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
|
// RUN: mlir-opt %s -test-linalg-transform-patterns=test-patterns | FileCheck %s
// CHECK-DAG: #[[$STRIDED_1D:.*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s0)>
// Map corresponding to a 2D memory access where the stride along the last dim is known to be 1.
// CHECK-DAG: #[[$STRIDED_2D_u_1:.*]] = affine_map<(d0, d1)[s0, s1] -> (d0 * s1 + s0 + d1)>
// Map corresponding to a 2D memory access where the stride along all dims are unknown.
// CHECK-DAG: #[[$STRIDED_2D:.*]] = affine_map<(d0, d1)[s0, s1, s2] -> (d0 * s1 + s0 + d1 * s2)>
// CHECK-DAG: #[[$mk:.*]] = affine_map<(d0, d1, d2) -> (d0, d2)>
// CHECK-DAG: #[[$kn:.*]] = affine_map<(d0, d1, d2) -> (d2, d1)>
// CHECK-DAG: #[[$mn:.*]] = affine_map<(d0, d1, d2) -> (d0, d1)>
// CHECK-DAG: #[[$nm:.*]] = affine_map<(d0, d1, d2) -> (d1, d0)>
// CHECK-DAG: #[[$km:.*]] = affine_map<(d0, d1, d2) -> (d2, d0)>
func @dot(%x: memref<?xf32, offset: ?, strides: [1]>,
%y: memref<?xf32, offset: ?, strides: [1]>,
%v: memref<f32>) {
linalg.dot { __internal_linalg_transform__ = "MEM" }
ins(%x, %y: memref<?xf32, offset: ?, strides: [1]>,
memref<?xf32, offset: ?, strides: [1]>)
outs(%v: memref<f32>)
return
}
// CHECK-LABEL: func @dot
// CHECK-DAG: %[[c0:.*]] = constant 0 : index
// CHECK-DAG: %[[c1:.*]] = constant 1 : index
// CHECK-DAG: %[[c8000:.*]] = constant 8000 : index
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c8000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c1]] {
// CHECK: load
// CHECK: load
// CHECK: load
// CHECK: mulf
// CHECK: addf
// CHECK: store
func @matvec(%A: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%x: memref<?xf32, offset: ?, strides: [1]>,
%y: memref<?xf32, offset: ?, strides: [1]>) {
linalg.matvec
ins(%A, %x: memref<?x?xf32, offset: ?, strides: [?, 1]>,
memref<?xf32, offset: ?, strides: [1]>)
outs(%y: memref<?xf32, offset: ?, strides: [1]>)
return
}
// CHECK-LABEL: func @matvec
// CHECK-DAG: %[[c0:.*]] = constant 0 : index
// CHECK-DAG: %[[c5:.*]] = constant 5 : index
// CHECK-DAG: %[[c6:.*]] = constant 6 : index
// CHECK: scf.parallel {{.*}} step (%[[c5]])
// CHECK: scf.for {{.*}} step %[[c6]]
// CHECK: linalg.matvec
// CHECK: ins({{.*}}, {{.*}}: memref<?x?xf32, #[[$STRIDED_2D]]>, memref<?xf32, #[[$STRIDED_1D]]>)
// CHECK: outs({{.*}}: memref<?xf32, #[[$STRIDED_1D]]>)
func @matmul(%A: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%B: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%C: memref<?x?xf32, offset: ?, strides: [?, 1]>) {
linalg.matmul { __internal_linalg_transform__ = "MEM" }
ins(%A, %B: memref<?x?xf32, offset: ?, strides: [?, 1]>,
memref<?x?xf32, offset: ?, strides: [?, 1]>)
outs(%C: memref<?x?xf32, offset: ?, strides: [?, 1]>)
return
}
// CHECK-LABEL: func @matmul
// CHECK-DAG: %[[c0:.*]] = constant 0 : index
// CHECK-DAG: %[[c2:.*]] = constant 2 : index
// CHECK-DAG: %[[c3:.*]] = constant 3 : index
// CHECK-DAG: %[[c4:.*]] = constant 4 : index
// CHECK-DAG: %[[c20:.*]] = constant 20 : index
// CHECK-DAG: %[[c30:.*]] = constant 30 : index
// CHECK-DAG: %[[c40:.*]] = constant 40 : index
// CHECK-DAG: %[[c200:.*]] = constant 200 : index
// CHECK-DAG: %[[c300:.*]] = constant 300 : index
