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
|
/*
* Copyright (c) 2019 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/CLGEMMMatrixMultiplyKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
#include "arm_compute/core/KernelDescriptors.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/GEMMFixture.h"
namespace arm_compute
{
namespace test
{
namespace validation
{
using namespace arm_compute::misc::shape_calculator;
// Create function for CLGEMMReshapeLHSMatrixKernel
using CLGEMMReshapeLHSMatrix = CLSynthetizeFunction<CLGEMMReshapeLHSMatrixKernel>;
// Create function for CLGEMMReshapeRHSMatrixKernel
using CLGEMMReshapeRHSMatrix = CLSynthetizeFunction<CLGEMMReshapeRHSMatrixKernel>;
// Create function for CLGEMMMatrixMultiplyKernel
using CLGEMMMatrixMultiplyReshaped = CLSynthetizeFunction<CLGEMMMatrixMultiplyKernel>;
// Fixture for GEMMMatrixMultiplyInterleavedTransposedValidationFixture
template <typename T>
using CLGEMMMatrixMultiplyReshapedFixture =
GEMMMatrixMultiplyInterleavedTransposedValidationFixture<CLTensor, CLAccessor, T, CLGEMMReshapeLHSMatrix, CLGEMMReshapeRHSMatrix, CLGEMMMatrixMultiplyReshaped>;
// Fixture for GEMMMatrixMultiplyInterleavedTransposed3DValidationFixture
template <typename T>
using CLGEMMMatrixMultiplyReshaped3DFixture =
GEMMMatrixMultiplyInterleavedTransposed3DValidationFixture<CLTensor, CLAccessor, T, CLGEMMReshapeLHSMatrix, CLGEMMReshapeRHSMatrix, CLGEMMMatrixMultiplyReshaped>;
namespace
{
// *INDENT-OFF*
// clang-format off
RelativeTolerance<float> rel_tolerance_f32(0.001f);
constexpr float abs_tolerance_f32(0.0001f);
RelativeTolerance<half> rel_tolerance_f16(half(0.2));
constexpr float tolerance_num_f16 = 0.02f;
/** Alpha values to test - Precommit */
const auto alpha_values = framework::dataset::make("alpha", {1.0f, -0.75f} );
/** Beta values to test - Precommit */
const auto beta_values = framework::dataset::make("beta", {-0.35f, 0.0f} );
/** M values to test - Precommit */
const auto m_values_precommit = framework::dataset::make("M", 37);
/** N values to test - Precommit */
const auto n_values_precommit = framework::dataset::make("N", 51);
/** K values to test - Precommit */
const auto k_values_precommit = framework::dataset::make("K", 23);
/** M values to test - Nightly */
const auto m_values_nightly = framework::dataset::make("M", {421, 1});
/** N values to test - Nightly */
const auto n_values_nightly = framework::dataset::make("N", 323);
/** K values to test - Nightly */
const auto k_values_nightly = framework::dataset::make("K", 207);
/** M_W values to test - Precommit */
const auto m_w_values_precommit = framework::dataset::make("M_W", 5);
/** M_H values to test - Precommit */
const auto m_h_values_precommit = framework::dataset::make("M_H", 7);
/** M_W values to test - Nightly */
const auto m_w_values_nightly = framework::dataset::make("M_W", 13);
/** M_H values to test - Nightly */
const auto m_h_values_nightly = framework::dataset::make("M_H", 27);
/** Batch size values to test */
const auto b_values = framework::dataset::make("batch_size", 1, 3);
/** Activation values to test */
const auto act_values = framework::dataset::make("Activation",
{
ActivationLayerInfo(),
ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 8.f, 2.f),
});
/** V0 values to test - Precommit */
const auto v0_values_precommit = framework::dataset::make("V0", 2);
/** H0 values to test - Precommit */
const auto h0_values_precommit = framework::dataset::make("H0", 4);
/** V0 values to test - Nightly */
const auto v0_values_nightly = framework::dataset::make("V0", {2, 4});
/** H0 values to test - Nightly */
const auto h0_values_nightly = framework::dataset::make("H0", { 2, 4 });
/** Broadcast bias from vector to matrix */
const auto broadcast_bias_values = framework::dataset::make("broadcast_bias", {false, true} );
