[X86] Add support for STRICT_FP_ROUND/STRICT_FP_EXTEND from/to fp128 to/from f32/f64/f80 in 64-bit mode.

[craig.topper]

Summary

These need to emit a libcall like we do for the non-strict version.

32-bit mode needs to SoftenFloat support to be implemented for strict FP nodes.

[uweigand]

FWIW, this looks good to me ...

[pengfei]

llvm/test/CodeGen/X86/fp128-cast-strict.ll

3

We can add option -disable-strictnode-mutation now.

[craig.topper]

Rebase after changing D70503

[craig.topper]

Add -disable-strictnode-mutation to the test run lines

[pengfei]

LGTM.

[pengfei]

LGTM.

llvm/lib/Target/X86/X86ISelLowering.cpp

if (!Subtarget.useSoftFloat() && Subtarget.is64Bit() && Subtarget.hasSSE1()) {
     setOperationAction(ISD::FCOS,         MVT::f128, LibCall);
     setOperationAction(ISD::STRICT_FCOS,  MVT::f128, LibCall);
     setOperationAction(ISD::FSINCOS,      MVT::f128, LibCall);
     // No STRICT_FSINCOS
     setOperationAction(ISD::FSQRT,        MVT::f128, LibCall);
     setOperationAction(ISD::STRICT_FSQRT, MVT::f128, LibCall);
 
     setOperationAction(ISD::FP_EXTEND, MVT::f128, Custom);
+    setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f128, Custom);
     // We need to custom handle any FP_ROUND with an f128 input, but
     // LegalizeDAG uses the result type to know when to run a custom handler.
     // So we have to list all legal floating point result types here.
     if (isTypeLegal(MVT::f32)) {
       setOperationAction(ISD::FP_ROUND, MVT::f32, Custom);
       setOperationAction(ISD::STRICT_FP_ROUND, MVT::f32, Custom);
     }
     if (isTypeLegal(MVT::f64)) {
if (IsStrict)
       return DAG.getMergeValues({V, Chain}, dl);
     return V;
   }
 
   llvm_unreachable("Expected FP_TO_INTHelper to handle all remaining cases.");
 }
 
 SDValue X86TargetLowering::LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const {
+  bool IsStrict = Op->isStrictFPOpcode();
+
   SDLoc DL(Op);
   MVT VT = Op.getSimpleValueType();
-  SDValue In = Op.getOperand(0);
+  SDValue In = Op.getOperand(IsStrict ? 1 : 0);
   MVT SVT = In.getSimpleValueType();
 
   if (VT == MVT::f128) {
     RTLIB::Libcall LC = RTLIB::getFPEXT(SVT, VT);
     return LowerF128Call(Op, DAG, LC);
   }
 
   assert(SVT == MVT::v2f32 && "Only customize MVT::v2f32 type legalization!");
+  // FIXME: Strict fp.
+  assert(Op.getOpcode() != ISD::STRICT_FP_EXTEND &&
+         "Strict FP not supported yet!");
 
   return DAG.getNode(X86ISD::VFPEXT, DL, VT,
                      DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v4f32,
                                  In, DAG.getUNDEF(SVT)));
 }
 
 SDValue X86TargetLowering::LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const {
+  bool IsStrict = Op->isStrictFPOpcode();
+
   MVT VT = Op.getSimpleValueType();
-  SDValue In = Op.getOperand(0);
+  SDValue In = Op.getOperand(IsStrict ? 1 : 0);
   MVT SVT = In.getSimpleValueType();
 
   // It's legal except when f128 is involved
   if (SVT != MVT::f128)
     return Op;
 
   RTLIB::Libcall LC = RTLIB::getFPROUND(SVT, VT);
 
   // FP_ROUND node has a second operand indicating whether it is known to be
   // precise. That doesn't take part in the LibCall so we can't directly use
   // LowerF128Call.
+
+  SDLoc dl(Op);
+  SDValue Chain = IsStrict ? Op.getOperand(0) : SDValue();
   MakeLibCallOptions CallOptions;
-  return makeLibCall(DAG, LC, VT, In, CallOptions, SDLoc(Op)).first;
-}
+  std::pair<SDValue, SDValue> Tmp = makeLibCall(DAG, LC, VT, In, CallOptions,
+                                                dl, Chain);
 
-// FIXME: This is a hack to allow FP_ROUND to be marked Custom without breaking
-// the default expansion of STRICT_FP_ROUND.
-static SDValue LowerSTRICT_FP_ROUND(SDValue Op, SelectionDAG &DAG) {
-  // FIXME: Need to form a libcall with an input chain for f128.
-  assert(Op.getOperand(0).getValueType() != MVT::f128 &&
-         "Don't know how to handle f128 yet!");
-  return Op;
+  if (IsStrict)
+    return DAG.getMergeValues({ Tmp.first, Tmp.second }, dl);
+
+  return Tmp.first;
 }
 
