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
|
/*
* Copyright (c) 2018-2020 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include "arm_compute/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
#include "arm_compute/core/Types.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/runtime/CL/CLTensor.h"
#include "arm_compute/runtime/CL/CLTensorAllocator.h"
#include "tests/CL/CLAccessor.h"
#include "tests/CL/Helper.h"
#include "tests/PaddingCalculator.h"
#include "tests/datasets/ShapeDatasets.h"
#include "tests/framework/Asserts.h"
#include "tests/framework/Macros.h"
#include "tests/framework/datasets/Datasets.h"
#include "tests/validation/Validation.h"
#include "tests/validation/fixtures/GEMMReshapeRHSMatrixFixture.h"
namespace arm_compute
{
namespace test
{
namespace validation
{
namespace
{
// *INDENT-OFF*
// clang-format off
/** Data types */
const auto data_types = framework::dataset::make("DataType", { DataType::QASYMM8, DataType::F16, DataType::F32 });
/** Batch size values to test */
const auto b_values = framework::dataset::make("batchsize", 1, 3);
/** N0 values to test */
const auto n0_values_nt_s32 = framework::dataset::make("N0", { 1, 2, 3 });
const auto n0_values_nt_s16 = framework::dataset::make("N0", { 4, 8 });
const auto n0_values_nt_s8 = framework::dataset::make("N0", { 16 });
const auto n0_values_t_s32 = framework::dataset::make("N0", { 4, 8 });
const auto n0_values_t_s16 = framework::dataset::make("N0", { 16 });
const auto n0_values_t_s8 = framework::dataset::make("N0", { 2, 3 });
/** K0 values to test */
const auto k0_values_nt_s32 = framework::dataset::make("K0", { 1, 2 });
const auto k0_values_nt_s16 = framework::dataset::make("K0", { 16 });
const auto k0_values_nt_s8 = framework::dataset::make("K0", { 3,4 });
const auto k0_values_t_s32 = framework::dataset::make("K0", { 2, 3 });
const auto k0_values_t_s16 = framework::dataset::make("K0", { 4, 8 });
const auto k0_values_t_s8 = framework::dataset::make("K0", { 16 });
/** H0 values to test */
const auto h0_values = framework::dataset::make("H0", 1, 4);
/** Interleave values to test */
const auto i_values = framework::dataset::make("interleave", { true, false });
} // namespace
using namespace arm_compute::misc::shape_calculator;
// Initialize the output tensor with zero and fill the border with zero
using CLGEMMReshapeRHSMatrix = CLSynthetizeFunctionInitOutputWithZeroAndWithZeroConstantBorder<CLGEMMReshapeRHSMatrixKernel, 16>;
template <typename T>
using CLGEMMReshapeRHSMatrixFixture = GEMMReshapeRHSMatrixValidationFixture<CLTensor, CLAccessor, CLGEMMReshapeRHSMatrix, T>;
TEST_SUITE(CL)
TEST_SUITE(GEMMReshapeRHSMatrix)
// *INDENT-OFF*
// clang-format off
DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(
framework::dataset::make("InputInfo", { TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32), // Mismatching data types
TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32), // Wrong n0 value
TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32), // Wrong k0 value
TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32), // Wrong h0 value
TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32), // n0 > 16
TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32), // k0 > 16
TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32), // k0 == 1 && transpose
}),
framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(64U, 2U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 2U, 2U), 1, DataType::F16),
TensorInfo(TensorShape(32U, 2U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 2U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 2U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 2U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 2U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 2U, 2U), 1, DataType::F32),
})),
framework::dataset::make("N0",{ 4, 0, 4, 4, 4, 17, 4, 4 })),
framework::dataset::make("K0",{ 4, 4, 0, 4, 4, 4, 17, 1 })),
framework::dataset::make("H0",{ 4, 4, 4, 0, 4, 4, 4, 4 })),
framework::dataset::make("Expected", { false, false, false, false, false, false, false})),
input_info, output_info, n0, k0, h0, expected)
{
GEMMRHSMatrixInfo rhs_info;
rhs_info.n0 = n0;
rhs_info.k0 = k0;
rhs_info.h0 = h0;
rhs_info.transpose = true;
rhs_info.interleave = true;
bool has_error = bool(CLGEMMReshapeRHSMatrixKernel::validate(&input_info.clone()->set_is_resizable(false), (output_info.total_size() == 0) ? nullptr : &output_info.clone()->set_is_resizable(false), rhs_info));
