diff options
Diffstat (limited to 'llvm/lib/Target/X86')
| -rw-r--r-- | llvm/lib/Target/X86/X86.td | 3 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86FastISel.cpp | 2 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 46 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.h | 1 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86Subtarget.cpp | 1 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86Subtarget.h | 4 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86TargetMachine.cpp | 16 |
7 files changed, 44 insertions, 29 deletions
diff --git a/llvm/lib/Target/X86/X86.td b/llvm/lib/Target/X86/X86.td index d13f1551d20..c70e2e95463 100644 --- a/llvm/lib/Target/X86/X86.td +++ b/llvm/lib/Target/X86/X86.td @@ -192,6 +192,9 @@ def FeatureUseSqrtEst : SubtargetFeature<"use-sqrt-est", "UseSqrtEst", "true", "Use RSQRT* to optimize square root calculations">; def FeatureUseRecipEst : SubtargetFeature<"use-recip-est", "UseReciprocalEst", "true", "Use RCP* to optimize division calculations">; +def FeatureSoftFloat + : SubtargetFeature<"soft-float", "UseSoftFloat", "true", + "Use software floating point features.">; //===----------------------------------------------------------------------===// // X86 processors supported. diff --git a/llvm/lib/Target/X86/X86FastISel.cpp b/llvm/lib/Target/X86/X86FastISel.cpp index 3bf4fb824fd..9af0aebea23 100644 --- a/llvm/lib/Target/X86/X86FastISel.cpp +++ b/llvm/lib/Target/X86/X86FastISel.cpp @@ -2254,7 +2254,7 @@ bool X86FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) { default: return false; case Intrinsic::convert_from_fp16: case Intrinsic::convert_to_fp16: { - if (TM.Options.UseSoftFloat || !Subtarget->hasF16C()) + if (Subtarget->useSoftFloat() || !Subtarget->hasF16C()) return false; const Value *Op = II->getArgOperand(0); diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 68d777be4ce..f36e4a5e369 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -183,7 +183,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, if (Subtarget->is64Bit()) { setOperationAction(ISD::UINT_TO_FP , MVT::i32 , Promote); setOperationAction(ISD::UINT_TO_FP , MVT::i64 , Custom); - } else if (!TM.Options.UseSoftFloat) { + } else if (!Subtarget->useSoftFloat()) { // We have an algorithm for SSE2->double, and we turn this into a // 64-bit FILD followed by conditional FADD for other targets. setOperationAction(ISD::UINT_TO_FP , MVT::i64 , Custom); @@ -197,7 +197,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, setOperationAction(ISD::SINT_TO_FP , MVT::i1 , Promote); setOperationAction(ISD::SINT_TO_FP , MVT::i8 , Promote); - if (!TM.Options.UseSoftFloat) { + if (!Subtarget->useSoftFloat()) { // SSE has no i16 to fp conversion, only i32 if (X86ScalarSSEf32) { setOperationAction(ISD::SINT_TO_FP , MVT::i16 , Promote); @@ -240,7 +240,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, if (Subtarget->is64Bit()) { setOperationAction(ISD::FP_TO_UINT , MVT::i64 , Expand); setOperationAction(ISD::FP_TO_UINT , MVT::i32 , Promote); - } else if (!TM.Options.UseSoftFloat) { + } else if (!Subtarget->useSoftFloat()) { // Since AVX is a superset of SSE3, only check for SSE here. if (Subtarget->hasSSE1() && !Subtarget->hasSSE3()) // Expand FP_TO_UINT into a select. @@ -368,7 +368,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, // Special handling for half-precision floating point conversions. // If we don't have F16C support, then lower half float conversions // into library calls. - if (TM.Options.UseSoftFloat || !Subtarget->hasF16C()) { + if (Subtarget->useSoftFloat() || !Subtarget->hasF16C()) { setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand); setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand); } @@ -517,7 +517,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, setOperationAction(ISD::GC_TRANSITION_START, MVT::Other, Custom); setOperationAction(ISD::GC_TRANSITION_END, MVT::Other, Custom); - if (!TM.Options.UseSoftFloat && X86ScalarSSEf64) { + if (!Subtarget->useSoftFloat() && X86ScalarSSEf64) { // f32 and f64 use SSE. // Set up the FP register classes. addRegisterClass(MVT::f32, &X86::FR32RegClass); @@ -551,7 +551,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, // cases we handle. addLegalFPImmediate(APFloat(+0.0)); // xorpd addLegalFPImmediate(APFloat(+0.0f)); // xorps - } else if (!TM.Options.UseSoftFloat && X86ScalarSSEf32) { + } else if (!Subtarget->useSoftFloat() && X86ScalarSSEf32) { // Use