llvm.org GIT mirror llvm / stable lib / Target / X86 / X86InstrFMA.td
stable

Tree @stable (Download .tar.gz)

X86InstrFMA.td @stableraw · history · blame

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
//===-- X86InstrFMA.td - FMA Instruction Set ---------------*- tablegen -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file describes FMA (Fused Multiply-Add) instructions.
//
//===----------------------------------------------------------------------===//

//===----------------------------------------------------------------------===//
// FMA3 - Intel 3 operand Fused Multiply-Add instructions
//===----------------------------------------------------------------------===//

// For all FMA opcodes declared in fma3p_rm_* and fma3s_rm_* multiclasses
// defined below, both the register and memory variants are commutable.
// For the register form the commutable operands are 1, 2 and 3.
// For the memory variant the folded operand must be in 3. Thus,
// in that case, only the operands 1 and 2 can be swapped.
// Commuting some of operands may require the opcode change.
// FMA*213*:
//   operands 1 and 2 (memory & register forms): *213* --> *213*(no changes);
//   operands 1 and 3 (register forms only):     *213* --> *231*;
//   operands 2 and 3 (register forms only):     *213* --> *132*.
// FMA*132*:
//   operands 1 and 2 (memory & register forms): *132* --> *231*;
//   operands 1 and 3 (register forms only):     *132* --> *132*(no changes);
//   operands 2 and 3 (register forms only):     *132* --> *213*.
// FMA*231*:
//   operands 1 and 2 (memory & register forms): *231* --> *132*;
//   operands 1 and 3 (register forms only):     *231* --> *213*;
//   operands 2 and 3 (register forms only):     *231* --> *231*(no changes).

multiclass fma3p_rm_213<bits<8> opc, string OpcodeStr, RegisterClass RC,
                        ValueType VT, X86MemOperand x86memop, PatFrag MemFrag,
                        SDNode Op, X86FoldableSchedWrite sched> {
  def r     : FMA3<opc, MRMSrcReg, (outs RC:$dst),
                   (ins RC:$src1, RC:$src2, RC:$src3),
                   !strconcat(OpcodeStr,
                              "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                   [(set RC:$dst, (VT (Op RC:$src2, RC:$src1, RC:$src3)))]>,
                   Sched<[sched]>;

  let mayLoad = 1 in
  def m     : FMA3<opc, MRMSrcMem, (outs RC:$dst),
                   (ins RC:$src1, RC:$src2, x86memop:$src3),
                   !strconcat(OpcodeStr,
                              "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                   [(set RC:$dst, (VT (Op RC:$src2, RC:$src1,
                                          (MemFrag addr:$src3))))]>,
                   Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
}

multiclass fma3p_rm_231<bits<8> opc, string OpcodeStr, RegisterClass RC,
                        ValueType VT, X86MemOperand x86memop, PatFrag MemFrag,
                        SDNode Op, X86FoldableSchedWrite sched> {
  let hasSideEffects = 0 in
  def r     : FMA3<opc, MRMSrcReg, (outs RC:$dst),
                   (ins RC:$src1, RC:$src2, RC:$src3),
                   !strconcat(OpcodeStr,
                              "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                   []>, Sched<[sched]>;

  let mayLoad = 1 in
  def m     : FMA3<opc, MRMSrcMem, (outs RC:$dst),
                   (ins RC:$src1, RC:$src2, x86memop:$src3),
                   !strconcat(OpcodeStr,
                              "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                   [(set RC:$dst, (VT (Op RC:$src2, (MemFrag addr:$src3),
                                          RC:$src1)))]>,
                   Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
}

multiclass fma3p_rm_132<bits<8> opc, string OpcodeStr, RegisterClass RC,
                        ValueType VT, X86MemOperand x86memop, PatFrag MemFrag,
                        SDNode Op, X86FoldableSchedWrite sched> {
  let hasSideEffects = 0 in
  def r     : FMA3<opc, MRMSrcReg, (outs RC:$dst),
                   (ins RC:$src1, RC:$src2, RC:$src3),
                   !strconcat(OpcodeStr,
                              "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                   []>, Sched<[sched]>;

