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
|
// 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<{
dimLevelType = ["compressed"]
}>
#SparseMatrix = #sparse_tensor.encoding<{
dimLevelType = ["compressed", "compressed"]
}>
#Sparse3dTensor = #sparse_tensor.encoding<{
dimLevelType = ["compressed", "compressed", "compressed"]
}>
#Sparse4dTensor = #sparse_tensor.encoding<{
dimLevelType = ["compressed", "compressed", "compressed", "compressed"]
}>
//
// Test with various forms of the two most elementary reshape
// operations: expand
//
module {
func.func @expand_dense(%arg0: tensor<12xf64>) -> tensor<3x4xf64> {
%0 = tensor.expand_shape %arg0 [[0, 1]] : tensor<12xf64> into tensor<3x4xf64>
return %0 : tensor<3x4xf64>
}
func.func @expand_from_sparse(%arg0: tensor<12xf64, #SparseVector>) -> tensor<3x4xf64> {
%0 = tensor.expand_shape %arg0 [[0, 1]] : tensor<12xf64, #SparseVector> into tensor<3x4xf64>
return %0 : tensor<3x4xf64>
}
func.func @expand_to_sparse(%arg0: tensor<12xf64>) -> tensor<3x4xf64, #SparseMatrix> {
%0 = tensor.expand_shape %arg0 [[0, 1]] : tensor<12xf64> into tensor<3x4xf64, #SparseMatrix>
return %0 : tensor<3x4xf64, #SparseMatrix>
}
func.func @expand_sparse2sparse(%arg0: tensor<12xf64, #SparseVector>) -> tensor<3x4xf64, #SparseMatrix> {
%0 = tensor.expand_shape %arg0 [[0, 1]] : tensor<12xf64, #SparseVector> into tensor<3x4xf64, #SparseMatrix>
return %0 : tensor<3x4xf64, #SparseMatrix>
}
func.func @expand_dense_3x2x2(%arg0: tensor<3x4xf64>) -> tensor<3x2x2xf64> {
%0 = tensor.expand_shape %arg0 [[0], [1, 2]] : tensor<3x4xf64> into tensor<3x2x2xf64>
return %0 : tensor<3x2x2xf64>
}
func.func @expand_from_sparse_3x2x2(%arg0: tensor<3x4xf64, #SparseMatrix>) -> tensor<3x2x2xf64> {
%0 = tensor.expand_shape %arg0 [[0], [1, 2]] : tensor<3x4xf64, #SparseMatrix> into tensor<3x2x2xf64>
return %0 : tensor<3x2x2xf64>
}
func.func @expand_to_sparse_3x2x2(%arg0: tensor<3x4xf64>) -> tensor<3x2x2xf64, #Sparse3dTensor> {
%0 = tensor.expand_shape %arg0 [[0], [1, 2]] : tensor<3x4xf64> into tensor<3x2x2xf64, #Sparse3dTensor>
return %0 : tensor<3x2x2xf64, #Sparse3dTensor>
}
func.func @expand_sparse2sparse_3x2x2(%arg0: tensor<3x4xf64, #SparseMatrix>) -> tensor<3x2x2xf64, #Sparse3dTensor> {
%0 = tensor.expand_shape %arg0 [[0], [1, 2]] : tensor<3x4xf64, #SparseMatrix> into tensor<3x2x2xf64, #Sparse3dTensor>
return %0 : tensor<3x2x2xf64, #Sparse3dTensor>
}
func.func @expand_dense_dyn(%arg0: tensor<?x?xf64>) -> tensor<?x2x?xf64> {
%0 = tensor.expand_shape %arg0 [[0], [1, 2]] : tensor<?x?xf64> into tensor<?x2x?xf64>
return %0 : tensor<?x2x?xf64>
}
func.func @expand_from_sparse_dyn(%arg0: tensor<?x?xf64, #SparseMatrix>) -> tensor<?x2x?xf64> {
%0 = tensor.expand_shape %arg0 [[0], [1, 2]] : tensor<?x?xf64, #SparseMatrix> into tensor<?x2x?xf64>
return %0 : tensor<?x2x?xf64>
}
func.func @expand_to_sparse_dyn(%arg0: tensor<?x?xf64>) -> tensor<?x2x?xf64, #Sparse3dTensor> {
%0 = tensor.expand_shape %arg0 [[0], [1, 2]] : tensor<?x?xf64> into tensor<?x2x?xf64, #Sparse3dTensor>
return %0 : tensor<?x2x?xf64, #Sparse3dTensor>
}
func.func @expand_sparse2sparse_dyn(%arg0: tensor<?x?xf64, #SparseMatrix>) -> tensor<?x2x?xf64, #Sparse3dTensor> {
%0 = tensor.expand_shape %arg0 [[0], [1, 2]] : tensor<?x?xf64, #SparseMatrix> into tensor<?x2x?xf64, #Sparse3dTensor>
return %0 : tensor<?x2x?xf64, #Sparse3dTensor>
}
//
// Main driver.