SSE for f32, x87 for f64. // Set up the FP register classes. addRegisterClass(MVT::f32, &X86::FR32RegClass); @@ -586,7 +586,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, setOperationAction(ISD::FCOS , MVT::f64, Expand); setOperationAction(ISD::FSINCOS, MVT::f64, Expand); } - } else if (!TM.Options.UseSoftFloat) { + } else if (!Subtarget->useSoftFloat()) { // f32 and f64 in x87. // Set up the FP register classes. addRegisterClass(MVT::f64, &X86::RFP64RegClass); @@ -620,7 +620,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, setOperationAction(ISD::FMA, MVT::f32, Expand); // Long double always uses X87. - if (!TM.Options.UseSoftFloat) { + if (!Subtarget->useSoftFloat()) { addRegisterClass(MVT::f80, &X86::RFP80RegClass); setOperationAction(ISD::UNDEF, MVT::f80, Expand); setOperationAction(ISD::FCOPYSIGN, MVT::f80, Expand); @@ -760,7 +760,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, // FIXME: In order to prevent SSE instructions being expanded to MMX ones // with -msoft-float, disable use of MMX as well. - if (!TM.Options.UseSoftFloat && Subtarget->hasMMX()) { + if (!Subtarget->useSoftFloat() && Subtarget->hasMMX()) { addRegisterClass(MVT::x86mmx, &X86::VR64RegClass); // No operations on x86mmx supported, everything uses intrinsics. } @@ -778,7 +778,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, } setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v1i64, Expand); - if (!TM.Options.UseSoftFloat && Subtarget->hasSSE1()) { + if (!Subtarget->useSoftFloat() && Subtarget->hasSSE1()) { addRegisterClass(MVT::v4f32, &X86::VR128RegClass); setOperationAction(ISD::FADD, MVT::v4f32, Legal); @@ -797,7 +797,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, setOperationAction(ISD::UINT_TO_FP, MVT::v4i32, Custom); } - if (!TM.Options.UseSoftFloat && Subtarget->hasSSE2()) { + if (!Subtarget->useSoftFloat() && Subtarget->hasSSE2()) { addRegisterClass(MVT::v2f64, &X86::VR128RegClass); // FIXME: Unfortunately, -soft-float and -no-implicit-float mean XMM @@ -942,7 +942,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, setOperationAction(ISD::BITCAST, MVT::v8i8, Custom); } - if (!TM.Options.UseSoftFloat && Subtarget->hasSSE41()) { + if (!Subtarget->useSoftFloat() && Subtarget->hasSSE41()) { for (MVT RoundedTy : {MVT::f32, MVT::f64, MVT::v4f32, MVT::v2f64}) { setOperationAction(ISD::FFLOOR, RoundedTy, Legal); setOperationAction(ISD::FCEIL, RoundedTy, Legal); @@ -1024,7 +1024,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, setOperationAction(ISD::SRA, MVT::v4i32, Custom); } - if (!TM.Options.UseSoftFloat && Subtarget->hasFp256()) { + if (!Subtarget->useSoftFloat() && Subtarget->hasFp256()) { addRegisterClass(MVT::v32i8, &X86::VR256RegClass); addRegisterClass(MVT::v16i16, &X86::VR256RegClass); addRegisterClass(MVT::v8i32, &X86::VR256RegClass); @@ -1244,7 +1244,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, } } - if (!TM.Options.UseSoftFloat && Subtarget->hasAVX512()) { + if (!Subtarget->useSoftFloat() && Subtarget->hasAVX512()) { addRegisterClass(MVT::v16i32, &X86::VR512RegClass); addRegisterClass(MVT::v16f32, &X86::VR512RegClass); addRegisterClass(MVT::v8i64, &X86::VR512RegClass); @@ -1447,7 +1447,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, } }// has AVX-512 - if (!TM.Options.UseSoftFloat && Subtarget->hasBWI()) { + if (!Subtarget->useSoftFloat() && Subtarget->hasBWI()) { addRegisterClass(MVT::v32i16, &X86::VR512RegClass); addRegisterClass(MVT::v64i8, &X86::VR512RegClass); @@ -1484,7 +1484,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM, } } - if (!TM.Options.UseSoftFloat && Subtarget->hasVLX()) { + if (!Subtarget->useSoftFloat() && Subtarget->hasVLX()) { addRegisterClass(MVT::v4i1, &X86::VK4RegClass); addRegisterClass(MVT::v2i1, &X86::VK2RegClass); @@ -1789,6 +1789,10 @@ unsigned X86TargetLowering::getJumpTableEncoding() const { return TargetLowering::getJumpTableEncoding(); } +bool X86TargetLowering::useSoftFloat() const { + return Subtarget->useSoftFloat(); +} + const MCExpr * X86TargetLowering::LowerCustomJumpTableEntry(const MachineJumpTableInfo *MJTI, const MachineBasicBlock *MBB, @@ -2294,7 +2298,7 @@ static ArrayRef<MCPhysReg> get64BitArgumentXMMs(MachineFunction &MF, const Function *Fn = MF.getFunction(); bool NoImplicitFloatOps = Fn->hasFnAttribute(Attribute::NoImplicitFloat); - bool isSoftFloat = MF.getTarget().Options.UseSoftFloat; + bool isSoftFloat = Subtarget->useSoftFloat(); assert(!