// CHECK-DAG: %[[c400:.*]] = constant 400 : index
// CHECK-DAG: %[[c2000:.*]] = constant 2000 : index
// CHECK-DAG: %[[c3000:.*]] = constant 3000 : index
// CHECK-DAG: %[[c4000:.*]] = constant 4000 : index
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c2000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c3000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c4000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c200]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c300]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c400]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c20]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c30]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c40]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c2]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c3]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c4]] {
// CHECK: linalg.matmul
// CHECK: ins({{.*}}, {{.*}}: memref<?x?xf32, #[[$STRIDED_2D]]>, memref<?x?xf32, #[[$STRIDED_2D]]>)
// CHECK: outs({{.*}}: memref<?x?xf32, #[[$STRIDED_2D]]>)
#matmul_trait = {
args_in = 2,
args_out = 1,
indexing_maps = [
affine_map<(m, n, k) -> (m, k)>,
affine_map<(m, n, k) -> (k, n)>,
affine_map<(m, n, k) -> (m, n)>
],
iterator_types = ["parallel", "parallel", "reduction"],
__internal_linalg_transform__ = "VECTORIZE"
}
func @vectorization_test(%A: memref<8x16xf32>, %B: memref<16x32xf32>,
%C: memref<8x32xf32>) {
linalg.generic #matmul_trait
ins(%A, %B : memref<8x16xf32>, memref<16x32xf32>)
outs(%C : memref<8x32xf32>) {
^bb(%a: f32, %b: f32, %c: f32) :
%d = mulf %a, %b: f32
%e = addf %c, %d: f32
linalg.yield %e : f32
}
return
}
// CHECK-LABEL: func @vectorization_test
// CHECK: vector.transfer_read %{{.*}} : memref<8x16xf32>, vector<8x16xf32>
// CHECK: vector.transfer_read %{{.*}} : memref<16x32xf32>, vector<16x32xf32>
// CHECK: vector.transfer_read %{{.*}} : memref<8x32xf32>, vector<8x32xf32>
// CHECK: vector.contract {indexing_maps = [#[[$mk]], #[[$kn]], #[[$mn]]], iterator_types = ["parallel", "parallel", "reduction"]} %{{.*}}, %{{.*}}, %{{.*}} : vector<8x16xf32>, vector<16x32xf32> into vector<8x32xf32>
// CHECK: vector.transfer_write %{{.*}}, %{{.*}} : vector<8x32xf32>, memref<8x32xf32>
func @vectorization_test_integer(%A: memref<8x16xi32>, %B: memref<16x32xi32>,
%C: memref<8x32xi32>) {
linalg.generic #matmul_trait
ins(%A, %B : memref<8x16xi32>, memref<16x32xi32>)
outs(%C : memref<8x32xi32>) {
^bb(%a: i32, %b: i32, %c: i32) :
%d = muli %a, %b: i32
%e = addi %c, %d: i32
linalg.yield %e : i32
}
return
}
// CHECK-LABEL: func @vectorization_test_integer
// CHECK: vector.transfer_read %{{.*}} : memref<8x16xi32>, vector<8x16xi32>
// CHECK: vector.transfer_read %{{.*}} : memref<16x32xi32>, vector<16x32xi32>
// CHECK: vector.transfer_read %{{.*}} : memref<8x32xi32>, vector<8x32xi32>
// CHECK: vector.contract {indexing_maps = [#[[$mk]], #[[$kn]], #[[$mn]]], iterator_types = ["parallel", "parallel", "reduction"]} %{{.*}}, %{{.*}}, %{{.*}} : vector<8x16xi32>, vector<16x32xi32> into vector<8x32xi32>
// CHECK: vector.transfer_write %{{.*}}, %{{.*}} : vector<8x32xi32>, memref<8x32xi32>
func @vectorization_test_2(%A: memref<8x16xf32>, %B: memref<16x32xf32>,
