#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/Analysis/AliasAnalysis.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
SDValue visitOR(SDNode *N);
SDValue visitXOR(SDNode *N);
SDValue SimplifyVBinOp(SDNode *N);
+ SDValue SimplifyVUnaryOp(SDNode *N);
SDValue visitSHL(SDNode *N);
SDValue visitSRA(SDNode *N);
SDValue visitSRL(SDNode *N);
SDValue visitFP_EXTEND(SDNode *N);
SDValue visitFNEG(SDNode *N);
SDValue visitFABS(SDNode *N);
+ SDValue visitFCEIL(SDNode *N);
+ SDValue visitFTRUNC(SDNode *N);
+ SDValue visitFFLOOR(SDNode *N);
SDValue visitBRCOND(SDNode *N);
SDValue visitBR_CC(SDNode *N);
SDValue visitLOAD(SDNode *N);
/// looking for a better chain (aliasing node.)
SDValue FindBetterChain(SDNode *N, SDValue Chain);
+ /// Merge consecutive store operations into a wide store.
+ /// This optimization uses wide integers or vectors when possible.
+ /// \return True if some memory operations were changed.
+ bool MergeConsecutiveStores(StoreSDNode *N);
+
public:
DAGCombiner(SelectionDAG &D, AliasAnalysis &A, CodeGenOpt::Level OL)
: DAG(D), TLI(D.getTargetLoweringInfo()), Level(BeforeLegalizeTypes),
!TLI.isOperationLegalOrCustom(ISD::FSUB, Op.getValueType()))
return 0;
- // fold (fsub (fadd A, B)) -> (fsub (fneg A), B)
+ // fold (fneg (fadd A, B)) -> (fsub (fneg A), B)
if (char V = isNegatibleForFree(Op.getOperand(0), LegalOperations, TLI,
Options, Depth + 1))
return V;
case ISD::FP_EXTEND: return visitFP_EXTEND(N);
case ISD::FNEG: return visitFNEG(N);
case ISD::FABS: return visitFABS(N);
+ case ISD::FFLOOR: return visitFFLOOR(N);
+ case ISD::FCEIL: return visitFCEIL(N);
+ case ISD::FTRUNC: return visitFTRUNC(N);
case ISD::BRCOND: return visitBRCOND(N);
case ISD::BR_CC: return visitBR_CC(N);
case ISD::LOAD: return visitLOAD(N);
return N0.getOperand(0);
// fold C2-(A+C1) -> (C2-C1)-A
if (N1.getOpcode() == ISD::ADD && N0C && N1C1) {
- SDValue NewC = DAG.getConstant((N0C->getAPIntValue() - N1C1->getAPIntValue()), VT);
+ SDValue NewC = DAG.getConstant(N0C->getAPIntValue() - N1C1->getAPIntValue(),
+ VT);
return DAG.getNode(ISD::SUB, N->getDebugLoc(), VT, NewC,
N1.getOperand(0));
}
// we don't want to undo this promotion.
// We also handle SCALAR_TO_VECTOR because xor/or/and operations are cheaper
// on scalars.
- if ((N0.getOpcode() == ISD::BITCAST || N0.getOpcode() == ISD::SCALAR_TO_VECTOR)
- && Level == AfterLegalizeTypes) {
+ if ((N0.getOpcode() == ISD::BITCAST ||
+ N0.getOpcode() == ISD::SCALAR_TO_VECTOR) &&
+ Level == AfterLegalizeTypes) {
SDValue In0 = N0.getOperand(0);
SDValue In1 = N1.getOperand(0);
EVT In0Ty = In0.getValueType();
EVT In1Ty = In1.getValueType();
- // If both incoming values are integers, and the original types are the same.
+ DebugLoc DL = N->getDebugLoc();
+ // If both incoming values are integers, and the original types are the
+ // same.
if (In0Ty.isInteger() && In1Ty.isInteger() && In0Ty == In1Ty) {
- SDValue Op = DAG.getNode(N->getOpcode(), N->getDebugLoc(), In0Ty, In0, In1);
- SDValue BC = DAG.getNode(N0.getOpcode(), N->getDebugLoc(), VT, Op);
+ SDValue Op = DAG.getNode(N->getOpcode(), DL, In0Ty, In0, In1);
+ SDValue BC = DAG.getNode(N0.getOpcode(), DL, VT, Op);
AddToWorkList(Op.getNode());
return BC;
}
// lanes of the constant together.
EVT VT = Vector->getValueType(0);
unsigned BitWidth = VT.getVectorElementType().getSizeInBits();
+
+ // If the splat value has been compressed to a bitlength lower
+ // than the size of the vector lane, we need to re-expand it to
+ // the lane size.
+ if (BitWidth > SplatBitSize)
+ for (SplatValue = SplatValue.zextOrTrunc(BitWidth);
+ SplatBitSize < BitWidth;
+ SplatBitSize = SplatBitSize * 2)
+ SplatValue |= SplatValue.shl(SplatBitSize);
+
Constant = APInt::getAllOnesValue(BitWidth);
- for (unsigned i = 0, n = VT.getVectorNumElements(); i < n; ++i)
+ for (unsigned i = 0, n = SplatBitSize/BitWidth; i < n; ++i)
Constant &= SplatValue.lshr(i*BitWidth).zextOrTrunc(BitWidth);
}
}
SDValue ShAmt = DAG.getConstant(16, getShiftAmountTy(VT));
if (TLI.isOperationLegalOrCustom(ISD::ROTL, VT))
return DAG.getNode(ISD::ROTL, N->getDebugLoc(), VT, BSwap, ShAmt);
- else if (TLI.isOperationLegalOrCustom(ISD::ROTR, VT))
+ if (TLI.isOperationLegalOrCustom(ISD::ROTR, VT))
return DAG.getNode(ISD::ROTR, N->getDebugLoc(), VT, BSwap, ShAmt);
return DAG.getNode(ISD::OR, N->getDebugLoc(), VT,
DAG.getNode(ISD::SHL, N->getDebugLoc(), VT, BSwap, ShAmt),
if ((LShVal + RShVal) != OpSizeInBits)
return 0;
- SDValue Rot;
- if (HasROTL)
- Rot = DAG.getNode(ISD::ROTL, DL, VT, LHSShiftArg, LHSShiftAmt);
- else
- Rot = DAG.getNode(ISD::ROTR, DL, VT, LHSShiftArg, RHSShiftAmt);
+ SDValue Rot = DAG.getNode(HasROTL ? ISD::ROTL : ISD::ROTR, DL, VT,
+ LHSShiftArg, HasROTL ? LHSShiftAmt : RHSShiftAmt);
// If there is an AND of either shifted operand, apply it to the result.
