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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
|
// RUN: mlir-opt %s -test-linalg-data-layout-propagation -split-input-file | FileCheck %s
#map0 = affine_map<(d0, d1) -> (d0, d1)>
func.func @dynamic_elem_pack(%arg0: tensor<?x?xf32>, %dest: tensor<?x?x8x2xf32>) -> tensor<?x?x8x2xf32>
{
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%0 = tensor.dim %arg0, %c0 : tensor<?x?xf32>
%1 = tensor.dim %arg0, %c1 : tensor<?x?xf32>
%2 = tensor.empty(%0, %1) : tensor<?x?xf32>
%3 = linalg.generic {indexing_maps = [#map0, #map0], iterator_types = ["parallel", "parallel"]}
ins(%arg0 : tensor<?x?xf32>)
outs(%2 : tensor<?x?xf32>) {
^bb0(%arg3: f32, %arg4: f32):
%4 = arith.addf %arg3, %arg3 : f32
linalg.yield %4 : f32
} -> tensor<?x?xf32>
%4 = tensor.pack %3
inner_dims_pos = [0, 1]
inner_tiles = [8, 2]
into %dest : tensor<?x?xf32> -> tensor<?x?x8x2xf32>
return %4 : tensor<?x?x8x2xf32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<()[s0] -> (s0 ceildiv 8)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<()[s0] -> (s0 ceildiv 2)>
// CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
// CHECK-LABEL: func.func @dynamic_elem_pack
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK-DAG: %[[C0:.+]] = arith.constant 0 : index
// CHECK-DAG: %[[C1:.+]] = arith.constant 1 : index
// CHECK-DAG: %[[D0:.+]] = tensor.dim %[[ARG0]], %[[C0]]
// CHECK-DAG: %[[D1:.+]] = tensor.dim %[[ARG0]], %[[C1]]
// CHECK-DAG: %[[OUTER_D0:.+]] = affine.apply #[[$MAP0]]()[%[[D0]]]
// CHECK-DAG: %[[OUTER_D1:.+]] = affine.apply #[[$MAP1]]()[%[[D1]]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty(%[[OUTER_D0]], %[[OUTER_D1]]) : tensor<?x?x8x2xf32>
// CHECK: %[[PACK_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [0, 1] inner_tiles = [8, 2]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[ELEM:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP2]], #[[$MAP2]]]
// CHECK-SAME: iterator_types = ["parallel", "parallel", "parallel", "parallel"]
// CHECK-SAME: ins(%[[PACK_ARG0]]
// CHECK-SAME: outs(%[[DEST]]
// CHECK: return %[[ELEM]] : tensor<?x?x8x2xf32>
// -----
#map0 = affine_map<(d0, d1) -> (d0, d1)>
func.func @elem_pack_transpose_inner_dims(%arg0: tensor<128x256xi32>, %dest: tensor<4x16x16x32xi32>) -> tensor<4x16x16x32xi32>{
%init = tensor.empty() : tensor<128x256xi32>
%elem = linalg.generic {indexing_maps = [#map0, #map0], iterator_types = ["parallel", "parallel"]}
ins(%arg0 : tensor<128x256xi32>)
outs(%init : tensor<128x256xi32>) {
^bb0(%arg3: i32, %arg4: i32):
%4 = arith.addi %arg3, %arg3 : i32
linalg.yield %4 : i32
} -> tensor<128x256xi32>
%pack = tensor.pack %elem
inner_dims_pos = [1, 0]
inner_tiles = [16, 32]
into %dest : tensor<128x256xi32> -> tensor<4x16x16x32xi32>
return %pack : tensor<4x16x16x32xi32>
}
// CHECK-DAG: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
// CHECK-LABEL: func.func @elem_pack_transpose_inner_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<4x16x16x32xi32>
// CHECK: %[[PACK_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [1, 0] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[ELEM:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP]]]
// CHECK-SAME: iterator_types = ["parallel", "parallel", "parallel", "parallel"]
// CHECK-SAME: ins(%[[PACK_ARG0]]
// CHECK-SAME: outs(%[[DEST]]
// CHECK: return %[[ELEM]] : tensor<4x16x16x32xi32>
// -----
#map0 = affine_map<(d0, d1) -> (d0, d1)>
func.func @elem_pack_transpose_outer_dims(%arg0: tensor<128x256xi32>, %dest: tensor<16x4x32x16xi32>) -> tensor<16x4x32x16xi32>{
%init = tensor.empty() : tensor<128x256xi32>
%elem = linalg.generic {indexing_maps = [#map0, #map0], iterator_types = ["parallel", "parallel"]}
ins(%arg0 : tensor<128x256xi32>)
outs(%init : tensor<128x256xi32>) {
^bb0(%arg3: i32, %arg4: i32):
%4 = arith.addi %arg3, %arg3 : i32
linalg.yield %4 : i32
} -> tensor<128x256xi32>
%pack = tensor.pack %elem
outer_dims_perm = [1, 0]
inner_dims_pos = [0, 1]
inner_tiles = [32, 16]
into %dest : tensor<128x256xi32> -> tensor<16x4x32x16xi32>
return %pack : tensor<16x4x32x16xi32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
// CHECK-LABEL: func.func @elem_pack_transpose_outer_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<16x4x32x16xi32>
// CHECK: %[[PACK_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [1, 0] inner_dims_pos = [0, 1] inner_tiles = [32, 16]
// CHECK-SAME: into %[[ARG0_EMPTY]] : tensor<128x256xi32> -> tensor<16x4x32x16xi32>
// CHECK: %[[ELEM:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP0]], #[[$MAP0]]]
// CHECK-SAME: iterator_types = ["parallel", "parallel", "parallel", "parallel"]
// CHECK-SAME: ins(%[[PACK_ARG0]]
// CHECK-SAME: outs(%[[DEST]]
// CHECK: return %[[ELEM]] : tensor<16x4x32x16xi32>
// -----
#map0 = affine_map<(d0, d1) -> (d0, d1)>
func.func @elem_pack_transpose_inner_and_outer_dims(%arg0: tensor<128x256xi32>, %dest: tensor<16x4x16x32xi32>) -> tensor<16x4x16x32xi32>{
%init = tensor.empty() : tensor<128x256xi32>
%elem = linalg.generic {indexing_maps = [#map0, #map0], iterator_types = ["parallel", "parallel"]}
ins(%arg0 : tensor<128x256xi32>)
outs(%init : tensor<128x256xi32>) {
^bb0(%arg3: i32, %arg4: i32):
%4 = arith.addi %arg3, %arg3 : i32
linalg.yield %4 : i32
} -> tensor<128x256xi32>
%pack = tensor.pack %elem
outer_dims_perm = [1, 0]
inner_dims_pos = [1, 0]
inner_tiles = [16, 32]
into %dest : tensor<128x256xi32> -> tensor<16x4x16x32xi32>
return %pack : tensor<16x4x16x32xi32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
// CHECK-LABEL: func.func @elem_pack_transpose_inner_and_outer_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<16x4x16x32xi32>
// CHECK: %[[PACK_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [1, 0] inner_dims_pos = [1, 0] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[ELEM:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP0]], #[[$MAP0]]]
// CHECK-SAME: iterator_types = ["parallel", "parallel", "parallel", "parallel"]
// CHECK-SAME: ins(%[[PACK_ARG0]]
// CHECK-SAME: outs(%[[DEST]]
// CHECK: return %[[ELEM]] : tensor<16x4x16x32xi32>
// -----
#map0 = affine_map<(d0, d1) -> (d0, d1)>
#map1 = affine_map<(d0, d1) -> (d0)>
#map2 = affine_map<(d0, d1) -> (d1)>
func.func @dynamic_broadcast_pack(%arg0: tensor<?xf32>, %arg1: tensor<?xf32>, %dest: tensor<?x?x8x2xf32>) -> tensor<?x?x8x2xf32>
{
%c0 = arith.constant 0 : index
%0 = tensor.dim %arg0, %c0 : tensor<?xf32>
%1 = tensor.dim %arg1, %c0 : tensor<?xf32>
%2 = tensor.empty(%0, %1) : tensor<?x?xf32>
%3 = linalg.generic {indexing_maps = [#map1, #map2, #map0], iterator_types = ["parallel", "parallel"]}
ins(%arg0, %arg1 : tensor<?xf32>, tensor<?xf32>)
outs(%2 : tensor<?x?xf32>) {
^bb0(%arg3: f32, %arg4: f32, %arg5: f32):
%4 = arith.addf %arg3, %arg4 : f32
linalg.yield %4 : f32
} -> tensor<?x?xf32>
%4 = tensor.pack %3
inner_dims_pos = [0, 1]
inner_tiles = [8, 2]
into %dest : tensor<?x?xf32> -> tensor<?x?x8x2xf32>
return %4 : tensor<?x?x8x2xf32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<()[s0] -> (s0 ceildiv 8)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<()[s0] -> (s0 ceildiv 2)>
// CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d2)>
// CHECK-DAG: #[[$MAP3:.+]] = affine_map<(d0, d1, d2, d3) -> (d1, d3)>
// CHECK-DAG: #[[$MAP4:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
// CHECK-LABEL: func.func @dynamic_broadcast_pack
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK-DAG: %[[C0:.+]] = arith.constant 0 : index
// CHECK-DAG: %[[D0:.+]] = tensor.dim %[[ARG0]], %[[C0]]
// CHECK-DAG: %[[OUTER_D0:.+]] = affine.apply #[[$MAP0]]()[%[[D0]]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty(%[[OUTER_D0]]) : tensor<?x8xf32>
// CHECK: %[[PACK_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [8]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK-DAG: %[[D1:.+]] = tensor.dim %[[ARG1]], %[[C0]]
// CHECK-DAG: %[[OUTER_D1:.+]] = affine.apply #[[$MAP1]]()[%[[D1]]]
// CHECK: %[[ARG1_EMPTY:.+]] = tensor.empty(%[[OUTER_D1]]) : tensor<?x2xf32>
// CHECK: %[[PACK_ARG1:.+]] = tensor.pack %[[ARG1]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [2]
// CHECK-SAME: into %[[ARG1_EMPTY]]
// CHECK: %[[ELEM:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP2]], #[[$MAP3]], #[[$MAP4]]]
// CHECK-SAME: iterator_types = ["parallel", "parallel", "parallel", "parallel"]
// CHECK-SAME: ins(%[[PACK_ARG0]], %[[PACK_ARG0]]
// CHECK-SAME: outs(%[[DEST]]
// CHECK: return %[[ELEM]] : tensor<?x?x8x2xf32>
// -----
#map = affine_map<(d0, d1, d2, d3) -> (d3)>
#map1 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
