diff options
| author | Guy Blank <guy.blank@intel.com> | 2017-07-31 08:26:14 +0000 |
|---|---|---|
| committer | Guy Blank <guy.blank@intel.com> | 2017-07-31 08:26:14 +0000 |
| commit | b169d56dc36ada9aa137763fa50056790513ee7a (patch) | |
| tree | 1496d61315a8738d3b978929ec735abb821a2fce /llvm/lib | |
| parent | e9a5e7e407816f338b2aba97a5b92f729986556b (diff) | |
| download | bcm5719-llvm-b169d56dc36ada9aa137763fa50056790513ee7a.tar.gz bcm5719-llvm-b169d56dc36ada9aa137763fa50056790513ee7a.zip | |
[X86][AVX512] Add masked MOVS[S|D] patterns
Added patterns to recognize AND 1 on the mask of a scalar masked
move is not needed since only the lower bit is relevant for the
instruction.
Differential Revision:
https://reviews.llvm.org/D35897
llvm-svn: 309546
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrAVX512.td | 16 |
1 files changed, 16 insertions, 0 deletions
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td index 2b5bbc1c094..079116353bc 100644 --- a/llvm/lib/Target/X86/X86InstrAVX512.td +++ b/llvm/lib/Target/X86/X86InstrAVX512.td @@ -4002,10 +4002,26 @@ defm : avx512_load_scalar_lowering_subreg<"VMOVSSZ", avx512vl_f32_info, defm : avx512_load_scalar_lowering_subreg<"VMOVSDZ", avx512vl_f64_info, (v8i1 (bitconvert (i8 (and GR8:$mask, (i8 1))))), GR8, sub_8bit>; +def : Pat<(f32 (X86selects (scalar_to_vector (and GR8:$mask, (i8 1))), + (f32 FR32X:$src1), (f32 FR32X:$src2))), + (COPY_TO_REGCLASS + (VMOVSSZrrk (COPY_TO_REGCLASS FR32X:$src2, VR128X), + (COPY_TO_REGCLASS (i32 (INSERT_SUBREG (IMPLICIT_DEF), + GR8:$mask, sub_8bit)), VK1WM), + (v4f32 (IMPLICIT_DEF)), FR32X:$src1), FR32X)>; + def : Pat<(f32 (X86selects VK1WM:$mask, (f32 FR32X:$src1), (f32 FR32X:$src2))), (COPY_TO_REGCLASS (VMOVSSZrrk (COPY_TO_REGCLASS FR32X:$src2, VR128X), VK1WM:$mask, (v4f32 (IMPLICIT_DEF)), FR32X:$src1), FR32X)>; +def : Pat<(f64 (X86selects (scalar_to_vector (and GR8:$mask, (i8 1))), + (f64 FR64X:$src1), (f64 FR64X:$src2))), + (COPY_TO_REGCLASS + (VMOVSDZrrk (COPY_TO_REGCLASS FR64X:$src2, VR128X), + (COPY_TO_REGCLASS (i32 (INSERT_SUBREG (IMPLICIT_DEF), + GR8:$mask, sub_8bit)), VK1WM), + (v2f64 (IMPLICIT_DEF)), FR64X:$src1), FR64X)>; + def : Pat<(f64 (X86selects VK1WM:$mask, (f64 FR64X:$src1), (f64 FR64X:$src2))), (COPY_TO_REGCLASS (VMOVSDZrrk (COPY_TO_REGCLASS FR64X:$src2, VR128X), VK1WM:$mask, (v2f64 (IMPLICIT_DEF)), FR64X:$src1), FR64X)>; |