%C: memref<8x32xf32>) {
linalg.matmul { __internal_linalg_transform__ = "VECTORIZE"}
ins(%A, %B: memref<8x16xf32>, memref<16x32xf32>)
outs(%C: memref<8x32xf32>)
return
}
// CHECK-LABEL: func @vectorization_test_2
// CHECK: vector.contract {{.*}} :
// vector<8x16xf32>, vector<16x32xf32> into vector<8x32xf32>
func @test_vectorize_fill(%A : memref<8x16xf32>, %arg0 : f32) {
linalg.fill(%A, %arg0) { __internal_linalg_transform__ = "VECTORIZE"} : memref<8x16xf32>, f32
return
}
// CHECK-LABEL: func @test_vectorize_fill
// CHECK: %[[V:.*]] = vector.broadcast {{.*}} : f32 to vector<8x16xf32>
// CHECK: vector.transfer_write %[[V]], {{.*}} : vector<8x16xf32>, memref<8x16xf32>
func @test_vectorize_fill_scalar(%A : memref<f32>, %arg0 : f32) {
linalg.fill(%A, %arg0) { __internal_linalg_transform__ = "VECTORIZE"} : memref<f32>, f32
return
}
// CHECK-LABEL: func @test_vectorize_fill
// CHECK-SAME: (%[[M:.*]]: memref<f32>, %[[V:.*]]: f32)
// CHECK: store %[[V]], %[[M]][] : memref<f32>
func @test_vectorize_copy(%A : memref<8x16xf32>, %B : memref<8x16xf32>) {
linalg.copy(%A, %B) { __internal_linalg_transform__ = "VECTORIZE"} : memref<8x16xf32>, memref<8x16xf32>
return
}
// CHECK-LABEL: func @test_vectorize_copy
// CHECK: %[[V:.*]] = vector.transfer_read {{.*}} : memref<8x16xf32>, vector<8x16xf32>
// CHECK: vector.transfer_write %[[V]], {{.*}} : vector<8x16xf32>, memref<8x16xf32>
func @test_vectorize_copy_scalar(%A : memref<f32>, %B : memref<f32>) {
linalg.copy(%A, %B) { __internal_linalg_transform__ = "VECTORIZE"} : memref<f32>, memref<f32>
return
}
// CHECK-LABEL: func @test_vectorize_copy_scalar
// CHECK: %[[V:.*]] = load {{.*}} : memref<f32>
// CHECK: store %[[V]], {{.*}} : memref<f32>
#matmul_accesses = [
affine_map<(m, n, k) -> (m, k)>,
affine_map<(m, n, k) -> (k, n)>,
affine_map<(m, n, k) -> (m, n)>
]
#generic_matmul_trait = {
args_in = 2,
args_out = 1,
indexing_maps = #matmul_accesses,
library_call = "linalg_matmul",
iterator_types = ["parallel", "parallel", "reduction"]
}
func @permute_generic(%A: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%B: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%C: memref<?x?xf32, offset: ?, strides: [?, 1]>) {
linalg.generic #generic_matmul_trait
ins(%A, %B : memref<?x?xf32, offset: ?, strides: [?, 1]>,
memref<?x?xf32, offset: ?, strides: [?, 1]>)
outs(%C : memref<?x?xf32, offset: ?, strides: [?, 1]>) {
^bb(%a: f32, %b: f32, %c: f32):
%d = mulf %a, %b: f32
%e = addf %c, %d: f32
linalg.yield %e: f32
}
return
}
// CHECK-LABEL: func @permute_generic
// CHECK: linalg.generic {
// CHECK-SAME: indexing_maps = [#[[$kn]], #[[$nm]], #[[$km]]],
// CHECK-SAME: iterator_types = ["parallel", "reduction", "parallel"],
// CHECK-SAME: library_call = "linalg_matmul"}
// CHECK: memref<?x?xf32, #[[$STRIDED_2D_u_1]]>,
// CHECK-SAME: memref<?x?xf32, #[[$STRIDED_2D_u_1]]>
// CHECK-SAME: memref<?x?xf32, #[[$STRIDED_2D_u_1]]>
#indexed_matmul_trait = {
args_in = 2,
args_out = 1,
indexing_maps = #matmul_accesses,
library_call = "linalg_matmul_indexed",
iterator_types = ["parallel", "parallel", "reduction"]
}
func @permute_generic_indexed(