func.func @elem_pack_transpose_inner_and_outer_dims2(%arg0: tensor<64xf32>, %dest: tensor<1x2x56x57x32xf32>) -> tensor<1x2x56x57x32xf32> {
%0 = tensor.empty() : tensor<1x56x57x64xf32>
%1 = linalg.generic {
indexing_maps = [#map, #map1],
iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
ins(%arg0 : tensor<64xf32>)
outs(%0 : tensor<1x56x57x64xf32>) {
^bb0(%in: f32, %out: f32):
linalg.yield %in : f32
} -> tensor<1x56x57x64xf32>
%2 = tensor.pack %1 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %dest : tensor<1x56x57x64xf32> -> tensor<1x2x56x57x32xf32>
return %2 : tensor<1x2x56x57x32xf32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d1, d4)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
// CHECK-LABEL: func.func @elem_pack_transpose_inner_and_outer_dims2
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<2x32xf32>
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
// CHECK-SAME: ins(%[[PACKED_ARG0]]
// CHECK-SAME: outs(%[[DEST]]
// -----
func.func @transpose_pack(%arg0: tensor<100x128x200x256xi32>, %arg1: tensor<100xi32>, %arg2: tensor<128xi32>, %dest: tensor<100x200x4x16x16x32xi32>) -> tensor<100x200x4x16x16x32xi32>
{
%init_transpose = tensor.empty() : tensor<100x200x128x256xi32>
%transpose = linalg.generic {
indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>,
affine_map<(d0, d1, d2, d3) -> (d0)>,
affine_map<(d0, d1, d2, d3) -> (d1)>,
affine_map<(d0, d1, d2, d3) -> (d0, d2, d1, d3)>],
iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
ins(%arg0, %arg1, %arg2 : tensor<100x128x200x256xi32>, tensor<100xi32>, tensor<128xi32>)
outs(%init_transpose : tensor<100x200x128x256xi32>) {
^bb0(%b0 : i32, %b1 : i32, %b2 : i32, %b3 : i32):
%0 = arith.addi %b0, %b1 : i32
%1 = arith.addi %0, %b2 : i32
linalg.yield %1 : i32
} -> tensor<100x200x128x256xi32>
%4 = tensor.pack %transpose
inner_dims_pos = [3, 2]
inner_tiles = [16, 32]
into %dest : tensor<100x200x128x256xi32> -> tensor<100x200x4x16x16x32xi32>
return %4 : tensor<100x200x4x16x16x32xi32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2, d3, d4, d5)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0)>
// CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d1, d5)>
// CHECK-DAG: #[[$MAP3:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d2, d1, d3, d4, d5)>
// CHECK-LABEL: func.func @transpose_pack
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG2:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<100x4x200x16x16x32xi32>
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [3, 1] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[ARG2_EMPTY:.+]] = tensor.empty() : tensor<4x32xi32>
// CHECK: %[[PACKED_ARG2:.+]] = tensor.pack %[[ARG2]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [32]
// CHECK-SAME: into %[[ARG2_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]], #[[$MAP3]]]
// CHECK-SAME: ins(%[[PACKED_ARG0]], %[[ARG1]], %[[PACKED_ARG2]]
// CHECK-SAME: outs(%[[DEST]]
// -----
func.func @affine_constant_expr_pack(%arg0: tensor<100x128x200x256xi32>, %arg1: tensor<100x1x1x1xi32>, %arg2: tensor<1x128x1x1xi32>, %dest: tensor<100x200x4x16x16x32xi32>) -> tensor<100x200x4x16x16x32xi32>
{
%init_transpose = tensor.empty() : tensor<100x200x128x256xi32>
%transpose = linalg.generic {
indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>,
affine_map<(d0, d1, d2, d3) -> (d0, 0, 0, 0)>,
affine_map<(d0, d1, d2, d3) -> (0, d1, 0, 0)>,
affine_map<(d0, d1, d2, d3) -> (d0, d2, d1, d3)>],
iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
ins(%arg0, %arg1, %arg2 : tensor<100x128x200x256xi32>, tensor<100x1x1x1xi32>, tensor<1x128x1x1xi32>)
outs(%init_transpose : tensor<100x200x128x256xi32>) {
^bb0(%b0 : i32, %b1 : i32, %b2 : i32, %b3 : i32):
%0 = arith.addi %b0, %b1 : i32
%1 = arith.addi %0, %b2 : i32
linalg.yield %1 : i32
} -> tensor<100x200x128x256xi32>
%4 = tensor.pack %transpose
inner_dims_pos = [3, 2]
inner_tiles = [16, 32]
into %dest : tensor<100x200x128x256xi32> -> tensor<100x200x4x16x16x32xi32>
return %4 : tensor<100x200x4x16x16x32xi32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2, d3, d4, d5)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, 0, 0, 0)>
// CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (0, d1, 0, 0, d5)>
// CHECK-DAG: #[[$MAP3:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d2, d1, d3, d4, d5)>
// CHECK-LABEL: func.func @affine_constant_expr_pack
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG2:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<100x4x200x16x16x32xi32>
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [3, 1] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[ARG2_EMPTY:.+]] = tensor.empty() : tensor<1x4x1x1x32xi32>
// CHECK: %[[PACKED_ARG2:.+]] = tensor.pack %[[ARG2]]
// CHECK-SAME: inner_dims_pos = [1] inner_tiles = [32]
// CHECK-SAME: into %[[ARG2_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]], #[[$MAP3]]]
// CHECK-SAME: ins(%[[PACKED_ARG0]], %[[ARG1]], %[[PACKED_ARG2]]
// CHECK-SAME: outs(%[[DEST]]
// -----
func.func @transpose_pack_with_outer_dims(%arg0: tensor<100x128x200x256xi32>, %arg1: tensor<100xi32>, %arg2: tensor<128xi32>, %dest: tensor<200x4x16x100x16x32xi32>) -> tensor<200x4x16x100x16x32xi32>
{
%init_transpose = tensor.empty() : tensor<100x200x128x256xi32>
%transpose = linalg.generic {
indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>,
affine_map<(d0, d1, d2, d3) -> (d0)>,
affine_map<(d0, d1, d2, d3) -> (d1)>,
affine_map<(d0, d1, d2, d3) -> (d0, d2, d1, d3)>],
iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
ins(%arg0, %arg1, %arg2 : tensor<100x128x200x256xi32>, tensor<100xi32>, tensor<128xi32>)
outs(%init_transpose : tensor<100x200x128x256xi32>) {
^bb0(%b0 : i32, %b1 : i32, %b2 : i32, %b3 : i32):
%0 = arith.addi %b0, %b1 : i32
%1 = arith.addi %0, %b2 : i32
linalg.yield %1 : i32
} -> tensor<100x200x128x256xi32>
%4 = tensor.pack %transpose
outer_dims_perm = [1, 2, 3, 0]
inner_dims_pos = [3, 2]
inner_tiles = [16, 32]
into %dest : tensor<100x200x128x256xi32> -> tensor<200x4x16x100x16x32xi32>
return %4 : tensor<200x4x16x100x16x32xi32>
}
// CHECK-DAG: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2, d3, d4, d5)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d3)>
// CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d1, d5)>
// CHECK-LABEL: func.func @transpose_pack_with_outer_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG2:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[DEST:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<200x4x16x100x16x32xi32>
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [2, 1, 3, 0] inner_dims_pos = [3, 1] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[ARG2_EMPTY:.+]] = tensor.empty() : tensor<4x32xi32>
// CHECK: %[[PACKED_ARG2:.+]] = tensor.pack %[[ARG2]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [32]
// CHECK-SAME: into %[[ARG2_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP1]], #[[$MAP2]], #[[$MAP]]]
// CHECK-SAME: ins(%[[PACKED_ARG0]], %[[ARG1]], %[[PACKED_ARG2]]
// CHECK-SAME: outs(%[[DEST]]
// -----
#map0 = affine_map<(d0, d1) -> (d0, d1)>
func.func @elem_pack_transpose_outer_dims(%arg0: tensor<128x256xi32>, %init: tensor<128x256xi32>) -> tensor<16x4x32x16xi32>{
%elem = linalg.generic {indexing_maps = [#map0, #map0], iterator_types = ["parallel", "parallel"]}
ins(%arg0 : tensor<128x256xi32>)
outs(%init : tensor<128x256xi32>) {
^bb0(%arg3: i32, %arg4: i32):
%4 = arith.addi %arg3, %arg4 : i32
linalg.yield %4 : i32
} -> tensor<128x256xi32>
%empty = tensor.empty() : tensor<16x4x32x16xi32>
%pack = tensor.pack %elem
outer_dims_perm = [1, 0]
inner_dims_pos = [0, 1]
inner_tiles = [32, 16]
into %empty : tensor<128x256xi32> -> tensor<16x4x32x16xi32>
return %pack : tensor<16x4x32x16xi32>
}
// CHECK: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
// CHECK-LABEL: func.func @elem_pack_transpose_outer_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[ARG1_EMPTY:.+]] = tensor.empty() : tensor<16x4x32x16xi32>
// CHECK: %[[PACKED_ARG1:.+]] = tensor.pack %[[ARG1]]
// CHECK-SAME: outer_dims_perm = [1, 0] inner_dims_pos = [0, 1] inner_tiles = [32, 16]
// CHECK-SAME: into %[[ARG1_EMPTY]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<16x4x32x16xi32>
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [1, 0] inner_dims_pos = [0, 1] inner_tiles = [32, 16]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP]]]
// CHECK-SAME: ins(%[[PACKED_ARG0]]
// CHECK-SAME: outs(%[[PACKED_ARG1]]
// -----
#map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