if (LHSMask.getNode() || RHSMask.getNode()) {
if (ConstantSDNode *SUBC =
dyn_cast<ConstantSDNode>(RHSShiftAmt.getOperand(0))) {
if (SUBC->getAPIntValue() == OpSizeInBits) {
- if (HasROTL)
- return DAG.getNode(ISD::ROTL, DL, VT,
- LHSShiftArg, LHSShiftAmt).getNode();
- else
- return DAG.getNode(ISD::ROTR, DL, VT,
- LHSShiftArg, RHSShiftAmt).getNode();
+ return DAG.getNode(HasROTL ? ISD::ROTL : ISD::ROTR, DL, VT, LHSShiftArg,
+ HasROTL ? LHSShiftAmt : RHSShiftAmt).getNode();
}
}
}
if (ConstantSDNode *SUBC =
dyn_cast<ConstantSDNode>(LHSShiftAmt.getOperand(0))) {
if (SUBC->getAPIntValue() == OpSizeInBits) {
- if (HasROTR)
- return DAG.getNode(ISD::ROTR, DL, VT,
- LHSShiftArg, RHSShiftAmt).getNode();
- else
- return DAG.getNode(ISD::ROTL, DL, VT,
- LHSShiftArg, LHSShiftAmt).getNode();
+ return DAG.getNode(HasROTR ? ISD::ROTR : ISD::ROTL, DL, VT, LHSShiftArg,
+ HasROTR ? RHSShiftAmt : LHSShiftAmt).getNode();
}
}
}
// Look for sign/zext/any-extended or truncate cases:
- if ((LHSShiftAmt.getOpcode() == ISD::SIGN_EXTEND
- || LHSShiftAmt.getOpcode() == ISD::ZERO_EXTEND
- || LHSShiftAmt.getOpcode() == ISD::ANY_EXTEND
- || LHSShiftAmt.getOpcode() == ISD::TRUNCATE) &&
- (RHSShiftAmt.getOpcode() == ISD::SIGN_EXTEND
- || RHSShiftAmt.getOpcode() == ISD::ZERO_EXTEND
- || RHSShiftAmt.getOpcode() == ISD::ANY_EXTEND
- || RHSShiftAmt.getOpcode() == ISD::TRUNCATE)) {
+ if ((LHSShiftAmt.getOpcode() == ISD::SIGN_EXTEND ||
+ LHSShiftAmt.getOpcode() == ISD::ZERO_EXTEND ||
+ LHSShiftAmt.getOpcode() == ISD::ANY_EXTEND ||
+ LHSShiftAmt.getOpcode() == ISD::TRUNCATE) &&
+ (RHSShiftAmt.getOpcode() == ISD::SIGN_EXTEND ||
+ RHSShiftAmt.getOpcode() == ISD::ZERO_EXTEND ||
+ RHSShiftAmt.getOpcode() == ISD::ANY_EXTEND ||
+ RHSShiftAmt.getOpcode() == ISD::TRUNCATE)) {
SDValue LExtOp0 = LHSShiftAmt.getOperand(0);
SDValue RExtOp0 = RHSShiftAmt.getOperand(0);
if (RExtOp0.getOpcode() == ISD::SUB &&
if (VT.isInteger() &&
(VT0 == MVT::i1 ||
(VT0.isInteger() &&
- TLI.getBooleanContents(false) == TargetLowering::ZeroOrOneBooleanContent)) &&
+ TLI.getBooleanContents(false) ==
+ TargetLowering::ZeroOrOneBooleanContent)) &&
N1C && N2C && N1C->isNullValue() && N2C->getAPIntValue() == 1) {
SDValue XORNode;
if (VT == VT0)
// If the desired elements are smaller or larger than the source
// elements we can use a matching integer vector type and then
// truncate/sign extend
- else {
- EVT MatchingElementType =
- EVT::getIntegerVT(*DAG.getContext(),
- N0VT.getScalarType().getSizeInBits());
- EVT MatchingVectorType =
- EVT::getVectorVT(*DAG.getContext(), MatchingElementType,
- N0VT.getVectorNumElements());
+ EVT MatchingElementType =
+ EVT::getIntegerVT(*DAG.getContext(),
+ N0VT.getScalarType().getSizeInBits());
+ EVT MatchingVectorType =
+ EVT::getVectorVT(*DAG.getContext(), MatchingElementType,
+ N0VT.getVectorNumElements());
- if (SVT == MatchingVectorType) {
- SDValue VsetCC = DAG.getSetCC(N->getDebugLoc(), MatchingVectorType,
- N0.getOperand(0), N0.getOperand(1),
- cast<CondCodeSDNode>(N0.getOperand(2))->get());
- return DAG.getSExtOrTrunc(VsetCC, N->getDebugLoc(), VT);
- }
+ if (SVT == MatchingVectorType) {
+ SDValue VsetCC = DAG.getSetCC(N->getDebugLoc(), MatchingVectorType,
+ N0.getOperand(0), N0.getOperand(1),
+ cast<CondCodeSDNode>(N0.getOperand(2))->get());
+ return DAG.getSExtOrTrunc(VsetCC, N->getDebugLoc(), VT);
}
}
// if the source is smaller than the dest, we still need an extend
return DAG.getNode(N0.getOpcode(), N->getDebugLoc(), VT,
N0.getOperand(0));
- else if (N0.getOperand(0).getValueType().bitsGT(VT))
+ if (N0.getOperand(0).getValueType().bitsGT(VT))
// if the source is larger than the dest, than we just need the truncate
return DAG.getNode(ISD::TRUNCATE, N->getDebugLoc(), VT, N0.getOperand(0));
- else
- // if the source and dest are the same type, we can drop both the extend
- // and the truncate.
- return N0.getOperand(0);
+ // if the source and dest are the same type, we can drop both the extend
+ // and the truncate.
+ return N0.getOperand(0);
}
// Fold extract-and-trunc into a narrow extract. For example:
!LD2->isVolatile() &&
DAG.isConsecutiveLoad(LD2, LD1, LD1VT.getSizeInBits()/8, 1)) {
unsigned Align = LD1->getAlignment();
- unsigned NewAlign = TLI.getTargetData()->
+ unsigned NewAlign = TLI.getDataLayout()->
getABITypeAlignment(VT.getTypeForEVT(*DAG.getContext()));
if (NewAlign <= Align &&
!cast<LoadSDNode>(N0)->isVolatile() &&
(!LegalOperations || TLI.isOperationLegal(ISD::LOAD, VT))) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
- unsigned Align = TLI.getTargetData()->
+ unsigned Align = TLI.getDataLayout()->
getABITypeAlignment(VT.getTypeForEVT(*DAG.getContext()));
unsigned OrigAlign = LN0->getAlignment();
// This often reduces constant pool loads.
if (((N0.getOpcode() == ISD::FNEG && !TLI.isFNegFree(VT)) ||
(N0.getOpcode() == ISD::FABS && !TLI.isFAbsFree(VT))) &&
- N0.getNode()->hasOneUse() && VT.isInteger() && !VT.isVector()) {
+ N0.getNode()->hasOneUse() && VT.isInteger() &&
+ !VT.isVector() && !N0.getValueType().isVector()) {
SDValue NewConv = DAG.getNode(ISD::BITCAST, N0.getDebugLoc(), VT,
N0.getOperand(0));
AddToWorkList(NewConv.getNode());
return N0;
// fold (fadd A, (fneg B)) -> (fsub A, B)
if ((!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FSUB, VT)) &&
- isNegatibleForFree(N1, LegalOperations, TLI, &DAG.getTarget().Options) == 2)
+ isNegatibleForFree(N1, LegalOperations, TLI, &DAG.getTarget().Options) == 2)
return DAG.getNode(ISD::FSUB, N->getDebugLoc(), VT, N0,
GetNegatedExpression(N1, DAG, LegalOperations));
// fold (fadd (fneg A), B) -> (fsub B, A)
if ((!LegalOperations || TLI.isOperationLegalOrCustom(ISD::FSUB, VT)) &&
- isNegatibleForFree(N0, LegalOperations, TLI, &DAG.getTarget().Options) == 2)
+ isNegatibleForFree(N0, LegalOperations, TLI, &DAG.getTarget().Options) == 2)
return DAG.getNode(ISD::FSUB, N->getDebugLoc(), VT, N1,
GetNegatedExpression(N0, DAG, LegalOperations));
DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
N0.getOperand(1), N1));
+ // In unsafe math mode, we can fold chains of FADD's of the same value
+ // into multiplications. This transform is not safe in general because
+ // we are reducing the number of rounding steps.