(isSoftFloat && NoImplicitFloatOps) && "SSE register cannot be used when SSE is disabled!"); if (isSoftFloat || NoImplicitFloatOps || !Subtarget->hasSSE1()) @@ -2468,7 +2472,7 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain, bool IsWinEHParent = WinEHParent && WinEHParent == Fn; // Figure out if XMM registers are in use. - assert(!(MF.getTarget().Options.UseSoftFloat && + assert(!(Subtarget->useSoftFloat() && Fn->hasFnAttribute(Attribute::NoImplicitFloat)) && "SSE register cannot be used when SSE is disabled!"); @@ -14600,7 +14604,7 @@ SDValue X86TargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const { if (ArgMode == 2) { // Sanity Check: Make sure using fp_offset makes sense. - assert(!DAG.getTarget().Options.UseSoftFloat && + assert(!Subtarget->useSoftFloat() && !(DAG.getMachineFunction().getFunction()->hasFnAttribute( Attribute::NoImplicitFloat)) && Subtarget->hasSSE1()); @@ -23203,8 +23207,8 @@ static SDValue PerformSTORECombine(SDNode *N, SelectionDAG &DAG, const Function *F = DAG.getMachineFunction().getFunction(); bool NoImplicitFloatOps = F->hasFnAttribute(Attribute::NoImplicitFloat); - bool F64IsLegal = !DAG.getTarget().Options.UseSoftFloat && !NoImplicitFloatOps - && Subtarget->hasSSE2(); + bool F64IsLegal = + !Subtarget->useSoftFloat() && !NoImplicitFloatOps && Subtarget->hasSSE2(); if ((VT.isVector() || (VT == MVT::i64 && F64IsLegal && !Subtarget->is64Bit())) && isa<LoadSDNode>(St->getValue()) && diff --git a/llvm/lib/Target/X86/X86ISelLowering.h b/llvm/lib/Target/X86/X86ISelLowering.h index 90b49328f22..1c5f73ab379 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.h +++ b/llvm/lib/Target/X86/X86ISelLowering.h @@ -572,6 +572,7 @@ namespace llvm { const X86Subtarget &STI); unsigned getJumpTableEncoding() const override; + bool useSoftFloat() const override; MVT getScalarShiftAmountTy(EVT LHSTy) const override { return MVT::i8; } diff --git a/llvm/lib/Target/X86/X86Subtarget.cpp b/llvm/lib/Target/X86/X86Subtarget.cpp index de30c75ac86..1cdab14e034 100644 --- a/llvm/lib/Target/X86/X86Subtarget.cpp +++ b/llvm/lib/Target/X86/X86Subtarget.cpp @@ -278,6 +278,7 @@ void X86Subtarget::initializeEnvironment() { stackAlignment = 4; // FIXME: this is a known good value for Yonah. How about others? MaxInlineSizeThreshold = 128; + UseSoftFloat = false; } X86Subtarget &X86Subtarget::initializeSubtargetDependencies(StringRef CPU, diff --git a/llvm/lib/Target/X86/X86Subtarget.h b/llvm/lib/Target/X86/X86Subtarget.h index 4c31f789546..455dd7744d7 100644 --- a/llvm/lib/Target/X86/X86Subtarget.h +++ b/llvm/lib/Target/X86/X86Subtarget.h @@ -218,6 +218,9 @@ protected: /// Processor has AVX-512 Vector Length eXtenstions bool HasVLX; + /// Use software floating point for code generation. + bool UseSoftFloat; + /// The minimum alignment known to hold of the stack frame on /// entry to the function and which must be maintained by every function. unsigned stackAlignment; @@ -385,6 +388,7 @@ public: bool isAtom() const { return X86ProcFamily == IntelAtom; } bool isSLM() const { return X86ProcFamily == IntelSLM; } + bool useSoftFloat() const { return UseSoftFloat; } const Triple &getTargetTriple() const { return TargetTriple; } diff --git a/llvm/lib/Target/X86/X86TargetMachine.cpp b/llvm/lib/Target/X86/X86TargetMachine.cpp index 9484f678808..d5e7eac37c4 100644 --- a/llvm/lib/Target/X86/X86TargetMachine.cpp +++ b/llvm/lib/Target/X86/X86TargetMachine.cpp @@ -131,13 +131,15 @@ X86TargetMachine::getSubtargetImpl(const Function &F) const { // function before we can generate a subtarget. We also need to use // it as a key for the subtarget since that can be the only difference // between two functions. - Attribute SFAttr = F.getFnAttribute("use-soft-float"); - bool SoftFloat = !SFAttr.hasAttribute(Attribute::None) - ? SFAttr.getValueAsString() == "true" - : Options.UseSoftFloat; - - auto &I = SubtargetMap[CPU + FS + (SoftFloat ? "use-soft-float=true" - : "use-soft-float=false")]; + bool SoftFloat = + F.hasFnAttribute("use-soft-float") && + F.getFnAttribute("use-soft-float").getValueAsString() == "true"; + // If the soft float attribute is set on the function turn on the soft float + // subtarget feature. + if (SoftFloat) + FS += FS.empty() ? "+soft-float" : ",+soft-float"; + + auto &I = SubtargetMap[CPU + FS]; if (!I) { // This needs to be done before we create a new subtarget since any // creation will depend on the TM and the code generation flags on the |