%A: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%B: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%C: memref<?x?xf32, offset: ?, strides: [?, 1]>) {
linalg.indexed_generic #indexed_matmul_trait
ins(%A, %B : memref<?x?xf32, offset: ?, strides: [?, 1]>,
memref<?x?xf32, offset: ?, strides: [?, 1]>)
outs(%C : memref<?x?xf32, offset: ?, strides: [?, 1]>) {
^bb(%i: index, %j: index, %k: index, %a: f32, %b: f32, %c: f32):
%d = mulf %a, %b: f32
%e = addf %c, %d: f32
linalg.yield %e: f32
}
return
}
// CHECK-LABEL: func @permute_generic_indexed
// CHECK: linalg.indexed_generic {
// CHECK-SAME: indexing_maps = [#[[$kn]], #[[$nm]], #[[$km]]],
// CHECK-SAME: iterator_types = ["parallel", "reduction", "parallel"],
// CHECK-SAME: library_call = "linalg_matmul_indexed"}
// CHECK: memref<?x?xf32, #[[$STRIDED_2D_u_1]]>,
// CHECK-SAME: memref<?x?xf32, #[[$STRIDED_2D_u_1]]>
// CHECK-SAME: memref<?x?xf32, #[[$STRIDED_2D_u_1]]>
func @matvec_perm(%A: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%x: memref<?xf32, offset: ?, strides: [1]>,
%y: memref<?xf32, offset: ?, strides: [1]>) {
linalg.matvec {__internal_linalg_transform__ = "__with_perm__"}
ins(%A, %x: memref<?x?xf32, offset: ?, strides: [?, 1]>,
memref<?xf32, offset: ?, strides: [1]>)
outs(%y: memref<?xf32, offset: ?, strides: [1]>)
return
}
// CHECK-LABEL: func @matvec_perm
// CHECK-DAG: %[[c0:.*]] = constant 0 : index
// CHECK-DAG: %[[c5:.*]] = constant 5 : index
// CHECK-DAG: %[[c6:.*]] = constant 6 : index
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c6]]
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c5]]
// CHECK: linalg.matvec
// CHECK: ins({{.*}}, {{.*}}: memref<?x?xf32, #[[$STRIDED_2D]]>, memref<?xf32, #[[$STRIDED_1D]]>)
// CHECK: outs({{.*}}: memref<?xf32, #[[$STRIDED_1D]]>)
func @matmul_perm(%A: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%B: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%C: memref<?x?xf32, offset: ?, strides: [?, 1]>) {
linalg.matmul {__internal_linalg_transform__ = "__with_perm__"}
ins(%A, %B: memref<?x?xf32, offset: ?, strides: [?, 1]>,
memref<?x?xf32, offset: ?, strides: [?, 1]>)
outs(%C : memref<?x?xf32, offset: ?, strides: [?, 1]>)
return
}
// CHECK-LABEL: func @matmul_perm
// CHECK-DAG: %[[c0:.*]] = constant 0 : index
// CHECK-DAG: %[[c20:.*]] = constant 20 : index
// CHECK-DAG: %[[c30:.*]] = constant 30 : index
// CHECK-DAG: %[[c40:.*]] = constant 40 : index
// CHECK-DAG: %[[c200:.*]] = constant 200 : index
// CHECK-DAG: %[[c300:.*]] = constant 300 : index
// CHECK-DAG: %[[c400:.*]] = constant 400 : index
// CHECK-DAG: %[[c2000:.*]] = constant 2000 : index
// CHECK-DAG: %[[c3000:.*]] = constant 3000 : index
// CHECK-DAG: %[[c4000:.*]] = constant 4000 : index
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c3000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c4000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c2000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c300]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c200]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c400]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c20]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c30]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c40]] {
// CHECK: linalg.matmul
