diff options
Diffstat (limited to 'llvm/lib/Target/ARM/ARMISelLowering.cpp')
| -rw-r--r-- | llvm/lib/Target/ARM/ARMISelLowering.cpp | 179 |
1 files changed, 158 insertions, 21 deletions
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp index 24992c7d6f2..faf12fefffd 100644 --- a/llvm/lib/Target/ARM/ARMISelLowering.cpp +++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp @@ -445,8 +445,7 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() && !Subtarget->isThumb1Only()) { addRegisterClass(MVT::f32, &ARM::SPRRegClass); - if (!Subtarget->isFPOnlySP()) - addRegisterClass(MVT::f64, &ARM::DPRRegClass); + addRegisterClass(MVT::f64, &ARM::DPRRegClass); } for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE; @@ -628,6 +627,39 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) if (!Subtarget->isThumb1Only()) setTargetDAGCombine(ISD::ADDC); + if (Subtarget->isFPOnlySP()) { + // When targetting a floating-point unit with only single-precision + // operations, f64 is legal for the few double-precision instructions which + // are present However, no double-precision operations other than moves, + // loads and stores are provided by the hardware. + setOperationAction(ISD::FADD, MVT::f64, Expand); + setOperationAction(ISD::FSUB, MVT::f64, Expand); + setOperationAction(ISD::FMUL, MVT::f64, Expand); + setOperationAction(ISD::FMA, MVT::f64, Expand); + setOperationAction(ISD::FDIV, MVT::f64, Expand); + setOperationAction(ISD::FREM, MVT::f64, Expand); + setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand); + setOperationAction(ISD::FGETSIGN, MVT::f64, Expand); + setOperationAction(ISD::FNEG, MVT::f64, Expand); + setOperationAction(ISD::FABS, MVT::f64, Expand); + setOperationAction(ISD::FSQRT, MVT::f64, Expand); + setOperationAction(ISD::FSIN, MVT::f64, Expand); + setOperationAction(ISD::FCOS, MVT::f64, Expand); + setOperationAction(ISD::FPOWI, MVT::f64, Expand); + setOperationAction(ISD::FPOW, MVT::f64, Expand); + setOperationAction(ISD::FLOG, MVT::f64, Expand); + setOperationAction(ISD::FLOG2, MVT::f64, Expand); + setOperationAction(ISD::FLOG10, MVT::f64, Expand); + setOperationAction(ISD::FEXP, MVT::f64, Expand); + setOperationAction(ISD::FEXP2, MVT::f64, Expand); + setOperationAction(ISD::FCEIL, MVT::f64, Expand); + setOperationAction(ISD::FTRUNC, MVT::f64, Expand); + setOperationAction(ISD::FRINT, MVT::f64, Expand); + setOperationAction(ISD::FNEARBYINT, MVT::f64, Expand); + setOperationAction(ISD::FFLOOR, MVT::f64, Expand); + setOperationAction(ISD::FP_ROUND, MVT::f32, Custom); + setOperationAction(ISD::FP_EXTEND, MVT::f64, Custom); + } computeRegisterProperties(); @@ -3276,6 +3308,7 @@ ARMTargetLowering::getARMCmp(SDValue LHS, SDValue RHS, ISD::CondCode CC, SDValue ARMTargetLowering::getVFPCmp(SDValue LHS, SDValue RHS, SelectionDAG &DAG, SDLoc dl) const { + assert(!Subtarget->isFPOnlySP() || RHS.getValueType() != MVT::f64); SDValue Cmp; if (!isFloatingPointZero(RHS)) Cmp = DAG.getNode(ARMISD::CMPFP, dl, MVT::Glue, LHS, RHS); @@ -3391,9 +3424,8 @@ SDValue ARMTargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const { SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32); EVT VT = Op.getValueType(); - return DAG.getNode(ARMISD::CMOV, SDLoc(Op), VT, SelectTrue, SelectFalse, - ARMcc, CCR, OverflowCmp); - + return getCMOV(SDLoc(Op), VT, SelectTrue, SelectFalse, ARMcc, CCR, + OverflowCmp, DAG); } // Convert: @@ -3427,7 +3459,7 @@ SDValue ARMTargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const { SDValue CCR = Cond.getOperand(3); SDValue Cmp = duplicateCmp(Cond.getOperand(4), DAG); assert(True.getValueType() == VT); - return DAG.getNode(ARMISD::CMOV, dl, VT, True, False, ARMcc, CCR, Cmp); + return getCMOV(dl, VT, True, False, ARMcc, CCR, Cmp, DAG); } } } @@ -3497,6 +3529,32 @@ static void checkVSELConstraints(ISD::CondCode CC, ARMCC::CondCodes &CondCode, } } +SDValue ARMTargetLowering::getCMOV(SDLoc dl, EVT VT, SDValue FalseVal, + SDValue TrueVal, SDValue ARMcc, SDValue CCR, + SDValue Cmp, SelectionDAG &DAG) const { + if (Subtarget->isFPOnlySP() && VT == MVT::f64) { + FalseVal = DAG.getNode(ARMISD::VMOVRRD, dl, + DAG.getVTList(MVT::i32, MVT::i32), FalseVal); + TrueVal = DAG.getNode(ARMISD::VMOVRRD, dl, + DAG.getVTList(MVT::i32, MVT::i32), TrueVal); + + SDValue TrueLow = TrueVal.getValue(0); + SDValue TrueHigh = TrueVal.getValue(1); + SDValue FalseLow = FalseVal.getValue(0); + SDValue FalseHigh = FalseVal.getValue(1); + + SDValue Low = DAG.getNode(ARMISD::CMOV, dl, MVT::i32, FalseLow, TrueLow, + ARMcc, CCR, Cmp); + SDValue High = DAG.getNode(ARMISD::CMOV, dl, MVT::i32, FalseHigh, TrueHigh, + ARMcc, CCR, duplicateCmp(Cmp, DAG)); + + return DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Low, High); + } else { + return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR, + Cmp); + } +} + SDValue ARMTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const { EVT VT = Op.getValueType(); SDValue LHS = Op.getOperand(0); @@ -3506,6 +3564,18 @@ SDValue ARMTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const { SDValue FalseVal = Op.getOperand(3); SDLoc dl(Op); + if (Subtarget->isFPOnlySP() && LHS.getValueType() == MVT::f64) { + DAG.getTargetLoweringInfo().softenSetCCOperands(DAG, MVT::f64, LHS, RHS, CC, + dl); + + // If softenSetCCOperands only returned one value, we should compare it to + // zero. + if (!RHS.getNode()) { + RHS = DAG.getConstant(0, LHS.getValueType()); + CC = ISD::SETNE; + } + } + if (LHS.getValueType() == MVT::i32) { // Try to generate VSEL on ARMv8. // The VSEL instruction can't use all the usual ARM condition @@ -3530,8 +3600,7 @@ SDValue ARMTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const { SDValue ARMcc; SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32); SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl); - return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR, - Cmp); + return getCMOV(dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp, DAG); } ARMCC::CondCodes CondCode, CondCode2; @@ -3570,14 +3639,12 @@ SDValue ARMTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const { SDValue ARMcc = DAG.getConstant(CondCode, MVT::i32); SDValue Cmp = getVFPCmp(LHS, RHS, DAG, dl); SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32); - SDValue Result = DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, - ARMcc, CCR, Cmp); + SDValue Result = getCMOV(dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp, DAG); if (CondCode2 != ARMCC::AL) { SDValue ARMcc2 = DAG.getConstant(CondCode2, MVT::i32); // FIXME: Needs another CMP because flag can have but one use. SDValue Cmp2 = getVFPCmp(LHS, RHS, DAG, dl); - Result = DAG.getNode(ARMISD::CMOV, dl, VT, - Result, TrueVal, ARMcc2, CCR, Cmp2); + Result = getCMOV(dl, VT, Result, TrueVal, ARMcc2, CCR, Cmp2, DAG); } return Result; } @@ -3710,6 +3777,18 @@ SDValue ARMTargetLowering::LowerBR_CC(SDValue Op, SelectionDAG &DAG) const { SDValue Dest = Op.getOperand(4); SDLoc dl(Op); + if (Subtarget->isFPOnlySP() && LHS.getValueType() == MVT::f64) { + DAG.getTargetLoweringInfo().softenSetCCOperands(DAG, MVT::f64, LHS, RHS, CC, + dl); + + // If softenSetCCOperands only returned one value, we should compare it to + // zero. + if (!RHS.getNode()) { + RHS = DAG.getConstant(0, LHS.getValueType()); + CC = ISD::SETNE; + } + } + if (LHS.getValueType() == MVT::i32) { SDValue ARMcc; SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl); @@ -3802,11 +3881,23 @@ static SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) { return DAG.getNode(ISD::TRUNCATE, dl, VT, Op); } -static SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) { +SDValue ARMTargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) const { EVT VT = Op.getValueType(); if (VT.isVector()) return LowerVectorFP_TO_INT(Op, DAG); + if (Subtarget->isFPOnlySP() && Op.getOperand(0).getValueType() == MVT::f64) { + RTLIB::Libcall LC; + if (Op.getOpcode() == ISD::FP_TO_SINT) + LC = RTLIB::getFPTOSINT(Op.getOperand(0).getValueType(), + Op.getValueType()); + else + LC = RTLIB::getFPTOUINT(Op.getOperand(0).getValueType(), + Op.getValueType()); + return makeLibCall(DAG, LC, Op.getValueType(), &Op.getOperand(0), 1, + /*isSigned*/ false, SDLoc(Op)).first; + } + SDLoc dl(Op); unsigned Opc; @@ -3856,11 +3947,23 @@ static SDValue LowerVectorINT_TO_FP(SDValue Op, SelectionDAG &DAG) { return DAG.getNode(Opc, dl, VT, Op); } -static SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) { +SDValue ARMTargetLowering::LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const { EVT VT = Op.getValueType(); if (VT.isVector()) return LowerVectorINT_TO_FP(Op, DAG); + if (Subtarget->isFPOnlySP() && Op.getValueType() == MVT::f64) { + RTLIB::Libcall LC; + if (Op.getOpcode() == ISD::SINT_TO_FP) + LC = RTLIB::getSINTTOFP(Op.getOperand(0).getValueType(), + Op.getValueType()); + else + LC = RTLIB::getUINTTOFP(Op.getOperand(0).getValueType(), + Op.getValueType()); + return makeLibCall(DAG, LC, Op.getValueType(), &Op.getOperand(0), 1, + /*isSigned*/ false, SDLoc(Op)).first; + } + SDLoc dl(Op); unsigned Opc; @@ -4369,7 +4472,7 @@ static SDValue LowerVSETCC(SDValue Op, SelectionDAG &DAG) { ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get(); SDLoc dl(Op); - if (Op.getOperand(1).getValueType().isFloatingPoint()) { + if (Op1.getValueType().isFloatingPoint()) { switch (SetCCOpcode) { default: llvm_unreachable("Illegal FP comparison"); case ISD::SETUNE: @@ -4633,6 +4736,11 @@ SDValue ARMTargetLowering::LowerConstantFP(SDValue Op, SelectionDAG &DAG, bool IsDouble = Op.getValueType() == MVT::f64; ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Op); + // Use the default (constant pool) lowering for double constants when we have + // an SP-only FPU + if (IsDouble && Subtarget->isFPOnlySP()) + return SDValue(); + // Try splatting with a VMOV.f32... APFloat FPVal = CFP->getValueAPF(); int ImmVal = IsDouble ? ARM_AM::getFP64Imm(FPVal) : ARM_AM::getFP32Imm(FPVal); @@ -6336,6 +6444,8 @@ SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const { if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment()) return LowerDYNAMIC_STACKALLOC(Op, DAG); llvm_unreachable("Don't know how to custom lower this!"); + case ISD::FP_ROUND: return LowerFP_ROUND(Op, DAG); + case ISD::FP_EXTEND: return LowerFP_EXTEND(Op, DAG); } } @@ -8479,10 +8589,11 @@ static SDValue PerformBFICombine(SDNode *N, /// PerformVMOVRRDCombine - Target-specific dag combine xforms for /// ARMISD::VMOVRRD. static SDValue PerformVMOVRRDCombine(SDNode *N, - TargetLowering::DAGCombinerInfo &DCI) { + TargetLowering::DAGCombinerInfo &DCI, + const ARMSubtarget *Subtarget) { // vmovrrd(vmovdrr x, y) -> x,y SDValue InDouble = N->getOperand(0); - if (InDouble.getOpcode() == ARMISD::VMOVDRR) + if (InDouble.getOpcode() == ARMISD::VMOVDRR && !Subtarget->isFPOnlySP()) return DCI.CombineTo(N, InDouble.getOperand(0), InDouble.getOperand(1)); // vmovrrd(load f64) -> (load i32), (load i32) @@ -8695,7 +8806,8 @@ static bool hasNormalLoadOperand(SDNode *N) { /// PerformBUILD_VECTORCombine - Target-specific dag combine xforms for /// ISD::BUILD_VECTOR. static SDValue PerformBUILD_VECTORCombine(SDNode *N, - TargetLowering::DAGCombinerInfo &DCI){ + TargetLowering::DAGCombinerInfo &DCI, + const ARMSubtarget *Subtarget) { // build_vector(N=ARMISD::VMOVRRD(X), N:1) -> bit_convert(X): // VMOVRRD is introduced when legalizing i64 types. It forces the i64 value // into a pair of GPRs, which is fine when the value is used as a scalar, @@ -9710,10 +9822,10 @@ SDValue ARMTargetLowering::PerformDAGCombine(SDNode *N, case ISD::XOR: return PerformXORCombine(N, DCI, Subtarget); case ISD::AND: return PerformANDCombine(N, DCI, Subtarget); case ARMISD::BFI: return PerformBFICombine(N, DCI); - case ARMISD::VMOVRRD: return PerformVMOVRRDCombine(N, DCI); + case ARMISD::VMOVRRD: return PerformVMOVRRDCombine(N, DCI, Subtarget); case ARMISD::VMOVDRR: return PerformVMOVDRRCombine(N, DCI.DAG); case ISD::STORE: return PerformSTORECombine(N, DCI); - case ISD::BUILD_VECTOR: return PerformBUILD_VECTORCombine(N, DCI); + case ISD::BUILD_VECTOR: return PerformBUILD_VECTORCombine(N, DCI, Subtarget); case ISD::INSERT_VECTOR_ELT: return PerformInsertEltCombine(N, DCI); case ISD::VECTOR_SHUFFLE: return PerformVECTOR_SHUFFLECombine(N, DCI.DAG); case ARMISD::VDUPLANE: return PerformVDUPLANECombine(N, DCI); @@ -10703,6 +10815,31 @@ ARMTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const return DAG.getMergeValues(Ops, DL); } +SDValue ARMTargetLowering::LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const { + assert(Op.getValueType() == MVT::f64 && Subtarget->isFPOnlySP() && + "Unexpected type for custom-lowering FP_EXTEND"); + + RTLIB::Libcall LC; + LC = RTLIB::getFPEXT(Op.getOperand(0).getValueType(), Op.getValueType()); + + SDValue SrcVal = Op.getOperand(0); + return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1, + /*isSigned*/ false, SDLoc(Op)).first; +} + +SDValue ARMTargetLowering::LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const { + assert(Op.getOperand(0).getValueType() == MVT::f64 && + Subtarget->isFPOnlySP() && + "Unexpected type for custom-lowering FP_ROUND"); + + RTLIB::Libcall LC; + LC = RTLIB::getFPROUND(Op.getOperand(0).getValueType(), Op.getValueType()); + + SDValue SrcVal = Op.getOperand(0); + return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1, + /*isSigned*/ false, SDLoc(Op)).first; +} + bool ARMTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const { // The ARM target isn't yet aware of offsets. @@ -10730,7 +10867,7 @@ bool ARMTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const { return false; if (VT == MVT::f32) return ARM_AM::getFP32Imm(Imm) != -1; - if (VT == MVT::f64) + if (VT == MVT::f64 && !Subtarget->isFPOnlySP()) return ARM_AM::getFP64Imm(Imm) != -1; return false; } |
