diff options
Diffstat (limited to 'lib/Target/X86/X86ISelLowering.cpp')
-rw-r--r-- | lib/Target/X86/X86ISelLowering.cpp | 380 |
1 files changed, 232 insertions, 148 deletions
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index 294a6a7..5096d9a 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -235,10 +235,16 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) // Setup Windows compiler runtime calls. setLibcallName(RTLIB::SDIV_I64, "_alldiv"); setLibcallName(RTLIB::UDIV_I64, "_aulldiv"); + setLibcallName(RTLIB::SREM_I64, "_allrem"); + setLibcallName(RTLIB::UREM_I64, "_aullrem"); + setLibcallName(RTLIB::MUL_I64, "_allmul"); setLibcallName(RTLIB::FPTOUINT_F64_I64, "_ftol2"); setLibcallName(RTLIB::FPTOUINT_F32_I64, "_ftol2"); setLibcallCallingConv(RTLIB::SDIV_I64, CallingConv::X86_StdCall); setLibcallCallingConv(RTLIB::UDIV_I64, CallingConv::X86_StdCall); + setLibcallCallingConv(RTLIB::SREM_I64, CallingConv::X86_StdCall); + setLibcallCallingConv(RTLIB::UREM_I64, CallingConv::X86_StdCall); + setLibcallCallingConv(RTLIB::MUL_I64, CallingConv::X86_StdCall); setLibcallCallingConv(RTLIB::FPTOUINT_F64_I64, CallingConv::C); setLibcallCallingConv(RTLIB::FPTOUINT_F32_I64, CallingConv::C); } @@ -646,6 +652,10 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) addLegalFPImmediate(APFloat(-1.0f)); // FLD1/FCHS } + // We don't support FMA. + setOperationAction(ISD::FMA, MVT::f64, Expand); + setOperationAction(ISD::FMA, MVT::f32, Expand); + // Long double always uses X87. if (!UseSoftFloat) { addRegisterClass(MVT::f80, X86::RFP80RegisterClass); @@ -670,6 +680,8 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setOperationAction(ISD::FSIN , MVT::f80 , Expand); setOperationAction(ISD::FCOS , MVT::f80 , Expand); } + + setOperationAction(ISD::FMA, MVT::f80, Expand); } // Always use a library call for pow. @@ -976,7 +988,6 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) addRegisterClass(MVT::v32i8, X86::VR256RegisterClass); setOperationAction(ISD::LOAD, MVT::v8f32, Legal); - setOperationAction(ISD::LOAD, MVT::v8i32, Legal); setOperationAction(ISD::LOAD, MVT::v4f64, Legal); setOperationAction(ISD::LOAD, MVT::v4i64, Legal); @@ -994,63 +1005,58 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setOperationAction(ISD::FSQRT, MVT::v4f64, Legal); setOperationAction(ISD::FNEG, MVT::v4f64, Custom); - // Custom lower build_vector, vector_shuffle, scalar_to_vector, - // insert_vector_elt extract_subvector and extract_vector_elt for - // 256-bit types. - for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE; - i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; - ++i) { - MVT::SimpleValueType VT = (MVT::SimpleValueType)i; - // Do not attempt to custom lower non-256-bit vectors - if (!isPowerOf2_32(MVT(VT).getVectorNumElements()) - || (MVT(VT).getSizeInBits() < 256)) - continue; - setOperationAction(ISD::BUILD_VECTOR, VT, Custom); - setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom); - setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom); - setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom); - setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Custom); - } - // Custom-lower insert_subvector and extract_subvector based on - // the result