VTName)), VTName));
PatFrag ScalarLdFrag = !cast<PatFrag>("load" # EltVT);
- // Load patterns used for memory operands. We only have this defined in
- // case of i64 element types for sub-512 integer vectors. For now, keep
- // MemOpFrag undefined in these cases.
- PatFrag MemOpFrag =
- !if (!eq (NumElts#EltTypeName, "1f32"), !cast<PatFrag>("memopfsf32"),
- !if (!eq (NumElts#EltTypeName, "1f64"), !cast<PatFrag>("memopfsf64"),
- !if (!eq (TypeVariantName, "f"), !cast<PatFrag>("memop" # VTName),
- !if (!eq (EltTypeName, "i64"), !cast<PatFrag>("memop" # VTName),
- !if (!eq (VTName, "v16i32"), !cast<PatFrag>("memop" # VTName), ?)))));
-
// The corresponding float type, e.g. v16f32 for v16i32
// Note: For EltSize < 32, FloatVT is illegal and TableGen
// fails to compile, so we choose FloatVT = VT
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set _.RC:$dst,
- (_.VT (OpNode (_.MemOpFrag addr:$src1),
+ (_.VT (OpNode (_.LdFrag addr:$src1),
(i8 imm:$src2))))]>,
EVEX, EVEX_CD8<_.EltSize, CD8VF>;
}
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set _.RC:$dst,
(_.VT (X86VPermilpv _.RC:$src1,
- (Ctrl.VT (Ctrl.MemOpFrag addr:$src2)))))]>,
+ (Ctrl.VT (Ctrl.LdFrag addr:$src2)))))]>,
EVEX_4V;
}
}
EVEX_4V;
}
-defm VPERMDZ : avx512_perm<0x36, "vpermd", VR512, memopv16i32, i512mem,
+defm VPERMDZ : avx512_perm<0x36, "vpermd", VR512, loadv16i32, i512mem,
v16i32>, EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VPERMQZ : avx512_perm<0x36, "vpermq", VR512, memopv8i64, i512mem,
+defm VPERMQZ : avx512_perm<0x36, "vpermq", VR512, loadv8i64, i512mem,
v8i64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
let ExeDomain = SSEPackedSingle in
-defm VPERMPSZ : avx512_perm<0x16, "vpermps", VR512, memopv16f32, f512mem,
+defm VPERMPSZ : avx512_perm<0x16, "vpermps", VR512, loadv16f32, f512mem,
v16f32>, EVEX_V512, EVEX_CD8<32, CD8VF>;
let ExeDomain = SSEPackedDouble in
-defm VPERMPDZ : avx512_perm<0x16, "vpermpd", VR512, memopv8f64, f512mem,
+defm VPERMPDZ : avx512_perm<0x16, "vpermpd", VR512, loadv8f64, f512mem,
v8f64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
// -- VPERM2I - 3 source operands form --
EVEX_4V, EVEX_KZ;
}
}
-defm VPERMI2D : avx512_perm_3src<0x76, "vpermi2d", VR512, memopv16i32,
+defm VPERMI2D : avx512_perm_3src<0x76, "vpermi2d", VR512, loadv16i32,
i512mem, X86VPermiv3, v16i32, VK16WM>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VPERMI2Q : avx512_perm_3src<0x76, "vpermi2q", VR512, memopv8i64,
+defm VPERMI2Q : avx512_perm_3src<0x76, "vpermi2q", VR512, loadv8i64,
i512mem, X86VPermiv3, v8i64, VK8WM>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
-defm VPERMI2PS : avx512_perm_3src<0x77, "vpermi2ps", VR512, memopv16f32,
+defm VPERMI2PS : avx512_perm_3src<0x77, "vpermi2ps", VR512, loadv16f32,
i512mem, X86VPermiv3, v16f32, VK16WM>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VPERMI2PD : avx512_perm_3src<0x77, "vpermi2pd", VR512, memopv8f64,
+defm VPERMI2PD : avx512_perm_3src<0x77, "vpermi2pd", VR512, loadv8f64,
i512mem, X86VPermiv3, v8f64, VK8WM>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
(MaskVT (COPY_TO_REGCLASS MRC:$mask, KRC)), VR512:$idx, VR512:$src2)>;
}
-defm VPERMT2D : avx512_perm_table_3src<0x7E, "d", VR512, memopv16i32, i512mem,
+defm VPERMT2D : avx512_perm_table_3src<0x7E, "d", VR512, loadv16i32, i512mem,
X86VPermv3, v16i32, VK16WM, v16i1, GR16>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VPERMT2Q : avx512_perm_table_3src<0x7E, "q", VR512, memopv8i64, i512mem,
+defm VPERMT2Q : avx512_perm_table_3src<0x7E, "q", VR512, loadv8i64, i512mem,
X86VPermv3, v8i64, VK8WM, v8i1, GR8>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
-defm VPERMT2PS : avx512_perm_table_3src<0x7F, "ps", VR512, memopv16f32, i512mem,
+defm VPERMT2PS : avx512_perm_table_3src<0x7F, "ps", VR512, loadv16f32, i512mem,
X86VPermv3, v16f32, VK16WM, v16i1, GR16>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VPERMT2PD : avx512_perm_table_3src<0x7F, "pd", VR512, memopv8f64, i512mem,
+defm VPERMT2PD : avx512_perm_table_3src<0x7F, "pd", VR512, loadv8f64, i512mem,
