Use movsd to shuffle in the lowest two elements of a v4f32 / v4i32 vector when

movlps cannot be used (e.g. when load from m64 has multiple uses).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@28089 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Evan Cheng 2006-05-03 20:32:03 +00:00
parent 5a032de387
commit 9e062ed516
2 changed files with 19 additions and 0 deletions

View File

@ -2481,6 +2481,14 @@ def : Pat<(v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
MOVL_shuffle_mask)),
(v2i64 (MOVLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
// vector_shuffle v1, v2 <4, 5, 2, 3> using MOVLPDrr (movsd)
def : Pat<(v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
MOVLP_shuffle_mask)),
(v4f32 (MOVLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
MOVLP_shuffle_mask)),
(v4i32 (MOVLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
// Set lowest element and zero upper elements.
def : Pat<(bc_v2i64 (vector_shuffle immAllZerosV,
(v2f64 (scalar_to_vector (loadf64 addr:$src))),

View File

@ -0,0 +1,11 @@
; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep movsd | wc -l | grep 1
void %test() {
%tmp1 = load <4 x float>* null
%tmp2 = shufflevector <4 x float> %tmp1, <4 x float> < float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00 >, <4 x uint> < uint 0, uint 1, uint 6, uint 7 >
%tmp3 = shufflevector <4 x float> %tmp1, <4 x float> zeroinitializer, <4 x uint> < uint 2, uint 3, uint 6, uint 7 >
%tmp4 = add <4 x float> %tmp2, %tmp3
store <4 x float> %tmp4, <4 x float>* null
ret void
}