/// getShiftAmountTy - Returns a type large enough to hold any valid
/// shift amount - before type legalization these can be huge.
EVT getShiftAmountTy(EVT LHSTy) {
- return LegalTypes ? TLI.getShiftAmountTy(LHSTy) : TLI.getPointerTy();
+ assert(LHSTy.isInteger() && "Shift amount is not an integer type!");
+ if (LHSTy.isVector())
+ return LHSTy;
+ return LegalTypes ? TLI.getScalarShiftAmountTy(LHSTy) : TLI.getPointerTy();
}
/// isTypeLegal - This method returns true if we are running before type
return SDValue();
}
+/// isConstantSplatVector - Returns true if N is a BUILD_VECTOR node whose elements are
+/// all the same constant or undefined.
+static bool isConstantSplatVector(SDNode *N, APInt& SplatValue) {
+ BuildVectorSDNode *C = dyn_cast<BuildVectorSDNode>(N);
+ if (!C)
+ return false;
+
+ APInt SplatUndef;
+ unsigned SplatBitSize;
+ bool HasAnyUndefs;
+ EVT EltVT = N->getValueType(0).getVectorElementType();
+ return (C->isConstantSplat(SplatValue, SplatUndef, SplatBitSize,
+ HasAnyUndefs) &&
+ EltVT.getSizeInBits() >= SplatBitSize);
+}
+
SDValue DAGCombiner::visitMUL(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
- ConstantSDNode *N0C = dyn_cast<ConstantSDNode>(N0);
- ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
EVT VT = N0.getValueType();
+ // fold (mul x, undef) -> 0
+ if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
+ return DAG.getConstant(0, VT);
+
+ bool N0IsConst = false;
+ bool N1IsConst = false;
+ APInt ConstValue0, ConstValue1;
// fold vector ops
if (VT.isVector()) {
SDValue FoldedVOp = SimplifyVBinOp(N);
if (FoldedVOp.getNode()) return FoldedVOp;
+
+ N0IsConst = isConstantSplatVector(N0.getNode(), ConstValue0);
+ N1IsConst = isConstantSplatVector(N1.getNode(), ConstValue1);
+ } else {
+ N0IsConst = dyn_cast<ConstantSDNode>(N0) != 0;
+ ConstValue0 = N0IsConst? (dyn_cast<ConstantSDNode>(N0))->getAPIntValue() : APInt();
+ N1IsConst = dyn_cast<ConstantSDNode>(N1) != 0;
+ ConstValue1 = N1IsConst? (dyn_cast<ConstantSDNode>(N1))->getAPIntValue() : APInt();
}
- // fold (mul x, undef) -> 0
- if (N0.getOpcode() == ISD::UNDEF || N1.getOpcode() == ISD::UNDEF)
- return DAG.getConstant(0, VT);
// fold (mul c1, c2) -> c1*c2
- if (N0C && N1C)
- return DAG.FoldConstantArithmetic(ISD::MUL, VT, N0C, N1C);
+ if (N0IsConst && N1IsConst)
+ return DAG.FoldConstantArithmetic(ISD::MUL, VT, N0.getNode(), N1.getNode());
+
// canonicalize constant to RHS
- if (N0C && !N1C)
+ if (N0IsConst && !N1IsConst)
return DAG.getNode(ISD::MUL, SDLoc(N), VT, N1, N0);
// fold (mul x, 0) -> 0
- if (N1C && N1C->isNullValue())
+ if (N1IsConst && ConstValue1 == 0)
return N1;
+ // fold (mul x, 1) -> x
+ if (N1IsConst && ConstValue1 == 1)
+ return N0;
// fold (mul x, -1) -> 0-x
- if (N1C && N1C->isAllOnesValue())
+ if (N1IsConst && ConstValue1.isAllOnesValue())
return DAG.getNode(ISD::SUB, SDLoc(N), VT,
DAG.getConstant(0, VT), N0);
// fold (mul x, (1 << c)) -> x << c
- if (N1C && N1C->getAPIntValue().isPowerOf2())
+ if (N1IsConst && ConstValue1.isPowerOf2())
return DAG.getNode(ISD::SHL, SDLoc(N), VT, N0,
- DAG.getConstant(N1C->getAPIntValue().logBase2(),
+ DAG.getConstant(ConstValue1.logBase2(),
getShiftAmountTy(N0.getValueType())));
// fold (mul x, -(1 << c)) -> -(x << c) or (-x) << c
- if (N1C && (-N1C->getAPIntValue()).isPowerOf2()) {
- unsigned Log2Val = (-N1C->getAPIntValue()).logBase2();
+ if (N1IsConst && (-ConstValue1).isPowerOf2()) {
+ unsigned Log2Val = (-ConstValue1).logBase2();
// FIXME: If the input is something that is easily negated (e.g. a
// single-use add), we should put the negate there.
