diff options
Diffstat (limited to 'lib/Target/X86/X86InstrSSE.td')
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 50 |
1 files changed, 31 insertions, 19 deletions
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 408ab167..65e3c1e 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -6742,6 +6742,16 @@ let Predicates = [HasAVX] in { def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2), (imm:$mask))), (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>; + + def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2), + (imm:$mask))), + (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>; + def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2), + (imm:$mask))), + (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>; + def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2), + (imm:$mask))), + (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>; } let Predicates = [HasAVX2] in { @@ -6802,13 +6812,13 @@ let Predicates = [HasSSE41] in { def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2), (imm:$mask))), - (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>; + (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>; def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2), (imm:$mask))), - (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>; + (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>; def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2), (imm:$mask))), - (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>; + (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>; } @@ -7725,45 +7735,47 @@ def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))), // multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag, - Intrinsic Int> { + ValueType OpVT> { def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src1, VR256:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V; + [(set VR256:$dst, + (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V; def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR256:$dst, (Int VR256:$src1, - (bitconvert (mem_frag addr:$src2))))]>, + [(set VR256:$dst, + (OpVT (X86VPermv VR256:$src1, + (bitconvert (mem_frag addr:$src2)))))]>, VEX_4V; } -defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>; +defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>; let ExeDomain = SSEPackedSingle in -defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>; +defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>; multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag, - Intrinsic Int> { - def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst), + ValueType OpVT> { + def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX; - def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst), + [(set VR256:$dst, + (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX; + def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>, - VEX; + [(set VR256:$dst, + (OpVT (X86VPermi (mem_frag addr:$src1), + (i8 imm:$src2))))]>, VEX; } -defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>, - VEX_W; +defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W; let ExeDomain = SSEPackedDouble in -defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>, - VEX_W; +defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W; //===----------------------------------------------------------------------===// // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks |
