setOperationAction(ISD::ADD, MVT::v8i16, Legal);
setOperationAction(ISD::ADD, MVT::v4i32, Legal);
setOperationAction(ISD::ADD, MVT::v2i64, Legal);
+ setOperationAction(ISD::MUL, MVT::v4i32, Custom);
setOperationAction(ISD::MUL, MVT::v2i64, Custom);
setOperationAction(ISD::SUB, MVT::v16i8, Legal);
setOperationAction(ISD::SUB, MVT::v8i16, Legal);
// FIXME: Why this routine is here? Move to RegInfo!
std::pair<const TargetRegisterClass*, uint8_t>
-X86TargetLowering::findRepresentativeClass(EVT VT) const{
+X86TargetLowering::findRepresentativeClass(MVT VT) const{
const TargetRegisterClass *RRC = 0;
uint8_t Cost = 1;
- switch (VT.getSimpleVT().SimpleTy) {
+ switch (VT.SimpleTy) {
default:
return TargetLowering::findRepresentativeClass(VT);
case MVT::i8: case MVT::i16: case MVT::i32: case MVT::i64:
return true;
}
-EVT
-X86TargetLowering::getTypeForExtArgOrReturn(LLVMContext &Context, EVT VT,
+MVT
+X86TargetLowering::getTypeForExtArgOrReturn(MVT VT,
ISD::NodeType ExtendKind) const {
MVT ReturnMVT;
// TODO: Is this also valid on 32-bit?
else
ReturnMVT = MVT::i32;
- EVT MinVT = getRegisterType(Context, ReturnMVT);
+ MVT MinVT = getRegisterType(ReturnMVT);
return VT.bitsLT(MinVT) ? MinVT : VT;
}
Op.getOperand(1), Op.getOperand(2));
}
+ // SSE2/SSE41/AVX2 integer max/min intrinsics.
+ case Intrinsic::x86_sse2_pmaxu_b:
+ case Intrinsic::x86_sse41_pmaxuw:
+ case Intrinsic::x86_sse41_pmaxud:
+ case Intrinsic::x86_avx2_pmaxu_b:
+ case Intrinsic::x86_avx2_pmaxu_w:
+ case Intrinsic::x86_avx2_pmaxu_d:
+ return DAG.getNode(X86ISD::UMAX, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
+ case Intrinsic::x86_sse2_pminu_b:
+ case Intrinsic::x86_sse41_pminuw:
+ case Intrinsic::x86_sse41_pminud:
+ case Intrinsic::x86_avx2_pminu_b:
+ case Intrinsic::x86_avx2_pminu_w:
+ case Intrinsic::x86_avx2_pminu_d:
+ return DAG.getNode(X86ISD::UMIN, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
+ case Intrinsic::x86_sse41_pmaxsb:
+ case Intrinsic::x86_sse2_pmaxs_w:
+ case Intrinsic::x86_sse41_pmaxsd:
+ case Intrinsic::x86_avx2_pmaxs_b:
+ case Intrinsic::x86_avx2_pmaxs_w:
+ case Intrinsic::x86_avx2_pmaxs_d:
+ return DAG.getNode(X86ISD::SMAX, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
+ case Intrinsic::x86_sse41_pminsb:
+ case Intrinsic::x86_sse2_pmins_w:
+ case Intrinsic::x86_sse41_pminsd:
+ case Intrinsic::x86_avx2_pmins_b:
+ case Intrinsic::x86_avx2_pmins_w:
+ case Intrinsic::x86_avx2_pmins_d:
+ return DAG.getNode(X86ISD::SMIN, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
+
// AVX2 variable shift intrinsics
case Intrinsic::x86_avx2_psllv_d:
case Intrinsic::x86_avx2_psllv_q:
static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
SelectionDAG &DAG) {
+ DebugLoc dl = Op.getDebugLoc();
EVT VT = Op.getValueType();
// Decompose 256-bit ops into smaller 128-bit ops.
if (VT.is256BitVector() && !Subtarget->hasInt256())
return Lower256IntArith(Op, DAG);
+ SDValue A = Op.getOperand(0);
+ SDValue B = Op.getOperand(1);
+
+ // Lower v4i32 mul as 2x shuffle, 2x pmuludq, 2x shuffle.
+ if (VT == MVT::v4i32) {
+ assert(Subtarget->hasSSE2() && !Subtarget->hasSSE41() &&
+ "Should not custom lower when pmuldq is available!");
+
+ // Extract the odd parts.
+ const int UnpackMask[] = { 1, -1, 3, -1 };
+ SDValue Aodds = DAG.getVectorShuffle(VT, dl, A, A, UnpackMask);
+ SDValue Bodds = DAG.getVectorShuffle(VT, dl, B, B, UnpackMask);
+
+ // Multiply the even parts.
+ SDValue Evens = DAG.getNode(X86ISD::PMULUDQ, dl, MVT::v2i64, A, B);
+ // Now multiply odd parts.
