forked from OSchip/llvm-project
[X86][SSE] Use SSE_PACK OpndItins in PACKSS/PACKUS instruction definitions
Update multi-classes to take the scheduling OpndItins instead of hard coding it. SSE_PACK will be reused in the AVX512 equivalents. llvm-svn: 319243
This commit is contained in:
Simon Pilgrim
parent
80fb55625b
commit
1bc7b0e148
|
|
@ -3935,8 +3935,8 @@ defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw,
|
|||
|
||||
let ExeDomain = SSEPackedInt in {
|
||||
multiclass sse2_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
|
||||
ValueType ArgVT, SDNode OpNode, PatFrag ld_frag,
|
||||
bit Is2Addr = 1> {
|
||||
ValueType ArgVT, SDNode OpNode, OpndItins itins,
|
||||
PatFrag ld_frag, bit Is2Addr = 1> {
|
||||
def rr : PDI<opc, MRMSrcReg,
|
||||
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
|
||||
!if(Is2Addr,
|
||||
|
|
@ -3945,7 +3945,7 @@ multiclass sse2_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
|
|||
"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
|
||||
[(set VR128:$dst,
|
||||
(OutVT (OpNode (ArgVT VR128:$src1), VR128:$src2)))],
|
||||
IIC_SSE_PACK>, Sched<[WriteShuffle]>;
|
||||
itins.rr>, Sched<[itins.Sched]>;
|
||||
def rm : PDI<opc, MRMSrcMem,
|
||||
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
|
||||
!if(Is2Addr,
|
||||
|
|
@ -3955,31 +3955,31 @@ multiclass sse2_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
|
|||
[(set VR128:$dst,
|
||||
(OutVT (OpNode (ArgVT VR128:$src1),
|
||||
(bitconvert (ld_frag addr:$src2)))))],
|
||||
IIC_SSE_PACK>, Sched<[WriteShuffleLd, ReadAfterLd]>;
|
||||
itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
|
||||
}
|
||||
|
||||
multiclass sse2_pack_y<bits<8> opc, string OpcodeStr, ValueType OutVT,
|
||||
ValueType ArgVT, SDNode OpNode> {
|
||||
ValueType ArgVT, SDNode OpNode, OpndItins itins> {
|
||||
def Yrr : PDI<opc, MRMSrcReg,
|
||||
(outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
|
||||
!strconcat(OpcodeStr,
|
||||
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
|
||||
[(set VR256:$dst,
|
||||
(OutVT (OpNode (ArgVT VR256:$src1), VR256:$src2)))]>,
|
||||
Sched<[WriteShuffle]>;
|
||||
(OutVT (OpNode (ArgVT VR256:$src1), VR256:$src2)))],
|
||||
itins.rr>, Sched<[itins.Sched]>;
|
||||
def Yrm : PDI<opc, MRMSrcMem,
|
||||
(outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
|
||||
!strconcat(OpcodeStr,
|
||||
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
|
||||
[(set VR256:$dst,
|
||||
(OutVT (OpNode (ArgVT VR256:$src1),
|
||||
(bitconvert (loadv4i64 addr:$src2)))))]>,
|
||||
Sched<[WriteShuffleLd, ReadAfterLd]>;
|
||||
(bitconvert (loadv4i64 addr:$src2)))))],
|
||||
itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
|
||||
}
|
||||
|
||||
multiclass sse4_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
|
||||
ValueType ArgVT, SDNode OpNode, PatFrag ld_frag,
|
||||
bit Is2Addr = 1> {
|
||||
ValueType ArgVT, SDNode OpNode, OpndItins itins,
|
||||
PatFrag ld_frag, bit Is2Addr = 1> {
|
||||
def rr : SS48I<opc, MRMSrcReg,
|
||||
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
|
||||
!if(Is2Addr,
|
||||
|
|
@ -3988,7 +3988,7 @@ multiclass sse4_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
|
|||
"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
|
||||
[(set VR128:$dst,
|
||||
(OutVT (OpNode (ArgVT VR128:$src1), VR128:$src2)))],
|
||||
IIC_SSE_PACK>, Sched<[WriteShuffle]>;
|
||||
itins.rr>, Sched<[itins.Sched]>;
|
||||
def rm : SS48I<opc, MRMSrcMem,
|
||||
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
|
||||
!if(Is2Addr,
|
||||
|
|
@ -3998,62 +3998,62 @@ multiclass sse4_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
|
|||
[(set VR128:$dst,
|
||||
(OutVT (OpNode (ArgVT VR128:$src1),
|
||||
