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

1 files changed, 6 insertions, 7 deletions

diff --git a/test/CodeGen/X86/2007-06-15-IntToMMX.ll b/test/CodeGen/X86/2007-06-15-IntToMMX.ll
index 6128d8b92d..660d4fe7b1 100644
--- a/test/CodeGen/X86/2007-06-15-IntToMMX.ll
+++ b/test/CodeGen/X86/2007-06-15-IntToMMX.ll
@@ -1,17 +1,16 @@
 ; RUN: llc < %s -march=x86-64 -mattr=+mmx | grep paddusw
-@R = external global <1 x i64>          ; <<1 x i64>*> [#uses=1]
+@R = external global x86_mmx          ; <x86_mmx*> [#uses=1]
 
 define void @foo(<1 x i64> %A, <1 x i64> %B) {
 entry:
-        %tmp4 = bitcast <1 x i64> %B to <4 x i16>               ; <<4 x i16>> [#uses=1]
-        %tmp6 = bitcast <1 x i64> %A to <4 x i16>               ; <<4 x i16>> [#uses=1]
-        %tmp7 = tail call <4 x i16> @llvm.x86.mmx.paddus.w( <4 x i16> %tmp6, <4 x i16> %tmp4 )   ; <<4 x i16>> [#uses=1]
-        %tmp8 = bitcast <4 x i16> %tmp7 to <1 x i64>            ; <<1 x i64>> [#uses=1]
-        store <1 x i64> %tmp8, <1 x i64>* @R
+        %tmp2 = bitcast <1 x i64> %A to x86_mmx
+        %tmp3 = bitcast <1 x i64> %B to x86_mmx
+        %tmp7 = tail call x86_mmx @llvm.x86.mmx.paddus.w( x86_mmx %tmp2, x86_mmx %tmp3 )   ; <x86_mmx> [#uses=1]
+        store x86_mmx %tmp7, x86_mmx* @R
         tail call void @llvm.x86.mmx.emms( )
         ret void
 }
 
-declare <4 x i16> @llvm.x86.mmx.paddus.w(<4 x i16>, <4 x i16>)
+declare x86_mmx @llvm.x86.mmx.paddus.w(x86_mmx, x86_mmx)
 
 declare void @llvm.x86.mmx.emms()