  // Pattern is 312 order so that the load is in a different place from the
  // 213 and 231 patterns this helps tablegen's duplicate pattern detection.
  let mayLoad = 1 in
  def m     : FMA3<opc, MRMSrcMem, (outs RC:$dst),
                   (ins RC:$src1, RC:$src2, x86memop:$src3),
                   !strconcat(OpcodeStr,
                              "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                   [(set RC:$dst, (VT (Op (MemFrag addr:$src3), RC:$src1,
                                          RC:$src2)))]>,
                   Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
}

let Constraints = "$src1 = $dst", hasSideEffects = 0, isCommutable = 1 in
multiclass fma3p_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                       string OpcodeStr, string PackTy, string Suff,
                       PatFrag MemFrag128, PatFrag MemFrag256,
                       SDNode Op, ValueType OpTy128, ValueType OpTy256,
                       X86SchedWriteWidths sched> {
  defm NAME#213#Suff : fma3p_rm_213<opc213, !strconcat(OpcodeStr, "213", PackTy),
                                    VR128, OpTy128, f128mem, MemFrag128, Op, sched.XMM>;
  defm NAME#231#Suff : fma3p_rm_231<opc231, !strconcat(OpcodeStr, "231", PackTy),
                                    VR128, OpTy128, f128mem, MemFrag128, Op, sched.XMM>;
  defm NAME#132#Suff : fma3p_rm_132<opc132, !strconcat(OpcodeStr, "132", PackTy),
                                    VR128, OpTy128, f128mem, MemFrag128, Op, sched.XMM>;

  defm NAME#213#Suff#Y : fma3p_rm_213<opc213, !strconcat(OpcodeStr, "213", PackTy),
                                      VR256, OpTy256, f256mem, MemFrag256, Op, sched.YMM>,
                                      VEX_L;
  defm NAME#231#Suff#Y : fma3p_rm_231<opc231, !strconcat(OpcodeStr, "231", PackTy),
                                      VR256, OpTy256, f256mem, MemFrag256, Op, sched.YMM>,
                                      VEX_L;
  defm NAME#132#Suff#Y : fma3p_rm_132<opc132, !strconcat(OpcodeStr, "132", PackTy),
                                      VR256, OpTy256, f256mem, MemFrag256, Op, sched.YMM>,
                                      VEX_L;
}

// Fused Multiply-Add
let ExeDomain = SSEPackedSingle in {
  defm VFMADD    : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "ps", "PS",
                               loadv4f32, loadv8f32, X86Fmadd, v4f32, v8f32,
                               SchedWriteFMA>;
  defm VFMSUB    : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "ps", "PS",
                               loadv4f32, loadv8f32, X86Fmsub, v4f32, v8f32,
                               SchedWriteFMA>;
  defm VFMADDSUB : fma3p_forms<0x96, 0xA6, 0xB6, "vfmaddsub", "ps", "PS",
                               loadv4f32, loadv8f32, X86Fmaddsub, v4f32, v8f32,
                               SchedWriteFMA>;
  defm VFMSUBADD : fma3p_forms<0x97, 0xA7, 0xB7, "vfmsubadd", "ps", "PS",
                               loadv4f32, loadv8f32, X86Fmsubadd, v4f32, v8f32,
                               SchedWriteFMA>;
}

let ExeDomain = SSEPackedDouble in {
  defm VFMADD    : fma3p_forms<0x98, 0xA8, 0xB8, "vfmadd", "pd", "PD",
                               loadv2f64, loadv4f64, X86Fmadd, v2f64,
                               v4f64, SchedWriteFMA>, VEX_W;
  defm VFMSUB    : fma3p_forms<0x9A, 0xAA, 0xBA, "vfmsub", "pd", "PD",
                               loadv2f64, loadv4f64, X86Fmsub, v2f64,
                               v4f64, SchedWriteFMA>, VEX_W;
  defm VFMADDSUB : fma3p_forms<0x96, 0xA6, 0xB6, "vfmaddsub", "pd", "PD",
                               loadv2f64, loadv4f64, X86Fmaddsub,
                               v2f64, v4f64, SchedWriteFMA>, VEX_W;
  defm VFMSUBADD : fma3p_forms<0x97, 0xA7, 0xB7, "vfmsubadd", "pd", "PD",
                               loadv2f64, loadv4f64, X86Fmsubadd,
                               v2f64, v4f64, SchedWriteFMA>, VEX_W;
}

// Fused Negative Multiply-Add
let ExeDomain = SSEPackedSingle in {
  defm VFNMADD : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "ps", "PS", loadv4f32,
                             loadv8f32, X86Fnmadd, v4f32, v8f32, SchedWriteFMA>;
  defm VFNMSUB : fma3p_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "ps", "PS", loadv4f32,
                             loadv8f32, X86Fnmsub, v4f32, v8f32, SchedWriteFMA>;
}
let ExeDomain = SSEPackedDouble in {
  defm VFNMADD : fma3p_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "pd", "PD", loadv2f64,
                             loadv4f64, X86Fnmadd, v2f64, v4f64, SchedWriteFMA>, VEX_W;
  defm VFNMSUB : fma3p_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "pd", "PD", loadv2f64,
                             loadv4f64, X86Fnmsub, v2f64, v4f64, SchedWriteFMA>, VEX_W;
}