//
func.func @entry() {
%c0 = arith.constant 0 : index
%df = arith.constant -1.0 : f64
// Setup test vectors and matrices..
%v = arith.constant dense <[ 1.0, 0.0, 3.0, 0.0, 5.0, 0.0,
7.0, 0.0, 9.0, 0.0, 11.0, 0.0]> : tensor<12xf64>
%m = arith.constant dense <[ [ 1.1, 1.2, 1.3, 1.4 ],
[ 2.1, 2.2, 2.3, 2.4 ],
[ 3.1, 3.2, 3.3, 3.4 ]]> : tensor<3x4xf64>
%sv = sparse_tensor.convert %v : tensor<12xf64> to tensor<12xf64, #SparseVector>
%sm = sparse_tensor.convert %m : tensor<3x4xf64> to tensor<3x4xf64, #SparseMatrix>
%dm = tensor.cast %m : tensor<3x4xf64> to tensor<?x?xf64>
%sdm = sparse_tensor.convert %dm : tensor<?x?xf64> to tensor<?x?xf64, #SparseMatrix>
// Call the kernels.
%expand0 = call @expand_dense(%v) : (tensor<12xf64>) -> tensor<3x4xf64>
%expand1 = call @expand_from_sparse(%sv) : (tensor<12xf64, #SparseVector>) -> tensor<3x4xf64>
%expand2 = call @expand_to_sparse(%v) : (tensor<12xf64>) -> tensor<3x4xf64, #SparseMatrix>
%expand3 = call @expand_sparse2sparse(%sv) : (tensor<12xf64, #SparseVector>) -> tensor<3x4xf64, #SparseMatrix>
%expand4 = call @expand_dense_3x2x2(%m) : (tensor<3x4xf64>) -> tensor<3x2x2xf64>
%expand5 = call @expand_from_sparse_3x2x2(%sm) : (tensor<3x4xf64, #SparseMatrix>) -> tensor<3x2x2xf64>
%expand6 = call @expand_to_sparse_3x2x2(%m) : (tensor<3x4xf64>) -> tensor<3x2x2xf64, #Sparse3dTensor>
%expand7 = call @expand_sparse2sparse_3x2x2(%sm) : (tensor<3x4xf64, #SparseMatrix>) -> tensor<3x2x2xf64, #Sparse3dTensor>
%expand8 = call @expand_dense_dyn(%dm) : (tensor<?x?xf64>) -> tensor<?x2x?xf64>
%expand9 = call @expand_from_sparse_dyn(%sdm) : (tensor<?x?xf64, #SparseMatrix>) -> tensor<?x2x?xf64>
%expand10 = call @expand_to_sparse_dyn(%dm) : (tensor<?x?xf64>) -> tensor<?x2x?xf64, #Sparse3dTensor>
%expand11 = call @expand_sparse2sparse_dyn(%sdm) : (tensor<?x?xf64, #SparseMatrix>) -> tensor<?x2x?xf64, #Sparse3dTensor>
//
// Verify results of expand
//
// CHECK: ( ( 1, 0, 3, 0 ), ( 5, 0, 7, 0 ), ( 9, 0, 11, 0 ) )
// CHECK-NEXT: ( ( 1, 0, 3, 0 ), ( 5, 0, 7, 0 ), ( 9, 0, 11, 0 ) )
// CHECK-NEXT: ( 1, 3, 5, 7, 9,
// CHECK-NEXT: ( 1, 3, 5, 7, 9,
// CHECK-NEXT: ( ( ( 1.1, 1.2 ), ( 1.3, 1.4 ) ), ( ( 2.1, 2.2 ), ( 2.3, 2.4 ) ), ( ( 3.1, 3.2 ), ( 3.3, 3.4 ) ) )
// CHECK-NEXT: ( ( ( 1.1, 1.2 ), ( 1.3, 1.4 ) ), ( ( 2.1, 2.2 ), ( 2.3, 2.4 ) ), ( ( 3.1, 3.2 ), ( 3.3, 3.4 ) ) )
// CHECK-NEXT: ( 1.1, 1.2, 1.3, 1.4, 2.1, 2.2, 2.3, 2.4, 3.1, 3.2, 3.3, 3.4 )
// CHECK-NEXT: ( 1.1, 1.2, 1.3, 1.4, 2.1, 2.2, 2.3, 2.4, 3.1, 3.2, 3.3, 3.4 )
// CHECK-NEXT: ( ( ( 1.1, 1.2 ), ( 1.3, 1.4 ) ), ( ( 2.1, 2.2 ), ( 2.3, 2.4 ) ), ( ( 3.1, 3.2 ), ( 3.3, 3.4 ) ) )
// CHECK-NEXT: ( ( ( 1.1, 1.2 ), ( 1.3, 1.4 ) ), ( ( 2.1, 2.2 ), ( 2.3, 2.4 ) ), ( ( 3.1, 3.2 ), ( 3.3, 3.4 ) ) )
// CHECK-NEXT: 12
// CHECK-NEXT: ( 1.1, 1.2, 1.3, 1.4, 2.1, 2.2, 2.3, 2.4, 3.1, 3.2, 3.3, 3.4 )
// CHECK-NEXT: 12
// CHECK-NEXT: ( 1.1, 1.2, 1.3, 1.4, 2.1, 2.2, 2.3, 2.4, 3.1, 3.2, 3.3, 3.4 )
//
%m0 = vector.transfer_read %expand0[%c0, %c0], %df: tensor<3x4xf64>, vector<3x4xf64>