// CHECK: ins({{.*}}, {{.*}}: memref<?x?xf32, #[[$STRIDED_2D]]>, memref<?x?xf32, #[[$STRIDED_2D]]>)
// CHECK: outs({{.*}}: memref<?x?xf32, #[[$STRIDED_2D]]>)
func @promote_subview_matmul(%arg0: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%arg1: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%arg2: memref<?x?xf32, offset: ?, strides: [?, 1]>) {
%c2000 = constant 2000 : index
%c3000 = constant 3000 : index
%c4000 = constant 4000 : index
%c0 = constant 0 : index
%c1 = constant 1 : index
%0 = dim %arg0, %c0 : memref<?x?xf32, offset: ?, strides: [?, 1]>
%1 = dim %arg0, %c1 : memref<?x?xf32, offset: ?, strides: [?, 1]>
%2 = dim %arg1, %c1 : memref<?x?xf32, offset: ?, strides: [?, 1]>
scf.for %arg3 = %c0 to %0 step %c2000 {
scf.for %arg4 = %c0 to %2 step %c3000 {
scf.for %arg5 = %c0 to %1 step %c4000 {
%3 = subview %arg0[%arg3, %arg5][%c2000, %c4000][%c1, %c1] :
memref<?x?xf32, offset: ?, strides: [?, 1]> to memref<?x?xf32, offset: ?, strides: [?, ?]>
%4 = subview %arg1[%arg5, %arg4][%c4000, %c3000][%c1, %c1] :
memref<?x?xf32, offset: ?, strides: [?, 1]> to memref<?x?xf32, offset: ?, strides: [?, ?]>
%5 = subview %arg2[%arg3, %arg4][%c2000, %c3000][%c1, %c1] :
memref<?x?xf32, offset: ?, strides: [?, 1]> to memref<?x?xf32, offset: ?, strides: [?, ?]>
linalg.matmul {__internal_linalg_transform__ = "_promote_views_"}
ins(%3, %4: memref<?x?xf32, offset: ?, strides: [?, ?]>,
memref<?x?xf32, offset: ?, strides: [?, ?]>)
outs(%5: memref<?x?xf32, offset: ?, strides: [?, ?]>)
}
}
}
return
}
// CHECK-LABEL: func @promote_subview_matmul
// CHECK-DAG: %[[c0:.*]] = constant 0 : index
// CHECK-DAG: %[[c2000:.*]] = constant 2000 : index
// CHECK-DAG: %[[c3000:.*]] = constant 3000 : index
// CHECK-DAG: %[[c4000:.*]] = constant 4000 : index
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c2000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c3000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c4000]] {
// CHECK: %[[s0:.*]] = subview {{%.*}}[{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32, #map{{.*}}> to memref<?x?xf32, #map{{.*}}>
// CHECK: %[[s1:.*]] = subview {{%.*}}[{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32, #map{{.*}}> to memref<?x?xf32, #map{{.*}}>
// CHECK: %[[s2:.*]] = subview {{%.*}}[{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32, #map{{.*}}> to memref<?x?xf32, #map{{.*}}>
// CHECK: %[[a0:.*]] = alloc({{%.*}}) : memref<?xi8>
// CHECK: %[[v0:.*]] = std.view %[[a0]][{{.*}}][{{%.*}}, {{%.*}}] : memref<?xi8> to memref<?x?xf32>
// CHECK: %[[l0:.*]] = subview %[[v0]][{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32> to memref<?x?xf32, #[[$STRIDED_2D]]>
// CHECK: %[[a1:.*]] = alloc({{%.*}}) : memref<?xi8>
// CHECK: %[[v1:.*]] = std.view %[[a1]][{{.*}}][{{%.*}}, {{%.*}}] : memref<?xi8> to memref<?x?xf32>
// CHECK: %[[l1:.*]] = subview %[[v1]][{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32> to memref<?x?xf32, #[[$STRIDED_2D]]>
// CHECK: %[[a2:.*]] = alloc({{%.*}}) : memref<?xi8>
// CHECK: %[[v2:.*]] = std.view %[[a2]][{{.*}}][{{%.*}}, {{%.*}}] : memref<?xi8> to memref<?x?xf32>
// CHECK: %[[l2:.*]] = subview %[[v2]][{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32> to memref<?x?xf32, #[[$STRIDED_2D]]>
