index 6093c943b2f46fda2507dcd549d95d8abcf6efe8..1b25ee8c55e77a6f7f375b6388b291bf560b67de 100644 (file)
// EltVT). These are things like the register class for the writemask, etc.
// The idea is to pass one of these as the template argument rather than the
// individual arguments.
-class X86VectorVTInfo<int numelts, ValueType EltVT, RegisterClass rc,
+// The template is also used for scalar types, in this case numelts is 1.
+class X86VectorVTInfo<int numelts, ValueType eltvt, RegisterClass rc,
string suffix = ""> {
RegisterClass RC = rc;
+ ValueType EltVT = eltvt;
int NumElts = numelts;
// Corresponding mask register class.
// Suffix used in the instruction mnemonic.
string Suffix = suffix;
- string VTName = "v" # NumElts # EltVT;
+ // VTName is a string name for vector VT. For vector types it will be
+ // v # NumElts # EltVT, so for vector of 8 elements of i32 it will be v8i32
+ // It is a little bit complex for scalar types, where NumElts = 1.
+ // In this case we build v4f32 or v2f64
+ string VTName = "v" # !if (!eq (NumElts, 1),
+ !if (!eq (EltVT.Size, 32), 4,
+ !if (!eq (EltVT.Size, 64), 2, NumElts)), NumElts) # EltVT;
// The vector VT.
ValueType VT = !cast<ValueType>(VTName);
// 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), ?)));
+ !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
def v4f32x_info : X86VectorVTInfo<4, f32, VR128X, "ps">;
def v2f64x_info : X86VectorVTInfo<2, f64, VR128X, "pd">;
+// We map scalar types to the smallest (128-bit) vector type
+// with the appropriate element type. This allows to use the same masking logic.
+def f32x_info : X86VectorVTInfo<1, f32, VR128X, "ss">;
+def f64x_info : X86VectorVTInfo<1, f64, VR128X, "sd">;
+
class AVX512VLVectorVTInfo<X86VectorVTInfo i512, X86VectorVTInfo i256,
X86VectorVTInfo i128> {
X86VectorVTInfo info512 = i512;
string OpcodeStr,
string AttSrcAsm, string IntelSrcAsm,
dag RHS, dag MaskingRHS,
- string Round = "",
+ SDNode Select = vselect, string Round = "",
string MaskingConstraint = "",
InstrItinClass itin = NoItinerary,
bit IsCommutable = 0> :
[(set _.RC:$dst, RHS)],
[(set _.RC:$dst, MaskingRHS)],
[(set _.RC:$dst,
- (vselect _.KRCWM:$mask, RHS, _.ImmAllZerosV))],
+ (Select _.KRCWM:$mask, RHS, _.ImmAllZerosV))],
Round, MaskingConstraint, NoItinerary, IsCommutable>;
// This multiclass generates the unconditional/non-masking, the masking and
-// the zero-masking variant of the instruction. In the masking case, the
+// the zero-masking variant of the vector instruction. In the masking case, the
// perserved vector elements come from a new dummy input operand tied to $dst.
multiclass AVX512_maskable<bits<8> O, Format F, X86VectorVTInfo _,
dag Outs, dag Ins, string OpcodeStr,
!con((ins _.RC:$src0, _.KRCWM:$mask), Ins),
!con((ins _.KRCWM:$mask), Ins),
OpcodeStr, AttSrcAsm, IntelSrcAsm, RHS,
- (vselect _.KRCWM:$mask, RHS, _.RC:$src0), Round,
- "$src0 = $dst", itin, IsCommutable>;
+ (vselect _.KRCWM:$mask, RHS, _.RC:$src0), vselect,
+ Round, "$src0 = $dst", itin, IsCommutable>;
+
+// This multiclass generates the unconditional/non-masking, the masking and
+// the zero-masking variant of the scalar instruction.
+multiclass AVX512_maskable_scalar<bits<8> O, Format F, X86VectorVTInfo _,
+ dag Outs, dag Ins, string OpcodeStr,
+ string AttSrcAsm, string IntelSrcAsm,
+ dag RHS, string Round = "",
+ InstrItinClass itin = NoItinerary,
+ bit IsCommutable = 0> :
+ AVX512_maskable_common<O, F, _, Outs, Ins,
+ !con((ins _.RC:$src0, _.KRCWM:$mask), Ins),
+ !con((ins _.KRCWM:$mask), Ins),
+ OpcodeStr, AttSrcAsm, IntelSrcAsm, RHS,
+ (X86select _.KRCWM:$mask, RHS, _.RC:$src0), X86select,
+ Round, "$src0 = $dst", itin, IsCommutable>;
// Similar to AVX512_maskable but in this case one of the source operands
// ($src1) is already tied to $dst so we just use that for the preserved
SDNodeXForm INSERT_get_vinsert_imm> {
let hasSideEffects = 0, ExeDomain = To.ExeDomain in {
def rr : AVX512AIi8<Opcode, MRMSrcReg, (outs VR512:$dst),
- (ins VR512:$src1, From.RC:$src2, i8imm:$src3),
+ (ins VR512:$src1, From.RC:$src2, u8imm:$src3),
"vinsert" # From.EltTypeName # "x" # From.NumElts #
"\t{$src3, $src2, $src1, $dst|"
"$dst, $src1, $src2, $src3}",
let mayLoad = 1 in
def rm : AVX512AIi8<Opcode, MRMSrcMem, (outs VR512:$dst),
- (ins VR512:$src1, From.MemOp:$src2, i8imm:$src3),
+ (ins VR512:$src1, From.MemOp:$src2, u8imm:$src3),
"vinsert" # From.EltTypeName # "x" # From.NumElts #
"\t{$src3, $src2, $src1, $dst|"
"$dst, $src1, $src2, $src3}",
// vinsertps - insert f32 to XMM
def VINSERTPSzrr : AVX512AIi8<0x21, MRMSrcReg, (outs VR128X:$dst),
- (ins VR128X:$src1, VR128X:$src2, i8imm:$src3),
+ (ins VR128X:$src1, VR128X:$src2, u8imm:$src3),
"vinsertps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128X:$dst, (X86insertps VR128X:$src1, VR128X:$src2, imm:$src3))]>,
EVEX_4V;
def VINSERTPSzrm: AVX512AIi8<0x21, MRMSrcMem, (outs VR128X:$dst),
- (ins VR128X:$src1, f32mem:$src2, i8imm:$src3),
+ (ins VR128X:$src1, f32mem:$src2, u8imm:$src3),
"vinsertps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128X:$dst, (X86insertps VR128X:$src1,
(v4f32 (scalar_to_vector (loadf32 addr:$src2))),
SDNodeXForm EXTRACT_get_vextract_imm> {
let hasSideEffects = 0, ExeDomain = To.ExeDomain in {
defm rr : AVX512_maskable_in_asm<Opcode, MRMDestReg, To, (outs To.RC:$dst),
- (ins VR512:$src1, i8imm:$idx),
+ (ins VR512:$src1, u8imm:$idx),
"vextract" # To.EltTypeName # "x4",
"$idx, $src1", "$src1, $idx",
[(set To.RC:$dst, (vextract_extract:$idx (From.VT VR512:$src1),
AVX512AIi8Base, EVEX, EVEX_V512;
let mayStore = 1 in
def rm : AVX512AIi8<Opcode, MRMDestMem, (outs),
- (ins To.MemOp:$dst, VR512:$src1, i8imm:$src2),
+ (ins To.MemOp:$dst, VR512:$src1, u8imm:$src2),
"vextract" # To.EltTypeName # "x4\t{$src2, $src1, $dst|"
"$dst, $src1, $src2}",
[]>, EVEX, EVEX_V512, EVEX_CD8<To.EltSize, CD8VT4>;
avx512vl_f64_info>, VEX_W, EVEX_CD8<64, CD8VT1>;
}
+// avx512_broadcast_pat introduces patterns for broadcast with a scalar argument.
+// Later, we can canonize broadcast instructions before ISel phase and
+// eliminate additional patterns on ISel.