// All source register operands of FMA opcodes defined in fma3s_rm multiclass
// can be commuted. In many cases such commute transformation requres an opcode
// adjustment, for example, commuting the operands 1 and 2 in FMA*132 form
// would require an opcode change to FMA*231:
//     FMA*132* reg1, reg2, reg3; // reg1 * reg3 + reg2;
//     -->
//     FMA*231* reg2, reg1, reg3; // reg1 * reg3 + reg2;
// Please see more detailed comment at the very beginning of the section
// defining FMA3 opcodes above.
multiclass fma3s_rm_213<bits<8> opc, string OpcodeStr,
                        X86MemOperand x86memop, RegisterClass RC,
                        SDPatternOperator OpNode,
                        X86FoldableSchedWrite sched> {
  def r : FMA3S<opc, MRMSrcReg, (outs RC:$dst),
                (ins RC:$src1, RC:$src2, RC:$src3),
                !strconcat(OpcodeStr,
                           "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                [(set RC:$dst, (OpNode RC:$src2, RC:$src1, RC:$src3))]>,
                Sched<[sched]>;

  let mayLoad = 1 in
  def m : FMA3S<opc, MRMSrcMem, (outs RC:$dst),
                (ins RC:$src1, RC:$src2, x86memop:$src3),
                !strconcat(OpcodeStr,
                           "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                [(set RC:$dst,
                  (OpNode RC:$src2, RC:$src1, (load addr:$src3)))]>,
                Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
}

multiclass fma3s_rm_231<bits<8> opc, string OpcodeStr,
                        X86MemOperand x86memop, RegisterClass RC,
                        SDPatternOperator OpNode, X86FoldableSchedWrite sched> {
  let hasSideEffects = 0 in
  def r : FMA3S<opc, MRMSrcReg, (outs RC:$dst),
                (ins RC:$src1, RC:$src2, RC:$src3),
                !strconcat(OpcodeStr,
                           "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                []>, Sched<[sched]>;

  let mayLoad = 1 in
  def m : FMA3S<opc, MRMSrcMem, (outs RC:$dst),
                (ins RC:$src1, RC:$src2, x86memop:$src3),
                !strconcat(OpcodeStr,
                           "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                [(set RC:$dst,
                  (OpNode RC:$src2, (load addr:$src3), RC:$src1))]>,
                Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
}

multiclass fma3s_rm_132<bits<8> opc, string OpcodeStr,
                        X86MemOperand x86memop, RegisterClass RC,
                        SDPatternOperator OpNode, X86FoldableSchedWrite sched> {
  let hasSideEffects = 0 in
  def r : FMA3S<opc, MRMSrcReg, (outs RC:$dst),
                (ins RC:$src1, RC:$src2, RC:$src3),
                !strconcat(OpcodeStr,
                           "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                []>, Sched<[sched]>;

  // Pattern is 312 order so that the load is in a different place from the
  // 213 and 231 patterns this helps tablegen's duplicate pattern detection.
  let mayLoad = 1 in
  def m : FMA3S<opc, MRMSrcMem, (outs RC:$dst),
                (ins RC:$src1, RC:$src2, x86memop:$src3),
                !strconcat(OpcodeStr,
                           "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                [(set RC:$dst,
                  (OpNode (load addr:$src3), RC:$src1, RC:$src2))]>,
                Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
}

let Constraints = "$src1 = $dst", isCommutable = 1, isCodeGenOnly = 1,
    hasSideEffects = 0 in
multiclass fma3s_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                       string OpStr, string PackTy, string Suff,
                       SDNode OpNode, RegisterClass RC,
                       X86MemOperand x86memop, X86FoldableSchedWrite sched> {
  defm NAME#213#Suff : fma3s_rm_213<opc213, !strconcat(OpStr, "213", PackTy),
                                    x86memop, RC, OpNode, sched>;
  defm NAME#231#Suff : fma3s_rm_231<opc231, !strconcat(OpStr, "231", PackTy),
                                    x86memop, RC, OpNode, sched>;
  defm NAME#132#Suff : fma3s_rm_132<opc132, !strconcat(OpStr, "132", PackTy),
                                    x86memop, RC, OpNode, sched>;
}