// CHECK: linalg.copy(%[[s0]], %[[l0]]) : memref<?x?xf32, #map{{.*}}>, memref<?x?xf32, #map{{.*}}>
// CHECK: linalg.copy(%[[s1]], %[[l1]]) : memref<?x?xf32, #map{{.*}}>, memref<?x?xf32, #map{{.*}}>
// CHECK: linalg.copy(%[[s2]], %[[l2]]) : memref<?x?xf32, #map{{.*}}>, memref<?x?xf32, #map{{.*}}>
// CHECK: linalg.matmul
// CHECK-SAME: ins(%[[v0]], %[[v1]] : memref<?x?xf32>, memref<?x?xf32>)
// CHECK-SAME: outs(%[[v2]] : memref<?x?xf32>)
func @promote_first_subview_matmul(%arg0: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%arg1: memref<?x?xf32, offset: ?, strides: [?, 1]>,
%arg2: memref<?x?xf32, offset: ?, strides: [?, 1]>) {
%c2000 = constant 2000 : index
%c3000 = constant 3000 : index
%c4000 = constant 4000 : index
%c0 = constant 0 : index
%c1 = constant 1 : index
%0 = dim %arg0, %c0 : memref<?x?xf32, offset: ?, strides: [?, 1]>
%1 = dim %arg0, %c1 : memref<?x?xf32, offset: ?, strides: [?, 1]>
%2 = dim %arg1, %c1 : memref<?x?xf32, offset: ?, strides: [?, 1]>
scf.for %arg3 = %c0 to %0 step %c2000 {
scf.for %arg4 = %c0 to %2 step %c3000 {
scf.for %arg5 = %c0 to %1 step %c4000 {
%3 = std.subview %arg0[%arg3, %arg5][%c2000, %c4000][%c1, %c1] :
memref<?x?xf32, offset: ?, strides: [?, 1]> to memref<?x?xf32, offset: ?, strides: [?, ?]>
%4 = std.subview %arg1[%arg5, %arg4][%c4000, %c3000][%c1, %c1] :
memref<?x?xf32, offset: ?, strides: [?, 1]> to memref<?x?xf32, offset: ?, strides: [?, ?]>
%5 = std.subview %arg2[%arg3, %arg4][%c2000, %c3000][%c1, %c1] :
memref<?x?xf32, offset: ?, strides: [?, 1]> to memref<?x?xf32, offset: ?, strides: [?, ?]>
linalg.matmul {__internal_linalg_transform__ = "_promote_first_view_"}
ins(%3, %4: memref<?x?xf32, offset: ?, strides: [?, ?]>,
memref<?x?xf32, offset: ?, strides: [?, ?]>)
outs(%5: memref<?x?xf32, offset: ?, strides: [?, ?]>)
}
}
}
return
}
// CHECK-LABEL: func @promote_first_subview_matmul
// CHECK-DAG: %[[c0:.*]] = constant 0 : index
// CHECK-DAG: %[[c2000:.*]] = constant 2000 : index
// CHECK-DAG: %[[c3000:.*]] = constant 3000 : index
// CHECK-DAG: %[[c4000:.*]] = constant 4000 : index
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c2000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c3000]] {
// CHECK: scf.for {{.*}} = %[[c0]] to {{.*}} step %[[c4000]] {
// CHECK: %[[s0:.*]] = subview {{%.*}}[{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32, #map{{.*}}> to memref<?x?xf32, #map{{.*}}>
// CHECK: %[[s1:.*]] = subview {{%.*}}[{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32, #map{{.*}}> to memref<?x?xf32, #map{{.*}}>
// CHECK: %[[s2:.*]] = subview {{%.*}}[{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32, #map{{.*}}> to memref<?x?xf32, #map{{.*}}>
// CHECK: %[[a0:.*]] = alloc({{%.*}}) : memref<?xi8>
// CHECK: %[[v0:.*]] = std.view %[[a0]][{{.*}}][{{%.*}}, {{%.*}}] : memref<?xi8> to memref<?x?xf32>
// CHECK: %[[l0:.*]] = subview %[[v0]][{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32> to memref<?x?xf32, #[[$STRIDED_2D]]>
// CHECK-NOT: %[[a1:.*]] = alloc({{%.*}}) : memref<?xi8>
// CHECK-NOT: %[[v1:.*]] = std.view %[[a1]][{{.*}}][{{%.*}}, {{%.*}}] : memref<?xi8> to memref<?x?xf32>
