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
|
// DEFINE: %{option} = enable-runtime-library=true
// DEFINE: %{compile} = mlir-opt %s --sparse-compiler=%{option}
// DEFINE: %{run} = mlir-cpu-runner \
// DEFINE: -e entry -entry-point-result=void \
// DEFINE: -shared-libs=%mlir_c_runner_utils | \
// DEFINE: FileCheck %s
//
// RUN: %{compile} | %{run}
//
// Do the same run, but now with direct IR generation.
// REDEFINE: %{option} = enable-runtime-library=false
// RUN: %{compile} | %{run}
//
// Do the same run, but now with direct IR generation and vectorization.
// REDEFINE: %{option} = "enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true"
// RUN: %{compile} | %{run}
// Do the same run, but now with direct IR generation and, if available, VLA
// vectorization.
// REDEFINE: %{option} = "enable-runtime-library=false vl=4 enable-arm-sve=%ENABLE_VLA"
// REDEFINE: %{run} = %lli_host_or_aarch64_cmd \
// REDEFINE: --entry-function=entry_lli \
// REDEFINE: --extra-module=%S/Inputs/main_for_lli.ll \
// REDEFINE: %VLA_ARCH_ATTR_OPTIONS \
// REDEFINE: --dlopen=%mlir_native_utils_lib_dir/libmlir_c_runner_utils%shlibext | \
// REDEFINE: FileCheck %s
// RUN: %{compile} | mlir-translate -mlir-to-llvmir | %{run}
#SparseVector = #sparse_tensor.encoding<{
lvlTypes = ["compressed"]
}>
#SparseMatrix = #sparse_tensor.encoding<{
lvlTypes = ["compressed", "compressed"]
}>
#Sparse3dTensor = #sparse_tensor.encoding<{
lvlTypes = ["compressed", "compressed", "compressed"]
}>
#Sparse4dTensor = #sparse_tensor.encoding<{
lvlTypes = ["compressed", "compressed", "compressed", "compressed"]
}>
//
// Test with various forms of the two most elementary reshape
// operations: collapse.
//
module {
func.func @collapse_dense(%arg0: tensor<3x4xf64>) -> tensor<12xf64> {
%0 = tensor.collapse_shape %arg0 [[0, 1]] : tensor<3x4xf64> into tensor<12xf64>
return %0 : tensor<12xf64>
}
func.func @collapse_from_sparse(%arg0: tensor<3x4xf64, #SparseMatrix>) -> tensor<12xf64> {
%0 = tensor.collapse_shape %arg0 [[0, 1]] : tensor<3x4xf64, #SparseMatrix> into tensor<12xf64>
return %0 : tensor<12xf64>
}
func.func @collapse_to_sparse(%arg0: tensor<3x4xf64>) -> tensor<12xf64, #SparseVector> {
%0 = tensor.collapse_shape %arg0 [[0, 1]] : tensor<3x4xf64> into tensor<12xf64, #SparseVector>
return %0 : tensor<12xf64, #SparseVector>
}
func.func @collapse_sparse2sparse(%arg0: tensor<3x4xf64, #SparseMatrix>) -> tensor<12xf64, #SparseVector> {
%0 = tensor.collapse_shape %arg0 [[0, 1]] : tensor<3x4xf64, #SparseMatrix> into tensor<12xf64, #SparseVector>
return %0 : tensor<12xf64, #SparseVector>
}
func.func @collapse_dense_6x10(%arg0: tensor<2x3x5x2xf64>) -> tensor<6x10xf64> {
%0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<2x3x5x2xf64> into tensor<6x10xf64>
return %0 : tensor<6x10xf64>
}