+ if (DAG.getTarget().Options.UnsafeFPMath &&
+ TLI.isOperationLegalOrCustom(ISD::FMUL, VT) &&
+ !N0CFP && !N1CFP) {
+ if (N0.getOpcode() == ISD::FMUL) {
+ ConstantFPSDNode *CFP00 = dyn_cast<ConstantFPSDNode>(N0.getOperand(0));
+ ConstantFPSDNode *CFP01 = dyn_cast<ConstantFPSDNode>(N0.getOperand(1));
+
+ // (fadd (fmul c, x), x) -> (fmul c+1, x)
+ if (CFP00 && !CFP01 && N0.getOperand(1) == N1) {
+ SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+ SDValue(CFP00, 0),
+ DAG.getConstantFP(1.0, VT));
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N1, NewCFP);
+ }
+
+ // (fadd (fmul x, c), x) -> (fmul c+1, x)
+ if (CFP01 && !CFP00 && N0.getOperand(0) == N1) {
+ SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+ SDValue(CFP01, 0),
+ DAG.getConstantFP(1.0, VT));
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N1, NewCFP);
+ }
+
+ // (fadd (fadd x, x), x) -> (fmul 3.0, x)
+ if (!CFP00 && !CFP01 && N0.getOperand(0) == N0.getOperand(1) &&
+ N0.getOperand(0) == N1) {
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N1, DAG.getConstantFP(3.0, VT));
+ }
+
+ // (fadd (fmul c, x), (fadd x, x)) -> (fmul c+2, x)
+ if (CFP00 && !CFP01 && N1.getOpcode() == ISD::FADD &&
+ N1.getOperand(0) == N1.getOperand(1) &&
+ N0.getOperand(1) == N1.getOperand(0)) {
+ SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+ SDValue(CFP00, 0),
+ DAG.getConstantFP(2.0, VT));
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0.getOperand(1), NewCFP);
+ }
+
+ // (fadd (fmul x, c), (fadd x, x)) -> (fmul c+2, x)
+ if (CFP01 && !CFP00 && N1.getOpcode() == ISD::FADD &&
+ N1.getOperand(0) == N1.getOperand(1) &&
+ N0.getOperand(0) == N1.getOperand(0)) {
+ SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+ SDValue(CFP01, 0),
+ DAG.getConstantFP(2.0, VT));
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0.getOperand(0), NewCFP);
+ }
+ }
+
+ if (N1.getOpcode() == ISD::FMUL) {
+ ConstantFPSDNode *CFP10 = dyn_cast<ConstantFPSDNode>(N1.getOperand(0));
+ ConstantFPSDNode *CFP11 = dyn_cast<ConstantFPSDNode>(N1.getOperand(1));
+
+ // (fadd x, (fmul c, x)) -> (fmul c+1, x)
+ if (CFP10 && !CFP11 && N1.getOperand(1) == N0) {
+ SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+ SDValue(CFP10, 0),
+ DAG.getConstantFP(1.0, VT));
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0, NewCFP);
+ }
+
+ // (fadd x, (fmul x, c)) -> (fmul c+1, x)
+ if (CFP11 && !CFP10 && N1.getOperand(0) == N0) {
+ SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+ SDValue(CFP11, 0),
+ DAG.getConstantFP(1.0, VT));
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0, NewCFP);
+ }
+
+ // (fadd x, (fadd x, x)) -> (fmul 3.0, x)
+ if (!CFP10 && !CFP11 && N1.getOperand(0) == N1.getOperand(1) &&
+ N1.getOperand(0) == N0) {
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0, DAG.getConstantFP(3.0, VT));
+ }
+
+ // (fadd (fadd x, x), (fmul c, x)) -> (fmul c+2, x)
+ if (CFP10 && !CFP11 && N1.getOpcode() == ISD::FADD &&
+ N1.getOperand(0) == N1.getOperand(1) &&
+ N0.getOperand(1) == N1.getOperand(0)) {
+ SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+ SDValue(CFP10, 0),
+ DAG.getConstantFP(2.0, VT));
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0.getOperand(1), NewCFP);
+ }
+
+ // (fadd (fadd x, x), (fmul x, c)) -> (fmul c+2, x)
+ if (CFP11 && !CFP10 && N1.getOpcode() == ISD::FADD &&
+ N1.getOperand(0) == N1.getOperand(1) &&
+ N0.getOperand(0) == N1.getOperand(0)) {
+ SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+ SDValue(CFP11, 0),
+ DAG.getConstantFP(2.0, VT));
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0.getOperand(0), NewCFP);
+ }
+ }
+
+ // (fadd (fadd x, x), (fadd x, x)) -> (fmul 4.0, x)
+ if (N0.getOpcode() == ISD::FADD && N1.getOpcode() == ISD::FADD &&
+ N0.getOperand(0) == N0.getOperand(1) &&
+ N1.getOperand(0) == N1.getOperand(1) &&
+ N0.getOperand(0) == N1.getOperand(0)) {
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0.getOperand(0),
+ DAG.getConstantFP(4.0, VT));
+ }
+ }
+
// FADD -> FMA combines:
if ((DAG.getTarget().Options.AllowFPOpFusion == FPOpFusion::Fast ||
DAG.getTarget().Options.UnsafeFPMath) &&
DAG.getTarget().getTargetLowering()->isFMAFasterThanMulAndAdd(VT) &&
- TLI.isOperationLegal(ISD::FMA, VT)) {
+ TLI.isOperationLegalOrCustom(ISD::FMA, VT)) {
// fold (fadd (fmul x, y), z) -> (fma x, y, z)
if (N0.getOpcode() == ISD::FMUL && N0->hasOneUse()) {
return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT,
N0.getOperand(0), N0.getOperand(1), N1);
}
-
- // fold (fadd x, (fmul y, z)) -> (fma x, y, z)
+
+ // fold (fadd x, (fmul y, z)) -> (fma y, z, x)
// Note: Commutes FADD operands.
if (N1.getOpcode() == ISD::FMUL && N1->hasOneUse()) {
return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT,
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
EVT VT = N->getValueType(0);
+ DebugLoc dl = N->getDebugLoc();
// fold vector ops
if (VT.isVector()) {
if (isNegatibleForFree(N1, LegalOperations, TLI, &DAG.getTarget().Options))
return GetNegatedExpression(N1, DAG, LegalOperations);
if (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT))
- return DAG.getNode(ISD::FNEG, N->getDebugLoc(), VT, N1);
+ return DAG.getNode(ISD::FNEG, dl, VT, N1);
}
// fold (fsub A, (fneg B)) -> (fadd A, B)
if (isNegatibleForFree(N1, LegalOperations, TLI, &DAG.getTarget().Options))
- return DAG.getNode(ISD::FADD, N->getDebugLoc(), VT, N0,
+ return DAG.getNode(ISD::FADD, dl, VT, N0,
GetNegatedExpression(N1, DAG, LegalOperations));
// If 'unsafe math' is enabled, fold
if ((DAG.getTarget().Options.AllowFPOpFusion == FPOpFusion::Fast ||
DAG.getTarget().Options.UnsafeFPMath) &&
DAG.getTarget().getTargetLowering()->isFMAFasterThanMulAndAdd(VT) &&
- TLI.isOperationLegal(ISD::FMA, VT)) {
+ TLI.isOperationLegalOrCustom(ISD::FMA, VT)) {
// fold (fsub (fmul x, y), z) -> (fma x, y, (fneg z))
if (N0.getOpcode() == ISD::FMUL && N0->hasOneUse()) {
- return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT,
+ return DAG.getNode(ISD::FMA, dl, VT,
N0.getOperand(0), N0.getOperand(1),
- DAG.getNode(ISD::FNEG, N1->getDebugLoc(), VT, N1));
+ DAG.getNode(ISD::FNEG, dl, VT, N1));
}
// fold (fsub x, (fmul y, z)) -> (fma (fneg y), z, x)
// Note: Commutes FSUB operands.