// These FMA*_Int instructions are defined specially for being used when
// the scalar FMA intrinsics are lowered to machine instructions, and in that
// sense, they are similar to existing ADD*_Int, SUB*_Int, MUL*_Int, etc.
// instructions.
//
// All of the FMA*_Int opcodes are defined as commutable here.
// Commuting the 2nd and 3rd source register operands of FMAs is quite trivial
// and the corresponding optimizations have been developed.
// Commuting the 1st operand of FMA*_Int requires some additional analysis,
// the commute optimization is legal only if all users of FMA*_Int use only
// the lowest element of the FMA*_Int instruction. Even though such analysis
// may be not implemented yet we allow the routines doing the actual commute
// transformation to decide if one or another instruction is commutable or not.
let Constraints = "$src1 = $dst", isCommutable = 1, hasSideEffects = 0 in
multiclass fma3s_rm_int<bits<8> opc, string OpcodeStr,
                        Operand memopr, RegisterClass RC,
                        X86FoldableSchedWrite sched> {
  def r_Int : FMA3S_Int<opc, MRMSrcReg, (outs RC:$dst),
                        (ins RC:$src1, RC:$src2, RC:$src3),
                        !strconcat(OpcodeStr,
                                   "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                        []>, Sched<[sched]>;

  let mayLoad = 1 in
  def m_Int : FMA3S_Int<opc, MRMSrcMem, (outs RC:$dst),
                        (ins RC:$src1, RC:$src2, memopr:$src3),
                        !strconcat(OpcodeStr,
                                   "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
                        []>, Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
}

// The FMA 213 form is created for lowering of scalar FMA intrinscis
// to machine instructions.
// The FMA 132 form can trivially be get by commuting the 2nd and 3rd operands
// of FMA 213 form.
// The FMA 231 form can be get only by commuting the 1st operand of 213 or 132
// forms and is possible only after special analysis of all uses of the initial
// instruction. Such analysis do not exist yet and thus introducing the 231
// form of FMA*_Int instructions is done using an optimistic assumption that
// such analysis will be implemented eventually.
multiclass fma3s_int_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                           string OpStr, string PackTy, string Suff,
                           RegisterClass RC, Operand memop,
                           X86FoldableSchedWrite sched> {
  defm NAME#132#Suff : fma3s_rm_int<opc132, !strconcat(OpStr, "132", PackTy),
                                    memop, RC, sched>;
  defm NAME#213#Suff : fma3s_rm_int<opc213, !strconcat(OpStr, "213", PackTy),
                                    memop, RC, sched>;
  defm NAME#231#Suff : fma3s_rm_int<opc231, !strconcat(OpStr, "231", PackTy),
                                    memop, RC, sched>;
}

multiclass fma3s<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                 string OpStr, SDNode OpNode, X86FoldableSchedWrite sched> {
  let ExeDomain = SSEPackedSingle in
  defm NAME : fma3s_forms<opc132, opc213, opc231, OpStr, "ss", "SS", OpNode,
                          FR32, f32mem, sched>,
              fma3s_int_forms<opc132, opc213, opc231, OpStr, "ss", "SS",
                              VR128, ssmem, sched>;

  let ExeDomain = SSEPackedDouble in
  defm NAME : fma3s_forms<opc132, opc213, opc231, OpStr, "sd", "SD", OpNode,
                        FR64, f64mem, sched>,
              fma3s_int_forms<opc132, opc213, opc231, OpStr, "sd", "SD",
                              VR128, sdmem, sched>, VEX_W;
}

defm VFMADD : fma3s<0x99, 0xA9, 0xB9, "vfmadd", X86Fmadd,
                    SchedWriteFMA.Scl>, VEX_LIG;
defm VFMSUB : fma3s<0x9B, 0xAB, 0xBB, "vfmsub", X86Fmsub,
                    SchedWriteFMA.Scl>, VEX_LIG;

defm VFNMADD : fma3s<0x9D, 0xAD, 0xBD, "vfnmadd", X86Fnmadd,
                     SchedWriteFMA.Scl>, VEX_LIG;
defm VFNMSUB : fma3s<0x9F, 0xAF, 0xBF, "vfnmsub", X86Fnmsub,
                     SchedWriteFMA.Scl>, VEX_LIG;

multiclass scalar_fma_patterns<SDNode Op, string Prefix, string Suffix,
                               SDNode Move, ValueType VT, ValueType EltVT,
                               RegisterClass RC, PatFrag mem_frag> {
  let Predicates = [HasFMA, NoAVX512] in {
    def : Pat<(VT (Move (VT VR128:$src1), (VT (scalar_to_vector
                (Op RC:$src2,
                    (EltVT (extractelt (VT VR128:$src1), (iPTR 0))),
                    RC:$src3))))),
              (!cast<Instruction>(Prefix#"213"#Suffix#"r_Int")
               VR128:$src1, (VT (COPY_TO_REGCLASS RC:$src2, VR128)),
               (VT (COPY_TO_REGCLASS RC:$src3, VR128)))>;

    def : Pat<(VT (Move (VT VR128:$src1), (VT (scalar_to_vector
                (Op RC:$src2, RC:$src3,
                    (EltVT (extractelt (VT VR128:$src1), (iPTR 0)))))))),
              (!cast<Instruction>(Prefix#"231"#Suffix#"r_Int")
               VR128:$src1, (VT (COPY_TO_REGCLASS RC:$src2, VR128)),
               (VT (COPY_TO_REGCLASS RC:$src3, VR128)))>;

    def : Pat<(VT (Move (VT VR128:$src1), (VT (scalar_to_vector
                (Op RC:$src2,
                    (EltVT (extractelt (VT VR128:$src1), (iPTR 0))),
                    (mem_frag addr:$src3)))))),
              (!cast<Instruction>(Prefix#"213"#Suffix#"m_Int")
               VR128:$src1, (VT (COPY_TO_REGCLASS RC:$src2, VR128)),
               addr:$src3)>;

    def : Pat<(VT (Move (VT VR128:$src1), (VT (scalar_to_vector
                (Op (EltVT (extractelt (VT VR128:$src1), (iPTR 0))),
                    (mem_frag addr:$src3), RC:$src2))))),
              (!cast<Instruction>(Prefix#"132"#Suffix#"m_Int")
               VR128:$src1, (VT (COPY_TO_REGCLASS RC:$src2, VR128)),
               addr:$src3)>;

    def : Pat<(VT (Move (VT VR128:$src1), (VT (scalar_to_vector
                (Op RC:$src2, (mem_frag addr:$src3),
                    (EltVT (extractelt (VT VR128:$src1), (iPTR 0)))))))),
              (!cast<Instruction>(Prefix#"231"#Suffix#"m_Int")
               VR128:$src1, (VT (COPY_TO_REGCLASS RC:$src2, VR128)),
               addr:$src3)>;
  }
}

defm : scalar_fma_patterns<X86Fmadd, "VFMADD", "SS", X86Movss, v4f32, f32, FR32, loadf32>;
defm : scalar_fma_patterns<X86Fmsub, "VFMSUB", "SS", X86Movss, v4f32, f32, FR32, loadf32>;
defm : scalar_fma_patterns<X86Fnmadd, "VFNMADD", "SS", X86Movss, v4f32, f32, FR32, loadf32>;
defm : scalar_fma_patterns<X86Fnmsub, "VFNMSUB", "SS", X86Movss, v4f32, f32, FR32, loadf32>;

defm : scalar_fma_patterns<X86Fmadd, "VFMADD", "SD", X86Movsd, v2f64, f64, FR64, loadf64>;
defm : scalar_fma_patterns<X86Fmsub, "VFMSUB", "SD", X86Movsd, v2f64, f64, FR64, loadf64>;
defm : scalar_fma_patterns<X86Fnmadd, "VFNMADD", "SD", X86Movsd, v2f64, f64, FR64, loadf64>;
defm : scalar_fma_patterns<X86Fnmsub, "VFNMSUB", "SD", X86Movsd, v2f64, f64, FR64, loadf64>;

//===----------------------------------------------------------------------===//
// FMA4 - AMD 4 operand Fused Multiply-Add instructions
//===----------------------------------------------------------------------===//

multiclass fma4s<bits<8> opc, string OpcodeStr, RegisterClass RC,
                 X86MemOperand x86memop, ValueType OpVT, SDNode OpNode,
                 PatFrag mem_frag, X86FoldableSchedWrite sched> {
  let isCommutable = 1 in
  def rr : FMA4S<opc, MRMSrcRegOp4, (outs RC:$dst),
           (ins RC:$src1, RC:$src2, RC:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set RC:$dst,
             (OpVT (OpNode RC:$src1, RC:$src2, RC:$src3)))]>, VEX_W, VEX_LIG,
           Sched<[sched]>;
  def rm : FMA4S<opc, MRMSrcMemOp4, (outs RC:$dst),
           (ins RC:$src1, RC:$src2, x86memop:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set RC:$dst, (OpNode RC:$src1, RC:$src2,
                           (mem_frag addr:$src3)))]>, VEX_W, VEX_LIG,
           Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
  def mr : FMA4S<opc, MRMSrcMem, (outs RC:$dst),
           (ins RC:$src1, x86memop:$src2, RC:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set RC:$dst,
             (OpNode RC:$src1, (mem_frag addr:$src2), RC:$src3))]>, VEX_LIG,
           Sched<[sched.Folded, sched.ReadAfterFold,
                  // x86memop:$src2
                  ReadDefault, ReadDefault, ReadDefault, ReadDefault,
                  ReadDefault,
                  // RC:$src3
                  sched.ReadAfterFold]>;
// For disassembler
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
  def rr_REV : FMA4S<opc, MRMSrcReg, (outs RC:$dst),
               (ins RC:$src1, RC:$src2, RC:$src3),
               !strconcat(OpcodeStr,
               "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>,
               VEX_LIG, FoldGenData<NAME#rr>, Sched<[sched]>;
}