func.func @collapse_from_sparse_6x10(%arg0: tensor<2x3x5x2xf64, #Sparse4dTensor>) -> tensor<6x10xf64> {
%0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<2x3x5x2xf64, #Sparse4dTensor> into tensor<6x10xf64>
return %0 : tensor<6x10xf64>
}
func.func @collapse_to_sparse_6x10(%arg0: tensor<2x3x5x2xf64>) -> tensor<6x10xf64, #SparseMatrix> {
%0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<2x3x5x2xf64> into tensor<6x10xf64, #SparseMatrix>
return %0 : tensor<6x10xf64, #SparseMatrix>
}
func.func @collapse_sparse2sparse_6x10(%arg0: tensor<2x3x5x2xf64, #Sparse4dTensor>) -> tensor<6x10xf64, #SparseMatrix> {
%0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<2x3x5x2xf64, #Sparse4dTensor> into tensor<6x10xf64, #SparseMatrix>
return %0 : tensor<6x10xf64, #SparseMatrix>
}
func.func @collapse_dense_dyn(%arg0: tensor<?x?x?x?xf64>) -> tensor<?x?xf64> {
%0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<?x?x?x?xf64> into tensor<?x?xf64>
return %0 : tensor<?x?xf64>
}
func.func @collapse_from_sparse_dyn(%arg0: tensor<?x?x?x?xf64, #Sparse4dTensor>) -> tensor<?x?xf64> {
%0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<?x?x?x?xf64, #Sparse4dTensor> into tensor<?x?xf64>
return %0 : tensor<?x?xf64>
}
func.func @collapse_to_sparse_dyn(%arg0: tensor<?x?x?x?xf64>) -> tensor<?x?xf64, #SparseMatrix> {
%0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<?x?x?x?xf64> into tensor<?x?xf64, #SparseMatrix>
return %0 : tensor<?x?xf64, #SparseMatrix>
}
func.func @collapse_sparse2sparse_dyn(%arg0: tensor<?x?x?x?xf64, #Sparse4dTensor>) -> tensor<?x?xf64, #SparseMatrix> {
%0 = tensor.collapse_shape %arg0 [[0, 1], [2, 3]] : tensor<?x?x?x?xf64, #Sparse4dTensor> into tensor<?x?xf64, #SparseMatrix>
return %0 : tensor<?x?xf64, #SparseMatrix>
}
//
// Main driver.
//
func.func @entry() {
%c0 = arith.constant 0 : index
%df = arith.constant -1.0 : f64
// Setup test vectors and matrices..
%m = arith.constant dense <[ [ 1.1, 0.0, 1.3, 0.0 ],
[ 2.1, 0.0, 2.3, 0.0 ],
[ 3.1, 0.0, 3.3, 0.0 ]]> : tensor<3x4xf64>
%n = arith.constant dense <[
[ [[ 1.0, 0.0], [ 3.0, 0.0], [ 5.0, 0.0], [ 7.0, 0.0], [ 9.0, 0.0]],
[[ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0]],
[[21.0, 0.0], [23.0, 0.0], [25.0, 0.0], [27.0, 0.0], [29.0, 0.0]] ],
[ [[ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0]],
[[41.0, 0.0], [43.0, 0.0], [45.0, 0.0], [47.0, 0.0], [49.0, 0.0]],
[[ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0], [ 0.0, 0.0]] ] ]> : tensor<2x3x5x2xf64>
%sm = sparse_tensor.convert %m : tensor<3x4xf64> to tensor<3x4xf64, #SparseMatrix>
%sn = sparse_tensor.convert %n : tensor<2x3x5x2xf64> to tensor<2x3x5x2xf64, #Sparse4dTensor>
%dm = tensor.cast %m : tensor<3x4xf64> to tensor<?x?xf64>
%dn = tensor.cast %n : tensor<2x3x5x2xf64> to tensor<?x?x?x?xf64>
%sdn = sparse_tensor.convert %dn : tensor<?x?x?x?xf64> to tensor<?x?x?x?xf64, #Sparse4dTensor>
// Call the kernels.
