diff options
| author | Hal Finkel <hfinkel@anl.gov> | 2013-11-16 21:29:08 +0000 |
|---|---|---|
| committer | Hal Finkel <hfinkel@anl.gov> | 2013-11-16 21:29:08 +0000 |
| commit | 64fa501b1081b5d5c25e5e9639075abb0cb724d9 (patch) | |
| tree | a190f3ea48a03753cd6f44e382cf97464e79b63b /test | |
| parent | e6e811277f045ee3d61cd62622d71005c47eb48d (diff) | |
| download | external_llvm-64fa501b1081b5d5c25e5e9639075abb0cb724d9.zip external_llvm-64fa501b1081b5d5c25e5e9639075abb0cb724d9.tar.gz external_llvm-64fa501b1081b5d5c25e5e9639075abb0cb724d9.tar.bz2 | |
Apply the InstCombine fptrunc sqrt optimization to llvm.sqrt
InstCombine, in visitFPTrunc, applies the following optimization to sqrt calls:
(fptrunc (sqrt (fpext x))) -> (sqrtf x)
but does not apply the same optimization to llvm.sqrt. This is a problem
because, to enable vectorization, Clang generates llvm.sqrt instead of sqrt in
fast-math mode, and because this optimization is being applied to sqrt and not
applied to llvm.sqrt, sometimes the fast-math code is slower.
This change makes InstCombine apply this optimization to llvm.sqrt as well.
This fixes the specific problem in PR17758, although the same underlying issue
(optimizations applied to libcalls are not applied to intrinsics) exists for
other optimizations in SimplifyLibCalls.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194935 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test')
| -rw-r--r-- | test/Transforms/InstCombine/double-float-shrink-1.ll | 13 |
1 files changed, 13 insertions, 0 deletions
diff --git a/test/Transforms/InstCombine/double-float-shrink-1.ll b/test/Transforms/InstCombine/double-float-shrink-1.ll index e5448ee..5cacb59 100644 --- a/test/Transforms/InstCombine/double-float-shrink-1.ll +++ b/test/Transforms/InstCombine/double-float-shrink-1.ll @@ -263,6 +263,7 @@ define double @sin_test2(float %f) nounwind readnone { ret double %call ; CHECK: call double @sin(double %conv) } + define float @sqrt_test(float %f) nounwind readnone { ; CHECK: sqrt_test %conv = fpext float %f to double @@ -272,6 +273,15 @@ define float @sqrt_test(float %f) nounwind readnone { ; CHECK: call float @sqrtf(float %f) } +define float @sqrt_int_test(float %f) nounwind readnone { +; CHECK: sqrt_int_test + %conv = fpext float %f to double + %call = call double @llvm.sqrt.f64(double %conv) + %conv1 = fptrunc double %call to float + ret float %conv1 +; CHECK: call float @llvm.sqrt.f32(float %f) +} + define double @sqrt_test2(float %f) nounwind readnone { ; CHECK: sqrt_test2 %conv = fpext float %f to double @@ -331,3 +341,6 @@ declare double @acos(double) nounwind readnone declare double @acosh(double) nounwind readnone declare double @asin(double) nounwind readnone declare double @asinh(double) nounwind readnone + +declare double @llvm.sqrt.f64(double) nounwind readnone + |