#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsAMDGPU.h"
#include "llvm/IR/IntrinsicsX86.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/Type.h"
case Intrinsic::convert_from_fp16:
case Intrinsic::convert_to_fp16:
case Intrinsic::bitreverse:
+ case Intrinsic::amdgcn_fmul_legacy:
case Intrinsic::x86_sse_cvtss2si:
case Intrinsic::x86_sse_cvtss2si64:
case Intrinsic::x86_sse_cvttss2si:
return ConstantFP::get(Ty->getContext(), maximum(C1, C2));
}
+ if (IntrinsicID == Intrinsic::amdgcn_fmul_legacy) {
+ const APFloat &C1 = Op1->getValueAPF();
+ const APFloat &C2 = Op2->getValueAPF();
+ // The legacy behaviour is that multiplying zero by anything, even NaN
+ // or infinity, gives +0.0.
+ if (C1.isZero() || C2.isZero())
+ return ConstantFP::getNullValue(Ty);
+ return ConstantFP::get(Ty->getContext(), C1 * C2);
+ }
+
if (!TLI)
return nullptr;
--- /dev/null
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instsimplify -S | FileCheck %s
+
+declare float @llvm.amdgcn.fmul.legacy(float, float)
+
+define void @test(float* %p) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT: store volatile float 6.000000e+00, float* [[P:%.*]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT: ret void
+;
+ %a = call float @llvm.amdgcn.fmul.legacy(float +2.0, float +3.0)
+ store volatile float %a, float* %p
+ %b = call float @llvm.amdgcn.fmul.legacy(float +2.0, float +0.0)
+ store volatile float %b, float* %p
+ %c = call float @llvm.amdgcn.fmul.legacy(float +2.0, float -0.0)
+ store volatile float %c, float* %p
+ %d = call float @llvm.amdgcn.fmul.legacy(float +0.0, float +0.0)
+ store volatile float %d, float* %p
+ %e = call float @llvm.amdgcn.fmul.legacy(float +0.0, float -0.0)
+ store volatile float %e, float* %p
+ %f = call float @llvm.amdgcn.fmul.legacy(float -0.0, float +0.0)
+ store volatile float %f, float* %p
+ %g = call float @llvm.amdgcn.fmul.legacy(float -0.0, float -0.0)
+ store volatile float %g, float* %p
+ %h = call float @llvm.amdgcn.fmul.legacy(float +0.0, float 0x7ff0000000000000) ; +inf
+ store volatile float %h, float* %p
+ %i = call float @llvm.amdgcn.fmul.legacy(float 0xfff0000000000000, float +0.0) ; -inf
+ store volatile float %i, float* %p
+ %j = call float @llvm.amdgcn.fmul.legacy(float 0x7ff0001000000000, float -0.0) ; +nan
+ store volatile float %j, float* %p
+ %k = call float @llvm.amdgcn.fmul.legacy(float -0.0, float 0xfff0000100000000) ; -nan
+ store volatile float %k, float* %p
+ ret void
+}