+// SrcRC_v and SrcRC_s are RegisterClasses for vector and scalar
+// representations of source
+multiclass avx512_broadcast_pat<string InstName, SDNode OpNode,
+ X86VectorVTInfo _, RegisterClass SrcRC_v,
+ RegisterClass SrcRC_s> {
+ def : Pat<(_.VT (OpNode (_.EltVT SrcRC_s:$src))),
+ (!cast<Instruction>(InstName##"r")
+ (COPY_TO_REGCLASS SrcRC_s:$src, SrcRC_v))>;
+
+ let AddedComplexity = 30 in {
+ def : Pat<(_.VT (vselect _.KRCWM:$mask,
+ (OpNode (_.EltVT SrcRC_s:$src)), _.RC:$src0)),
+ (!cast<Instruction>(InstName##"rk") _.RC:$src0, _.KRCWM:$mask,
+ (COPY_TO_REGCLASS SrcRC_s:$src, SrcRC_v))>;
+
+ def : Pat<(_.VT(vselect _.KRCWM:$mask,
+ (OpNode (_.EltVT SrcRC_s:$src)), _.ImmAllZerosV)),
+ (!cast<Instruction>(InstName##"rkz") _.KRCWM:$mask,
+ (COPY_TO_REGCLASS SrcRC_s:$src, SrcRC_v))>;
+ }
+}
+
+defm : avx512_broadcast_pat<"VBROADCASTSSZ", X86VBroadcast, v16f32_info,
+ VR128X, FR32X>;
+defm : avx512_broadcast_pat<"VBROADCASTSDZ", X86VBroadcast, v8f64_info,
+ VR128X, FR64X>;
+
+let Predicates = [HasVLX] in {
+ defm : avx512_broadcast_pat<"VBROADCASTSSZ256", X86VBroadcast,
+ v8f32x_info, VR128X, FR32X>;
+ defm : avx512_broadcast_pat<"VBROADCASTSSZ128", X86VBroadcast,
+ v4f32x_info, VR128X, FR32X>;
+ defm : avx512_broadcast_pat<"VBROADCASTSDZ256", X86VBroadcast,
+ v4f64x_info, VR128X, FR64X>;
+}
+
def : Pat<(v16f32 (X86VBroadcast (loadf32 addr:$src))),
(VBROADCASTSSZm addr:$src)>;
def : Pat<(v8f64 (X86VBroadcast (loadf64 addr:$src))),
def : Pat<(int_x86_avx512_vbroadcast_sd_512 addr:$src),
(VBROADCASTSDZm addr:$src)>;
-multiclass avx512_int_broadcast_reg<bits<8> opc, string OpcodeStr,
- RegisterClass SrcRC, RegisterClass KRC> {
- def Zrr : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst), (ins SrcRC:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- []>, EVEX, EVEX_V512;
- def Zkrr : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst),
- (ins KRC:$mask, SrcRC:$src),
- !strconcat(OpcodeStr,
- "\t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}"),
- []>, EVEX, EVEX_V512, EVEX_KZ;
-}
-
-defm VPBROADCASTDr : avx512_int_broadcast_reg<0x7C, "vpbroadcastd", GR32, VK16WM>;
-defm VPBROADCASTQr : avx512_int_broadcast_reg<0x7C, "vpbroadcastq", GR64, VK8WM>,
- VEX_W;
-
+multiclass avx512_int_broadcast_reg<bits<8> opc, X86VectorVTInfo _,
+ RegisterClass SrcRC> {
+ defm r : AVX512_maskable_in_asm<opc, MRMSrcReg, _, (outs _.RC:$dst),
+ (ins SrcRC:$src), "vpbroadcast"##_.Suffix,
+ "$src", "$src", []>, T8PD, EVEX;
+}
+
+multiclass avx512_int_broadcast_reg_vl<bits<8> opc, AVX512VLVectorVTInfo _,
+ RegisterClass SrcRC, Predicate prd> {
+ let Predicates = [prd] in
+ defm Z : avx512_int_broadcast_reg<opc, _.info512, SrcRC>, EVEX_V512;
+ let Predicates = [prd, HasVLX] in {
+ defm Z256 : avx512_int_broadcast_reg<opc, _.info256, SrcRC>, EVEX_V256;
+ defm Z128 : avx512_int_broadcast_reg<opc, _.info128, SrcRC>, EVEX_V128;
+ }
+}
+
+defm VPBROADCASTBr : avx512_int_broadcast_reg_vl<0x7A, avx512vl_i8_info, GR32,
+ HasBWI>;
+defm VPBROADCASTWr : avx512_int_broadcast_reg_vl<0x7B, avx512vl_i16_info, GR32,
+ HasBWI>;
+defm VPBROADCASTDr : avx512_int_broadcast_reg_vl<0x7C, avx512vl_i32_info, GR32,
+ HasAVX512>;
+defm VPBROADCASTQr : avx512_int_broadcast_reg_vl<0x7C, avx512vl_i64_info, GR64,
+ HasAVX512>, VEX_W;
+
def : Pat <(v16i32 (X86vzext VK16WM:$mask)),
- (VPBROADCASTDrZkrr VK16WM:$mask, (i32 (MOV32ri 0x1)))>;
+ (VPBROADCASTDrZrkz VK16WM:$mask, (i32 (MOV32ri 0x1)))>;
def : Pat <(v8i64 (X86vzext VK8WM:$mask)),
- (VPBROADCASTQrZkrr VK8WM:$mask, (i64 (MOV64ri 0x1)))>;
+ (VPBROADCASTQrZrkz VK8WM:$mask, (i64 (MOV64ri 0x1)))>;
def : Pat<(v16i32 (X86VBroadcast (i32 GR32:$src))),
- (VPBROADCASTDrZrr GR32:$src)>;
+ (VPBROADCASTDrZr GR32:$src)>;
def : Pat<(v16i32 (X86VBroadcastm VK16WM:$mask, (i32 GR32:$src))),
- (VPBROADCASTDrZkrr VK16WM:$mask, GR32:$src)>;
+ (VPBROADCASTDrZrkz VK16WM:$mask, GR32:$src)>;
def : Pat<(v8i64 (X86VBroadcast (i64 GR64:$src))),
- (VPBROADCASTQrZrr GR64:$src)>;
+ (VPBROADCASTQrZr GR64:$src)>;
def : Pat<(v8i64 (X86VBroadcastm VK8WM:$mask, (i64 GR64:$src))),
- (VPBROADCASTQrZkrr VK8WM:$mask, GR64:$src)>;
+ (VPBROADCASTQrZrkz VK8WM:$mask, GR64:$src)>;
def : Pat<(v16i32 (int_x86_avx512_pbroadcastd_i32_512 (i32 GR32:$src))),
- (VPBROADCASTDrZrr GR32:$src)>;
+ (VPBROADCASTDrZr GR32:$src)>;
def : Pat<(v8i64 (int_x86_avx512_pbroadcastq_i64_512 (i64 GR64:$src))),
- (VPBROADCASTQrZrr GR64:$src)>;
+ (VPBROADCASTQrZr GR64:$src)>;
def : Pat<(v16i32 (int_x86_avx512_mask_pbroadcast_d_gpr_512 (i32 GR32:$src),
(v16i32 immAllZerosV), (i16 GR16:$mask))),
- (VPBROADCASTDrZkrr (COPY_TO_REGCLASS GR16:$mask, VK16WM), GR32:$src)>;
+ (VPBROADCASTDrZrkz (COPY_TO_REGCLASS GR16:$mask, VK16WM), GR32:$src)>;
def : Pat<(v8i64 (int_x86_avx512_mask_pbroadcast_q_gpr_512 (i64 GR64:$src),
(bc_v8i64 (v16i32 immAllZerosV)), (i8 GR8:$mask))),
- (VPBROADCASTQrZkrr (COPY_TO_REGCLASS GR8:$mask, VK8WM), GR64:$src)>;
+ (VPBROADCASTQrZrkz (COPY_TO_REGCLASS GR8:$mask, VK8WM), GR64:$src)>;
multiclass avx512_int_broadcast_rm<bits<8> opc, string OpcodeStr,
X86MemOperand x86memop, PatFrag ld_frag,
(OpVT (X86VBroadcast (SrcVT VR128X:$src))))]>, EVEX;
def krr : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst), (ins KRC:$mask,
VR128X:$src),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src}"),
[(set DstRC:$dst,
(OpVT (X86VBroadcastm KRC:$mask, (SrcVT VR128X:$src))))]>,
let mayLoad = 1 in {
def rm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set DstRC:$dst,
+ [(set DstRC:$dst,
(OpVT (X86VBroadcast (ld_frag addr:$src))))]>, EVEX;
def krm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst), (ins KRC:$mask,
x86memop:$src),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src}"),
- [(set DstRC:$dst, (OpVT (X86VBroadcastm KRC:$mask,
+ [(set DstRC:$dst, (OpVT (X86VBroadcastm KRC:$mask,
(ld_frag addr:$src))))]>, EVEX, EVEX_KZ;
}
}
(VBROADCASTSSZr VR128X:$src)>;
def : Pat<(v8f64 (int_x86_avx512_vbroadcast_sd_pd_512 (v2f64 VR128X:$src))),
(VBROADCASTSDZr VR128X:$src)>;
-
+
// Provide fallback in case the load node that is used in the patterns above
// is used by additional users, which prevents the pattern selection.
def : Pat<(v16f32 (X86VBroadcast FR32X:$src)),
let Predicates = [HasAVX512] in {
def : Pat<(v8i32 (X86VBroadcastm (v8i1 VK8WM:$mask), (loadi32 addr:$src))),
- (EXTRACT_SUBREG
+ (EXTRACT_SUBREG
(v16i32 (VPBROADCASTDZkrm (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
addr:$src)), sub_ymm)>;
}
def Zrr : AVX512XS8I<opc, MRMSrcReg, (outs VR512:$dst), (ins KRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[]>, EVEX, EVEX_V512;
-
+
let Predicates = [HasCDI, HasVLX] in {
def Z128rr : AVX512XS8I<opc, MRMSrcReg, (outs VR128:$dst), (ins KRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
X86VectorVTInfo _> {
let ExeDomain = _.ExeDomain in {
def ri : AVX512AIi8<opc, MRMSrcReg, (outs _.RC:$dst),
- (ins _.RC:$src1, i8imm:$src2),
+ (ins _.RC:$src1, u8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set _.RC:$dst,
(_.VT (OpNode _.RC:$src1, (i8 imm:$src2))))]>,
EVEX;
def mi : AVX512AIi8<opc, MRMSrcMem, (outs _.RC:$dst),
- (ins _.MemOp:$src1, i8imm:$src2),
+ (ins _.MemOp:$src1, u8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set _.RC:$dst,
(VPERMILPDZri VR512:$src1, imm:$imm)>;
// -- VPERM - register form --
-multiclass avx512_perm<bits<8> opc, string OpcodeStr, RegisterClass RC,
+multiclass avx512_perm<bits<8> opc, string OpcodeStr, RegisterClass RC,
PatFrag mem_frag, X86MemOperand x86memop, ValueType OpVT> {
def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst),
defm VPERMDZ : avx512_perm<0x36, "vpermd", VR512, memopv16i32, i512mem,
v16i32>, EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VPERMQZ : avx512_perm<0x36, "vpermq", VR512, memopv8i64, i512mem,
+defm VPERMQZ : avx512_perm<0x36, "vpermq", VR512, memopv8i64, i512mem,
v8i64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
let ExeDomain = SSEPackedSingle in
defm VPERMPSZ : avx512_perm<0x16, "vpermps", VR512, memopv16f32, f512mem,
v16f32>, EVEX_V512, EVEX_CD8<32, CD8VF>;
let ExeDomain = SSEPackedDouble in
-defm VPERMPDZ : avx512_perm<0x16, "vpermpd", VR512, memopv8f64, f512mem,
+defm VPERMPDZ : avx512_perm<0x16, "vpermpd", VR512, memopv8f64, f512mem,
v8f64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
// -- VPERM2I - 3 source operands form --
@@ -1006,77 +1085,110 @@ defm VPERMT2PD : avx512_perm_table_3src<0x7F, "pd", VR512, memopv8f64, i512mem,
//===----------------------------------------------------------------------===//
// AVX-512 - BLEND using mask
//
-multiclass avx512_blendmask<bits<8> opc, string OpcodeStr,
- RegisterClass KRC, RegisterClass RC,
- X86MemOperand x86memop, PatFrag mem_frag,
- SDNode OpNode, ValueType vt> {
- def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, RC:$src2),