func.func @unpack_on_output(%arg0: tensor<12x2x56x56x32xf32>) -> tensor<12x56x56x64xf32> {
%0 = tensor.empty() : tensor<12x56x56x64xf32>
%1 = tensor.unpack %arg0 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %0 : tensor<12x2x56x56x32xf32> -> tensor<12x56x56x64xf32>
%2 = linalg.generic {indexing_maps = [#map], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} outs(%1 : tensor<12x56x56x64xf32>) {
^bb0(%out: f32):
%3 = arith.addf %out, %out : f32
linalg.yield %3 : f32
} -> tensor<12x56x56x64xf32>
return %2 : tensor<12x56x56x64xf32>
}
// CHECK: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
// CHECK-LABEL: func.func @unpack_on_output
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_EMPTY_UNPACK:.+]] = tensor.empty() : tensor<12x56x56x64xf32>
// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_EMPTY_UNPACK]]
// CHECK: %[[ARG0_EMPTY_PACK:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[UNPACKED_ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_EMPTY_PACK]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]]]
// CHECK-SAME: outs(%[[PACKED_ARG0]]
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[RES]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_EMPTY_UNPACK]]
// -----
#map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
func.func @unpack_on_input(%arg0: tensor<12x2x56x56x32xf32>, %init: tensor<12x56x56x64xf32>) -> tensor<12x56x56x64xf32> {
%0 = tensor.empty() : tensor<12x56x56x64xf32>
%1 = tensor.unpack %arg0 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %0 : tensor<12x2x56x56x32xf32> -> tensor<12x56x56x64xf32>
%2 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%1: tensor<12x56x56x64xf32>) outs(%init : tensor<12x56x56x64xf32>) {
^bb0(%in: f32, %out: f32):
%3 = arith.addf %in, %out : f32
linalg.yield %3 : f32
} -> tensor<12x56x56x64xf32>
return %2 : tensor<12x56x56x64xf32>
}
// CHECK: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
// CHECK-LABEL: func.func @unpack_on_input
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_UNPACK_EMPTY:.+]] = tensor.empty() : tensor<12x56x56x64xf32>
// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// CHECK: %[[ARG1_PACK_EMPTY:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
// CHECK: %[[ARG1_PACK:.+]] = tensor.pack %[[ARG1]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG1_PACK_EMPTY]]
// CHECK: %[[ARG0_PACK_EMPTY:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
// CHECK: %[[ARG0_PACK:.+]] = tensor.pack %[[UNPACKED_ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_PACK_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP]]]
// CHECK-SAME: ins(%[[ARG0_PACK]]
// CHECK-SAME: outs(%[[ARG1_PACK]]
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[RES]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// -----
#map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
func.func @unpack_element_type_change(%arg0: tensor<12x2x56x56x32xf32>, %init: tensor<12x56x56x64xf16>) -> tensor<12x56x56x64xf16> {
%0 = tensor.empty() : tensor<12x56x56x64xf32>
%1 = tensor.unpack %arg0 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %0 : tensor<12x2x56x56x32xf32> -> tensor<12x56x56x64xf32>
%2 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%1: tensor<12x56x56x64xf32>) outs(%init : tensor<12x56x56x64xf16>) {
^bb0(%in: f32, %out: f16):
%3 = arith.truncf %in : f32 to f16
linalg.yield %3 : f16
} -> tensor<12x56x56x64xf16>
return %2 : tensor<12x56x56x64xf16>
}
// CHECK: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
// CHECK-LABEL: func.func @unpack_element_type_change
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_UNPACK_EMPTY:.+]] = tensor.empty() : tensor<12x56x56x64xf32>
// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// CHECK: %[[ARG1_PACK_EMPTY:.+]] = tensor.empty() : tensor<12x2x56x56x32xf16>
// CHECK: %[[ARG1_PACK:.+]] = tensor.pack %[[ARG1]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG1_PACK_EMPTY]]
// CHECK: %[[ARG0_PACK_EMPTY:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
// CHECK: %[[ARG0_PACK:.+]] = tensor.pack %[[UNPACKED_ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_PACK_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP]]]
// CHECK-SAME: ins(%[[ARG0_PACK]]
// CHECK-SAME: outs(%[[ARG1_PACK]]
// CHECK: %[[ARG0_NEW_EMPTY_UNPACK:.+]] = tensor.empty() : tensor<12x56x56x64xf16>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[RES]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_NEW_EMPTY_UNPACK]]
// -----
#map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
func.func @forward_tensor_empty(%arg0: tensor<12x2x56x56x32xf32>) -> tensor<12x56x56x64xf32> {
%init = tensor.empty() : tensor<12x56x56x64xf32>
%0 = tensor.empty() : tensor<12x56x56x64xf32>
%1 = tensor.unpack %arg0 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %0 : tensor<12x2x56x56x32xf32> -> tensor<12x56x56x64xf32>
%2 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%1: tensor<12x56x56x64xf32>) outs(%init : tensor<12x56x56x64xf32>) {
^bb0(%in: f32, %out: f32):
%3 = arith.addf %in, %in : f32
linalg.yield %3 : f32
} -> tensor<12x56x56x64xf32>
return %2 : tensor<12x56x56x64xf32>
}
// CHECK: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
// CHECK-LABEL: func.func @forward_tensor_empty
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_UNPACK_EMPTY:.+]] = tensor.empty() : tensor<12x56x56x64xf32>
// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// CHECK: %[[DEST:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
// CHECK: %[[ARG0_PACK_EMPTY:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[UNPACKED_ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_PACK_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP]]]
// CHECK-SAME: ins(%[[PACKED_ARG0]]
// CHECK-SAME: outs(%[[DEST]]
// CHECK: %[[UNPACKED:.+]] = tensor.unpack %[[RES]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// -----
func.func @pad_valid_unpack_propagation(%arg0: tensor<1x2x56x56x32xf32>) -> tensor<1x58x58x64xf32> {
%cst = arith.constant 0.000000e+00 : f32
%0 = tensor.empty() : tensor<1x56x56x64xf32>
%1 = tensor.unpack %arg0 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %0 : tensor<1x2x56x56x32xf32> -> tensor<1x56x56x64xf32>
%padded = tensor.pad %1 low[0, 1, 1, 0] high[0, 1, 1, 0] {
^bb0(%arg3: index, %arg4: index, %arg5: index, %arg6: index):
tensor.yield %cst : f32
} : tensor<1x56x56x64xf32> to tensor<1x58x58x64xf32>
return %padded : tensor<1x58x58x64xf32>
}
// CHECK-LABEL: func.func @pad_valid_unpack_propagation(
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x2x56x56x32xf32>)
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[PADDED:.+]] = tensor.pad %[[ARG0]] low[0, 0, 1, 1, 0] high[0, 0, 1, 1, 0]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x58x58x64xf32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[PADDED]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<1x2x58x58x32xf32> -> tensor<1x58x58x64xf32>
// -----
func.func @pad_valid_unpack_propagation(%arg0: tensor<1x2x56x56x32xf32>) -> tensor<2x58x58x64xf32> {
%cst = arith.constant 0.000000e+00 : f32
%0 = tensor.empty() : tensor<1x56x56x64xf32>
%1 = tensor.unpack %arg0 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %0 : tensor<1x2x56x56x32xf32> -> tensor<1x56x56x64xf32>
%padded = tensor.pad %1 low[1, 1, 1, 0] high[0, 1, 1, 0] {
^bb0(%arg3: index, %arg4: index, %arg5: index, %arg6: index):
tensor.yield %cst : f32
} : tensor<1x56x56x64xf32> to tensor<2x58x58x64xf32>
return %padded : tensor<2x58x58x64xf32>
}
// CHECK-LABEL: func.func @pad_valid_unpack_propagation(
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x2x56x56x32xf32>)
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[PADDED:.+]] = tensor.pad %[[ARG0]] low[1, 0, 1, 1, 0] high[0, 0, 1, 1, 0]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<2x58x58x64xf32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[PADDED]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<2x2x58x58x32xf32> -> tensor<2x58x58x64xf32>
// -----
func.func @pad_along_unpacked_dim(%arg0: tensor<1x2x56x56x32xf32>) -> tensor<1x58x58x66xf32> {
%cst = arith.constant 0.000000e+00 : f32
%0 = tensor.empty() : tensor<1x56x56x64xf32>
%1 = tensor.unpack %arg0 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %0 : tensor<1x2x56x56x32xf32> -> tensor<1x56x56x64xf32>
%padded = tensor.pad %1 low[0, 1, 1, 1] high[0, 1, 1, 1] {
^bb0(%arg3: index, %arg4: index, %arg5: index, %arg6: index):
tensor.yield %cst : f32
} : tensor<1x56x56x64xf32> to tensor<1x58x58x66xf32>
return %padded : tensor<1x58x58x66xf32>
}
// CHECK-LABEL: func.func @pad_along_unpacked_dim(