 /// Depending on uarch and/or optimizing for size, we might prefer to use a
 /// vector operation in place of the typical scalar operation.
 static SDValue lowerAddSubToHorizontalOp(SDValue Op, SelectionDAG &DAG,
                                          const X86Subtarget &Subtarget) {
   // If both operands have other uses, this is probably not profitable.
   SDValue LHS = Op.getOperand(0);
SDValue X86TargetLowering::LowerGC_TRANSITION_END(SDValue Op,
   SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue);
   SDValue NOOP(DAG.getMachineNode(X86::NOOP, SDLoc(Op), VTs, Ops), 0);
 
   return NOOP;
 }
 
 SDValue X86TargetLowering::LowerF128Call(SDValue Op, SelectionDAG &DAG,
                                          RTLIB::Libcall Call) const {
-  SmallVector<SDValue, 2> Ops(Op->op_begin(), Op->op_end());
+
+  bool IsStrict = Op->isStrictFPOpcode();
+  unsigned Offset = IsStrict ? 1 : 0;
+  SmallVector<SDValue, 2> Ops(Op->op_begin() + Offset, Op->op_end());
+
+  SDLoc dl(Op);
+  SDValue Chain = IsStrict ? Op.getOperand(0) : SDValue();
   MakeLibCallOptions CallOptions;
-  return makeLibCall(DAG, Call, MVT::f128, Ops, CallOptions, SDLoc(Op)).first;
+  std::pair<SDValue, SDValue> Tmp = makeLibCall(DAG, Call, MVT::f128, Ops,
+                                                CallOptions, dl, Chain);
+
+  if (IsStrict)
+    return DAG.getMergeValues({ Tmp.first, Tmp.second }, dl);
+
+  return Tmp.first;
 }
 
 /// Provide custom lowering hooks for some operations.
 SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   switch (Op.getOpcode()) {
   default: llvm_unreachable("Should not custom lower this!");
   case ISD::ATOMIC_FENCE:       return LowerATOMIC_FENCE(Op, Subtarget, DAG);
   case ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS:
SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::ANY_EXTEND:         return LowerANY_EXTEND(Op, Subtarget, DAG);
   case ISD::ZERO_EXTEND_VECTOR_INREG:
   case ISD::SIGN_EXTEND_VECTOR_INREG:
     return LowerEXTEND_VECTOR_INREG(Op, Subtarget, DAG);
   case ISD::FP_TO_SINT:
   case ISD::STRICT_FP_TO_SINT:
   case ISD::FP_TO_UINT:
   case ISD::STRICT_FP_TO_UINT:  return LowerFP_TO_INT(Op, DAG);
-  case ISD::FP_EXTEND:          return LowerFP_EXTEND(Op, DAG);
-  case ISD::FP_ROUND:           return LowerFP_ROUND(Op, DAG);
-  case ISD::STRICT_FP_ROUND:    return LowerSTRICT_FP_ROUND(Op, DAG);
+  case ISD::FP_EXTEND:
+  case ISD::STRICT_FP_EXTEND:   return LowerFP_EXTEND(Op, DAG);
+  case ISD::FP_ROUND:
+  case ISD::STRICT_FP_ROUND:    return LowerFP_ROUND(Op, DAG);
   case ISD::LOAD:               return LowerLoad(Op, Subtarget, DAG);
   case ISD::STORE:              return LowerStore(Op, Subtarget, DAG);
   case ISD::FADD:
   case ISD::FSUB:               return lowerFaddFsub(Op, DAG);
   case ISD::FABS:
   case ISD::FNEG:               return LowerFABSorFNEG(Op, DAG);
   case ISD::FCOPYSIGN:          return LowerFCOPYSIGN(Op, DAG);
   case ISD::FGETSIGN:           return LowerFGETSIGN(Op, DAG);