if (N1.getOpcode() == ISD::FMUL && N1->hasOneUse()) {
- return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT,
- DAG.getNode(ISD::FNEG, N1->getDebugLoc(), VT,
+ return DAG.getNode(ISD::FMA, dl, VT,
+ DAG.getNode(ISD::FNEG, dl, VT,
N1.getOperand(0)),
N1.getOperand(1), N0);
}
+
+ // fold (fsub (-(fmul, x, y)), z) -> (fma (fneg x), y, (fneg z))
+ if (N0.getOpcode() == ISD::FNEG &&
+ N0.getOperand(0).getOpcode() == ISD::FMUL &&
+ N0->hasOneUse() && N0.getOperand(0).hasOneUse()) {
+ SDValue N00 = N0.getOperand(0).getOperand(0);
+ SDValue N01 = N0.getOperand(0).getOperand(1);
+ return DAG.getNode(ISD::FMA, dl, VT,
+ DAG.getNode(ISD::FNEG, dl, VT, N00), N01,
+ DAG.getNode(ISD::FNEG, dl, VT, N1));
+ }
}
return SDValue();
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
EVT VT = N->getValueType(0);
+ DebugLoc dl = N->getDebugLoc();
if (N0CFP && N0CFP->isExactlyValue(1.0))
return DAG.getNode(ISD::FADD, N->getDebugLoc(), VT, N1, N2);
if (N0CFP && !N1CFP)
return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT, N1, N0, N2);
+ // (fma x, c1, (fmul x, c2)) -> (fmul x, c1+c2)
+ if (DAG.getTarget().Options.UnsafeFPMath && N1CFP &&
+ N2.getOpcode() == ISD::FMUL &&
+ N0 == N2.getOperand(0) &&
+ N2.getOperand(1).getOpcode() == ISD::ConstantFP) {
+ return DAG.getNode(ISD::FMUL, dl, VT, N0,
+ DAG.getNode(ISD::FADD, dl, VT, N1, N2.getOperand(1)));
+ }
+
+
+ // (fma (fmul x, c1), c2, y) -> (fma x, c1*c2, y)
+ if (DAG.getTarget().Options.UnsafeFPMath &&
+ N0.getOpcode() == ISD::FMUL && N1CFP &&
+ N0.getOperand(1).getOpcode() == ISD::ConstantFP) {
+ return DAG.getNode(ISD::FMA, dl, VT,
+ N0.getOperand(0),
+ DAG.getNode(ISD::FMUL, dl, VT, N1, N0.getOperand(1)),
+ N2);
+ }
+
+ // (fma x, 1, y) -> (fadd x, y)
+ // (fma x, -1, y) -> (fadd (fneg x), y)
+ if (N1CFP) {
+ if (N1CFP->isExactlyValue(1.0))
+ return DAG.getNode(ISD::FADD, dl, VT, N0, N2);
+
+ if (N1CFP->isExactlyValue(-1.0) &&
+ (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT))) {
+ SDValue RHSNeg = DAG.getNode(ISD::FNEG, dl, VT, N0);
+ AddToWorkList(RHSNeg.getNode());
+ return DAG.getNode(ISD::FADD, dl, VT, N2, RHSNeg);
+ }
+ }
+
+ // (fma x, c, x) -> (fmul x, (c+1))
+ if (DAG.getTarget().Options.UnsafeFPMath && N1CFP && N0 == N2) {
+ return DAG.getNode(ISD::FMUL, dl, VT,
+ N0,
+ DAG.getNode(ISD::FADD, dl, VT,
+ N1, DAG.getConstantFP(1.0, VT)));
+ }
+
+ // (fma x, c, (fneg x)) -> (fmul x, (c-1))
+ if (DAG.getTarget().Options.UnsafeFPMath && N1CFP &&
+ N2.getOpcode() == ISD::FNEG && N2.getOperand(0) == N0) {
+ return DAG.getNode(ISD::FMUL, dl, VT,
+ N0,
+ DAG.getNode(ISD::FADD, dl, VT,
+ N1, DAG.getConstantFP(-1.0, VT)));
+ }
+
+
return SDValue();
}
return DAG.getNode(ISD::UINT_TO_FP, N->getDebugLoc(), VT, N0);
}
- // fold (sint_to_fp (setcc x, y, cc)) -> (select_cc x, y, -1.0, 0.0,, cc)
- if (N0.getOpcode() == ISD::SETCC && N0.getValueType() == MVT::i1 &&
- !VT.isVector() &&
- (!LegalOperations ||
- TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
- SDValue Ops[] =
- { N0.getOperand(0), N0.getOperand(1),
- DAG.getConstantFP(-1.0, VT) , DAG.getConstantFP(0.0, VT),
- N0.getOperand(2) };
- return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(), VT, Ops, 5);
- }
-
- // fold (sint_to_fp (zext (setcc x, y, cc))) ->
- // (select_cc x, y, 1.0, 0.0,, cc)
- if (N0.getOpcode() == ISD::ZERO_EXTEND &&
- N0.getOperand(0).getOpcode() == ISD::SETCC &&!VT.isVector() &&
- (!LegalOperations ||
- TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
- SDValue Ops[] =
- { N0.getOperand(0).getOperand(0), N0.getOperand(0).getOperand(1),
- DAG.getConstantFP(1.0, VT) , DAG.getConstantFP(0.0, VT),
- N0.getOperand(0).getOperand(2) };
- return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(), VT, Ops, 5);
+ // The next optimizations are desireable only if SELECT_CC can be lowered.
