forked from llvm/llvm-project
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathWebAssemblyInstrSIMD.td
1419 lines (1237 loc) · 57.5 KB
/
WebAssemblyInstrSIMD.td
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
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// WebAssemblyInstrSIMD.td - WebAssembly SIMD codegen support -*- 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
//
//===----------------------------------------------------------------------===//
///
/// \file
/// WebAssembly SIMD operand code-gen constructs.
///
//===----------------------------------------------------------------------===//
// Instructions using the SIMD opcode prefix and requiring one of the SIMD
// feature predicates.
multiclass ABSTRACT_SIMD_I<dag oops_r, dag iops_r, dag oops_s, dag iops_s,
list<dag> pattern_r, string asmstr_r,
string asmstr_s, bits<32> simdop,
Predicate simd_level> {
defm "" : I<oops_r, iops_r, oops_s, iops_s, pattern_r, asmstr_r, asmstr_s,
!if(!ge(simdop, 0x100),
!or(0xfd0000, !and(0xffff, simdop)),
!or(0xfd00, !and(0xff, simdop)))>,
Requires<[simd_level]>;
}
multiclass SIMD_I<dag oops_r, dag iops_r, dag oops_s, dag iops_s,
list<dag> pattern_r, string asmstr_r = "",
string asmstr_s = "", bits<32> simdop = -1> {
defm "" : ABSTRACT_SIMD_I<oops_r, iops_r, oops_s, iops_s, pattern_r, asmstr_r,
asmstr_s, simdop, HasSIMD128>;
}
multiclass RELAXED_I<dag oops_r, dag iops_r, dag oops_s, dag iops_s,
list<dag> pattern_r, string asmstr_r = "",
string asmstr_s = "", bits<32> simdop = -1> {
defm "" : ABSTRACT_SIMD_I<oops_r, iops_r, oops_s, iops_s, pattern_r, asmstr_r,
asmstr_s, simdop, HasRelaxedSIMD>;
}
defm "" : ARGUMENT<V128, v16i8>;
defm "" : ARGUMENT<V128, v8i16>;
defm "" : ARGUMENT<V128, v4i32>;
defm "" : ARGUMENT<V128, v2i64>;
defm "" : ARGUMENT<V128, v4f32>;
defm "" : ARGUMENT<V128, v2f64>;
// Constrained immediate argument types
foreach SIZE = [8, 16] in
def ImmI#SIZE : ImmLeaf<i32,
"return -(1 << ("#SIZE#" - 1)) <= Imm && Imm < (1 << ("#SIZE#" - 1));"
>;
foreach SIZE = [2, 4, 8, 16, 32] in
def LaneIdx#SIZE : ImmLeaf<i32, "return 0 <= Imm && Imm < "#SIZE#";">;
// Create vector with identical lanes: splat
def splat2 : PatFrag<(ops node:$x), (build_vector $x, $x)>;
def splat4 : PatFrag<(ops node:$x), (build_vector $x, $x, $x, $x)>;
def splat8 : PatFrag<(ops node:$x), (build_vector $x, $x, $x, $x,
$x, $x, $x, $x)>;
def splat16 : PatFrag<(ops node:$x),
(build_vector $x, $x, $x, $x, $x, $x, $x, $x,
$x, $x, $x, $x, $x, $x, $x, $x)>;
class Vec {
ValueType vt;
ValueType int_vt;
ValueType lane_vt;
WebAssemblyRegClass lane_rc;
int lane_bits;
ImmLeaf lane_idx;
PatFrag splat;
string prefix;
Vec split;
}
def I8x16 : Vec {
let vt = v16i8;
let int_vt = vt;
let lane_vt = i32;
let lane_rc = I32;
let lane_bits = 8;
let lane_idx = LaneIdx16;
let splat = splat16;
let prefix = "i8x16";
}
def I16x8 : Vec {
let vt = v8i16;
let int_vt = vt;
let lane_vt = i32;
let lane_rc = I32;
let lane_bits = 16;
let lane_idx = LaneIdx8;
let splat = splat8;
let prefix = "i16x8";
let split = I8x16;
}
def I32x4 : Vec {
let vt = v4i32;
let int_vt = vt;
let lane_vt = i32;
let lane_rc = I32;
let lane_bits = 32;
let lane_idx = LaneIdx4;
let splat = splat4;
let prefix = "i32x4";
let split = I16x8;
}
def I64x2 : Vec {
let vt = v2i64;
let int_vt = vt;
let lane_vt = i64;
let lane_rc = I64;
let lane_bits = 64;
let lane_idx = LaneIdx2;
let splat = splat2;
let prefix = "i64x2";
let split = I32x4;
}
def F32x4 : Vec {
let vt = v4f32;
let int_vt = v4i32;
let lane_vt = f32;
let lane_rc = F32;
let lane_bits = 32;
let lane_idx = LaneIdx4;
let splat = splat4;
let prefix = "f32x4";
}
def F64x2 : Vec {
let vt = v2f64;
let int_vt = v2i64;
let lane_vt = f64;
let lane_rc = F64;
let lane_bits = 64;
let lane_idx = LaneIdx2;
let splat = splat2;
let prefix = "f64x2";
}
defvar AllVecs = [I8x16, I16x8, I32x4, I64x2, F32x4, F64x2];
defvar IntVecs = [I8x16, I16x8, I32x4, I64x2];
//===----------------------------------------------------------------------===//
// Load and store
//===----------------------------------------------------------------------===//
// Load: v128.load
let mayLoad = 1, UseNamedOperandTable = 1 in {
defm LOAD_V128_A32 :
SIMD_I<(outs V128:$dst), (ins P2Align:$p2align, offset32_op:$off, I32:$addr),
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
"v128.load\t$dst, ${off}(${addr})$p2align",
"v128.load\t$off$p2align", 0>;
defm LOAD_V128_A64 :
SIMD_I<(outs V128:$dst), (ins P2Align:$p2align, offset64_op:$off, I64:$addr),
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