X86VPermv3, v8f64, VK8WM, v8i1, GR8>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
multiclass avx512_icmp_cc<bits<8> opc, string Suffix, SDNode OpNode,
X86VectorVTInfo _> {
def rri : AVX512AIi8<opc, MRMSrcReg,
- (outs _.KRC:$dst), (ins _.RC:$src1, _.RC:$src2, AVXCC:$cc),
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.RC:$src2, AVX512ICC:$cc),
!strconcat("vpcmp${cc}", Suffix,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set _.KRC:$dst, (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2),
IIC_SSE_ALU_F32P_RR>, EVEX_4V;
let mayLoad = 1 in
def rmi : AVX512AIi8<opc, MRMSrcMem,
- (outs _.KRC:$dst), (ins _.RC:$src1, _.MemOp:$src2, AVXCC:$cc),
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.MemOp:$src2, AVX512ICC:$cc),
!strconcat("vpcmp${cc}", Suffix,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set _.KRC:$dst, (OpNode (_.VT _.RC:$src1),
IIC_SSE_ALU_F32P_RM>, EVEX_4V;
def rrik : AVX512AIi8<opc, MRMSrcReg,
(outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1, _.RC:$src2,
- AVXCC:$cc),
+ AVX512ICC:$cc),
!strconcat("vpcmp${cc}", Suffix,
"\t{$src2, $src1, $dst {${mask}}|",
"$dst {${mask}}, $src1, $src2}"),
let mayLoad = 1 in
def rmik : AVX512AIi8<opc, MRMSrcMem,
(outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1, _.MemOp:$src2,
- AVXCC:$cc),
+ AVX512ICC:$cc),
!strconcat("vpcmp${cc}", Suffix,
"\t{$src2, $src1, $dst {${mask}}|",
"$dst {${mask}}, $src1, $src2}"),
avx512_icmp_cc<opc, Suffix, OpNode, _> {
def rmib : AVX512AIi8<opc, MRMSrcMem,
(outs _.KRC:$dst), (ins _.RC:$src1, _.ScalarMemOp:$src2,
- AVXCC:$cc),
+ AVX512ICC:$cc),
!strconcat("vpcmp${cc}", Suffix,
"\t{${src2}", _.BroadcastStr, ", $src1, $dst|",
"$dst, $src1, ${src2}", _.BroadcastStr, "}"),
IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_B;
def rmibk : AVX512AIi8<opc, MRMSrcMem,
(outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1,
- _.ScalarMemOp:$src2, AVXCC:$cc),
+ _.ScalarMemOp:$src2, AVX512ICC:$cc),
!strconcat("vpcmp${cc}", Suffix,
"\t{${src2}", _.BroadcastStr, ", $src1, $dst {${mask}}|",
"$dst {${mask}}, $src1, ${src2}", _.BroadcastStr, "}"),
!strconcat("vcmp${cc}", suffix,
"\t{$src2, $src1, $dst|$dst, $src1, $src2, $cc}"),
[(set KRC:$dst,
- (X86cmpm (vt RC:$src1), (memop addr:$src2), imm:$cc))], d>;
+ (X86cmpm (vt RC:$src1), (load addr:$src2), imm:$cc))], d>;
// Accept explicit immediate argument form instead of comparison code.
let isAsmParserOnly = 1, hasSideEffects = 0 in {
//
multiclass avx512_mask_mov<bits<8> opc_kk, bits<8> opc_km, bits<8> opc_mk,
string OpcodeStr, RegisterClass KRC,
- ValueType vvt, ValueType ivt, X86MemOperand x86memop> {
+ ValueType vvt, X86MemOperand x86memop> {
let hasSideEffects = 0 in {
def kk : I<opc_kk, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
let mayLoad = 1 in
def km : I<opc_km, MRMSrcMem, (outs KRC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set KRC:$dst, (vvt (bitconvert (ivt (load addr:$src)))))]>;
+ [(set KRC:$dst, (vvt (load addr:$src)))]>;
let mayStore = 1 in
def mk : I<opc_mk, MRMDestMem, (outs), (ins x86memop:$dst, KRC:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(store KRC:$src, addr:$dst)]>;
}
}
}
let Predicates = [HasDQI] in
- defm KMOVB : avx512_mask_mov<0x90, 0x90, 0x91, "kmovb", VK8, v8i1, i8,
- i8mem>,
+ defm KMOVB : avx512_mask_mov<0x90, 0x90, 0x91, "kmovb", VK8, v8i1, i8mem>,
avx512_mask_mov_gpr<0x92, 0x93, "kmovb", VK8, GR32>,
VEX, PD;
let Predicates = [HasAVX512] in
- defm KMOVW : avx512_mask_mov<0x90, 0x90, 0x91, "kmovw", VK16, v16i1, i16,
- i16mem>,
+ defm KMOVW : avx512_mask_mov<0x90, 0x90, 0x91, "kmovw", VK16, v16i1, i16mem>,
avx512_mask_mov_gpr<0x92, 0x93, "kmovw", VK16, GR32>,
VEX, PS;
let Predicates = [HasBWI] in {
- defm KMOVD : avx512_mask_mov<0x90, 0x90, 0x91, "kmovd", VK32, v32i1, i32,
- i32mem>, VEX, PD, VEX_W;
+ defm KMOVD : avx512_mask_mov<0x90, 0x90, 0x91, "kmovd", VK32, v32i1,i32mem>,
+ VEX, PD, VEX_W;
defm KMOVD : avx512_mask_mov_gpr<0x92, 0x93, "kmovd", VK32, GR32>,
VEX, XD;
}
let Predicates = [HasBWI] in {
- defm KMOVQ : avx512_mask_mov<0x90, 0x90, 0x91, "kmovq", VK64, v64i1, i64,
- i64mem>, VEX, PS, VEX_W;