+multiclass avx512_blendmask<bits<8> opc, string OpcodeStr, X86VectorVTInfo _> {
+ let ExeDomain = _.ExeDomain in {
+ def rr : AVX5128I<opc, MRMSrcReg, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.RC:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, ${dst} |${dst}, $src1, $src2}"),
+ []>, EVEX_4V;
+ def rrk : AVX5128I<opc, MRMSrcReg, (outs _.RC:$dst),
+ (ins _.KRCWM:$mask, _.RC:$src1, _.RC:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, ${dst} {${mask}}|${dst} {${mask}}, $src1, $src2}"),
- [(set RC:$dst, (OpNode KRC:$mask, (vt RC:$src2),
- (vt RC:$src1)))]>, EVEX_4V, EVEX_K;
- let mayLoad = 1 in
- def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, x86memop:$src2),
+ [(set _.RC:$dst, (X86select _.KRCWM:$mask, (_.VT _.RC:$src1),
+ (_.VT _.RC:$src2)))]>, EVEX_4V, EVEX_K;
+ def rrkz : AVX5128I<opc, MRMSrcReg, (outs _.RC:$dst),
+ (ins _.KRCWM:$mask, _.RC:$src1, _.RC:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src1, $src2}"),
+ []>, EVEX_4V, EVEX_KZ;
+ let mayLoad = 1 in {
+ def rm : AVX5128I<opc, MRMSrcMem, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.MemOp:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, ${dst} |${dst}, $src1, $src2}"),
+ []>, EVEX_4V, EVEX_CD8<_.EltSize, CD8VF>;
+ def rmk : AVX5128I<opc, MRMSrcMem, (outs _.RC:$dst),
+ (ins _.KRCWM:$mask, _.RC:$src1, _.MemOp:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, ${dst} {${mask}}|${dst} {${mask}}, $src1, $src2}"),
- []>, EVEX_4V, EVEX_K;
+ [(set _.RC:$dst, (X86select _.KRCWM:$mask, (_.VT _.RC:$src1),
+ (_.VT (bitconvert (_.LdFrag addr:$src2)))))]>,
+ EVEX_4V, EVEX_K, EVEX_CD8<_.EltSize, CD8VF>;
+ def rmkz : AVX5128I<opc, MRMSrcMem, (outs _.RC:$dst),
+ (ins _.KRCWM:$mask, _.RC:$src1, _.MemOp:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src1, $src2}"),
+ []>, EVEX_4V, EVEX_KZ, EVEX_CD8<_.EltSize, CD8VF>;
+ }
+ }
}
+multiclass avx512_blendmask_rmb<bits<8> opc, string OpcodeStr, X86VectorVTInfo _> {
+
+ def rmbk : AVX5128I<opc, MRMSrcMem, (outs _.RC:$dst),
+ (ins _.KRCWM:$mask, _.RC:$src1, _.ScalarMemOp:$src2),
+ !strconcat(OpcodeStr,
+ "\t{${src2}", _.BroadcastStr, ", $src1, $dst {${mask}}|",
+ "$dst {${mask}}, $src1, ${src2}", _.BroadcastStr, "}"),
+ [(set _.RC:$dst,(X86select _.KRCWM:$mask, (_.VT _.RC:$src1),
+ (X86VBroadcast (_.ScalarLdFrag addr:$src2))))]>,
+ EVEX_4V, EVEX_K, EVEX_B, EVEX_CD8<_.EltSize, CD8VF>;
+
+ def rmb : AVX5128I<opc, MRMSrcMem, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.ScalarMemOp:$src2),
+ !strconcat(OpcodeStr,
+ "\t{${src2}", _.BroadcastStr, ", $src1, $dst|",
+ "$dst, $src1, ${src2}", _.BroadcastStr, "}"),
+ []>, EVEX_4V, EVEX_B, EVEX_CD8<_.EltSize, CD8VF>;
+
+}
+
+multiclass blendmask_dq <bits<8> opc, string OpcodeStr,
+ AVX512VLVectorVTInfo VTInfo> {
+ defm Z : avx512_blendmask <opc, OpcodeStr, VTInfo.info512>,
+ avx512_blendmask_rmb <opc, OpcodeStr, VTInfo.info512>, EVEX_V512;
+
+ let Predicates = [HasVLX] in {
+ defm Z256 : avx512_blendmask<opc, OpcodeStr, VTInfo.info256>,
+ avx512_blendmask_rmb <opc, OpcodeStr, VTInfo.info256>, EVEX_V256;
+ defm Z128 : avx512_blendmask<opc, OpcodeStr, VTInfo.info128>,
+ avx512_blendmask_rmb <opc, OpcodeStr, VTInfo.info128>, EVEX_V128;
+ }
+}
+
+multiclass blendmask_bw <bits<8> opc, string OpcodeStr,
+ AVX512VLVectorVTInfo VTInfo> {
+ let Predicates = [HasBWI] in
+ defm Z : avx512_blendmask <opc, OpcodeStr, VTInfo.info512>, EVEX_V512;
+
+ let Predicates = [HasBWI, HasVLX] in {
+ defm Z256 : avx512_blendmask <opc, OpcodeStr, VTInfo.info256>, EVEX_V256;
+ defm Z128 : avx512_blendmask <opc, OpcodeStr, VTInfo.info128>, EVEX_V128;
+ }
+}
+
+
+defm VBLENDMPS : blendmask_dq <0x65, "vblendmps", avx512vl_f32_info>;
+defm VBLENDMPD : blendmask_dq <0x65, "vblendmpd", avx512vl_f64_info>, VEX_W;
+defm VPBLENDMD : blendmask_dq <0x64, "vpblendmd", avx512vl_i32_info>;
+defm VPBLENDMQ : blendmask_dq <0x64, "vpblendmq", avx512vl_i64_info>, VEX_W;
+defm VPBLENDMB : blendmask_bw <0x66, "vpblendmb", avx512vl_i8_info>;
+defm VPBLENDMW : blendmask_bw <0x66, "vpblendmw", avx512vl_i16_info>, VEX_W;
-let ExeDomain = SSEPackedSingle in
-defm VBLENDMPSZ : avx512_blendmask<0x65, "vblendmps",
- VK16WM, VR512, f512mem,
- memopv16f32, vselect, v16f32>,
- EVEX_CD8<32, CD8VF>, EVEX_V512;
-let ExeDomain = SSEPackedDouble in
-defm VBLENDMPDZ : avx512_blendmask<0x65, "vblendmpd",
- VK8WM, VR512, f512mem,
- memopv8f64, vselect, v8f64>,
- VEX_W, EVEX_CD8<64, CD8VF>, EVEX_V512;
-
-def : Pat<(v16f32 (int_x86_avx512_mask_blend_ps_512 (v16f32 VR512:$src1),
- (v16f32 VR512:$src2), (i16 GR16:$mask))),
- (VBLENDMPSZrr (COPY_TO_REGCLASS GR16:$mask, VK16WM),
- VR512:$src1, VR512:$src2)>;
-
-def : Pat<(v8f64 (int_x86_avx512_mask_blend_pd_512 (v8f64 VR512:$src1),
- (v8f64 VR512:$src2), (i8 GR8:$mask))),
- (VBLENDMPDZrr (COPY_TO_REGCLASS GR8:$mask, VK8WM),
- VR512:$src1, VR512:$src2)>;
-
-defm VPBLENDMDZ : avx512_blendmask<0x64, "vpblendmd",
- VK16WM, VR512, f512mem,
- memopv16i32, vselect, v16i32>,
- EVEX_CD8<32, CD8VF>, EVEX_V512;
-
-defm VPBLENDMQZ : avx512_blendmask<0x64, "vpblendmq",
- VK8WM, VR512, f512mem,
- memopv8i64, vselect, v8i64>,
- VEX_W, EVEX_CD8<64, CD8VF>, EVEX_V512;
-
-def : Pat<(v16i32 (int_x86_avx512_mask_blend_d_512 (v16i32 VR512:$src1),
- (v16i32 VR512:$src2), (i16 GR16:$mask))),
- (VPBLENDMDZrr (COPY_TO_REGCLASS GR16:$mask, VK16),
- VR512:$src1, VR512:$src2)>;
-
-def : Pat<(v8i64 (int_x86_avx512_mask_blend_q_512 (v8i64 VR512:$src1),
- (v8i64 VR512:$src2), (i8 GR8:$mask))),
- (VPBLENDMQZrr (COPY_TO_REGCLASS GR8:$mask, VK8),
- VR512:$src1, VR512:$src2)>;
let Predicates = [HasAVX512] in {
def : Pat<(v8f32 (vselect (v8i1 VK8WM:$mask), (v8f32 VR256X:$src1),
(v8f32 VR256X:$src2))),
- (EXTRACT_SUBREG
- (v16f32 (VBLENDMPSZrr (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
+ (EXTRACT_SUBREG
+ (v16f32 (VBLENDMPSZrrk (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
(v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)),
(v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
def : Pat<(v8i32 (vselect (v8i1 VK8WM:$mask), (v8i32 VR256X:$src1),
(v8i32 VR256X:$src2))),
- (EXTRACT_SUBREG
- (v16i32 (VPBLENDMDZrr (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
+ (EXTRACT_SUBREG
+ (v16i32 (VPBLENDMDZrrk (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
(v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)),
(v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
}
(ld_frag addr:$src2), imm:$cc))], IIC_SSE_ALU_F32P_RM>, EVEX_4V;
let isAsmParserOnly = 1, hasSideEffects = 0 in {
def rri_alt : AVX512Ii8<0xC2, MRMSrcReg,
- (outs VK1:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
+ (outs VK1:$dst), (ins RC:$src1, RC:$src2, u8imm:$cc),
asm_alt, [], IIC_SSE_ALU_F32S_RR>, EVEX_4V;
+ let mayLoad = 1 in
def rmi_alt : AVX512Ii8<0xC2, MRMSrcMem,
- (outs VK1:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
+ (outs VK1:$dst), (ins RC:$src1, x86memop:$src2, u8imm:$cc),
asm_alt, [], IIC_SSE_ALU_F32P_RM>, EVEX_4V;
}
}
// Accept explicit immediate argument form instead of comparison code.
let isAsmParserOnly = 1, hasSideEffects = 0 in {
def rri_alt : AVX512AIi8<opc, MRMSrcReg,
- (outs _.KRC:$dst), (ins _.RC:$src1, _.RC:$src2, i8imm:$cc),
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.RC:$src2, u8imm:$cc),
!strconcat("vpcmp", Suffix, "\t{$cc, $src2, $src1, $dst|",
"$dst, $src1, $src2, $cc}"),
[], IIC_SSE_ALU_F32P_RR>, EVEX_4V;
+ let mayLoad = 1 in
def rmi_alt : AVX512AIi8<opc, MRMSrcMem,
- (outs _.KRC:$dst), (ins _.RC:$src1, _.MemOp:$src2, i8imm:$cc),
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.MemOp:$src2, u8imm:$cc),
!strconcat("vpcmp", Suffix, "\t{$cc, $src2, $src1, $dst|",
"$dst, $src1, $src2, $cc}"),
[], IIC_SSE_ALU_F32P_RM>, EVEX_4V;
def rrik_alt : AVX512AIi8<opc, MRMSrcReg,
(outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1, _.RC:$src2,
- i8imm:$cc),
+ u8imm:$cc),
!strconcat("vpcmp", Suffix,
"\t{$cc, $src2, $src1, $dst {${mask}}|",
"$dst {${mask}}, $src1, $src2, $cc}"),
[], IIC_SSE_ALU_F32P_RR>, EVEX_4V, EVEX_K;
+ let mayLoad = 1 in
def rmik_alt : AVX512AIi8<opc, MRMSrcMem,
(outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1, _.MemOp:$src2,
- i8imm:$cc),
+ u8imm:$cc),
!strconcat("vpcmp", Suffix,
"\t{$cc, $src2, $src1, $dst {${mask}}|",
"$dst {${mask}}, $src1, $src2, $cc}"),
multiclass avx512_icmp_cc_rmb<bits<8> opc, string Suffix, SDNode OpNode,
X86VectorVTInfo _> :
avx512_icmp_cc<opc, Suffix, OpNode, _> {
- let mayLoad = 1 in {
def rmib : AVX512AIi8<opc, MRMSrcMem,
(outs _.KRC:$dst), (ins _.RC:$src1, _.ScalarMemOp:$src2,
AVXCC:$cc),
(X86VBroadcast (_.ScalarLdFrag addr:$src2)),
imm:$cc)))],
IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_K, EVEX_B;
- }
// Accept explicit immediate argument form instead of comparison code.
