diff options
| author | Manman Ren <mren@apple.com> | 2012-10-30 23:53:59 +0000 |
|---|---|---|
| committer | Manman Ren <mren@apple.com> | 2012-10-30 23:53:59 +0000 |
| commit | dfd0b9b460686ca9491e49dd3647beec5e748a1a (patch) | |
| tree | 1713bc57a1af86db5c6db4072b03a90e02bfd179 /test | |
| parent | bbc6e671b1a902c96aff152cc524a5ee6e253907 (diff) | |
X86 SSE: update rsqrtss and rcpss to use two source operands and
the first source operand is tied to the destination operand.
This is to accurately model the corresponding instructions where the upper
bits are unmodified.
rdar://12558838
PR14221
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167064 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test')
| -rw-r--r-- | test/CodeGen/X86/sse_partial_update.ll | 36 |
1 files changed, 36 insertions, 0 deletions
diff --git a/test/CodeGen/X86/sse_partial_update.ll b/test/CodeGen/X86/sse_partial_update.ll new file mode 100644 index 0000000000..655f75800c --- /dev/null +++ b/test/CodeGen/X86/sse_partial_update.ll @@ -0,0 +1,36 @@ +; RUN: llc < %s -mtriple=x86_64-apple-macosx -mattr=+sse2 -mcpu=nehalem | FileCheck %s + +; rdar: 12558838 +; PR14221 +; There is a mismatch between the intrinsic and the actual instruction. +; The actual instruction has a partial update of dest, while the intrinsic +; passes through the upper FP values. Here, we make sure the source and +; destination of rsqrtss are the same. +define void @t1(<4 x float> %a) nounwind uwtable ssp { +entry: +; CHECK: t1: +; CHECK: rsqrtss %xmm0, %xmm0 + %0 = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> %a) nounwind + %a.addr.0.extract = extractelement <4 x float> %0, i32 0 + %conv = fpext float %a.addr.0.extract to double + %a.addr.4.extract = extractelement <4 x float> %0, i32 1 + %conv3 = fpext float %a.addr.4.extract to double + tail call void @callee(double %conv, double %conv3) nounwind + ret void +} +declare void @callee(double, double) +declare <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float>) nounwind readnone + +define void @t2(<4 x float> %a) nounwind uwtable ssp { +entry: +; CHECK: t2: +; CHECK: rcpss %xmm0, %xmm0 + %0 = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> %a) nounwind + %a.addr.0.extract = extractelement <4 x float> %0, i32 0 + %conv = fpext float %a.addr.0.extract to double + %a.addr.4.extract = extractelement <4 x float> %0, i32 1 + %conv3 = fpext float %a.addr.4.extract to double + tail call void @callee(double %conv, double %conv3) nounwind + ret void +} +declare <4 x float> @llvm.x86.sse.rcp.ss(<4 x float>) nounwind readnone |