+ // Check against MVT::Other for SELECT_CC, which is a workaround for targets
+ // having to say they don't support SELECT_CC on every type the DAG knows
+ // about, since there is no way to mark an opcode illegal at all value types
+ // (See also visitSELECT)
+ if (TLI.isOperationLegalOrCustom(ISD::SELECT_CC, MVT::Other)) {
+ // fold (sint_to_fp (setcc x, y, cc)) -> (select_cc x, y, -1.0, 0.0,, cc)
+ if (N0.getOpcode() == ISD::SETCC && N0.getValueType() == MVT::i1 &&
+ !VT.isVector() &&
+ (!LegalOperations ||
+ TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
+ SDValue Ops[] =
+ { N0.getOperand(0), N0.getOperand(1),
+ DAG.getConstantFP(-1.0, VT) , DAG.getConstantFP(0.0, VT),
+ N0.getOperand(2) };
+ return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(), VT, Ops, 5);
+ }
+
+ // fold (sint_to_fp (zext (setcc x, y, cc))) ->
+ // (select_cc x, y, 1.0, 0.0,, cc)
+ if (N0.getOpcode() == ISD::ZERO_EXTEND &&
+ N0.getOperand(0).getOpcode() == ISD::SETCC &&!VT.isVector() &&
+ (!LegalOperations ||
+ TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
+ SDValue Ops[] =
+ { N0.getOperand(0).getOperand(0), N0.getOperand(0).getOperand(1),
+ DAG.getConstantFP(1.0, VT) , DAG.getConstantFP(0.0, VT),
+ N0.getOperand(0).getOperand(2) };
+ return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(), VT, Ops, 5);
+ }
}
return SDValue();
return DAG.getNode(ISD::SINT_TO_FP, N->getDebugLoc(), VT, N0);
}
- // fold (uint_to_fp (setcc x, y, cc)) -> (select_cc x, y, -1.0, 0.0,, cc)
- if (N0.getOpcode() == ISD::SETCC && !VT.isVector() &&
- (!LegalOperations ||
- TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
- SDValue Ops[] =
- { N0.getOperand(0), N0.getOperand(1),
- DAG.getConstantFP(1.0, VT), DAG.getConstantFP(0.0, VT),
- N0.getOperand(2) };
- return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(), VT, Ops, 5);
- }
+ // The next optimizations are desireable only if SELECT_CC can be lowered.
+ // Check against MVT::Other for SELECT_CC, which is a workaround for targets
+ // having to say they don't support SELECT_CC on every type the DAG knows
+ // about, since there is no way to mark an opcode illegal at all value types
+ // (See also visitSELECT)
+ if (TLI.isOperationLegalOrCustom(ISD::SELECT_CC, MVT::Other)) {
+ // fold (uint_to_fp (setcc x, y, cc)) -> (select_cc x, y, -1.0, 0.0,, cc)
+ if (N0.getOpcode() == ISD::SETCC && !VT.isVector() &&
+ (!LegalOperations ||
+ TLI.isOperationLegalOrCustom(llvm::ISD::ConstantFP, VT))) {
+ SDValue Ops[] =
+ { N0.getOperand(0), N0.getOperand(1),
+ DAG.getConstantFP(1.0, VT), DAG.getConstantFP(0.0, VT),
+ N0.getOperand(2) };
+ return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(), VT, Ops, 5);
+ }
+ }
return SDValue();
}
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
+ if (VT.isVector()) {
+ SDValue FoldedVOp = SimplifyVUnaryOp(N);
+ if (FoldedVOp.getNode()) return FoldedVOp;
+ }
+
if (isNegatibleForFree(N0, LegalOperations, DAG.getTargetLoweringInfo(),
&DAG.getTarget().Options))
return GetNegatedExpression(N0, DAG, LegalOperations);
}
}
+ // (fneg (fmul c, x)) -> (fmul -c, x)
+ if (N0.getOpcode() == ISD::FMUL) {
+ ConstantFPSDNode *CFP1 = dyn_cast<ConstantFPSDNode>(N0.getOperand(1));
+ if (CFP1) {
+ return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+ N0.getOperand(0),
+ DAG.getNode(ISD::FNEG, N->getDebugLoc(), VT,
+ N0.getOperand(1)));
+ }
+ }
+
+ return SDValue();
+}
+
+SDValue DAGCombiner::visitFCEIL(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
+ EVT VT = N->getValueType(0);
+
+ // fold (fceil c1) -> fceil(c1)
+ if (N0CFP && VT != MVT::ppcf128)
+ return DAG.getNode(ISD::FCEIL, N->getDebugLoc(), VT, N0);
+
+ return SDValue();
+}
+
+SDValue DAGCombiner::visitFTRUNC(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
+ EVT VT = N->getValueType(0);
+
+ // fold (ftrunc c1) -> ftrunc(c1)
+ if (N0CFP && VT != MVT::ppcf128)
+ return DAG.getNode(ISD::FTRUNC, N->getDebugLoc(), VT, N0);
+
+ return SDValue();
+}
+
+SDValue DAGCombiner::visitFFLOOR(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
+ EVT VT = N->getValueType(0);
+
+ // fold (ffloor c1) -> ffloor(c1)
+ if (N0CFP && VT != MVT::ppcf128)
+ return DAG.getNode(ISD::FFLOOR, N->getDebugLoc(), VT, N0);
+
return SDValue();
}
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
EVT VT = N->getValueType(0);
+ if (VT.isVector()) {
+ SDValue FoldedVOp = SimplifyVUnaryOp(N);
+ if (FoldedVOp.getNode()) return FoldedVOp;
+ }
+
// fold (fabs c1) -> fabs(c1)
if (N0CFP && VT != MVT::ppcf128)
return DAG.getNode(ISD::FABS, N->getDebugLoc(), VT, N0);
} else
return false;
- TargetLowering::AddrMode AM;
+ AddrMode AM;
if (N->getOpcode() == ISD::ADD) {
ConstantSDNode *Offset = dyn_cast<ConstantSDNode>(N->getOperand(1));
if (Offset)
unsigned NewAlign = MinAlign(LD->getAlignment(), PtrOff);
Type *NewVTTy = NewVT.getTypeForEVT(*DAG.getContext());
- if (NewAlign < TLI.getTargetData()->getABITypeAlignment(NewVTTy))
+ if (NewAlign < TLI.getDataLayout()->getABITypeAlignment(NewVTTy))
return SDValue();
SDValue NewPtr = DAG.getNode(ISD::ADD, LD->getDebugLoc(),
unsigned LDAlign = LD->getAlignment();
unsigned STAlign = ST->getAlignment();
Type *IntVTTy = IntVT.getTypeForEVT(*DAG.getContext());
- unsigned ABIAlign = TLI.getTargetData()->getABITypeAlignment(IntVTTy);
+ unsigned ABIAlign = TLI.getDataLayout()->getABITypeAlignment(IntVTTy);
if (LDAlign < ABIAlign || STAlign < ABIAlign)
return SDValue();
return SDValue();
}
+/// Returns the base pointer and an integer offset from that object.
+static std::pair<SDValue, int64_t> GetPointerBaseAndOffset(SDValue Ptr) {
+ if (Ptr->getOpcode() == ISD::ADD && isa<ConstantSDNode>(Ptr->getOperand(1))) {
+ int64_t Offset = cast<ConstantSDNode>(Ptr->getOperand(1))->getSExtValue();
+ SDValue Base = Ptr->getOperand(0);
+ return std::make_pair(Base, Offset);
+ }
+
+ return std::make_pair(Ptr, 0);
+}
+
+/// Holds a pointer to an LSBaseSDNode as well as information on where it
+/// is located in a sequence of memory operations connected by a chain.
+struct MemOpLink {
+ MemOpLink (LSBaseSDNode *N, int64_t Offset, unsigned Seq):
+ MemNode(N), OffsetFromBase(Offset), SequenceNum(Seq) { }
+ // Ptr to the mem node.
+ LSBaseSDNode *MemNode;
+ // Offset from the base ptr.
+ int64_t OffsetFromBase;
+ // What is the sequence number of this mem node.
+ // Lowest mem operand in the DAG starts at zero.
+ unsigned SequenceNum;
+};
+
+/// Sorts store nodes in a link according to their offset from a shared
+// base ptr.