- let isAsmParserOnly = 1, hasSideEffects = 0 in {
+ let isAsmParserOnly = 1, hasSideEffects = 0, mayLoad = 1 in {
def rmib_alt : AVX512AIi8<opc, MRMSrcMem,
(outs _.KRC:$dst), (ins _.RC:$src1, _.ScalarMemOp:$src2,
- i8imm:$cc),
+ u8imm:$cc),
!strconcat("vpcmp", Suffix,
"\t{$cc, ${src2}", _.BroadcastStr, ", $src1, $dst|",
"$dst, $src1, ${src2}", _.BroadcastStr, ", $cc}"),
[], IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_B;
def rmibk_alt : AVX512AIi8<opc, MRMSrcMem,
(outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1,
- _.ScalarMemOp:$src2, i8imm:$cc),
+ _.ScalarMemOp:$src2, u8imm:$cc),
!strconcat("vpcmp", Suffix,
"\t{$cc, ${src2}", _.BroadcastStr, ", $src1, $dst {${mask}}|",
"$dst {${mask}}, $src1, ${src2}", _.BroadcastStr, ", $cc}"),
!strconcat("vcmp${cc}", suffix,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set KRC:$dst, (X86cmpm (vt RC:$src1), (vt RC:$src2), imm:$cc))], d>;
+ let hasSideEffects = 0 in
def rrib: AVX512PIi8<0xC2, MRMSrcReg,
(outs KRC:$dst), (ins RC:$src1, RC:$src2, AVXCC:$cc),
!strconcat("vcmp${cc}", suffix,
// Accept explicit immediate argument form instead of comparison code.
let isAsmParserOnly = 1, hasSideEffects = 0 in {
def rri_alt : AVX512PIi8<0xC2, MRMSrcReg,
- (outs KRC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
+ (outs KRC:$dst), (ins RC:$src1, RC:$src2, u8imm:$cc),
!strconcat("vcmp", suffix,
"\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}"), [], d>;
+ let mayLoad = 1 in
def rmi_alt : AVX512PIi8<0xC2, MRMSrcMem,
- (outs KRC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
+ (outs KRC:$dst), (ins RC:$src1, x86memop:$src2, u8imm:$cc),
!strconcat("vcmp", suffix,
"\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}"), [], d>;
}
@@ -1465,25 +1580,25 @@ def : Pat<(v8i1 (X86cmpmu (v8i32 VR256X:$src1), (v8i32 VR256X:$src2), imm:$cc)),
imm:$cc), VK8)>;
def : Pat<(i16 (int_x86_avx512_mask_cmp_ps_512 (v16f32 VR512:$src1),
- (v16f32 VR512:$src2), imm:$cc, (i16 -1),
+ (v16f32 VR512:$src2), i8immZExt5:$cc, (i16 -1),
FROUND_NO_EXC)),
(COPY_TO_REGCLASS (VCMPPSZrrib VR512:$src1, VR512:$src2,
(I8Imm imm:$cc)), GR16)>;
-
+
def : Pat<(i8 (int_x86_avx512_mask_cmp_pd_512 (v8f64 VR512:$src1),
- (v8f64 VR512:$src2), imm:$cc, (i8 -1),
+ (v8f64 VR512:$src2), i8immZExt5:$cc, (i8 -1),
FROUND_NO_EXC)),
(COPY_TO_REGCLASS (VCMPPDZrrib VR512:$src1, VR512:$src2,
(I8Imm imm:$cc)), GR8)>;
def : Pat<(i16 (int_x86_avx512_mask_cmp_ps_512 (v16f32 VR512:$src1),
- (v16f32 VR512:$src2), imm:$cc, (i16 -1),
+ (v16f32 VR512:$src2), i8immZExt5:$cc, (i16 -1),
FROUND_CURRENT)),
(COPY_TO_REGCLASS (VCMPPSZrri VR512:$src1, VR512:$src2,
(I8Imm imm:$cc)), GR16)>;
def : Pat<(i8 (int_x86_avx512_mask_cmp_pd_512 (v8f64 VR512:$src1),
- (v8f64 VR512:$src2), imm:$cc, (i8 -1),
+ (v8f64 VR512:$src2), i8immZExt5:$cc, (i8 -1),
FROUND_CURRENT)),
(COPY_TO_REGCLASS (VCMPPDZrri VR512:$src1, VR512:$src2,
(I8Imm imm:$cc)), GR8)>;
multiclass avx512_mask_shiftop<bits<8> opc, string OpcodeStr, RegisterClass KRC,
SDNode OpNode> {
let Predicates = [HasAVX512] in
- def ri : Ii8<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src, i8imm:$imm),
+ def ri : Ii8<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src, u8imm:$imm),
!strconcat(OpcodeStr,
"\t{$imm, $src, $dst|$dst, $src, $imm}"),
[(set KRC:$dst, (OpNode KRC:$src, (i8 imm:$imm)))]>;
multiclass avx512_store<bits<8> opc, string OpcodeStr, PatFrag st_frag,
ValueType OpVT, RegisterClass KRC, RegisterClass RC,
X86MemOperand memop, Domain d> {
- let isAsmParserOnly = 1, hasSideEffects = 0 in {
+ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in {
def rr_alt : AVX512PI<opc, MRMDestReg, (outs RC:$dst), (ins RC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [], d>,
EVEX;
(bc_v16f32 (v16i32 immAllZerosV)), GR16:$mask)),
(VMOVUPSZrmkz (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)), addr:$ptr)>;
+def: Pat<(v8f64 (int_x86_avx512_mask_load_pd_512 addr:$ptr,
+ (bc_v8f64 (v16i32 immAllZerosV)), GR8:$mask)),
+ (VMOVAPDZrmkz (v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)), addr:$ptr)>;
+
+def: Pat<(v16f32 (int_x86_avx512_mask_load_ps_512 addr:$ptr,
+ (bc_v16f32 (v16i32 immAllZerosV)), GR16:$mask)),
+ (VMOVAPSZrmkz (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)), addr:$ptr)>;
+
+def: Pat<(v8f64 (int_x86_avx512_mask_load_pd_512 addr:$ptr,
+ (bc_v8f64 (v16i32 immAllZerosV)), (i8 -1))),
+ (VMOVAPDZrm addr:$ptr)>;
+
+def: Pat<(v16f32 (int_x86_avx512_mask_load_ps_512 addr:$ptr,
+ (bc_v16f32 (v16i32 immAllZerosV)), (i16 -1))),
+ (VMOVAPSZrm addr:$ptr)>;
+
def: Pat<(int_x86_avx512_mask_storeu_ps_512 addr:$ptr, (v16f32 VR512:$src),
GR16:$mask),
(VMOVUPSZmrk addr:$ptr, (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)),
(VMOVUPDZmrk addr:$ptr, (v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)),
VR512:$src)>;
+def: Pat<(int_x86_avx512_mask_store_ps_512 addr:$ptr, (v16f32 VR512:$src),
+ GR16:$mask),
+ (VMOVAPSZmrk addr:$ptr, (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)),
+ VR512:$src)>;
+def: Pat<(int_x86_avx512_mask_store_pd_512 addr:$ptr, (v8f64 VR512:$src),
+ GR8:$mask),
+ (VMOVAPDZmrk addr:$ptr, (v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)),
+ VR512:$src)>;
+
def: Pat<(masked_store addr:$ptr, VK8WM:$mask, (v8f32 VR256:$src)),
(VMOVUPSZmrk addr:$ptr,
(v16i1 (COPY_TO_REGCLASS VK8WM:$mask, VK16WM)),
def: Pat<(v8f64 (masked_load addr:$ptr, VK8WM:$mask, (v8f64 VR512:$src0))),
(VMOVUPDZrmk VR512:$src0, VK8WM:$mask, addr:$ptr)>;
+def: Pat<(v8f32 (masked_load addr:$ptr, VK8WM:$mask, (v8f32 VR256:$src0))),
+ (v8f32 (EXTRACT_SUBREG (v16f32 (VMOVUPSZrmk
+ (INSERT_SUBREG (v16f32 (IMPLICIT_DEF)), VR256:$src0, sub_ymm),
+ (v16i1 (COPY_TO_REGCLASS VK8WM:$mask, VK16WM)), addr:$ptr)), sub_ymm))>;
+
defm VMOVDQA32 : avx512_load_vl<0x6F, "vmovdqa32", "alignedload", "i", "32",
"16", "8", "4", SSEPackedInt, HasAVX512>,
avx512_store_vl<0x7F, "vmovdqa32", "alignedstore",
// AVX-512 MOVSS, MOVSD
//===----------------------------------------------------------------------===//
-multiclass avx512_move_scalar <string asm, RegisterClass RC,
+multiclass avx512_move_scalar <string asm, RegisterClass RC,
SDNode OpNode, ValueType vt,
X86MemOperand x86memop, PatFrag mem_pat> {
let hasSideEffects = 0 in {
- def rr : SI<0x10, MRMSrcReg, (outs VR128X:$dst), (ins VR128X:$src1, RC:$src2),
+ def rr : SI<0x10, MRMSrcReg, (outs VR128X:$dst), (ins VR128X:$src1, RC:$src2),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128X:$dst, (vt (OpNode VR128X:$src1,
(scalar_to_vector RC:$src2))))],
// Move low f32 and clear high bits.