multiclass fma4s_int<bits<8> opc, string OpcodeStr, Operand memop,
                     ValueType VT, X86FoldableSchedWrite sched> {
let isCodeGenOnly = 1, hasSideEffects = 0 in {
  def rr_Int : FMA4S_Int<opc, MRMSrcRegOp4, (outs VR128:$dst),
               (ins VR128:$src1, VR128:$src2, VR128:$src3),
               !strconcat(OpcodeStr,
               "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
               []>, VEX_W, VEX_LIG, Sched<[sched]>;
  let mayLoad = 1 in
  def rm_Int : FMA4S_Int<opc, MRMSrcMemOp4, (outs VR128:$dst),
               (ins VR128:$src1, VR128:$src2, memop:$src3),
               !strconcat(OpcodeStr,
               "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
               []>, VEX_W, VEX_LIG,
               Sched<[sched.Folded, sched.ReadAfterFold, sched.ReadAfterFold]>;
  let mayLoad = 1 in
  def mr_Int : FMA4S_Int<opc, MRMSrcMem, (outs VR128:$dst),
               (ins VR128:$src1, memop:$src2, VR128:$src3),
               !strconcat(OpcodeStr,
               "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
               []>,
               VEX_LIG, Sched<[sched.Folded, sched.ReadAfterFold,
                               // memop:$src2
                               ReadDefault, ReadDefault, ReadDefault,
                               ReadDefault, ReadDefault,
                               // VR128::$src3
                               sched.ReadAfterFold]>;
  def rr_Int_REV : FMA4S_Int<opc, MRMSrcReg, (outs VR128:$dst),
               (ins VR128:$src1, VR128:$src2, VR128:$src3),
               !strconcat(OpcodeStr,
               "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
               []>, VEX_LIG, FoldGenData<NAME#rr_Int>, Sched<[sched]>;
} // isCodeGenOnly = 1
}

multiclass fma4p<bits<8> opc, string OpcodeStr, SDNode OpNode,
                 ValueType OpVT128, ValueType OpVT256,
                 PatFrag ld_frag128, PatFrag ld_frag256,
                 X86SchedWriteWidths sched> {
  let isCommutable = 1 in
  def rr : FMA4<opc, MRMSrcRegOp4, (outs VR128:$dst),
           (ins VR128:$src1, VR128:$src2, VR128:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst,
             (OpVT128 (OpNode VR128:$src1, VR128:$src2, VR128:$src3)))]>,
           VEX_W, Sched<[sched.XMM]>;
  def rm : FMA4<opc, MRMSrcMemOp4, (outs VR128:$dst),
           (ins VR128:$src1, VR128:$src2, f128mem:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst, (OpNode VR128:$src1, VR128:$src2,
                              (ld_frag128 addr:$src3)))]>, VEX_W,
           Sched<[sched.XMM.Folded, sched.XMM.ReadAfterFold, sched.XMM.ReadAfterFold]>;
  def mr : FMA4<opc, MRMSrcMem, (outs VR128:$dst),
           (ins VR128:$src1, f128mem:$src2, VR128:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR128:$dst,
             (OpNode VR128:$src1, (ld_frag128 addr:$src2), VR128:$src3))]>,
           Sched<[sched.XMM.Folded, sched.XMM.ReadAfterFold,
                  // f128mem:$src2
                  ReadDefault, ReadDefault, ReadDefault, ReadDefault,
                  ReadDefault,
                  // VR128::$src3
                  sched.XMM.ReadAfterFold]>;
  let isCommutable = 1 in
  def Yrr : FMA4<opc, MRMSrcRegOp4, (outs VR256:$dst),
           (ins VR256:$src1, VR256:$src2, VR256:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR256:$dst,
             (OpVT256 (OpNode VR256:$src1, VR256:$src2, VR256:$src3)))]>,
           VEX_W, VEX_L, Sched<[sched.YMM]>;
  def Yrm : FMA4<opc, MRMSrcMemOp4, (outs VR256:$dst),
           (ins VR256:$src1, VR256:$src2, f256mem:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR256:$dst, (OpNode VR256:$src1, VR256:$src2,
                              (ld_frag256 addr:$src3)))]>, VEX_W, VEX_L,
           Sched<[sched.YMM.Folded, sched.YMM.ReadAfterFold, sched.YMM.ReadAfterFold]>;
  def Ymr : FMA4<opc, MRMSrcMem, (outs VR256:$dst),
           (ins VR256:$src1, f256mem:$src2, VR256:$src3),
           !strconcat(OpcodeStr,
           "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
           [(set VR256:$dst, (OpNode VR256:$src1,
                              (ld_frag256 addr:$src2), VR256:$src3))]>, VEX_L,
           Sched<[sched.YMM.Folded, sched.YMM.ReadAfterFold,
                  // f256mem:$src2
                  ReadDefault, ReadDefault, ReadDefault, ReadDefault,
                  ReadDefault,
                  // VR256::$src3
                  sched.YMM.ReadAfterFold]>;
// For disassembler
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in {
  def rr_REV : FMA4<opc, MRMSrcReg, (outs VR128:$dst),
               (ins VR128:$src1, VR128:$src2, VR128:$src3),
               !strconcat(OpcodeStr,
               "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>,
               Sched<[sched.XMM]>, FoldGenData<NAME#rr>;
  def Yrr_REV : FMA4<opc, MRMSrcReg, (outs VR256:$dst),
                (ins VR256:$src1, VR256:$src2, VR256:$src3),
                !strconcat(OpcodeStr,
                "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>,
                VEX_L, Sched<[sched.YMM]>, FoldGenData<NAME#Yrr>;
} // isCodeGenOnly = 1
}

