xref: /freebsd/contrib/llvm-project/llvm/lib/Target/X86/X86InstrXOP.td (revision d4eeb02986980bf33dd56c41ceb9fc5f180c0d47)
1//===-- X86InstrXOP.td - XOP Instruction Set ---------------*- tablegen -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file describes XOP (eXtended OPerations)
10//
11//===----------------------------------------------------------------------===//
12
13multiclass xop2op<bits<8> opc, string OpcodeStr, Intrinsic Int> {
14  def rr : IXOP<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
15           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
16           [(set VR128:$dst, (Int VR128:$src))]>, XOP, Sched<[SchedWritePHAdd.XMM]>;
17  def rm : IXOP<opc, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
18           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
19           [(set VR128:$dst, (Int (load addr:$src)))]>, XOP,
20           Sched<[SchedWritePHAdd.XMM.Folded, SchedWritePHAdd.XMM.ReadAfterFold]>;
21}
22
23let ExeDomain = SSEPackedInt in {
24  defm VPHSUBWD  : xop2op<0xE2, "vphsubwd", int_x86_xop_vphsubwd>;
25  defm VPHSUBDQ  : xop2op<0xE3, "vphsubdq", int_x86_xop_vphsubdq>;
26  defm VPHSUBBW  : xop2op<0xE1, "vphsubbw", int_x86_xop_vphsubbw>;
27  defm VPHADDWQ  : xop2op<0xC7, "vphaddwq", int_x86_xop_vphaddwq>;
28  defm VPHADDWD  : xop2op<0xC6, "vphaddwd", int_x86_xop_vphaddwd>;
29  defm VPHADDUWQ : xop2op<0xD7, "vphadduwq", int_x86_xop_vphadduwq>;
30  defm VPHADDUWD : xop2op<0xD6, "vphadduwd", int_x86_xop_vphadduwd>;
31  defm VPHADDUDQ : xop2op<0xDB, "vphaddudq", int_x86_xop_vphaddudq>;
32  defm VPHADDUBW : xop2op<0xD1, "vphaddubw", int_x86_xop_vphaddubw>;
33  defm VPHADDUBQ : xop2op<0xD3, "vphaddubq", int_x86_xop_vphaddubq>;
34  defm VPHADDUBD : xop2op<0xD2, "vphaddubd", int_x86_xop_vphaddubd>;
35  defm VPHADDDQ  : xop2op<0xCB, "vphadddq", int_x86_xop_vphadddq>;
36  defm VPHADDBW  : xop2op<0xC1, "vphaddbw", int_x86_xop_vphaddbw>;
37  defm VPHADDBQ  : xop2op<0xC3, "vphaddbq", int_x86_xop_vphaddbq>;
38  defm VPHADDBD  : xop2op<0xC2, "vphaddbd", int_x86_xop_vphaddbd>;
39}
40
41// Scalar load 2 addr operand instructions
42multiclass xop2opsld<bits<8> opc, string OpcodeStr, Intrinsic Int,
43                     Operand memop, PatFrags mem_frags,
44                     X86FoldableSchedWrite sched> {
45  def rr : IXOP<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
46           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
47           [(set VR128:$dst, (Int VR128:$src))]>, XOP, Sched<[sched]>;
48  def rm : IXOP<opc, MRMSrcMem, (outs VR128:$dst), (ins memop:$src),
49           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
50           [(set VR128:$dst, (Int (mem_frags addr:$src)))]>, XOP,
51           Sched<[sched.Folded, sched.ReadAfterFold]>;
52}
53
54multiclass xop2op128<bits<8> opc, string OpcodeStr, Intrinsic Int,
55                     X86FoldableSchedWrite sched> {
56  def rr : IXOP<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
57           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
58           [(set VR128:$dst, (Int VR128:$src))]>, XOP, Sched<[sched]>;
59  def rm : IXOP<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
60           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
61           [(set VR128:$dst, (Int (load addr:$src)))]>, XOP,
62           Sched<[sched.Folded, sched.ReadAfterFold]>;
63}
64
65multiclass xop2op256<bits<8> opc, string OpcodeStr, Intrinsic Int,
66                     X86FoldableSchedWrite sched> {
67  def Yrr : IXOP<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
68           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
69           [(set VR256:$dst, (Int VR256:$src))]>, XOP, VEX_L, Sched<[sched]>;
70  def Yrm : IXOP<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
71           !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