+struct ConsecutiveMemoryChainSorter {
+ bool operator()(MemOpLink LHS, MemOpLink RHS) {
+ return LHS.OffsetFromBase < RHS.OffsetFromBase;
+ }
+};
+
+bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) {
+ EVT MemVT = St->getMemoryVT();
+ int64_t ElementSizeBytes = MemVT.getSizeInBits()/8;
+
+ // Don't merge vectors into wider inputs.
+ if (MemVT.isVector() || !MemVT.isSimple())
+ return false;
+
+ // Perform an early exit check. Do not bother looking at stored values that
+ // are not constants or loads.
+ SDValue StoredVal = St->getValue();
+ bool IsLoadSrc = isa<LoadSDNode>(StoredVal);
+ if (!isa<ConstantSDNode>(StoredVal) && !isa<ConstantFPSDNode>(StoredVal) &&
+ !IsLoadSrc)
+ return false;
+
+ // Only look at ends of store sequences.
+ SDValue Chain = SDValue(St, 1);
+ if (Chain->hasOneUse() && Chain->use_begin()->getOpcode() == ISD::STORE)
+ return false;
+
+ // This holds the base pointer and the offset in bytes from the base pointer.
+ std::pair<SDValue, int64_t> BasePtr =
+ GetPointerBaseAndOffset(St->getBasePtr());
+
+ // We must have a base and an offset.
+ if (!BasePtr.first.getNode())
+ return false;
+
+ // Do not handle stores to undef base pointers.
+ if (BasePtr.first.getOpcode() == ISD::UNDEF)
+ return false;
+
+ SmallVector<MemOpLink, 8> StoreNodes;
+ // Walk up the chain and look for nodes with offsets from the same
+ // base pointer. Stop when reaching an instruction with a different kind
+ // or instruction which has a different base pointer.
+ unsigned Seq = 0;
+ StoreSDNode *Index = St;
+ while (Index) {
+ // If the chain has more than one use, then we can't reorder the mem ops.
+ if (Index != St && !SDValue(Index, 1)->hasOneUse())
+ break;
+
+ // Find the base pointer and offset for this memory node.
+ std::pair<SDValue, int64_t> Ptr =
+ GetPointerBaseAndOffset(Index->getBasePtr());
+
+ // Check that the base pointer is the same as the original one.
+ if (Ptr.first.getNode() != BasePtr.first.getNode())
+ break;
+
+ // Check that the alignment is the same.
+ if (Index->getAlignment() != St->getAlignment())
+ break;
+
+ // The memory operands must not be volatile.
+ if (Index->isVolatile() || Index->isIndexed())
+ break;
+
+ // No truncation.
+ if (StoreSDNode *St = dyn_cast<StoreSDNode>(Index))
+ if (St->isTruncatingStore())
+ break;
+
+ // The stored memory type must be the same.
+ if (Index->getMemoryVT() != MemVT)
+ break;
+
+ // We do not allow unaligned stores because we want to prevent overriding
+ // stores.
+ if (Index->getAlignment()*8 != MemVT.getSizeInBits())
+ break;
+
+ // We found a potential memory operand to merge.
+ StoreNodes.push_back(MemOpLink(Index, Ptr.second, Seq++));
+
+ // Move up the chain to the next memory operation.
+ Index = dyn_cast<StoreSDNode>(Index->getChain().getNode());
+ }
+
+ // Check if there is anything to merge.
+ if (StoreNodes.size() < 2)
+ return false;
+
+ // Sort the memory operands according to their distance from the base pointer.
+ std::sort(StoreNodes.begin(), StoreNodes.end(),
+ ConsecutiveMemoryChainSorter());
+
+ // Scan the memory operations on the chain and find the first non-consecutive
+ // store memory address.
+ unsigned LastConsecutiveStore = 0;
+ int64_t StartAddress = StoreNodes[0].OffsetFromBase;
+ for (unsigned i=1; i<StoreNodes.size(); ++i) {
+ int64_t CurrAddress = StoreNodes[i].OffsetFromBase;
+ if (CurrAddress - StartAddress != (ElementSizeBytes * i))
+ break;
+
+ // Mark this node as useful.
+ LastConsecutiveStore = i;
+ }
+
+ // The node with the lowest store address.
+ LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode;
+
+ // Store the constants into memory as one consecutive store.
+ if (!IsLoadSrc) {
+ unsigned LastLegalType = 0;
+ unsigned LastLegalVectorType = 0;
+ bool NonZero = false;
+ for (unsigned i=0; i<LastConsecutiveStore+1; ++i) {
+ StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
+ SDValue StoredVal = St->getValue();
+
+ if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(StoredVal)) {
+ NonZero |= !C->isNullValue();
+ } else if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(StoredVal)) {
+ NonZero |= !C->getConstantFPValue()->isNullValue();
+ } else {
+ // Non constant.
+ break;
+ }
+
+ // Find a legal type for the constant store.
+ unsigned StoreBW = (i+1) * ElementSizeBytes * 8;
+ EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ if (TLI.isTypeLegal(StoreTy))
+ LastLegalType = i+1;
+
+ // Find a legal type for the vector store.
+ EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1);
+ if (TLI.isTypeLegal(Ty))
+ LastLegalVectorType = i + 1;
+ }
+
+ // We only use vectors if the constant is known to be zero.
+ if (NonZero)
+ LastLegalVectorType = 0;
+
+ // Check if we found a legal integer type to store.
+ if (LastLegalType == 0 && LastLegalVectorType == 0)
+ return false;
+
+ bool UseVector = LastLegalVectorType > LastLegalType;
+ unsigned NumElem = UseVector ? LastLegalVectorType : LastLegalType;
+
+ // Make sure we have something to merge.
+ if (NumElem < 2)
+ return false;
+
+ unsigned EarliestNodeUsed = 0;
+ for (unsigned i=0; i < NumElem; ++i) {
+ // Find a chain for the new wide-store operand. Notice that some
+ // of the store nodes that we found may not be selected for inclusion
+ // in the wide store. The chain we use needs to be the chain of the
+ // earliest store node which is *used* and replaced by the wide store.
+ if (StoreNodes[i].SequenceNum > StoreNodes[EarliestNodeUsed].SequenceNum)
+ EarliestNodeUsed = i;
+ }
+
+ // The earliest Node in the DAG.
+ LSBaseSDNode *EarliestOp = StoreNodes[EarliestNodeUsed].MemNode;
+ DebugLoc DL = StoreNodes[0].MemNode->getDebugLoc();
+
+ SDValue StoredVal;
+ if (UseVector) {
+ // Find a legal type for the vector store.
+ EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem);
+ assert(TLI.isTypeLegal(Ty) && "Illegal vector store");
+ StoredVal = DAG.getConstant(0, Ty);
+ } else {
+ unsigned StoreBW = NumElem * ElementSizeBytes * 8;
+ APInt StoreInt(StoreBW, 0);
+
+ // Construct a single integer constant which is made of the smaller
+ // constant inputs.
+ bool IsLE = TLI.isLittleEndian();
+ for (unsigned i = 0; i < NumElem ; ++i) {
+ unsigned Idx = IsLE ?(NumElem - 1 - i) : i;
+ StoreSDNode *St = cast<StoreSDNode>(StoreNodes[Idx].MemNode);
+ SDValue Val = St->getValue();
+ StoreInt<<=ElementSizeBytes*8;
+ if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Val)) {
+ StoreInt|=C->getAPIntValue().zext(StoreBW);
+ } else if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Val)) {
+ StoreInt|= C->getValueAPF().bitcastToAPInt().zext(StoreBW);
+ } else {
+ assert(false && "Invalid constant element type");
+ }
+ }
+
+ // Create the new Load and Store operations.
