diff options
| author | Mon P Wang <wangmp@apple.com> | 2008-11-16 05:06:27 +0000 |
|---|---|---|
| committer | Mon P Wang <wangmp@apple.com> | 2008-11-16 05:06:27 +0000 |
| commit | c7849c22f4804075c0c972e20f9cd701bdb6ab6f (patch) | |
| tree | 9f4dda703c98f0b9f5a3db4fdc973b15b3e15ac5 /test/CodeGen/X86/vec_shuffle-26.ll | |
| parent | 29cd5ba621268c44c8be72397e00c79899ecdf0b (diff) | |
Improved shuffle normalization to avoid using extract/build when we
can extract using different indexes for two vectors. Added a few tests
for vector shuffles.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59399 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/X86/vec_shuffle-26.ll')
| -rw-r--r-- | test/CodeGen/X86/vec_shuffle-26.ll | 29 |
1 files changed, 29 insertions, 0 deletions
diff --git a/test/CodeGen/X86/vec_shuffle-26.ll b/test/CodeGen/X86/vec_shuffle-26.ll new file mode 100644 index 0000000000..abd6e90b79 --- /dev/null +++ b/test/CodeGen/X86/vec_shuffle-26.ll @@ -0,0 +1,29 @@ +; RUN: llvm-as < %s | llc -march=x86 -mattr=sse41 -o %t -f +; RUN: grep unpcklps %t | count 1 +; RUN: grep unpckhps %t | count 3 + +; Transpose example using the more generic vector shuffle. Return float8 +; instead of float16 +; ModuleID = 'transpose2_opt.bc' +target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32" +target triple = "i386-apple-cl.1.0" +@r0 = common global <4 x float> zeroinitializer, align 16 ; <<4 x float>*> [#uses=1] +@r1 = common global <4 x float> zeroinitializer, align 16 ; <<4 x float>*> [#uses=1] +@r2 = common global <4 x float> zeroinitializer, align 16 ; <<4 x float>*> [#uses=1] +@r3 = common global <4 x float> zeroinitializer, align 16 ; <<4 x float>*> [#uses=1] + +define <8 x float> @__transpose2(<4 x float> %p0, <4 x float> %p1, <4 x float> %p2, <4 x float> %p3) nounwind { +entry: + %unpcklps = shufflevector <4 x float> %p0, <4 x float> %p2, <4 x i32> < i32 0, i32 4, i32 1, i32 5 > ; <<4 x float>> [#uses=2] + %unpckhps = shufflevector <4 x float> %p0, <4 x float> %p2, <4 x i32> < i32 2, i32 6, i32 3, i32 7 > ; <<4 x float>> [#uses=2] + %unpcklps8 = shufflevector <4 x float> %p1, <4 x float> %p3, <4 x i32> < i32 0, i32 4, i32 1, i32 5 > ; <<4 x float>> [#uses=2] + %unpckhps11 = shufflevector <4 x float> %p1, <4 x float> %p3, <4 x i32> < i32 2, i32 6, i32 3, i32 7 > ; <<4 x float>> [#uses=2] + %unpcklps14 = shufflevector <4 x float> %unpcklps, <4 x float> %unpcklps8, <4 x i32> < i32 0, i32 4, i32 1, i32 5 > ; <<4 x float>> [#uses=1] + %unpckhps17 = shufflevector <4 x float> %unpcklps, <4 x float> %unpcklps8, <4 x i32> < i32 2, i32 6, i32 3, i32 7 > ; <<4 x float>> [#uses=1] + %r1 = shufflevector <4 x float> %unpcklps14, <4 x float> %unpckhps17, <8 x i32> < i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 >; + %unpcklps20 = shufflevector <4 x float> %unpckhps, <4 x float> %unpckhps11, <4 x i32> < i32 0, i32 4, i32 1, i32 5 > ; <<4 x float>> [#uses=1] + %unpckhps23 = shufflevector <4 x float> %unpckhps, <4 x float> %unpckhps11, <4 x i32> < i32 2, i32 6, i32 3, i32 7 > ; <<4 x float>> [#uses=1] + %r2 = shufflevector <4 x float> %unpcklps20, <4 x float> %unpckhps23, <8 x i32> < i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 >; +; %r3 = shufflevector <8 x float> %r1, <8 x float> %r2, <16 x i32> < i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15 >; + ret <8 x float> %r2; +} |