShiftVT = IntVT;
if (SignBit.getValueSizeInBits() < ClearedSign.getValueSizeInBits()) {
SignBit = DAG.getNode(ISD::ZERO_EXTEND, DL, MagVT, SignBit);
ShiftVT = MagVT;
}
if (ShiftAmount > 0) {
SDValue ShiftCnst = DAG.getConstant(ShiftAmount, DL, ShiftVT);
SignBit = DAG.getNode(ISD::SRL, DL, ShiftVT, SignBit, ShiftCnst);
} else if (ShiftAmount < 0) {
SDValue ShiftCnst = DAG.getConstant(-ShiftAmount, DL, ShiftVT);
SignBit = DAG.getNode(ISD::SHL, DL, ShiftVT, SignBit, ShiftCnst);
}
if (SignBit.getValueSizeInBits() > ClearedSign.getValueSizeInBits()) {
SignBit = DAG.getNode(ISD::TRUNCATE, DL, MagVT, SignBit);
}
// Store the part with the modified sign and convert back to float.
SDValue CopiedSign = DAG.getNode(ISD::OR, DL, MagVT, ClearedSign, SignBit);
return modifySignAsInt(MagAsInt, DL, CopiedSign);
}
SDValue SelectionDAGLegalize::ExpandFABS(SDNode *Node) const {
SDLoc DL(Node);
SDValue Value = Node->getOperand(0);
// Transform FABS(x) => FCOPYSIGN(x, 0.0) if FCOPYSIGN is legal.
EVT FloatVT = Value.getValueType();
if (TLI.isOperationLegalOrCustom(ISD::FCOPYSIGN, FloatVT)) {
SDValue Zero = DAG.getConstantFP(0.0, DL, FloatVT);
return DAG.getNode(ISD::FCOPYSIGN, DL, FloatVT, Value, Zero);
}
// Transform value to integer, clear the sign bit and transform back.
FloatSignAsInt ValueAsInt;
getSignAsIntValue(ValueAsInt, DL, Value);
EVT IntVT = ValueAsInt.IntValue.getValueType();
SDValue ClearSignMask = DAG.getConstant(~ValueAsInt.SignMask, DL, IntVT);
SDValue ClearedSign = DAG.getNode(ISD::AND, DL, IntVT, ValueAsInt.IntValue,
ClearSignMask);
return modifySignAsInt(ValueAsInt, DL, ClearedSign);
}
void SelectionDAGLegalize::ExpandDYNAMIC_STACKALLOC(SDNode* Node,
SmallVectorImpl<SDValue> &Results) {
unsigned SPReg = TLI.getStackPointerRegisterToSaveRestore();
assert(SPReg && "Target cannot require DYNAMIC_STACKALLOC expansion and"
" not tell us which reg is the stack pointer!");
SDLoc dl(Node);
EVT VT = Node->getValueType(0);
SDValue Tmp1 = SDValue(Node, 0);
SDValue Tmp2 = SDValue(Node, 1);
SDValue Tmp3 = Node->getOperand(2);
SDValue Chain = Tmp1.getOperand(0);
// Chain the dynamic stack allocation so that it doesn't modify the stack
// pointer when other instructions are using the stack.
Chain = DAG.getCALLSEQ_START(Chain, 0, 0, dl);
SDValue Size = Tmp2.getOperand(1);
SDValue SP = DAG.getCopyFromReg(Chain, dl, SPReg, VT);
Chain = SP.getValue(1);
unsigned Align = cast<ConstantSDNode>(Tmp3)->getZExtValue();
unsigned StackAlign =
DAG.getSubtarget().getFrameLowering()->getStackAlignment();
Tmp1 = DAG.getNode(ISD::SUB, dl, VT, SP, Size); // Value
if (Align > StackAlign)
Tmp1 = DAG.getNode(ISD::AND, dl, VT, Tmp1,
DAG.getConstant(-(uint64_t)Align, dl, VT));
Chain = DAG.getCopyToReg(Chain, dl, SPReg, Tmp1); // Output chain
Tmp2 = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(0, dl, true),
DAG.getIntPtrConstant(0, dl, true), SDValue(), dl);
Results.push_back(Tmp1);
Results.push_back(Tmp2);
}
/// Legalize a SETCC with given LHS and RHS and condition code CC on the current
/// target.
///
/// If the SETCC has been legalized using AND / OR, then the legalized node
/// will be stored in LHS. RHS and CC will be set to SDValue(). NeedInvert
/// will be set to false.
///
/// If the SETCC has been legalized by using getSetCCSwappedOperands(),
/// then the values of LHS and RHS will be swapped, CC will be set to the
/// new condition, and NeedInvert will be set to false.
///
/// If the SETCC has been legalized using the inverse condcode, then LHS and
/// RHS will be unchanged, CC will set to the inverted condcode, and NeedInvert
/// will be set to true. The caller must invert the result of the SETCC with
/// SelectionDAG::getLogicalNOT() or take equivalent action to swap the effect
/// of a true/false result.