return DAG.getNode(ISD::SUB, SDLoc(N), VT,
DAG.getConstant(Log2Val,
getShiftAmountTy(N0.getValueType()))));
}
+
+ APInt Val;
// (mul (shl X, c1), c2) -> (mul X, c2 << c1)
- if (N1C && N0.getOpcode() == ISD::SHL &&
- isa<ConstantSDNode>(N0.getOperand(1))) {
+ if (N1IsConst && N0.getOpcode() == ISD::SHL &&
+ (isConstantSplatVector(N0.getOperand(1).getNode(), Val) ||
+ isa<ConstantSDNode>(N0.getOperand(1)))) {
SDValue C3 = DAG.getNode(ISD::SHL, SDLoc(N), VT,
N1, N0.getOperand(1));
AddToWorkList(C3.getNode());
{
SDValue Sh(0,0), Y(0,0);
// Check for both (mul (shl X, C), Y) and (mul Y, (shl X, C)).
- if (N0.getOpcode() == ISD::SHL && isa<ConstantSDNode>(N0.getOperand(1)) &&
+ if (N0.getOpcode() == ISD::SHL &&
+ (isConstantSplatVector(N0.getOperand(1).getNode(), Val) ||
+ isa<ConstantSDNode>(N0.getOperand(1))) &&
N0.getNode()->hasOneUse()) {
Sh = N0; Y = N1;
} else if (N1.getOpcode() == ISD::SHL &&
}
// fold (mul (add x, c1), c2) -> (add (mul x, c2), c1*c2)
- if (N1C && N0.getOpcode() == ISD::ADD && N0.getNode()->hasOneUse() &&
- isa<ConstantSDNode>(N0.getOperand(1)))
+ if (N1IsConst && N0.getOpcode() == ISD::ADD && N0.getNode()->hasOneUse() &&
+ (isConstantSplatVector(N0.getOperand(1).getNode(), Val) ||
+ isa<ConstantSDNode>(N0.getOperand(1))))
return DAG.getNode(ISD::ADD, SDLoc(N), VT,
DAG.getNode(ISD::MUL, SDLoc(N0), VT,
N0.getOperand(0), N1),
return SDValue(N, 0); // Return N so it doesn't get rechecked!
}
}
- // similarly fold (and (X (load ([non_ext|any_ext|zero_ext] V))), c) ->
+ // similarly fold (and (X (load ([non_ext|any_ext|zero_ext] V))), c) ->
// (X (load ([non_ext|zero_ext] V))) if 'and' only clears top bits which must
// already be zero by virtue of the width of the base type of the load.
//
? cast<LoadSDNode>(N0.getOperand(0))
: cast<LoadSDNode>(N0);
if (LN0->getExtensionType() != ISD::SEXTLOAD &&
- LN0->isUnindexed() && N0.hasOneUse() && LN0->hasOneUse()) {
+ LN0->isUnindexed() && N0.hasOneUse() && SDValue(LN0, 0).hasOneUse()) {
uint32_t ActiveBits = N1C->getAPIntValue().getActiveBits();
if (ActiveBits > 0 && APIntOps::isMask(ActiveBits, N1C->getAPIntValue())){
EVT ExtVT = EVT::getIntegerVT(*DAG.getContext(), ActiveBits);
DAG.getConstant(~0ULL >> ShAmt, VT));
}
-
- // fold (srl (anyextend x), c) -> (anyextend (srl x, c))
+ // fold (srl (anyextend x), c) -> (and (anyextend (srl x, c)), mask)
if (N1C && N0.getOpcode() == ISD::ANY_EXTEND) {
// Shifting in all undef bits?
EVT SmallVT = N0.getOperand(0).getValueType();
N0.getOperand(0),
DAG.getConstant(ShiftAmt, getShiftAmountTy(SmallVT)));
AddToWorkList(SmallShift.getNode());
- return DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), VT, SmallShift);
+ APInt Mask = APInt::getAllOnesValue(VT.getSizeInBits()).lshr(ShiftAmt);
+ return DAG.getNode(ISD::AND, SDLoc(N), VT,
+ DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), VT, SmallShift),
+ DAG.getConstant(Mask, VT));
}
}
// sext(setcc) -> sext_in_reg(vsetcc) for vectors.