72           [(set VR256:$dst, (Int (load addr:$src)))]>, XOP, VEX_L,
73           Sched<[sched.Folded, sched.ReadAfterFold]>;
74}
75
76let ExeDomain = SSEPackedSingle in {
77  defm VFRCZSS : xop2opsld<0x82, "vfrczss", int_x86_xop_vfrcz_ss,
78                           ssmem, sse_load_f32, SchedWriteFRnd.Scl>;
79  defm VFRCZPS : xop2op128<0x80, "vfrczps", int_x86_xop_vfrcz_ps,
80                           SchedWriteFRnd.XMM>;
81  defm VFRCZPS : xop2op256<0x80, "vfrczps", int_x86_xop_vfrcz_ps_256,
82                           SchedWriteFRnd.YMM>;
83}
84
85let ExeDomain = SSEPackedDouble in {
86  defm VFRCZSD : xop2opsld<0x83, "vfrczsd", int_x86_xop_vfrcz_sd,
87                           sdmem, sse_load_f64, SchedWriteFRnd.Scl>;
88  defm VFRCZPD : xop2op128<0x81, "vfrczpd", int_x86_xop_vfrcz_pd,
89                           SchedWriteFRnd.XMM>;
90  defm VFRCZPD : xop2op256<0x81, "vfrczpd", int_x86_xop_vfrcz_pd_256,
91                           SchedWriteFRnd.YMM>;
92}
93
94multiclass xop3op<bits<8> opc, string OpcodeStr, SDNode OpNode,
95                  ValueType vt128, X86FoldableSchedWrite sched> {
96  def rr : IXOP<opc, MRMSrcReg4VOp3, (outs VR128:$dst),
97           (ins VR128:$src1, VR128:$src2),
98           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
99           [(set VR128:$dst,
100              (vt128 (OpNode (vt128 VR128:$src1), (vt128 VR128:$src2))))]>,
101           XOP, Sched<[sched]>;
102  def rm : IXOP<opc, MRMSrcMem, (outs VR128:$dst),
103           (ins VR128:$src1, i128mem:$src2),
104           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
105           [(set VR128:$dst,
106              (vt128 (OpNode (vt128 VR128:$src1),
107                             (vt128 (load addr:$src2)))))]>,
108           XOP_4V, VEX_W, Sched<[sched.Folded, sched.ReadAfterFold]>;
109  def mr : IXOP<opc, MRMSrcMem4VOp3, (outs VR128:$dst),
110           (ins i128mem:$src1, VR128:$src2),
111           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
112           [(set VR128:$dst,
113              (vt128 (OpNode (vt128 (load addr:$src1)),
114                             (vt128 VR128:$src2))))]>,
115             XOP, Sched<[sched.Folded, sched.ReadAfterFold]>;
116  // For disassembler
117  let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
118  def rr_REV : IXOP<opc, MRMSrcReg, (outs VR128:$dst),
119               (ins VR128:$src1, VR128:$src2),
120               !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
121               []>,
122               XOP_4V, VEX_W, Sched<[sched]>, FoldGenData<NAME#rr>;
123}
124
125let ExeDomain = SSEPackedInt in {
126  defm VPROTB : xop3op<0x90, "vprotb", rotl, v16i8, SchedWriteVarVecShift.XMM>;
127  defm VPROTD : xop3op<0x92, "vprotd", rotl, v4i32, SchedWriteVarVecShift.XMM>;
128  defm VPROTQ : xop3op<0x93, "vprotq", rotl, v2i64, SchedWriteVarVecShift.XMM>;
129  defm VPROTW : xop3op<0x91, "vprotw", rotl, v8i16, SchedWriteVarVecShift.XMM>;
130  defm VPSHAB : xop3op<0x98, "vpshab", X86vpsha, v16i8, SchedWriteVarVecShift.XMM>;
131  defm VPSHAD : xop3op<0x9A, "vpshad", X86vpsha, v4i32, SchedWriteVarVecShift.XMM>;
132  defm VPSHAQ : xop3op<0x9B, "vpshaq", X86vpsha, v2i64, SchedWriteVarVecShift.XMM>;
133  defm VPSHAW : xop3op<0x99, "vpshaw", X86vpsha, v8i16, SchedWriteVarVecShift.XMM>;
134  defm VPSHLB : xop3op<0x94, "vpshlb", X86vpshl, v16i8, SchedWriteVarVecShift.XMM>;
135  defm VPSHLD : xop3op<0x96, "vpshld", X86vpshl, v4i32, SchedWriteVarVecShift.XMM>;
136  defm VPSHLQ : xop3op<0x97, "vpshlq", X86vpshl, v2i64, SchedWriteVarVecShift.XMM>;
137  defm VPSHLW : xop3op<0x95, "vpshlw", X86vpshl, v8i16, SchedWriteVarVecShift.XMM>;
138}
139
140multiclass xop3opimm<bits<8> opc, string OpcodeStr, SDNode OpNode,
141                     ValueType vt128, X86FoldableSchedWrite sched> {
142  def ri : IXOPi8<opc, MRMSrcReg, (outs VR128:$dst),
143           (ins VR128:$src1, u8imm:$src2),
144           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
145           [(set VR128:$dst,
146              (vt128 (OpNode (vt128 VR128:$src1), timm:$src2)))]>,
147           XOP, Sched<[sched]>;
148  def mi : IXOPi8<opc, MRMSrcMem, (outs VR128:$dst),
149           (ins i128mem:$src1, u8imm:$src2),
150           !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
151           [(set VR128:$dst,
152              (vt128 (OpNode (vt128 (load addr:$src1)), timm:$src2)))]>,
153           XOP, Sched<[sched.Folded, sched.ReadAfterFold]>;
154}
155
156let ExeDomain = SSEPackedInt in {
157  defm VPROTB : xop3opimm<0xC0, "vprotb", X86vrotli, v16i8,
158                          SchedWriteVecShiftImm.XMM>;
159  defm VPROTD : xop3opimm<0xC2, "vprotd", X86vrotli, v4i32,
160                          SchedWriteVecShiftImm.XMM>;
161  defm VPROTQ : xop3opimm<0xC3, "vprotq", X86vrotli, v2i64,
162                          SchedWriteVecShiftImm.XMM>;
163  defm VPROTW : xop3opimm<0xC1, "vprotw", X86vrotli, v8i16,
164                          SchedWriteVecShiftImm.XMM>;
165}
166
167// Instruction where second source can be memory, but third must be register
168multiclass xop4opm2<bits<8> opc, string OpcodeStr, Intrinsic Int,
169                    X86FoldableSchedWrite sched> {
170  let isCommutable = 1 in
171  def rr : IXOPi8Reg<opc, MRMSrcReg, (outs VR128:$dst),
172           (ins VR128:$src1, VR128:$src2, VR128:$src3),
173           !strconcat(OpcodeStr,
174           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
175           [(set VR128:$dst,
176              (Int VR128:$src1, VR128:$src2, VR128:$src3))]>, XOP_4V,
177           Sched<[sched]>;
178  def rm : IXOPi8Reg<opc, MRMSrcMem, (outs VR128:$dst),
179           (ins VR128:$src1, i128mem:$src2, VR128:$src3),
180           !strconcat(OpcodeStr,
181           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
182           [(set VR128:$dst,
183              (Int VR128:$src1, (load addr:$src2),
184              VR128:$src3))]>, XOP_4V, Sched<[sched.Folded, sched.ReadAfterFold]>;
185}
186
187let ExeDomain = SSEPackedInt in {
188  defm VPMADCSWD  : xop4opm2<0xB6, "vpmadcswd",
189                             int_x86_xop_vpmadcswd, SchedWriteVecIMul.XMM>;
190  defm VPMADCSSWD : xop4opm2<0xA6, "vpmadcsswd",
191                             int_x86_xop_vpmadcsswd, SchedWriteVecIMul.XMM>;
192  defm VPMACSWW   : xop4opm2<0x95, "vpmacsww",
193                             int_x86_xop_vpmacsww, SchedWriteVecIMul.XMM>;
194  defm VPMACSWD   : xop4opm2<0x96, "vpmacswd",
195                             int_x86_xop_vpmacswd, SchedWriteVecIMul.XMM>;
196  defm VPMACSSWW  : xop4opm2<0x85, "vpmacssww",
197                             int_x86_xop_vpmacssww, SchedWriteVecIMul.XMM>;
198  defm VPMACSSWD  : xop4opm2<0x86, "vpmacsswd",
199                             int_x86_xop_vpmacsswd, SchedWriteVecIMul.XMM>;
200  defm VPMACSSDQL : xop4opm2<0x87, "vpmacssdql",
201                             int_x86_xop_vpmacssdql, SchedWritePMULLD.XMM>;
202  defm VPMACSSDQH : xop4opm2<0x8F, "vpmacssdqh",
203                             int_x86_xop_vpmacssdqh, SchedWritePMULLD.XMM>;
204  defm VPMACSSDD  : xop4opm2<0x8E, "vpmacssdd",
205                             int_x86_xop_vpmacssdd, SchedWritePMULLD.XMM>;
206  defm VPMACSDQL  : xop4opm2<0x97, "vpmacsdql",
207                             int_x86_xop_vpmacsdql, SchedWritePMULLD.XMM>;
208  defm VPMACSDQH  : xop4opm2<0x9F, "vpmacsdqh",
209                             int_x86_xop_vpmacsdqh, SchedWritePMULLD.XMM>;
210  defm VPMACSDD   : xop4opm2<0x9E, "vpmacsdd",
211                             int_x86_xop_vpmacsdd, SchedWritePMULLD.XMM>;
212}
213
214// IFMA patterns - for cases where we can safely ignore the overflow bits from
215// the multiply or easily match with existing intrinsics.