///
/// \returns true if the SetCC has been legalized, false if it hasn't.
bool SelectionDAGLegalize::LegalizeSetCCCondCode(
EVT VT, SDValue &LHS, SDValue &RHS, SDValue &CC, bool &NeedInvert,
const SDLoc &dl, SDValue &Chain, bool IsSignaling) {
MVT OpVT = LHS.getSimpleValueType();
ISD::CondCode CCCode = cast<CondCodeSDNode>(CC)->get();
NeedInvert = false;
switch (TLI.getCondCodeAction(CCCode, OpVT)) {
default: llvm_unreachable("Unknown condition code action!");
case TargetLowering::Legal:
// Nothing to do.
break;
case TargetLowering::Expand: {
ISD::CondCode InvCC = ISD::getSetCCSwappedOperands(CCCode);
if (TLI.isCondCodeLegalOrCustom(InvCC, OpVT)) {
std::swap(LHS, RHS);
CC = DAG.getCondCode(InvCC);
return true;
}
// Swapping operands didn't work. Try inverting the condition.
bool NeedSwap = false;
InvCC = getSetCCInverse(CCCode, OpVT);
if (!TLI.isCondCodeLegalOrCustom(InvCC, OpVT)) {
// If inverting the condition is not enough, try swapping operands
// on top of it.
InvCC = ISD::getSetCCSwappedOperands(InvCC);
NeedSwap = true;
}
if (TLI.isCondCodeLegalOrCustom(InvCC, OpVT)) {
CC = DAG.getCondCode(InvCC);
NeedInvert = true;
if (NeedSwap)
std::swap(LHS, RHS);
return true;
}
ISD::CondCode CC1 = ISD::SETCC_INVALID, CC2 = ISD::SETCC_INVALID;
unsigned Opc = 0;
switch (CCCode) {
default: llvm_unreachable("Don't know how to expand this condition!");
case ISD::SETO:
assert(TLI.isCondCodeLegal(ISD::SETOEQ, OpVT)
&& "If SETO is expanded, SETOEQ must be legal!");
CC1 = ISD::SETOEQ; CC2 = ISD::SETOEQ; Opc = ISD::AND; break;
case ISD::SETUO:
assert(TLI.isCondCodeLegal(ISD::SETUNE, OpVT)
&& "If SETUO is expanded, SETUNE must be legal!");
CC1 = ISD::SETUNE; CC2 = ISD::SETUNE; Opc = ISD::OR; break;
case ISD::SETOEQ:
case ISD::SETOGT:
case ISD::SETOGE:
case ISD::SETOLT:
case ISD::SETOLE:
case ISD::SETONE:
case ISD::SETUEQ:
case ISD::SETUNE:
case ISD::SETUGT:
case ISD::SETUGE:
case ISD::SETULT:
case ISD::SETULE:
// If we are floating point, assign and break, otherwise fall through.
if (!OpVT.isInteger()) {
// We can use the 4th bit to tell if we are the unordered
// or ordered version of the opcode.
CC2 = ((unsigned)CCCode & 0x8U) ? ISD::SETUO : ISD::SETO;
Opc = ((unsigned)CCCode & 0x8U) ? ISD::OR : ISD::AND;
CC1 = (ISD::CondCode)(((int)CCCode & 0x7) | 0x10);
break;
}
// Fallthrough if we are unsigned integer.
LLVM_FALLTHROUGH;
case ISD::SETLE:
case ISD::SETGT:
case ISD::SETGE:
case ISD::SETLT:
case ISD::SETNE:
case ISD::SETEQ:
// If all combinations of inverting the condition and swapping operands
// didn't work then we have no means to expand the condition.
llvm_unreachable("Don't know how to expand this condition!");
}
SDValue SetCC1, SetCC2;
if (CCCode != ISD::SETO && CCCode != ISD::SETUO) {
// If we aren't the ordered or unorder operation,
// then the pattern is (LHS CC1 RHS) Opc (LHS CC2 RHS).
SetCC1 = DAG.getSetCC(dl, VT, LHS, RHS, CC1, Chain, IsSignaling);
SetCC2 = DAG.getSetCC(dl, VT, LHS, RHS, CC2, Chain, IsSignaling);
} else {
// Otherwise, the pattern is (LHS CC1 LHS) Opc (RHS CC2 RHS)
SetCC1 = DAG.getSetCC(dl, VT, LHS, LHS, CC1, Chain, IsSignaling);
SetCC2 = DAG.getSetCC(dl, VT, RHS, RHS, CC2, Chain, IsSignaling);
}
if (Chain)
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, SetCC1.getValue(1),
SetCC2.getValue(1));
LHS = DAG.getNode(Opc, dl, VT, SetCC1, SetCC2);
RHS = SDValue();
CC = SDValue();
return true;
}
}
return false;
}
/// Emit a store/load combination to the stack. This stores
/// SrcOp to a stack slot of type SlotVT, truncating it if needed. It then does
/// a load from the stack slot to DestVT, extending it if needed.
/// The resultant code need not be legal.