"v128.load\t$dst, ${off}(${addr})$p2align",
"v128.load\t$off$p2align", 0>;
}
// Def load patterns from WebAssemblyInstrMemory.td for vector types
foreach vec = AllVecs in {
defm : LoadPatNoOffset<vec.vt, load, "LOAD_V128">;
defm : LoadPatImmOff<vec.vt, load, regPlusImm, "LOAD_V128">;
defm : LoadPatImmOff<vec.vt, load, or_is_add, "LOAD_V128">;
defm : LoadPatOffsetOnly<vec.vt, load, "LOAD_V128">;
defm : LoadPatGlobalAddrOffOnly<vec.vt, load, "LOAD_V128">;
}
// v128.loadX_splat
multiclass SIMDLoadSplat<int size, bits<32> simdop> {
let mayLoad = 1, UseNamedOperandTable = 1 in {
defm LOAD#size#_SPLAT_A32 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset32_op:$off, I32:$addr),
(outs),
(ins P2Align:$p2align, offset32_op:$off), [],
"v128.load"#size#"_splat\t$dst, ${off}(${addr})$p2align",
"v128.load"#size#"_splat\t$off$p2align", simdop>;
defm LOAD#size#_SPLAT_A64 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset64_op:$off, I64:$addr),
(outs),
(ins P2Align:$p2align, offset64_op:$off), [],
"v128.load"#size#"_splat\t$dst, ${off}(${addr})$p2align",
"v128.load"#size#"_splat\t$off$p2align", simdop>;
}
}
defm "" : SIMDLoadSplat<8, 7>;
defm "" : SIMDLoadSplat<16, 8>;
defm "" : SIMDLoadSplat<32, 9>;
defm "" : SIMDLoadSplat<64, 10>;
def wasm_load_splat_t : SDTypeProfile<1, 1, [SDTCisPtrTy<1>]>;
def wasm_load_splat : SDNode<"WebAssemblyISD::LOAD_SPLAT", wasm_load_splat_t,
[SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
def load_splat : PatFrag<(ops node:$addr), (wasm_load_splat node:$addr)>;
foreach vec = AllVecs in {
defvar inst = "LOAD"#vec.lane_bits#"_SPLAT";
defm : LoadPatNoOffset<vec.vt, load_splat, inst>;
defm : LoadPatImmOff<vec.vt, load_splat, regPlusImm, inst>;
defm : LoadPatImmOff<vec.vt, load_splat, or_is_add, inst>;
defm : LoadPatOffsetOnly<vec.vt, load_splat, inst>;
defm : LoadPatGlobalAddrOffOnly<vec.vt, load_splat, inst>;
}
// Load and extend
multiclass SIMDLoadExtend<Vec vec, string loadPat, bits<32> simdop> {
defvar signed = vec.prefix#".load"#loadPat#"_s";
defvar unsigned = vec.prefix#".load"#loadPat#"_u";
let mayLoad = 1, UseNamedOperandTable = 1 in {
defm LOAD_EXTEND_S_#vec#_A32 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset32_op:$off, I32:$addr),
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
signed#"\t$dst, ${off}(${addr})$p2align",
signed#"\t$off$p2align", simdop>;
defm LOAD_EXTEND_U_#vec#_A32 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset32_op:$off, I32:$addr),
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
unsigned#"\t$dst, ${off}(${addr})$p2align",
unsigned#"\t$off$p2align", !add(simdop, 1)>;
defm LOAD_EXTEND_S_#vec#_A64 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset64_op:$off, I64:$addr),
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
signed#"\t$dst, ${off}(${addr})$p2align",
signed#"\t$off$p2align", simdop>;
defm LOAD_EXTEND_U_#vec#_A64 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset64_op:$off, I64:$addr),
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
unsigned#"\t$dst, ${off}(${addr})$p2align",
unsigned#"\t$off$p2align", !add(simdop, 1)>;
}
}
defm "" : SIMDLoadExtend<I16x8, "8x8", 1>;
defm "" : SIMDLoadExtend<I32x4, "16x4", 3>;
defm "" : SIMDLoadExtend<I64x2, "32x2", 5>;
foreach vec = [I16x8, I32x4, I64x2] in
foreach exts = [["sextloadvi", "_S"],
["zextloadvi", "_U"],
["extloadvi", "_U"]] in {
defvar loadpat = !cast<PatFrag>(exts[0]#vec.split.lane_bits);
defvar inst = "LOAD_EXTEND"#exts[1]#"_"#vec;
defm : LoadPatNoOffset<vec.vt, loadpat, inst>;
defm : LoadPatImmOff<vec.vt, loadpat, regPlusImm, inst>;
defm : LoadPatImmOff<vec.vt, loadpat, or_is_add, inst>;
defm : LoadPatOffsetOnly<vec.vt, loadpat, inst>;
defm : LoadPatGlobalAddrOffOnly<vec.vt, loadpat, inst>;
}
// Load lane into zero vector
multiclass SIMDLoadZero<Vec vec, bits<32> simdop> {
defvar name = "v128.load"#vec.lane_bits#"_zero";
let mayLoad = 1, UseNamedOperandTable = 1 in {
defm LOAD_ZERO_#vec#_A32 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset32_op:$off, I32:$addr),
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
name#"\t$dst, ${off}(${addr})$p2align",
name#"\t$off$p2align", simdop>;
defm LOAD_ZERO_#vec#_A64 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset64_op:$off, I64:$addr),
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
name#"\t$dst, ${off}(${addr})$p2align",
name#"\t$off$p2align", simdop>;
} // mayLoad = 1, UseNamedOperandTable = 1
}
defm "" : SIMDLoadZero<I32x4, 0x5c>;
defm "" : SIMDLoadZero<I64x2, 0x5d>;
// Use load_zero to load scalars into vectors as well where possible.