+ defm KMOVQ : avx512_mask_mov<0x90, 0x90, 0x91, "kmovq", VK64, v64i1, i64mem>,
+ VEX, PS, VEX_W;
defm KMOVQ : avx512_mask_mov_gpr<0x92, 0x93, "kmovq", VK64, GR64>,
VEX, XD, VEX_W;
}
let Predicates = [HasDQI] in {
def : Pat<(store (i8 (bitconvert (v8i1 VK8:$src))), addr:$dst),
(KMOVBmk addr:$dst, VK8:$src)>;
+ def : Pat<(v8i1 (bitconvert (i8 (load addr:$src)))),
+ (KMOVBkm addr:$src)>;
+}
+let Predicates = [HasAVX512, NoDQI] in {
+ def : Pat<(store (i8 (bitconvert (v8i1 VK8:$src))), addr:$dst),
+ (KMOVWmk addr:$dst, (COPY_TO_REGCLASS VK8:$src, VK16))>;
+ def : Pat<(v8i1 (bitconvert (i8 (load addr:$src)))),
+ (COPY_TO_REGCLASS (KMOVWkm addr:$src), VK8)>;
}
let Predicates = [HasAVX512] in {
def : Pat<(store (i16 (bitconvert (v16i1 VK16:$src))), addr:$dst),
(KMOVWmk addr:$dst, VK16:$src)>;
- def : Pat<(store (i8 (bitconvert (v8i1 VK8:$src))), addr:$dst),
- (KMOVWmk addr:$dst, (COPY_TO_REGCLASS VK8:$src, VK16))>;
def : Pat<(i1 (load addr:$src)),
(COPY_TO_REGCLASS (KMOVWkm addr:$src), VK1)>;
- def : Pat<(v8i1 (bitconvert (i8 (load addr:$src)))),
- (COPY_TO_REGCLASS (KMOVWkm addr:$src), VK8)>;
+ def : Pat<(v16i1 (bitconvert (i16 (load addr:$src)))),
+ (KMOVWkm addr:$src)>;
}
let Predicates = [HasBWI] in {
def : Pat<(store (i32 (bitconvert (v32i1 VK32:$src))), addr:$dst),
(KMOVDmk addr:$dst, VK32:$src)>;
+ def : Pat<(v32i1 (bitconvert (i32 (load addr:$src)))),
+ (KMOVDkm addr:$src)>;
}
let Predicates = [HasBWI] in {
def : Pat<(store (i64 (bitconvert (v64i1 VK64:$src))), addr:$dst),
(KMOVQmk addr:$dst, VK64:$src)>;
+ def : Pat<(v64i1 (bitconvert (i64 (load addr:$src)))),
+ (KMOVQkm addr:$src)>;
}
let Predicates = [HasAVX512] in {
def : Pat<(xor VK64:$src1, (v64i1 immAllOnesV)), (KNOTQrr VK64:$src1)>;
// KNL does not support KMOVB, 8-bit mask is promoted to 16-bit
-let Predicates = [HasAVX512] in {
+let Predicates = [HasAVX512, NoDQI] in {
def : Pat<(xor VK8:$src1, (v8i1 immAllOnesV)),
(COPY_TO_REGCLASS (KNOTWrr (COPY_TO_REGCLASS VK8:$src1, VK16)), VK8)>;
multiclass avx512_mask_testop_w<bits<8> opc, string OpcodeStr, SDNode OpNode> {
defm W : avx512_mask_testop<opc, !strconcat(OpcodeStr, "w"), VK16, OpNode>,
VEX, PS;
+ let Predicates = [HasDQI] in
+ defm B : avx512_mask_testop<opc, !strconcat(OpcodeStr, "b"), VK8, OpNode>,
+ VEX, PD;
+ let Predicates = [HasBWI] in {
+ defm Q : avx512_mask_testop<opc, !strconcat(OpcodeStr, "q"), VK64, OpNode>,
+ VEX, PS, VEX_W;
+ defm D : avx512_mask_testop<opc, !strconcat(OpcodeStr, "d"), VK32, OpNode>,
+ VEX, PD, VEX_W;
+ }
}
defm KORTEST : avx512_mask_testop_w<0x98, "kortest", X86kortest>;
-def : Pat<(X86cmp VK1:$src1, (i1 0)),
- (KORTESTWrr (COPY_TO_REGCLASS VK1:$src1, VK16),
- (COPY_TO_REGCLASS VK1:$src1, VK16))>;
-
// Mask shift
multiclass avx512_mask_shiftop<bits<8> opc, string OpcodeStr, RegisterClass KRC,
SDNode OpNode> {
multiclass avx512_mask_shiftop_w<bits<8> opc1, bits<8> opc2, string OpcodeStr,
SDNode OpNode> {
defm W : avx512_mask_shiftop<opc1, !strconcat(OpcodeStr, "w"), VK16, OpNode>,
- VEX, TAPD, VEX_W;
+ VEX, TAPD, VEX_W;
+ let Predicates = [HasDQI] in
+ defm B : avx512_mask_shiftop<opc1, !strconcat(OpcodeStr, "b"), VK8, OpNode>,
+ VEX, TAPD;
+ let Predicates = [HasBWI] in {
+ defm Q : avx512_mask_shiftop<opc2, !strconcat(OpcodeStr, "q"), VK64, OpNode>,
+ VEX, TAPD, VEX_W;
+ let Predicates = [HasDQI] in
+ defm D : avx512_mask_shiftop<opc2, !strconcat(OpcodeStr, "d"), VK32, OpNode>,
+ VEX, TAPD;
+ }
}
defm KSHIFTL : avx512_mask_shiftop_w<0x32, 0x33, "kshiftl", X86vshli>;
}
def : Pat<(v8i1 (X86vshli VK8:$src, (i8 imm:$imm))),
- (v8i1 (COPY_TO_REGCLASS (KSHIFTLWri (COPY_TO_REGCLASS VK8:$src, VK16), (I8Imm $imm)), VK8))>;
+ (v8i1 (COPY_TO_REGCLASS
+ (KSHIFTLWri (COPY_TO_REGCLASS VK8:$src, VK16),
+ (I8Imm $imm)), VK8))>, Requires<[HasAVX512, NoDQI]>;
def : Pat<(v8i1 (X86vsrli VK8:$src, (i8 imm:$imm))),
- (v8i1 (COPY_TO_REGCLASS (KSHIFTRWri (COPY_TO_REGCLASS VK8:$src, VK16), (I8Imm $imm)), VK8))>;
+ (v8i1 (COPY_TO_REGCLASS
+ (KSHIFTRWri (COPY_TO_REGCLASS VK8:$src, VK16),
+ (I8Imm $imm)), VK8))>, Requires<[HasAVX512, NoDQI]>;
//===----------------------------------------------------------------------===//