// CHECK: %[[ARG0:.+]]: tensor<1x2x56x56x32xf32>)
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x56x56x64xf32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<1x2x56x56x32xf32> -> tensor<1x56x56x64xf32>
// CHECK: %[[PADDED:.+]] = tensor.pad %[[UNPACK]] low[0, 1, 1, 1] high[0, 1, 1, 1]
// -----
func.func @pad_valid_pack_propagation(%arg0: tensor<1x64x56x56xf32>) -> tensor<1x2x58x58x32xf32> {
%cst = arith.constant 0.000000e+00 : f32
%padded = tensor.pad %arg0 low[0, 0, 1, 1] high[0, 0, 1, 1] {
^bb0(%arg3: index, %arg4: index, %arg5: index, %arg6: index):
tensor.yield %cst : f32
} : tensor<1x64x56x56xf32> to tensor<1x64x58x58xf32>
%0 = tensor.empty() : tensor<1x2x58x58x32xf32>
%1 = tensor.pack %padded inner_dims_pos = [1] inner_tiles = [32] into %0 : tensor<1x64x58x58xf32> -> tensor<1x2x58x58x32xf32>
return %1 : tensor<1x2x58x58x32xf32>
}
// CHECK-LABEL: func.func @pad_valid_pack_propagation(
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x64x56x56xf32>)
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x2x56x56x32xf32>
// CHECK: %[[PACKED:.+]] = tensor.pack %[[ARG0]] inner_dims_pos = [1] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<1x64x56x56xf32> -> tensor<1x2x56x56x32xf32>
// CHECK: %[[PADDED:.+]] = tensor.pad %[[PACKED]] low[0, 0, 1, 1, 0] high[0, 0, 1, 1, 0]
// CHECK: return %[[PADDED]]
// -----
func.func @pad_valid_outer_dims_pack_propagation(%arg0: tensor<1x64x56x56xf32>) -> tensor<1x58x58x2x32xf32> {
%cst = arith.constant 0.000000e+00 : f32
%padded = tensor.pad %arg0 low[0, 0, 1, 1] high[0, 0, 1, 1] {
^bb0(%arg3: index, %arg4: index, %arg5: index, %arg6: index):
tensor.yield %cst : f32
} : tensor<1x64x56x56xf32> to tensor<1x64x58x58xf32>
%0 = tensor.empty() : tensor<1x58x58x2x32xf32>
%1 = tensor.pack %padded outer_dims_perm = [0, 3, 2, 1] inner_dims_pos = [1] inner_tiles = [32] into %0 : tensor<1x64x58x58xf32> -> tensor<1x58x58x2x32xf32>
return %1 : tensor<1x58x58x2x32xf32>
}
// CHECK-LABEL: func.func @pad_valid_outer_dims_pack_propagation(
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x64x56x56xf32>)
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x56x56x2x32xf32>
// CHECK: %[[PACKED:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 2, 1] inner_dims_pos = [1] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<1x64x56x56xf32> -> tensor<1x56x56x2x32xf32>
// CHECK: %[[PADDED:.+]] = tensor.pad %[[PACKED]] low[0, 1, 1, 0, 0] high[0, 1, 1, 0, 0]
// CHECK: return %[[PADDED]]
// -----
func.func @pad_along_packed_dim(%arg0: tensor<1x60x56x56xf32>) -> tensor<1x2x58x58x32xf32> {
%cst = arith.constant 0.000000e+00 : f32
%padded = tensor.pad %arg0 low[0, 2, 1, 1] high[0, 2, 1, 1] {
^bb0(%arg3: index, %arg4: index, %arg5: index, %arg6: index):
tensor.yield %cst : f32
} : tensor<1x60x56x56xf32> to tensor<1x64x58x58xf32>
%0 = tensor.empty() : tensor<1x2x58x58x32xf32>
%1 = tensor.pack %padded inner_dims_pos = [1] inner_tiles = [32] into %0 : tensor<1x64x58x58xf32> -> tensor<1x2x58x58x32xf32>
return %1 : tensor<1x2x58x58x32xf32>
}
// CHECK-LABEL: func.func @pad_along_packed_dim(
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x60x56x56xf32>)
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[PADDED:.+]] = tensor.pad %[[ARG0]] low[0, 2, 1, 1] high[0, 2, 1, 1]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x2x58x58x32xf32>
// CHECK: tensor.pack %[[PADDED]] inner_dims_pos = [1] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<1x64x58x58xf32> -> tensor<1x2x58x58x32xf32>
// -----
func.func @multi_use_pad_pack_propagation(%arg0: tensor<1x64x56x56xf32>) -> (tensor<1x64x58x58xf32>, tensor<1x2x58x58x32xf32>) {
%cst = arith.constant 0.000000e+00 : f32
%padded = tensor.pad %arg0 low[0, 0, 1, 1] high[0, 0, 1, 1] {
^bb0(%arg3: index, %arg4: index, %arg5: index, %arg6: index):
tensor.yield %cst : f32
} : tensor<1x64x56x56xf32> to tensor<1x64x58x58xf32>
%0 = tensor.empty() : tensor<1x2x58x58x32xf32>
%1 = tensor.pack %padded inner_dims_pos = [1] inner_tiles = [32] into %0 : tensor<1x64x58x58xf32> -> tensor<1x2x58x58x32xf32>
return %padded, %1 : tensor<1x64x58x58xf32>, tensor<1x2x58x58x32xf32>
}
// CHECK-LABEL: func.func @multi_use_pad_pack_propagation(
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x64x56x56xf32>)
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x2x56x56x32xf32>
// CHECK: %[[PACKED:.+]] = tensor.pack %[[ARG0]] inner_dims_pos = [1] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<1x64x56x56xf32> -> tensor<1x2x56x56x32xf32>
// CHECK: %[[PADDED:.+]] = tensor.pad %[[PACKED]] low[0, 0, 1, 1, 0] high[0, 0, 1, 1, 0]
// CHECK: %[[UNPACKED:.+]] = tensor.unpack %[[PADDED]] inner_dims_pos = [1] inner_tiles = [32]
// CHECK: return %[[UNPACKED]], %[[PADDED]]
// -----
#map0 = affine_map<(d0, d1) -> (d0, d1)>
func.func @would_break_dominance(%arg0: tensor<128x256xi32>) -> tensor<4x16x16x32xi32>{
%init = tensor.empty() : tensor<128x256xi32>
%elem = linalg.generic {indexing_maps = [#map0, #map0], iterator_types = ["parallel", "parallel"]}
ins(%arg0 : tensor<128x256xi32>)
outs(%init : tensor<128x256xi32>) {
^bb0(%arg3: i32, %arg4: i32):
%4 = arith.addi %arg3, %arg3 : i32
linalg.yield %4 : i32
} -> tensor<128x256xi32>
%dest = bufferization.alloc_tensor() : tensor<4x16x16x32xi32>
%pack = tensor.pack %elem
inner_dims_pos = [1, 0]
inner_tiles = [16, 32]
into %dest : tensor<128x256xi32> -> tensor<4x16x16x32xi32>
return %pack : tensor<4x16x16x32xi32>
}
// CHECK-LABEL: func.func @would_break_dominance(
// CHECK-SAME: %[[ARG0:.+]]: tensor<128x256xi32>)
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<128x256xi32>
// CHECK-NEXT: %[[GEN:.+]] = linalg.generic
// CHECK-SAME: ins(%[[ARG0]]
// CHECK-SAME: outs(%[[EMPTY]]
// CHECK: %[[ALLOC:.+]] = bufferization.alloc_tensor() : tensor<4x16x16x32xi32>
// CHECK-NEXT: %{{.+}} = tensor.pack %[[GEN]]
// CHECK-SAME: inner_dims_pos = [1, 0] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ALLOC]]
// -----
#map0 = affine_map<(d0, d1, d2, d3) -> ()>
#map1 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
func.func @scalar_tensor(%arg0 : tensor<f32>) -> tensor<1x32x7x7x32xf32> {
%empty_gen = tensor.empty() : tensor<1x7x7x1024xf32>
%gen = linalg.generic {indexing_maps = [#map0, #map1], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%arg0 : tensor<f32>) outs(%empty_gen : tensor<1x7x7x1024xf32>) {
^bb0(%in: f32, %out: f32):
linalg.yield %in : f32
} -> tensor<1x7x7x1024xf32>
%empty_pack = tensor.empty() : tensor<1x32x7x7x32xf32>
%pack = tensor.pack %gen outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32] into %empty_pack : tensor<1x7x7x1024xf32> -> tensor<1x32x7x7x32xf32>
return %pack : tensor<1x32x7x7x32xf32>
}
// CHECK-DAG: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3, d4) -> ()>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
// CHECK-LABEL: func.func @scalar_tensor
// CHECK-SAME: %[[ARG0:.+]]: tensor<f32>)
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x32x7x7x32xf32>
// CHECK: linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP1]]]
// CHECK-SAME: iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel"]
// CHECK-SAME: ins(%[[ARG0]]
// CHECK-SAME: outs(%[[EMPTY]]
// -----
#map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
func.func @unpack_empty_inner_dims(%arg0: tensor<12x64x56x56xf32>) -> tensor<12x56x56x64xf32> {
%init = tensor.empty() : tensor<12x56x56x64xf32>
%0 = tensor.empty() : tensor<12x56x56x64xf32>
%1 = tensor.unpack %arg0 outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = [] into %0 : tensor<12x64x56x56xf32> -> tensor<12x56x56x64xf32>
%2 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%1: tensor<12x56x56x64xf32>) outs(%init : tensor<12x56x56x64xf32>) {
^bb0(%in: f32, %out: f32):
%3 = arith.addf %in, %in : f32
linalg.yield %3 : f32
} -> tensor<12x56x56x64xf32>
return %2 : tensor<12x56x56x64xf32>
}
// CHECK-LABEL: func.func @unpack_empty_inner_dims
// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[UNPACKED_ARG0]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: ins(%[[PACKED_ARG0]]
// CHECK: %[[UNPACKED:.+]] = tensor.unpack %[[RES]]
// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
// -----
#map0 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
#map1 = affine_map<(d0, d1, d2) -> (d0, d1)>
func.func @reduction_pack_transpose_inner_dims(%arg0: tensor<128x256x32xi32>,
%arg1: tensor<128x256xi32>) -> tensor<4x16x16x32xi32>{
%elem = linalg.generic {indexing_maps = [#map0, #map1], iterator_types = ["parallel", "parallel", "reduction"]}
ins(%arg0 : tensor<128x256x32xi32>)
outs(%arg1 : tensor<128x256xi32>) {