// Only do this before legalize for now.
if (VT.isVector() && !LegalOperations &&
- TLI.getBooleanContents(true) ==
+ TLI.getBooleanContents(true) ==
TargetLowering::ZeroOrNegativeOneBooleanContent) {
EVT N0VT = N0.getOperand(0).getValueType();
// On some architectures (such as SSE/NEON/etc) the SETCC result type is
// For the transform to be legal, the load must produce only two values
// (the value loaded and the chain). Don't transform a pre-increment
- // load, for example, which produces an extra value. Otherwise the
+ // load, for example, which produces an extra value. Otherwise the
// transformation is not equivalent, and the downstream logic to replace
// uses gets things wrong.
if (LN0->getNumValues() > 2)
return SDValue();
+ // If the load that we're shrinking is an extload and we're not just
+ // discarding the extension we can't simply shrink the load. Bail.
+ // TODO: It would be possible to merge the extensions in some cases.
+ if (LN0->getExtensionType() != ISD::NON_EXTLOAD &&
+ LN0->getMemoryVT().getSizeInBits() < ExtVT.getSizeInBits() + ShAmt)
+ return SDValue();
+
EVT PtrType = N0.getOperand(1).getValueType();
if (PtrType == MVT::Untyped || PtrType.isExtended())
// We don't need test this condition for transformation like following, as
// the DAG being transformed implies it is legal to take FP constant as
// operand.
- //
+ //
// (fadd (fmul c, x), x) -> (fmul c+1, x)
- //
+ //
bool AllowNewFpConst = (Level < AfterLegalizeDAG);
// If allow, fold (fadd (fneg x), x) -> 0.0
}
// fold (fsub (-(fmul, x, y)), z) -> (fma (fneg x), y, (fneg z))
- if (N0.getOpcode() == ISD::FNEG &&
+ if (N0.getOpcode() == ISD::FNEG &&
N0.getOperand(0).getOpcode() == ISD::FMUL &&
N0->hasOneUse() && N0.getOperand(0).hasOneUse()) {
SDValue N00 = N0.getOperand(0).getOperand(0);
// fold (fmul (fneg X), (fneg Y)) -> (fmul X, Y)
if (char LHSNeg = isNegatibleForFree(N0, LegalOperations, TLI,
&DAG.getTarget().Options)) {
- if (char RHSNeg = isNegatibleForFree(N1, LegalOperations, TLI,
+ if (char RHSNeg = isNegatibleForFree(N1, LegalOperations, TLI,
&DAG.getTarget().Options)) {
// Both can be negated for free, check to see if at least one is cheaper
// negated.
// Transform fabs(bitconvert(x)) -> bitconvert(x&~sign) to avoid loading
// constant pool values.
- if (!TLI.isFAbsFree(VT) &&
+ if (!TLI.isFAbsFree(VT) &&
N0.getOpcode() == ISD::BITCAST && N0.getNode()->hasOneUse() &&
N0.getOperand(0).getValueType().isInteger() &&
!N0.getOperand(0).getValueType().isVector()) {
// x0 * offset0 + y0 * ptr0 = t0
// knowing that
// x1 * offset1 + y1 * ptr0 = t1 (the indexed load/store)
- //
+ //
// where x0, x1, y0 and y1 in {-1, 1} are given by the types of the
// indexed load/store and the expresion that needs to be re-written.
//
for (SDNode::use_iterator III = Use->use_begin(),
EEE = Use->use_end(); III != EEE; ++III) {
SDNode *UseUse = *III;
- if (!canFoldInAddressingMode(Use, UseUse, DAG, TLI))
+ if (!canFoldInAddressingMode(Use, UseUse, DAG, TLI))
RealUse = true;
}
} else {
Load = DAG.getLoad(LVT, SDLoc(N), LN0->getChain(), NewPtr,
LN0->getPointerInfo().getWithOffset(PtrOff),
- LN0->isVolatile(), LN0->isNonTemporal(),
+ LN0->isVolatile(), LN0->isNonTemporal(),
LN0->isInvariant(), Align);
Chain = Load.getValue(1);
if (NVT.bitsLT(LVT))
// The extract index must be constant.
if (!CS)
return SDValue();
-
+
// Check that we are reading from the identity index.
if (CS->getZExtValue() != IdentityIndex)
return SDValue();
if (SingleSource.getNode())
return SingleSource;
-
+
return SDValue();
}