diff options
| author | Ahmed Bougacha <ahmed.bougacha@gmail.com> | 2014-12-06 01:31:07 +0000 |
|---|---|---|
| committer | Ahmed Bougacha <ahmed.bougacha@gmail.com> | 2014-12-06 01:31:07 +0000 |
| commit | 8b54286d1c31dc6a870823ce9532b38b9477ad52 (patch) | |
| tree | 2e44cd8e4859563b6a8e78070410c47930ca69b3 /llvm/lib | |
| parent | 08de833c1cc559eb21a7ffd62224c78c5fc77e4f (diff) | |
| download | bcm5719-llvm-8b54286d1c31dc6a870823ce9532b38b9477ad52.tar.gz bcm5719-llvm-8b54286d1c31dc6a870823ce9532b38b9477ad52.zip | |
[X86] Refactor PMOV[SZ]Xrm to add missing AVX2 patterns.
Most patterns will go away once the extload legalization changes land.
Differential Revision: http://reviews.llvm.org/D6125
llvm-svn: 223567
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrSSE.td | 733 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86IntrinsicsInfo.h | 24 |
2 files changed, 226 insertions, 531 deletions
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td index 547e3835a72..ba444af4882 100644 --- a/llvm/lib/Target/X86/X86InstrSSE.td +++ b/llvm/lib/Target/X86/X86InstrSSE.td @@ -6094,552 +6094,223 @@ def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>, // SSE4.1 - Packed Move with Sign/Zero Extend //===----------------------------------------------------------------------===// -multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId, - OpndItins itins = DEFAULT_ITINS> { - def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, - Sched<[itins.Sched]>; - - def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, - (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))], - itins.rm>, Sched<[itins.Sched.Folded]>; -} - -multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr, - Intrinsic IntId, X86FoldableSchedWrite Sched> { - def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (IntId VR128:$src))]>, Sched<[Sched]>; - - def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (IntId (load addr:$src)))]>, - Sched<[Sched.Folded]>; -} - -let Predicates = [HasAVX] in { -defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", - int_x86_sse41_pmovsxbw, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", - int_x86_sse41_pmovsxwd, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", - int_x86_sse41_pmovsxdq, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", - int_x86_sse41_pmovzxbw, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", - int_x86_sse41_pmovzxwd, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", - int_x86_sse41_pmovzxdq, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -} - -let Predicates = [HasAVX2] in { -defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw", - int_x86_avx2_pmovsxbw, - WriteShuffle>, VEX, VEX_L; -defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd", - int_x86_avx2_pmovsxwd, - WriteShuffle>, VEX, VEX_L; -defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq", - int_x86_avx2_pmovsxdq, - WriteShuffle>, VEX, VEX_L; -defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw", - int_x86_avx2_pmovzxbw, - WriteShuffle>, VEX, VEX_L; -defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd", - int_x86_avx2_pmovzxwd, - WriteShuffle>, VEX, VEX_L; -defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq", - int_x86_avx2_pmovzxdq, - WriteShuffle>, VEX, VEX_L; -} - -defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw, - SSE_INTALU_ITINS_SHUFF_P>; -defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd, - SSE_INTALU_ITINS_SHUFF_P>; -defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq, - SSE_INTALU_ITINS_SHUFF_P>; -defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw, - SSE_INTALU_ITINS_SHUFF_P>; -defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd, - SSE_INTALU_ITINS_SHUFF_P>; -defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq, - SSE_INTALU_ITINS_SHUFF_P>; - -let Predicates = [HasAVX] in { - // Common patterns involving scalar load. - def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)), - (VPMOVSXBWrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)), - (VPMOVSXBWrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))), - (VPMOVSXBWrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)), - (VPMOVSXWDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)), - (VPMOVSXWDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))), - (VPMOVSXWDrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)), - (VPMOVSXDQrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)), - (VPMOVSXDQrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))), - (VPMOVSXDQrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)), - (VPMOVZXBWrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)), - (VPMOVZXBWrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))), - (VPMOVZXBWrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)), - (VPMOVZXWDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)), - (VPMOVZXWDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))), - (VPMOVZXWDrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)), - (VPMOVZXDQrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)), - (VPMOVZXDQrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))), - (VPMOVZXDQrm addr:$src)>; -} - -let Predicates = [UseSSE41] in { - // Common patterns involving scalar load. - def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)), - (PMOVSXBWrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)), - (PMOVSXBWrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))), - (PMOVSXBWrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)), - (PMOVSXWDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)), - (PMOVSXWDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))), - (PMOVSXWDrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)), - (PMOVSXDQrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)), - (PMOVSXDQrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))), - (PMOVSXDQrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)), - (PMOVZXBWrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)), - (PMOVZXBWrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))), - (PMOVZXBWrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)), - (PMOVZXWDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)), - (PMOVZXWDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))), - (PMOVZXWDrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)), - (PMOVZXDQrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)), - (PMOVZXDQrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))), - (PMOVZXDQrm addr:$src)>; -} - -multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId, - OpndItins itins = DEFAULT_ITINS> { - def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), +multiclass SS41I_pmovx_rrrm<bits<8> opc, string OpcodeStr, X86MemOperand MemOp, + RegisterClass OutRC, RegisterClass InRC, + OpndItins itins> { + def rr : SS48I<opc, MRMSrcReg, (outs OutRC:$dst), (ins InRC:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, + [], itins.rr>, Sched<[itins.Sched]>; - def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), + def rm : SS48I<opc, MRMSrcMem, (outs OutRC:$dst), (ins MemOp:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, - (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))], - itins.rm>, Sched<[itins.Sched.Folded]>; + [], + itins.rm>, Sched<[itins.Sched.Folded]>; } -multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr, - Intrinsic IntId, X86FoldableSchedWrite Sched> { - def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (IntId VR128:$src))]>, Sched<[Sched]>; - - def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, - (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>, - Sched<[Sched.Folded]>; -} - -let Predicates = [HasAVX] in { -defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq, - DEFAULT_ITINS_SHUFFLESCHED>, VEX; -} - -let Predicates = [HasAVX2] in { -defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd", - int_x86_avx2_pmovsxbd, WriteShuffle>, - VEX, VEX_L; -defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq", - int_x86_avx2_pmovsxwq, WriteShuffle>, - VEX, VEX_L; -defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd", - int_x86_avx2_pmovzxbd, WriteShuffle>, - VEX, VEX_L; -defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq", - int_x86_avx2_pmovzxwq, WriteShuffle>, - VEX, VEX_L; -} - -defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd, - SSE_INTALU_ITINS_SHUFF_P>; -defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq, - SSE_INTALU_ITINS_SHUFF_P>; -defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd, - SSE_INTALU_ITINS_SHUFF_P>; -defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq, - SSE_INTALU_ITINS_SHUFF_P>; - -let Predicates = [HasAVX] in { - // Common patterns involving scalar load - def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)), - (VPMOVSXBDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)), - (VPMOVSXWQrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)), - (VPMOVZXBDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)), - (VPMOVZXWQrm