type. + // Custom lower several nodes for 256-bit types. for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE; - i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; - ++i) { - MVT::SimpleValueType VT = (MVT::SimpleValueType)i; - // Do not attempt to custom lower non-256-bit vectors - if (!isPowerOf2_32(MVT(VT).getVectorNumElements())) + i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) { + MVT::SimpleValueType SVT = (MVT::SimpleValueType)i; + EVT VT = SVT; + + // Extract subvector is special because the value type + // (result) is 128-bit but the source is 256-bit wide. + if (VT.is128BitVector()) + setOperationAction(ISD::EXTRACT_SUBVECTOR, SVT, Custom); + + // Do not attempt to custom lower other non-256-bit vectors + if (!VT.is256BitVector()) continue; - if (MVT(VT).getSizeInBits() == 128) { - setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Custom); - } - else if (MVT(VT).getSizeInBits() == 256) { - setOperationAction(ISD::INSERT_SUBVECTOR, VT, Custom); - } + setOperationAction(ISD::BUILD_VECTOR, SVT, Custom); + setOperationAction(ISD::VECTOR_SHUFFLE, SVT, Custom); + setOperationAction(ISD::INSERT_VECTOR_ELT, SVT, Custom); + setOperationAction(ISD::EXTRACT_VECTOR_ELT, SVT, Custom); + setOperationAction(ISD::SCALAR_TO_VECTOR, SVT, Custom); + setOperationAction(ISD::INSERT_SUBVECTOR, SVT, Custom); } // Promote v32i8, v16i16, v8i32 select, and, or, xor to v4i64. - // Don't promote loads because we need them for VPERM vector index versions. + for (unsigned i = (unsigned)MVT::v32i8; i != (unsigned)MVT::v4i64; ++i) { + MVT::SimpleValueType SVT = (MVT::SimpleValueType)i; + EVT VT = SVT; - for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE; - VT != (unsigned)MVT::LAST_VECTOR_VALUETYPE; - VT++) { - if (!isPowerOf2_32(MVT((MVT::SimpleValueType)VT).getVectorNumElements()) - || (MVT((MVT::SimpleValueType)VT).getSizeInBits() < 256)) + // Do not attempt to promote non-256-bit vectors + if (!VT.is256BitVector()) continue; - setOperationAction(ISD::AND, (MVT::SimpleValueType)VT, Promote); - AddPromotedToType (ISD::AND, (MVT::SimpleValueType)VT, MVT::v4i64); - setOperationAction(ISD::OR, (MVT::SimpleValueType)VT, Promote); - AddPromotedToType (ISD::OR, (MVT::SimpleValueType)VT, MVT::v4i64); - setOperationAction(ISD::XOR, (MVT::SimpleValueType)VT, Promote); - AddPromotedToType (ISD::XOR, (MVT::SimpleValueType)VT, MVT::v4i64); - //setOperationAction(ISD::LOAD, (MVT::SimpleValueType)VT, Promote); - //AddPromotedToType (ISD::LOAD, (MVT::SimpleValueType)VT, MVT::v4i64); - setOperationAction(ISD::SELECT, (MVT::SimpleValueType)VT, Promote); - AddPromotedToType (ISD::SELECT, (MVT::SimpleValueType)VT, MVT::v4i64); + + setOperationAction(ISD::AND, SVT, Promote); + AddPromotedToType (ISD::AND, SVT, MVT::v4i64); + setOperationAction(ISD::OR, SVT, Promote); + AddPromotedToType (ISD::OR, SVT, MVT::v4i64); + setOperationAction(ISD::XOR, SVT, Promote); + AddPromotedToType (ISD::XOR, SVT, MVT::v4i64); + setOperationAction(ISD::LOAD, SVT, Promote); + AddPromotedToType (ISD::LOAD, SVT, MVT::v4i64); + setOperationAction(ISD::SELECT, SVT, Promote); + AddPromotedToType (ISD::SELECT, SVT, MVT::v4i64); } } + // SIGN_EXTEND_INREGs are evaluated by the extend type. Handle the expansion + // of this type with custom code. + for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE; + VT != (unsigned)MVT::LAST_VECTOR_VALUETYPE; VT++) { + setOperationAction(ISD::SIGN_EXTEND_INREG, (MVT::SimpleValueType)VT, Custom); + } + // We want to custom lower some of our intrinsics. setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom); @@ -1511,20 +1517,15 @@ X86TargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag, // If this is a call to a function that returns an fp value on the floating // point stack, we must guarantee the the value is popped from the stack, so // a CopyFromReg is not good enough - the copy instruction may be eliminated - // if the return value is not used. We use the FpGET_ST0 instructions + // if the return value is not used. We use the FpPOP_RETVAL instruction // instead. if (VA.getLocReg() == X86::ST0 || VA.getLocReg() == X86::ST1) { // If we prefer to use the value in xmm registers, copy it out as f80 and // use a truncate to move it from fp stack reg to xmm reg. if (isScalarFPTypeInSSEReg(VA.getValVT())) CopyVT = MVT::f80; - bool isST0 = VA.getLocReg() == X86::ST0; - unsigned Opc = 0; - if (CopyVT == MVT::f32) Opc = isST0 ? X86::FpGET_ST0_32:X86::FpGET_ST1_32; - if (CopyVT == MVT::f64) Opc = isST0 ? X86::FpGET_ST0_64:X86::FpGET_ST1_64; - if (CopyVT == MVT::f80) Opc = isST0 ? X86::FpGET_ST0_80:X86::FpGET_ST1_80; SDValue Ops[] = { Chain, InFlag }; - Chain = SDValue(DAG.getMachineNode(Opc, dl, CopyVT, MVT::Other, MVT::Glue, - Ops, 2), 1); + Chain = SDValue(DAG.getMachineNode(X86::FpPOP_RETVAL, dl, CopyVT, + MVT::Other, MVT::Glue, Ops, 2), 1); Val = Chain.getValue(0); // Round the f80 to the right size, which also moves it to the appropriate @@ -1898,7 +1899,7 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain, } // Some CCs need callee pop. - if (Subtarget->IsCalleePop(isVarArg, CallConv)) { + if (X86::isCalleePop(CallConv, Is64Bit, isVarArg, GuaranteedTailCallOpt)) { FuncInfo->setBytesToPopOnReturn(StackSize); // Callee pops everything. } else { FuncInfo->setBytesToPopOnReturn(0); // Callee pops nothing. @@ -2271,6 +2272,8 @@ X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee, const GlobalValue *GV = G->getGlobal(); if (!GV->hasDLLImportLinkage()) { unsigned char OpFlags = 0; + bool ExtraLoad = false; + unsigned WrapperKind = ISD::DELETED_NODE; // On ELF targets, in both X86-64 and X86-32 mode, direct calls to // external symbols most go through the PLT in PIC mode. If the symbol @@ -2288,10 +2291,28 @@ X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee, // unless we're building with the leopard linker or later, which // automatically synthesizes these stubs. OpFlags = X86II::MO_DARWIN_STUB; + } else if (Subtarget->isPICStyleRIPRel() && + isa<Function>(GV) && + cast<Function>(GV)->hasFnAttr(Attribute::NonLazyBind)) { + // If the function is marked as non-lazy, generate an indirect call + // which loads from the GOT directly. This avoids runtime overhead + // at the cost of eager binding (and one extra byte of encoding). + OpFlags = X86II::MO_GOTPCREL; + WrapperKind = X86ISD::WrapperRIP; + ExtraLoad = true; } Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), G->getOffset(), OpFlags); + + // Add a wrapper if needed. + if (WrapperKind != ISD::DELETED_NODE) + Callee = DAG.getNode(X86ISD::WrapperRIP, dl, getPointerTy(), Callee); + // Add extra indirection if needed. + if (ExtraLoad) + Callee = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(), Callee, + MachinePointerInfo::getGOT(), + false, false, 0); } } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) { unsigned char OpFlags = 0; @@ -2363,7 +2384,7 @@ X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee, // Create the CALLSEQ_END node. unsigned NumBytesForCalleeToPush; - if (Subtarget->IsCalleePop(isVarArg, CallConv)) + if (X86::isCalleePop(CallConv, Is64Bit, isVarArg, GuaranteedTailCallOpt)) NumBytesForCalleeToPush = NumBytes; // Callee pops everything else if (!Is64Bit && !IsTailCallConvention(CallConv) && IsStructRet) // If this is a call to a struct-return function, the callee @@ -2485,6 +2506,10 @@ bool MatchingStackOffset(SDValue Arg, unsigned Offset, ISD::ArgFlagsTy Flags, if (!FINode) return false; FI = FINode->getIndex(); + } else if (Arg.getOpcode() == ISD::FrameIndex && Flags.isByVal()) { + FrameIndexSDNode *FINode = cast<FrameIndexSDNode>(Arg); + FI = FINode->getIndex(); + Bytes = Flags.getByValSize(); } else return false; @@ -2536,6 +2561,11 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee, if (isCalleeStructRet || isCallerStructRet) return false; + // An stdcall caller is expected to clean up its arguments; the callee + // isn't going to do that. + if (!CCMatch && CallerCC==CallingConv::X86_StdCall) + return false; + // Do not sibcall optimize vararg calls unless all arguments are passed via // registers. if (isVarArg && !Outs.empty()) { @@ -2672,11 +2702,6 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee, } } - // An stdcall caller is expected to clean up its arguments; the callee - // isn't going to do that. - if (!CCMatch && CallerCC==CallingConv::X86_StdCall) - return false; - return true; } @@ -2856,6 +2881,29 @@ bool X86::isOffsetSuitableForCodeModel(int64_t Offset, CodeModel::Model M, return false; } +/// isCalleePop - Determines whether the callee is required to pop its +/// own arguments. Callee pop is necessary to support tail calls. +bool X86::isCalleePop(CallingConv::ID CallingConv, + bool is64Bit, bool IsVarArg, bool TailCallOpt) { + if (IsVarArg) + return false; + + switch (CallingConv) { + default: + return false; + case CallingConv::X86_StdCall: + return !is64Bit; + case CallingConv::X86_FastCall: + return !is64Bit; + case CallingConv::X86_ThisCall: + return !is64Bit; + case CallingConv::Fast: + return TailCallOpt; + case CallingConv::GHC: + return TailCallOpt; + } +} + /// TranslateX86CC - do a one to one translation of a ISD::CondCode to the X86 /// specific condition code, returning the condition code and the LHS/RHS of the /// comparison to make. @@ -3790,19 +3838,24 @@ static SDValue getZeroVector(EVT VT, bool HasSSE2, SelectionDAG &DAG, } /// getOnesVector - Returns a vector of specified type with all bits set. -/// +/// Always build ones vectors as <4 x i32> or <8 x i32> bitcasted to +/// their original type, ensuring they get CSE'd. static SDValue getOnesVector(EVT VT, SelectionDAG &DAG, DebugLoc dl) { assert(VT.isVector() && "Expected a vector