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
|
// RUN: mlir-opt --split-input-file --test-transform-dialect-interpreter %s | FileCheck %s
func.func @matmul_split(%A : tensor<16x256xf32>, %B: tensor<256x32xf32>, %C: tensor<16x32xf32>) -> tensor<16x32xf32> {
%0 = linalg.matmul ins(%A, %B: tensor<16x256xf32>, tensor<256x32xf32>)
outs(%C: tensor<16x32xf32>) -> tensor<16x32xf32>
return %0: tensor<16x32xf32>
}
// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d2, d3)>
// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d2, d3, d1)>
// CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2)>
// CHECK-DAG: #[[$MAP3:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
// CHECK-DAG: #[[$MAP4:.*]] = affine_map<(d0, d1, d2) -> (d0, d1)>
// CHECK-LABEL: @matmul_split
// CHECK-DAG: %[[ID:.*]] = arith.constant 0.000000e+00 : f32
// CHECK-DAG: %[[I1:.*]] = tensor.expand_shape %{{.*}}[0], [1, 2]] : tensor<16x256xf32> into tensor<16x4x64xf32>
// CHECK-DAG: %[[I2:.*]] = tensor.expand_shape %{{.*}}[0, 1], [2]] : tensor<256x32xf32> into tensor<4x64x32xf32>
// CHECK-DAG: %[[INI:.*]] = tensor.empty() : tensor<16x32x4xf32>
// CHECK: %[[F:.*]] = linalg.fill ins(%[[ID]] : f32) outs(%[[INI]] : tensor<16x32x4xf32>) -> tensor<16x32x4xf32>
// CHECK: %[[G:.*]] = linalg.generic {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]]]
// CHECK-SAME: , iterator_types = ["parallel", "parallel", "parallel", "reduction"]}
// CHECK-SAME: ins(%[[I1]], %[[I2]] : tensor<16x4x64xf32>, tensor<4x64x32xf32>) outs(%[[F]] : tensor<16x32x4xf32>) {
// CHECK: arith.mulf
// CHECK: arith.addf
// CHECK: linalg.yield
// CHECK: } -> tensor<16x32x4xf32>
// CHECK: %[[R:.*]] = linalg.generic {indexing_maps = [#[[$MAP3]], #[[$MAP4]]],
// CHECK-SAME: iterator_types = ["parallel", "parallel", "reduction"]} ins(%[[G]] : tensor<16x32x4xf32>) outs(%{{.*}} : tensor<16x32xf32>) {
// CHECK: arith.addf
// CHECK: linalg.yield %{{.*}} : f32
// CHECK: } -> tensor<16x32xf32>
// CHECK: return %[[R]] : tensor<16x32xf32>
transform.sequence failures(propagate) {
^bb1(%arg1: !pdl.operation):
%0 = transform.structured.match ops{["linalg.matmul"]} in %arg1 : (!pdl.operation) -> !pdl.operation
%1:4 = transform.structured.split_reduction %0 { split_factor = 4, insert_split_dimension = 2}
}
// -----
func.func @generic_split_1d(%arg0: tensor<32xf32>, %arg1: tensor<f32>, %out: tensor<f32>) -> tensor<f32> {
%red = linalg.generic {indexing_maps = [affine_map<(d0) -> (d0)>,
affine_map<(d0) -> ()>,
affine_map<(d0) -> ()>],
iterator_types = ["reduction"]}
ins(%arg0, %arg1 : tensor<32xf32>, tensor<f32>)
outs(%out : tensor<f32>) {
^bb0(%arg7: f32, %arg8: f32, %arg9: f32):
%40 = arith.subf %arg7, %arg8 : f32
%41 = math.exp %40 : f32
%42 = arith.mulf %41, %arg9 : f32
linalg.yield %42 : f32
} -> tensor<f32>
return %red : tensor<f32>
}
// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, d1)>
// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1) -> ()>
// CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0, d1) -> (d0)>
// CHECK-DAG: #[[$MAP3:.*]] = affine_map<(d0) -> (d0)>
// CHECK-DAG: #[[$MAP4:.*]] = affine_map<(d0) -> ()>
//CHECK-LABEL: @generic_split_1d
// CHECK-DAG: %[[ID:.*]] = arith.constant 1.000000e+00 : f32
// CHECK-DAG: %[[I1:.*]] = tensor.expand_shape %{{.*}}[0, 1]] : tensor<32xf32> into tensor<4x8xf32>
// CHECK-DAG: %[[INI:.*]] = tensor.empty() : tensor<4xf32>