def : Pat<(v8f32 (X86vzmovl (v8f32 VR256X:$src))),
(SUBREG_TO_REG (i32 0),
- (VMOVSSZrr (v4f32 (V_SET0)),
+ (VMOVSSZrr (v4f32 (V_SET0)),
(EXTRACT_SUBREG (v8f32 VR256X:$src), sub_xmm)), sub_xmm)>;
def : Pat<(v8i32 (X86vzmovl (v8i32 VR256X:$src))),
(SUBREG_TO_REG (i32 0),
def VMOVZPQILo2PQIZrr : AVX512XSI<0x7E, MRMSrcReg, (outs VR128X:$dst),
(ins VR128X:$src),
"vmovq\t{$src, $dst|$dst, $src}",
- [(set VR128X:$dst, (v2i64 (X86vzmovl
+ [(set VR128X:$dst, (v2i64 (X86vzmovl
(v2i64 VR128X:$src))))],
IIC_SSE_MOVQ_RR>, EVEX, VEX_W;
(VMOV64toPQIZrr GR64:$src)>;
def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
(VMOVDI2PDIZrr GR32:$src)>;
-
+
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
(VMOVDI2PDIZrm addr:$src)>;
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
def rr : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (OpVT (OpNode (OpVT RC:$src1), (OpVT RC:$src2))))],
+ [(set RC:$dst, (OpVT (OpNode (OpVT RC:$src1), (OpVT RC:$src2))))],
IIC_SSE_UNPCK>, EVEX_4V;
def rm : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2),
@@ -3022,17 +3167,17 @@ defm VPUNPCKHQDQZ : avx512_unpack_int<0x6D, "vpunpckhqdq", X86Unpckh, v8i64,
//
multiclass avx512_pshuf_imm<bits<8> opc, string OpcodeStr, RegisterClass RC,
- SDNode OpNode, PatFrag mem_frag,
+ SDNode OpNode, PatFrag mem_frag,
X86MemOperand x86memop, ValueType OpVT> {
def ri : AVX512Ii8<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, i8imm:$src2),
+ (ins RC:$src1, u8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
(OpVT (OpNode RC:$src1, (i8 imm:$src2))))]>,
EVEX;
def mi : AVX512Ii8<opc, MRMSrcMem, (outs RC:$dst),
- (ins x86memop:$src1, i8imm:$src2),
+ (ins x86memop:$src1, u8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
// AVX-512 VPTESTM instructions
//===----------------------------------------------------------------------===//
-multiclass avx512_vptest<bits<8> opc, string OpcodeStr, RegisterClass KRC,
- RegisterClass RC, X86MemOperand x86memop, PatFrag memop_frag,
+multiclass avx512_vptest<bits<8> opc, string OpcodeStr, RegisterClass KRC,
+ RegisterClass RC, X86MemOperand x86memop, PatFrag memop_frag,
SDNode OpNode, ValueType vt> {
def rr : AVX512PI<opc, MRMSrcReg,
- (outs KRC:$dst), (ins RC:$src1, RC:$src2),
+ (outs KRC:$dst), (ins RC:$src1, RC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set KRC:$dst, (OpNode (vt RC:$src1), (vt RC:$src2)))],
SSEPackedInt>, EVEX_4V;
def rm : AVX512PI<opc, MRMSrcMem,
- (outs KRC:$dst), (ins RC:$src1, x86memop:$src2),
+ (outs KRC:$dst), (ins RC:$src1, x86memop:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set KRC:$dst, (OpNode (vt RC:$src1),
+ [(set KRC:$dst, (OpNode (vt RC:$src1),
(bitconvert (memop_frag addr:$src2))))], SSEPackedInt>, EVEX_4V;
}
// AVX-512 Shift instructions
//===----------------------------------------------------------------------===//
multiclass avx512_shift_rmi<bits<8> opc, Format ImmFormR, Format ImmFormM,
- string OpcodeStr, SDNode OpNode, X86VectorVTInfo _> {
+ string OpcodeStr, SDNode OpNode, X86VectorVTInfo _> {
defm ri : AVX512_maskable<opc, ImmFormR, _, (outs _.RC:$dst),
- (ins _.RC:$src1, i8imm:$src2), OpcodeStr,
+ (ins _.RC:$src1, u8imm:$src2), OpcodeStr,
"$src2, $src1", "$src1, $src2",
(_.VT (OpNode _.RC:$src1, (i8 imm:$src2))),
" ", SSE_INTSHIFT_ITINS_P.rr>, AVX512BIi8Base, EVEX_4V;
defm mi : AVX512_maskable<opc, ImmFormM, _, (outs _.RC:$dst),
- (ins _.MemOp:$src1, i8imm:$src2), OpcodeStr,
+ (ins _.MemOp:$src1, u8imm:$src2), OpcodeStr,
"$src2, $src1", "$src1, $src2",
(_.VT (OpNode (_.MemOpFrag addr:$src1), (i8 imm:$src2))),
" ", SSE_INTSHIFT_ITINS_P.rm>, AVX512BIi8Base, EVEX_4V;
" ", SSE_INTSHIFT_ITINS_P.rm>, AVX512BIBase, EVEX_4V;
}
-multiclass avx512_varshift_sizes<bits<8> opc, string OpcodeStr, SDNode OpNode,
+multiclass avx512_shift_sizes<bits<8> opc, string OpcodeStr, SDNode OpNode,
ValueType SrcVT, PatFrag bc_frag, X86VectorVTInfo _> {
defm Z : avx512_shift_rrm<opc, OpcodeStr, OpNode, SrcVT, bc_frag, _>, EVEX_V512;
}
-multiclass avx512_varshift_types<bits<8> opcd, bits<8> opcq, string OpcodeStr,
+multiclass avx512_shift_types<bits<8> opcd, bits<8> opcq, string OpcodeStr,
SDNode OpNode> {
- defm D : avx512_varshift_sizes<opcd, OpcodeStr#"d", OpNode, v4i32, bc_v4i32,
- v16i32_info>, EVEX_CD8<32, CD8VQ>;
- defm Q : avx512_varshift_sizes<opcq, OpcodeStr#"q", OpNode, v2i64, bc_v2i64,
+ defm D : avx512_shift_sizes<opcd, OpcodeStr#"d", OpNode, v4i32, bc_v4i32,
+ v16i32_info>, EVEX_CD8<32, CD8VQ>;
+ defm Q : avx512_shift_sizes<opcq, OpcodeStr#"q", OpNode, v2i64, bc_v2i64,
v8i64_info>, EVEX_CD8<64, CD8VQ>, VEX_W;
}
v8i64_info>, EVEX_V512,
EVEX_CD8<64, CD8VF>, VEX_W;
-defm VPSRL : avx512_varshift_types<0xD2, 0xD3, "vpsrl", X86vsrl>;
-defm VPSLL : avx512_varshift_types<0xF2, 0xF3, "vpsll", X86vshl>;
-defm VPSRA : avx512_varshift_types<0xE2, 0xE2, "vpsra", X86vsra>;
+defm VPSLL : avx512_shift_types<0xF2, 0xF3, "vpsll", X86vshl>;
+defm VPSRA : avx512_shift_types<0xE2, 0xE2, "vpsra", X86vsra>;
+defm VPSRL : avx512_shift_types<0xD2, 0xD3, "vpsrl", X86vsrl>;
//===-------------------------------------------------------------------===//
// Variable Bit Shifts
//===-------------------------------------------------------------------===//
multiclass avx512_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
- RegisterClass RC, ValueType vt,
- X86MemOperand x86memop, PatFrag mem_frag> {
- def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2),
- !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst,
- (vt (OpNode RC:$src1, (vt RC:$src2))))]>,
- EVEX_4V;
- def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2),
- !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst,
- (vt (OpNode RC:$src1, (mem_frag addr:$src2))))]>,
- EVEX_4V;
+ X86VectorVTInfo _> {
+ defm rr : AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.RC:$src2), OpcodeStr,
+ "$src2, $src1", "$src1, $src2",
+ (_.VT (OpNode _.RC:$src1, (_.VT _.RC:$src2))),
+ " ", SSE_INTSHIFT_ITINS_P.rr>, AVX5128IBase, EVEX_4V;
+ defm rm : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.MemOp:$src2), OpcodeStr,
+ "$src2, $src1", "$src1, $src2",
+ (_.VT (OpNode _.RC:$src1, (_.MemOpFrag addr:$src2))),
+ " ", SSE_INTSHIFT_ITINS_P.rm>, AVX5128IBase, EVEX_4V;
+}
+
+multiclass avx512_var_shift_sizes<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ AVX512VLVectorVTInfo _> {
+ defm Z : avx512_var_shift<opc, OpcodeStr, OpNode, _.info512>, EVEX_V512;
+}
+
+multiclass avx512_var_shift_types<bits<8> opc, string OpcodeStr,
+ SDNode OpNode> {
+ defm D : avx512_var_shift_sizes<opc, OpcodeStr#"d", OpNode,
+ avx512vl_i32_info>, EVEX_CD8<32, CD8VQ>;
+ defm Q : avx512_var_shift_sizes<opc, OpcodeStr#"q", OpNode,
+ avx512vl_i64_info>, EVEX_CD8<64, CD8VQ>, VEX_W;
}
-defm VPSLLVDZ : avx512_var_shift<0x47, "vpsllvd", shl, VR512, v16i32,
- i512mem, memopv16i32>, EVEX_V512,
- EVEX_CD8<32, CD8VF>;
-defm VPSLLVQZ : avx512_var_shift<0x47, "vpsllvq", shl, VR512, v8i64,
- i512mem, memopv8i64>, EVEX_V512, VEX_W,
- EVEX_CD8<64, CD8VF>;
-defm VPSRLVDZ : avx512_var_shift<0x45, "vpsrlvd", srl, VR512, v16i32,
- i512mem, memopv16i32>, EVEX_V512,
- EVEX_CD8<32, CD8VF>;
-defm VPSRLVQZ : avx512_var_shift<0x45, "vpsrlvq", srl, VR512, v8i64,
- i512mem, memopv8i64>, EVEX_V512, VEX_W,
- EVEX_CD8<64, CD8VF>;
-defm VPSRAVDZ : avx512_var_shift<0x46, "vpsravd", sra, VR512, v16i32,
- i512mem, memopv16i32>, EVEX_V512,
- EVEX_CD8<32, CD8VF>;
-defm VPSRAVQZ : avx512_var_shift<0x46, "vpsravq", sra, VR512, v8i64,
- i512mem, memopv8i64>, EVEX_V512, VEX_W,
- EVEX_CD8<64, CD8VF>;
+defm VPSLLV : avx512_var_shift_types<0x47, "vpsllv", shl>;
+defm VPSRAV : avx512_var_shift_types<0x46, "vpsrav", sra>;
+defm VPSRLV : avx512_var_shift_types<0x45, "vpsrlv", srl>;
//===----------------------------------------------------------------------===//
// AVX-512 - MOVDDUP
//===----------------------------------------------------------------------===//
