diff options
Diffstat (limited to 'lib/Target/X86/X86InstrSSE.td')
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 27 |
1 files changed, 25 insertions, 2 deletions
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 408ab16..8210965 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -7746,12 +7746,12 @@ defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>; multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag, Intrinsic Int> { - def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst), + def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$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:$dst), + def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), @@ -7765,6 +7765,29 @@ let ExeDomain = SSEPackedDouble in defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>, VEX_W; +let Predicates = [HasAVX2] in { +def : Pat<(v8i32 (X86VPermd VR256:$src1, VR256:$src2)), + (VPERMDYrr VR256:$src1, VR256:$src2)>; +def : Pat<(v8f32 (X86VPermps VR256:$src1, VR256:$src2)), + (VPERMPSYrr VR256:$src1, VR256:$src2)>; + +def : Pat<(v4i64 (X86VPermq VR256:$src1, (i8 imm:$imm))), + (VPERMQYri VR256:$src1, imm:$imm)>; +def : Pat<(v4f64 (X86VPermpd VR256:$src1, (i8 imm:$imm))), + (VPERMPDYri VR256:$src1, imm:$imm)>; + +def : Pat<(v8i32 (X86VPermps VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))), + (VPERMDYrm VR256:$src1, addr:$src2)>; +def : Pat<(v8f32 (X86VPermps VR256:$src1, (memopv8f32 addr:$src2))), + (VPERMPSYrm VR256:$src1, addr:$src2)>; + +def : Pat<(v4i64 (X86VPermq (memopv4i64 addr:$src1), (i8 imm:$imm))), + (VPERMQYmi addr:$src1, imm:$imm)>; +def : Pat<(v4f64 (X86VPermpd (memopv4f64 addr:$src1), (i8 imm:$imm))), + (VPERMPDYmi addr:$src1, imm:$imm)>; + +} + //===----------------------------------------------------------------------===// // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks // |