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
|
/*******************************************************************************
*
* MIT License
*
* Copyright (c) 2017 Advanced Micro Devices, Inc.
*
* 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 <miopen/tensor_ops.hpp>
#include <tensor_util.hpp>
#include "gtest_common.hpp"
namespace {
std::vector<std::vector<size_t>> tensorALensArr = {{32, 16, 8, 4, 4}, // tensor A
{16, 20, 16, 8},
{20, 16, 8},
{1, 16, 8},
{16, 8},
{8}};
std::vector<std::vector<size_t>> tensorBLensArr = {{32, 16, 8, 4, 4}, // tensor B
{32, 16, 1, 1, 1},
{1, 16, 8, 1, 1},
{1, 1, 8, 4, 1},
{16, 20, 16, 8},
{16, 20, 16, 1},
{16, 20, 1, 1},
{16, 1, 1, 1},
{1, 20, 16, 8},
{1, 20, 16, 1},
{1, 20, 1, 1},
{1, 1, 16, 8},
{1, 1, 1, 8},
{20, 16, 8},
{20, 16, 1},
{1, 16, 8},
{1, 16, 1},
{20, 1, 1},
{16, 8},
{16, 1},
{1, 8},
{8},
{1}};
std::vector<std::vector<int64_t>> offsetsArr = {
{0, 0, 0}, {64, 32, 16}, {32, 16, 32}, {32, 16, 32}};
std::vector<std::vector<float>> alphabetaArr = {{1, 1, 0}, {-1, 1, 1}, {1.0, 0.5, 0.3}};
std::vector<std::vector<size_t>> stridesArr = {{8 * 16 * 20 * 16, 8 * 16 * 20, 8 * 16, 8, 1}};
std::vector<bool> packedArr = {true, false};
std::vector<miopenTensorOp_t> operationArr = {
miopenTensorOpAdd, miopenTensorOpMul, miopenTensorOpMin, miopenTensorOpMax};
} // namespace
struct TestCase
{
std::vector<size_t> tensorlens_ac;
std::vector<size_t> tensorlens_b;
std::vector<int64_t> offsets;
std::vector<size_t> stride_a;
std::vector<size_t> stride_b;
std::vector<size_t> stride_c;
std::vector<float> alphabeta;
bool packed;
miopenTensorOp_t operation;
};
template <typename T>
struct TensorOpsCommon : public testing::TestWithParam<TestCase>
{
void SetUp() override { prng::reset_seed(); }
void Run()
{
CreateTensors();
std::vector<T> tensorGPUData = CalculateOnGPU();
std::vector<T> tensorCPUData = CalculateOnCPU();
CompareResults(tensorGPUData, tensorCPUData);
}
private:
void CreateTensors()
{
const TestCase& testCase = GetParam();
tensorA = CreateTensor(
testCase.tensorlens_ac, testCase.stride_a, testCase.offsets[0], testCase.packed);
tensorB = CreateTensor(
testCase.tensorlens_b, testCase.stride_b, testCase.offsets[1], testCase.packed);
tensorC = CreateTensor(
testCase.tensorlens_ac, testCase.stride_c, testCase.offsets[2], testCase.packed);
}
tensor<T> CreateTensor(const std::vector<size_t>& lens,
const std::vector<size_t>& strides,
int64_t offset,
bool isPacked)
{
uint64_t max_value = miopen_type<T>{} == miopenHalf ? 5 : 17;
if(!isPacked)
{
std::vector<size_t> real_strides(strides.begin() + (strides.size() - lens.size()),
strides.end());
auto r = tensor<T>{lens, real_strides}.generate(tensor_elem_gen_integer{max_value});
r.data.resize(r.data.size() + offset);
return r;
}
else
{
return tensor<T>{lens}.generate(tensor_elem_gen_integer{max_value});
}
}
std::vector<T> CalculateOnGPU() const
{
const TestCase& testCase = GetParam();
auto&& handle = get_handle();
auto a_dev = handle.Write(tensorA.data);
auto b_dev = handle.Write(tensorB.data);
auto c_dev = handle.Write(tensorC.data);
miopen::OpTensor(handle,
testCase.operation,
&testCase.alphabeta[0],
tensorA.desc,
a_dev.get(),
&testCase.alphabeta[1],
tensorB.desc,
b_dev.get(),
&testCase.alphabeta[2],
tensorC.desc,
c_dev.get(),
testCase.offsets[0],
testCase.offsets[1],
testCase.offsets[2],
false); // it does not verify non-standard behaviour
return handle.Read<T>(c_dev, tensorC.data.size());
}
std::vector<T> CalculateOnCPU()
{
const TestCase& testCase = GetParam();
float alpha1 = testCase.alphabeta[0];
float alpha2 = testCase.alphabeta[1];
float beta = testCase.alphabeta[2];
if(testCase.operation == miopenTensorOpAdd)
{
return CalculateOnCPUDataOp([alpha1, alpha2, beta](auto& C, auto A, auto B) {
C = A * alpha1 + B * alpha2 + C * beta;
});
}
else if(testCase.operation == miopenTensorOpMul)
{
return CalculateOnCPUDataOp([alpha1, alpha2, beta](auto& C, auto A, auto B) {
C = A * alpha1 * B * alpha2 + C * beta;
});
}
else if(testCase.operation == miopenTensorOpMin)
{
return CalculateOnCPUDataOp([alpha1, alpha2, beta](auto& C, auto A, auto B) {
C = std::min(A * alpha1, B * alpha2) + C * beta;
});
}
else
{
return CalculateOnCPUDataOp([alpha1, alpha2, beta](auto& C, auto A, auto B) {
C = std::max(A * alpha1, B * alpha2) + C * beta;
});
}
}
template <typename DataOp>
std::vector<T> CalculateOnCPUDataOp(DataOp&& dataOp)
{
const TestCase& testCase = GetParam();
auto r = tensorC;
operate_over_subtensor<>(dataOp,
r.data,
tensorA.data,
tensorB.data,
r.desc,
tensorA.desc,
tensorB.desc,
testCase.offsets[2],
testCase.offsets[0],
testCase.offsets[1]);
return r.data;
}
void CompareResults(const std::vector<T>& tensorGPUData, const std::vector<T>& tensorCPUData)
{
const TestCase& testCase = GetParam();
double tolerance = 1;
if(std::is_same_v<T, half_float::half>)
{
// taken from original c-test
tolerance = 80;
}
double threshold = std::numeric_limits<T>::epsilon() * tolerance;
double error = miopen::rms_range(tensorCPUData, tensorGPUData);
ASSERT_LE(error, threshold)
<< "TensorOp: " << testCase.operation << std::endl
<< "A tensor: " << tensorA.desc.ToString() << std::endl
<< "B tensor: " << tensorB.desc.ToString() << std::endl
<< "IsPacked: " << testCase.packed << std::endl
<< "Offsets: " << testCase.offsets[0] << "," << testCase.offsets[1] << ","
<< testCase.offsets[2] << std::endl;
}
private:
tensor<T> tensorA;
tensor<T> tensorB;
tensor<T> tensorC;
};
using GPU_TernaryTensorOps_FP32 = TensorOpsCommon<float>;
using GPU_TernaryTensorOps_FP16 = TensorOpsCommon<half_float::half>;
using GPU_TernaryTensorOps_FP64 = TensorOpsCommon<double>;
namespace {
bool checkTensorsCompatibility(const std::vector<size_t>& tensorALens,
const std::vector<size_t>& tensorBLens)
{
if(tensorALens.size() != tensorBLens.size())
{
return false;
}
for(size_t idx = 0; idx < tensorBLens.size(); ++idx)
{
if((tensorBLens[idx] != 1) && (tensorALens[idx] != tensorBLens[idx]))
{
return false;
}
}
return true;
}
void AddTestCases(std::vector<TestCase>& testCases,
const std::vector<size_t>& tensorALens,
const std::vector<size_t>& tensorBLens)
{
const auto& stride_a = stridesArr[0];
const auto& stride_b = stridesArr[0];
const auto& stride_c = stridesArr[0];
for(bool packed : packedArr)
for(const auto& offsets : offsetsArr)
{
std::vector<int64_t> final_offsets{0, 0, 0};
if(!packed)
{
if(std::any_of(offsets.begin(), offsets.end(), [](int64_t o) { return o < 0; }))
continue;
final_offsets = offsets;
}
auto checkStride = [p = packed](const std::vector<size_t>& lens,
const std::vector<size_t>& strides) {
if(p)
return true;
if(lens.size() > strides.size())
return false;
// only sparsed case allowed, since all the kernels do not support the last
// dimension strides
if(strides.back() == 1)
{
// we use float here for all types because strides are independent to type
auto packedStrides =
miopen::TensorDescriptor(miopen_type<float>{}, lens).GetStrides();
return std::equal(packedStrides.rbegin(),
packedStrides.rend(),
strides.rbegin(),
[](size_t ps, size_t s) { return s >= ps; });
}
// currently tensor operations do not support non-one stride in the last dimention.
return false;
};
if(!checkStride(tensorALens, stride_a))
continue;
if(!checkStride(tensorBLens, stride_b))
continue;
if(!checkStride(tensorALens, stride_c))
continue;
for(const auto& alphabeta : alphabetaArr)
for(const auto& operation : operationArr)
{
TestCase& testCase = testCases.emplace_back();
testCase.tensorlens_ac = tensorALens;
testCase.tensorlens_b = tensorBLens;
testCase.alphabeta = alphabeta;
testCase.offsets = final_offsets;
testCase.packed = packed;
testCase.operation = operation;
testCase.stride_a = stride_a;
testCase.stride_b = stride_b;
testCase.stride_c = stride_c;
}
}
}
std::vector<TestCase> GenCases()
{
std::vector<TestCase> testCases;
for(const auto& tensorALens : tensorALensArr)
for(const auto& tensorBLens : tensorBLensArr)
{
if(!checkTensorsCompatibility(tensorALens, tensorBLens))
{
continue;
}
AddTestCases(testCases, tensorALens, tensorBLens);
}
return testCases;
}
inline auto GetCases()
{
static const auto cases = testing::ValuesIn(GenCases());
return cases;
}
} // namespace
TEST_P(GPU_TernaryTensorOps_FP32, TestFloat) { this->Run(); }
TEST_P(GPU_TernaryTensorOps_FP16, TestFloat16) { this->Run(); }
TEST_P(GPU_TernaryTensorOps_FP64, TestDouble) { this->Run(); }
INSTANTIATE_TEST_SUITE_P(Smoke, GPU_TernaryTensorOps_FP32, GetCases());
INSTANTIATE_TEST_SUITE_P(Full, GPU_TernaryTensorOps_FP64, GetCases());
INSTANTIATE_TEST_SUITE_P(Full, GPU_TernaryTensorOps_FP16, GetCases());
|
