diff options
| author | Dale Johannesen <dalej@apple.com> | 2010-09-30 23:57:10 +0000 |
|---|---|---|
| committer | Dale Johannesen <dalej@apple.com> | 2010-09-30 23:57:10 +0000 |
| commit | 0488fb649a56b7fc89a5814df5308813f9e5a85d (patch) | |
| tree | 21913bb81960866b73c6b49f29782c142563304c /test/CodeGen/X86/vec_insert-7.ll | |
| parent | a7e3b564773cc45a1b08383c09fe86b17b3ffe92 (diff) | |
| download | external_llvm-0488fb649a56b7fc89a5814df5308813f9e5a85d.zip external_llvm-0488fb649a56b7fc89a5814df5308813f9e5a85d.tar.gz external_llvm-0488fb649a56b7fc89a5814df5308813f9e5a85d.tar.bz2 | |
Massive rewrite of MMX:
The x86_mmx type is used for MMX intrinsics, parameters and
return values where these use MMX registers, and is also
supported in load, store, and bitcast.
Only the above operations generate MMX instructions, and optimizations
do not operate on or produce MMX intrinsics.
MMX-sized vectors <2 x i32> etc. are lowered to XMM or split into
smaller pieces. Optimizations may occur on these forms and the
result casted back to x86_mmx, provided the result feeds into a
previous existing x86_mmx operation.
The point of all this is prevent optimizations from introducing
MMX operations, which is unsafe due to the EMMS problem.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115243 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/X86/vec_insert-7.ll')
| -rw-r--r-- | test/CodeGen/X86/vec_insert-7.ll | 12 |
1 files changed, 8 insertions, 4 deletions
diff --git a/test/CodeGen/X86/vec_insert-7.ll b/test/CodeGen/X86/vec_insert-7.ll index 9ede10f..ea7f919 100644 --- a/test/CodeGen/X86/vec_insert-7.ll +++ b/test/CodeGen/X86/vec_insert-7.ll @@ -1,8 +1,12 @@ -; RUN: llc < %s -march=x86 -mattr=+mmx -mtriple=i686-apple-darwin9 -o - | grep punpckldq +; RUN: llc < %s -march=x86 -mattr=+mmx,+sse -mtriple=i686-apple-darwin9 -o - | grep pinsrd | count 2 +; MMX insertelement is not available; these are promoted to XMM. +; (Without SSE they are split to two ints, and the code is much better.) -define <2 x i32> @mmx_movzl(<2 x i32> %x) nounwind { +define x86_mmx @mmx_movzl(x86_mmx %x) nounwind { entry: - %tmp3 = insertelement <2 x i32> %x, i32 32, i32 0 ; <<2 x i32>> [#uses=1] + %tmp = bitcast x86_mmx %x to <2 x i32> + %tmp3 = insertelement <2 x i32> %tmp, i32 32, i32 0 ; <<2 x i32>> [#uses=1] %tmp8 = insertelement <2 x i32> %tmp3, i32 0, i32 1 ; <<2 x i32>> [#uses=1] - ret <2 x i32> %tmp8 + %tmp9 = bitcast <2 x i32> %tmp8 to x86_mmx + ret x86_mmx %tmp9 } |