+ SDValue Odds = DAG.getNode(X86ISD::PMULUDQ, dl, MVT::v2i64, Aodds, Bodds);
+
+ Evens = DAG.getNode(ISD::BITCAST, dl, VT, Evens);
+ Odds = DAG.getNode(ISD::BITCAST, dl, VT, Odds);
+
+ // Merge the two vectors back together with a shuffle. This expands into 2
+ // shuffles.
+ const int ShufMask[] = { 0, 4, 2, 6 };
+ return DAG.getVectorShuffle(VT, dl, Evens, Odds, ShufMask);
+ }
+
assert((VT == MVT::v2i64 || VT == MVT::v4i64) &&
"Only know how to lower V2I64/V4I64 multiply");
- DebugLoc dl = Op.getDebugLoc();
-
// Ahi = psrlqi(a, 32);
// Bhi = psrlqi(b, 32);
//
// AhiBlo = psllqi(AhiBlo, 32);
// return AloBlo + AloBhi + AhiBlo;
- SDValue A = Op.getOperand(0);
- SDValue B = Op.getOperand(1);
-
SDValue ShAmt = DAG.getConstant(32, MVT::i32);
SDValue Ahi = DAG.getNode(X86ISD::VSRLI, dl, VT, A, ShAmt);
case X86ISD::HSUB: return "X86ISD::HSUB";
case X86ISD::FHADD: return "X86ISD::FHADD";
case X86ISD::FHSUB: return "X86ISD::FHSUB";
+ case X86ISD::UMAX: return "X86ISD::UMAX";
+ case X86ISD::UMIN: return "X86ISD::UMIN";
+ case X86ISD::SMAX: return "X86ISD::SMAX";
+ case X86ISD::SMIN: return "X86ISD::SMIN";
case X86ISD::FMAX: return "X86ISD::FMAX";
case X86ISD::FMIN: return "X86ISD::FMIN";
case X86ISD::FMAXC: return "X86ISD::FMAXC";
return SDValue();
}
+/// \brief Matches a VSELECT onto min/max or return 0 if the node doesn't match.
+static unsigned matchIntegerMINMAX(SDValue Cond, EVT VT, SDValue LHS,
+ SDValue RHS, SelectionDAG &DAG,
+ const X86Subtarget *Subtarget) {
+ if (!VT.isVector())
+ return 0;
+
+ switch (VT.getSimpleVT().SimpleTy) {
+ default: return 0;
+ case MVT::v32i8:
+ case MVT::v16i16:
+ case MVT::v8i32:
+ if (!Subtarget->hasAVX2())
+ return 0;
+ case MVT::v16i8:
+ case MVT::v8i16:
+ case MVT::v4i32:
+ if (!Subtarget->hasSSE2())
+ return 0;
+ }
+
+ // SSE2 has only a small subset of the operations.
+ bool hasUnsigned = Subtarget->hasSSE41() ||
+ (Subtarget->hasSSE2() && VT == MVT::v16i8);
+ bool hasSigned = Subtarget->hasSSE41() ||
+ (Subtarget->hasSSE2() && VT == MVT::v8i16);
+
+ ISD::CondCode CC = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
+
+ // Check for x CC y ? x : y.
+ if (DAG.isEqualTo(LHS, Cond.getOperand(0)) &&
+ DAG.isEqualTo(RHS, Cond.getOperand(1))) {
+ switch (CC) {
+ default: break;
+ case ISD::SETULT:
+ case ISD::SETULE:
+ return hasUnsigned ? X86ISD::UMIN : 0;
+ case ISD::SETUGT:
+ case ISD::SETUGE:
+ return hasUnsigned ? X86ISD::UMAX : 0;
+ case ISD::SETLT:
+ case ISD::SETLE:
+ return hasSigned ? X86ISD::SMIN : 0;
+ case ISD::SETGT:
+ case ISD::SETGE:
+ return hasSigned ? X86ISD::SMAX : 0;
+ }
+ // Check for x CC y ? y : x -- a min/max with reversed arms.
+ } else if (DAG.isEqualTo(LHS, Cond.getOperand(1)) &&
+ DAG.isEqualTo(RHS, Cond.getOperand(0))) {
+ switch (CC) {
+ default: break;
+ case ISD::SETULT:
+ case ISD::SETULE:
+ return hasUnsigned ? X86ISD::UMAX : 0;
+ case ISD::SETUGT:
+ case ISD::SETUGE:
+ return hasUnsigned ? X86ISD::UMIN : 0;
+ case ISD::SETLT:
+ case ISD::SETLE:
+ return hasSigned ? X86ISD::SMAX : 0;
+ case ISD::SETGT:
+ case ISD::SETGE:
+ return hasSigned ? X86ISD::SMIN : 0;
+ }
+ }
+
+ return 0;
+}
+
/// PerformSELECTCombine - Do target-specific dag combines on SELECT and VSELECT
/// nodes.
static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG,
}
}
+ // Try to match a min/max vector operation.