// CHECK: %[[F:.*]] = linalg.fill ins(%[[ID]] : f32) outs(%[[INI]] : tensor<4xf32>) -> tensor<4xf32>
// CHECK: %[[G:.*]] = linalg.generic
// CHECK: {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]]],
// CHECK: iterator_types = ["parallel", "reduction"]} ins(%[[I1]], %{{.*}} : tensor<4x8xf32>, tensor<f32>) outs(%[[F]] : tensor<4xf32>) {
// CHECK: arith.subf
// CHECK: math.exp
// CHECK: arith.mulf
// CHECK: linalg.yield
// CHECK: } -> tensor<4xf32>
// CHECK: %[[R:.*]] = linalg.generic {indexing_maps = [#[[$MAP3]], #[[$MAP4]]], iterator_types = ["reduction"]} ins(%[[G]] : tensor<4xf32>) outs(%{{.*}} : tensor<f32>) {
// CHECK: arith.mulf
// CHECK: linalg.yield
// CHECK: } -> tensor<f32>
// CHECK: return %[[R]] : tensor<f32>
transform.sequence failures(propagate) {
^bb1(%arg1: !pdl.operation):
%0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!pdl.operation) -> !pdl.operation
%1:4 = transform.structured.split_reduction %0 { split_factor = 4, insert_split_dimension = 0}
}
// -----
func.func @generic_split_3d(%input: tensor<32x2xf32>, %input_2: tensor<5x32xf32>, %output: tensor<5x2xf32>)
-> tensor<5x2xf32>
{
%0 = linalg.generic {
indexing_maps = [
affine_map<(d0, d1, d2) -> (d1, d0)>,
affine_map<(d0, d1, d2) -> (d2, d1)>,
affine_map<(d0, d1, d2) -> (d2, d0)>
],
iterator_types = ["parallel", "reduction", "parallel"]
} ins(%input, %input_2 : tensor<32x2xf32>, tensor<5x32xf32>) outs(%output : tensor<5x2xf32>) {
^bb0(%arg0: f32, %arg1: f32, %arg2: f32):
%3 = arith.addf %arg0, %arg1 : f32
%4 = arith.maxf %3, %arg2 : f32
linalg.yield %4 : f32
} -> tensor<5x2xf32>
return %0 : tensor<5x2xf32>
}
// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d2, d1, d0)>
// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d3, d2, d1)>
// CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0, d1, d2, d3) -> (d3, d0, d2)>
// CHECK-DAG: #[[$MAP3:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
// CHECK-DAG: #[[$MAP4:.*]] = affine_map<(d0, d1, d2) -> (d0, d1)>
// CHECK-LABEL: func @generic_split_3d
// CHECK-DAG: %[[ID:.*]] = arith.constant 0xFF800000 : f32
// CHECK-DAG: %[[I1:.*]] = tensor.expand_shape %{{.*}}[0, 1], [2]] : tensor<32x2xf32> into tensor<4x8x2xf32>
// CHECK-DAG: %[[I2:.*]] = tensor.expand_shape %{{.*}}[0], [1, 2]] : tensor<5x32xf32> into tensor<5x4x8xf32>
// CHECK-DAG: %[[INI:.*]] = tensor.empty() : tensor<5x2x4xf32>
// CHECK: %[[F:.*]] = linalg.fill ins(%[[ID]] : f32) outs(%[[INI]] : tensor<5x2x4xf32>) -> tensor<5x2x4xf32>
// CHECK: %[[G:.*]] = linalg.generic {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]]], iterator_types = ["parallel", "reduction", "parallel", "parallel"]}
// CHECK-SAME: ins(%[[I1]], %[[I2]] : tensor<4x8x2xf32>, tensor<5x4x8xf32>) outs(%[[F]] : tensor<5x2x4xf32>) {
// CHECK: arith.addf
// CHECK: arith.maxf
// CHECK: linalg.yield
// CHECK: } -> tensor<5x2x4xf32>
// CHECK: %[[R:.*]] = linalg.generic {indexing_maps = [#[[$MAP3]], #[[$MAP4]]], iterator_types = ["parallel", "parallel", "reduction"]}
// CHECK-SAME: ins(%[[G]] : tensor<5x2x4xf32>) outs(%{{.*}} : tensor<5x2xf32>) {
// CHECK: arith.maxf
// CHECK: linalg.yield
// CHECK: } -> tensor<5x2xf32>
// CHECK: return %[[R]] : tensor<5x2xf32>
transform.sequence failures(propagate) {
^bb1(%arg1: !pdl.operation):
%0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!pdl.operation) -> !pdl.operation
%1:4 = transform.structured.split_reduction %0 { split_factor = 4, insert_split_dimension = 2}