addr:$src)>; -} - -let Predicates = [UseSSE41] in { - // Common patterns involving scalar load - def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)), - (PMOVSXBDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)), - (PMOVSXWQrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)), - (PMOVZXBDrm addr:$src)>; - def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)), - (PMOVZXWQrm addr:$src)>; -} - -multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId, - X86FoldableSchedWrite Sched> { - def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (IntId VR128:$src))]>, Sched<[Sched]>; - - // Expecting a i16 load any extended to i32 value. - def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (IntId (bitconvert - (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>, - Sched<[Sched.Folded]>; -} - -multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr, - Intrinsic IntId, X86FoldableSchedWrite Sched> { - def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (IntId VR128:$src))]>, Sched<[Sched]>; - - // Expecting a i16 load any extended to i32 value. - def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (IntId (bitconvert - (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>, - Sched<[Sched.Folded]>; -} - -let Predicates = [HasAVX] in { -defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq, - WriteShuffle>, VEX; -defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq, - WriteShuffle>, VEX; -} -let Predicates = [HasAVX2] in { -defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq", int_x86_avx2_pmovsxbq, - WriteShuffle>, VEX, VEX_L; -defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq", int_x86_avx2_pmovzxbq, - WriteShuffle>, VEX, VEX_L; -} -defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq, - WriteShuffle>; -defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq, - WriteShuffle>; - -let Predicates = [HasAVX2] in { - def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>; - def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>; - def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>; - - def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>; - def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>; - - def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>; - - def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))), - (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))), - (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))), - (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - - def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))), - (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))), - (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - - def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))), - (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - - def : Pat<(v8i32 (X86vsext (v8i16 (bitconvert (v2i64 (load addr:$src)))))), - (VPMOVSXWDYrm addr:$src)>; - def : Pat<(v4i64 (X86vsext (v4i32 (bitconvert (v2i64 (load addr:$src)))))), - (VPMOVSXDQYrm addr:$src)>; - - def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64 - (scalar_to_vector (loadi64 addr:$src))))))), - (VPMOVSXBDYrm addr:$src)>; - def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64 - (scalar_to_vector (loadf64 addr:$src))))))), - (VPMOVSXBDYrm addr:$src)>; - - def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64 - (scalar_to_vector (loadi64 addr:$src))))))), - (VPMOVSXWQYrm addr:$src)>; - def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64 - (scalar_to_vector (loadf64 addr:$src))))))), - (VPMOVSXWQYrm addr:$src)>; - - def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32 - (scalar_to_vector (loadi32 addr:$src))))))), - (VPMOVSXBQYrm addr:$src)>; -} - -let Predicates = [HasAVX] in { - // Common patterns involving scalar load - def : Pat<(int_x86_sse41_pmovsxbq - (bitconvert (v4i32 (X86vzmovl - (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (VPMOVSXBQrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxbq - (bitconvert (v4i32 (X86vzmovl - (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (VPMOVZXBQrm addr:$src)>; -} - -let Predicates = [UseSSE41] in { - def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>; - def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>; - def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>; - - def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>; - def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>; - - def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>; - - // Common