AMDGPU: Turn class x, p_zero|n_zero into fcmp oeq x, 0
The library does use this for some reason.
llvm-svn: 339461
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 310b87e..f0c4ceb 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -3195,6 +3195,15 @@
return replaceInstUsesWith(*II, FCmp);
}
+ if (Mask == (N_ZERO | P_ZERO)) {
+ // Equivalent of == 0.
+ Value *FCmp = Builder.CreateFCmpOEQ(
+ Src0, ConstantFP::get(Src0->getType(), 0.0));
+
+ FCmp->takeName(II);
+ return replaceInstUsesWith(*II, FCmp);
+ }
+
const ConstantFP *CVal = dyn_cast<ConstantFP>(Src0);
if (!CVal) {
if (isa<UndefValue>(Src0))
diff --git a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
index 1fad1d8..e7b2fe3 100644
--- a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
+++ b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
@@ -457,6 +457,13 @@
ret i1 %val
}
+; CHECK-LABEL: @test_class_is_p0_n0_f32(
+; CHECK: %val = fcmp oeq float %x, 0.000000e+00
+define i1 @test_class_is_p0_n0_f32(float %x) nounwind {
+ %val = call i1 @llvm.amdgcn.class.f32(float %x, i32 96)
+ ret i1 %val
+}
+
; CHECK-LABEL: @test_constant_class_snan_test_snan_f64(
; CHECK: ret i1 true
define i1 @test_constant_class_snan_test_snan_f64() nounwind {