}
// -----
func.func @matmul_split(%A : tensor<16x256xf32>, %B: tensor<256x32xf32>, %C: tensor<16x32xf32>) -> tensor<16x32xf32> {
%0 = linalg.matmul ins(%A, %B: tensor<16x256xf32>, tensor<256x32xf32>)
outs(%C: tensor<16x32xf32>) -> tensor<16x32xf32>
return %0: tensor<16x32xf32>
}
// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d2, d3)>
// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d2, d3, d1)>
// CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d3)>
// CHECK-DAG: #[[$MAP3:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
// CHECK-DAG: #[[$MAP4:.*]] = affine_map<(d0, d1, d2) -> (d0, d1)>
// CHECK-LABEL: @matmul_split
// CHECK-DAG: %[[ID:.*]] = arith.constant 0.000000e+00 : f32
// CHECK-DAG: %[[I1:.*]] = tensor.expand_shape %{{.*}}[0], [1, 2]] : tensor<16x256xf32> into tensor<16x64x4xf32>
// CHECK-DAG: %[[I2:.*]] = tensor.expand_shape %{{.*}}[0, 1], [2]] : tensor<256x32xf32> into tensor<64x4x32xf32>
// CHECK-DAG: %[[INI:.*]] = tensor.empty() : tensor<16x32x4xf32>
// CHECK: %[[F:.*]] = linalg.fill ins(%[[ID]] : f32) outs(%[[INI]] : tensor<16x32x4xf32>) -> tensor<16x32x4xf32>
// CHECK: %[[G:.*]] = linalg.generic {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]]]
// CHECK-SAME: , iterator_types = ["parallel", "parallel", "reduction", "parallel"]}
// CHECK-SAME: ins(%[[I1]], %[[I2]] : tensor<16x64x4xf32>, tensor<64x4x32xf32>) outs(%[[F]] : tensor<16x32x4xf32>) {
// CHECK: arith.mulf
// CHECK: arith.addf
// CHECK: linalg.yield
// CHECK: } -> tensor<16x32x4xf32>
// CHECK: %[[R:.*]] = linalg.generic {indexing_maps = [#[[$MAP3]], #[[$MAP4]]],
// CHECK-SAME: iterator_types = ["parallel", "parallel", "reduction"]} ins(%[[G]] : tensor<16x32x4xf32>) outs(%{{.*}} : tensor<16x32xf32>) {
// CHECK: arith.addf
// CHECK: linalg.yield %{{.*}} : f32
// CHECK: } -> tensor<16x32xf32>
// CHECK: return %[[R]] : tensor<16x32xf32>
transform.sequence failures(propagate) {
^bb1(%arg1: !pdl.operation):
%0 = transform.structured.match ops{["linalg.matmul"]} in %arg1 : (!pdl.operation) -> !pdl.operation
%1:4 = transform.structured.split_reduction %0 { split_factor = 4, insert_split_dimension = 2, inner_parallel}
}
// -----
func.func @generic_split_1d(%arg0: tensor<32xf32>, %arg1: tensor<f32>, %out: tensor<f32>) -> tensor<f32> {
%red = linalg.generic {indexing_maps = [affine_map<(d0) -> (d0)>,
affine_map<(d0) -> ()>,
affine_map<(d0) -> ()>],
iterator_types = ["reduction"]}
ins(%arg0, %arg1 : tensor<32xf32>, tensor<f32>)
outs(%out : tensor<f32>) {
^bb0(%arg7: f32, %arg8: f32, %arg9: f32):
%40 = arith.subf %arg7, %arg8 : f32
%41 = math.exp %40 : f32
%42 = arith.mulf %41, %arg9 : f32
linalg.yield %42 : f32
} -> tensor<f32>
return %red : tensor<f32>
}
// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, d1)>
// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1) -> ()>
// CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0, d1) -> (d1)>
// CHECK-DAG: #[[$MAP3:.*]] = affine_map<(d0) -> (d0)>
// CHECK-DAG: #[[$MAP4:.*]] = affine_map<(d0) -> ()>
//CHECK-LABEL: @generic_split_1d
// CHECK-DAG: %[[ID:.*]] = arith.constant 1.000000e+00 : f32
// CHECK-DAG: %[[I1:.*]] = tensor.expand_shape %{{.*}}[0, 1]] : tensor<32xf32> into tensor<8x4xf32>
// CHECK-DAG: %[[INI:.*]] = tensor.empty() : tensor<4xf32>
// CHECK: %[[F:.*]] = linalg.fill ins(%[[ID]] : f32) outs(%[[INI]] : tensor<4xf32>) -> tensor<4xf32>