SDValue SelectionDAGLegalize::EmitStackConvert(SDValue SrcOp, EVT SlotVT,
EVT DestVT, const SDLoc &dl) {
return EmitStackConvert(SrcOp, SlotVT, DestVT, dl, DAG.getEntryNode());
}
SDValue SelectionDAGLegalize::EmitStackConvert(SDValue SrcOp, EVT SlotVT,
EVT DestVT, const SDLoc &dl,
SDValue Chain) {
// Create the stack frame object.
unsigned SrcAlign = DAG.getDataLayout().getPrefTypeAlignment(
SrcOp.getValueType().getTypeForEVT(*DAG.getContext()));
SDValue FIPtr = DAG.CreateStackTemporary(SlotVT, SrcAlign);
FrameIndexSDNode *StackPtrFI = cast<FrameIndexSDNode>(FIPtr);
int SPFI = StackPtrFI->getIndex();
MachinePointerInfo PtrInfo =
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SPFI);
unsigned SrcSize = SrcOp.getValueSizeInBits();
unsigned SlotSize = SlotVT.getSizeInBits();
unsigned DestSize = DestVT.getSizeInBits();
Type *DestType = DestVT.getTypeForEVT(*DAG.getContext());
unsigned DestAlign = DAG.getDataLayout().getPrefTypeAlignment(DestType);
// Emit a store to the stack slot. Use a truncstore if the input value is
// later than DestVT.
SDValue Store;
if (SrcSize > SlotSize)
Store = DAG.getTruncStore(Chain, dl, SrcOp, FIPtr, PtrInfo,
SlotVT, SrcAlign);
else {
assert(SrcSize == SlotSize && "Invalid store");
Store =
DAG.getStore(Chain, dl, SrcOp, FIPtr, PtrInfo, SrcAlign);
}
// Result is a load from the stack slot.
if (SlotSize == DestSize)
return DAG.getLoad(DestVT, dl, Store, FIPtr, PtrInfo, DestAlign);
assert(SlotSize < DestSize && "Unknown extension!");
return DAG.getExtLoad(ISD::EXTLOAD, dl, DestVT, Store, FIPtr, PtrInfo, SlotVT,
DestAlign);
}
SDValue SelectionDAGLegalize::ExpandSCALAR_TO_VECTOR(SDNode *Node) {
SDLoc dl(Node);
// Create a vector sized/aligned stack slot, store the value to element #0,
// then load the whole vector back out.
SDValue StackPtr = DAG.CreateStackTemporary(Node->getValueType(0));
FrameIndexSDNode *StackPtrFI = cast<FrameIndexSDNode>(StackPtr);
int SPFI = StackPtrFI->getIndex();
SDValue Ch = DAG.getTruncStore(
DAG.getEntryNode(), dl, Node->getOperand(0), StackPtr,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SPFI),
Node->getValueType(0).getVectorElementType());
return DAG.getLoad(
Node->getValueType(0), dl, Ch, StackPtr,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SPFI));
}
static bool
ExpandBVWithShuffles(SDNode *Node, SelectionDAG &DAG,
const TargetLowering &TLI, SDValue &Res) {
unsigned NumElems = Node->getNumOperands();
SDLoc dl(Node);
EVT VT = Node->getValueType(0);
// Try to group the scalars into pairs, shuffle the pairs together, then
// shuffle the pairs of pairs together, etc. until the vector has
// been built. This will work only if all of the necessary shuffle masks
// are legal.
// We do this in two phases; first to check the legality of the shuffles,
// and next, assuming that all shuffles are legal, to create the new nodes.
for (int Phase = 0; Phase < 2; ++Phase) {
SmallVector<std::pair<SDValue, SmallVector<int, 16>>, 16> IntermedVals,
NewIntermedVals;
for (unsigned i = 0; i < NumElems; ++i) {
SDValue V = Node->getOperand(i);
if (V.isUndef())
continue;
SDValue Vec;
if (Phase)
Vec = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, V);
IntermedVals.push_back(std::make_pair(Vec, SmallVector<int, 16>(1, i)));
}
while (IntermedVals.size() > 2) {
NewIntermedVals.clear();
for (unsigned i = 0, e = (IntermedVals.size() & ~1u); i < e; i += 2) {
// This vector and the next vector are shuffled together (simply to
// append the one to the other).
