diff options
| author | Dan Gohman <gohman@apple.com> | 2007-08-02 21:17:01 +0000 |
|---|---|---|
| committer | Dan Gohman <gohman@apple.com> | 2007-08-02 21:17:01 +0000 |
| commit | 7f55fcbc6bec56662e46d35ca604553effb7a525 (patch) | |
| tree | 0c686baf843b7a9d50dd3ff3408eddd43e0b0def /test/CodeGen/X86/sse-align-12.ll | |
| parent | f3372d1d6416349e42d90d75b912d31e61409bed (diff) | |
| download | llvm-7f55fcbc6bec56662e46d35ca604553effb7a525.tar.gz llvm-7f55fcbc6bec56662e46d35ca604553effb7a525.tar.bz2 llvm-7f55fcbc6bec56662e46d35ca604553effb7a525.tar.xz | |
Fix the alignment requirements of several unpck and shuf instructions.
Generalize isPSHUFDMask and add a unary SHUFPD pattern so that SHUFPD's
memory operand alignment can be tested as well, with a fix to avoid
breaking MMX's use of isPSHUFDMask.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@40756 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/X86/sse-align-12.ll')
| -rw-r--r-- | test/CodeGen/X86/sse-align-12.ll | 50 |
1 files changed, 50 insertions, 0 deletions
diff --git a/test/CodeGen/X86/sse-align-12.ll b/test/CodeGen/X86/sse-align-12.ll new file mode 100644 index 0000000000..731d429608 --- /dev/null +++ b/test/CodeGen/X86/sse-align-12.ll @@ -0,0 +1,50 @@ +; RUN: llvm-as < %s | llc -march=x86-64 | grep unpck | wc -l | grep 2 +; RUN: llvm-as < %s | llc -march=x86-64 | grep shuf | wc -l | grep 2 +; RUN: llvm-as < %s | llc -march=x86-64 | grep ps | wc -l | grep 4 +; RUN: llvm-as < %s | llc -march=x86-64 | grep pd | wc -l | grep 4 +; RUN: llvm-as < %s | llc -march=x86-64 | grep movup | wc -l | grep 4 + +define <4 x float> @a(<4 x float>* %y) +{ + %x = load <4 x float>* %y, align 4 + %a = extractelement <4 x float> %x, i32 0 + %b = extractelement <4 x float> %x, i32 1 + %c = extractelement <4 x float> %x, i32 2 + %d = extractelement <4 x float> %x, i32 3 + %p = insertelement <4 x float> undef, float %d, i32 0 + %q = insertelement <4 x float> %p, float %c, i32 1 + %r = insertelement <4 x float> %q, float %b, i32 2 + %s = insertelement <4 x float> %r, float %a, i32 3 + ret <4 x float> %s +} +define <4 x float> @b(<4 x float>* %y, <4 x float> %z) +{ + %x = load <4 x float>* %y, align 4 + %a = extractelement <4 x float> %x, i32 2 + %b = extractelement <4 x float> %x, i32 3 + %c = extractelement <4 x float> %z, i32 2 + %d = extractelement <4 x float> %z, i32 3 + %p = insertelement <4 x float> undef, float %c, i32 0 + %q = insertelement <4 x float> %p, float %a, i32 1 + %r = insertelement <4 x float> %q, float %d, i32 2 + %s = insertelement <4 x float> %r, float %b, i32 3 + ret <4 x float> %s +} +define <2 x double> @c(<2 x double>* %y) +{ + %x = load <2 x double>* %y, align 8 + %a = extractelement <2 x double> %x, i32 0 + %c = extractelement <2 x double> %x, i32 1 + %p = insertelement <2 x double> undef, double %c, i32 0 + %r = insertelement <2 x double> %p, double %a, i32 1 + ret <2 x double> %r +} +define <2 x double> @d(<2 x double>* %y, <2 x double> %z) +{ + %x = load <2 x double>* %y, align 8 + %a = extractelement <2 x double> %x, i32 1 + %c = extractelement <2 x double> %z, i32 1 + %p = insertelement <2 x double> undef, double %c, i32 0 + %r = insertelement <2 x double> %p, double %a, i32 1 + ret <2 x double> %r +} |