^bb0(%arg3: i32, %arg4: i32):
%4 = arith.addi %arg3, %arg4 : i32
linalg.yield %4 : i32
} -> tensor<128x256xi32>
%dest = tensor.empty() : tensor<4x16x16x32xi32>
%pack = tensor.pack %elem
inner_dims_pos = [1, 0]
inner_tiles = [16, 32]
into %dest : tensor<128x256xi32> -> tensor<4x16x16x32xi32>
return %pack : tensor<4x16x16x32xi32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d3, d4)>
// CHECK-LABEL: func.func @reduction_pack_transpose_inner_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[ARG1_EMPTY:.+]] = tensor.empty() : tensor<4x16x16x32xi32>
// CHECK: %[[PACK_ARG1:.+]] = tensor.pack %[[ARG1]]
// CHECK-SAME: inner_dims_pos = [1, 0] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG1_EMPTY]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<4x16x32x16x32xi32>
// CHECK: %[[PACK_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [1, 0] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[RED:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
// CHECK-SAME: iterator_types = ["parallel", "parallel", "reduction", "parallel", "parallel"]
// CHECK-SAME: ins(%[[PACK_ARG0]]
// CHECK-SAME: outs(%[[PACK_ARG1]]
// CHECK: return %[[RED]] : tensor<4x16x16x32xi32>
// -----
func.func @reduction_pack_with_outer_dims(%arg0: tensor<100x128x200x256xi32>, %arg1: tensor<100xi32>,
%arg2: tensor<128xi32>, %init_reduction: tensor<100x128x256xi32>) -> tensor<4x16x100x16x32xi32>
{
%reduction = linalg.generic {
indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>,
affine_map<(d0, d1, d2, d3) -> (d0)>,
affine_map<(d0, d1, d2, d3) -> (d1)>,
affine_map<(d0, d1, d2, d3) -> (d0, d1, d3)>],
iterator_types = ["parallel", "parallel", "reduction", "parallel"]}
ins(%arg0, %arg1, %arg2 : tensor<100x128x200x256xi32>, tensor<100xi32>, tensor<128xi32>)
outs(%init_reduction : tensor<100x128x256xi32>) {
^bb0(%b0 : i32, %b1 : i32, %b2 : i32, %b3 : i32):
%0 = arith.addi %b0, %b1 : i32
%1 = arith.addi %0, %b2 : i32
%2 = arith.addi %1, %b3 : i32
linalg.yield %2 : i32
} -> tensor<100x128x256xi32>
%init_pack = tensor.empty() : tensor<4x16x100x16x32xi32>
%4 = tensor.pack %reduction
outer_dims_perm = [1, 2, 0]
inner_dims_pos = [2, 1]
inner_tiles = [16, 32]
into %init_pack : tensor<100x128x256xi32> -> tensor<4x16x100x16x32xi32>
return %4 : tensor<4x16x100x16x32xi32>
}
// CHECK-DAG: #[[$MAP:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2, d3, d4, d5)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d3)>
// CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d5)>
// CHECK-DAG: #[[$MAP3:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d3, d4, d5)>
// CHECK-LABEL: func.func @reduction_pack_with_outer_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG2:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG3:[a-zA-Z0-9]+]]
// CHECK: %[[ARG3_EMPTY:.+]] = tensor.empty() : tensor<4x16x100x16x32xi32>
// CHECK: %[[PACKED_ARG3:.+]] = tensor.pack %[[ARG3]]
// CHECK-SAME: outer_dims_perm = [1, 2, 0] inner_dims_pos = [2, 1] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG3_EMPTY]]
// CHECK: %[[ARG0_EMPTY:.+]] = tensor.empty() : tensor<4x16x200x100x16x32xi32>
// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: outer_dims_perm = [1, 3, 2, 0] inner_dims_pos = [3, 1] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ARG0_EMPTY]]
// CHECK: %[[ARG2_EMPTY:.+]] = tensor.empty() : tensor<4x32xi32>
// CHECK: %[[PACKED_ARG2:.+]] = tensor.pack %[[ARG2]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [32]
// CHECK-SAME: into %[[ARG2_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP1]], #[[$MAP2]], #[[$MAP3]]]
// CHECK-SAME: ins(%[[PACKED_ARG0]], %[[ARG1]], %[[PACKED_ARG2]]
// CHECK-SAME: outs(%[[PACKED_ARG3]]
// -----
#map0 = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2 * 2 + d4, d3 * 2 + d5)>
#map1 = affine_map<(d0, d1, d2, d3, d4, d5) -> (d4, d5)>
#map2 = affine_map<(d0, d1, d2, d3, d4, d5) -> (d1, d2, d3)>
func.func @unpack_different_destination_shape(%arg0: tensor<1x1x1080x1920x16xi32>,
%filter: tensor<2x2xi32>) -> tensor<16x540x960xi32>{
%init = tensor.empty() : tensor<16x540x960xi32>
%empty = tensor.empty() : tensor<1x16x1080x1920xi32>
%unpack = tensor.unpack %arg0
inner_dims_pos = [1]
inner_tiles = [16]
into %empty : tensor<1x1x1080x1920x16xi32> -> tensor<1x16x1080x1920xi32>
%pool = linalg.generic {indexing_maps = [#map0, #map1, #map2], iterator_types = ["parallel", "parallel", "parallel", "parallel", "reduction", "reduction"]}
ins(%unpack, %filter : tensor<1x16x1080x1920xi32>, tensor<2x2xi32>)
outs(%init : tensor<16x540x960xi32>) {
^bb0(%in: i32, %in_1: i32, %out: i32):
%max = arith.maxui %in, %in_1 : i32
linalg.yield %max : i32
} -> tensor<16x540x960xi32>
return %pool : tensor<16x540x960xi32>
}
// CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d0, d1, d2 * 2 + d4, d3 * 2 + d5, d6)>
// CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d4, d5)>
// CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0, d1, d2, d3, d4, d5, d6) -> (d1, d2, d3, d6)>
// CHECK-LABEL: func.func @unpack_different_destination_shape
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[INIT:.+]] = tensor.empty() : tensor<1x540x960x16xi32>
// CHECK: %[[PACK_EMPTY:.+]] = tensor.empty() : tensor<1x1x1080x1920x16xi32>
// CHECK: %[[PACK_ARG0:.+]] = tensor.pack
// CHECK-SAME: inner_dims_pos = [1] inner_tiles = [16]
// CHECK-SAME: into %[[PACK_EMPTY]]
// CHECK: %[[POOL:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]]]
// CHECK-SAME: iterator_types = ["parallel", "parallel", "parallel", "parallel", "reduction", "reduction", "parallel"]
// CHECK-SAME: ins(%[[PACK_ARG0]], %[[ARG1]]
// CHECK-SAME: outs(%[[INIT]]
// CHECK: %[[UNPACK_NEW_DEST:.+]] = tensor.empty() : tensor<16x540x960xi32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[POOL]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [16]
// CHECK-SAME: into %[[UNPACK_NEW_DEST]]
// CHECK: return %[[UNPACK]] : tensor<16x540x960xi32>
// -----
func.func @bubble_up_pack_through_collapse(%1: tensor<?x16x4xf32>, %dim : index) -> tensor<?x4x8x1xf32> {
%collapsed = tensor.collapse_shape %1 [[0, 1], [2]] : tensor<?x16x4xf32> into tensor<?x4xf32>
%2 = tensor.empty(%dim) : tensor<?x4x8x1xf32>
%pack = tensor.pack %collapsed outer_dims_perm = [0, 1] inner_dims_pos = [0, 1] inner_tiles = [8, 1] into %2 : tensor<?x4xf32> -> tensor<?x4x8x1xf32>
func.return %pack : tensor<?x4x8x1xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_through_collapse
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[C0:.+]] = arith.constant 0 : index
// CHECK: %[[DIM:.+]] = tensor.dim %[[ARG0]], %[[C0]] : tensor<?x16x4xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty(%[[DIM]]) : tensor<?x2x4x8x1xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]] outer_dims_perm = [0, 1, 2] inner_dims_pos = [1, 2] inner_tiles = [8, 1] into %[[EMPTY]] : tensor<?x16x4xf32> -> tensor<?x2x4x8x1xf32>
// CHECK: %[[COLLAPSED:.+]] = tensor.collapse_shape %[[PACK]] {{\[}}[0, 1], [2], [3], [4]] : tensor<?x2x4x8x1xf32> into tensor<?x4x8x1xf32>
// CHECK: return %[[COLLAPSED]] : tensor<?x4x8x1xf32>
// -----
func.func @bubble_up_pack_through_collapse_empty_outer_dims_perm(%1: tensor<?x16x4xf32>, %dim : index) -> tensor<?x4x8x1xf32> {
%collapsed = tensor.collapse_shape %1 [[0, 1], [2]] : tensor<?x16x4xf32> into tensor<?x4xf32>
%2 = tensor.empty(%dim) : tensor<?x4x8x1xf32>
%pack = tensor.pack %collapsed inner_dims_pos = [0, 1] inner_tiles = [8, 1] into %2 : tensor<?x4xf32> -> tensor<?x4x8x1xf32>
func.return %pack : tensor<?x4x8x1xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_through_collapse_empty_outer_dims_perm
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[C0:.+]] = arith.constant 0 : index
// CHECK: %[[DIM:.+]] = tensor.dim %[[ARG0]], %[[C0]] : tensor<?x16x4xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty(%[[DIM]]) : tensor<?x2x4x8x1xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]] inner_dims_pos = [1, 2] inner_tiles = [8, 1] into %[[EMPTY]] : tensor<?x16x4xf32> -> tensor<?x2x4x8x1xf32>
// CHECK: %[[COLLAPSED:.+]] = tensor.collapse_shape %[[PACK]] {{\[}}[0, 1], [2], [3], [4]] : tensor<?x2x4x8x1xf32> into tensor<?x4x8x1xf32>
// CHECK: return %[[COLLAPSED]] : tensor<?x4x8x1xf32>
// -----
func.func @bubble_up_permuted_pack_through_collapse(%1: tensor<4x192x16x256xf32>) -> tensor<4x32x3072x8x1xf32> {
%collapsed = tensor.collapse_shape %1 [[0], [1, 2], [3]] : tensor<4x192x16x256xf32> into tensor<4x3072x256xf32>
%2 = tensor.empty() : tensor<4x32x3072x8x1xf32>
%pack = tensor.pack %collapsed outer_dims_perm = [0, 2, 1] inner_dims_pos = [2, 1] inner_tiles = [8, 1] into %2 : tensor<4x3072x256xf32> -> tensor<4x32x3072x8x1xf32>