patterns involving scalar load - def : Pat<(int_x86_sse41_pmovsxbq - (bitconvert (v4i32 (X86vzmovl - (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (PMOVSXBQrm addr:$src)>; - - def : Pat<(int_x86_sse41_pmovzxbq - (bitconvert (v4i32 (X86vzmovl - (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (PMOVZXBQrm addr:$src)>; - - def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64 - (scalar_to_vector (loadi64 addr:$src))))))), - (PMOVSXWDrm addr:$src)>; - def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64 - (scalar_to_vector (loadf64 addr:$src))))))), - (PMOVSXWDrm addr:$src)>; - def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32 - (scalar_to_vector (loadi32 addr:$src))))))), - (PMOVSXBDrm addr:$src)>; - def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32 - (scalar_to_vector (loadi32 addr:$src))))))), - (PMOVSXWQrm addr:$src)>; - def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32 - (scalar_to_vector (extloadi32i16 addr:$src))))))), - (PMOVSXBQrm addr:$src)>; - def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64 - (scalar_to_vector (loadi64 addr:$src))))))), - (PMOVSXDQrm addr:$src)>; - def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64 - (scalar_to_vector (loadf64 addr:$src))))))), - (PMOVSXDQrm addr:$src)>; - def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64 - (scalar_to_vector (loadi64 addr:$src))))))), - (PMOVSXBWrm addr:$src)>; - def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64 - (scalar_to_vector (loadf64 addr:$src))))))), - (PMOVSXBWrm addr:$src)>; +multiclass SS41I_pmovx_rm_all<bits<8> opc, string OpcodeStr, + X86MemOperand MemOp, X86MemOperand MemYOp, + OpndItins SSEItins, OpndItins AVXItins, + OpndItins AVX2Itins> { + defm NAME : SS41I_pmovx_rrrm<opc, OpcodeStr, MemOp, VR128, VR128, SSEItins>; + let Predicates = [HasAVX] in + defm V#NAME : SS41I_pmovx_rrrm<opc, !strconcat("v", OpcodeStr), MemOp, + VR128, VR128, AVXItins>, VEX; + let Predicates = [HasAVX2] in + defm V#NAME#Y : SS41I_pmovx_rrrm<opc, !strconcat("v", OpcodeStr), MemYOp, + VR256, VR128, AVX2Itins>, VEX, VEX_L; +} + +multiclass SS41I_pmovx_rm<bits<8> opc, string OpcodeStr, + X86MemOperand MemOp, X86MemOperand MemYOp> { + defm PMOVSX#NAME : SS41I_pmovx_rm_all<opc, !strconcat("pmovsx", OpcodeStr), + MemOp, MemYOp, + SSE_INTALU_ITINS_SHUFF_P, + DEFAULT_ITINS_SHUFFLESCHED, + DEFAULT_ITINS_SHUFFLESCHED>; + defm PMOVZX#NAME : SS41I_pmovx_rm_all<!add(opc, 0x10), + !strconcat("pmovzx", OpcodeStr), + MemOp, MemYOp, + SSE_INTALU_ITINS_SHUFF_P, + DEFAULT_ITINS_SHUFFLESCHED, + DEFAULT_ITINS_SHUFFLESCHED>; +} + +defm BW : SS41I_pmovx_rm<0x20, "bw", i64mem, i128mem>; +defm WD : SS41I_pmovx_rm<0x23, "wd", i64mem, i128mem>; +defm DQ : SS41I_pmovx_rm<0x25, "dq", i64mem, i128mem>; + +defm BD : SS41I_pmovx_rm<0x21, "bd", i32mem, i64mem>; +defm WQ : SS41I_pmovx_rm<0x24, "wq", i32mem, i64mem>; + +defm BQ : SS41I_pmovx_rm<0x22, "bq", i16mem, i32mem>; + +// AVX2 Patterns +multiclass SS41I_pmovx_avx2_patterns<string OpcPrefix, SDNode ExtOp> { + // Register-Register patterns + def : Pat<(v16i16 (ExtOp (v16i8 VR128:$src))), + (!cast<I>(OpcPrefix#BWYrr) VR128:$src)>; + def : Pat<(v8i32 (ExtOp (v16i8 VR128:$src))), + (!cast<I>(OpcPrefix#BDYrr) VR128:$src)>; + def : Pat<(v4i64 (ExtOp (v16i8 VR128:$src))), + (!cast<I>(OpcPrefix#BQYrr) VR128:$src)>; + + def : Pat<(v8i32 (ExtOp (v8i16 VR128:$src))), + (!cast<I>(OpcPrefix#WDYrr) VR128:$src)>; + def : Pat<(v4i64 (ExtOp (v8i16 VR128:$src))), + (!cast<I>(OpcPrefix#WQYrr) VR128:$src)>; + + def : Pat<(v4i64 (ExtOp (v4i32 VR128:$src))), + (!cast<I>(OpcPrefix#DQYrr) VR128:$src)>; + + // AVX2 Register-Memory patterns + def : Pat<(v16i16 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BWYrm) addr:$src)>; + def : Pat<(v16i16 (ExtOp (v16i8 (vzmovl_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BWYrm) addr:$src)>; + def : Pat<(v16i16 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BWYrm) addr:$src)>; + def : Pat<(v16i16 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BWYrm) addr:$src)>; + + def : Pat<(v8i32 (ExtOp (bc_v16i8 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))), + (!cast<I>(OpcPrefix#BDYrm) addr:$src)>; + def : Pat<(v8i32 (ExtOp (v16i8 (vzmovl_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BDYrm) addr:$src)>; + def : Pat<(v8i32 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BDYrm) addr:$src)>; + def : Pat<(v8i32 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BDYrm) addr:$src)>; + + def : Pat<(v4i64 (ExtOp (bc_v16i8 (v4i32 (scalar_to_vector (loadi32 addr:$src)))))), + (!cast<I>(OpcPrefix#BQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (v16i8 (vzmovl_v4i32 addr:$src)))), + (!cast<I>(OpcPrefix#BQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BQYrm) addr:$src)>; + + def : Pat<(v8i32 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WDYrm) addr:$src)>; + def : Pat<(v8i32 (ExtOp (v8i16 (vzmovl_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WDYrm) addr:$src)>; + def : Pat<(v8i32 (ExtOp (v8i16 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WDYrm) addr:$src)>; + def : Pat<(v8i32 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WDYrm) addr:$src)>; + + def : Pat<(v4i64 (ExtOp (bc_v8i16 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))), + (!cast<I>(OpcPrefix#WQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (v8i16 (vzmovl_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (v8i16 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WQYrm) addr:$src)>; + + def : Pat<(v4i64 (ExtOp (bc_v4i32 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#DQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (v4i32 (vzmovl_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#DQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (v4i32 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#DQYrm) addr:$src)>; + def : Pat<(v4i64 (ExtOp (bc_v4i32 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#DQYrm) addr:$src)>; } let Predicates = [HasAVX2] in { - def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>; - def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>; - def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>; - - def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>; - def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>; - - def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>; - - def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))), - (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))), - (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))), - (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - - def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))), - (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))), - (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; - - def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))), - (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; + defm : SS41I_pmovx_avx2_patterns<"VPMOVSX", X86vsext>; + defm : SS41I_pmovx_avx2_patterns<"VPMOVZX", X86vzext>; +} + +// SSE4.1/AVX patterns. +multiclass SS41I_pmovx_patterns<string OpcPrefix, SDNode ExtOp, + PatFrag ExtLoad16> { + def : Pat<(v8i16 (ExtOp (v16i8 VR128:$src))), + (!cast<I>(OpcPrefix#BWrr) VR128:$src)>; + def : Pat<(v4i32 (ExtOp (v16i8 VR128:$src))), + (!cast<I>(OpcPrefix#BDrr) VR128:$src)>; + def : Pat<(v2i64 (ExtOp (v16i8 VR128:$src))), + (!cast<I>(OpcPrefix#BQrr) VR128:$src)>; + + def : Pat<(v4i32 (ExtOp (v8i16 VR128:$src))), + (!cast<I>(OpcPrefix#WDrr) VR128:$src)>; + def : Pat<(v2i64 (ExtOp (v8i16 VR128:$src))), + (!cast<I>(OpcPrefix#WQrr) VR128:$src)>; + + def : Pat<(v2i64 (ExtOp (v4i32 VR128:$src))), + (!cast<I>(OpcPrefix#DQrr) VR128:$src)>; + + def : Pat<(v8i16 (ExtOp (bc_v16i8 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))), + (!cast<I>(OpcPrefix#BWrm) addr:$src)>; + def : Pat<(v8i16 (ExtOp (bc_v16i8 (v2f64 (scalar_to_vector (loadf64 addr:$src)))))), + (!cast<I>(OpcPrefix#BWrm) addr:$src)>; + def : Pat<(v8i16 (ExtOp (v16i8 (vzmovl_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BWrm) addr:$src)>; + def : Pat<(v8i16 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BWrm) addr:$src)>; + def : Pat<(v8i16 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BWrm) addr:$src)>; + + def : Pat<(v4i32 (ExtOp (bc_v16i8 (v4i32 (scalar_to_vector (loadi32 addr:$src)))))), + (!cast<I>(OpcPrefix#BDrm) addr:$src)>; + def : Pat<(v4i32 (ExtOp (v16i8 (vzmovl_v4i32 addr:$src)))), + (!cast<I>(OpcPrefix#BDrm) addr:$src)>; + def : Pat<(v4i32 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BDrm) addr:$src)>; + def : Pat<(v4i32 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BDrm) addr:$src)>; + + def : Pat<(v2i64 (ExtOp (bc_v16i8 (v4i32 (scalar_to_vector (ExtLoad16 