diff options
| author | Simon Pilgrim <llvm-dev@redking.me.uk> | 2014-12-02 22:31:23 +0000 |
|---|---|---|
| committer | Simon Pilgrim <llvm-dev@redking.me.uk> | 2014-12-02 22:31:23 +0000 |
| commit | 6b988ad8f2285e8bb3c56db38c0416f6186d3889 (patch) | |
| tree | 4d91ac43a414d292c08e53221c42bd662eff0f2d /llvm/lib | |
| parent | 656c29b08f1d32071b63d413a209cd418b5240fa (diff) | |
| download | bcm5719-llvm-6b988ad8f2285e8bb3c56db38c0416f6186d3889.tar.gz bcm5719-llvm-6b988ad8f2285e8bb3c56db38c0416f6186d3889.zip | |
[X86][SSE] Keep 4i32 vector insertions in integer domain on SSE4.1 targets
4i32 shuffles for single insertions into zero vectors lowers to X86vzmovl which was using (v)blendps - causing domain switch stalls. This patch fixes this by using (v)pblendw instead.
The updated tests on test/CodeGen/X86/sse41.ll still contain a domain stall due to the use of insertps - I'm looking at fixing this in a future patch.
Differential Revision: http://reviews.llvm.org/D6458
llvm-svn: 223165
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrSSE.td | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td index 1b07e874837..0bf765b53dd 100644 --- a/llvm/lib/Target/X86/X86InstrSSE.td +++ b/llvm/lib/Target/X86/X86InstrSSE.td @@ -7734,7 +7734,7 @@ let Predicates = [UseAVX] in { def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))), (VBLENDPSrri (v4f32 (V_SET0)), VR128:$src, (i8 1))>; def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))), - (VBLENDPSrri (v4i32 (V_SET0)), VR128:$src, (i8 1))>; + (VPBLENDWrri (v4i32 (V_SET0)), VR128:$src, (i8 3))>; def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))), (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>; @@ -7769,7 +7769,7 @@ let Predicates = [UseSSE41] in { def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))), (BLENDPSrri (v4f32 (V_SET0)), VR128:$src, (i8 1))>; def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))), - (BLENDPSrri (v4f32 (V_SET0)), VR128:$src, (i8 1))>; + (PBLENDWrri (v4i32 (V_SET0)), VR128:$src, (i8 3))>; def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))), (BLENDPDrri (v2f64 (V_SET0)), VR128:$src, (i8 1))>; } |