func.return %pack : tensor<4x32x3072x8x1xf32>
}
// CHECK-LABEL: func.func @bubble_up_permuted_pack_through_collapse
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<4x32x192x16x8x1xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]] outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3, 2] inner_tiles = [8, 1] into %[[EMPTY]] : tensor<4x192x16x256xf32> -> tensor<4x32x192x16x8x1xf32>
// CHECK: %[[COLLAPSED:.+]] = tensor.collapse_shape %pack {{\[}}[0], [1], [2, 3], [4], [5]] : tensor<4x32x192x16x8x1xf32> into tensor<4x32x3072x8x1xf32>
// CHECK: return %[[COLLAPSED]] : tensor<4x32x3072x8x1xf32>
// -----
func.func @bubble_up_pack_through_unit_collapse(%1: tensor<1x64x1x4xf32>) -> tensor<8x4x8x1xf32> {
%collapsed = tensor.collapse_shape %1 [[0, 1, 2], [3]] : tensor<1x64x1x4xf32> into tensor<64x4xf32>
%2 = tensor.empty() : tensor<8x4x8x1xf32>
%pack = tensor.pack %collapsed outer_dims_perm = [0, 1] inner_dims_pos = [0, 1] inner_tiles = [8, 1] into %2 : tensor<64x4xf32> -> tensor<8x4x8x1xf32>
func.return %pack : tensor<8x4x8x1xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_through_unit_collapse
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x8x1x4x8x1xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]] outer_dims_perm = [0, 1, 2, 3] inner_dims_pos = [1, 3] inner_tiles = [8, 1] into %[[EMPTY]] : tensor<1x64x1x4xf32> -> tensor<1x8x1x4x8x1xf32>
// CHECK: %[[COLLAPSED:.+]] = tensor.collapse_shape %[[PACK]] {{\[}}[0, 1, 2], [3], [4], [5]] : tensor<1x8x1x4x8x1xf32> into tensor<8x4x8x1xf32>
// CHECK: return %[[COLLAPSED]] : tensor<8x4x8x1xf32>
// -----
func.func @bubble_up_pack_through_collapse_on_outer_dims(%1: tensor<?x16x4xf32>, %dim : index) -> tensor<?x1x4xf32> {
%collapsed = tensor.collapse_shape %1 [[0, 1], [2]] : tensor<?x16x4xf32> into tensor<?x4xf32>
%2 = tensor.empty(%dim) : tensor<?x1x4xf32>
%pack = tensor.pack %collapsed outer_dims_perm = [0, 1] inner_dims_pos = [1] inner_tiles = [4] into %2 : tensor<?x4xf32> -> tensor<?x1x4xf32>
func.return %pack : tensor<?x1x4xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_through_collapse_on_outer_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[C0:.+]] = arith.constant 0 : index
// CHECK: %[[DIM:.+]] = tensor.dim %[[ARG0]], %[[C0]] : tensor<?x16x4xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty(%[[DIM]]) : tensor<?x16x1x4xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]] outer_dims_perm = [0, 1, 2] inner_dims_pos = [2] inner_tiles = [4] into %[[EMPTY]] : tensor<?x16x4xf32> -> tensor<?x16x1x4xf32>
// CHECK: %[[COLLAPSED:.+]] = tensor.collapse_shape %[[PACK]] {{\[}}[0, 1], [2], [3]] : tensor<?x16x1x4xf32> into tensor<?x1x4xf32>
// CHECK: return %[[COLLAPSED]] : tensor<?x1x4xf32>
// -----
func.func @no_bubble_up_pack_through_non_divisible_collapse(%1: tensor<3072x64x4xf32>) -> tensor<384x32x8x8xf32> {
%collapsed = tensor.collapse_shape %1 [[0], [1, 2]] : tensor<3072x64x4xf32> into tensor<3072x256xf32>
%2 = tensor.empty() : tensor<384x32x8x8xf32>
%pack = tensor.pack %collapsed outer_dims_perm = [0, 1] inner_dims_pos = [0, 1] inner_tiles = [8, 8] into %2 : tensor<3072x256xf32> -> tensor<384x32x8x8xf32>
func.return %pack : tensor<384x32x8x8xf32>
}
// CHECK-LABEL: func.func @no_bubble_up_pack_through_non_divisible_collapse
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[COLLAPSED:.+]] = tensor.collapse_shape %[[ARG0]] {{\[}}[0], [1, 2]] : tensor<3072x64x4xf32> into tensor<3072x256xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[COLLAPSED]]
// CHECK: return %[[PACK]] : tensor<384x32x8x8xf32>
// -----
func.func @bubble_up_pack_outer_expanded_through_expand(%arg0: tensor<32x64xf32>) -> tensor<4x2x64x4xf32> {
%empty = tensor.empty() : tensor<4x2x64x4xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2]] output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
%pack = tensor.pack %expanded inner_dims_pos = [1] inner_tiles = [4] into %empty : tensor<4x8x64xf32> -> tensor<4x2x64x4xf32>
return %pack : tensor<4x2x64x4xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_outer_expanded_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<8x64x4xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [4] into %[[EMPTY]] : tensor<32x64xf32> -> tensor<8x64x4xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0, 1], [2], [3]]
// CHECK-SAME: output_shape [4, 2, 64, 4] : tensor<8x64x4xf32> into tensor<4x2x64x4xf32>
// CHECK: return %[[EXPANDED]] : tensor<4x2x64x4xf32>
// -----
func.func @bubble_up_pack_inner_expanded_through_expand(%arg0: tensor<32x64xf32>) -> tensor<32x4x4x4xf32> {
%empty = tensor.empty() : tensor<32x4x4x4xf32>
%expanded = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [32, 4, 16] : tensor<32x64xf32> into tensor<32x4x16xf32>
%pack = tensor.pack %expanded inner_dims_pos = [2] inner_tiles = [4] into %empty : tensor<32x4x16xf32> -> tensor<32x4x4x4xf32>
return %pack : tensor<32x4x4x4xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_inner_expanded_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<32x16x4xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [1] inner_tiles = [4] into %[[EMPTY]]
// CHECK-SAME: : tensor<32x64xf32> -> tensor<32x16x4xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0], [1, 2], [3]]
// CHECK-SAME: output_shape [32, 4, 4, 4] : tensor<32x16x4xf32> into tensor<32x4x4x4xf32>
// CHECK: return %[[EXPANDED]] : tensor<32x4x4x4xf32>
// -----
func.func @bubble_up_pack_non_expanded_dims_through_expand(%arg0: tensor<32x64x16xf32>) -> tensor<8x2x32x16x4xf32> {
%empty = tensor.empty() : tensor<8x2x32x16x4xf32>
%expanded = tensor.expand_shape %arg0 [[0], [1, 2], [3]] output_shape [32, 2, 32, 16] : tensor<32x64x16xf32> into tensor<32x2x32x16xf32>
%pack = tensor.pack %expanded inner_dims_pos = [0] inner_tiles = [4] into %empty : tensor<32x2x32x16xf32> -> tensor<8x2x32x16x4xf32>
return %pack : tensor<8x2x32x16x4xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_non_expanded_dims_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<8x64x16x4xf32>
// CHECK: %[[PACK:.+]] = tensor.pack
// CHECK-SAME: %[[ARG0]] inner_dims_pos = [0] inner_tiles = [4] into %[[EMPTY]]
// CHECK-SAME: : tensor<32x64x16xf32> -> tensor<8x64x16x4xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0], [1, 2], [3], [4]]
// CHECK-SAME: output_shape [8, 2, 32, 16, 4] : tensor<8x64x16x4xf32> into tensor<8x2x32x16x4xf32>
// CHECK: return %[[EXPANDED]] : tensor<8x2x32x16x4xf32>
// -----
func.func @bubble_up_pack_through_expand_dynamic(%arg0: tensor<?x64xf32>) -> tensor<?x4x2x8xf32> {
%c0 = arith.constant 0 : index
%dim = tensor.dim %arg0, %c0 : tensor<?x64xf32>
%empty = tensor.empty(%dim) : tensor<?x4x2x8xf32>
%expanded = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [%dim, 4, 16] : tensor<?x64xf32> into tensor<?x4x16xf32>
%pack = tensor.pack %expanded inner_dims_pos = [2] inner_tiles = [8] into %empty : tensor<?x4x16xf32> -> tensor<?x4x2x8xf32>
return %pack : tensor<?x4x2x8xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_through_expand_dynamic(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-DAG: %[[C0:.+]] = arith.constant 0 : index
// CHECK: %[[DIM_INPUT:.+]] = tensor.dim %[[ARG0]], %[[C0]] : tensor<?x64xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty(%[[DIM_INPUT]]) : tensor<?x8x8xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [1] inner_tiles = [8] into %[[EMPTY]]
// CHECK-SAME: : tensor<?x64xf32> -> tensor<?x8x8xf32>
// CHECK: %[[DIM_PACK:.+]] = tensor.dim %[[PACK]], %[[C0]] : tensor<?x8x8xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0], [1, 2], [3]]
// CHECK-SAME: output_shape [%[[DIM_PACK]], 4, 2, 8] : tensor<?x8x8xf32> into tensor<?x4x2x8xf32>
// CHECK: return %[[EXPANDED]] : tensor<?x4x2x8xf32>
// -----
func.func @bubble_up_pack_non_expanded_padding_through_expand(%arg0: tensor<32x60xf32>) -> tensor<4x2x8x4x8xf32> {
%cst = arith.constant 3.000000e+00 : f32
%empty = tensor.empty() : tensor<4x2x8x4x8xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2]] output_shape [4, 8, 64] : tensor<32x60xf32> into tensor<4x8x60xf32>
%pack = tensor.pack %expanded padding_value(%cst : f32) inner_dims_pos = [1, 2] inner_tiles = [4, 8] into %empty : tensor<4x8x60xf32> -> tensor<4x2x8x4x8xf32>
return %pack : tensor<4x2x8x4x8xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_non_expanded_padding_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-DAG: %[[CST:.+]] = arith.constant 3.000000e+00 : f32