// TODO: i32, i16, and i8 scalars
def load_scalar :
PatFrag<(ops node:$addr), (scalar_to_vector (i64 (load $addr)))>;
defm : LoadPatNoOffset<v2i64, load_scalar, "LOAD_ZERO_I64x2">;
defm : LoadPatImmOff<v2i64, load_scalar, regPlusImm, "LOAD_ZERO_I64x2">;
defm : LoadPatImmOff<v2i64, load_scalar, or_is_add, "LOAD_ZERO_I64x2">;
defm : LoadPatOffsetOnly<v2i64, load_scalar, "LOAD_ZERO_I64x2">;
defm : LoadPatGlobalAddrOffOnly<v2i64, load_scalar, "LOAD_ZERO_I64x2">;
// TODO: f32x4 and f64x2 as well
foreach vec = [I32x4, I64x2] in {
defvar inst = "LOAD_ZERO_"#vec;
defvar pat = PatFrag<(ops node:$ptr),
(vector_insert (vec.splat (vec.lane_vt 0)), (vec.lane_vt (load $ptr)), 0)>;
defm : LoadPatNoOffset<vec.vt, pat, inst>;
defm : LoadPatImmOff<vec.vt, pat, regPlusImm, inst>;
defm : LoadPatImmOff<vec.vt, pat, or_is_add, inst>;
defm : LoadPatOffsetOnly<vec.vt, pat, inst>;
defm : LoadPatGlobalAddrOffOnly<vec.vt, pat, inst>;
}
// Load lane
multiclass SIMDLoadLane<Vec vec, bits<32> simdop> {
defvar name = "v128.load"#vec.lane_bits#"_lane";
let mayLoad = 1, UseNamedOperandTable = 1 in {
defm LOAD_LANE_#vec#_A32 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset32_op:$off, vec_i8imm_op:$idx,
I32:$addr, V128:$vec),
(outs), (ins P2Align:$p2align, offset32_op:$off, vec_i8imm_op:$idx),
[], name#"\t$dst, ${off}(${addr})$p2align, $vec, $idx",
name#"\t$off$p2align, $idx", simdop>;
defm LOAD_LANE_#vec#_A64 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset64_op:$off, vec_i8imm_op:$idx,
I64:$addr, V128:$vec),
(outs), (ins P2Align:$p2align, offset64_op:$off, vec_i8imm_op:$idx),
[], name#"\t$dst, ${off}(${addr})$p2align, $vec, $idx",
name#"\t$off$p2align, $idx", simdop>;
} // mayLoad = 1, UseNamedOperandTable = 1
}
defm "" : SIMDLoadLane<I8x16, 0x54>;
defm "" : SIMDLoadLane<I16x8, 0x55>;
defm "" : SIMDLoadLane<I32x4, 0x56>;
defm "" : SIMDLoadLane<I64x2, 0x57>;
// Select loads with no constant offset.
multiclass LoadLanePatNoOffset<Vec vec, SDPatternOperator kind> {
defvar load_lane_a32 = !cast<NI>("LOAD_LANE_"#vec#"_A32");
defvar load_lane_a64 = !cast<NI>("LOAD_LANE_"#vec#"_A64");
def : Pat<(vec.vt (kind (i32 I32:$addr),
(vec.vt V128:$vec), (i32 vec.lane_idx:$idx))),
(load_lane_a32 0, 0, imm:$idx, $addr, $vec)>,
Requires<[HasAddr32]>;
def : Pat<(vec.vt (kind (i64 I64:$addr),
(vec.vt V128:$vec), (i32 vec.lane_idx:$idx))),
(load_lane_a64 0, 0, imm:$idx, $addr, $vec)>,
Requires<[HasAddr64]>;
}
def load8_lane :
PatFrag<(ops node:$ptr, node:$vec, node:$idx),
(vector_insert $vec, (i32 (extloadi8 $ptr)), $idx)>;
def load16_lane :
PatFrag<(ops node:$ptr, node:$vec, node:$idx),
(vector_insert $vec, (i32 (extloadi16 $ptr)), $idx)>;
def load32_lane :
PatFrag<(ops node:$ptr, node:$vec, node:$idx),
(vector_insert $vec, (i32 (load $ptr)), $idx)>;
def load64_lane :
PatFrag<(ops node:$ptr, node:$vec, node:$idx),
(vector_insert $vec, (i64 (load $ptr)), $idx)>;
// TODO: floating point lanes as well
defm : LoadLanePatNoOffset<I8x16, load8_lane>;
defm : LoadLanePatNoOffset<I16x8, load16_lane>;
defm : LoadLanePatNoOffset<I32x4, load32_lane>;
defm : LoadLanePatNoOffset<I64x2, load64_lane>;
// TODO: Also support the other load patterns for load_lane once the instructions
// are merged to the proposal.
// Store: v128.store
let mayStore = 1, UseNamedOperandTable = 1 in {
defm STORE_V128_A32 :
SIMD_I<(outs), (ins P2Align:$p2align, offset32_op:$off, I32:$addr, V128:$vec),
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
"v128.store\t${off}(${addr})$p2align, $vec",
"v128.store\t$off$p2align", 11>;
defm STORE_V128_A64 :
SIMD_I<(outs), (ins P2Align:$p2align, offset64_op:$off, I64:$addr, V128:$vec),
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
"v128.store\t${off}(${addr})$p2align, $vec",
"v128.store\t$off$p2align", 11>;
}
// Def store patterns from WebAssemblyInstrMemory.td for vector types
foreach vec = AllVecs in {
defm : StorePatNoOffset<vec.vt, store, "STORE_V128">;
defm : StorePatImmOff<vec.vt, store, regPlusImm, "STORE_V128">;
defm : StorePatImmOff<vec.vt, store, or_is_add, "STORE_V128">;
defm : StorePatOffsetOnly<vec.vt, store, "STORE_V128">;
defm : StorePatGlobalAddrOffOnly<vec.vt, store, "STORE_V128">;
}
// Store lane
multiclass SIMDStoreLane<Vec vec, bits<32> simdop> {
defvar name = "v128.store"#vec.lane_bits#"_lane";
let mayStore = 1, UseNamedOperandTable = 1 in {
defm STORE_LANE_#vec#_A32 :
SIMD_I<(outs),
(ins P2Align:$p2align, offset32_op:$off, vec_i8imm_op:$idx,
I32:$addr, V128:$vec),
(outs), (ins P2Align:$p2align, offset32_op:$off, vec_i8imm_op:$idx),
[], name#"\t${off}(${addr})$p2align, $vec, $idx",
name#"\t$off$p2align, $idx", simdop>;
defm STORE_LANE_#vec#_A64 :
SIMD_I<(outs V128:$dst),
(ins P2Align:$p2align, offset64_op:$off, vec_i8imm_op:$idx,
I64:$addr, V128:$vec),
(outs), (ins P2Align:$p2align, offset64_op:$off, vec_i8imm_op:$idx),
[], name#"\t${off}(${addr})$p2align, $vec, $idx",
name#"\t$off$p2align, $idx", simdop>;
} // mayStore = 1, UseNamedOperandTable = 1
}
defm "" : SIMDStoreLane<I8x16, 0x58>;
defm "" : SIMDStoreLane<I16x8, 0x59>;
defm "" : SIMDStoreLane<I32x4, 0x5a>;
defm "" : SIMDStoreLane<I64x2, 0x5b>;
// Select stores with no constant offset.