SmallVector<int, 16> ShuffleVec(NumElems, -1);
SmallVector<int, 16> FinalIndices;
FinalIndices.reserve(IntermedVals[i].second.size() +
IntermedVals[i+1].second.size());
int k = 0;
for (unsigned j = 0, f = IntermedVals[i].second.size(); j != f;
++j, ++k) {
ShuffleVec[k] = j;
FinalIndices.push_back(IntermedVals[i].second[j]);
}
for (unsigned j = 0, f = IntermedVals[i+1].second.size(); j != f;
++j, ++k) {
ShuffleVec[k] = NumElems + j;
FinalIndices.push_back(IntermedVals[i+1].second[j]);
}
SDValue Shuffle;
if (Phase)
Shuffle = DAG.getVectorShuffle(VT, dl, IntermedVals[i].first,
IntermedVals[i+1].first,
ShuffleVec);
else if (!TLI.isShuffleMaskLegal(ShuffleVec, VT))
return false;
NewIntermedVals.push_back(
std::make_pair(Shuffle, std::move(FinalIndices)));
}
// If we had an odd number of defined values, then append the last
// element to the array of new vectors.
if ((IntermedVals.size() & 1) != 0)
NewIntermedVals.push_back(IntermedVals.back());
IntermedVals.swap(NewIntermedVals);
}
assert(IntermedVals.size() <= 2 && IntermedVals.size() > 0 &&
"Invalid number of intermediate vectors");
SDValue Vec1 = IntermedVals[0].first;
SDValue Vec2;
if (IntermedVals.size() > 1)
Vec2 = IntermedVals[1].first;
else if (Phase)
Vec2 = DAG.getUNDEF(VT);
SmallVector<int, 16> ShuffleVec(NumElems, -1);
for (unsigned i = 0, e = IntermedVals[0].second.size(); i != e; ++i)
ShuffleVec[IntermedVals[0].second[i]] = i;
for (unsigned i = 0, e = IntermedVals[1].second.size(); i != e; ++i)
ShuffleVec[IntermedVals[1].second[i]] = NumElems + i;
if (Phase)
Res = DAG.getVectorShuffle(VT, dl, Vec1, Vec2, ShuffleVec);
else if (!TLI.isShuffleMaskLegal(ShuffleVec, VT))
return false;
}
return true;
}
/// Expand a BUILD_VECTOR node on targets that don't
/// support the operation, but do support the resultant vector type.
SDValue SelectionDAGLegalize::ExpandBUILD_VECTOR(SDNode *Node) {
unsigned NumElems = Node->getNumOperands();
SDValue Value1, Value2;
SDLoc dl(Node);
EVT VT = Node->getValueType(0);
EVT OpVT = Node->getOperand(0).getValueType();
EVT EltVT = VT.getVectorElementType();
// If the only non-undef value is the low element, turn this into a
// SCALAR_TO_VECTOR node. If this is { X, X, X, X }, determine X.
bool isOnlyLowElement = true;
bool MoreThanTwoValues = false;
bool isConstant = true;
for (unsigned i = 0; i < NumElems; ++i) {
SDValue V = Node->getOperand(i);
if (V.isUndef())
continue;
if (i > 0)
isOnlyLowElement = false;
if (!isa<ConstantFPSDNode>(V) && !isa<ConstantSDNode>(V))
isConstant = false;
if (!Value1.getNode()) {
Value1 = V;
} else if (!Value2.getNode()) {
if (V != Value1)
Value2 = V;
} else if (V != Value1 && V != Value2) {
MoreThanTwoValues = true;
}
}
if (!Value1.getNode())
return DAG.getUNDEF(VT);
if (isOnlyLowElement)
return DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Node->getOperand(0));
// If all elements are constants, create a load from the constant pool.
if (isConstant) {
SmallVector<Constant*, 16> CV;
for (unsigned i = 0, e = NumElems; i != e; ++i) {
if (ConstantFPSDNode *V =
dyn_cast<ConstantFPSDNode>(Node->getOperand(i))) {
CV.push_back(const_cast<ConstantFP *>(V->getConstantFPValue()));
} else if (ConstantSDNode *V =
dyn_cast<ConstantSDNode>(Node->getOperand(i))) {
if (OpVT==EltVT)
CV.push_back(const_cast<ConstantInt *>(V->getConstantIntValue()));
else {
// If OpVT and EltVT don't match, EltVT is not legal and the
// element values have been promoted/truncated earlier. Undo this;
// we don't want a v16i8 to become a v16i32 for example.
const ConstantInt *CI = V->getConstantIntValue();
CV.push_back(ConstantInt::get(EltVT.getTypeForEVT(*DAG.getContext()),
CI->getZExtValue()));
}
} else {
assert(Node->getOperand(i).isUndef());
Type *OpNTy = EltVT.getTypeForEVT(*DAG.getContext());
CV.push_back(UndefValue::get(OpNTy));
}
}
Constant *CP = ConstantVector::get(CV);
SDValue CPIdx =
DAG.getConstantPool(CP, TLI.getPointerTy(DAG.getDataLayout()));
unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
return DAG.getLoad(
VT, dl, DAG.getEntryNode(), CPIdx,
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
Alignment);
}
SmallSet<SDValue, 16> DefinedValues;
for (unsigned i = 0; i < NumElems; ++i) {
if (Node->getOperand(i).isUndef())
continue;
DefinedValues.insert(Node->getOperand(i));
}
if (TLI.shouldExpandBuildVectorWithShuffles(VT, DefinedValues.size())) {
if (!MoreThanTwoValues) {
SmallVector<int, 8> ShuffleVec(NumElems, -1);
for (unsigned i = 0; i < NumElems; ++i) {
SDValue V = Node->getOperand(i);
if (V.isUndef())
continue;
ShuffleVec[i] = V == Value1 ? 0 : NumElems;
}
if (TLI.isShuffleMaskLegal(ShuffleVec, Node->getValueType(0))) {
// Get the splatted value into the low element of a vector register.