216let Predicates = [HasXOP] in {
217  def : Pat<(v8i16 (add (mul (v8i16 VR128:$src1), (v8i16 VR128:$src2)),
218                        (v8i16 VR128:$src3))),
219            (VPMACSWWrr VR128:$src1, VR128:$src2, VR128:$src3)>;
220  def : Pat<(v4i32 (add (mul (v4i32 VR128:$src1), (v4i32 VR128:$src2)),
221                        (v4i32 VR128:$src3))),
222            (VPMACSDDrr VR128:$src1, VR128:$src2, VR128:$src3)>;
223  def : Pat<(v2i64 (add (X86pmuldq (bc_v2i64 (X86PShufd (v4i32 VR128:$src1), (i8 -11))),
224                                   (bc_v2i64 (X86PShufd (v4i32 VR128:$src2), (i8 -11)))),
225                        (v2i64 VR128:$src3))),
226            (VPMACSDQHrr VR128:$src1, VR128:$src2, VR128:$src3)>;
227  def : Pat<(v2i64 (add (X86pmuldq (v2i64 VR128:$src1), (v2i64 VR128:$src2)),
228                        (v2i64 VR128:$src3))),
229            (VPMACSDQLrr VR128:$src1, VR128:$src2, VR128:$src3)>;
230  def : Pat<(v4i32 (add (X86vpmaddwd (v8i16 VR128:$src1), (v8i16 VR128:$src2)),
231                        (v4i32 VR128:$src3))),
232            (VPMADCSWDrr VR128:$src1, VR128:$src2, VR128:$src3)>;
233}
234
235// Transforms to swizzle an immediate to help matching memory operand in first
236// operand.
237def CommuteVPCOMCC : SDNodeXForm<imm, [{
238  uint8_t Imm = N->getZExtValue() & 0x7;
239  Imm = X86::getSwappedVPCOMImm(Imm);
240  return getI8Imm(Imm, SDLoc(N));
241}]>;
242
243// Instruction where second source can be memory, third must be imm8
244multiclass xopvpcom<bits<8> opc, string Suffix, SDNode OpNode, ValueType vt128,
245                    X86FoldableSchedWrite sched> {
246  let ExeDomain = SSEPackedInt in { // SSE integer instructions
247    let isCommutable = 1 in
248    def ri : IXOPi8<opc, MRMSrcReg, (outs VR128:$dst),
249             (ins VR128:$src1, VR128:$src2, u8imm:$cc),
250             !strconcat("vpcom", Suffix,
251             "\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}"),
252             [(set VR128:$dst,
253                (vt128 (OpNode (vt128 VR128:$src1), (vt128 VR128:$src2),
254                               timm:$cc)))]>,
255             XOP_4V, Sched<[sched]>;
256    def mi : IXOPi8<opc, MRMSrcMem, (outs VR128:$dst),
257             (ins VR128:$src1, i128mem:$src2, u8imm:$cc),
258             !strconcat("vpcom", Suffix,
259             "\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}"),
260             [(set VR128:$dst,
261                (vt128 (OpNode (vt128 VR128:$src1),
262                               (vt128 (load addr:$src2)),
263                                timm:$cc)))]>,
264             XOP_4V, Sched<[sched.Folded, sched.ReadAfterFold]>;
265  }
266
267  def : Pat<(OpNode (load addr:$src2),
268                    (vt128 VR128:$src1), timm:$cc),
269            (!cast<Instruction>(NAME#"mi") VR128:$src1, addr:$src2,
270                                           (CommuteVPCOMCC timm:$cc))>;
271}
272
273defm VPCOMB  : xopvpcom<0xCC, "b", X86vpcom, v16i8, SchedWriteVecALU.XMM>;
274defm VPCOMW  : xopvpcom<0xCD, "w", X86vpcom, v8i16, SchedWriteVecALU.XMM>;
275defm VPCOMD  : xopvpcom<0xCE, "d", X86vpcom, v4i32, SchedWriteVecALU.XMM>;
276defm VPCOMQ  : xopvpcom<0xCF, "q", X86vpcom, v2i64, SchedWriteVecALU.XMM>;
277defm VPCOMUB : xopvpcom<0xEC, "ub", X86vpcomu, v16i8, SchedWriteVecALU.XMM>;
278defm VPCOMUW : xopvpcom<0xED, "uw", X86vpcomu, v8i16, SchedWriteVecALU.XMM>;
279defm VPCOMUD : xopvpcom<0xEE, "ud", X86vpcomu, v4i32, SchedWriteVecALU.XMM>;
280defm VPCOMUQ : xopvpcom<0xEF, "uq", X86vpcomu, v2i64, SchedWriteVecALU.XMM>;
281
282multiclass xop4op<bits<8> opc, string OpcodeStr, SDNode OpNode,
283                  ValueType vt128, X86FoldableSchedWrite sched> {
284  def rrr : IXOPi8Reg<opc, MRMSrcReg, (outs VR128:$dst),
285            (ins VR128:$src1, VR128:$src2, VR128:$src3),
286            !strconcat(OpcodeStr,
287            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
288            [(set VR128:$dst,
289              (vt128 (OpNode (vt128 VR128:$src1), (vt128 VR128:$src2),
290                             (vt128 VR128:$src3))))]>,
291            XOP_4V, Sched<[sched]>;
292  def rrm : IXOPi8Reg<opc, MRMSrcMemOp4, (outs VR128:$dst),
293            (ins VR128:$src1, VR128:$src2, i128mem:$src3),
294            !strconcat(OpcodeStr,
295            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
296            [(set VR128:$dst,
297              (vt128 (OpNode (vt128 VR128:$src1), (vt128 VR128:$src2),
298                             (vt128 (load addr:$src3)))))]>,
299            XOP_4V, VEX_W, Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
300  def rmr : IXOPi8Reg<opc, MRMSrcMem, (outs VR128:$dst),
301            (ins VR128:$src1, i128mem:$src2, VR128:$src3),
302            !strconcat(OpcodeStr,
303            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
304            [(set VR128:$dst,
305              (v16i8 (OpNode (vt128 VR128:$src1), (vt128 (load addr:$src2)),
306                             (vt128 VR128:$src3))))]>,
307            XOP_4V, Sched<[sched.Folded, sched.ReadAfterFold,
308                           // 128mem:$src2
309                           ReadDefault, ReadDefault, ReadDefault, ReadDefault,
310                           ReadDefault,
311                           // VR128:$src3
312                           sched.ReadAfterFold]>;
313  // For disassembler
314  let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
315  def rrr_REV : IXOPi8Reg<opc, MRMSrcRegOp4, (outs VR128:$dst),
316                (ins VR128:$src1, VR128:$src2, VR128:$src3),
317                !strconcat(OpcodeStr,
318                "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
319                []>, XOP_4V, VEX_W, Sched<[sched]>, FoldGenData<NAME#rrr>;
320}
321
322let ExeDomain = SSEPackedInt in {
323  defm VPPERM : xop4op<0xA3, "vpperm", X86vpperm, v16i8,
324                       SchedWriteVarShuffle.XMM>;
325}
326
327// Instruction where either second or third source can be memory
328multiclass xop4op_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
329                      X86MemOperand x86memop, ValueType VT,
330                      X86FoldableSchedWrite sched> {
331  def rrr : IXOPi8Reg<opc, MRMSrcReg, (outs RC:$dst),
332            (ins RC:$src1, RC:$src2, RC:$src3),
333            !strconcat(OpcodeStr,
334            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
335            [(set RC:$dst, (VT (or (and RC:$src3, RC:$src1),
336                                   (X86andnp RC:$src3, RC:$src2))))]>, XOP_4V,
337            Sched<[sched]>;
338  // FIXME: We can't write a pattern for this in tablegen.
339  let hasSideEffects = 0, mayLoad = 1 in
340  def rrm : IXOPi8Reg<opc, MRMSrcMemOp4, (outs RC:$dst),
341            (ins RC:$src1, RC:$src2, x86memop:$src3),
342            !strconcat(OpcodeStr,
343            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
344            []>,
345            XOP_4V, VEX_W, Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
346  def rmr : IXOPi8Reg<opc, MRMSrcMem, (outs RC:$dst),
347            (ins RC:$src1, x86memop:$src2, RC:$src3),
348            !strconcat(OpcodeStr,
349            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
350            [(set RC:$dst, (VT (or (and RC:$src3, RC:$src1),
351                                   (X86andnp RC:$src3, (load addr:$src2)))))]>,
352            XOP_4V, Sched<[sched.Folded, sched.ReadAfterFold,
353                           // x86memop:$src2
354                           ReadDefault, ReadDefault, ReadDefault, ReadDefault,
355                           ReadDefault,
356                           // RC::$src3
357                           sched.ReadAfterFold]>;
358  // For disassembler
359  let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
360  def rrr_REV : IXOPi8Reg<opc, MRMSrcRegOp4, (outs RC:$dst),
361            (ins RC:$src1, RC:$src2, RC:$src3),
362            !strconcat(OpcodeStr,
363            "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
364            []>, XOP_4V, VEX_W, Sched<[sched]>, FoldGenData<NAME#rrr>;
365}
366
367let ExeDomain = SSEPackedInt in {
368  defm VPCMOV : xop4op_int<0xA2, "vpcmov", VR128, i128mem, v2i64,
369                           SchedWriteShuffle.XMM>;
370  defm VPCMOVY : xop4op_int<0xA2, "vpcmov", VR256, i256mem, v4i64,
371                            SchedWriteShuffle.YMM>, VEX_L;
372}
373
374let Predicates = [HasXOP] in {
375  def : Pat<(v16i8 (or (and VR128:$src3, VR128:$src1),
376                   (X86andnp VR128:$src3, VR128:$src2))),
377            (VPCMOVrrr VR128:$src1, VR128:$src2, VR128:$src3)>;
378  def : Pat<(v8i16 (or (and VR128:$src3, VR128:$src1),
379                   (X86andnp VR128:$src3, VR128:$src2))),
380            (VPCMOVrrr VR128:$src1, VR128:$src2, VR128:$src3)>;
381  def : Pat<(v4i32 (or (and VR128:$src3, VR128:$src1),
382                   (X86andnp VR128:$src3, VR128:$src2))),
383            (VPCMOVrrr VR128:$src1, VR128:$src2, VR128:$src3)>;
384
385  def : Pat<(or (and VR128:$src3, VR128:$src1),
386                (X86andnp VR128:$src3, (loadv16i8 addr:$src2))),
387            (VPCMOVrmr VR128:$src1, addr:$src2, VR128:$src3)>;
388  def : Pat<(or (and VR128:$src3, VR128:$src1),
389                (X86andnp VR128:$src3, (loadv8i16 addr:$src2))),
390            (VPCMOVrmr VR128:$src1, addr:$src2, VR128:$src3)>;
391  def : Pat<(or (and VR128:$src3, VR128:$src1),
392                (X86andnp VR128:$src3, (loadv4i32 addr:$src2))),
393            (VPCMOVrmr VR128:$src1, addr:$src2, VR128:$src3)>;
394
395  def : Pat<(v32i8 (or (and VR256:$src3, VR256:$src1),
396                   (X86andnp VR256:$src3, VR256:$src2))),
397            (VPCMOVYrrr VR256:$src1, VR256:$src2, VR256:$src3)>;
398  def : Pat<(v16i16 (or (and VR256:$src3, VR256:$src1),
399                    (X86andnp VR256:$src3, VR256:$src2))),
400            (VPCMOVYrrr VR256:$src1, VR256:$src2, VR256:$src3)>;
401  def : Pat<(v8i32 (or (and VR256:$src3, VR256:$src1),
402                   (X86andnp VR256:$src3, VR256:$src2))),
403            (VPCMOVYrrr VR256:$src1, VR256:$src2, VR256:$src3)>;
404
405  def : Pat<(or (and VR256:$src3, VR256:$src1),