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<8x8x4x8xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]] padding_value(%[[CST]] : f32)
// CHECK-SAME: inner_dims_pos = [0, 1] inner_tiles = [4, 8] into %[[EMPTY]]
// CHECK-SAME: : tensor<32x60xf32> -> tensor<8x8x4x8xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0, 1], [2], [3], [4]]
// CHECK-SAME: output_shape [4, 2, 8, 4, 8] : tensor<8x8x4x8xf32> into tensor<4x2x8x4x8xf32>
// CHECK: return %[[EXPANDED]] : tensor<4x2x8x4x8xf32>
// -----
func.func @bubble_up_pack_outer_dims_perm_identity_through_expand(%arg0: tensor<32x64xf32>) -> tensor<4x2x32x4x2xf32> {
%empty = tensor.empty() : tensor<4x2x32x4x2xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2]] output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
%pack = tensor.pack %expanded outer_dims_perm = [0, 1, 2] inner_dims_pos = [1, 2] inner_tiles = [4, 2] into %empty : tensor<4x8x64xf32> -> tensor<4x2x32x4x2xf32>
return %pack : tensor<4x2x32x4x2xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_outer_dims_perm_identity_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<8x32x4x2xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [0, 1] inner_tiles = [4, 2] into %[[EMPTY]]
// CHECK-SAME: : tensor<32x64xf32> -> tensor<8x32x4x2xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0, 1], [2], [3], [4]]
// CHECK-SAME: output_shape [4, 2, 32, 4, 2] : tensor<8x32x4x2xf32> into tensor<4x2x32x4x2xf32>
// CHECK: return %[[EXPANDED]] : tensor<4x2x32x4x2xf32>
// -----
func.func @bubble_up_pack_multiple_dims_through_expand(%arg0: tensor<32x64x16xf32>) -> tensor<8x2x4x8x4x8x2xf32> {
%empty = tensor.empty() : tensor<8x2x4x8x4x8x2xf32>
%expanded = tensor.expand_shape %arg0 [[0], [1, 2], [3]] output_shape [32, 2, 32, 16] : tensor<32x64x16xf32> into tensor<32x2x32x16xf32>
%pack = tensor.pack %expanded inner_dims_pos = [0, 2, 3] inner_tiles = [4, 8, 2] into %empty : tensor<32x2x32x16xf32> -> tensor<8x2x4x8x4x8x2xf32>
return %pack : tensor<8x2x4x8x4x8x2xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_multiple_dims_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<8x8x8x4x8x2xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [0, 1, 2] inner_tiles = [4, 8, 2] into %[[EMPTY]]
// CHECK-SAME: : tensor<32x64x16xf32> -> tensor<8x8x8x4x8x2xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0], [1, 2], [3], [4], [5], [6]]
// CHECK-SAME: output_shape [8, 2, 4, 8, 4, 8, 2] : tensor<8x8x8x4x8x2xf32> into tensor<8x2x4x8x4x8x2xf32>
// CHECK: return %[[EXPANDED]] : tensor<8x2x4x8x4x8x2xf32>
// -----
func.func @bubble_up_pack_inner_dims_reorder_through_expand(%arg0: tensor<32x64xf32>) -> tensor<4x2x4x16x4xf32> {
%empty = tensor.empty() : tensor<4x2x4x16x4xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2]] output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
%pack = tensor.pack %expanded inner_dims_pos = [2, 1] inner_tiles = [16, 4] into %empty : tensor<4x8x64xf32> -> tensor<4x2x4x16x4xf32>
return %pack : tensor<4x2x4x16x4xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_inner_dims_reorder_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<8x4x16x4xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [1, 0] inner_tiles = [16, 4] into %[[EMPTY]]
// CHECK-SAME: : tensor<32x64xf32> -> tensor<8x4x16x4xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0, 1], [2], [3], [4]]
// CHECK-SAME: output_shape [4, 2, 4, 16, 4] : tensor<8x4x16x4xf32> into tensor<4x2x4x16x4xf32>
// CHECK: return %[[EXPANDED]] : tensor<4x2x4x16x4xf32>
// -----
func.func @bubble_up_pack_multiple_different_expanded_dims_through_expand(%arg0: tensor<32x64x16xf32>) -> tensor<4x2x2x8x16x4x4xf32> {
%empty = tensor.empty() : tensor<4x2x2x8x16x4x4xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2, 3], [4]] output_shape [4, 8, 2, 32, 16] : tensor<32x64x16xf32> into tensor<4x8x2x32x16xf32>
%pack = tensor.pack %expanded inner_dims_pos = [1, 3] inner_tiles = [4, 4] into %empty : tensor<4x8x2x32x16xf32> -> tensor<4x2x2x8x16x4x4xf32>
return %pack : tensor<4x2x2x8x16x4x4xf32>
}
// CHECK-LABEL: func.func @bubble_up_pack_multiple_different_expanded_dims_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<8x16x16x4x4xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[ARG0]]
// CHECK-SAME: inner_dims_pos = [0, 1] inner_tiles = [4, 4] into %[[EMPTY]]
// CHECK-SAME: : tensor<32x64x16xf32> -> tensor<8x16x16x4x4xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[PACK]] {{\[}}[0, 1], [2, 3], [4], [5], [6]]
// CHECK-SAME: output_shape [4, 2, 2, 8, 16, 4, 4] : tensor<8x16x16x4x4xf32> into tensor<4x2x2x8x16x4x4xf32>
// CHECK: return %[[EXPANDED]] : tensor<4x2x2x8x16x4x4xf32>
// -----
func.func @no_bubble_up_pack_outer_dims_permutation_through_expand(%arg0: tensor<32x64xf32>) -> tensor<32x4x2x4x2xf32> {
%empty = tensor.empty() : tensor<32x4x2x4x2xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2]] output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
%pack = tensor.pack %expanded outer_dims_perm = [2, 0, 1] inner_dims_pos = [1, 2] inner_tiles = [4, 2] into %empty : tensor<4x8x64xf32> -> tensor<32x4x2x4x2xf32>
return %pack : tensor<32x4x2x4x2xf32>
}
// CHECK-LABEL: func.func @no_bubble_up_pack_outer_dims_permutation_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<32x4x2x4x2xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0, 1], [2]]
// CHECK-SAME: output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[EXPANDED]]
// CHECK-SAME: outer_dims_perm = [2, 0, 1] inner_dims_pos = [1, 2] inner_tiles = [4, 2] into %[[EMPTY]]
// CHECK-SAME: : tensor<4x8x64xf32> -> tensor<32x4x2x4x2xf32>
// CHECK: return %[[PACK]] : tensor<32x4x2x4x2xf32>
// -----
func.func @no_bubble_up_pack_multiple_same_expanded_dim_through_expand(%arg0: tensor<32x64xf32>) -> tensor<2x2x64x2x4xf32> {
%empty = tensor.empty() : tensor<2x2x64x2x4xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2]] output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
%pack = tensor.pack %expanded inner_dims_pos = [0, 1] inner_tiles = [2, 4] into %empty : tensor<4x8x64xf32> -> tensor<2x2x64x2x4xf32>
return %pack : tensor<2x2x64x2x4xf32>
}
// CHECK-LABEL: func.func @no_bubble_up_pack_multiple_same_expanded_dim_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<2x2x64x2x4xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0, 1], [2]]
// CHECK-SAME: output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[EXPANDED]]
// CHECK-SAME: inner_dims_pos = [0, 1] inner_tiles = [2, 4] into %[[EMPTY]]
// CHECK-SAME: : tensor<4x8x64xf32> -> tensor<2x2x64x2x4xf32>
// CHECK: return %[[PACK]] : tensor<2x2x64x2x4xf32>
// -----
func.func @no_bubble_up_pack_non_innermost_expanded_dim_through_expand(%arg0: tensor<32x64xf32>) -> tensor<2x8x64x2xf32> {
%empty = tensor.empty() : tensor<2x8x64x2xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2]] output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
%pack = tensor.pack %expanded inner_dims_pos = [0] inner_tiles = [2] into %empty : tensor<4x8x64xf32> -> tensor<2x8x64x2xf32>
return %pack : tensor<2x8x64x2xf32>
}
// CHECK-LABEL: func.func @no_bubble_up_pack_non_innermost_expanded_dim_through_expand(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<2x8x64x2xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0, 1], [2]]
// CHECK-SAME: output_shape [4, 8, 64] : tensor<32x64xf32> into tensor<4x8x64xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[EXPANDED]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [2] into %[[EMPTY]]
// CHECK-SAME: : tensor<4x8x64xf32> -> tensor<2x8x64x2xf32>
// CHECK: return %[[PACK]] : tensor<2x8x64x2xf32>
// -----
func.func @no_bubble_up_pack_expanded_padding_through_expand_cannot_reassociate(%arg0: tensor<30x60xf32>) -> tensor<3x2x60x8xf32> {
%cst = arith.constant 3.000000e+00 : f32
%empty = tensor.empty() : tensor<3x2x60x8xf32>
%expanded = tensor.expand_shape %arg0 [[0, 1], [2]] output_shape [3, 10, 60] : tensor<30x60xf32> into tensor<3x10x60xf32>
%pack = tensor.pack %expanded padding_value(%cst : f32) inner_dims_pos = [1] inner_tiles = [8] into %empty : tensor<3x10x60xf32> -> tensor<3x2x60x8xf32>
return %pack : tensor<3x2x60x8xf32>
}
// CHECK-LABEL: func.func @no_bubble_up_pack_expanded_padding_through_expand_cannot_reassociate(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-DAG: %[[CST:.+]] = arith.constant 3.000000e+00 : f32
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<3x2x60x8xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0, 1], [2]]
// CHECK-SAME: output_shape [3, 10, 60] : tensor<30x60xf32> into tensor<3x10x60xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[EXPANDED]] padding_value(%[[CST]] : f32)