// AVX-512 - Aligned and unaligned load and store
//
SSE_INTALU_ITINS_P, HasDQI, 1>, T8PD;
defm VPMULDQZ : avx512_binop_rm2<0x28, "vpmuldq", v8i64, v16i32, VK8WM, VR512,
- memopv8i64, i512mem, loadi64, i64mem, "{1to8}",
+ loadv8i64, i512mem, loadi64, i64mem, "{1to8}",
SSE_INTALU_ITINS_P, 1>, T8PD, EVEX_V512,
EVEX_CD8<64, CD8VF>, VEX_W;
defm VPMULUDQZ : avx512_binop_rm2<0xF4, "vpmuludq", v8i64, v16i32, VK8WM, VR512,
- memopv8i64, i512mem, loadi64, i64mem, "{1to8}",
+ loadv8i64, i512mem, loadi64, i64mem, "{1to8}",
SSE_INTMUL_ITINS_P, 1>, EVEX_V512, EVEX_CD8<64, CD8VF>, VEX_W;
def : Pat<(v8i64 (X86pmuludq (v16i32 VR512:$src1), (v16i32 VR512:$src2))),
d>, EVEX_4V;
}
-defm VUNPCKHPSZ: avx512_unpack_fp<0x15, X86Unpckh, v16f32, memopv8f64,
+defm VUNPCKHPSZ: avx512_unpack_fp<0x15, X86Unpckh, v16f32, loadv8f64,
VR512, f512mem, "vunpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
SSEPackedSingle>, PS, EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VUNPCKHPDZ: avx512_unpack_fp<0x15, X86Unpckh, v8f64, memopv8f64,
+defm VUNPCKHPDZ: avx512_unpack_fp<0x15, X86Unpckh, v8f64, loadv8f64,
VR512, f512mem, "vunpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
SSEPackedDouble>, PD, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
-defm VUNPCKLPSZ: avx512_unpack_fp<0x14, X86Unpckl, v16f32, memopv8f64,
+defm VUNPCKLPSZ: avx512_unpack_fp<0x14, X86Unpckl, v16f32, loadv8f64,
VR512, f512mem, "vunpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
SSEPackedSingle>, PS, EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VUNPCKLPDZ: avx512_unpack_fp<0x14, X86Unpckl, v8f64, memopv8f64,
+defm VUNPCKLPDZ: avx512_unpack_fp<0x14, X86Unpckl, v8f64, loadv8f64,
VR512, f512mem, "vunpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
SSEPackedDouble>, PD, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
IIC_SSE_UNPCK>, EVEX_4V;
}
defm VPUNPCKLDQZ : avx512_unpack_int<0x62, "vpunpckldq", X86Unpckl, v16i32,
- VR512, memopv16i32, i512mem>, EVEX_V512,
+ VR512, loadv16i32, i512mem>, EVEX_V512,
EVEX_CD8<32, CD8VF>;
defm VPUNPCKLQDQZ : avx512_unpack_int<0x6C, "vpunpcklqdq", X86Unpckl, v8i64,
- VR512, memopv8i64, i512mem>, EVEX_V512,
+ VR512, loadv8i64, i512mem>, EVEX_V512,
VEX_W, EVEX_CD8<64, CD8VF>;
defm VPUNPCKHDQZ : avx512_unpack_int<0x6A, "vpunpckhdq", X86Unpckh, v16i32,
- VR512, memopv16i32, i512mem>, EVEX_V512,
+ VR512, loadv16i32, i512mem>, EVEX_V512,
EVEX_CD8<32, CD8VF>;
defm VPUNPCKHQDQZ : avx512_unpack_int<0x6D, "vpunpckhqdq", X86Unpckh, v8i64,
- VR512, memopv8i64, i512mem>, EVEX_V512,
+ VR512, loadv8i64, i512mem>, EVEX_V512,
VEX_W, EVEX_CD8<64, CD8VF>;
//===----------------------------------------------------------------------===//
// AVX-512 - PSHUFD
(i8 imm:$src2))))]>, EVEX;
}
-defm VPSHUFDZ : avx512_pshuf_imm<0x70, "vpshufd", VR512, X86PShufd, memopv16i32,
+defm VPSHUFDZ : avx512_pshuf_imm<0x70, "vpshufd", VR512, X86PShufd, loadv16i32,
i512mem, v16i32>, PD, EVEX_V512, EVEX_CD8<32, CD8VF>;
//===----------------------------------------------------------------------===//
}//let mayLoad = 1
}
-multiclass avx512_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
+multiclass avx512_fp_round_packed<bits<8> opc, string OpcodeStr, SDNode OpNodeRnd,
+ X86VectorVTInfo _, bit IsCommutable> {
+ defm rb: AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.RC:$src2, AVX512RC:$rc), OpcodeStr##_.Suffix,
+ "$rc, $src2, $src1", "$src1, $src2, $rc",
+ (_.VT (OpNodeRnd _.RC:$src1, _.RC:$src2, (i32 imm:$rc)))>,
+ EVEX_4V, EVEX_B, EVEX_RC;
+}
+
+multiclass avx512_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
bit IsCommutable = 0> {
defm PSZ : avx512_fp_packed<opc, OpcodeStr, OpNode, v16f32_info,
IsCommutable>, EVEX_V512, PS,
}
}
-defm VADD : avx512_fp_binop_p<0x58, "vadd", fadd, 1>;
-defm VMUL : avx512_fp_binop_p<0x59, "vmul", fmul, 1>;
+multiclass avx512_fp_binop_p_round<bits<8> opc, string OpcodeStr, SDNode OpNodeRnd> {
+ defm PSZ : avx512_fp_round_packed<opc, OpcodeStr, OpNodeRnd, v16f32_info, 0>,
+ EVEX_V512, PS, EVEX_CD8<32, CD8VF>;