llvm/test/CodeGen/X86/fp128-cast-strict.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -O2 -mtriple=x86_64-linux-android -mattr=+sse | FileCheck %s --check-prefixes=X64,X64-SSE
+; RUN: llc < %s -O2 -mtriple=x86_64-linux-gnu -mattr=+sse | FileCheck %s --check-prefixes=X64,X64-SSE
+; RUN: llc < %s -O2 -mtriple=x86_64-linux-android -mattr=+avx | FileCheck %s --check-prefixes=X64,X64-AVX
+; RUN: llc < %s -O2 -mtriple=x86_64-linux-gnu -mattr=+avx | FileCheck %s --check-prefixes=X64,X64-AVX
+; RUN: llc < %s -O2 -mtriple=x86_64-linux-android -mattr=+avx512f | FileCheck %s --check-prefixes=X64,X64-AVX
+; RUN: llc < %s -O2 -mtriple=x86_64-linux-gnu -mattr=+avx512f | FileCheck %s --check-prefixes=X64,X64-AVX
+
+; Check soft floating point conversion function calls.
+
+@vf32 = common global float 0.000000e+00, align 4
+@vf64 = common global double 0.000000e+00, align 8
+@vf80 = common global x86_fp80 0xK00000000000000000000, align 8
+@vf128 = common global fp128 0xL00000000000000000000000000000000, align 16
+
+define void @TestFPExtF32_F128() nounwind strictfp {
+; X64-SSE-LABEL: TestFPExtF32_F128:
+; X64-SSE:       # %bb.0: # %entry
+; X64-SSE-NEXT:    pushq %rax
+; X64-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X64-SSE-NEXT:    callq __extendsftf2
+; X64-SSE-NEXT:    movaps %xmm0, {{.*}}(%rip)
+; X64-SSE-NEXT:    popq %rax
+; X64-SSE-NEXT:    retq
+;
+; X64-AVX-LABEL: TestFPExtF32_F128:
+; X64-AVX:       # %bb.0: # %entry
+; X64-AVX-NEXT:    pushq %rax
+; X64-AVX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X64-AVX-NEXT:    callq __extendsftf2
+; X64-AVX-NEXT:    vmovaps %xmm0, {{.*}}(%rip)
+; X64-AVX-NEXT:    popq %rax
+; X64-AVX-NEXT:    retq
+entry:
+  %0 = load float, float* @vf32, align 4
+  %conv = call fp128 @llvm.experimental.constrained.fpext.f128.f32(float %0, metadata !"fpexcept.strict") #0
+  store fp128 %conv, fp128* @vf128, align 16
+  ret void
+}
+
+define void @TestFPExtF64_F128() nounwind strictfp {
+; X64-SSE-LABEL: TestFPExtF64_F128:
+; X64-SSE:       # %bb.0: # %entry
+; X64-SSE-NEXT:    pushq %rax
+; X64-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; X64-SSE-NEXT:    callq __extenddftf2
+; X64-SSE-NEXT:    movaps %xmm0, {{.*}}(%rip)
+; X64-SSE-NEXT:    popq %rax
+; X64-SSE-NEXT:    retq
+;
+; X64-AVX-LABEL: TestFPExtF64_F128:
+; X64-AVX:       # %bb.0: # %entry
+; X64-AVX-NEXT:    pushq %rax
+; X64-AVX-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X64-AVX-NEXT:    callq __extenddftf2
+; X64-AVX-NEXT:    vmovaps %xmm0, {{.*}}(%rip)
+; X64-AVX-NEXT:    popq %rax
+; X64-AVX-NEXT:    retq
+entry:
+  %0 = load double, double* @vf64, align 8
+  %conv = call fp128 @llvm.experimental.constrained.fpext.f128.f64(double %0, metadata !"fpexcept.strict") #0
+  store fp128 %conv, fp128* @vf128, align 16
+  ret void
+}
+
+define void @TestFPExtF80_F128() nounwind strictfp {
+; X64-SSE-LABEL: TestFPExtF80_F128:
+; X64-SSE:       # %bb.0: # %entry
+; X64-SSE-NEXT:    subq $24, %rsp
+; X64-SSE-NEXT:    fldt {{.*}}(%rip)
+; X64-SSE-NEXT:    fstpt (%rsp)
+; X64-SSE-NEXT:    callq __extendxftf2
+; X64-SSE-NEXT:    movaps %xmm0, {{.*}}(%rip)
+; X64-SSE-NEXT:    addq $24, %rsp
+; X64-SSE-NEXT:    retq
+;
+; X64-AVX-LABEL: TestFPExtF80_F128:
+; X64-AVX:       # %bb.0: # %entry
+; X64-AVX-NEXT:    subq $24, %rsp
+; X64-AVX-NEXT:    fldt {{.*}}(%rip)
+; X64-AVX-NEXT:    fstpt (%rsp)
+; X64-AVX-NEXT:    callq __extendxftf2
+; X64-AVX-NEXT:    vmovaps %xmm0, {{.*}}(%rip)
+; X64-AVX-NEXT:    addq $24, %rsp
+; X64-AVX-NEXT:    retq
+entry:
+  %0 = load x86_fp80, x86_fp80* @vf80, align 8
+  %conv = call fp128 @llvm.experimental.constrained.fpext.f128.f80(x86_fp80 %0, metadata !"fpexcept.strict") #0
+  store fp128 %conv, fp128* @vf128, align 16
+  ret void
+}
+
+define void @TestFPTruncF128_F32() nounwind strictfp {
+; X64-SSE-LABEL: TestFPTruncF128_F32:
+; X64-SSE:       # %bb.0: # %entry
+; X64-SSE-NEXT:    pushq %rax
+; X64-SSE-NEXT:    movaps {{.*}}(%rip), %xmm0
+; X64-SSE-NEXT:    callq __trunctfsf2
+; X64-SSE-NEXT:    movss %xmm0, {{.*}}(%rip)
+; X64-SSE-NEXT:    popq %rax
+; X64-SSE-NEXT:    retq
+;
+; X64-AVX-LABEL: TestFPTruncF128_F32:
+; X64-AVX:       # %bb.0: # %entry
+; X64-AVX-NEXT:    pushq %rax
+; X64-AVX-NEXT:    vmovaps {{.*}}(%rip), %xmm0
+; X64-AVX-NEXT:    callq __trunctfsf2
+; X64-AVX-NEXT:    vmovss %xmm0, {{.*}}(%rip)
+; X64-AVX-NEXT:    popq %rax
+; X64-AVX-NEXT:    retq
+entry:
+  %0 = load fp128, fp128* @vf128, align 16
+  %conv = call float @llvm.experimental.constrained.fptrunc.f32.f128(fp128 %0, metadata !"round.dynamic", metadata !"fpexcept.strict") #0
+  store float %conv, float* @vf32, align 4
+  ret void
+}
+
+define void @TestFPTruncF128_F64() nounwind strictfp {
+; X64-SSE-LABEL: TestFPTruncF128_F64:
+; X64-SSE:       # %bb.0: # %entry
+; X64-SSE-NEXT:    pushq %rax
+; X64-SSE-NEXT:    movaps {{.*}}(%rip), %xmm0
+; X64-SSE-NEXT:    callq __trunctfdf2
+; X64-SSE-NEXT:    movsd %xmm0, {{.*}}(%rip)
+; X64-SSE-NEXT:    popq %rax
+; X64-SSE-NEXT:    retq
+;
+; X64-AVX-LABEL: TestFPTruncF128_F64:
+; X64-AVX:       # %bb.0: # %entry
+; X64-AVX-NEXT:    pushq %rax
+; X64-AVX-NEXT:    vmovaps {{.*}}(%rip), %xmm0
+; X64-AVX-NEXT:    callq __trunctfdf2
+; X64-AVX-NEXT:    vmovsd %xmm0, {{.*}}(%rip)
+; X64-AVX-NEXT:    popq %rax
+; X64-AVX-NEXT:    retq
+entry:
+  %0 = load fp128, fp128* @vf128, align 16
+  %conv = call double @llvm.experimental.constrained.fptrunc.f64.f128(fp128 %0, metadata !"round.dynamic", metadata !"fpexcept.strict") #0
+  store double %conv, double* @vf64, align 8
+  ret void
+}
+
+define void @TestFPTruncF128_F80() nounwind strictfp {
+; X64-SSE-LABEL: TestFPTruncF128_F80:
+; X64-SSE:       # %bb.0: # %entry
+; X64-SSE-NEXT:    pushq %rax
+; X64-SSE-NEXT:    movaps {{.*}}(%rip), %xmm0
+; X64-SSE-NEXT:    callq __trunctfxf2
+; X64-SSE-NEXT:    fstpt {{.*}}(%rip)
+; X64-SSE-NEXT:    popq %rax
+; X64-SSE-NEXT:    retq
+;
+; X64-AVX-LABEL: TestFPTruncF128_F80:
+; X64-AVX:       # %bb.0: # %entry
+; X64-AVX-NEXT:    pushq %rax
+; X64-AVX-NEXT:    vmovaps {{.*}}(%rip), %xmm0
+; X64-AVX-NEXT:    callq __trunctfxf2
+; X64-AVX-NEXT:    fstpt {{.*}}(%rip)
+; X64-AVX-NEXT:    popq %rax
+; X64-AVX-NEXT:    retq
+entry:
+  %0 = load fp128, fp128* @vf128, align 16
+  %conv = call x86_fp80 @llvm.experimental.constrained.fptrunc.f80.f128(fp128 %0, metadata !"round.dynamic", metadata !"fpexcept.strict") #0
+  store x86_fp80 %conv, x86_fp80* @vf80, align 8
+  ret void
+}
+
+attributes #0 = { strictfp }
+
+declare float @llvm.experimental.constrained.fptrunc.f32.f128(fp128, metadata, metadata)
+declare double @llvm.experimental.constrained.fptrunc.f64.f128(fp128, metadata, metadata)
+declare x86_fp80 @llvm.experimental.constrained.fptrunc.f80.f128(fp128, metadata, metadata)
+declare fp128 @llvm.experimental.constrained.fpext.f128.f32(float, metadata)
+declare fp128 @llvm.experimental.constrained.fpext.f128.f64(double, metadata)
+declare fp128 @llvm.experimental.constrained.fpext.f128.f80(x86_fp80, metadata)