diff options
| author | Evan Cheng <evan.cheng@apple.com> | 2008-05-23 21:23:16 +0000 |
|---|---|---|
| committer | Evan Cheng <evan.cheng@apple.com> | 2008-05-23 21:23:16 +0000 |
| commit | cd0baf21a1f3582ada94d11cf1ccbf342b504077 (patch) | |
| tree | 57d4d609a1f5d2cfc38d5c26f9ff669ef7b7598e /lib/Target/X86/X86InstrSSE.td | |
| parent | 3bf19e4c42203a8691d8d3fbdb5fcf8ab2f6597b (diff) | |
Use movlps / movhps to modify low / high half of 16-byet memory location.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@51501 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target/X86/X86InstrSSE.td')
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 37 |
1 files changed, 35 insertions, 2 deletions
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index e9c6481fd7..06b7a3ab2e 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -2977,13 +2977,15 @@ def : Pat<(v2f64 (vector_shuffle VR128:$src1, (memop addr:$src2), MOVHP_shuffle_mask)), (MOVHPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>; -def : Pat<(v4i32 (vector_shuffle VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)), +def : Pat<(v4i32 (vector_shuffle VR128:$src1, + (bc_v4i32 (memopv2i64 addr:$src2)), MOVLP_shuffle_mask)), (MOVLPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>; def : Pat<(v2i64 (vector_shuffle VR128:$src1, (memop addr:$src2), MOVLP_shuffle_mask)), (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>; -def : Pat<(v4i32 (vector_shuffle VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)), +def : Pat<(v4i32 (vector_shuffle VR128:$src1, + (bc_v4i32 (memopv2i64 addr:$src2)), MOVHP_shuffle_mask)), (MOVHPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>; def : Pat<(v2i64 (vector_shuffle VR128:$src1, (memop addr:$src2), @@ -2991,6 +2993,37 @@ def : Pat<(v2i64 (vector_shuffle VR128:$src1, (memop addr:$src2), (MOVHPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>; } +// (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS +// (store (vector_shuffle (load addr), v2, <0, 1, 4, 5>), addr) using MOVHPS +def : Pat<(store (v4f32 (vector_shuffle (memop addr:$src1), VR128:$src2, + MOVLP_shuffle_mask)), addr:$src1), + (MOVLPSmr addr:$src1, VR128:$src2)>, Requires<[HasSSE1]>; +def : Pat<(store (v2f64 (vector_shuffle (memop addr:$src1), VR128:$src2, + MOVLP_shuffle_mask)), addr:$src1), + (MOVLPDmr addr:$src1, VR128:$src2)>, Requires<[HasSSE2]>; +def : Pat<(store (v4f32 (vector_shuffle (memop addr:$src1), VR128:$src2, + MOVHP_shuffle_mask)), addr:$src1), + (MOVHPSmr addr:$src1, VR128:$src2)>, Requires<[HasSSE1]>; +def : Pat<(store (v2f64 (vector_shuffle (memop addr:$src1), VR128:$src2, + MOVHP_shuffle_mask)), addr:$src1), + (MOVHPDmr addr:$src1, VR128:$src2)>, Requires<[HasSSE2]>; + +def : Pat<(store (v4i32 (vector_shuffle + (bc_v4i32 (memopv2i64 addr:$src1)), VR128:$src2, + MOVLP_shuffle_mask)), addr:$src1), + (MOVLPSmr addr:$src1, VR128:$src2)>, Requires<[HasSSE1]>; +def : Pat<(store (v2i64 (vector_shuffle (memop addr:$src1), VR128:$src2, + MOVLP_shuffle_mask)), addr:$src1), + (MOVLPDmr addr:$src1, VR128:$src2)>, Requires<[HasSSE2]>; +def : Pat<(store (v4i32 (vector_shuffle + (bc_v4i32 (memopv2i64 addr:$src1)), VR128:$src2, + MOVHP_shuffle_mask)), addr:$src1), + (MOVHPSmr addr:$src1, VR128:$src2)>, Requires<[HasSSE1]>; +def : Pat<(store (v2i64 (vector_shuffle (memop addr:$src1), VR128:$src2, + MOVHP_shuffle_mask)), addr:$src1), + (MOVHPDmr addr:$src1, VR128:$src2)>, Requires<[HasSSE2]>; + + let AddedComplexity = 15 in { // Setting the lowest element in the vector. def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2, |