+ defm PDZ : avx512_fp_round_packed<opc, OpcodeStr, OpNodeRnd, v8f64_info, 0>,
+ EVEX_V512, PD, VEX_W,EVEX_CD8<64, CD8VF>;
+}
+
+defm VADD : avx512_fp_binop_p<0x58, "vadd", fadd, 1>,
+ avx512_fp_binop_p_round<0x58, "vadd", X86faddRnd>;
+defm VMUL : avx512_fp_binop_p<0x59, "vmul", fmul, 1>,
+ avx512_fp_binop_p_round<0x59, "vmul", X86fmulRnd>;
+defm VSUB : avx512_fp_binop_p<0x5C, "vsub", fsub>,
+ avx512_fp_binop_p_round<0x5C, "vsub", X86fsubRnd>;
+defm VDIV : avx512_fp_binop_p<0x5E, "vdiv", fdiv>,
+ avx512_fp_binop_p_round<0x5E, "vdiv", X86fdivRnd>;
defm VMIN : avx512_fp_binop_p<0x5D, "vmin", X86fmin, 1>;
defm VMAX : avx512_fp_binop_p<0x5F, "vmax", X86fmax, 1>;
-defm VSUB : avx512_fp_binop_p<0x5C, "vsub", fsub>;
-defm VDIV : avx512_fp_binop_p<0x5E, "vdiv", fdiv>;
def : Pat<(v16f32 (int_x86_avx512_mask_max_ps_512 (v16f32 VR512:$src1),
(v16f32 VR512:$src2), (bc_v16f32 (v16i32 immAllZerosV)),
}
defm VPTESTMDZ : avx512_vptest<0x27, "vptestmd", VK16, VR512, f512mem,
- memopv16i32, X86testm, v16i32>, T8PD, EVEX_V512,
+ loadv16i32, X86testm, v16i32>, T8PD, EVEX_V512,
EVEX_CD8<32, CD8VF>;
defm VPTESTMQZ : avx512_vptest<0x27, "vptestmq", VK8, VR512, f512mem,
- memopv8i64, X86testm, v8i64>, T8PD, EVEX_V512, VEX_W,
+ loadv8i64, X86testm, v8i64>, T8PD, EVEX_V512, VEX_W,
EVEX_CD8<64, CD8VF>;
let Predicates = [HasCDI] in {
defm VPTESTNMDZ : avx512_vptest<0x27, "vptestnmd", VK16, VR512, f512mem,
- memopv16i32, X86testnm, v16i32>, T8XS, EVEX_V512,
+ loadv16i32, X86testnm, v16i32>, T8XS, EVEX_V512,
EVEX_CD8<32, CD8VF>;
defm VPTESTNMQZ : avx512_vptest<0x27, "vptestnmq", VK8, VR512, f512mem,
- memopv8i64, X86testnm, v8i64>, T8XS, EVEX_V512, VEX_W,
+ loadv8i64, X86testnm, v8i64>, T8XS, EVEX_V512, VEX_W,
EVEX_CD8<64, CD8VF>;
}
defm mi : AVX512_maskable<opc, ImmFormM, _, (outs _.RC:$dst),
(ins _.MemOp:$src1, u8imm:$src2), OpcodeStr,
"$src2, $src1", "$src1, $src2",
- (_.VT (OpNode (_.MemOpFrag addr:$src1), (i8 imm:$src2))),
+ (_.VT (OpNode (_.LdFrag addr:$src1), (i8 imm:$src2))),
" ", SSE_INTSHIFT_ITINS_P.rm>, AVX512BIi8Base, EVEX_4V;
}
defm rm : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
(ins _.RC:$src1, i128mem:$src2), OpcodeStr,
"$src2, $src1", "$src1, $src2",
- (_.VT (OpNode _.RC:$src1, (bc_frag (memopv2i64 addr:$src2)))),
+ (_.VT (OpNode _.RC:$src1, (bc_frag (loadv2i64 addr:$src2)))),
" ", SSE_INTSHIFT_ITINS_P.rm>, AVX512BIBase, EVEX_4V;
}
defm rm : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
(ins _.RC:$src1, _.MemOp:$src2), OpcodeStr,
"$src2, $src1", "$src1, $src2",
- (_.VT (OpNode _.RC:$src1, (_.MemOpFrag addr:$src2))),
+ (_.VT (OpNode _.RC:$src1, (_.LdFrag addr:$src2))),
" ", SSE_INTSHIFT_ITINS_P.rm>, AVX5128IBase, EVEX_4V;
}
(VT (X86Movddup (memop_frag addr:$src))))]>, EVEX;
}
-defm VMOVDDUPZ : avx512_movddup<"vmovddup", VR512, v8f64, f512mem, memopv8f64>,
+defm VMOVDDUPZ : avx512_movddup<"vmovddup", VR512, v8f64, f512mem, loadv8f64>,
VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;
def : Pat<(X86Movddup (v8f64 (scalar_to_vector (loadf64 addr:$src)))),
(VMOVDDUPZrm addr:$src)>;
}
defm VMOVSHDUPZ : avx512_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
- v16f32, VR512, memopv16f32, f512mem>, EVEX_V512,
+ v16f32, VR512, loadv16f32, f512mem>, EVEX_V512,
EVEX_CD8<32, CD8VF>;
defm VMOVSLDUPZ : avx512_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
- v16f32, VR512, memopv16f32, f512mem>, EVEX_V512,
+ v16f32, VR512, loadv16f32, f512mem>, EVEX_V512,
EVEX_CD8<32, CD8VF>;
def : Pat<(v16i32 (X86Movshdup VR512:$src)), (VMOVSHDUPZrr VR512:$src)>;
-def : Pat<(v16i32 (X86Movshdup (memopv16i32 addr:$src))),
+def : Pat<(v16i32 (X86Movshdup (loadv16i32 addr:$src))),
(VMOVSHDUPZrm addr:$src)>;
def : Pat<(v16i32 (X86Movsldup VR512:$src)), (VMOVSLDUPZrr VR512:$src)>;
-def : Pat<(v16i32 (X86Movsldup (memopv16i32 addr:$src))),
+def : Pat<(v16i32 (X86Movsldup (loadv16i32 addr:$src))),
(VMOVSLDUPZrm addr:$src)>;
//===----------------------------------------------------------------------===//
}
} // Constraints = "$src1 = $dst"
+let Constraints = "$src1 = $dst" in {
+// Omitting the parameter OpNode (= null_frag) disables ISel pattern matching.
