From dd224d23336695a765356a94ace7cad2d5428760 Mon Sep 17 00:00:00 2001 From: Dale Johannesen Date: Thu, 30 Sep 2010 23:57:10 +0000 Subject: 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. llvm-svn: 115243 --- llvm/test/CodeGen/X86/vec_insert-7.ll | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-) (limited to 'llvm/test/CodeGen/X86/vec_insert-7.ll') diff --git a/llvm/test/CodeGen/X86/vec_insert-7.ll b/llvm/test/CodeGen/X86/vec_insert-7.ll index 9ede10f63d3..ea7f9193042 100644 --- a/llvm/test/CodeGen/X86/vec_insert-7.ll +++ b/llvm/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 } -- cgit v1.2.3