/** GPU architectures values to test */
const auto gpu_arch_values = framework::dataset::make("GPUArch",
{
GPUTarget::MIDGARD,
GPUTarget::BIFROST
});
/** Data types values to test in the configuration */
const auto data_type_values = framework::dataset::make("DataType",
{
DataType::F32,
DataType::F16
});
/** M values to test */
const auto fp16_mixed_precision_values = framework::dataset::make("fp16_mixed_precision", {true, false});
/** Configuration test */
void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned int k_value, unsigned int b_value, unsigned int v0_value, unsigned int h0_value, bool broadcast_bias, bool fp16_mixed_precision, const ActivationLayerInfo &act_info, DataType data_type, GPUTarget gpu_arch_value)
{
GEMMLHSMatrixInfo lhs_info;
lhs_info.m0 = 4;
lhs_info.k0 = 4;
lhs_info.v0 = v0_value;
lhs_info.interleave = true;
lhs_info.transpose = true;
GEMMRHSMatrixInfo rhs_info;
rhs_info.n0 = data_type == DataType::F32? 4 : 8;
rhs_info.k0 = 1;
rhs_info.h0 = h0_value;
rhs_info.interleave = false;
rhs_info.transpose = false;
GEMMReshapeInfo reshape_info(m_value, n_value, k_value, rhs_info.h0, lhs_info.v0, 0, false, broadcast_bias);
const TensorShape lhs_shape(k_value, m_value, b_value);
const TensorShape lhs_shape_reshaped = compute_lhs_reshaped_shape(TensorInfo(lhs_shape, 1, data_type),
lhs_info,
false);
const TensorShape rhs_shape(n_value, k_value, b_value);
const TensorShape rhs_shape_reshaped = compute_rhs_reshaped_shape(TensorInfo(rhs_shape, 1, data_type),
rhs_info);
const TensorShape dst_shape = compute_mm_shape(TensorInfo(lhs_shape_reshaped, 1, data_type),
TensorInfo(rhs_shape_reshaped, 1, data_type),
reshape_info);
const TensorShape bias_shape(n_value,
broadcast_bias? 1 : m_value,
broadcast_bias? 1 : b_value);
// Create tensors
CLTensor lhs_reshaped = create_tensor<CLTensor>(lhs_shape_reshaped, data_type);
CLTensor rhs_reshaped = create_tensor<CLTensor>(rhs_shape_reshaped, data_type);
CLTensor bias = create_tensor<CLTensor>(bias_shape, data_type);
CLTensor dst = create_tensor<CLTensor>(dst_shape, data_type);
ARM_COMPUTE_EXPECT(lhs_reshaped.info()->is_resizable(), framework::LogLevel::ERRORS);
ARM_COMPUTE_EXPECT(rhs_reshaped.info()->is_resizable(), framework::LogLevel::ERRORS);
ARM_COMPUTE_EXPECT(bias.info()->is_resizable(), framework::LogLevel::ERRORS);
ARM_COMPUTE_EXPECT(dst.info()->is_resizable(), framework::LogLevel::ERRORS);
// Create and configure function
CLGEMMMatrixMultiplyReshaped gemm;
gemm.configure(gpu_arch_value, &lhs_reshaped, &rhs_reshaped, &bias, &dst, 1.0f, 2.0f, true, reshape_info, fp16_mixed_precision, act_info);
}
} // namespace
TEST_SUITE(CL)
TEST_SUITE(GEMMMatrixMultiplyInterleavedTransposed)
TEST_SUITE(Float)
TEST_SUITE(FP32)
DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_values_precommit,
n_values_precommit),
k_values_precommit),
framework::dataset::make("batch_size", 1)),
v0_values_precommit),
h0_values_precommit),
broadcast_bias_values),
framework::dataset::make("fp16_mixed_precision", false)),
act_values),
data_type_values),
gpu_arch_values),
m_value, n_value, k_value, b_value, v0_value, h0_value, broadcast_bias, fp16_mixed_precision_value, act_value, data_type_value, gpu_arch_value)
{
validate_configuration(m_value, n_value, k_value, b_value, v0_value, h0_value, broadcast_bias, fp16_mixed_precision_value, act_value, data_type_value, gpu_arch_value);
}
FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<float>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_values_precommit,
n_values_precommit),
k_values_precommit),
b_values),
alpha_values),
beta_values),
v0_values_precommit),
h0_values_precommit),
broadcast_bias_values),
framework::dataset::make("fp16_mixed_precision", false)),
act_values),
framework::dataset::make("DataType", DataType::F32)),