addr:$src)))))), + (!cast<I>(OpcPrefix#BQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (v16i8 (vzmovl_v4i32 addr:$src)))), + (!cast<I>(OpcPrefix#BQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#BQrm) addr:$src)>; + + def : Pat<(v4i32 (ExtOp (bc_v8i16 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))), + (!cast<I>(OpcPrefix#WDrm) addr:$src)>; + def : Pat<(v4i32 (ExtOp (bc_v8i16 (v2f64 (scalar_to_vector (loadf64 addr:$src)))))), + (!cast<I>(OpcPrefix#WDrm) addr:$src)>; + def : Pat<(v4i32 (ExtOp (v8i16 (vzmovl_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WDrm) addr:$src)>; + def : Pat<(v4i32 (ExtOp (v8i16 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WDrm) addr:$src)>; + def : Pat<(v4i32 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WDrm) addr:$src)>; + + def : Pat<(v2i64 (ExtOp (bc_v8i16 (v4i32 (scalar_to_vector (loadi32 addr:$src)))))), + (!cast<I>(OpcPrefix#WQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (v8i16 (vzmovl_v4i32 addr:$src)))), + (!cast<I>(OpcPrefix#WQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (v8i16 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#WQrm) addr:$src)>; + + def : Pat<(v2i64 (ExtOp (bc_v4i32 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))), + (!cast<I>(OpcPrefix#DQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (bc_v4i32 (v2f64 (scalar_to_vector (loadf64 addr:$src)))))), + (!cast<I>(OpcPrefix#DQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (v4i32 (vzmovl_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#DQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (v4i32 (vzload_v2i64 addr:$src)))), + (!cast<I>(OpcPrefix#DQrm) addr:$src)>; + def : Pat<(v2i64 (ExtOp (bc_v4i32 (loadv2i64 addr:$src)))), + (!cast<I>(OpcPrefix#DQrm) addr:$src)>; } let Predicates = [HasAVX] in { - def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>; - def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>; - def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>; - - def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>; - def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>; - - def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>; - - def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))), - (VPMOVZXBWrm addr:$src)>; - def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))), - (VPMOVZXBWrm addr:$src)>; - def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (VPMOVZXBDrm addr:$src)>; - def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))), - (VPMOVZXBQrm addr:$src)>; - - def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))), - (VPMOVZXWDrm addr:$src)>; - def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))), - (VPMOVZXWDrm addr:$src)>; - def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (VPMOVZXWQrm addr:$src)>; - - def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))), - (VPMOVZXDQrm addr:$src)>; - def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))), - (VPMOVZXDQrm addr:$src)>; - def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))), - (VPMOVZXDQrm addr:$src)>; - - def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>; - def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>; - def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>; - - def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>; - def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>; - - def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>; - - def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64 - (scalar_to_vector (loadi64 addr:$src))))))), - (VPMOVSXWDrm addr:$src)>; - def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64 - (scalar_to_vector (loadi64 addr:$src))))))), - (VPMOVSXDQrm addr:$src)>; - def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64 - (scalar_to_vector (loadf64 addr:$src))))))), - (VPMOVSXWDrm addr:$src)>; - def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64 - (scalar_to_vector (loadf64 addr:$src))))))), - (VPMOVSXDQrm addr:$src)>; - def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64 - (scalar_to_vector (loadi64 addr:$src))))))), - (VPMOVSXBWrm addr:$src)>; - def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64 - (scalar_to_vector (loadf64 addr:$src))))))), - (VPMOVSXBWrm addr:$src)>; - - def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32 - (scalar_to_vector (loadi32 addr:$src))))))), - (VPMOVSXBDrm addr:$src)>; - def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32 - (scalar_to_vector (loadi32 