/// 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();
}
+/// isSplatVector - Returns true if N is a BUILD_VECTOR node whose elements are
+/// all the same 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),
ret <4 x i64> %x
}
+; CHECK: mul_const1
+; CHECK: vpaddd
+; CHECK: ret
+define <8 x i32> @mul_const1(<8 x i32> %x) {
+ %y = mul <8 x i32> %x, <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
+ ret <8 x i32> %y
+}
+
+; CHECK: mul_const2
+; CHECK: vpsllq $2
+; CHECK: ret
+define <4 x i64> @mul_const2(<4 x i64> %x) {
+ %y = mul <4 x i64> %x, <i64 4, i64 4, i64 4, i64 4>
+ ret <4 x i64> %y
+}
+
+; CHECK: mul_const3
+; CHECK: vpsllw $3
+; CHECK: ret
+define <16 x i16> @mul_const3(<16 x i16> %x) {
+ %y = mul <16 x i16> %x, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8>
+ ret <16 x i16> %y
+}
+
+; CHECK: mul_const4
+; CHECK: vpxor
+; CHECK: vpsubq
+; CHECK: ret
+define <4 x i64> @mul_const4(<4 x i64> %x) {
+ %y = mul <4 x i64> %x, <i64 -1, i64 -1, i64 -1, i64 -1>
+ ret <4 x i64> %y
+}
+
+; CHECK: mul_const5
+; CHECK: vxorps
+; CHECK-NEXT: ret
+define <8 x i32> @mul_const5(<8 x i32> %x) {
+ %y = mul <8 x i32> %x, <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
+ ret <8 x i32> %y
+}
+
+; CHECK: mul_const6
+; CHECK: vpmulld
+; CHECK: ret
+define <8 x i32> @mul_const6(<8 x i32> %x) {
+ %y = mul <8 x i32> %x, <i32 0, i32 0, i32 0, i32 2, i32 0, i32 2, i32 0, i32 0>
+ ret <8 x i32> %y
+}
+
+; CHECK: mul_const7
+; CHECK: vpaddq
+; CHECK: vpaddq
+; CHECK: ret
+define <8 x i64> @mul_const7(<8 x i64> %x) {
+ %y = mul <8 x i64> %x, <i64 2, i64 2, i64 2, i64 2, i64 2, i64 2, i64 2, i64 2>
+ ret <8 x i64> %y
+}
+
+; CHECK: mul_const8
+; CHECK: vpsllw $3
+; CHECK: ret
+define <8 x i16> @mul_const8(<8 x i16> %x) {
+ %y = mul <8 x i16> %x, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8>
+ ret <8 x i16> %y
+}
+
+; CHECK: mul_const9
+; CHECK: vpmulld
+; CHECK: ret
+define <8 x i32> @mul_const9(<8 x i32> %x) {
+ %y = mul <8 x i32> %x, <i32 2, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
+ ret <8 x i32> %y
+}
\ No newline at end of file