+multiclass avx512_fma3_round_rrb<bits<8> opc, string OpcodeStr, X86VectorVTInfo _,
+ SDPatternOperator OpNode> {
+ defm rb: AVX512_maskable_3src<opc, MRMSrcReg, _, (outs _.RC:$dst),
+ (ins _.RC:$src2, _.RC:$src3, AVX512RC:$rc),
+ OpcodeStr, "$rc, $src3, $src2", "$src2, $src3, $rc",
+ (_.VT ( OpNode _.RC:$src1, _.RC:$src2, _.RC:$src3, (i32 imm:$rc)))>,
+ AVX512FMA3Base, EVEX_B, EVEX_RC;
+ }
+} // Constraints = "$src1 = $dst"
+
+multiclass avx512_fma3_round_forms<bits<8> opc213, string OpcodeStr,
+ X86VectorVTInfo VTI, SDPatternOperator OpNode> {
+ defm v213r : avx512_fma3_round_rrb<opc213, !strconcat(OpcodeStr, "213", VTI.Suffix),
+ VTI, OpNode>, EVEX_CD8<VTI.EltSize, CD8VF>;
+}
+
multiclass avx512_fma3p_forms<bits<8> opc213, bits<8> opc231,
string OpcodeStr, X86VectorVTInfo VTI,
SDPatternOperator OpNode> {
multiclass avx512_fma3p<bits<8> opc213, bits<8> opc231,
string OpcodeStr,
- SDPatternOperator OpNode> {
+ SDPatternOperator OpNode,
+ SDPatternOperator OpNodeRnd> {
let ExeDomain = SSEPackedSingle in {
defm NAME##PSZ : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
- v16f32_info, OpNode>, EVEX_V512;
+ v16f32_info, OpNode>,
+ avx512_fma3_round_forms<opc213, OpcodeStr,
+ v16f32_info, OpNodeRnd>, EVEX_V512;
defm NAME##PSZ256 : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
v8f32x_info, OpNode>, EVEX_V256;
defm NAME##PSZ128 : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
}
let ExeDomain = SSEPackedDouble in {
defm NAME##PDZ : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
- v8f64_info, OpNode>, EVEX_V512, VEX_W;
+ v8f64_info, OpNode>,
+ avx512_fma3_round_forms<opc213, OpcodeStr,
+ v8f64_info, OpNodeRnd>, EVEX_V512, VEX_W;
defm NAME##PDZ256 : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
v4f64x_info, OpNode>, EVEX_V256, VEX_W;
defm NAME##PDZ128 : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
}
}
-defm VFMADD : avx512_fma3p<0xA8, 0xB8, "vfmadd", X86Fmadd>;
-defm VFMSUB : avx512_fma3p<0xAA, 0xBA, "vfmsub", X86Fmsub>;
-defm VFMADDSUB : avx512_fma3p<0xA6, 0xB6, "vfmaddsub", X86Fmaddsub>;
-defm VFMSUBADD : avx512_fma3p<0xA7, 0xB7, "vfmsubadd", X86Fmsubadd>;
-defm VFNMADD : avx512_fma3p<0xAC, 0xBC, "vfnmadd", X86Fnmadd>;
-defm VFNMSUB : avx512_fma3p<0xAE, 0xBE, "vfnmsub", X86Fnmsub>;
+defm VFMADD : avx512_fma3p<0xA8, 0xB8, "vfmadd", X86Fmadd, X86FmaddRnd>;
+defm VFMSUB : avx512_fma3p<0xAA, 0xBA, "vfmsub", X86Fmsub, X86FmsubRnd>;
+defm VFMADDSUB : avx512_fma3p<0xA6, 0xB6, "vfmaddsub", X86Fmaddsub, X86FmaddsubRnd>;
+defm VFMSUBADD : avx512_fma3p<0xA7, 0xB7, "vfmsubadd", X86Fmsubadd, X86FmsubaddRnd>;
+defm VFNMADD : avx512_fma3p<0xAC, 0xBC, "vfnmadd", X86Fnmadd, X86FnmaddRnd>;
+defm VFNMSUB : avx512_fma3p<0xAE, 0xBE, "vfnmsub", X86Fnmsub, X86FnmsubRnd>;
let Constraints = "$src1 = $dst" in {
multiclass avx512_fma3p_m132<bits<8> opc, string OpcodeStr, SDNode OpNode,
def m: AVX512FMA3<opc, MRMSrcMem, (outs _.RC:$dst),
(ins _.RC:$src1, _.RC:$src3, _.MemOp:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src3, $dst|$dst, $src3, $src2}"),
- [(set _.RC:$dst, (_.VT (OpNode _.RC:$src1, (_.MemOpFrag addr:$src2),
+ [(set _.RC:$dst, (_.VT (OpNode _.RC:$src1, (_.LdFrag addr:$src2),
_.RC:$src3)))]>;
def mb: AVX512FMA3<opc, MRMSrcMem, (outs _.RC:$dst),
(ins _.RC:$src1, _.RC:$src3, _.ScalarMemOp:$src2),
}
defm VCVTPD2PSZ : avx512_vcvt_fp_with_rc<0x5A, "vcvtpd2ps", VR512, VR256X, fround,
- memopv8f64, f512mem, v8f32, v8f64,
+ loadv8f64, f512mem, v8f32, v8f64,
SSEPackedSingle>, EVEX_V512, VEX_W, PD,
EVEX_CD8<64, CD8VF>;
defm VCVTPS2PDZ : avx512_vcvt_fp<0x5A, "vcvtps2pd", VR256X, VR512, fextend,
- memopv4f64, f256mem, v8f64, v8f32,
+ loadv4f64, f256mem, v8f64, v8f32,
SSEPackedDouble>, EVEX_V512, PS,
EVEX_CD8<32, CD8VH>;
def : Pat<(v8f64 (extloadv8f32 addr:$src)),
//===----------------------------------------------------------------------===//