%collapse0 = call @collapse_dense(%m) : (tensor<3x4xf64>) -> tensor<12xf64>
%collapse1 = call @collapse_from_sparse(%sm) : (tensor<3x4xf64, #SparseMatrix>) -> tensor<12xf64>
%collapse2 = call @collapse_to_sparse(%m) : (tensor<3x4xf64>) -> tensor<12xf64, #SparseVector>
%collapse3 = call @collapse_sparse2sparse(%sm) : (tensor<3x4xf64, #SparseMatrix>) -> tensor<12xf64, #SparseVector>
%collapse4 = call @collapse_dense_6x10(%n) : (tensor<2x3x5x2xf64>) -> tensor<6x10xf64>
%collapse5 = call @collapse_from_sparse_6x10(%sn) : (tensor<2x3x5x2xf64, #Sparse4dTensor>) -> tensor<6x10xf64>
%collapse6 = call @collapse_to_sparse_6x10(%n) : (tensor<2x3x5x2xf64>) -> tensor<6x10xf64, #SparseMatrix>
%collapse7 = call @collapse_sparse2sparse_6x10(%sn) : (tensor<2x3x5x2xf64, #Sparse4dTensor>) -> tensor<6x10xf64, #SparseMatrix>
%collapse8 = call @collapse_dense_dyn(%dn) : (tensor<?x?x?x?xf64>) -> tensor<?x?xf64>
%collapse9 = call @collapse_from_sparse_dyn(%sdn) : (tensor<?x?x?x?xf64, #Sparse4dTensor>) -> tensor<?x?xf64>
%collapse10 = call @collapse_to_sparse_dyn(%dn) : (tensor<?x?x?x?xf64>) -> tensor<?x?xf64, #SparseMatrix>
%collapse11 = call @collapse_sparse2sparse_dyn(%sdn) : (tensor<?x?x?x?xf64, #Sparse4dTensor>) -> tensor<?x?xf64, #SparseMatrix>
//
// Verify results of collapse
//
// CHECK: ( 1.1, 0, 1.3, 0, 2.1, 0, 2.3, 0, 3.1, 0, 3.3, 0 )
// CHECK-NEXT: ( 1.1, 0, 1.3, 0, 2.1, 0, 2.3, 0, 3.1, 0, 3.3, 0 )
// CHECK-NEXT: ( 1.1, 1.3, 2.1, 2.3, 3.1, 3.3
// CHECK-NEXT: ( 1.1, 1.3, 2.1, 2.3, 3.1, 3.3
// CHECK-NEXT: ( ( 1, 0, 3, 0, 5, 0, 7, 0, 9, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ),
// CHECK-SAME: ( 21, 0, 23, 0, 25, 0, 27, 0, 29, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ),
// CHECK-SAME: ( 41, 0, 43, 0, 45, 0, 47, 0, 49, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ) )
// CHECK-NEXT: ( ( 1, 0, 3, 0, 5, 0, 7, 0, 9, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ),
// CHECK-SAME: ( 21, 0, 23, 0, 25, 0, 27, 0, 29, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ),
// CHECK-SAME: ( 41, 0, 43, 0, 45, 0, 47, 0, 49, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ) )
// CHECK-NEXT: ( 1, 3, 5, 7, 9, 21, 23, 25, 27, 29, 41, 43, 45, 47
// CHECK-NEXT: ( 1, 3, 5, 7, 9, 21, 23, 25, 27, 29, 41, 43, 45, 47
// CHECK-NEXT: ( ( 1, 0, 3, 0, 5, 0, 7, 0, 9, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ),
// CHECK-SAME: ( 21, 0, 23, 0, 25, 0, 27, 0, 29, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ),
// CHECK-SAME: ( 41, 0, 43, 0, 45, 0, 47, 0, 49, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ) )
// CHECK-NEXT: ( ( 1, 0, 3, 0, 5, 0, 7, 0, 9, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ),
// CHECK-SAME: ( 21, 0, 23, 0, 25, 0, 27, 0, 29, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ),
// CHECK-SAME: ( 41, 0, 43, 0, 45, 0, 47, 0, 49, 0 ),
// CHECK-SAME: ( 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ) )