addr:$src))))))), - (VPMOVSXWQrm addr:$src)>; - def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32 - (scalar_to_vector (extloadi32i16 addr:$src))))))), - (VPMOVSXBQrm addr:$src)>; + defm : SS41I_pmovx_patterns<"VPMOVSX", X86vsext, extloadi32i16>; + defm : SS41I_pmovx_patterns<"VPMOVZX", X86vzext, loadi16_anyext>; } let Predicates = [UseSSE41] in { - def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>; - def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>; - def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>; - - def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>; - def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>; - - def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>; - - def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))), - (PMOVZXBWrm addr:$src)>; - def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))), - (PMOVZXBWrm addr:$src)>; - def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (PMOVZXBDrm addr:$src)>; - def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))), - (PMOVZXBQrm addr:$src)>; - - def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))), - (PMOVZXWDrm addr:$src)>; - def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))), - (PMOVZXWDrm addr:$src)>; - def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (PMOVZXWQrm addr:$src)>; - - def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))), - (PMOVZXDQrm addr:$src)>; - def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))), - (PMOVZXDQrm addr:$src)>; - def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))), - (PMOVZXDQrm addr:$src)>; + defm : SS41I_pmovx_patterns<"PMOVSX", X86vsext, extloadi32i16>; + defm : SS41I_pmovx_patterns<"PMOVZX", X86vzext, loadi16_anyext>; } //===----------------------------------------------------------------------===// diff --git a/llvm/lib/Target/X86/X86IntrinsicsInfo.h b/llvm/lib/Target/X86/X86IntrinsicsInfo.h index d4b1b5e0db3..f20ef4b21a1 100644 --- a/llvm/lib/Target/X86/X86IntrinsicsInfo.h +++ b/llvm/lib/Target/X86/X86IntrinsicsInfo.h @@ -138,6 +138,18 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx2_pminu_b, INTR_TYPE_2OP, X86ISD::UMIN, 0), X86_INTRINSIC_DATA(avx2_pminu_d, INTR_TYPE_2OP, X86ISD::UMIN, 0), X86_INTRINSIC_DATA(avx2_pminu_w, INTR_TYPE_2OP, X86ISD::UMIN, 0), + X86_INTRINSIC_DATA(avx2_pmovsxbd, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovsxbq, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovsxbw, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovsxdq, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovsxwd, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovsxwq, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovzxbd, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovzxbq, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovzxbw, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovzxdq, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovzxwd, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(avx2_pmovzxwq, INTR_TYPE_1OP, X86ISD::VZEXT, 0), X86_INTRINSIC_DATA(avx2_psll_d, INTR_TYPE_2OP, X86ISD::VSHL, 0), X86_INTRINSIC_DATA(avx2_psll_q, INTR_TYPE_2OP, X86ISD::VSHL, 0), X86_INTRINSIC_DATA(avx2_psll_w, INTR_TYPE_2OP, X86ISD::VSHL, 0), @@ -284,6 +296,18 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(sse41_pminsd, INTR_TYPE_2OP, X86ISD::SMIN, 0), X86_INTRINSIC_DATA(sse41_pminud, INTR_TYPE_2OP, X86ISD::UMIN, 0), X86_INTRINSIC_DATA(sse41_pminuw, INTR_TYPE_2OP, X86ISD::UMIN, 0), + X86_INTRINSIC_DATA(sse41_pmovsxbd, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovsxbq, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovsxbw, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovsxdq, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovsxwd, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovsxwq, INTR_TYPE_1OP, X86ISD::VSEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovzxbd, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovzxbq, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovzxbw, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovzxdq, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovzxwd, INTR_TYPE_1OP, X86ISD::VZEXT, 0), + X86_INTRINSIC_DATA(sse41_pmovzxwq, INTR_TYPE_1OP, X86ISD::VZEXT, 0), X86_INTRINSIC_DATA(sse_comieq_ss, COMI, X86ISD::COMI, ISD::SETEQ), X86_INTRINSIC_DATA(sse_comige_ss, COMI, X86ISD::COMI, ISD::SETGE), X86_INTRINSIC_DATA(sse_comigt_ss, COMI, X86ISD::COMI, ISD::SETGT), |