SDValue Vec1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value1);
SDValue Vec2;
if (Value2.getNode())
Vec2 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value2);
else
Vec2 = DAG.getUNDEF(VT);
// Return shuffle(LowValVec, undef, <0,0,0,0>)
return DAG.getVectorShuffle(VT, dl, Vec1, Vec2, ShuffleVec);
}
} else {
SDValue Res;
if (ExpandBVWithShuffles(Node, DAG, TLI, Res))
return Res;
}
}
// Otherwise, we can't handle this case efficiently.
return ExpandVectorBuildThroughStack(Node);
}
SDValue SelectionDAGLegalize::ExpandSPLAT_VECTOR(SDNode *Node) {
SDLoc DL(Node);
EVT VT = Node->getValueType(0);
SDValue SplatVal = Node->getOperand(0);
return DAG.getSplatBuildVector(VT, DL, SplatVal);
}
// Expand a node into a call to a libcall. If the result value
// does not fit into a register, return the lo part and set the hi part to the
// by-reg argument. If it does fit into a single register, return the result
// and leave the Hi part unset.
SDValue SelectionDAGLegalize::ExpandLibCall(RTLIB::Libcall LC, SDNode *Node,
bool isSigned) {
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
for (const SDValue &Op : Node->op_values()) {
EVT ArgVT = Op.getValueType();
Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
Entry.Node = Op;
Entry.Ty = ArgTy;
Entry.IsSExt = TLI.shouldSignExtendTypeInLibCall(ArgVT, isSigned);
Entry.IsZExt = !TLI.shouldSignExtendTypeInLibCall(ArgVT, isSigned);
Args.push_back(Entry);
}
SDValue Callee = DAG.getExternalSymbol(TLI.getLibcallName(LC),
TLI.getPointerTy(DAG.getDataLayout()));
EVT RetVT = Node->getValueType(0);
Type *RetTy = RetVT.getTypeForEVT(*DAG.getContext());
// By default, the input chain to this libcall is the entry node of the
// function. If the libcall is going to be emitted as a tail call then
// TLI.isUsedByReturnOnly will change it to the right chain if the return
// node which is being folded has a non-entry input chain.
SDValue InChain = DAG.getEntryNode();
// isTailCall may be true since the callee does not reference caller stack
// frame. Check if it's in the right position and that the return types match.
SDValue TCChain = InChain;
const Function &F = DAG.getMachineFunction().getFunction();
bool isTailCall =
TLI.isInTailCallPosition(DAG, Node, TCChain) &&
(RetTy == F.getReturnType() || F.getReturnType()->isVoidTy());
if (isTailCall)
InChain = TCChain;
TargetLowering::CallLoweringInfo CLI(DAG);
bool signExtend = TLI.shouldSignExtendTypeInLibCall(RetVT, isSigned);
CLI.setDebugLoc(SDLoc(Node))
.setChain(InChain)
.setLibCallee(TLI.getLibcallCallingConv(LC), RetTy, Callee,
std::move(Args))
.setTailCall(isTailCall)
.setSExtResult(signExtend)
.setZExtResult(!signExtend)
.setIsPostTypeLegalization(true);
std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
if (!CallInfo.second.getNode()) {
LLVM_DEBUG(dbgs() << "Created tailcall: "; DAG.getRoot().dump(&DAG));
// It's a tailcall, return the chain (which is the DAG root).
return DAG.getRoot();
}
LLVM_DEBUG(dbgs() << "Created libcall: "; CallInfo.first.dump(&DAG));
return CallInfo.first;
}
void SelectionDAGLegalize::ExpandFPLibCall(SDNode* Node,
RTLIB::Libcall Call_F32,
RTLIB::Libcall Call_F64,
RTLIB::Libcall Call_F80,
RTLIB::Libcall Call_F128,
RTLIB::Libcall Call_PPCF128,
SmallVectorImpl<SDValue> &Results) {
RTLIB::Libcall LC;
switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::f32: LC = Call_F32; break;
case MVT::f64: LC = Call_F64; break;
case MVT::f80: LC = Call_F80; break;
case MVT::f128: LC = Call_F128; break;
case MVT::ppcf128: LC = Call_PPCF128; break;
}
if (Node->isStrictFPOpcode()) {
EVT RetVT = Node->getValueType(0);
SmallVector<SDValue, 4> Ops(Node->op_begin() + 1, Node->op_end());
TargetLowering::MakeLibCallOptions CallOptions;
// FIXME: This doesn't support tail calls.
std::pair<SDValue, SDValue> Tmp = TLI.makeLibCall(DAG, LC, RetVT,
Ops, CallOptions,
SDLoc(Node),
Node->getOperand(0));
Results.push_back(Tmp.first);
Results.push_back(Tmp.second);
} else {
SDValue Tmp = ExpandLibCall(LC, Node, false);
Results.push_back(Tmp);
}
}
SDValue SelectionDAGLegalize::ExpandIntLibCall(SDNode* Node, bool isSigned,
RTLIB::Libcall Call_I8,
RTLIB::Libcall Call_I16,
RTLIB::Libcall Call_I32,
RTLIB::Libcall Call_I64,
RTLIB::Libcall Call_I128) {
RTLIB::Libcall LC;
switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::i8: LC = Call_I8; break;
case MVT::i16: LC = Call_I16; break;
case MVT::i32: LC = Call_I32; break;
case MVT::i64: LC = Call_I64; break;
case MVT::i128: LC = Call_I128; break;
}
return ExpandLibCall(LC, Node, isSigned);
}
/// Expand the node to a libcall based on first argument type (for instance
/// lround and its variant).
void SelectionDAGLegalize::ExpandArgFPLibCall(SDNode* Node,
RTLIB::Libcall Call_F32,
RTLIB::Libcall Call_F64,
RTLIB::Libcall Call_F80,
RTLIB::Libcall Call_F128,
RTLIB::Libcall Call_PPCF128,
SmallVectorImpl<SDValue> &Results) {
EVT InVT = Node->getOperand(Node->isStrictFPOpcode() ? 1 : 0).getValueType();
RTLIB::Libcall LC;
switch (InVT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::f32: LC = Call_F32; break;
case MVT::f64: LC = Call_F64; break;
case MVT::f80: LC = Call_F80; break;
case MVT::f128: LC = Call_F128; break;
case MVT::ppcf128: LC = Call_PPCF128; break;
}
if (Node->isStrictFPOpcode()) {
EVT RetVT = Node->getValueType(0);
SmallVector<SDValue, 4> Ops(Node->op_begin() + 1, Node->op_end());
TargetLowering::MakeLibCallOptions CallOptions;
// FIXME: This doesn't support tail calls.
std::pair<SDValue, SDValue> Tmp = TLI.makeLibCall(DAG, LC, RetVT,
Ops, CallOptions,
SDLoc(Node),
Node->getOperand(0));
Results.push_back(Tmp.first);
Results.push_back(Tmp.second);
} else {
SDValue Tmp = ExpandLibCall(LC, Node, false);
Results.push_back(Tmp);
}
}
/// Issue libcalls to __{u}divmod to compute div / rem pairs.
void
SelectionDAGLegalize::ExpandDivRemLibCall(SDNode *Node,
SmallVectorImpl<SDValue> &Results) {
unsigned Opcode = Node->getOpcode();
bool isSigned = Opcode == ISD::SDIVREM;
RTLIB::Libcall LC;
switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::i8: LC= isSigned ? RTLIB::SDIVREM_I8 : RTLIB::UDIVREM_I8; break;
case MVT::i16: LC= isSigned ? RTLIB::SDIVREM_I16 : RTLIB::UDIVREM_I16; break;
case MVT::i32: LC= isSigned ? RTLIB::SDIVREM_I32 : RTLIB::UDIVREM_I32; break;
case MVT::i64: LC= isSigned ? RTLIB::SDIVREM_I64 : RTLIB::UDIVREM_I64; break;
case MVT::i128: LC= isSigned ? RTLIB::SDIVREM_I128:RTLIB::UDIVREM_I128; break;
}
// The input chain to this libcall is the entry node of the function.
// Legalizing the call will automatically add the previous call to the
// dependence.
SDValue InChain = DAG.getEntryNode();
EVT RetVT = Node->getValueType(0);
Type *RetTy = RetVT.getTypeForEVT(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
for (const SDValue &Op : Node->op_values()) {
EVT ArgVT = Op.getValueType();
Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
Entry.Node = Op;
Entry.Ty = ArgTy;
Entry.IsSExt = isSigned;
Entry.IsZExt = !isSigned;
Args.push_back(Entry);
}
// Also pass the return address of the remainder.
SDValue FIPtr = DAG.CreateStackTemporary(RetVT);
Entry.Node = FIPtr;
Entry.Ty = RetTy->getPointerTo();
Entry.IsSExt = isSigned;
Entry.IsZExt = !isSigned;
Args.push_back(Entry);
SDValue Callee = DAG.getExternalSymbol(TLI.getLibcallName(LC),
TLI.getPointerTy(DAG.getDataLayout()));
SDLoc dl(Node);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl)
.setChain(InChain)
.setLibCallee(TLI.getLibcallCallingConv(LC), RetTy, Callee,
std::move(Args))
.setSExtResult(isSigned)
.setZExtResult(!isSigned);
std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
// Remainder is loaded back from the stack frame.
SDValue Rem =
DAG.getLoad(RetVT, dl, CallInfo.second, FIPtr, MachinePointerInfo());
Results.push_back(CallInfo.first);
Results.push_back(Rem);
}
/// Return true if sincos libcall is available.
static bool isSinCosLibcallAvailable(SDNode *Node, const TargetLowering &TLI) {
RTLIB::Libcall LC;
switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::f32: LC = RTLIB::SINCOS_F32; break;
case MVT::f64: LC = RTLIB::SINCOS_F64; break;
case MVT::f80: LC = RTLIB::SINCOS_F80; break;
case MVT::f128: LC = RTLIB::SINCOS_F128; break;
case MVT::ppcf128: LC = RTLIB::SINCOS_PPCF128; break;
}
return TLI.getLibcallName(LC) != nullptr;
}
/// Only issue sincos libcall if both sin and cos are needed.
static bool useSinCos(SDNode *Node) {
unsigned OtherOpcode = Node->getOpcode() == ISD::FSIN
? ISD::FCOS : ISD::FSIN;
SDValue Op0 = Node->getOperand(0);
for (SDNode::use_iterator UI = Op0.getNode()->use_begin(),
UE = Op0.getNode()->use_end(); UI != UE; ++UI) {
SDNode *User = *UI;
if (User == Node)
continue;
// The other user might have been turned into sincos already.
if (User->getOpcode() == OtherOpcode || User->getOpcode() == ISD::FSINCOS)
return true;
}
return false;
}
/// Issue libcalls to sincos to compute sin / cos pairs.
void
SelectionDAGLegalize::ExpandSinCosLibCall(SDNode *Node,
SmallVectorImpl<SDValue> &Results) {
RTLIB::Libcall LC;
switch (Node->getSimpleValueType(0).SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::f32: LC = RTLIB::SINCOS_F32; break;
case MVT::f64: LC = RTLIB::SINCOS_F64; break;
case MVT::f80: LC = RTLIB::SINCOS_F80; break;
case MVT::f128: LC = RTLIB::SINCOS_F128; break;
case MVT::ppcf128: LC = RTLIB::SINCOS_PPCF128; break;
}
// The input chain to this libcall is the entry node of the function.
// Legalizing the call will automatically add the previous call to the
// dependence.
SDValue InChain = DAG.getEntryNode();
EVT RetVT = Node->getValueType(0);
Type *RetTy = RetVT.getTypeForEVT(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
// Pass the argument.
Entry.Node = Node->getOperand(0);
Entry.Ty = RetTy;
Entry.IsSExt = false;
Entry.IsZExt = false;
Args.push_back(Entry);
// Pass the return address of sin.
SDValue SinPtr = DAG.CreateStackTemporary(RetVT);
Entry.Node = SinPtr;
Entry.Ty = RetTy->getPointerTo();
Entry.IsSExt = false;
Entry.IsZExt = false;
Args.push_back(Entry);
// Also pass the return address of the cos.
SDValue CosPtr = DAG.CreateStackTemporary(RetVT);
Entry.Node = CosPtr;
Entry.Ty = RetTy->getPointerTo();
Entry.IsSExt = false;
Entry.IsZExt = false;
Args.push_back(Entry);
SDValue Callee = DAG.getExternalSymbol(TLI.getLibcallName(LC),
TLI.getPointerTy(DAG.getDataLayout()));
SDLoc dl(Node);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(InChain).setLibCallee(
TLI.getLibcallCallingConv(LC), Type::getVoidTy(*DAG.getContext()), Callee,
std::move(Args));
std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
Results.push_back(
DAG.getLoad(RetVT, dl, CallInfo.second, SinPtr, MachinePointerInfo()));
Results.push_back(
DAG.getLoad(RetVT, dl, CallInfo.second, CosPtr, MachinePointerInfo()));
}
/// This function is responsible for legalizing a
/// INT_TO_FP operation of the specified operand when the target requests that
/// we expand it. At this point, we know that the result and operand types are
/// legal for the target.
SDValue SelectionDAGLegalize::ExpandLegalINT_TO_FP(SDNode *Node,
SDValue &Chain) {
bool isSigned = (Node->getOpcode() == ISD::STRICT_SINT_TO_FP ||
Node->getOpcode() == ISD::SINT_TO_FP);
EVT DestVT = Node->getValueType(0);
SDLoc dl(Node);
unsigned OpNo = Node->isStrictFPOpcode() ? 1 : 0;
SDValue Op0 = Node->getOperand(OpNo);
EVT SrcVT = Op0.getValueType();
// TODO: Should any fast-math-flags be set for the created nodes?
LLVM_DEBUG(dbgs() << "Legalizing INT_TO_FP\n");
if (SrcVT == MVT::i32 && TLI.isTypeLegal(MVT::f64)) {
LLVM_DEBUG(dbgs() << "32-bit [signed|unsigned] integer to float/double "
"expansion\n");
// Get the stack frame index of a 8 byte buffer.