gpu_arch_values))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
}
FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<float>, framework::DatasetMode::DISABLED,
combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_values_nightly,
n_values_nightly),
k_values_nightly),
b_values),
alpha_values),
beta_values),
v0_values_nightly),
h0_values_nightly),
broadcast_bias_values),
framework::dataset::make("fp16_mixed_precision", false)),
act_values),
framework::dataset::make("DataType", DataType::F32)),
gpu_arch_values))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
}
FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_w_values_precommit,
m_h_values_precommit),
n_values_precommit),
k_values_precommit),
b_values),
alpha_values),
beta_values),
v0_values_precommit),
h0_values_precommit),
broadcast_bias_values),
framework::dataset::make("fp16_mixed_precision", false)),
act_values),
framework::dataset::make("DataType", DataType::F32)),
gpu_arch_values))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
}
FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, framework::DatasetMode::DISABLED,
combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_w_values_nightly,
m_h_values_nightly),
n_values_nightly),
k_values_nightly),
b_values),
alpha_values),
beta_values),
v0_values_nightly),
h0_values_nightly),
broadcast_bias_values),
framework::dataset::make("fp16_mixed_precision", false)),
act_values),
framework::dataset::make("DataType", DataType::F32)),
gpu_arch_values))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
}
TEST_SUITE_END() // FP32
TEST_SUITE(FP16)
FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<half>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_values_precommit,
n_values_precommit),
k_values_precommit),
b_values),
alpha_values),
beta_values),
v0_values_precommit),
h0_values_precommit),
broadcast_bias_values),
fp16_mixed_precision_values),
act_values),
framework::dataset::make("DataType", DataType::F16)),
gpu_arch_values))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f16, tolerance_num_f16);
}
FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<half>, framework::DatasetMode::DISABLED,
combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_values_nightly,
n_values_nightly),
k_values_nightly),
b_values),
alpha_values),
beta_values),
v0_values_nightly),
h0_values_nightly),
broadcast_bias_values),
fp16_mixed_precision_values),
act_values),
framework::dataset::make("DataType", DataType::F16)),
gpu_arch_values))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f16, tolerance_num_f16);
}
FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_w_values_precommit,
m_h_values_precommit),
n_values_precommit),
k_values_precommit),
b_values),
alpha_values),
beta_values),
v0_values_precommit),
h0_values_precommit),
broadcast_bias_values),
fp16_mixed_precision_values),
act_values),
framework::dataset::make("DataType", DataType::F16)),
gpu_arch_values))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f16, tolerance_num_f16);
}
FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, framework::DatasetMode::DISABLED,
combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
m_w_values_nightly,
m_h_values_nightly),
n_values_nightly),
k_values_nightly),
b_values),
alpha_values),
beta_values),
v0_values_nightly),
h0_values_nightly),
broadcast_bias_values),
fp16_mixed_precision_values),
act_values),
framework::dataset::make("DataType", DataType::F16)),
gpu_arch_values))
{
// Validate output
validate(CLAccessor(_target), _reference, rel_tolerance_f16, tolerance_num_f16);
}
TEST_SUITE_END() // FP16
TEST_SUITE_END() // Float
TEST_SUITE_END() // GEMMMatrixMulipltyInterleavedTransposed
TEST_SUITE_END() // CL
} // namespace validation
} // namespace test
} // namespace arm_compute
|