// CHECK-SAME: inner_dims_pos = [1] inner_tiles = [8] into %[[EMPTY]]
// CHECK-SAME: : tensor<3x10x60xf32> -> tensor<3x2x60x8xf32>
// CHECK: return %[[PACK]] : tensor<3x2x60x8xf32>
// -----
func.func @no_bubble_up_pack_extending_dimension_through_expand_cannot_reassociate(%arg0: tensor<32x64xf32>) -> tensor<8x4x16x8xf32> {
%empty = tensor.empty() : tensor<8x4x16x8xf32>
%expanded = tensor.expand_shape %arg0 [[0], [1, 2]] output_shape [32, 4, 16] : tensor<32x64xf32> into tensor<32x4x16xf32>
%pack = tensor.pack %expanded inner_dims_pos = [0] inner_tiles = [8] into %empty : tensor<32x4x16xf32> -> tensor<8x4x16x8xf32>
return %pack : tensor<8x4x16x8xf32>
}
// CHECK-LABEL: func.func @no_bubble_up_pack_extending_dimension_through_expand_cannot_reassociate(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<8x4x16x8xf32>
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0], [1, 2]]
// CHECK-SAME: output_shape [32, 4, 16] : tensor<32x64xf32> into tensor<32x4x16xf32>
// CHECK: %[[PACK:.+]] = tensor.pack %[[EXPANDED]]
// CHECK-SAME: inner_dims_pos = [0] inner_tiles = [8] into %[[EMPTY]]
// CHECK-SAME: : tensor<32x4x16xf32> -> tensor<8x4x16x8xf32>
// CHECK: return %[[PACK]] : tensor<8x4x16x8xf32>
// -----
func.func @push_down_unpack_through_expand(%5: tensor<?x32x8x8xf32>, %dim: index, %sz0: index) -> tensor<?x256x256xf32> {
%6 = tensor.empty(%dim) : tensor<?x256xf32>
%unpack = tensor.unpack %5 outer_dims_perm = [0, 1] inner_dims_pos = [0, 1] inner_tiles = [8, 8] into %6 : tensor<?x32x8x8xf32> -> tensor<?x256xf32>
%expanded = tensor.expand_shape %unpack [[0, 1], [2]] output_shape [%sz0, 256, 256] : tensor<?x256xf32> into tensor<?x256x256xf32>
func.return %expanded : tensor<?x256x256xf32>
}
// CHECK-LABEL: func.func @push_down_unpack_through_expand
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[C32:.+]] = arith.constant 32 : index
// CHECK: %[[C0:.+]] = arith.constant 0 : index
// CHECK: %[[DIM0:.+]] = tensor.dim %[[ARG0]], %[[C0]] : tensor<?x32x8x8xf32>
// CHECK: %[[SZ0:.+]] = arith.divui %[[DIM0]], %[[C32]] : index
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0, 1], [2], [3], [4]] output_shape [%[[SZ0]], 32, 32, 8, 8] : tensor<?x32x8x8xf32> into tensor<?x32x32x8x8xf32>
// CHECK: %[[DIM:.+]] = tensor.dim %[[EXPANDED]], %[[C0]] : tensor<?x32x32x8x8xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty(%[[DIM]]) : tensor<?x256x256xf32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[EXPANDED:.+]] outer_dims_perm = [0, 1, 2] inner_dims_pos = [1, 2] inner_tiles = [8, 8] into %[[EMPTY]] : tensor<?x32x32x8x8xf32> -> tensor<?x256x256xf32>
// CHECK: return %[[UNPACK]] : tensor<?x256x256xf32>
// -----
func.func @push_down_unpack_through_expand_empty_outer_dims_perm(%5: tensor<?x32x8x8xf32>, %dim: index, %sz0: index) -> tensor<?x256x256xf32> {
%6 = tensor.empty(%dim) : tensor<?x256xf32>
%unpack = tensor.unpack %5 inner_dims_pos = [0, 1] inner_tiles = [8, 8] into %6 : tensor<?x32x8x8xf32> -> tensor<?x256xf32>
%expanded = tensor.expand_shape %unpack [[0, 1], [2]] output_shape [%sz0, 256, 256] : tensor<?x256xf32> into tensor<?x256x256xf32>
func.return %expanded : tensor<?x256x256xf32>
}
// CHECK-LABEL: func.func @push_down_unpack_through_expand_empty_outer_dims_perm
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[C32:.+]] = arith.constant 32 : index
// CHECK: %[[C0:.+]] = arith.constant 0 : index
// CHECK: %[[DIM0:.+]] = tensor.dim %[[ARG0]], %[[C0]] : tensor<?x32x8x8xf32>
// CHECK: %[[SZ0:.+]] = arith.divui %[[DIM0]], %[[C32]] : index
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0, 1], [2], [3], [4]] output_shape [%[[SZ0]], 32, 32, 8, 8] : tensor<?x32x8x8xf32> into tensor<?x32x32x8x8xf32>
// CHECK: %[[DIM:.+]] = tensor.dim %[[EXPANDED]], %[[C0]] : tensor<?x32x32x8x8xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty(%[[DIM]]) : tensor<?x256x256xf32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[EXPANDED:.+]] inner_dims_pos = [1, 2] inner_tiles = [8, 8] into %[[EMPTY]] : tensor<?x32x32x8x8xf32> -> tensor<?x256x256xf32>
// CHECK: return %[[UNPACK]] : tensor<?x256x256xf32>
// -----
func.func @push_down_permuted_unpack_through_expand(%5: tensor<4x32x384x8x8xf32>) -> tensor<4x12x256x256xf32> {
%6 = tensor.empty() : tensor<4x3072x256xf32>
%unpack = tensor.unpack %5 outer_dims_perm = [0, 2, 1] inner_dims_pos = [2, 1] inner_tiles = [8, 8] into %6 : tensor<4x32x384x8x8xf32> -> tensor<4x3072x256xf32>
%expanded = tensor.expand_shape %unpack [[0], [1, 2], [3]] output_shape [4, 12, 256, 256] : tensor<4x3072x256xf32> into tensor<4x12x256x256xf32>
func.return %expanded : tensor<4x12x256x256xf32>
}
// CHECK-LABEL: @push_down_permuted_unpack_through_expand
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0], [1], [2, 3], [4], [5]] output_shape [4, 32, 12, 32, 8, 8] : tensor<4x32x384x8x8xf32> into tensor<4x32x12x32x8x8xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<4x12x256x256xf32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[EXPANDED]] outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3, 2] inner_tiles = [8, 8] into %[[EMPTY]] : tensor<4x32x12x32x8x8xf32> -> tensor<4x12x256x256xf32>
// CHECK: return %[[UNPACK]] : tensor<4x12x256x256xf32>
// -----
func.func @push_down_unpack_through_unit_expand(%5: tensor<6x32x8x8xf32>) -> tensor<3x16x1x256xf32> {
%6 = tensor.empty() : tensor<48x256xf32>
%unpack = tensor.unpack %5 outer_dims_perm = [0, 1] inner_dims_pos = [0, 1] inner_tiles = [8, 8] into %6 : tensor<6x32x8x8xf32> -> tensor<48x256xf32>
%expanded = tensor.expand_shape %unpack [[0, 1, 2], [3]] output_shape [3, 16, 1, 256] : tensor<48x256xf32> into tensor<3x16x1x256xf32>
func.return %expanded : tensor<3x16x1x256xf32>
}
// CHECK-LABEL: func.func @push_down_unpack_through_unit_expand
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0, 1, 2], [3], [4], [5]] output_shape [3, 2, 1, 32, 8, 8] : tensor<6x32x8x8xf32> into tensor<3x2x1x32x8x8xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<3x16x1x256xf32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[EXPANDED]] outer_dims_perm = [0, 1, 2, 3] inner_dims_pos = [1, 3] inner_tiles = [8, 8] into %[[EMPTY]] : tensor<3x2x1x32x8x8xf32> -> tensor<3x16x1x256xf32>
// CHECK: return %[[UNPACK]] : tensor<3x16x1x256xf32>
// -----
func.func @push_down_unpack_through_expand_on_outer_dims(%5: tensor<?x32x8xf32>, %dim: index, %sz0: index) -> tensor<?x256x256xf32> {
%6 = tensor.empty(%dim) : tensor<?x256xf32>
%unpack = tensor.unpack %5 outer_dims_perm = [0, 1] inner_dims_pos = [1] inner_tiles = [8] into %6 : tensor<?x32x8xf32> -> tensor<?x256xf32>
%expanded = tensor.expand_shape %unpack [[0, 1], [2]] output_shape [%sz0, 256, 256] : tensor<?x256xf32> into tensor<?x256x256xf32>
func.return %expanded : tensor<?x256x256xf32>
}
// CHECK-LABEL: func.func @push_down_unpack_through_expand_on_outer_dims
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[C256:.+]] = arith.constant 256 : index
// CHECK: %[[C0:.+]] = arith.constant 0 : index
// CHECK: %[[DIM0:.+]] = tensor.dim %[[ARG0]], %[[C0]] : tensor<?x32x8xf32>
// CHECK: %[[SZ0:.+]] = arith.divui %[[DIM0]], %[[C256]] : index
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[ARG0]] {{\[}}[0, 1], [2], [3]] output_shape [%[[SZ0]], 256, 32, 8] : tensor<?x32x8xf32> into tensor<?x256x32x8xf32>
// CHECK: %[[DIM:.+]] = tensor.dim %[[EXPANDED]], %[[C0]] : tensor<?x256x32x8xf32>
// CHECK: %[[EMPTY:.+]] = tensor.empty(%[[DIM]]) : tensor<?x256x256xf32>
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[EXPANDED:.+]] outer_dims_perm = [0, 1, 2] inner_dims_pos = [2] inner_tiles = [8] into %[[EMPTY]] : tensor<?x256x32x8xf32> -> tensor<?x256x256xf32>
// CHECK: return %[[UNPACK]] : tensor<?x256x256xf32>
// -----
func.func @no_push_down_unpack_through_non_divisible_expand(%5: tensor<384x32x8x8xf32>) -> tensor<256x12x256xf32> {
%6 = tensor.empty() : tensor<3072x256xf32>
%unpack = tensor.unpack %5 outer_dims_perm = [0, 1] inner_dims_pos = [0, 1] inner_tiles = [8, 8] into %6 : tensor<384x32x8x8xf32> -> tensor<3072x256xf32>
%expanded = tensor.expand_shape %unpack [[0, 1], [2]] output_shape [256, 12, 256] : tensor<3072x256xf32> into tensor<256x12x256xf32>
func.return %expanded : tensor<256x12x256xf32>
}
// CHECK-LABEL: func.func @no_push_down_unpack_through_non_divisible_expand
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[ARG0]]
// CHECK: %[[EXPANDED:.+]] = tensor.expand_shape %[[UNPACK]] {{\[}}[0, 1], [2]] output_shape [256, 12, 256] : tensor<3072x256xf32> into tensor<256x12x256xf32>
// CHECK: return %[[EXPANDED]] : tensor<256x12x256xf32>
|