-multiclass avx512_movddup<string OpcodeStr, RegisterClass RC, ValueType VT,
+multiclass avx512_movddup<string OpcodeStr, RegisterClass RC, ValueType VT,
X86MemOperand x86memop, PatFrag memop_frag> {
def rr : AVX512PDI<0x12, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
@@ -3396,61 +3536,58 @@ multiclass avx512_fma3p_rm<bits<8> opc, string OpcodeStr, X86VectorVTInfo _,
AVX512FMA3Base;
let mayLoad = 1 in
- def m: AVX512FMA3<opc, MRMSrcMem, (outs _.RC:$dst),
- (ins _.RC:$src1, _.RC:$src2, _.MemOp:$src3),
- !strconcat(OpcodeStr, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set _.RC:$dst, (_.VT (OpNode _.RC:$src1, _.RC:$src2,
- (_.MemOpFrag addr:$src3))))]>;
- def mb: AVX512FMA3<opc, MRMSrcMem, (outs _.RC:$dst),
- (ins _.RC:$src1, _.RC:$src2, _.ScalarMemOp:$src3),
- !strconcat(OpcodeStr, "\t{${src3}", _.BroadcastStr,
- ", $src2, $dst|$dst, $src2, ${src3}", _.BroadcastStr, "}"),
- [(set _.RC:$dst, (OpNode _.RC:$src1, _.RC:$src2,
- (_.VT (X86VBroadcast (_.ScalarLdFrag addr:$src3)))))]>, EVEX_B;
-}
+ defm m: AVX512_maskable_3src<opc, MRMSrcMem, _, (outs _.RC:$dst),
+ (ins _.RC:$src2, _.MemOp:$src3),
+ OpcodeStr, "$src3, $src2", "$src2, $src3",
+ (_.VT (OpNode _.RC:$src1, _.RC:$src2, (_.LdFrag addr:$src3)))>,
+ AVX512FMA3Base;
+
+ defm mb: AVX512_maskable_3src<opc, MRMSrcMem, _, (outs _.RC:$dst),
+ (ins _.RC:$src2, _.ScalarMemOp:$src3),
+ OpcodeStr, !strconcat("${src3}", _.BroadcastStr,", $src2"), !strconcat("$src2, ${src3}", _.BroadcastStr ),
+ (OpNode _.RC:$src1, _.RC:$src2,(_.VT (X86VBroadcast (_.ScalarLdFrag addr:$src3))))>,
+ AVX512FMA3Base, EVEX_B;
+ }
} // Constraints = "$src1 = $dst"
multiclass avx512_fma3p_forms<bits<8> opc213, bits<8> opc231,
string OpcodeStr, X86VectorVTInfo VTI,
SDPatternOperator OpNode> {
- defm v213 : avx512_fma3p_rm<opc213, !strconcat(OpcodeStr, "213", VTI.Suffix),
- VTI, OpNode>,
- EVEX_V512, EVEX_CD8<VTI.EltSize, CD8VF>;
+ defm v213r : avx512_fma3p_rm<opc213, !strconcat(OpcodeStr, "213", VTI.Suffix),
+ VTI, OpNode>, EVEX_CD8<VTI.EltSize, CD8VF>;
- defm v231 : avx512_fma3p_rm<opc231, !strconcat(OpcodeStr, "231", VTI.Suffix),
- VTI>,
- EVEX_V512, EVEX_CD8<VTI.EltSize, CD8VF>;
+ defm v231r : avx512_fma3p_rm<opc231, !strconcat(OpcodeStr, "231", VTI.Suffix),
+ VTI>, EVEX_CD8<VTI.EltSize, CD8VF>;
}
+multiclass avx512_fma3p<bits<8> opc213, bits<8> opc231,
+ string OpcodeStr,
+ SDPatternOperator OpNode> {
let ExeDomain = SSEPackedSingle in {
- defm VFMADDPSZ : avx512_fma3p_forms<0xA8, 0xB8, "vfmadd",
- v16f32_info, X86Fmadd>;
- defm VFMSUBPSZ : avx512_fma3p_forms<0xAA, 0xBA, "vfmsub",
- v16f32_info, X86Fmsub>;
- defm VFMADDSUBPSZ : avx512_fma3p_forms<0xA6, 0xB6, "vfmaddsub",
- v16f32_info, X86Fmaddsub>;
- defm VFMSUBADDPSZ : avx512_fma3p_forms<0xA7, 0xB7, "vfmsubadd",
- v16f32_info, X86Fmsubadd>;
- defm VFNMADDPSZ : avx512_fma3p_forms<0xAC, 0xBC, "vfnmadd",
- v16f32_info, X86Fnmadd>;
- defm VFNMSUBPSZ : avx512_fma3p_forms<0xAE, 0xBE, "vfnmsub",
- v16f32_info, X86Fnmsub>;
-}
+ defm NAME##PSZ : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
+ v16f32_info, OpNode>, 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,
+ v4f32x_info, OpNode>, EVEX_V128;
+ }
let ExeDomain = SSEPackedDouble in {
- defm VFMADDPDZ : avx512_fma3p_forms<0xA8, 0xB8, "vfmadd",
- v8f64_info, X86Fmadd>, VEX_W;
- defm VFMSUBPDZ : avx512_fma3p_forms<0xAA, 0xBA, "vfmsub",
- v8f64_info, X86Fmsub>, VEX_W;
- defm VFMADDSUBPDZ : avx512_fma3p_forms<0xA6, 0xB6, "vfmaddsub",
- v8f64_info, X86Fmaddsub>, VEX_W;
- defm VFMSUBADDPDZ : avx512_fma3p_forms<0xA7, 0xB7, "vfmsubadd",
- v8f64_info, X86Fmsubadd>, VEX_W;
- defm VFNMADDPDZ : avx512_fma3p_forms<0xAC, 0xBC, "vfnmadd",
- v8f64_info, X86Fnmadd>, VEX_W;
- defm VFNMSUBPDZ : avx512_fma3p_forms<0xAE, 0xBE, "vfnmsub",
- v8f64_info, X86Fnmsub>, VEX_W;
+ defm NAME##PDZ : avx512_fma3p_forms<opc213, opc231, OpcodeStr,
+ v8f64_info, OpNode>, 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,
+ v2f64x_info, OpNode>, EVEX_V128, VEX_W;
+ }
}
+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>;
+
let Constraints = "$src1 = $dst" in {
multiclass avx512_fma3p_m132<bits<8> opc, string OpcodeStr, SDNode OpNode,
X86VectorVTInfo _> {
} // Constraints = "$src1 = $dst"
+multiclass avx512_fma3p_m132_f<bits<8> opc,
+ string OpcodeStr,
+ SDNode OpNode> {
+
let ExeDomain = SSEPackedSingle in {
- defm VFMADD132PSZ : avx512_fma3p_m132<0x98, "vfmadd132ps", X86Fmadd,
- v16f32_info>,
- EVEX_V512, EVEX_CD8<32, CD8VF>;
- defm VFMSUB132PSZ : avx512_fma3p_m132<0x9A, "vfmsub132ps", X86Fmsub,
- v16f32_info>,
- EVEX_V512, EVEX_CD8<32, CD8VF>;
- defm VFMADDSUB132PSZ : avx512_fma3p_m132<0x96, "vfmaddsub132ps", X86Fmaddsub,
- v16f32_info>,
- EVEX_V512, EVEX_CD8<32, CD8VF>;
- defm VFMSUBADD132PSZ : avx512_fma3p_m132<0x97, "vfmsubadd132ps", X86Fmsubadd,
- v16f32_info>,
- EVEX_V512, EVEX_CD8<32, CD8VF>;
- defm VFNMADD132PSZ : avx512_fma3p_m132<0x9C, "vfnmadd132ps", X86Fnmadd,
- v16f32_info>,
- EVEX_V512, EVEX_CD8<32, CD8VF>;
- defm VFNMSUB132PSZ : avx512_fma3p_m132<0x9E, "vfnmsub132ps", X86Fnmsub,
- v16f32_info>,
- EVEX_V512, EVEX_CD8<32, CD8VF>;
-}
+ defm NAME##PSZ : avx512_fma3p_m132<opc, OpcodeStr##ps,
+ OpNode,v16f32_info>, EVEX_V512, EVEX_CD8<32, CD8VF>;
+ defm NAME##PSZ256 : avx512_fma3p_m132<opc, OpcodeStr##ps,
+ OpNode, v8f32x_info>, EVEX_V256, EVEX_CD8<32, CD8VF>;
+ defm NAME##PSZ128 : avx512_fma3p_m132<opc, OpcodeStr##ps,
+ OpNode, v4f32x_info>, EVEX_V128, EVEX_CD8<32, CD8VF>;
+ }
let ExeDomain = SSEPackedDouble in {
- defm VFMADD132PDZ : avx512_fma3p_m132<0x98, "vfmadd132pd", X86Fmadd,
- v8f64_info>,
- EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
- defm VFMSUB132PDZ : avx512_fma3p_m132<0x9A, "vfmsub132pd", X86Fmsub,
- v8f64_info>,
- EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
- defm VFMADDSUB132PDZ : avx512_fma3p_m132<0x96, "vfmaddsub132pd", X86Fmaddsub,
- v8f64_info>,
- EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
- defm VFMSUBADD132PDZ : avx512_fma3p_m132<0x97, "vfmsubadd132pd", X86Fmsubadd,
- v8f64_info>,
- EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
- defm VFNMADD132PDZ : avx512_fma3p_m132<0x9C, "vfnmadd132pd", X86Fnmadd,
- v8f64_info>,
- EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
- defm VFNMSUB132PDZ : avx512_fma3p_m132<0x9E, "vfnmsub132pd", X86Fnmsub,
- v8f64_info>,
- EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+ defm NAME##PDZ : avx512_fma3p_m132<opc, OpcodeStr##pd,
+ OpNode, v8f64_info>, EVEX_V512, VEX_W, EVEX_CD8<32, CD8VF>;
+ defm NAME##PDZ256 : avx512_fma3p_m132<opc, OpcodeStr##pd,
+ OpNode, v4f64x_info>, EVEX_V256, VEX_W, EVEX_CD8<32, CD8VF>;
+ defm NAME##PDZ128 : avx512_fma3p_m132<opc, OpcodeStr##pd,
+ OpNode, v2f64x_info>, EVEX_V128, VEX_W, EVEX_CD8<32, CD8VF>;
+ }
}
+defm VFMADD132 : avx512_fma3p_m132_f<0x98, "vfmadd132", X86Fmadd>;
+defm VFMSUB132 : avx512_fma3p_m132_f<0x9A, "vfmsub132", X86Fmsub>;
+defm VFMADDSUB132 : avx512_fma3p_m132_f<0x96, "vfmaddsub132", X86Fmaddsub>;
+defm VFMSUBADD132 : avx512_fma3p_m132_f<0x97, "vfmsubadd132", X86Fmsubadd>;
+defm VFNMADD132 : avx512_fma3p_m132_f<0x9C, "vfnmadd132", X86Fnmadd>;
+defm VFNMSUB132 : avx512_fma3p_m132_f<0x9E, "vfnmsub132", X86Fnmsub>;
+
+
// Scalar FMA
let Constraints = "$src1 = $dst" in {
-multiclass avx512_fma3s_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
- RegisterClass RC, ValueType OpVT,
- X86MemOperand x86memop, Operand memop,
+multiclass avx512_fma3s_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ RegisterClass RC, ValueType OpVT,
+ X86MemOperand x86memop, Operand memop,
PatFrag mem_frag> {
let isCommutable = 1 in
def r : AVX512FMA3<opc, MRMSrcReg, (outs RC:$dst),
def : Pat<(f32 (fround FR64X:$src)), (VCVTSD2SSZrr FR64X:$src, FR64X:$src)>,
Requires<[HasAVX512]>;
-multiclass avx512_vcvt_fp_with_rc<bits<8> opc, string asm, RegisterClass SrcRC,
- RegisterClass DstRC, SDNode OpNode, PatFrag mem_frag,
+multiclass avx512_vcvt_fp_with_rc<bits<8> opc, string asm, RegisterClass SrcRC,