multiclass StoreLanePatNoOffset<Vec vec, SDPatternOperator kind> {
def : Pat<(kind (i32 I32:$addr), (vec.vt V128:$vec), (i32 vec.lane_idx:$idx)),
(!cast<NI>("STORE_LANE_"#vec#"_A32") 0, 0, imm:$idx, $addr, $vec)>,
Requires<[HasAddr32]>;
def : Pat<(kind (i64 I64:$addr), (vec.vt V128:$vec), (i32 vec.lane_idx:$idx)),
(!cast<NI>("STORE_LANE_"#vec#"_A64") 0, 0, imm:$idx, $addr, $vec)>,
Requires<[HasAddr64]>;
}
def store8_lane :
PatFrag<(ops node:$ptr, node:$vec, node:$idx),
(truncstorei8 (i32 (vector_extract $vec, $idx)), $ptr)>;
def store16_lane :
PatFrag<(ops node:$ptr, node:$vec, node:$idx),
(truncstorei16 (i32 (vector_extract $vec, $idx)), $ptr)>;
def store32_lane :
PatFrag<(ops node:$ptr, node:$vec, node:$idx),
(store (i32 (vector_extract $vec, $idx)), $ptr)>;
def store64_lane :
PatFrag<(ops node:$ptr, node:$vec, node:$idx),
(store (i64 (vector_extract $vec, $idx)), $ptr)>;
// TODO: floating point lanes as well
let AddedComplexity = 1 in {
defm : StoreLanePatNoOffset<I8x16, store8_lane>;
defm : StoreLanePatNoOffset<I16x8, store16_lane>;
defm : StoreLanePatNoOffset<I32x4, store32_lane>;
defm : StoreLanePatNoOffset<I64x2, store64_lane>;
}
//===----------------------------------------------------------------------===//
// Constructing SIMD values
//===----------------------------------------------------------------------===//
// Constant: v128.const
multiclass ConstVec<Vec vec, dag ops, dag pat, string args> {
let isMoveImm = 1, isReMaterializable = 1 in
defm CONST_V128_#vec : SIMD_I<(outs V128:$dst), ops, (outs), ops,
[(set V128:$dst, (vec.vt pat))],
"v128.const\t$dst, "#args,
"v128.const\t"#args, 12>;
}
defm "" : ConstVec<I8x16,
(ins vec_i8imm_op:$i0, vec_i8imm_op:$i1,
vec_i8imm_op:$i2, vec_i8imm_op:$i3,
vec_i8imm_op:$i4, vec_i8imm_op:$i5,
vec_i8imm_op:$i6, vec_i8imm_op:$i7,
vec_i8imm_op:$i8, vec_i8imm_op:$i9,
vec_i8imm_op:$iA, vec_i8imm_op:$iB,
vec_i8imm_op:$iC, vec_i8imm_op:$iD,
vec_i8imm_op:$iE, vec_i8imm_op:$iF),
(build_vector ImmI8:$i0, ImmI8:$i1, ImmI8:$i2, ImmI8:$i3,
ImmI8:$i4, ImmI8:$i5, ImmI8:$i6, ImmI8:$i7,
ImmI8:$i8, ImmI8:$i9, ImmI8:$iA, ImmI8:$iB,
ImmI8:$iC, ImmI8:$iD, ImmI8:$iE, ImmI8:$iF),
!strconcat("$i0, $i1, $i2, $i3, $i4, $i5, $i6, $i7, ",
"$i8, $i9, $iA, $iB, $iC, $iD, $iE, $iF")>;
defm "" : ConstVec<I16x8,
(ins vec_i16imm_op:$i0, vec_i16imm_op:$i1,
vec_i16imm_op:$i2, vec_i16imm_op:$i3,
vec_i16imm_op:$i4, vec_i16imm_op:$i5,
vec_i16imm_op:$i6, vec_i16imm_op:$i7),
(build_vector
ImmI16:$i0, ImmI16:$i1, ImmI16:$i2, ImmI16:$i3,
ImmI16:$i4, ImmI16:$i5, ImmI16:$i6, ImmI16:$i7),
"$i0, $i1, $i2, $i3, $i4, $i5, $i6, $i7">;
let IsCanonical = 1 in
defm "" : ConstVec<I32x4,
(ins vec_i32imm_op:$i0, vec_i32imm_op:$i1,
vec_i32imm_op:$i2, vec_i32imm_op:$i3),
(build_vector (i32 imm:$i0), (i32 imm:$i1),
(i32 imm:$i2), (i32 imm:$i3)),
"$i0, $i1, $i2, $i3">;
defm "" : ConstVec<I64x2,
(ins vec_i64imm_op:$i0, vec_i64imm_op:$i1),
(build_vector (i64 imm:$i0), (i64 imm:$i1)),
"$i0, $i1">;
defm "" : ConstVec<F32x4,
(ins f32imm_op:$i0, f32imm_op:$i1,
f32imm_op:$i2, f32imm_op:$i3),
(build_vector (f32 fpimm:$i0), (f32 fpimm:$i1),
(f32 fpimm:$i2), (f32 fpimm:$i3)),
"$i0, $i1, $i2, $i3">;
defm "" : ConstVec<F64x2,
(ins f64imm_op:$i0, f64imm_op:$i1),
(build_vector (f64 fpimm:$i0), (f64 fpimm:$i1)),
"$i0, $i1">;
// Shuffle lanes: shuffle
defm SHUFFLE :
SIMD_I<(outs V128:$dst),
(ins V128:$x, V128:$y,
vec_i8imm_op:$m0, vec_i8imm_op:$m1,
vec_i8imm_op:$m2, vec_i8imm_op:$m3,
vec_i8imm_op:$m4, vec_i8imm_op:$m5,
vec_i8imm_op:$m6, vec_i8imm_op:$m7,
vec_i8imm_op:$m8, vec_i8imm_op:$m9,
vec_i8imm_op:$mA, vec_i8imm_op:$mB,
vec_i8imm_op:$mC, vec_i8imm_op:$mD,
vec_i8imm_op:$mE, vec_i8imm_op:$mF),
(outs),
(ins
vec_i8imm_op:$m0, vec_i8imm_op:$m1,
vec_i8imm_op:$m2, vec_i8imm_op:$m3,
vec_i8imm_op:$m4, vec_i8imm_op:$m5,
vec_i8imm_op:$m6, vec_i8imm_op:$m7,
vec_i8imm_op:$m8, vec_i8imm_op:$m9,
vec_i8imm_op:$mA, vec_i8imm_op:$mB,
vec_i8imm_op:$mC, vec_i8imm_op:$mD,
vec_i8imm_op:$mE, vec_i8imm_op:$mF),