SDValue StackSlot = DAG.CreateStackTemporary(MVT::f64);
// word offset constant for Hi/Lo address computation
SDValue WordOff = DAG.getConstant(sizeof(int), dl,
StackSlot.getValueType());
// set up Hi and Lo (into buffer) address based on endian
SDValue Hi = StackSlot;
SDValue Lo = DAG.getNode(ISD::ADD, dl, StackSlot.getValueType(),
StackSlot, WordOff);
if (DAG.getDataLayout().isLittleEndian())
std::swap(Hi, Lo);
// if signed map to unsigned space
SDValue Op0Mapped;
if (isSigned) {
// constant used to invert sign bit (signed to unsigned mapping)
SDValue SignBit = DAG.getConstant(0x80000000u, dl, MVT::i32);
Op0Mapped = DAG.getNode(ISD::XOR, dl, MVT::i32, Op0, SignBit);
} else {
Op0Mapped = Op0;
}
// store the lo of the constructed double - based on integer input
SDValue Store1 = DAG.getStore(DAG.getEntryNode(), dl, Op0Mapped, Lo,
MachinePointerInfo());
// initial hi portion of constructed double
SDValue InitialHi = DAG.getConstant(0x43300000u, dl, MVT::i32);
// store the hi of the constructed double - biased exponent
SDValue Store2 =
DAG.getStore(Store1, dl, InitialHi, Hi, MachinePointerInfo());
// load the constructed double
SDValue Load =
DAG.getLoad(MVT::f64, dl, Store2, StackSlot, MachinePointerInfo());
// FP constant to bias correct the final result
SDValue Bias = DAG.getConstantFP(isSigned ?
BitsToDouble(0x4330000080000000ULL) :
BitsToDouble(0x4330000000000000ULL),
dl, MVT::f64);
// Subtract the bias and get the final result.
SDValue Sub;
SDValue Result;
if (Node->isStrictFPOpcode()) {
Sub = DAG.getNode(ISD::STRICT_FSUB, dl, {MVT::f64, MVT::Other},
{Node->getOperand(0), Load, Bias});
Chain = Sub.getValue(1);
if (DestVT != Sub.getValueType()) {
std::pair<SDValue, SDValue> ResultPair;
ResultPair =
DAG.getStrictFPExtendOrRound(Sub, Chain, dl, DestVT);
Result = ResultPair.first;
Chain = ResultPair.second;
}
else
Result = Sub;
} else {
Sub = DAG.getNode(ISD::FSUB, dl, MVT::f64, Load, Bias);
Result = DAG.getFPExtendOrRound(Sub, dl, DestVT);
}
return Result;
}
assert(!isSigned && "Legalize cannot Expand SINT_TO_FP for i64 yet");
// Code below here assumes !isSigned without checking again.
// FIXME: This can produce slightly incorrect results. See details in
// FIXME: https://reviews.llvm.org/D69275
SDValue Tmp1;
if (Node->isStrictFPOpcode()) {
Tmp1 = DAG.getNode(ISD::STRICT_SINT_TO_FP, dl, { DestVT, MVT::Other },
{ Node->getOperand(0), Op0 });
} else
Tmp1 = DAG.getNode(ISD::SINT_TO_FP, dl, DestVT, Op0);
SDValue SignSet = DAG.getSetCC(dl, getSetCCResultType(SrcVT), Op0,
DAG.getConstant(0, dl, SrcVT), ISD::SETLT);
SDValue Zero = DAG.getIntPtrConstant(0, dl),
Four = DAG.getIntPtrConstant(4, dl);
SDValue CstOffset = DAG.getSelect(dl, Zero.getValueType(),
SignSet, Four, Zero);
// If the sign bit of the integer is set, the large number will be treated
// as a negative number. To counteract this, the dynamic code adds an
// offset depending on the data type.
uint64_t FF;
switch (SrcVT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unsupported integer type!");
case MVT::i8 : FF = 0x43800000ULL; break; // 2^8 (as a float)
case MVT::i16: FF = 0x47800000ULL; break; // 2^16 (as a float)
case MVT::i32: FF = 0x4F800000ULL; break; // 2^32 (as a float)
case MVT::i64: FF = 0x5F800000ULL; break; // 2^64 (as a float)
}
if (DAG.getDataLayout().isLittleEndian())
FF <<= 32;
Constant *FudgeFactor = ConstantInt::get(
Type::getInt64Ty(*DAG.getContext()), FF);
SDValue CPIdx =
DAG.getConstantPool(FudgeFactor, TLI.getPointerTy(DAG.getDataLayout()));
unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
CPIdx = DAG.getNode(ISD::ADD, dl, CPIdx.getValueType(), CPIdx, CstOffset);
Alignment = std::min(Alignment, 4u);
SDValue FudgeInReg;
if (DestVT == MVT::f32)
FudgeInReg = DAG.getLoad(
MVT::f32, dl, DAG.getEntryNode(), CPIdx,
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
Alignment);
else {
SDValue Load = DAG.getExtLoad(
ISD::EXTLOAD, dl, DestVT, DAG.getEntryNode(), CPIdx,
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()), MVT::f32,
Alignment);
HandleSDNode Handle(Load);
LegalizeOp(Load.getNode());
FudgeInReg = Handle.getValue();
}
if (Node->isStrictFPOpcode()) {
SDValue Result = DAG.getNode(ISD::STRICT_FADD, dl, { DestVT, MVT::Other },