// CHECK: %[[G:.*]] = linalg.generic
// CHECK: {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]]],
// CHECK: iterator_types = ["reduction", "parallel"]} ins(%[[I1]], %{{.*}} : tensor<8x4xf32>, tensor<f32>) outs(%[[F]] : tensor<4xf32>) {
// CHECK: arith.subf
// CHECK: math.exp
// CHECK: arith.mulf
// CHECK: linalg.yield
// CHECK: } -> tensor<4xf32>
// CHECK: %[[R:.*]] = linalg.generic {indexing_maps = [#[[$MAP3]], #[[$MAP4]]], iterator_types = ["reduction"]} ins(%[[G]] : tensor<4xf32>) outs(%{{.*}} : tensor<f32>) {
// CHECK: arith.mulf
// CHECK: linalg.yield
// CHECK: } -> tensor<f32>
// CHECK: return %[[R]] : tensor<f32>
transform.sequence failures(propagate) {
^bb1(%arg1: !pdl.operation):
%0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!pdl.operation) -> !pdl.operation
%1:4 = transform.structured.split_reduction %0 { split_factor = 4, insert_split_dimension = 0, inner_parallel}
}
// -----
func.func @generic_split_3d(%input: tensor<32x2xf32>, %input_2: tensor<5x32xf32>, %output: tensor<5x2xf32>)
-> tensor<5x2xf32>
{
%0 = linalg.generic {
indexing_maps = [
affine_map<(d0, d1, d2) -> (d1, d0)>,
affine_map<(d0, d1, d2) -> (d2, d1)>,
affine_map<(d0, d1, d2) -> (d2, d0)>
],
iterator_types = ["parallel", "reduction", "parallel"]
} ins(%input, %input_2 : tensor<32x2xf32>, tensor<5x32xf32>) outs(%output : tensor<5x2xf32>) {
^bb0(%arg0: f32, %arg1: f32, %arg2: f32):
%3 = arith.addf %arg0, %arg1 : f32
%4 = arith.minf %3, %arg2 : f32
linalg.yield %4 : f32
} -> tensor<5x2xf32>
return %0 : tensor<5x2xf32>
}
// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d1, d2, d0)>
// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d3, d1, d2)>
// CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0, d1, d2, d3) -> (d3, d0, d2)>
// CHECK-DAG: #[[$MAP3:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
// CHECK-DAG: #[[$MAP4:.*]] = affine_map<(d0, d1, d2) -> (d0, d1)>
// CHECK-LABEL: func @generic_split_3d
// CHECK-DAG: %[[ID:.*]] = arith.constant 0x7F800000 : f32
// CHECK-DAG: %[[I1:.*]] = tensor.expand_shape %{{.*}}[0, 1], [2]] : tensor<32x2xf32> into tensor<8x4x2xf32>
// CHECK-DAG: %[[I2:.*]] = tensor.expand_shape %{{.*}}[0], [1, 2]] : tensor<5x32xf32> into tensor<5x8x4xf32>
// CHECK-DAG: %[[INI:.*]] = tensor.empty() : tensor<5x2x4xf32>
// CHECK: %[[F:.*]] = linalg.fill ins(%[[ID]] : f32) outs(%[[INI]] : tensor<5x2x4xf32>) -> tensor<5x2x4xf32>
// CHECK: %[[G:.*]] = linalg.generic {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP2]]], iterator_types = ["parallel", "reduction", "parallel", "parallel"]}
// CHECK-SAME: ins(%[[I1]], %[[I2]] : tensor<8x4x2xf32>, tensor<5x8x4xf32>) outs(%[[F]] : tensor<5x2x4xf32>) {
// CHECK: arith.addf
// CHECK: arith.minf
// CHECK: linalg.yield
// CHECK: } -> tensor<5x2x4xf32>
// CHECK: %[[R:.*]] = linalg.generic {indexing_maps = [#[[$MAP3]], #[[$MAP4]]], iterator_types = ["parallel", "parallel", "reduction"]}
// CHECK-SAME: ins(%[[G]] : tensor<5x2x4xf32>) outs(%{{.*}} : tensor<5x2xf32>) {
// CHECK: arith.minf
// CHECK: linalg.yield
// CHECK: } -> tensor<5x2xf32>
// CHECK: return %[[R]] : tensor<5x2xf32>
transform.sequence failures(propagate) {
^bb1(%arg1: !pdl.operation):
%0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!pdl.operation) -> !pdl.operation
%1:4 = transform.structured.split_reduction %0 { split_factor = 4, insert_split_dimension = 2, inner_parallel}
}
|