[],
"i8x16.shuffle\t$dst, $x, $y, "#
"$m0, $m1, $m2, $m3, $m4, $m5, $m6, $m7, "#
"$m8, $m9, $mA, $mB, $mC, $mD, $mE, $mF",
"i8x16.shuffle\t"#
"$m0, $m1, $m2, $m3, $m4, $m5, $m6, $m7, "#
"$m8, $m9, $mA, $mB, $mC, $mD, $mE, $mF",
13>;
// Shuffles after custom lowering
def wasm_shuffle_t : SDTypeProfile<1, 18, []>;
def wasm_shuffle : SDNode<"WebAssemblyISD::SHUFFLE", wasm_shuffle_t>;
foreach vec = AllVecs in {
def : Pat<(vec.vt (wasm_shuffle (vec.vt V128:$x), (vec.vt V128:$y),
(i32 LaneIdx32:$m0), (i32 LaneIdx32:$m1),
(i32 LaneIdx32:$m2), (i32 LaneIdx32:$m3),
(i32 LaneIdx32:$m4), (i32 LaneIdx32:$m5),
(i32 LaneIdx32:$m6), (i32 LaneIdx32:$m7),
(i32 LaneIdx32:$m8), (i32 LaneIdx32:$m9),
(i32 LaneIdx32:$mA), (i32 LaneIdx32:$mB),
(i32 LaneIdx32:$mC), (i32 LaneIdx32:$mD),
(i32 LaneIdx32:$mE), (i32 LaneIdx32:$mF))),
(SHUFFLE $x, $y,
imm:$m0, imm:$m1, imm:$m2, imm:$m3,
imm:$m4, imm:$m5, imm:$m6, imm:$m7,
imm:$m8, imm:$m9, imm:$mA, imm:$mB,
imm:$mC, imm:$mD, imm:$mE, imm:$mF)>;
}
// Swizzle lanes: i8x16.swizzle
def wasm_swizzle_t : SDTypeProfile<1, 2, []>;
def wasm_swizzle : SDNode<"WebAssemblyISD::SWIZZLE", wasm_swizzle_t>;
defm SWIZZLE :
SIMD_I<(outs V128:$dst), (ins V128:$src, V128:$mask), (outs), (ins),
[(set (v16i8 V128:$dst),
(wasm_swizzle (v16i8 V128:$src), (v16i8 V128:$mask)))],
"i8x16.swizzle\t$dst, $src, $mask", "i8x16.swizzle", 14>;
def : Pat<(int_wasm_swizzle (v16i8 V128:$src), (v16i8 V128:$mask)),
(SWIZZLE $src, $mask)>;
multiclass Splat<Vec vec, bits<32> simdop> {
defm SPLAT_#vec : SIMD_I<(outs V128:$dst), (ins vec.lane_rc:$x),
(outs), (ins),
[(set (vec.vt V128:$dst),
(vec.splat vec.lane_rc:$x))],
vec.prefix#".splat\t$dst, $x", vec.prefix#".splat",
simdop>;
}
defm "" : Splat<I8x16, 15>;
defm "" : Splat<I16x8, 16>;
defm "" : Splat<I32x4, 17>;
defm "" : Splat<I64x2, 18>;
defm "" : Splat<F32x4, 19>;
defm "" : Splat<F64x2, 20>;
// scalar_to_vector leaves high lanes undefined, so can be a splat
foreach vec = AllVecs in
def : Pat<(vec.vt (scalar_to_vector (vec.lane_vt vec.lane_rc:$x))),
(!cast<Instruction>("SPLAT_"#vec) $x)>;
//===----------------------------------------------------------------------===//
// Accessing lanes
//===----------------------------------------------------------------------===//
// Extract lane as a scalar: extract_lane / extract_lane_s / extract_lane_u
multiclass ExtractLane<Vec vec, bits<32> simdop, string suffix = ""> {
defm EXTRACT_LANE_#vec#suffix :
SIMD_I<(outs vec.lane_rc:$dst), (ins V128:$vec, vec_i8imm_op:$idx),
(outs), (ins vec_i8imm_op:$idx), [],
vec.prefix#".extract_lane"#suffix#"\t$dst, $vec, $idx",
vec.prefix#".extract_lane"#suffix#"\t$idx", simdop>;
}
defm "" : ExtractLane<I8x16, 21, "_s">;
defm "" : ExtractLane<I8x16, 22, "_u">;
defm "" : ExtractLane<I16x8, 24, "_s">;
defm "" : ExtractLane<I16x8, 25, "_u">;
defm "" : ExtractLane<I32x4, 27>;
defm "" : ExtractLane<I64x2, 29>;
defm "" : ExtractLane<F32x4, 31>;
defm "" : ExtractLane<F64x2, 33>;
def : Pat<(vector_extract (v16i8 V128:$vec), (i32 LaneIdx16:$idx)),
(EXTRACT_LANE_I8x16_u $vec, imm:$idx)>;
def : Pat<(vector_extract (v8i16 V128:$vec), (i32 LaneIdx8:$idx)),
(EXTRACT_LANE_I16x8_u $vec, imm:$idx)>;
def : Pat<(vector_extract (v4i32 V128:$vec), (i32 LaneIdx4:$idx)),
(EXTRACT_LANE_I32x4 $vec, imm:$idx)>;
def : Pat<(vector_extract (v4f32 V128:$vec), (i32 LaneIdx4:$idx)),
(EXTRACT_LANE_F32x4 $vec, imm:$idx)>;
def : Pat<(vector_extract (v2i64 V128:$vec), (i32 LaneIdx2:$idx)),
(EXTRACT_LANE_I64x2 $vec, imm:$idx)>;
def : Pat<(vector_extract (v2f64 V128:$vec), (i32 LaneIdx2:$idx)),
(EXTRACT_LANE_F64x2 $vec, imm:$idx)>;
def : Pat<
(sext_inreg (vector_extract (v16i8 V128:$vec), (i32 LaneIdx16:$idx)), i8),
(EXTRACT_LANE_I8x16_s $vec, imm:$idx)>;
def : Pat<
(and (vector_extract (v16i8 V128:$vec), (i32 LaneIdx16:$idx)), (i32 0xff)),
(EXTRACT_LANE_I8x16_u $vec, imm:$idx)>;
def : Pat<
(sext_inreg (vector_extract (v8i16 V128:$vec), (i32 LaneIdx8:$idx)), i16),