+ RegisterClass DstRC, SDNode OpNode, PatFrag mem_frag,
X86MemOperand x86memop, ValueType OpVT, ValueType InVT,
Domain d> {
let hasSideEffects = 0 in {
EVEX_CD8<32, CD8VH>;
def : Pat<(v8f64 (extloadv8f32 addr:$src)),
(VCVTPS2PDZrm addr:$src)>;
-
+
def : Pat<(v8f32 (int_x86_avx512_mask_cvtpd2ps_512 (v8f64 VR512:$src),
(bc_v8f32(v8i32 immAllZerosV)), (i8 -1), (i32 FROUND_CURRENT))),
(VCVTPD2PSZrr VR512:$src)>;
@@ -3892,7 +4018,7 @@ defm VCVTTPS2DQZ : avx512_vcvt_fp<0x5B, "vcvttps2dq", VR512, VR512, fp_to_sint,
EVEX_CD8<32, CD8VF>;
defm VCVTTPD2DQZ : avx512_vcvt_fp<0xE6, "vcvttpd2dq", VR512, VR256X, fp_to_sint,
- memopv8f64, f512mem, v8i32, v8f64,
+ memopv8f64, f512mem, v8i32, v8f64,
SSEPackedDouble>, EVEX_V512, PD, VEX_W,
EVEX_CD8<64, CD8VF>;
@@ -3910,7 +4036,7 @@ defm VCVTTPD2UDQZ : avx512_vcvt_fp<0x78, "vcvttpd2udq", VR512, VR256X, fp_to_uin
memopv8f64, f512mem, v8i32, v8f64,
SSEPackedDouble>, EVEX_V512, PS, VEX_W,
EVEX_CD8<64, CD8VF>;
-
+
// cvttpd2udq (src, 0, mask-all-ones, sae-current)
def : Pat<(v8i32 (int_x86_avx512_mask_cvttpd2udq_512 (v8f64 VR512:$src),
(v8i32 immAllZerosV), (i8 -1), FROUND_CURRENT)),
@@ -3920,16 +4046,16 @@ defm VCVTUDQ2PDZ : avx512_vcvt_fp<0x7A, "vcvtudq2pd", VR256X, VR512, uint_to_fp,
memopv4i64, 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,
SSEPackedSingle>, EVEX_V512, XD,
EVEX_CD8<32, CD8VF>;
def : Pat<(v8i32 (fp_to_uint (v8f32 VR256X:$src1))),
- (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
+ (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
(v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
-
+
def : Pat<(v4i32 (fp_to_uint (v4f32 VR128X:$src1))),
(EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
(v16f32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_xmm)>;
def : Pat<(v8f32 (uint_to_fp (v8i32 VR256X:$src1))),
(EXTRACT_SUBREG (v16f32 (VCVTUDQ2PSZrr
(v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
-
+
def : Pat<(v4f32 (uint_to_fp (v4i32 VR128X:$src1))),
(EXTRACT_SUBREG (v16f32 (VCVTUDQ2PSZrr
(v16i32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_xmm)>;
VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>;
}
}
-
+
/// avx512_fp14_s rcp14ss, rcp14sd, rsqrt14ss, rsqrt14sd
multiclass avx512_fp14_s<bits<8> opc, string OpcodeStr, RegisterClass RC,
X86MemOperand x86memop> {
(VRCP14PDZr VR512:$src)>;
/// avx512_fp28_s rcp28ss, rcp28sd, rsqrt28ss, rsqrt28sd
-multiclass avx512_fp28_s<bits<8> opc, string OpcodeStr, RegisterClass RC,
- X86MemOperand x86memop> {
- let hasSideEffects = 0, Predicates = [HasERI] in {
- def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, EVEX_4V;
- def rrb : AVX5128I<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2),
- !strconcat(OpcodeStr,
- "\t{{sae}, $src2, $src1, $dst|$dst, $src1, $src2, {sae}}"),
- []>, EVEX_4V, EVEX_B;
- let mayLoad = 1 in {
- def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, EVEX_4V;
- }
-}
-}
-
-defm VRCP28SS : avx512_fp28_s<0xCB, "vrcp28ss", FR32X, f32mem>,
- EVEX_CD8<32, CD8VT1>;
-defm VRCP28SD : avx512_fp28_s<0xCB, "vrcp28sd", FR64X, f64mem>,
- VEX_W, EVEX_CD8<64, CD8VT1>;
-defm VRSQRT28SS : avx512_fp28_s<0xCD, "vrsqrt28ss", FR32X, f32mem>,
- EVEX_CD8<32, CD8VT1>;
-defm VRSQRT28SD : avx512_fp28_s<0xCD, "vrsqrt28sd", FR64X, f64mem>,
- VEX_W, EVEX_CD8<64, CD8VT1>;
+multiclass avx512_fp28_s<bits<8> opc, string OpcodeStr,X86VectorVTInfo _,
+ SDNode OpNode> {
-def : Pat <(v4f32 (int_x86_avx512_rcp28_ss (v4f32 VR128X:$src1),
- (v4f32 VR128X:$src2), (bc_v4f32 (v4i32 immAllZerosV)), (i8 -1),
- FROUND_NO_EXC)),
- (COPY_TO_REGCLASS (VRCP28SSrrb (COPY_TO_REGCLASS VR128X:$src1, FR32X),
- (COPY_TO_REGCLASS VR128X:$src2, FR32X)), VR128X)>;
+ defm r : AVX512_maskable_scalar<opc, MRMSrcReg, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.RC:$src2), OpcodeStr,
+ "$src2, $src1", "$src1, $src2",
+ (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2),
+ (i32 FROUND_CURRENT))>;
-def : Pat <(v2f64 (int_x86_avx512_rcp28_sd (v2f64 VR128X:$src1),
- (v2f64 VR128X:$src2), (bc_v2f64 (v4i32 immAllZerosV)), (i8 -1),
- FROUND_NO_EXC)),
- (COPY_TO_REGCLASS (VRCP28SDrrb (COPY_TO_REGCLASS VR128X:$src1, FR64X),
- (COPY_TO_REGCLASS VR128X:$src2, FR64X)), VR128X)>;
+ defm rb : AVX512_maskable_scalar<opc, MRMSrcReg, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.RC:$src2), OpcodeStr,
+ "$src2, $src1", "$src1, $src2",
+ (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2),
+ (i32 FROUND_NO_EXC)), "{sae}">, EVEX_B;
-def : Pat <(v4f32 (int_x86_avx512_rsqrt28_ss (v4f32 VR128X:$src1),
- (v4f32 VR128X:$src2), (bc_v4f32 (v4i32 immAllZerosV)), (i8 -1),
- FROUND_NO_EXC)),
- (COPY_TO_REGCLASS (VRSQRT28SSrrb (COPY_TO_REGCLASS VR128X:$src1, FR32X),
- (COPY_TO_REGCLASS VR128X:$src2, FR32X)), VR128X)>;
+ defm m : AVX512_maskable_scalar<opc, MRMSrcMem, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.MemOp:$src2), OpcodeStr,
+ "$src2, $src1", "$src1, $src2",
+ (OpNode (_.VT _.RC:$src1),
+ (_.VT (scalar_to_vector (_.ScalarLdFrag addr:$src2))),
+ (i32 FROUND_CURRENT))>;
+}
-def : Pat <(v2f64 (int_x86_avx512_rsqrt28_sd (v2f64 VR128X:$src1),
- (v2f64 VR128X:$src2), (bc_v2f64 (v4i32 immAllZerosV)), (i8 -1),
- FROUND_NO_EXC)),
- (COPY_TO_REGCLASS (VRSQRT28SDrrb (COPY_TO_REGCLASS VR128X:$src1, FR64X),
- (COPY_TO_REGCLASS VR128X:$src2, FR64X)), VR128X)>;
+multiclass avx512_eri_s<bits<8> opc, string OpcodeStr, SDNode OpNode> {
+ defm SS : avx512_fp28_s<opc, OpcodeStr#"ss", f32x_info, OpNode>,
+ EVEX_CD8<32, CD8VT1>;
+ defm SD : avx512_fp28_s<opc, OpcodeStr#"sd", f64x_info, OpNode>,
+ EVEX_CD8<64, CD8VT1>, VEX_W;
+}
+let hasSideEffects = 0, Predicates = [HasERI] in {
+ defm VRCP28 : avx512_eri_s<0xCB, "vrcp28", X86rcp28s>, T8PD, EVEX_4V;
+ defm VRSQRT28 : avx512_eri_s<0xCD, "vrsqrt28", X86rsqrt28s>, T8PD, EVEX_4V;
+}
/// avx512_fp28_p rcp28ps, rcp28pd, rsqrt28ps, rsqrt28pd
multiclass avx512_fp28_p<bits<8> opc, string OpcodeStr, X86VectorVTInfo _,
defm rb : AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
(ins _.RC:$src), OpcodeStr,
"$src", "$src",
- (OpNode (_.VT _.RC:$src), (i32 FROUND_NO_EXC)), "{sae}">, EVEX_B;
+ (OpNode (_.VT _.RC:$src), (i32 FROUND_NO_EXC)),
+ "{sae}">, EVEX_B;
defm m : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
(ins _.MemOp:$src), OpcodeStr, "$src", "$src",
(OpNode (_.FloatVT
- (bitconvert (_.LdFrag addr:$src))), (i32 FROUND_CURRENT))>;
+ (bitconvert (_.LdFrag addr:$src))),
+ (i32 FROUND_CURRENT))>;
defm mb : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
(ins _.MemOp:$src), OpcodeStr, "$src", "$src",
}
let Predicates = [HasERI], hasSideEffects = 0 in {
-
+
defm VRSQRT28 : avx512_eri<0xCC, "vrsqrt28", X86rsqrt28>, EVEX, EVEX_V512, T8PD;
defm VRCP28 : avx512_eri<0xCA, "vrcp28", X86rcp28>, EVEX, EVEX_V512, T8PD;
defm VEXP2 : avx512_eri<0xC8, "vexp2", X86exp2>, EVEX, EVEX_V512, T8PD;
(ins VR128X:$src1, VR128X:$src2),
!strconcat(OpcodeStr,
"ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR128X:$dst,
+ [(set VR128X:$dst,
(F32Int VR128X:$src1, VR128X:$src2))],
itins_s.rr>, XS, EVEX_4V;
let mayLoad = 1 in {
(ins VR128X:$src1, ssmem:$src2),
!strconcat(OpcodeStr,
"ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR128X:$dst,
+ [(set VR128X:$dst,
(F32Int VR128X:$src1, sse_load_f32:$src2))],
itins_s.rm>, XS, EVEX_4V, EVEX_CD8<32, CD8VT1>;
}
(ins VR128X:$src1, VR128X:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR128X:$dst,
+ [(set VR128X:$dst,
(F64Int VR128X:$src1, VR128X:$src2))],
itins_s.rr>, XD, EVEX_4V, VEX_W;
let mayLoad = 1 in {
(ins VR128X:$src1, sdmem:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR128X:$dst,
- (F64Int VR128X:$src1, sse_load_f64:$src2))]>,
+ [(set VR128X:$dst,
+ (F64Int VR128X:$src1, sse_load_f64:$src2))]>,
XD, EVEX_4V, VEX_W, EVEX_CD8<64, CD8VT1>;
}
}
defm VSQRT : avx512_sqrt_packed_all<0x51, "vsqrt", fsqrt>;
-defm VSQRT : avx512_sqrt_scalar<0x51, "sqrt",
- int_x86_avx512_sqrt_ss, int_x86_avx512_sqrt_sd,
+defm VSQRT : avx512_sqrt_scalar<0x51, "sqrt",