(bitconvert (ld_frag addr:$src2)))))],
|
||||
IIC_SSE_PACK>, Sched<[WriteShuffleLd, ReadAfterLd]>;
|
||||
itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
|
||||
}
|
||||
|
||||
multiclass sse4_pack_y<bits<8> opc, string OpcodeStr, ValueType OutVT,
|
||||
ValueType ArgVT, SDNode OpNode> {
|
||||
ValueType ArgVT, SDNode OpNode, OpndItins itins> {
|
||||
def Yrr : SS48I<opc, MRMSrcReg,
|
||||
(outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
|
||||
!strconcat(OpcodeStr,
|
||||
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
|
||||
[(set VR256:$dst,
|
||||
(OutVT (OpNode (ArgVT VR256:$src1), VR256:$src2)))]>,
|
||||
Sched<[WriteShuffle]>;
|
||||
(OutVT (OpNode (ArgVT VR256:$src1), VR256:$src2)))],
|
||||
itins.rr>, Sched<[itins.Sched]>;
|
||||
def Yrm : SS48I<opc, MRMSrcMem,
|
||||
(outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
|
||||
!strconcat(OpcodeStr,
|
||||
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
|
||||
[(set VR256:$dst,
|
||||
(OutVT (OpNode (ArgVT VR256:$src1),
|
||||
(bitconvert (loadv4i64 addr:$src2)))))]>,
|
||||
Sched<[WriteShuffleLd, ReadAfterLd]>;
|
||||
(bitconvert (loadv4i64 addr:$src2)))))],
|
||||
itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
|
||||
}
|
||||
|
||||
let Predicates = [HasAVX, NoVLX_Or_NoBWI] in {
|
||||
defm VPACKSSWB : sse2_pack<0x63, "vpacksswb", v16i8, v8i16, X86Packss,
|
||||
loadv2i64, 0>, VEX_4V, VEX_WIG;
|
||||
SSE_PACK, loadv2i64, 0>, VEX_4V, VEX_WIG;
|
||||
defm VPACKSSDW : sse2_pack<0x6B, "vpackssdw", v8i16, v4i32, X86Packss,
|
||||
loadv2i64, 0>, VEX_4V, VEX_WIG;
|
||||
SSE_PACK, loadv2i64, 0>, VEX_4V, VEX_WIG;
|
||||
|
||||
defm VPACKUSWB : sse2_pack<0x67, "vpackuswb", v16i8, v8i16, X86Packus,
|
||||
loadv2i64, 0>, VEX_4V, VEX_WIG;
|
||||
SSE_PACK, loadv2i64, 0>, VEX_4V, VEX_WIG;
|
||||
defm VPACKUSDW : sse4_pack<0x2B, "vpackusdw", v8i16, v4i32, X86Packus,
|
||||
loadv2i64, 0>, VEX_4V;
|
||||
SSE_PACK, loadv2i64, 0>, VEX_4V;
|
||||
}
|
||||
|
||||
let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
|
||||
defm VPACKSSWB : sse2_pack_y<0x63, "vpacksswb", v32i8, v16i16, X86Packss>,
|
||||
defm VPACKSSWB : sse2_pack_y<0x63, "vpacksswb", v32i8, v16i16, X86Packss, SSE_PACK>,
|
||||
VEX_4V, VEX_L, VEX_WIG;
|
||||
defm VPACKSSDW : sse2_pack_y<0x6B, "vpackssdw", v16i16, v8i32, X86Packss>,
|
||||
defm VPACKSSDW : sse2_pack_y<0x6B, "vpackssdw", v16i16, v8i32, X86Packss, SSE_PACK>,
|
||||
VEX_4V, VEX_L, VEX_WIG;
|
||||
|
||||
defm VPACKUSWB : sse2_pack_y<0x67, "vpackuswb", v32i8, v16i16, X86Packus>,
|
||||
defm VPACKUSWB : sse2_pack_y<0x67, "vpackuswb", v32i8, v16i16, X86Packus, SSE_PACK>,
|
||||
VEX_4V, VEX_L, VEX_WIG;
|
||||
defm VPACKUSDW : sse4_pack_y<0x2B, "vpackusdw", v16i16, v8i32, X86Packus>,
|
||||
defm VPACKUSDW : sse4_pack_y<0x2B, "vpackusdw", v16i16, v8i32, X86Packus, SSE_PACK>,
|
||||
VEX_4V, VEX_L;
|
||||
}
|
||||
|
||||
let Constraints = "$src1 = $dst" in {
|
||||
defm PACKSSWB : sse2_pack<0x63, "packsswb", v16i8, v8i16, X86Packss,
|
||||
defm PACKSSWB : sse2_pack<0x63, "packsswb", v16i8, v8i16, X86Packss, SSE_PACK,
|
||||
memopv2i64>;
|
||||
defm PACKSSDW : sse2_pack<0x6B, "packssdw", v8i16, v4i32, X86Packss,
|
||||
defm PACKSSDW : sse2_pack<0x6B, "packssdw", v8i16, v4i32, X86Packss, SSE_PACK,
|
||||
memopv2i64>;
|
||||
|
||||
defm PACKUSWB : sse2_pack<0x67, "packuswb", v16i8, v8i16, X86Packus,
|
||||
defm PACKUSWB : sse2_pack<0x67, "packuswb", v16i8, v8i16, X86Packus, SSE_PACK,
|
||||
memopv2i64>;
|
||||
|
||||
defm PACKUSDW : sse4_pack<0x2B, "packusdw", v8i16, v4i32, X86Packus,
|
||||
defm PACKUSDW : sse4_pack<0x2B, "packusdw", v8i16, v4i32, X86Packus, SSE_PACK,
|
||||
memopv2i64>;
|
||||
}
|
||||
} // ExeDomain = SSEPackedInt
|
||||
|
|
|
|||
Loading…
Reference in New Issue