let ExeDomain = SSEPackedSingle in {
  // Scalar Instructions
  defm VFMADDSS4  : fma4s<0x6A, "vfmaddss", FR32, f32mem, f32, X86Fmadd, loadf32,
                          SchedWriteFMA.Scl>,
                    fma4s_int<0x6A, "vfmaddss", ssmem, v4f32,
                              SchedWriteFMA.Scl>;
  defm VFMSUBSS4  : fma4s<0x6E, "vfmsubss", FR32, f32mem, f32, X86Fmsub, loadf32,
                          SchedWriteFMA.Scl>,
                    fma4s_int<0x6E, "vfmsubss", ssmem, v4f32,
                              SchedWriteFMA.Scl>;
  defm VFNMADDSS4 : fma4s<0x7A, "vfnmaddss", FR32, f32mem, f32,
                          X86Fnmadd, loadf32, SchedWriteFMA.Scl>,
                    fma4s_int<0x7A, "vfnmaddss", ssmem, v4f32,
                              SchedWriteFMA.Scl>;
  defm VFNMSUBSS4 : fma4s<0x7E, "vfnmsubss", FR32, f32mem, f32,
                          X86Fnmsub, loadf32, SchedWriteFMA.Scl>,
                    fma4s_int<0x7E, "vfnmsubss", ssmem, v4f32,
                              SchedWriteFMA.Scl>;
  // Packed Instructions
  defm VFMADDPS4    : fma4p<0x68, "vfmaddps", X86Fmadd, v4f32, v8f32,
                            loadv4f32, loadv8f32, SchedWriteFMA>;
  defm VFMSUBPS4    : fma4p<0x6C, "vfmsubps", X86Fmsub, v4f32, v8f32,
                            loadv4f32, loadv8f32, SchedWriteFMA>;
  defm VFNMADDPS4   : fma4p<0x78, "vfnmaddps", X86Fnmadd, v4f32, v8f32,
                            loadv4f32, loadv8f32, SchedWriteFMA>;
  defm VFNMSUBPS4   : fma4p<0x7C, "vfnmsubps", X86Fnmsub, v4f32, v8f32,
                            loadv4f32, loadv8f32, SchedWriteFMA>;
  defm VFMADDSUBPS4 : fma4p<0x5C, "vfmaddsubps", X86Fmaddsub, v4f32, v8f32,
                            loadv4f32, loadv8f32, SchedWriteFMA>;
  defm VFMSUBADDPS4 : fma4p<0x5E, "vfmsubaddps", X86Fmsubadd, v4f32, v8f32,
                            loadv4f32, loadv8f32, SchedWriteFMA>;
}