(EXTRACT_LANE_I16x8_s $vec, imm:$idx)>;
def : Pat<
(and (vector_extract (v8i16 V128:$vec), (i32 LaneIdx8:$idx)), (i32 0xffff)),
(EXTRACT_LANE_I16x8_u $vec, imm:$idx)>;
// Replace lane value: replace_lane
multiclass ReplaceLane<Vec vec, bits<32> simdop> {
defm REPLACE_LANE_#vec :
SIMD_I<(outs V128:$dst), (ins V128:$vec, vec_i8imm_op:$idx, vec.lane_rc:$x),
(outs), (ins vec_i8imm_op:$idx),
[(set V128:$dst, (vector_insert
(vec.vt V128:$vec),
(vec.lane_vt vec.lane_rc:$x),
(i32 vec.lane_idx:$idx)))],
vec.prefix#".replace_lane\t$dst, $vec, $idx, $x",
vec.prefix#".replace_lane\t$idx", simdop>;
}
defm "" : ReplaceLane<I8x16, 23>;
defm "" : ReplaceLane<I16x8, 26>;
defm "" : ReplaceLane<I32x4, 28>;
defm "" : ReplaceLane<I64x2, 30>;
defm "" : ReplaceLane<F32x4, 32>;
defm "" : ReplaceLane<F64x2, 34>;
// Lower undef lane indices to zero
def : Pat<(vector_insert (v16i8 V128:$vec), I32:$x, undef),
(REPLACE_LANE_I8x16 $vec, 0, $x)>;
def : Pat<(vector_insert (v8i16 V128:$vec), I32:$x, undef),
(REPLACE_LANE_I16x8 $vec, 0, $x)>;
def : Pat<(vector_insert (v4i32 V128:$vec), I32:$x, undef),
(REPLACE_LANE_I32x4 $vec, 0, $x)>;
def : Pat<(vector_insert (v2i64 V128:$vec), I64:$x, undef),
(REPLACE_LANE_I64x2 $vec, 0, $x)>;
def : Pat<(vector_insert (v4f32 V128:$vec), F32:$x, undef),
(REPLACE_LANE_F32x4 $vec, 0, $x)>;
def : Pat<(vector_insert (v2f64 V128:$vec), F64:$x, undef),
(REPLACE_LANE_F64x2 $vec, 0, $x)>;
//===----------------------------------------------------------------------===//
// Comparisons
//===----------------------------------------------------------------------===//
multiclass SIMDCondition<Vec vec, string name, CondCode cond, bits<32> simdop> {
defm _#vec :
SIMD_I<(outs V128:$dst), (ins V128:$lhs, V128:$rhs), (outs), (ins),
[(set (vec.int_vt V128:$dst),
(setcc (vec.vt V128:$lhs), (vec.vt V128:$rhs), cond))],
vec.prefix#"."#name#"\t$dst, $lhs, $rhs",
vec.prefix#"."#name, simdop>;
}
multiclass SIMDConditionInt<string name, CondCode cond, bits<32> baseInst> {
defm "" : SIMDCondition<I8x16, name, cond, baseInst>;
defm "" : SIMDCondition<I16x8, name, cond, !add(baseInst, 10)>;
defm "" : SIMDCondition<I32x4, name, cond, !add(baseInst, 20)>;
}
multiclass SIMDConditionFP<string name, CondCode cond, bits<32> baseInst> {
defm "" : SIMDCondition<F32x4, name, cond, baseInst>;
defm "" : SIMDCondition<F64x2, name, cond, !add(baseInst, 6)>;
}
// Equality: eq
let isCommutable = 1 in {
defm EQ : SIMDConditionInt<"eq", SETEQ, 35>;
defm EQ : SIMDCondition<I64x2, "eq", SETEQ, 214>;
defm EQ : SIMDConditionFP<"eq", SETOEQ, 65>;
} // isCommutable = 1
// Non-equality: ne
let isCommutable = 1 in {
defm NE : SIMDConditionInt<"ne", SETNE, 36>;
defm NE : SIMDCondition<I64x2, "ne", SETNE, 215>;
defm NE : SIMDConditionFP<"ne", SETUNE, 66>;
} // isCommutable = 1
// Less than: lt_s / lt_u / lt
defm LT_S : SIMDConditionInt<"lt_s", SETLT, 37>;
defm LT_S : SIMDCondition<I64x2, "lt_s", SETLT, 216>;
defm LT_U : SIMDConditionInt<"lt_u", SETULT, 38>;
defm LT : SIMDConditionFP<"lt", SETOLT, 67>;
// Greater than: gt_s / gt_u / gt
defm GT_S : SIMDConditionInt<"gt_s", SETGT, 39>;
defm GT_S : SIMDCondition<I64x2, "gt_s", SETGT, 217>;
defm GT_U : SIMDConditionInt<"gt_u", SETUGT, 40>;
defm GT : SIMDConditionFP<"gt", SETOGT, 68>;
// Less than or equal: le_s / le_u / le
defm LE_S : SIMDConditionInt<"le_s", SETLE, 41>;
defm LE_S : SIMDCondition<I64x2, "le_s", SETLE, 218>;
defm LE_U : SIMDConditionInt<"le_u", SETULE, 42>;
defm LE : SIMDConditionFP<"le", SETOLE, 69>;
// Greater than or equal: ge_s / ge_u / ge
defm GE_S : SIMDConditionInt<"ge_s", SETGE, 43>;
defm GE_S : SIMDCondition<I64x2, "ge_s", SETGE, 219>;
defm GE_U : SIMDConditionInt<"ge_u", SETUGE, 44>;
defm GE : SIMDConditionFP<"ge", SETOGE, 70>;
// Lower float comparisons that don't care about NaN to standard WebAssembly
// float comparisons. These instructions are generated with nnan and in the
// target-independent expansion of unordered comparisons and ordered ne.