// CHECK-NEXT: ( 1, 3, 5, 7, 9, 21, 23, 25, 27, 29, 41, 43, 45, 47, 49
// CHECK-NEXT: ( 1, 3, 5, 7, 9, 21, 23, 25, 27, 29, 41, 43, 45, 47, 49
%v0 = vector.transfer_read %collapse0[%c0], %df: tensor<12xf64>, vector<12xf64>
vector.print %v0 : vector<12xf64>
%v1 = vector.transfer_read %collapse1[%c0], %df: tensor<12xf64>, vector<12xf64>
vector.print %v1 : vector<12xf64>
%b2 = sparse_tensor.values %collapse2 : tensor<12xf64, #SparseVector> to memref<?xf64>
%v2 = vector.transfer_read %b2[%c0], %df: memref<?xf64>, vector<12xf64>
vector.print %v2 : vector<12xf64>
%b3 = sparse_tensor.values %collapse3 : tensor<12xf64, #SparseVector> to memref<?xf64>
%v3 = vector.transfer_read %b3[%c0], %df: memref<?xf64>, vector<12xf64>
vector.print %v3 : vector<12xf64>
%v4 = vector.transfer_read %collapse4[%c0, %c0], %df: tensor<6x10xf64>, vector<6x10xf64>
vector.print %v4 : vector<6x10xf64>
%v5 = vector.transfer_read %collapse5[%c0, %c0], %df: tensor<6x10xf64>, vector<6x10xf64>
vector.print %v5 : vector<6x10xf64>
%b6 = sparse_tensor.values %collapse6 : tensor<6x10xf64, #SparseMatrix> to memref<?xf64>
%v6 = vector.transfer_read %b6[%c0], %df: memref<?xf64>, vector<60xf64>
vector.print %v6 : vector<60xf64>
%b7 = sparse_tensor.values %collapse7 : tensor<6x10xf64, #SparseMatrix> to memref<?xf64>
%v7 = vector.transfer_read %b7[%c0], %df: memref<?xf64>, vector<60xf64>
vector.print %v7 : vector<60xf64>
%v8 = vector.transfer_read %collapse8[%c0, %c0], %df: tensor<?x?xf64>, vector<6x10xf64>
vector.print %v8 : vector<6x10xf64>
%v9 = vector.transfer_read %collapse9[%c0, %c0], %df: tensor<?x?xf64>, vector<6x10xf64>
vector.print %v9 : vector<6x10xf64>
%b10 = sparse_tensor.values %collapse10 : tensor<?x?xf64, #SparseMatrix> to memref<?xf64>
%v10 = vector.transfer_read %b10[%c0], %df: memref<?xf64>, vector<60xf64>
vector.print %v10 : vector<60xf64>
%b11 = sparse_tensor.values %collapse11 : tensor<?x?xf64, #SparseMatrix> to memref<?xf64>
%v11 = vector.transfer_read %b11[%c0], %df: memref<?xf64>, vector<60xf64>
vector.print %v11 : vector<60xf64>
// Release sparse resources.
bufferization.dealloc_tensor %sm : tensor<3x4xf64, #SparseMatrix>
bufferization.dealloc_tensor %sn : tensor<2x3x5x2xf64, #Sparse4dTensor>
bufferization.dealloc_tensor %sdn : tensor<?x?x?x?xf64, #Sparse4dTensor>
bufferization.dealloc_tensor %collapse2 : tensor<12xf64, #SparseVector>
bufferization.dealloc_tensor %collapse3 : tensor<12xf64, #SparseVector>
bufferization.dealloc_tensor %collapse6 : tensor<6x10xf64, #SparseMatrix>
bufferization.dealloc_tensor %collapse7 : tensor<6x10xf64, #SparseMatrix>
bufferization.dealloc_tensor %collapse10 : tensor<?x?xf64, #SparseMatrix>
bufferization.dealloc_tensor %collapse11 : tensor<?x?xf64, #SparseMatrix>
// Release dense resources.
bufferization.dealloc_tensor %collapse1 : tensor<12xf64>
bufferization.dealloc_tensor %collapse5 : tensor<6x10xf64>
bufferization.dealloc_tensor %collapse9: tensor<?x?xf64>
return
}
}
|
