diff options
| author | Evan Cheng <evan.cheng@apple.com> | 2008-09-26 23:41:32 +0000 |
|---|---|---|
| committer | Evan Cheng <evan.cheng@apple.com> | 2008-09-26 23:41:32 +0000 |
| commit | b7a75a5a541a836d07cfc5ccebd70bb5e42f797f (patch) | |
| tree | 66181d36082ae4203edbca00c97e614e31fc7f3b /lib/Target/X86/X86InstrSSE.td | |
| parent | 19c874638d9478a5d5028854817a5ee72293bb2b (diff) | |
Implement "punpckldq %xmm0, $xmm0" as "pshufd $0x50, %xmm0, %xmm" unless optimizing for code size.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56711 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target/X86/X86InstrSSE.td')
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 29 |
1 files changed, 25 insertions, 4 deletions
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index af7694a1ee..8b4579517e 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -744,7 +744,7 @@ def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src), addr:$dst)]>; let Constraints = "$src1 = $dst" in { -let AddedComplexity = 15 in { +let AddedComplexity = 20 in { def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), "movlhps\t{$src2, $dst|$dst, $src2}", [(set VR128:$dst, @@ -759,7 +759,7 @@ def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128: } // AddedComplexity } // Constraints = "$src1 = $dst" -let AddedComplexity = 15 in +let AddedComplexity = 20 in def : Pat<(v4f32 (vector_shuffle VR128:$src, (undef), MOVDDUP_shuffle_mask)), (MOVLHPSrr VR128:$src, VR128:$src)>, Requires<[HasSSE1]>; @@ -2921,6 +2921,7 @@ def : Pat<(vector_shuffle (bc_v4i32 (memopv4f32 addr:$src1)), (undef), SHUFP_unary_shuffle_mask:$sm), (PSHUFDmi addr:$src1, SHUFP_unary_shuffle_mask:$sm)>, Requires<[HasSSE2]>; + // Special binary v4i32 shuffle cases with SHUFPS. def : Pat<(v4i32 (vector_shuffle VR128:$src1, (v4i32 VR128:$src2), PSHUFD_binary_shuffle_mask:$sm)), @@ -2937,11 +2938,21 @@ def : Pat<(v2i64 (vector_shuffle VR128:$src1, VR128:$src2, Requires<[HasSSE2]>; // Special unary SHUFPDrri case. def : Pat<(v2i64 (vector_shuffle VR128:$src1, (undef), - SHUFP_unary_shuffle_mask:$sm)), + SHUFP_unary_shuffle_mask:$sm)), (SHUFPDrri VR128:$src1, VR128:$src1, SHUFP_unary_shuffle_mask:$sm)>, Requires<[HasSSE2]>; // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...> +let AddedComplexity = 15 in { +def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef), + UNPCKL_v_undef_shuffle_mask:$sm)), + (PSHUFDri VR128:$src, PSHUFD_shuffle_mask:$sm)>, + Requires<[OptForSpeed, HasSSE2]>; +def : Pat<(v4f32 (vector_shuffle VR128:$src, (undef), + UNPCKL_v_undef_shuffle_mask:$sm)), + (PSHUFDri VR128:$src, PSHUFD_shuffle_mask:$sm)>, + Requires<[OptForSpeed, HasSSE2]>; +} let AddedComplexity = 10 in { def : Pat<(v4f32 (vector_shuffle VR128:$src, (undef), UNPCKL_v_undef_shuffle_mask)), @@ -2958,6 +2969,16 @@ def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef), } // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...> +let AddedComplexity = 15 in { +def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef), + UNPCKH_v_undef_shuffle_mask:$sm)), + (PSHUFDri VR128:$src, PSHUFD_shuffle_mask:$sm)>, + Requires<[OptForSpeed, HasSSE2]>; +def : Pat<(v4f32 (vector_shuffle VR128:$src, (undef), + UNPCKH_v_undef_shuffle_mask:$sm)), + (PSHUFDri VR128:$src, PSHUFD_shuffle_mask:$sm)>, + Requires<[OptForSpeed, HasSSE2]>; +} let AddedComplexity = 10 in { def : Pat<(v4f32 (vector_shuffle VR128:$src, (undef), UNPCKH_v_undef_shuffle_mask)), @@ -2973,7 +2994,7 @@ def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef), (PUNPCKHDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>; } -let AddedComplexity = 15 in { +let AddedComplexity = 20 in { // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2, MOVHP_shuffle_mask)), |