foreach nodes = [[seteq, EQ_F32x4], [setne, NE_F32x4], [setlt, LT_F32x4],
[setgt, GT_F32x4], [setle, LE_F32x4], [setge, GE_F32x4]] in
def : Pat<(v4i32 (nodes[0] (v4f32 V128:$lhs), (v4f32 V128:$rhs))),
(nodes[1] $lhs, $rhs)>;
foreach nodes = [[seteq, EQ_F64x2], [setne, NE_F64x2], [setlt, LT_F64x2],
[setgt, GT_F64x2], [setle, LE_F64x2], [setge, GE_F64x2]] in
def : Pat<(v2i64 (nodes[0] (v2f64 V128:$lhs), (v2f64 V128:$rhs))),
(nodes[1] $lhs, $rhs)>;
//===----------------------------------------------------------------------===//
// Bitwise operations
//===----------------------------------------------------------------------===//
multiclass SIMDBinary<Vec vec, SDPatternOperator node, string name, bits<32> simdop> {
defm _#vec : SIMD_I<(outs V128:$dst), (ins V128:$lhs, V128:$rhs),
(outs), (ins),
[(set (vec.vt V128:$dst),
(node (vec.vt V128:$lhs), (vec.vt V128:$rhs)))],
vec.prefix#"."#name#"\t$dst, $lhs, $rhs",
vec.prefix#"."#name, simdop>;
}
multiclass SIMDBitwise<SDPatternOperator node, string name, bits<32> simdop,
bit commutable = false> {
let isCommutable = commutable in
defm "" : SIMD_I<(outs V128:$dst), (ins V128:$lhs, V128:$rhs),
(outs), (ins), [],
"v128."#name#"\t$dst, $lhs, $rhs", "v128."#name, simdop>;
foreach vec = IntVecs in
def : Pat<(node (vec.vt V128:$lhs), (vec.vt V128:$rhs)),
(!cast<NI>(NAME) $lhs, $rhs)>;
}
multiclass SIMDUnary<Vec vec, SDPatternOperator node, string name, bits<32> simdop> {
defm _#vec : SIMD_I<(outs V128:$dst), (ins V128:$v), (outs), (ins),
[(set (vec.vt V128:$dst),
(vec.vt (node (vec.vt V128:$v))))],
vec.prefix#"."#name#"\t$dst, $v",
vec.prefix#"."#name, simdop>;
}
// Bitwise logic: v128.not
defm NOT : SIMD_I<(outs V128:$dst), (ins V128:$v), (outs), (ins), [],
"v128.not\t$dst, $v", "v128.not", 77>;
foreach vec = IntVecs in
def : Pat<(vnot (vec.vt V128:$v)), (NOT $v)>;
// Bitwise logic: v128.and / v128.or / v128.xor
defm AND : SIMDBitwise<and, "and", 78, true>;
defm OR : SIMDBitwise<or, "or", 80, true>;
defm XOR : SIMDBitwise<xor, "xor", 81, true>;
// Bitwise logic: v128.andnot
def andnot : PatFrag<(ops node:$left, node:$right), (and $left, (vnot $right))>;
defm ANDNOT : SIMDBitwise<andnot, "andnot", 79>;
// Bitwise select: v128.bitselect
defm BITSELECT :
SIMD_I<(outs V128:$dst), (ins V128:$v1, V128:$v2, V128:$c), (outs), (ins), [],
"v128.bitselect\t$dst, $v1, $v2, $c", "v128.bitselect", 82>;
foreach vec = AllVecs in
def : Pat<(vec.vt (int_wasm_bitselect
(vec.vt V128:$v1), (vec.vt V128:$v2), (vec.vt V128:$c))),
(BITSELECT $v1, $v2, $c)>;
// Bitselect is equivalent to (c & v1) | (~c & v2)
foreach vec = IntVecs in
def : Pat<(vec.vt (or (and (vec.vt V128:$c), (vec.vt V128:$v1)),
(and (vnot V128:$c), (vec.vt V128:$v2)))),
(BITSELECT $v1, $v2, $c)>;
// Also implement vselect in terms of bitselect
foreach vec = AllVecs in
def : Pat<(vec.vt (vselect
(vec.int_vt V128:$c), (vec.vt V128:$v1), (vec.vt V128:$v2))),
(BITSELECT $v1, $v2, $c)>;
// MVP select on v128 values
defm SELECT_V128 :
I<(outs V128:$dst), (ins V128:$lhs, V128:$rhs, I32:$cond), (outs), (ins), [],
"v128.select\t$dst, $lhs, $rhs, $cond", "v128.select", 0x1b>;
foreach vec = AllVecs in {
def : Pat<(select I32:$cond, (vec.vt V128:$lhs), (vec.vt V128:$rhs)),
(SELECT_V128 $lhs, $rhs, $cond)>;
// ISD::SELECT requires its operand to conform to getBooleanContents, but
// WebAssembly's select interprets any non-zero value as true, so we can fold
// a setne with 0 into a select.
def : Pat<(select
(i32 (setne I32:$cond, 0)), (vec.vt V128:$lhs), (vec.vt V128:$rhs)),
(SELECT_V128 $lhs, $rhs, $cond)>;
// And again, this time with seteq instead of setne and the arms reversed.
def : Pat<(select
(i32 (seteq I32:$cond, 0)), (vec.vt V128:$lhs), (vec.vt V128:$rhs)),
(SELECT_V128 $rhs, $lhs, $cond)>;
} // foreach vec
//===----------------------------------------------------------------------===//
// Integer unary arithmetic
//===----------------------------------------------------------------------===//
multiclass SIMDUnaryInt<SDPatternOperator node, string name, bits<32> baseInst> {
defm "" : SIMDUnary<I8x16, node, name, baseInst>;
defm "" : SIMDUnary<I16x8, node, name, !add(baseInst, 32)>;
defm "" : SIMDUnary<I32x4, node, name, !add(baseInst, 64)>;
defm "" : SIMDUnary<I64x2, node, name, !add(baseInst, 96)>;
}
// Integer vector negation
def ivneg : PatFrag<(ops node:$in), (sub immAllZerosV, $in)>;
// Integer absolute value: abs
defm ABS : SIMDUnaryInt<abs, "abs", 96>;
// Integer negation: neg
defm NEG : SIMDUnaryInt<ivneg, "neg", 97>;
// Population count: popcnt
defm POPCNT : SIMDUnary<I8x16, ctpop, "popcnt", 0x62>;
// Any lane true: any_true
defm ANYTRUE : SIMD_I<(outs I32:$dst), (ins V128:$vec), (outs), (ins), [],
"v128.any_true\t$dst, $vec", "v128.any_true", 0x53>;
foreach vec = IntVecs in
def : Pat<(int_wasm_anytrue (vec.vt V128:$vec)), (ANYTRUE V128:$vec)>;
// All lanes true: all_true
multiclass SIMDAllTrue<Vec vec, bits<32> simdop> {
defm ALLTRUE_#vec : SIMD_I<(outs I32:$dst), (ins V128:$vec), (outs), (ins),
[(set I32:$dst,
(i32 (int_wasm_alltrue (vec.vt V128:$vec))))],
vec.prefix#".all_true\t$dst, $vec",
vec.prefix#".all_true", simdop>;
}