defm VCVTDQ2PSZ : avx512_vcvt_fp_with_rc<0x5B, "vcvtdq2ps", VR512, VR512, sint_to_fp,
- memopv8i64, i512mem, v16f32, v16i32,
+ loadv8i64, i512mem, v16f32, v16i32,
SSEPackedSingle>, EVEX_V512, PS,
EVEX_CD8<32, CD8VF>;
defm VCVTDQ2PDZ : avx512_vcvt_fp<0xE6, "vcvtdq2pd", VR256X, VR512, sint_to_fp,
- memopv4i64, i256mem, v8f64, v8i32,
+ loadv4i64, i256mem, v8f64, v8i32,
SSEPackedDouble>, EVEX_V512, XS,
EVEX_CD8<32, CD8VH>;
defm VCVTTPS2DQZ : avx512_vcvt_fp<0x5B, "vcvttps2dq", VR512, VR512, fp_to_sint,
- memopv16f32, f512mem, v16i32, v16f32,
+ loadv16f32, f512mem, v16i32, v16f32,
SSEPackedSingle>, EVEX_V512, XS,
EVEX_CD8<32, CD8VF>;
defm VCVTTPD2DQZ : avx512_vcvt_fp<0xE6, "vcvttpd2dq", VR512, VR256X, fp_to_sint,
- memopv8f64, f512mem, v8i32, v8f64,
+ loadv8f64, f512mem, v8i32, v8f64,
SSEPackedDouble>, EVEX_V512, PD, VEX_W,
EVEX_CD8<64, CD8VF>;
defm VCVTTPS2UDQZ : avx512_vcvt_fp<0x78, "vcvttps2udq", VR512, VR512, fp_to_uint,
- memopv16f32, f512mem, v16i32, v16f32,
+ loadv16f32, f512mem, v16i32, v16f32,
SSEPackedSingle>, EVEX_V512, PS,
EVEX_CD8<32, CD8VF>;
(VCVTTPS2UDQZrr VR512:$src)>;
defm VCVTTPD2UDQZ : avx512_vcvt_fp<0x78, "vcvttpd2udq", VR512, VR256X, fp_to_uint,
- memopv8f64, f512mem, v8i32, v8f64,
+ loadv8f64, f512mem, v8i32, v8f64,
SSEPackedDouble>, EVEX_V512, PS, VEX_W,
EVEX_CD8<64, CD8VF>;
(VCVTTPD2UDQZrr VR512:$src)>;
defm VCVTUDQ2PDZ : avx512_vcvt_fp<0x7A, "vcvtudq2pd", VR256X, VR512, uint_to_fp,
- memopv4i64, f256mem, v8f64, v8i32,
+ loadv4i64, f256mem, v8f64, v8i32,
SSEPackedDouble>, EVEX_V512, XS,
EVEX_CD8<32, CD8VH>;
defm VCVTUDQ2PSZ : avx512_vcvt_fp_with_rc<0x7A, "vcvtudq2ps", VR512, VR512, uint_to_fp,
- memopv16i32, f512mem, v16f32, v16i32,
+ loadv16i32, f512mem, v16f32, v16i32,
SSEPackedSingle>, EVEX_V512, XD,
EVEX_CD8<32, CD8VF>;
}
defm VCVTPS2DQZ : avx512_vcvt_fp2int<0x5B, "vcvtps2dq", VR512, VR512,
- memopv16f32, f512mem, SSEPackedSingle>, PD,
+ loadv16f32, f512mem, SSEPackedSingle>, PD,
EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VCVTPD2DQZ : avx512_vcvt_fp2int<0xE6, "vcvtpd2dq", VR512, VR256X,
- memopv8f64, f512mem, SSEPackedDouble>, XD, VEX_W,
+ loadv8f64, f512mem, SSEPackedDouble>, XD, VEX_W,
EVEX_V512, EVEX_CD8<64, CD8VF>;
def : Pat <(v16i32 (int_x86_avx512_mask_cvtps2dq_512 (v16f32 VR512:$src),
(VCVTPD2DQZrrb VR512:$src, imm:$rc)>;
defm VCVTPS2UDQZ : avx512_vcvt_fp2int<0x79, "vcvtps2udq", VR512, VR512,
- memopv16f32, f512mem, SSEPackedSingle>,
+ loadv16f32, f512mem, SSEPackedSingle>,
PS, EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VCVTPD2UDQZ : avx512_vcvt_fp2int<0x79, "vcvtpd2udq", VR512, VR256X,
- memopv8f64, f512mem, SSEPackedDouble>, VEX_W,
+ loadv8f64, f512mem, SSEPackedDouble>, VEX_W,
PS, EVEX_V512, EVEX_CD8<64, CD8VF>;
def : Pat <(v16i32 (int_x86_avx512_mask_cvtps2udq_512 (v16f32 VR512:$src),
defm VRNDSCALEPSZ : avx512_rndscale<0x08, "vrndscaleps", f512mem, VR512,
- memopv16f32, SSEPackedSingle>, EVEX_V512,
+ loadv16f32, SSEPackedSingle>, EVEX_V512,
EVEX_CD8<32, CD8VF>;
def : Pat<(v16f32 (int_x86_avx512_mask_rndscale_ps_512 (v16f32 VR512:$src1),
defm VRNDSCALEPDZ : avx512_rndscale<0x09, "vrndscalepd", f512mem, VR512,
- memopv8f64, SSEPackedDouble>, EVEX_V512,
+ loadv8f64, SSEPackedDouble>, EVEX_V512,
VEX_W, EVEX_CD8<64, CD8VF>;
def : Pat<(v8f64 (int_x86_avx512_mask_rndscale_pd_512 (v8f64 VR512:$src1),
}
defm VPMOVZXBDZ: avx512_extend<0x31, "vpmovzxbd", VK16WM, VR512, VR128X, X86vzext,
- memopv2i64, i128mem, v16i32, v16i8>, EVEX_V512,
+ loadv2i64, i128mem, v16i32, v16i8>, EVEX_V512,
EVEX_CD8<8, CD8VQ>;
defm VPMOVZXBQZ: avx512_extend<0x32, "vpmovzxbq", VK8WM, VR512, VR128X, X86vzext,
- memopv2i64, i128mem, v8i64, v16i8>, EVEX_V512,
+ loadv2i64, i128mem, v8i64, v16i8>, EVEX_V512,
EVEX_CD8<8, CD8VO>;
defm VPMOVZXWDZ: avx512_extend<0x33, "vpmovzxwd", VK16WM, VR512, VR256X, X86vzext,