+ int_x86_avx512_sqrt_ss, int_x86_avx512_sqrt_sd,
SSE_SQRTSS, SSE_SQRTSD>;
let Predicates = [HasAVX512] in {
def : Pat<(v8f64 (int_x86_avx512_sqrt_pd_512 (v8f64 VR512:$src1),
(bc_v8f64 (v16i32 immAllZerosV)), (i8 -1), FROUND_CURRENT)),
(VSQRTPDZr VR512:$src1)>;
-
+
def : Pat<(f32 (fsqrt FR32X:$src)),
(VSQRTSSZr (f32 (IMPLICIT_DEF)), FR32X:$src)>;
def : Pat<(f32 (fsqrt (load addr:$src))),
(ins VR128X:$src1, ssmem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
- [(set VR128X:$dst, (F32Int VR128X:$src1,
+ [(set VR128X:$dst, (F32Int VR128X:$src1,
sse_load_f32:$src2, imm:$src3))]>,
EVEX_CD8<32, CD8VT1>;
defm VRNDSCALESS : avx512_rndscale_scalar<0x0A, "vrndscaless", ssmem, FR32X,
SSEPackedSingle>, EVEX_CD8<32, CD8VT1>;
-
+
defm VRNDSCALESD : avx512_rndscale_scalar<0x0B, "vrndscalesd", sdmem, FR64X,
SSEPackedDouble>, EVEX_CD8<64, CD8VT1>;
[]>, EVEX, EVEX_K;
}
-defm VPMOVQB : avx512_trunc_sat<0x32, "vpmovqb", VR128X, VR512, VK8WM,
+defm VPMOVQB : avx512_trunc_sat<0x32, "vpmovqb", VR128X, VR512, VK8WM,
i128mem>, EVEX_V512, EVEX_CD8<8, CD8VO>;
defm VPMOVSQB : avx512_trunc_sat<0x22, "vpmovsqb", VR128X, VR512, VK8WM,
i128mem>, EVEX_V512, EVEX_CD8<8, CD8VO>;
defm VGATHERQPSZ : avx512_gather<0x93, "vgatherqps", VK8WM, VR256X, vz64mem>,
EVEX_V512, EVEX_CD8<32, CD8VT1>;
}
-
+
defm VPGATHERDQZ : avx512_gather<0x90, "vpgatherdq", VK8WM, VR512, vy64xmem>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>;
defm VPGATHERDDZ : avx512_gather<0x90, "vpgatherdd", VK16WM, VR512, vz32mem>,
@@ -4898,7 +5006,7 @@ defm VGATHERPF0DPD: avx512_gather_scatter_prefetch<0xC6, MRM1m, "vgatherpf0dpd",
defm VGATHERPF0QPD: avx512_gather_scatter_prefetch<0xC7, MRM1m, "vgatherpf0qpd",
VK8WM, vz64mem>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>;
-
+
defm VGATHERPF1DPS: avx512_gather_scatter_prefetch<0xC6, MRM2m, "vgatherpf1dps",
VK16WM, vz32mem>, EVEX_V512, EVEX_CD8<32, CD8VT1>;
ValueType vt, string OpcodeStr, PatFrag mem_frag,
Domain d> {
def rmi : AVX512PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2, i8imm:$src3),
+ (ins RC:$src1, x86memop:$src2, u8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
(i8 imm:$src3))))], d, IIC_SSE_SHUFP>,
EVEX_4V, Sched<[WriteShuffleLd, ReadAfterLd]>;
def rri : AVX512PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2, i8imm:$src3),
+ (ins RC:$src1, RC:$src2, u8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
multiclass avx512_valign<X86VectorVTInfo _> {
defm rri : AVX512_maskable<0x03, MRMSrcReg, _, (outs _.RC:$dst),
- (ins _.RC:$src1, _.RC:$src2, i8imm:$src3),
+ (ins _.RC:$src1, _.RC:$src2, u8imm:$src3),
"valign"##_.Suffix,
"$src3, $src2, $src1", "$src1, $src2, $src3",
(_.VT (X86VAlign _.RC:$src2, _.RC:$src1,
let mayLoad = 1 in
def rmi : AVX512AIi8<0x03, MRMSrcMem, (outs _.RC:$dst),
- (ins _.RC:$src1, _.MemOp:$src2, i8imm:$src3),
+ (ins _.RC:$src1, _.MemOp:$src2, u8imm:$src3),
!strconcat("valign"##_.Suffix,
"\t{$src3, $src2, $src1, $dst|"
"$dst, $src1, $src2, $src3}"),
(bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
(VPABSQZrr VR512:$src)>;
-multiclass avx512_conflict<bits<8> opc, string OpcodeStr,
+multiclass avx512_conflict<bits<8> opc, string OpcodeStr,
RegisterClass RC, RegisterClass KRC,
X86MemOperand x86memop,
X86MemOperand x86scalar_mop, string BrdcstStr> {
", ${dst} {${mask}} {z}|${dst} {${mask}} {z}, ${src}",
BrdcstStr, "}"),
[]>, EVEX, EVEX_KZ, EVEX_B;
-
+
let Constraints = "$src1 = $dst" in {
def rrk : AVX5128I<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, KRC:$mask, RC:$src2),
!strconcat(OpcodeStr##Vec.Suffix, "\t{$src, $dst|$dst, $src}"),
[(set Vec.RC:$dst, (Vec.VT (X86vsext Vec.KRC:$src)))]>, EVEX;
}
-
+
multiclass cvt_mask_by_elt_width<bits<8> opc, AVX512VLVectorVTInfo VTInfo,
string OpcodeStr, Predicate prd> {
let Predicates = [prd] in
defm NAME##Q : cvt_mask_by_elt_width<0x38, avx512vl_i64_info, OpcodeStr,
HasDQI>, VEX_W;
}
-
+
defm VPMOVM2 : avx512_convert_mask_to_vector<"vpmovm2">;
+
+//===----------------------------------------------------------------------===//
+// AVX-512 - COMPRESS and EXPAND
+//
+multiclass compress_by_vec_width<bits<8> opc, X86VectorVTInfo _,
+ string OpcodeStr> {
+ def rrkz : AVX5128I<opc, MRMDestReg, (outs _.RC:$dst),
+ (ins _.KRCWM:$mask, _.RC:$src),
+ OpcodeStr # "\t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}",
+ [(set _.RC:$dst, (_.VT (X86compress _.KRCWM:$mask, _.RC:$src,
+ _.ImmAllZerosV)))]>, EVEX_KZ;
+
+ let Constraints = "$src0 = $dst" in
+ def rrk : AVX5128I<opc, MRMDestReg, (outs _.RC:$dst),
+ (ins _.RC:$src0, _.KRCWM:$mask, _.RC:$src),
+ OpcodeStr # "\t{$src, $dst {${mask}} |$dst {${mask}}, $src}",
+ [(set _.RC:$dst, (_.VT (X86compress _.KRCWM:$mask, _.RC:$src,
+ _.RC:$src0)))]>, EVEX_K;
+
+ let mayStore = 1 in {
+ def mrk : AVX5128I<opc, MRMDestMem, (outs),
+ (ins _.MemOp:$dst, _.KRCWM:$mask, _.RC:$src),
+ OpcodeStr # "\t{$src, $dst {${mask}} |$dst {${mask}}, $src}",
+ [(store (_.VT (X86compress _.KRCWM:$mask, _.RC:$src, undef)),
+ addr:$dst)]>,
+ EVEX_K, EVEX_CD8<_.EltSize, CD8VT1>;
+ }
+}
+
+multiclass compress_by_elt_width<bits<8> opc, string OpcodeStr,
+ AVX512VLVectorVTInfo VTInfo> {
+ defm Z : compress_by_vec_width<opc, VTInfo.info512, OpcodeStr>, EVEX_V512;
+
+ let Predicates = [HasVLX] in {
+ defm Z256 : compress_by_vec_width<opc, VTInfo.info256, OpcodeStr>, EVEX_V256;
+ defm Z128 : compress_by_vec_width<opc, VTInfo.info128, OpcodeStr>, EVEX_V128;
+ }
+}
+
+defm VPCOMPRESSD : compress_by_elt_width <0x8B, "vpcompressd", avx512vl_i32_info>,
+ EVEX;
+defm VPCOMPRESSQ : compress_by_elt_width <0x8B, "vpcompressq", avx512vl_i64_info>,
+ EVEX, VEX_W;
+defm VCOMPRESSPS : compress_by_elt_width <0x8A, "vcompressps", avx512vl_f32_info>,
+ EVEX;
+defm VCOMPRESSPD : compress_by_elt_width <0x8A, "vcompresspd", avx512vl_f64_info>,
+ EVEX, VEX_W;
+
+// expand
+multiclass expand_by_vec_width<bits<8> opc, X86VectorVTInfo _,
+ string OpcodeStr> {
+ def rrkz : AVX5128I<opc, MRMSrcReg, (outs _.RC:$dst),
+ (ins _.KRCWM:$mask, _.RC:$src),
+ OpcodeStr # "\t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}",
+ [(set _.RC:$dst, (_.VT (X86expand _.KRCWM:$mask, (_.VT _.RC:$src),
+ _.ImmAllZerosV)))]>, EVEX_KZ;
+
+ let Constraints = "$src0 = $dst" in
+ def rrk : AVX5128I<opc, MRMSrcReg, (outs _.RC:$dst),
+ (ins _.RC:$src0, _.KRCWM:$mask, _.RC:$src),
+ OpcodeStr # "\t{$src, $dst {${mask}} |$dst {${mask}}, $src}",
+ [(set _.RC:$dst, (_.VT (X86expand _.KRCWM:$mask,
+ (_.VT _.RC:$src), _.RC:$src0)))]>, EVEX_K;
+
+ let mayLoad = 1, Constraints = "$src0 = $dst" in
+ def rmk : AVX5128I<opc, MRMSrcMem, (outs _.RC:$dst),
+ (ins _.RC:$src0, _.KRCWM:$mask, _.MemOp:$src),
+ OpcodeStr # "\t{$src, $dst {${mask}} |$dst {${mask}}, $src}",
+ [(set _.RC:$dst, (_.VT (X86expand _.KRCWM:$mask,
+ (_.VT (bitconvert
+ (_.LdFrag addr:$src))),
+ _.RC:$src0)))]>,
+ EVEX_K, EVEX_CD8<_.EltSize, CD8VT1>;
+
+ let mayLoad = 1 in
+ def rmkz : AVX5128I<opc, MRMSrcMem, (outs _.RC:$dst),
+ (ins _.KRCWM:$mask, _.MemOp:$src),
+ OpcodeStr # "\t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}",
+ [(set _.RC:$dst, (_.VT (X86expand _.KRCWM:$mask,
+ (_.VT (bitconvert (_.LdFrag addr:$src))),
+ _.ImmAllZerosV)))]>,
+ EVEX_KZ, EVEX_CD8<_.EltSize, CD8VT1>;
+
+}
+
+multiclass expand_by_elt_width<bits<8> opc, string OpcodeStr,
+ AVX512VLVectorVTInfo VTInfo> {
+ defm Z : expand_by_vec_width<opc, VTInfo.info512, OpcodeStr>, EVEX_V512;
+
+ let Predicates = [HasVLX] in {
+ defm Z256 : expand_by_vec_width<opc, VTInfo.info256, OpcodeStr>, EVEX_V256;
+ defm Z128 : expand_by_vec_width<opc, VTInfo.info128, OpcodeStr>, EVEX_V128;
+ }
+}
+
+defm VPEXPANDD : expand_by_elt_width <0x89, "vpexpandd", avx512vl_i32_info>,
+ EVEX;
+defm VPEXPANDQ : expand_by_elt_width <0x89, "vpexpandq", avx512vl_i64_info>,
+ EVEX, VEX_W;
+defm VEXPANDPS : expand_by_elt_width <0x88, "vexpandps", avx512vl_f32_info>,
+ EVEX;
+defm VEXPANDPD : expand_by_elt_width <0x88, "vexpandpd", avx512vl_f64_info>,
+ EVEX, VEX_W;