let ExeDomain = SSEPackedDouble in {
  // Scalar Instructions
  defm VFMADDSD4  : fma4s<0x6B, "vfmaddsd", FR64, f64mem, f64, X86Fmadd, loadf64,
                          SchedWriteFMA.Scl>,
                    fma4s_int<0x6B, "vfmaddsd", sdmem, v2f64,
                              SchedWriteFMA.Scl>;
  defm VFMSUBSD4  : fma4s<0x6F, "vfmsubsd", FR64, f64mem, f64, X86Fmsub, loadf64,
                          SchedWriteFMA.Scl>,
                    fma4s_int<0x6F, "vfmsubsd", sdmem, v2f64,
                              SchedWriteFMA.Scl>;
  defm VFNMADDSD4 : fma4s<0x7B, "vfnmaddsd", FR64, f64mem, f64,
                          X86Fnmadd, loadf64, SchedWriteFMA.Scl>,
                    fma4s_int<0x7B, "vfnmaddsd", sdmem, v2f64,
                              SchedWriteFMA.Scl>;
  defm VFNMSUBSD4 : fma4s<0x7F, "vfnmsubsd", FR64, f64mem, f64,
                          X86Fnmsub, loadf64, SchedWriteFMA.Scl>,
                    fma4s_int<0x7F, "vfnmsubsd", sdmem, v2f64,
                              SchedWriteFMA.Scl>;
  // Packed Instructions
  defm VFMADDPD4    : fma4p<0x69, "vfmaddpd", X86Fmadd, v2f64, v4f64,
                            loadv2f64, loadv4f64, SchedWriteFMA>;
  defm VFMSUBPD4    : fma4p<0x6D, "vfmsubpd", X86Fmsub, v2f64, v4f64,
                            loadv2f64, loadv4f64, SchedWriteFMA>;
  defm VFNMADDPD4   : fma4p<0x79, "vfnmaddpd", X86Fnmadd, v2f64, v4f64,
                            loadv2f64, loadv4f64, SchedWriteFMA>;
  defm VFNMSUBPD4   : fma4p<0x7D, "vfnmsubpd", X86Fnmsub, v2f64, v4f64,
                            loadv2f64, loadv4f64, SchedWriteFMA>;
  defm VFMADDSUBPD4 : fma4p<0x5D, "vfmaddsubpd", X86Fmaddsub, v2f64, v4f64,
                            loadv2f64, loadv4f64, SchedWriteFMA>;
  defm VFMSUBADDPD4 : fma4p<0x5F, "vfmsubaddpd", X86Fmsubadd, v2f64, v4f64,
                            loadv2f64, loadv4f64, SchedWriteFMA>;
}

multiclass scalar_fma4_patterns<SDNode Op, string Name,
                               ValueType VT, ValueType EltVT,
                               RegisterClass RC, PatFrag mem_frag> {
  let Predicates = [HasFMA4] in {
    def : Pat<(VT (X86vzmovl (VT (scalar_to_vector
                                  (Op RC:$src1, RC:$src2, RC:$src3))))),
              (!cast<Instruction>(Name#"rr_Int")
               (VT (COPY_TO_REGCLASS RC:$src1, VR128)),
               (VT (COPY_TO_REGCLASS RC:$src2, VR128)),
               (VT (COPY_TO_REGCLASS RC:$src3, VR128)))>;

    def : Pat<(VT (X86vzmovl (VT (scalar_to_vector
                                  (Op RC:$src1, RC:$src2,
                                      (mem_frag addr:$src3)))))),
              (!cast<Instruction>(Name#"rm_Int")
               (VT (COPY_TO_REGCLASS RC:$src1, VR128)),
               (VT (COPY_TO_REGCLASS RC:$src2, VR128)), addr:$src3)>;

    def : Pat<(VT (X86vzmovl (VT (scalar_to_vector
                                  (Op RC:$src1, (mem_frag addr:$src2),
                                      RC:$src3))))),
              (!cast<Instruction>(Name#"mr_Int")
               (VT (COPY_TO_REGCLASS RC:$src1, VR128)), addr:$src2,
               (VT (COPY_TO_REGCLASS RC:$src3, VR128)))>;
  }
}

defm : scalar_fma4_patterns<X86Fmadd, "VFMADDSS4", v4f32, f32, FR32, loadf32>;
defm : scalar_fma4_patterns<X86Fmsub, "VFMSUBSS4", v4f32, f32, FR32, loadf32>;
defm : scalar_fma4_patterns<X86Fnmadd, "VFNMADDSS4", v4f32, f32, FR32, loadf32>;
defm : scalar_fma4_patterns<X86Fnmsub, "VFNMSUBSS4", v4f32, f32, FR32, loadf32>;

defm : scalar_fma4_patterns<X86Fmadd, "VFMADDSD4", v2f64, f64, FR64, loadf64>;
defm : scalar_fma4_patterns<X86Fmsub, "VFMSUBSD4", v2f64, f64, FR64, loadf64>;
defm : scalar_fma4_patterns<X86Fnmadd, "VFNMADDSD4", v2f64, f64, FR64, loadf64>;
defm : scalar_fma4_patterns<X86Fnmsub, "VFNMSUBSD4", v2f64, f64, FR64, loadf64>;