+ if (!DCI.isBeforeLegalize() &&
+ N->getOpcode() == ISD::VSELECT && Cond.getOpcode() == ISD::SETCC)
+ if (unsigned Op = matchIntegerMINMAX(Cond, VT, LHS, RHS, DAG, Subtarget))
+ return DAG.getNode(Op, DL, N->getValueType(0), LHS, RHS);
+
+
// If we know that this node is legal then we know that it is going to be
// matched by one of the SSE/AVX BLEND instructions. These instructions only
// depend on the highest bit in each word. Try to use SimplifyDemandedBits
ISD::LoadExtType Ext = Ld->getExtensionType();
// If this is a vector EXT Load then attempt to optimize it using a
- // shuffle. We need SSSE3 shuffles.
- // SEXT loads are suppoted starting SSE41.
- // We generate X86ISD::VSEXT for them.
+ // shuffle. If SSSE3 is not available we may emit an illegal shuffle but the
+ // expansion is still better than scalar code.
+ // We generate X86ISD::VSEXT for SEXTLOADs if it's available, otherwise we'll
+ // emit a shuffle and a arithmetic shift.
// TODO: It is possible to support ZExt by zeroing the undef values
// during the shuffle phase or after the shuffle.
- if (RegVT.isVector() && RegVT.isInteger() &&
- ((Ext == ISD::EXTLOAD && Subtarget->hasSSSE3()) ||
- (Ext == ISD::SEXTLOAD && Subtarget->hasSSE41()))){
+ if (RegVT.isVector() && RegVT.isInteger() && Subtarget->hasSSE2() &&
+ (Ext == ISD::EXTLOAD || Ext == ISD::SEXTLOAD)) {
assert(MemVT != RegVT && "Cannot extend to the same type");
assert(MemVT.isVector() && "Must load a vector from memory");
unsigned SizeRatio = RegSz/MemSz;
if (Ext == ISD::SEXTLOAD) {
- SDValue Sext = DAG.getNode(X86ISD::VSEXT, dl, RegVT, SlicedVec);
- return DCI.CombineTo(N, Sext, TF, true);
+ // If we have SSE4.1 we can directly emit a VSEXT node.
+ if (Subtarget->hasSSE41()) {
+ SDValue Sext = DAG.getNode(X86ISD::VSEXT, dl, RegVT, SlicedVec);
+ return DCI.CombineTo(N, Sext, TF, true);
+ }
+
+ // Otherwise we'll shuffle the small elements in the high bits of the
+ // larger type and perform an arithmetic shift. If the shift is not legal
+ // it's better to scalarize.
+ if (!TLI.isOperationLegalOrCustom(ISD::SRA, RegVT))
+ return SDValue();
+
+ // Redistribute the loaded elements into the different locations.
+ SmallVector<int, 8> ShuffleVec(NumElems * SizeRatio, -1);
+ for (unsigned i = 0; i != NumElems; ++i)
+ ShuffleVec[i*SizeRatio + SizeRatio-1] = i;
+
+ SDValue Shuff = DAG.getVectorShuffle(WideVecVT, dl, SlicedVec,
+ DAG.getUNDEF(WideVecVT),
+ &ShuffleVec[0]);
+
+ Shuff = DAG.getNode(ISD::BITCAST, dl, RegVT, Shuff);
+
+ // Build the arithmetic shift.
+ unsigned Amt = RegVT.getVectorElementType().getSizeInBits() -
+ MemVT.getVectorElementType().getSizeInBits();
+ SmallVector<SDValue, 8> C(NumElems,
+ DAG.getConstant(Amt, RegVT.getScalarType()));
+ SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, dl, RegVT, &C[0], C.size());
+ Shuff = DAG.getNode(ISD::SRA, dl, RegVT, Shuff, BV);
+
+ return DCI.CombineTo(N, Shuff, TF, true);
}
+
// Redistribute the loaded elements into the different locations.
SmallVector<int, 8> ShuffleVec(NumElems * SizeRatio, -1);
for (unsigned i = 0; i != NumElems; ++i)
return VectorTargetTransformImpl::getArithmeticInstrCost(Opcode, Ty);
}
+unsigned
+X86VectorTargetTransformInfo::getMemoryOpCost(unsigned Opcode, Type *Src,
+ unsigned Alignment,
+ unsigned AddressSpace) const {
+ // Legalize the type.
+ std::pair<unsigned, MVT> LT = getTypeLegalizationCost(Src);
+ assert(Opcode == Instruction::Load || Opcode == Instruction::Store &&
+ "Invalid Opcode");
+
+ const X86Subtarget &ST =
+ TLI->getTargetMachine().getSubtarget<X86Subtarget>();
+
+ // Each load/store unit costs 1.
+ unsigned Cost = LT.first * 1;
+
+ // On Sandybridge 256bit load/stores are double pumped
+ // (but not on Haswell).
+ if (LT.second.getSizeInBits() > 128 && !ST.hasAVX2())
+ Cost*=2;
+
+ return Cost;
+}
+
unsigned
X86VectorTargetTransformInfo::getVectorInstrCost(unsigned Opcode, Type *Val,
unsigned Index) const {