[AMDGPU][ConstantFolding] Fold llvm.amdgcn.fmul.legacy intrinsic

[arsenm]

llvm/test/Transforms/ConstProp/amdgpu.ll

2 ↗

(On Diff #245402)

You can test this just by running -instsimplify? This also should go in an AMDGPU test specific directory (although technically you don't need that right now)

[foad]

Move test.

[foad]

Add lit.local.cfg.

[arsenm]

LGTM. I would name the test file fmul_legacy.ll to be clearer

llvm/lib/Analysis/ConstantFolding.cpp

 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalValue.h"
 #include "llvm/IR/GlobalVariable.h"
 #include "llvm/IR/InstrTypes.h"
 #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"
 #include "llvm/IR/Value.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/KnownBits.h"
 #include "llvm/Support/MathExtras.h"
bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
   case Intrinsic::uadd_sat:
   case Intrinsic::ssub_sat:
   case Intrinsic::usub_sat:
   case Intrinsic::smul_fix:
   case Intrinsic::smul_fix_sat:
   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:
   case Intrinsic::x86_sse_cvttss2si64:
   case Intrinsic::x86_sse2_cvtsd2si:
   case Intrinsic::x86_sse2_cvtsd2si64:
   case Intrinsic::x86_sse2_cvttsd2si:
   case Intrinsic::x86_sse2_cvttsd2si64:
if (auto *Op2 = dyn_cast<ConstantFP>(Operands[1])) {
       }
 
       if (IntrinsicID == Intrinsic::maximum) {
         const APFloat &C1 = Op1->getValueAPF();
         const APFloat &C2 = Op2->getValueAPF();
         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;
 
       LibFunc Func = NotLibFunc;
       TLI->getLibFunc(Name, Func);
       switch (Func) {
       default:
         break;

llvm/test/Analysis/ConstantFolding/AMDGPU/fmul_legacy.ll

+; 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
+}

llvm/test/Analysis/ConstantFolding/AMDGPU/lit.local.cfg

+if not 'AMDGPU' in config.root.targets:
+    config.unsupported = True