type"); + assert((VT.is128BitVector() || VT.is256BitVector()) + && "Expected a 128-bit or 256-bit vector type"); - // Always build ones vectors as <4 x i32> or <2 x i32> bitcasted to their dest - // type. This ensures they get CSE'd. SDValue Cst = DAG.getTargetConstant(~0U, MVT::i32); + SDValue Vec; - Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst); + if (VT.is256BitVector()) { + SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst }; + Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops, 8); + } else + Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst); return DAG.getNode(ISD::BITCAST, dl, VT, Vec); } - /// NormalizeMask - V2 is a splat, modify the mask (if needed) so all elements /// that point to V2 points to its first element. static SDValue NormalizeMask(ShuffleVectorSDNode *SVOp, SelectionDAG &DAG) { @@ -4417,17 +4470,17 @@ X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const { return ConcatVectors(Lower, Upper, DAG); } - // All zero's are handled with pxor in SSE2 and above, xorps in SSE1. - // All one's are handled with pcmpeqd. In AVX, zero's are handled with - // vpxor in 128-bit and xor{pd,ps} in 256-bit, but no 256 version of pcmpeqd - // is present, so AllOnes is ignored. + // All zero's: + // - pxor (SSE2), xorps (SSE1), vpxor (128 AVX), xorp[s|d] (256 AVX) + // All one's: + // - pcmpeqd (SSE2 and 128 AVX), fallback to constant pools (256 AVX) if (ISD::isBuildVectorAllZeros(Op.getNode()) || - (Op.getValueType().getSizeInBits() != 256 && - ISD::isBuildVectorAllOnes(Op.getNode()))) { - // Canonicalize this to <4 x i32> (SSE) to + ISD::isBuildVectorAllOnes(Op.getNode())) { + // Canonicalize this to <4 x i32> or <8 x 32> (SSE) to // 1) ensure the zero vectors are CSE'd, and 2) ensure that i64 scalars are // eliminated on x86-32 hosts. - if (Op.getValueType() == MVT::v4i32) + if (Op.getValueType() == MVT::v4i32 || + Op.getValueType() == MVT::v8i32) return Op; if (ISD::isBuildVectorAllOnes(Op.getNode())) @@ -8874,8 +8927,8 @@ SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const { } // Lower SHL with variable shift amount. - // Cannot lower SHL without SSE4.1 or later. - if (!Subtarget->hasSSE41()) return SDValue(); + // Cannot lower SHL without SSE2 or later. + if (!Subtarget->hasSSE2()) return SDValue(); if (VT == MVT::v4i32 && Op->getOpcode() == ISD::SHL) { Op = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT, @@ -9022,13 +9075,66 @@ SDValue X86TargetLowering::LowerXALUO(SDValue Op, SelectionDAG &DAG) const { return Sum; } +SDValue X86TargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const{ + DebugLoc dl = Op.getDebugLoc(); + SDNode* Node = Op.getNode(); + EVT ExtraVT = cast<VTSDNode>(Node->getOperand(1))->getVT(); + EVT VT = Node->getValueType(0); + + if (Subtarget->hasSSE2() && VT.isVector()) { + unsigned BitsDiff = VT.getScalarType().getSizeInBits() - + ExtraVT.getScalarType().getSizeInBits(); + SDValue ShAmt = DAG.getConstant(BitsDiff, MVT::i32); + + unsigned SHLIntrinsicsID = 0; + unsigned SRAIntrinsicsID = 0; + switch (VT.getSimpleVT().SimpleTy) { + default: + return SDValue(); + case MVT::v2i64: { + SHLIntrinsicsID = Intrinsic::x86_sse2_pslli_q; + SRAIntrinsicsID = 0; + break; + } + case MVT::v4i32: { + SHLIntrinsicsID = Intrinsic::x86_sse2_pslli_d; + SRAIntrinsicsID = Intrinsic::x86_sse2_psrai_d; + break; + } + case MVT::v8i16: { + SHLIntrinsicsID = Intrinsic::x86_sse2_pslli_w; + SRAIntrinsicsID = Intrinsic::x86_sse2_psrai_w; + break; + } + } + + SDValue Tmp1 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT, + DAG.getConstant(SHLIntrinsicsID, MVT::i32), + Node->getOperand(0), ShAmt); + + // In case of 1 bit sext, no need to shr + if (ExtraVT.getScalarType().getSizeInBits() == 1) return Tmp1; + + if (SRAIntrinsicsID) { + Tmp1 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT, + DAG.getConstant(SRAIntrinsicsID, MVT::i32), + Tmp1, ShAmt); + } + return Tmp1; + } + + return SDValue(); +} + + SDValue X86TargetLowering::LowerMEMBARRIER(SDValue Op, SelectionDAG &DAG) const{ DebugLoc dl = Op.getDebugLoc(); - if (!Subtarget->hasSSE2()) { + // Go ahead and emit the fence on x86-64 even if we asked for no-sse2. + // There isn't any reason to disable it if the target processor supports it. + if (!Subtarget->hasSSE2() && !Subtarget->is64Bit()) { SDValue Chain = Op.getOperand(0); - SDValue Zero = DAG.getConstant(0, - Subtarget->is64Bit() ? MVT::i64 : MVT::i32); + SDValue Zero = DAG.getConstant(0, MVT::i32); SDValue Ops[] = { DAG.getRegister(X86::ESP, MVT::i32), // Base DAG.getTargetConstant(1, MVT::i8), // Scale @@ -9183,6 +9289,7 @@ static SDValue LowerADDC_ADDE_SUBC_SUBE(SDValue Op, SelectionDAG &DAG) { SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const { switch (Op.getOpcode()) { default: llvm_unreachable("Should not custom lower this!"); + case ISD::SIGN_EXTEND_INREG: return LowerSIGN_EXTEND_INREG(Op,DAG); case ISD::MEMBARRIER: return LowerMEMBARRIER(Op,DAG); case ISD::ATOMIC_CMP_SWAP: return LowerCMP_SWAP(Op,DAG); case ISD::ATOMIC_LOAD_SUB: return LowerLOAD_SUB(Op,DAG); @@ -9281,6 +9388,7 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N, default: assert(false && "Do not know how to custom type legalize this operation!"); return; + case ISD::SIGN_EXTEND_INREG: case ISD::ADDC: case ISD::ADDE: case ISD::SUBC: @@ -9415,7 +9523,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const { case X86ISD::PINSRB: return "X86ISD::PINSRB"; case X86ISD::PINSRW: return "X86ISD::PINSRW"; case X86ISD::PSHUFB: return "X86ISD::PSHUFB"; - case X86ISD::PANDN: return "X86ISD::PANDN"; + case X86ISD::ANDNP: return "X86ISD::ANDNP"; case X86ISD::PSIGNB: return "X86ISD::PSIGNB"; case X86ISD::PSIGNW: return "X86ISD::PSIGNW"; case X86ISD::PSIGND: return "X86ISD::PSIGND"; @@ -11766,10 +11874,12 @@ static SDValue PerformAndCombine(SDNode *N, SelectionDAG &DAG, if (R.getNode()) return R; - // Want to form PANDN nodes, in the hopes of then easily combining them with - // OR and AND nodes to form PBLEND/PSIGN. + // Want to form ANDNP nodes: + // 1) In the hopes of then easily combining them with OR and AND nodes + // to form PBLEND/PSIGN. + // 2) To match ANDN packed intrinsics EVT VT = N->getValueType(0); - if (VT != MVT::v2i64) + if (VT != MVT::v2i64 && VT != MVT::v4i64) return SDValue(); SDValue N0 = N->getOperand(0); @@ -11779,12 +11889,12 @@ static SDValue PerformAndCombine(SDNode *N, SelectionDAG &DAG, // Check LHS for vnot if (N0.getOpcode() == ISD::XOR && ISD::isBuildVectorAllOnes(N0.getOperand(1).getNode())) - return DAG.getNode(X86ISD::PANDN, DL, VT, N0.getOperand(0), N1); + return DAG.getNode(X86ISD::ANDNP, DL, VT, N0.getOperand(0), N1); // Check RHS for vnot if (N1.getOpcode() == ISD::XOR && ISD::isBuildVectorAllOnes(N1.getOperand(1).getNode())) - return DAG.getNode(X86ISD::PANDN, DL, VT, N1.getOperand(0), N0); + return DAG.getNode(X86ISD::ANDNP, DL, VT, N1.getOperand(0), N0); return SDValue(); } @@ -11810,10 +11920,10 @@ static SDValue PerformOrCombine(SDNode *N, SelectionDAG &DAG, if (Subtarget->hasSSSE3()) { if (VT == MVT::v2i64) { // Canonicalize pandn to RHS - if (N0.getOpcode() == X86ISD::PANDN) + if (N0.getOpcode() == X86ISD::ANDNP) std::swap(N0, N1); // or (and (m, x), (pandn m, y)) - if (N0.getOpcode() == ISD::AND && N1.getOpcode() == X86ISD::PANDN) { + if (N0.getOpcode() == ISD::AND && N1.getOpcode() == X86ISD::ANDNP) { SDValue Mask = N1.getOperand(0); SDValue X = N1.getOperand(1); SDValue Y; @@ -11822,7 +11932,7 @@ static SDValue PerformOrCombine(SDNode *N, SelectionDAG &DAG, if (N0.getOperand(1) == Mask) Y = N0.getOperand(0); - // Check to see if the mask appeared in both the AND and PANDN and + // Check to see if the mask appeared in both the AND and ANDNP and if (!Y.getNode()) return SDValue(); @@ -12166,8 +12276,8 @@ static SDValue PerformSETCCCombine(SDNode *N, SelectionDAG &DAG) { return SDValue(); } -static SDValue PerformSINT_TO_FPCombine(SDNode *N, SelectionDAG &DAG, const X86TargetLowering *XTLI) { - DebugLoc dl = N->getDebugLoc(); +static SDValue PerformSINT_TO_FPCombine(SDNode *N, SelectionDAG &DAG, + const X86TargetLowering *XTLI) { SDValue Op0 = N->getOperand(0); // Transform (SINT_TO_FP (i64 ...)) into an x87 operation if we have // a 32-bit target where SSE doesn't support i64->FP operations. @@ -12178,7 +12288,8 @@ static SDValue PerformSINT_TO_FPCombine(SDNode *N, SelectionDAG &DAG, const X86T ISD::isNON_EXTLoad(Op0.getNode()) && Op0.hasOneUse() && !XTLI->getSubtarget()->is64Bit() && !DAG.getTargetLoweringInfo().isTypeLegal(VT)) { - SDValue FILDChain = XTLI->BuildFILD(SDValue(N, 0), Ld->getValueType(0), Ld->getChain(), Op0, DAG); + SDValue FILDChain = XTLI->BuildFILD(SDValue(N, 0), Ld->getValueType(0), + Ld->getChain(), Op0, DAG); DAG.ReplaceAllUsesOfValueWith(Op0.getValue(1), FILDChain.getValue(1)); return FILDChain; } @@ -12549,6 +12660,7 @@ X86TargetLowering::getConstraintType(const std::string &Constraint) const { case 'y': case 'x': case 'Y': + case 'l': return C_RegisterClass; case 'a': case 'b': @@ -12832,60 +12944,6 @@ void X86TargetLowering::LowerAsmOperandForConstraint(SDValue Op, return TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG); } -std::vector<unsigned> X86TargetLowering:: -getRegClassForInlineAsmConstraint(const std::string &Constraint, - EVT VT) const { - if (Constraint.size() == 1) { - // FIXME: not handling fp-stack yet! - switch (Constraint[0]) { // GCC X86 Constraint Letters - default: break; // Unknown constraint letter - case 'q': // GENERAL_REGS in 64-bit mode, Q_REGS in 32-bit mode. - if (Subtarget->is64Bit()) { - if (VT == MVT::i32) - return make_vector<unsigned>(X86::EAX, X86::EDX, X86::ECX, X86::EBX, - X86::ESI, X86::EDI, X86::R8D, X86::R9D, - X86::R10D,X86::R11D,X86::R12D, - X86::R13D,X86::R14D,X86::R15D, - X86::EBP, X86::ESP, 