+ EVT StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ StoredVal = DAG.getConstant(StoreInt, StoreTy);
+ }
+
+ SDValue NewStore = DAG.getStore(EarliestOp->getChain(), DL, StoredVal,
+ FirstInChain->getBasePtr(),
+ FirstInChain->getPointerInfo(),
+ false, false,
+ FirstInChain->getAlignment());
+
+ // Replace the first store with the new store
+ CombineTo(EarliestOp, NewStore);
+ // Erase all other stores.
+ for (unsigned i = 0; i < NumElem ; ++i) {
+ if (StoreNodes[i].MemNode == EarliestOp)
+ continue;
+ StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
+ DAG.ReplaceAllUsesOfValueWith(SDValue(St, 0), St->getChain());
+ removeFromWorkList(St);
+ DAG.DeleteNode(St);
+ }
+
+ return true;
+ }
+
+ // Below we handle the case of multiple consecutive stores that
+ // come from multiple consecutive loads. We merge them into a single
+ // wide load and a single wide store.
+
+ // Look for load nodes which are used by the stored values.
+ SmallVector<MemOpLink, 8> LoadNodes;
+
+ // Find acceptable loads. Loads need to have the same chain (token factor),
+ // must not be zext, volatile, indexed, and they must be consecutive.
+ SDValue LdBasePtr;
+ for (unsigned i=0; i<LastConsecutiveStore+1; ++i) {
+ StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
+ LoadSDNode *Ld = dyn_cast<LoadSDNode>(St->getValue());
+ if (!Ld) break;
+
+ // Loads must only have one use.
+ if (!Ld->hasNUsesOfValue(1, 0))
+ break;
+
+ // Check that the alignment is the same as the stores.
+ if (Ld->getAlignment() != St->getAlignment())
+ break;
+
+ // The memory operands must not be volatile.
+ if (Ld->isVolatile() || Ld->isIndexed())
+ break;
+
+ // We do not accept ext loads.
+ if (Ld->getExtensionType() != ISD::NON_EXTLOAD)
+ break;
+
+ // The stored memory type must be the same.
+ if (Ld->getMemoryVT() != MemVT)
+ break;
+
+ std::pair<SDValue, int64_t> LdPtr =
+ GetPointerBaseAndOffset(Ld->getBasePtr());
+
+ // If this is not the first ptr that we check.
+ if (LdBasePtr.getNode()) {
+ // The base ptr must be the same.
+ if (LdPtr.first != LdBasePtr)
+ break;
+ } else {
+ // Check that all other base pointers are the same as this one.
+ LdBasePtr = LdPtr.first;
+ }
+
+ // We found a potential memory operand to merge.
+ LoadNodes.push_back(MemOpLink(Ld, LdPtr.second, 0));
+ }
+
+ if (LoadNodes.size() < 2)
+ return false;
+
+ // Scan the memory operations on the chain and find the first non-consecutive
+ // load memory address. These variables hold the index in the store node
+ // array.
+ unsigned LastConsecutiveLoad = 0;
+ // This variable refers to the size and not index in the array.
+ unsigned LastLegalVectorType = 0;
+ unsigned LastLegalIntegerType = 0;
+ StartAddress = LoadNodes[0].OffsetFromBase;
+ SDValue FirstChain = LoadNodes[0].MemNode->getChain();
+ for (unsigned i = 1; i < LoadNodes.size(); ++i) {
+ // All loads much share the same chain.
+ if (LoadNodes[i].MemNode->getChain() != FirstChain)
+ break;
+
+ int64_t CurrAddress = LoadNodes[i].OffsetFromBase;
+ if (CurrAddress - StartAddress != (ElementSizeBytes * i))
+ break;
+ LastConsecutiveLoad = i;
+
+ // Find a legal type for the vector store.
+ EVT StoreTy = EVT::getVectorVT(*DAG.getContext(), MemVT, i+1);
+ if (TLI.isTypeLegal(StoreTy))
+ LastLegalVectorType = i + 1;
+
+ // Find a legal type for the integer store.
+ unsigned StoreBW = (i+1) * ElementSizeBytes * 8;
+ StoreTy = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ if (TLI.isTypeLegal(StoreTy))
+ LastLegalIntegerType = i + 1;
+ }
+
+ // Only use vector types if the vector type is larger than the integer type.
+ // If they are the same, use integers.
+ bool UseVectorTy = LastLegalVectorType > LastLegalIntegerType;
+ unsigned LastLegalType = std::max(LastLegalVectorType, LastLegalIntegerType);
+
+ // We add +1 here because the LastXXX variables refer to location while
+ // the NumElem refers to array/index size.
+ unsigned NumElem = std::min(LastConsecutiveStore, LastConsecutiveLoad) + 1;
+ NumElem = std::min(LastLegalType, NumElem);
+
+ if (NumElem < 2)
+ return false;
+
+ // The earliest Node in the DAG.
+ unsigned EarliestNodeUsed = 0;
+ LSBaseSDNode *EarliestOp = StoreNodes[EarliestNodeUsed].MemNode;
+ for (unsigned i=1; i<NumElem; ++i) {
+ // Find a chain for the new wide-store operand. Notice that some
+ // of the store nodes that we found may not be selected for inclusion
+ // in the wide store. The chain we use needs to be the chain of the
+ // earliest store node which is *used* and replaced by the wide store.
+ if (StoreNodes[i].SequenceNum > StoreNodes[EarliestNodeUsed].SequenceNum)
+ EarliestNodeUsed = i;
+ }
+
+ // Find if it is better to use vectors or integers to load and store
+ // to memory.
+ EVT JointMemOpVT;
+ if (UseVectorTy) {
+ JointMemOpVT = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem);
+ } else {
+ unsigned StoreBW = NumElem * ElementSizeBytes * 8;
+ JointMemOpVT = EVT::getIntegerVT(*DAG.getContext(), StoreBW);
+ }
+
+ DebugLoc LoadDL = LoadNodes[0].MemNode->getDebugLoc();
+ DebugLoc StoreDL = StoreNodes[0].MemNode->getDebugLoc();
+
+ LoadSDNode *FirstLoad = cast<LoadSDNode>(LoadNodes[0].MemNode);
+ SDValue NewLoad = DAG.getLoad(JointMemOpVT, LoadDL,
+ FirstLoad->getChain(),
+ FirstLoad->getBasePtr(),
+ FirstLoad->getPointerInfo(),
+ false, false, false,
+ FirstLoad->getAlignment());
+
+ SDValue NewStore = DAG.getStore(EarliestOp->getChain(), StoreDL, NewLoad,
+ FirstInChain->getBasePtr(),
+ FirstInChain->getPointerInfo(), false, false,
+ FirstInChain->getAlignment());
+
+ // Replace one of the loads with the new load.
+ LoadSDNode *Ld = cast<LoadSDNode>(LoadNodes[0].MemNode);
+ DAG.ReplaceAllUsesOfValueWith(SDValue(Ld, 1),
+ SDValue(NewLoad.getNode(), 1));
+
+ // Remove the rest of the load chains.
+ for (unsigned i = 1; i < NumElem ; ++i) {
+ // Replace all chain users of the old load nodes with the chain of the new
+ // load node.