- memopv4i64, i256mem, v16i32, v16i16>, EVEX_V512,
+ loadv4i64, i256mem, v16i32, v16i16>, EVEX_V512,
EVEX_CD8<16, CD8VH>;
defm VPMOVZXWQZ: avx512_extend<0x34, "vpmovzxwq", VK8WM, VR512, VR128X, X86vzext,
- memopv2i64, i128mem, v8i64, v8i16>, EVEX_V512,
+ loadv2i64, i128mem, v8i64, v8i16>, EVEX_V512,
EVEX_CD8<16, CD8VQ>;
defm VPMOVZXDQZ: avx512_extend<0x35, "vpmovzxdq", VK8WM, VR512, VR256X, X86vzext,
- memopv4i64, i256mem, v8i64, v8i32>, EVEX_V512,
+ loadv4i64, i256mem, v8i64, v8i32>, EVEX_V512,
EVEX_CD8<32, CD8VH>;
defm VPMOVSXBDZ: avx512_extend<0x21, "vpmovsxbd", VK16WM, VR512, VR128X, X86vsext,
- memopv2i64, i128mem, v16i32, v16i8>, EVEX_V512,
+ loadv2i64, i128mem, v16i32, v16i8>, EVEX_V512,
EVEX_CD8<8, CD8VQ>;
defm VPMOVSXBQZ: avx512_extend<0x22, "vpmovsxbq", VK8WM, VR512, VR128X, X86vsext,
- memopv2i64, i128mem, v8i64, v16i8>, EVEX_V512,
+ loadv2i64, i128mem, v8i64, v16i8>, EVEX_V512,
EVEX_CD8<8, CD8VO>;
defm VPMOVSXWDZ: avx512_extend<0x23, "vpmovsxwd", VK16WM, VR512, VR256X, X86vsext,
- memopv4i64, i256mem, v16i32, v16i16>, EVEX_V512,
+ loadv4i64, i256mem, v16i32, v16i16>, EVEX_V512,
EVEX_CD8<16, CD8VH>;
defm VPMOVSXWQZ: avx512_extend<0x24, "vpmovsxwq", VK8WM, VR512, VR128X, X86vsext,
- memopv2i64, i128mem, v8i64, v8i16>, EVEX_V512,
+ loadv2i64, i128mem, v8i64, v8i16>, EVEX_V512,
EVEX_CD8<16, CD8VQ>;
defm VPMOVSXDQZ: avx512_extend<0x25, "vpmovsxdq", VK8WM, VR512, VR256X, X86vsext,
- memopv4i64, i256mem, v8i64, v8i32>, EVEX_V512,
+ loadv4i64, i256mem, v8i64, v8i32>, EVEX_V512,
EVEX_CD8<32, CD8VH>;
//===----------------------------------------------------------------------===//
EVEX_4V, Sched<[WriteShuffle]>;
}
-defm VSHUFPSZ : avx512_shufp<VR512, f512mem, v16f32, "vshufps", memopv16f32,
+defm VSHUFPSZ : avx512_shufp<VR512, f512mem, v16f32, "vshufps", loadv16f32,
SSEPackedSingle>, PS, EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VSHUFPDZ : avx512_shufp<VR512, f512mem, v8f64, "vshufpd", memopv8f64,
+defm VSHUFPDZ : avx512_shufp<VR512, f512mem, v8f64, "vshufpd", loadv8f64,
SSEPackedDouble>, PD, VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;
def : Pat<(v16i32 (X86Shufp VR512:$src1, VR512:$src2, (i8 imm:$imm))),
(VSHUFPSZrri VR512:$src1, VR512:$src2, imm:$imm)>;
def : Pat<(v16i32 (X86Shufp VR512:$src1,
- (memopv16i32 addr:$src2), (i8 imm:$imm))),
+ (loadv16i32 addr:$src2), (i8 imm:$imm))),
(VSHUFPSZrmi VR512:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v8i64 (X86Shufp VR512:$src1, VR512:$src2, (i8 imm:$imm))),
(VSHUFPDZrri VR512:$src1, VR512:$src2, imm:$imm)>;
def : Pat<(v8i64 (X86Shufp VR512:$src1,
- (memopv8i64 addr:$src2), (i8 imm:$imm))),
+ (loadv8i64 addr:$src2), (i8 imm:$imm))),
(VSHUFPDZrmi VR512:$src1, addr:$src2, imm:$imm)>;
multiclass avx512_valign<X86VectorVTInfo _> {
(VPLZCNTQrrk VR512:$src1,
(v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)), VR512:$src2)>;
-def : Pat<(v16i32 (ctlz (memopv16i32 addr:$src))),
+def : Pat<(v16i32 (ctlz (loadv16i32 addr:$src))),
(VPLZCNTDrm addr:$src)>;
def : Pat<(v16i32 (ctlz (v16i32 VR512:$src))),
(VPLZCNTDrr VR512:$src)>;
-def : Pat<(v8i64 (ctlz (memopv8i64 addr:$src))),
+def : Pat<(v8i64 (ctlz (loadv8i64 addr:$src))),
(VPLZCNTQrm addr:$src)>;
def : Pat<(v8i64 (ctlz (v8i64 VR512:$src))),
(VPLZCNTQrr VR512:$src)>;
def : Pat<(store (i1 0), addr:$dst), (MOV8mi addr:$dst, (i8 0))>;
def : Pat<(store VK1:$src, addr:$dst),
- (KMOVWmk addr:$dst, (COPY_TO_REGCLASS VK1:$src, VK16))>;
+ (MOV8mr addr:$dst,
+ (EXTRACT_SUBREG (KMOVWrk (COPY_TO_REGCLASS VK1:$src, VK16)),
+ sub_8bit))>, Requires<[HasAVX512, NoDQI]>;
+
+def : Pat<(store VK8:$src, addr:$dst),
+ (MOV8mr addr:$dst,
+ (EXTRACT_SUBREG (KMOVWrk (COPY_TO_REGCLASS VK8:$src, VK16)),
+ sub_8bit))>, Requires<[HasAVX512, NoDQI]>;
def truncstorei1 : PatFrag<(ops node:$val, node:$ptr),
(truncstore node:$val, node:$ptr), [{