0); - else if (VT == MVT::i16) - return make_vector<unsigned>(X86::AX, X86::DX, X86::CX, X86::BX, - X86::SI, X86::DI, X86::R8W,X86::R9W, - X86::R10W,X86::R11W,X86::R12W, - X86::R13W,X86::R14W,X86::R15W, - X86::BP, X86::SP, 0); - else if (VT == MVT::i8) - return make_vector<unsigned>(X86::AL, X86::DL, X86::CL, X86::BL, - X86::SIL, X86::DIL, X86::R8B,X86::R9B, - X86::R10B,X86::R11B,X86::R12B, - X86::R13B,X86::R14B,X86::R15B, - X86::BPL, X86::SPL, 0); - - else if (VT == MVT::i64) - return make_vector<unsigned>(X86::RAX, X86::RDX, X86::RCX, X86::RBX, - X86::RSI, X86::RDI, X86::R8, X86::R9, - X86::R10, X86::R11, X86::R12, - X86::R13, X86::R14, X86::R15, - X86::RBP, X86::RSP, 0); - - break; - } - // 32-bit fallthrough - case 'Q': // Q_REGS - if (VT == MVT::i32) - return make_vector<unsigned>(X86::EAX, X86::EDX, X86::ECX, X86::EBX, 0); - else if (VT == MVT::i16) - return make_vector<unsigned>(X86::AX, X86::DX, X86::CX, X86::BX, 0); - else if (VT == MVT::i8) - return make_vector<unsigned>(X86::AL, X86::DL, X86::CL, X86::BL, 0); - else if (VT == MVT::i64) - return make_vector<unsigned>(X86::RAX, X86::RDX, X86::RCX, X86::RBX, 0); - break; - } - } - - return std::vector<unsigned>(); -} - std::pair<unsigned, const TargetRegisterClass*> X86TargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, EVT VT) const { @@ -12895,9 +12953,35 @@ X86TargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, // GCC Constraint Letters switch (Constraint[0]) { default: break; + // TODO: Slight differences here in allocation order and leaving + // RIP in the class. Do they matter any more here than they do + // in the normal allocation? + case 'q': // GENERAL_REGS in 64-bit mode, Q_REGS in 32-bit mode. + if (Subtarget->is64Bit()) { + if (VT == MVT::i32 || VT == MVT::f32) + return std::make_pair(0U, X86::GR32RegisterClass); + else if (VT == MVT::i16) + return std::make_pair(0U, X86::GR16RegisterClass); + else if (VT == MVT::i8 || VT == MVT::i1) + return std::make_pair(0U, X86::GR8RegisterClass); + else if (VT == MVT::i64 || VT == MVT::f64) + return std::make_pair(0U, X86::GR64RegisterClass); + break; + } + // 32-bit fallthrough + case 'Q': // Q_REGS + if (VT == MVT::i32 || VT == MVT::f32) + return std::make_pair(0U, X86::GR32_ABCDRegisterClass); + else if (VT == MVT::i16) + return std::make_pair(0U, X86::GR16_ABCDRegisterClass); + else if (VT == MVT::i8 || VT == MVT::i1) + return std::make_pair(0U, X86::GR8_ABCD_LRegisterClass); + else if (VT == MVT::i64) + return std::make_pair(0U, X86::GR64_ABCDRegisterClass); + break; case 'r': // GENERAL_REGS case 'l': // INDEX_REGS - if (VT == MVT::i8) + if (VT == MVT::i8 || VT == MVT::i1) return std::make_pair(0U, X86::GR8RegisterClass); if (VT == MVT::i16) return std::make_pair(0U, X86::GR16RegisterClass); @@ -12905,7 +12989,7 @@ X86TargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, return std::make_pair(0U, X86::GR32RegisterClass); return std::make_pair(0U, X86::GR64RegisterClass); case 'R': // LEGACY_REGS - if (VT == MVT::i8) + if (VT == MVT::i8 || VT == MVT::i1) return std::make_pair(0U, X86::GR8_NOREXRegisterClass); if (VT == MVT::i16) return std::make_pair(0U, X86::GR16_NOREXRegisterClass); |