+ LoadSDNode *Ld = cast<LoadSDNode>(LoadNodes[i].MemNode);
+ DAG.ReplaceAllUsesOfValueWith(SDValue(Ld, 1), Ld->getChain());
+ }
+
+ // Replace the first store with the new store.
+ CombineTo(EarliestOp, NewStore);
+ // Erase all other stores.
+ for (unsigned i = 0; i < NumElem ; ++i) {
+ // Remove all Store nodes.
+ if (StoreNodes[i].MemNode == EarliestOp)
+ continue;
+ StoreSDNode *St = cast<StoreSDNode>(StoreNodes[i].MemNode);
+ DAG.ReplaceAllUsesOfValueWith(SDValue(St, 0), St->getChain());
+ removeFromWorkList(St);
+ DAG.DeleteNode(St);
+ }
+
+ return true;
+}
+
SDValue DAGCombiner::visitSTORE(SDNode *N) {
StoreSDNode *ST = cast<StoreSDNode>(N);
SDValue Chain = ST->getChain();
ST->isUnindexed()) {
unsigned OrigAlign = ST->getAlignment();
EVT SVT = Value.getOperand(0).getValueType();
- unsigned Align = TLI.getTargetData()->
+ unsigned Align = TLI.getDataLayout()->
getABITypeAlignment(SVT.getTypeForEVT(*DAG.getContext()));
if (Align <= OrigAlign &&
((!LegalOperations && !ST->isVolatile()) ||
ST->getAlignment());
}
+ // Only perform this optimization before the types are legal, because we
+ // don't want to perform this optimization on every DAGCombine invocation.
+ if (!LegalTypes && MergeConsecutiveStores(ST))
+ return SDValue(N, 0);
+
return ReduceLoadOpStoreWidth(N);
}
// Transform: (EXTRACT_VECTOR_ELT( VECTOR_SHUFFLE )) -> EXTRACT_VECTOR_ELT.
// We only perform this optimization before the op legalization phase because
- // we may introduce new vector instructions which are not backed by TD patterns.
- // For example on AVX, extracting elements from a wide vector without using
- // extract_subvector.
+ // we may introduce new vector instructions which are not backed by TD
+ // patterns. For example on AVX, extracting elements from a wide vector
+ // without using extract_subvector.
if (InVec.getOpcode() == ISD::VECTOR_SHUFFLE
&& ConstEltNo && !LegalOperations) {
int Elt = cast<ConstantSDNode>(EltNo)->getZExtValue();
// Check the resultant load doesn't need a higher alignment than the
// original load.
unsigned NewAlign =
- TLI.getTargetData()
+ TLI.getDataLayout()
->getABITypeAlignment(LVT.getTypeForEVT(*DAG.getContext()));
if (NewAlign > Align || !TLI.isOperationLegalOrCustom(ISD::LOAD, LVT))
if (VecIn1.getValueType().getSizeInBits()*2 != VT.getSizeInBits())
return SDValue();
+ // If the input vector type has a different base type to the output
+ // vector type, bail out.
+ if (VecIn1.getValueType().getVectorElementType() !=
+ VT.getVectorElementType())
+ return SDValue();
+
// Widen the input vector by adding undef values.
VecIn1 = DAG.getNode(ISD::CONCAT_VECTORS, N->getDebugLoc(), VT,
VecIn1, DAG.getUNDEF(VecIn1.getValueType()));
// If VecIn2 is unused then change it to undef.
VecIn2 = VecIn2.getNode() ? VecIn2 : DAG.getUNDEF(VT);
- // Check that we were able to transform all incoming values to the same type.
+ // Check that we were able to transform all incoming values to the same
+ // type.
if (VecIn2.getValueType() != VecIn1.getValueType() ||
VecIn1.getValueType() != VT)
return SDValue();
return SDValue();
}
+/// SimplifyVUnaryOp - Visit a binary vector operation, like FABS/FNEG.
+SDValue DAGCombiner::SimplifyVUnaryOp(SDNode *N) {
+ // After legalize, the target may be depending on adds and other
+ // binary ops to provide legal ways to construct constants or other
+ // things. Simplifying them may result in a loss of legality.
+ if (LegalOperations) return SDValue();
+
+ assert(N->getValueType(0).isVector() &&
+ "SimplifyVUnaryOp only works on vectors!");
+
+ SDValue N0 = N->getOperand(0);
+
+ if (N0.getOpcode() != ISD::BUILD_VECTOR)
+ return SDValue();
+
+ // Operand is a BUILD_VECTOR node, see if we can constant fold it.
+ SmallVector<SDValue, 8> Ops;
+ for (unsigned i = 0, e = N0.getNumOperands(); i != e; ++i) {
+ SDValue Op = N0.getOperand(i);
+ if (Op.getOpcode() != ISD::UNDEF &&
+ Op.getOpcode() != ISD::ConstantFP)
+ break;
+ EVT EltVT = Op.getValueType();
+ SDValue FoldOp = DAG.getNode(N->getOpcode(), N0.getDebugLoc(), EltVT, Op);
+ if (FoldOp.getOpcode() != ISD::UNDEF &&
+ FoldOp.getOpcode() != ISD::ConstantFP)
+ break;
+ Ops.push_back(FoldOp);
+ AddToWorkList(FoldOp.getNode());
+ }
+
+ if (Ops.size() != N0.getNumOperands())
+ return SDValue();
+
+ return DAG.getNode(ISD::BUILD_VECTOR, N->getDebugLoc(),
+ N0.getValueType(), &Ops[0], Ops.size());
+}
+
SDValue DAGCombiner::SimplifySelect(DebugLoc DL, SDValue N0,
SDValue N1, SDValue N2){
assert(N0.getOpcode() ==ISD::SETCC && "First argument must be a SetCC node!");
const_cast<ConstantFP*>(TV->getConstantFPValue())
};
Type *FPTy = Elts[0]->getType();
- const TargetData &TD = *TLI.getTargetData();
+ const DataLayout &TD = *TLI.getDataLayout();
// Create a ConstantArray of the two constants.
Constant *CA = ConstantArray::get(ArrayType::get(FPTy, 2), Elts);
// to alias with anything but itself. Provides base object and offset as
// results.
static bool FindBaseOffset(SDValue Ptr, SDValue &Base, int64_t &Offset,
- const GlobalValue *&GV, void *&CV) {
+ const GlobalValue *&GV, const void *&CV) {
// Assume it is a primitive operation.
Base = Ptr; Offset = 0; GV = 0; CV = 0;
// for ConstantSDNodes since the same constant pool entry may be represented
// by multiple nodes with different offsets.
if (ConstantPoolSDNode *C = dyn_cast<ConstantPoolSDNode>(Base)) {
- CV = C->isMachineConstantPoolEntry() ? (void *)C->getMachineCPVal()
- : (void *)C->getConstVal();
+ CV = C->isMachineConstantPoolEntry() ? (const void *)C->getMachineCPVal()
+ : (const void *)C->getConstVal();
Offset += C->getOffset();
return false;
}
SDValue Base1, Base2;
int64_t Offset1, Offset2;
const GlobalValue *GV1, *GV2;
- void *CV1, *CV2;
+ const void *CV1, *CV2;
bool isFrameIndex1 = FindBaseOffset(Ptr1, Base1, Offset1, GV1, CV1);
bool isFrameIndex2 = FindBaseOffset(Ptr2, Base2, Offset2, GV2, CV2);
projects / oota-llvm.git / blobdiff