defm "" : SIMDAllTrue<I8x16, 0x63>;
defm "" : SIMDAllTrue<I16x8, 0x83>;
defm "" : SIMDAllTrue<I32x4, 0xa3>;
defm "" : SIMDAllTrue<I64x2, 0xc3>;
// Reductions already return 0 or 1, so and 1, setne 0, and seteq 1
// can be folded out
foreach reduction =
[["int_wasm_anytrue", "ANYTRUE", "I8x16"],
["int_wasm_anytrue", "ANYTRUE", "I16x8"],
["int_wasm_anytrue", "ANYTRUE", "I32x4"],
["int_wasm_anytrue", "ANYTRUE", "I64x2"],
["int_wasm_alltrue", "ALLTRUE_I8x16", "I8x16"],
["int_wasm_alltrue", "ALLTRUE_I16x8", "I16x8"],
["int_wasm_alltrue", "ALLTRUE_I32x4", "I32x4"],
["int_wasm_alltrue", "ALLTRUE_I64x2", "I64x2"]] in {
defvar intrinsic = !cast<Intrinsic>(reduction[0]);
defvar inst = !cast<NI>(reduction[1]);
defvar vec = !cast<Vec>(reduction[2]);
def : Pat<(i32 (and (i32 (intrinsic (vec.vt V128:$x))), (i32 1))), (inst $x)>;
def : Pat<(i32 (setne (i32 (intrinsic (vec.vt V128:$x))), (i32 0))), (inst $x)>;
def : Pat<(i32 (seteq (i32 (intrinsic (vec.vt V128:$x))), (i32 1))), (inst $x)>;
}
multiclass SIMDBitmask<Vec vec, bits<32> simdop> {
defm _#vec : SIMD_I<(outs I32:$dst), (ins V128:$vec), (outs), (ins),
[(set I32:$dst,
(i32 (int_wasm_bitmask (vec.vt V128:$vec))))],
vec.prefix#".bitmask\t$dst, $vec", vec.prefix#".bitmask",
simdop>;
}
defm BITMASK : SIMDBitmask<I8x16, 100>;
defm BITMASK : SIMDBitmask<I16x8, 132>;
defm BITMASK : SIMDBitmask<I32x4, 164>;
defm BITMASK : SIMDBitmask<I64x2, 196>;
//===----------------------------------------------------------------------===//
// Bit shifts
//===----------------------------------------------------------------------===//
multiclass SIMDShift<Vec vec, SDNode node, string name, bits<32> simdop> {
defm _#vec : SIMD_I<(outs V128:$dst), (ins V128:$vec, I32:$x), (outs), (ins),
[(set (vec.vt V128:$dst), (node V128:$vec, I32:$x))],
vec.prefix#"."#name#"\t$dst, $vec, $x",
vec.prefix#"."#name, simdop>;
}
multiclass SIMDShiftInt<SDNode node, string name, bits<32> baseInst> {
defm "" : SIMDShift<I8x16, node, name, baseInst>;
defm "" : SIMDShift<I16x8, node, name, !add(baseInst, 32)>;
defm "" : SIMDShift<I32x4, node, name, !add(baseInst, 64)>;
defm "" : SIMDShift<I64x2, node, name, !add(baseInst, 96)>;
}
// WebAssembly SIMD shifts are nonstandard in that the shift amount is
// an i32 rather than a vector, so they need custom nodes.
def wasm_shift_t :
SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>, SDTCisVT<2, i32>]>;
def wasm_shl : SDNode<"WebAssemblyISD::VEC_SHL", wasm_shift_t>;
def wasm_shr_s : SDNode<"WebAssemblyISD::VEC_SHR_S", wasm_shift_t>;
def wasm_shr_u : SDNode<"WebAssemblyISD::VEC_SHR_U", wasm_shift_t>;
// Left shift by scalar: shl
defm SHL : SIMDShiftInt<wasm_shl, "shl", 107>;
// Right shift by scalar: shr_s / shr_u
defm SHR_S : SIMDShiftInt<wasm_shr_s, "shr_s", 108>;
defm SHR_U : SIMDShiftInt<wasm_shr_u, "shr_u", 109>;
// Optimize away an explicit mask on a shift count.
def : Pat<(wasm_shl (v16i8 V128:$lhs), (and I32:$rhs, 7)),
(SHL_I8x16 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_s (v16i8 V128:$lhs), (and I32:$rhs, 7)),
(SHR_S_I8x16 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_u (v16i8 V128:$lhs), (and I32:$rhs, 7)),
(SHR_U_I8x16 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shl (v8i16 V128:$lhs), (and I32:$rhs, 15)),
(SHL_I16x8 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_s (v8i16 V128:$lhs), (and I32:$rhs, 15)),
(SHR_S_I16x8 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_u (v8i16 V128:$lhs), (and I32:$rhs, 15)),
(SHR_U_I16x8 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shl (v4i32 V128:$lhs), (and I32:$rhs, 31)),
(SHL_I32x4 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_s (v4i32 V128:$lhs), (and I32:$rhs, 31)),
(SHR_S_I32x4 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shr_u (v4i32 V128:$lhs), (and I32:$rhs, 31)),
(SHR_U_I32x4 V128:$lhs, I32:$rhs)>;
def : Pat<(wasm_shl (v2i64 V128:$lhs), (trunc (and I64:$rhs, 63))),
(SHL_I64x2 V128:$lhs, (I32_WRAP_I64 I64:$rhs))>;
def : Pat<(wasm_shr_s (v2i64 V128:$lhs), (trunc (and I64:$rhs, 63))),
(SHR_S_I64x2 V128:$lhs, (I32_WRAP_I64 I64:$rhs))>;
def : Pat<(wasm_shr_u (v2i64 V128:$lhs), (trunc (and I64:$rhs, 63))),
(SHR_U_I64x2 V128:$lhs, (I32_WRAP_I64 I64:$rhs))>;
//===----------------------------------------------------------------------===//
// Integer binary arithmetic
//===----------------------------------------------------------------------===//
multiclass SIMDBinaryIntNoI8x16<SDPatternOperator node, string name, bits<32> baseInst> {
defm "" : SIMDBinary<I16x8, node, name, !add(baseInst, 32)>;
defm "" : SIMDBinary<I32x4, node, name, !add(baseInst, 64)>;
defm "" : SIMDBinary<I64x2, node, name, !add(baseInst, 96)>;
}
multiclass SIMDBinaryIntSmall<SDPatternOperator node, string name, bits<32> baseInst> {
defm "" : SIMDBinary<I8x16, node, name, baseInst>;
defm "" : SIMDBinary<I16x8, node, name, !add(baseInst, 32)>;
}
multiclass SIMDBinaryIntNoI64x2<SDPatternOperator node, string name, bits<32> baseInst> {
defm "" : SIMDBinaryIntSmall<node, name, baseInst>;
defm "" : SIMDBinary<I32x4, node, name, !add(baseInst, 64)>;
}
multiclass SIMDBinaryInt<SDPatternOperator node, string name, bits<32> baseInst> {