ARM_COMPUTE_EXPECT(has_error == expected, framework::LogLevel::ERRORS);
}
DATA_TEST_CASE(ValidatePadding, framework::DatasetMode::ALL, combine(combine(combine(
framework::dataset::make("InputShape", { TensorShape(32U, 16U, 1U),
TensorShape(32U, 16U, 2U)
}),
framework::dataset::make("N0",{ 4 })),
framework::dataset::make("K0",{ 4, 8, 16 })),
framework::dataset::make("H0",{ 1, 2, 4 })),
input_shape, n0, k0, h0)
{
CLTensor input;
CLTensor output;
input.info()->init(input_shape, 1, DataType::F32);
unsigned int padding = 0;
GEMMRHSMatrixInfo rhs_info;
rhs_info.n0 = n0;
rhs_info.k0 = k0;
rhs_info.h0 = h0;
rhs_info.transpose = true;
rhs_info.interleave = true;
rhs_info.export_to_cl_image = image2d_from_buffer_supported(CLKernelLibrary::get().get_device()) && (get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device()) != 0);
if(rhs_info.export_to_cl_image)
{
TensorShape output_shape = compute_rhs_reshaped_shape(*input.info(), rhs_info);
constexpr unsigned int num_floats_per_pixel = 4;
const unsigned int pixel_aligment = get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device());
const unsigned int row_pitch_alignment = pixel_aligment * num_floats_per_pixel;
const unsigned int round_up_width = ((output_shape[0] + row_pitch_alignment - 1) / row_pitch_alignment) * row_pitch_alignment;
padding = round_up_width - output_shape[0];
}
CLGEMMReshapeRHSMatrixKernel kernel;
kernel.configure(&input, &output, rhs_info);
ARM_COMPUTE_EXPECT((output.info()->padding().right == padding), framework::LogLevel::ERRORS);
}
// clang-format on
// *INDENT-ON*
// Run S32 tests only for transpose = false
FIXTURE_DATA_TEST_CASE(S32_NT, CLGEMMReshapeRHSMatrixFixture<int>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(datasets::SmallGEMMReshape2DShapes(),
b_values),
framework::dataset::make("DataType", DataType::S32)),
n0_values_nt_s32),
k0_values_nt_s32),
h0_values),
i_values),
framework::dataset::make("transpose", false)))
{
// Validate output
validate(CLAccessor(_target), _reference);
}
// Run S32 tests only for transpose = true
FIXTURE_DATA_TEST_CASE(S32_T, CLGEMMReshapeRHSMatrixFixture<int>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(datasets::SmallGEMMReshape2DShapes(),
b_values),
framework::dataset::make("DataType", DataType::S32)),
n0_values_t_s32),
k0_values_t_s32),
h0_values),
i_values),
framework::dataset::make("transpose", true)))
{
// Validate output
validate(CLAccessor(_target), _reference);
}
// Run S16 tests only for transpose = false
FIXTURE_DATA_TEST_CASE(S16_NT, CLGEMMReshapeRHSMatrixFixture<short>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(datasets::SmallGEMMReshape2DShapes(),
b_values),
framework::dataset::make("DataType", DataType::S16)),
n0_values_nt_s16),
k0_values_nt_s16),
h0_values),
i_values),
framework::dataset::make("transpose", false)))
{
// Validate output
validate(CLAccessor(_target), _reference);
}
// Run S16 tests only for transpose = true
FIXTURE_DATA_TEST_CASE(S16_T, CLGEMMReshapeRHSMatrixFixture<short>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(datasets::SmallGEMMReshape2DShapes(),
b_values),
framework::dataset::make("DataType", DataType::S16)),
n0_values_t_s16),
k0_values_t_s16),
h0_values),
i_values),
framework::dataset::make("transpose", true)))
{
// Validate output
validate(CLAccessor(_target), _reference);
}
// Run S8 tests only for transpose = false
FIXTURE_DATA_TEST_CASE(S8_NT, CLGEMMReshapeRHSMatrixFixture<char>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(datasets::SmallGEMMReshape2DShapes(),
b_values),
framework::dataset::make("DataType", DataType::S8)),
n0_values_nt_s8),
k0_values_nt_s8),
h0_values),
i_values),
framework::dataset::make("transpose", false)))
{
// Validate output
validate(CLAccessor(_target), _reference);
}
// Run S8 tests only for transpose = true
FIXTURE_DATA_TEST_CASE(S8_T, CLGEMMReshapeRHSMatrixFixture<char>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(datasets::SmallGEMMReshape2DShapes(),
b_values),
framework::dataset::make("DataType", DataType::S8)),
n0_values_t_s8),
k0_values_t_s8),
h0_values),
i_values),
framework::dataset::make("transpose", true)))
{
// Validate output
validate(CLAccessor(_target), _reference);
}
TEST_SUITE_END() // GEMMReshapeRHSMatrix
TEST_SUITE_END() // CL
} // namespace validation
} // namespace test
} // namespace arm_compute
|