vector.print %m0 : vector<3x4xf64>
%m1 = vector.transfer_read %expand1[%c0, %c0], %df: tensor<3x4xf64>, vector<3x4xf64>
vector.print %m1 : vector<3x4xf64>
%a2 = sparse_tensor.values %expand2 : tensor<3x4xf64, #SparseMatrix> to memref<?xf64>
%m2 = vector.transfer_read %a2[%c0], %df: memref<?xf64>, vector<12xf64>
vector.print %m2 : vector<12xf64>
%a3 = sparse_tensor.values %expand3 : tensor<3x4xf64, #SparseMatrix> to memref<?xf64>
%m3 = vector.transfer_read %a3[%c0], %df: memref<?xf64>, vector<12xf64>
vector.print %m3 : vector<12xf64>
%m4 = vector.transfer_read %expand4[%c0, %c0, %c0], %df: tensor<3x2x2xf64>, vector<3x2x2xf64>
vector.print %m4 : vector<3x2x2xf64>
%m5 = vector.transfer_read %expand5[%c0, %c0, %c0], %df: tensor<3x2x2xf64>, vector<3x2x2xf64>
vector.print %m5 : vector<3x2x2xf64>
%a6 = sparse_tensor.values %expand6 : tensor<3x2x2xf64, #Sparse3dTensor> to memref<?xf64>
%m6 = vector.transfer_read %a6[%c0], %df: memref<?xf64>, vector<12xf64>
vector.print %m6 : vector<12xf64>
%a7 = sparse_tensor.values %expand7 : tensor<3x2x2xf64, #Sparse3dTensor> to memref<?xf64>
%m7 = vector.transfer_read %a7[%c0], %df: memref<?xf64>, vector<12xf64>
vector.print %m7 : vector<12xf64>
%m8 = vector.transfer_read %expand8[%c0, %c0, %c0], %df: tensor<?x2x?xf64>, vector<3x2x2xf64>
vector.print %m8 : vector<3x2x2xf64>
%m9 = vector.transfer_read %expand9[%c0, %c0, %c0], %df: tensor<?x2x?xf64>, vector<3x2x2xf64>
vector.print %m9 : vector<3x2x2xf64>
%n10 = sparse_tensor.number_of_entries %expand10 : tensor<?x2x?xf64, #Sparse3dTensor>
vector.print %n10 : index
%a10 = sparse_tensor.values %expand10 : tensor<?x2x?xf64, #Sparse3dTensor> to memref<?xf64>
%m10 = vector.transfer_read %a10[%c0], %df: memref<?xf64>, vector<12xf64>
vector.print %m10 : vector<12xf64>
%n11 = sparse_tensor.number_of_entries %expand11 : tensor<?x2x?xf64, #Sparse3dTensor>
vector.print %n11 : index
%a11 = sparse_tensor.values %expand11 : tensor<?x2x?xf64, #Sparse3dTensor> to memref<?xf64>
%m11 = vector.transfer_read %a11[%c0], %df: memref<?xf64>, vector<12xf64>
vector.print %m11 : vector<12xf64>
// Release sparse resources.
bufferization.dealloc_tensor %sv : tensor<12xf64, #SparseVector>
bufferization.dealloc_tensor %sm : tensor<3x4xf64, #SparseMatrix>
bufferization.dealloc_tensor %sdm : tensor<?x?xf64, #SparseMatrix>
bufferization.dealloc_tensor %expand2 : tensor<3x4xf64, #SparseMatrix>
bufferization.dealloc_tensor %expand3 : tensor<3x4xf64, #SparseMatrix>
bufferization.dealloc_tensor %expand6 : tensor<3x2x2xf64, #Sparse3dTensor>
bufferization.dealloc_tensor %expand7 : tensor<3x2x2xf64, #Sparse3dTensor>
bufferization.dealloc_tensor %expand10 : tensor<?x2x?xf64, #Sparse3dTensor>
bufferization.dealloc_tensor %expand11 : tensor<?x2x?xf64, #Sparse3dTensor>
// Release dense resources.
bufferization.dealloc_tensor %expand1 : tensor<3x4xf64>
bufferization.dealloc_tensor %expand5 : tensor<3x2x2xf64>
bufferization.dealloc_tensor %expand9 : tensor<?x2x?xf64>
return
}
}
|