406                (X86andnp VR256:$src3, (loadv32i8 addr:$src2))),
407            (VPCMOVYrmr VR256:$src1, addr:$src2, VR256:$src3)>;
408  def : Pat<(or (and VR256:$src3, VR256:$src1),
409                (X86andnp VR256:$src3, (loadv16i16 addr:$src2))),
410            (VPCMOVYrmr VR256:$src1, addr:$src2, VR256:$src3)>;
411  def : Pat<(or (and VR256:$src3, VR256:$src1),
412                (X86andnp VR256:$src3, (loadv8i32 addr:$src2))),
413            (VPCMOVYrmr VR256:$src1, addr:$src2, VR256:$src3)>;
414}
415
416multiclass xop_vpermil2<bits<8> Opc, string OpcodeStr, RegisterClass RC,
417                        X86MemOperand intmemop, X86MemOperand fpmemop,
418                        ValueType VT, PatFrag FPLdFrag, PatFrag IntLdFrag,
419                        X86FoldableSchedWrite sched> {
420  def rr : IXOP5<Opc, MRMSrcReg, (outs RC:$dst),
421        (ins RC:$src1, RC:$src2, RC:$src3, u4imm:$src4),
422        !strconcat(OpcodeStr,
423        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
424        [(set RC:$dst,
425           (VT (X86vpermil2 RC:$src1, RC:$src2, RC:$src3, (i8 timm:$src4))))]>,
426        Sched<[sched]>;
427  def rm : IXOP5<Opc, MRMSrcMemOp4, (outs RC:$dst),
428        (ins RC:$src1, RC:$src2, intmemop:$src3, u4imm:$src4),
429        !strconcat(OpcodeStr,
430        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
431        [(set RC:$dst,
432          (VT (X86vpermil2 RC:$src1, RC:$src2, (IntLdFrag addr:$src3),
433                           (i8 timm:$src4))))]>, VEX_W,
434        Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
435  def mr : IXOP5<Opc, MRMSrcMem, (outs RC:$dst),
436        (ins RC:$src1, fpmemop:$src2, RC:$src3, u4imm:$src4),
437        !strconcat(OpcodeStr,
438        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
439        [(set RC:$dst,
440          (VT (X86vpermil2 RC:$src1, (FPLdFrag addr:$src2),
441                           RC:$src3, (i8 timm:$src4))))]>,
442        Sched<[sched.Folded, sched.ReadAfterFold,
443               // fpmemop:$src2
444               ReadDefault, ReadDefault, ReadDefault, ReadDefault, ReadDefault,
445               // RC:$src3
446               sched.ReadAfterFold]>;
447  // For disassembler
448  let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
449  def rr_REV : IXOP5<Opc, MRMSrcRegOp4, (outs RC:$dst),
450        (ins RC:$src1, RC:$src2, RC:$src3, u4imm:$src4),
451        !strconcat(OpcodeStr,
452        "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"),
453        []>, VEX_W, Sched<[sched]>, FoldGenData<NAME#rr>;
454}
455
456let ExeDomain = SSEPackedDouble in {
457  defm VPERMIL2PD : xop_vpermil2<0x49, "vpermil2pd", VR128, i128mem, f128mem,
458                                 v2f64, loadv2f64, loadv2i64,
459                                 SchedWriteFVarShuffle.XMM>;
460  defm VPERMIL2PDY : xop_vpermil2<0x49, "vpermil2pd", VR256, i256mem, f256mem,
461                                  v4f64, loadv4f64, loadv4i64,
462                                  SchedWriteFVarShuffle.YMM>, VEX_L;
463}
464
465let ExeDomain = SSEPackedSingle in {
466  defm VPERMIL2PS : xop_vpermil2<0x48, "vpermil2ps", VR128, i128mem, f128mem,
467                                 v4f32, loadv4f32, loadv4i32,
468                                 SchedWriteFVarShuffle.XMM>;
469  defm VPERMIL2PSY : xop_vpermil2<0x48, "vpermil2ps", VR256, i256mem, f256mem,
470                                  v8f32, loadv8f32, loadv8i32,
471                                  SchedWriteFVarShuffle.YMM>, VEX_L;
472}
473
474