// CHECK-NOT: %[[l0:.*]] = subview %[[v1]][{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32> to memref<?x?xf32, #[[$STRIDED_2D]]>
// CHECK-NOT: %[[a2:.*]] = alloc({{%.*}}) : memref<?xi8>
// CHECK-NOT: %[[v2:.*]] = std.view %[[a2]][{{.*}}][{{%.*}}, {{%.*}}] : memref<?xi8> to memref<?x?xf32>
// CHECK-NOT: %[[l0:.*]] = subview %[[v2]][{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32> to memref<?x?xf32, #[[$STRIDED_2D]]>
// CHECK: linalg.copy(%[[s0]], %[[l0]]) : memref<?x?xf32, #map{{.*}}>, memref<?x?xf32, #map{{.*}}>
// CHECK-NOT: linalg.copy(%[[s1]], %[[l1]]) : memref<?x?xf32, #map{{.*}}>, memref<?x?xf32, #map{{.*}}>
// CHECK-NOT: linalg.copy(%[[s2]], %[[l2]]) : memref<?x?xf32, #map{{.*}}>, memref<?x?xf32, #map{{.*}}>^
// CHECK: linalg.matmul
// CHECK-SAME: ins(%[[v0]], %[[s1]] : memref<?x?xf32>, memref<?x?xf32, #[[$STRIDED_2D]]>)
// CHECK-SAME: outs(%[[s2]] : memref<?x?xf32, #[[$STRIDED_2D]]>)
func @aligned_promote_fill(%arg0: memref<?x?xf32, offset: ?, strides: [?, 1]>) {
%c2000 = constant 2000 : index
%c4000 = constant 4000 : index
%c0 = constant 0 : index
%c1 = constant 1 : index
%cf = constant 1.0 : f32
%3 = std.subview %arg0[%c0, %c0][%c2000, %c4000][%c1, %c1] :
memref<?x?xf32, offset: ?, strides: [?, 1]> to memref<?x?xf32, offset: ?, strides: [?, ?]>
linalg.fill(%3, %cf) { __internal_linalg_transform__ = "_promote_views_aligned_"}
: memref<?x?xf32, offset: ?, strides: [?, ?]>, f32
return
}
// CHECK-LABEL: func @aligned_promote_fill
// CHECK: %[[cf:.*]] = constant {{.*}} : f32
// CHECK: %[[s0:.*]] = subview {{%.*}}[{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32, #map{{.*}}> to memref<?x?xf32, #map{{.*}}>
// CHECK: %[[a0:.*]] = alloc({{%.*}}) {alignment = 32 : i64} : memref<?xi8>
// CHECK: %[[v0:.*]] = std.view %[[a0]][{{.*}}][{{%.*}}, {{%.*}}] : memref<?xi8> to memref<?x?xf32>
// CHECK: %[[l0:.*]] = subview %[[v0]][{{%.*}}, {{%.*}}] [{{%.*}}, {{%.*}}] : memref<?x?xf32> to memref<?x?xf32, #[[$STRIDED_2D]]>
// CHECK: linalg.fill(%[[v0]], {{%.*}}) : memref<?x?xf32>, f32
// CHECK: linalg.copy(%[[s0]], %[[l0]]) : memref<?x?xf32, #map{{.*}}>, memref<?x?xf32, #map{{.*}}>
// CHECK: linalg.fill(%[[v0]], %[[cf]]) : memref<?x?xf32>, f32
func @tile_permute_parallel_loop(%arg0: memref<?x?xf32>,
%arg1: memref<?x?xf32>,
%arg2: memref<?x?xf32>) {
linalg.matmul {__internal_linalg_transform__ = "par__with_perm__"}
ins(%arg0, %arg1: memref<?x?xf32>, memref<?x?xf32>)
outs(%arg2: memref<?x?xf32>)
return
}
// CHECK-LABEL: func @tile_permute_parallel_loop
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]+]]: memref<?x?xf32>
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9_]+]]: memref<?x?xf32>
// CHECK-SAME: %[[ARG2:[a-zA-Z0-9_]+]]: memref<?x?xf32>
// CHECK-DAG: %[[C16:.*]] = constant 16 : index
// CHECK-DAG: %[[C8:.*]] = constant 8 : index
// CHECK-DAG: %[[C4:.*]] = constant 4 : index
// CHECK-DAG: %[[C0:.*]] = constant 0 : index
// CHECK-DAG: %[[D0:.*]] = dim %[[ARG0]], %c0
// CHECK-DAG: %[[D1:.*]] = dim %[[ARG0]], %c1
// CHECK-DAG: %[[D2:.*]] = dim %[[ARG1]], %c1
// CHECK: scf.parallel (%{{.*}}) = (%[[C0]]) to (%[[D2]]) step (%[[C8]])
// CHECK: scf.for %{{.*}} = %[[C0]] to %[[D1]] step %[[C4]]
// CHECK: scf.parallel (%{{.*}}) = (%[[C0]]) to (%[[D0]]) step (%[[C16]])
|
