File: conv_for_implicit_gemm.cpp

package info (click to toggle)
miopen 6.4.3%2Bdfsg-1
  • links: PTS, VCS
  • area: main
  • in suites: sid
  • size: 66,788 kB
  • sloc: cpp: 300,511; lisp: 29,731; ansic: 2,683; sh: 471; python: 323; makefile: 155
file content (283 lines) | stat: -rw-r--r-- 13,617 bytes parent folder | download
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
/*******************************************************************************
 *
 * MIT License
 *
 * Copyright (c) 2023 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 <tuple>

#include <miopen/miopen.h>
#include <gtest/gtest.h>
#include <miopen/env.hpp>
#include "../conv2d.hpp"
#include "get_handle.hpp"

using TestCase = std::tuple<std::vector<std::string>, std::string>;

MIOPEN_DECLARE_ENV_VAR_BOOL(IMPLICITGEMM_TESTING_ENV)

namespace env = miopen::env;

namespace test_conv_for_implicit_gemm {

static bool SkipTest()
{
    if(env::enabled(IMPLICITGEMM_TESTING_ENV))
        return false;
    return true;
}

void GetArgs(const TestCase& param, std::vector<std::string>& tokens)
{
    auto env_vars = std::get<0>(param);
    for(auto& elem : env_vars)
    {
        putenv(elem.data());
    }

    auto cmd = std::get<1>(param);

    std::stringstream ss(cmd);
    std::istream_iterator<std::string> begin(ss);
    std::istream_iterator<std::string> end;
    while(begin != end)
        tokens.push_back(*begin++);
}

class GPU_ConvImplicitGemm_FP16 : public testing::TestWithParam<std::vector<TestCase>>
{
};

class GPU_ConvImplicitGemm_BFP16 : public testing::TestWithParam<std::vector<TestCase>>
{
};

void Run2dDriver(miopenDataType_t prec)
{
    std::vector<TestCase> params;

    switch(prec)
    {
    case miopenHalf: params = GPU_ConvImplicitGemm_FP16::GetParam(); break;
    case miopenBFloat16: params = GPU_ConvImplicitGemm_BFP16::GetParam(); break;
    case miopenFloat:
    case miopenInt8:
    case miopenInt32:
    case miopenInt64:
    case miopenDouble:
    case miopenFloat8:
    case miopenBFloat8:
    default:
        FAIL() << "miopenFloat, miopenInt8, miopenInt32, miopenDouble, miopenFloat8, miopenBFloat8 "
                  "data type not supported by test_conv_for_implicit_gemm test";
    }

    for(const auto& test_value : params)
    {
        std::vector<std::string> tokens;
        GetArgs(test_value, tokens);
        std::vector<const char*> ptrs;

        std::transform(tokens.begin(), tokens.end(), std::back_inserter(ptrs), [](const auto& str) {
            return str.data();
        });

        testing::internal::CaptureStderr();
        test_drive<conv2d_driver>(ptrs.size(), ptrs.data(), "test_conv2d");
        auto capture = testing::internal::GetCapturedStderr();
        EXPECT_FALSE(capture.find("No suitable algorithm was found") != std::string::npos);
        std::cout << capture;
    }
}

bool IsTestSupportedForDevice(const miopen::Handle& handle)
{
    std::string devName = handle.GetDeviceName();
    if(devName == "gfx900" || devName == "gfx906" || devName == "gfx908" || devName == "gfx90a" ||
       miopen::StartsWith(devName, "gfx94") || miopen::StartsWith(devName, "gfx103") ||
       miopen::StartsWith(devName, "gfx110"))
    {
        return true;
    }
    return false;
}

std::vector<TestCase> GetTestCases(const std::string& precision)
{

    std::vector<std::string> env = {
        "MIOPEN_FIND_MODE=normal",
        "MIOPEN_DEBUG_CONV_WINOGRAD=0",
        "MIOPEN_DEBUG_CONV_GEMM=0",
        "MIOPEN_DEBUG_CONV_DIRECT=0",
        "MIOPEN_DEBUG_CONV_IMPLICIT_GEMM=1",
        "MIOPEN_DEBUG_CONV_FFT=0",
    };

    std::string flags = " --verbose ";

    std::string psd0 = " --pads_strides_dilations 0 0 2 2 1 1";
    std::string psd1 = " --pads_strides_dilations 0 0 1 1 1 1";
    std::string psd2 = " --pads_strides_dilations 1 1 2 2 1 1";
    std::string psd3 = " --pads_strides_dilations 0 0 1 1 2 2";
    std::string psd4 = " --pads_strides_dilations 1 1 1 1 1 1";
    std::string psd5 = " --pads_strides_dilations 2 2 2 2 1 1";

    const std::vector<TestCase> test_cases = {

        // clang-format off
        TestCase{env, precision + flags + "--input 64 16 28 28 --weights 192 16 3 3 " + psd0},
        TestCase{env, precision + flags + "--input 64 16 14 14 --weights 160 16 3 3 " + psd0},
        TestCase{env, precision + flags +   "--input 64 16 7 7 --weights 128 16 3 3 " + psd0},
        TestCase{env, precision + flags +   "--input 64 16 55 55 --weights 96 16 1 7 " + psd0},
        TestCase{env, precision + flags +   "--input 64 16 28 28 --weights 64 16 1 7 " + psd0},
        TestCase{env, precision + flags +   "--input 64 16 14 14 --weights 32 16 1 7 " + psd0},
        TestCase{env, precision + flags +   "--input 64 32 28 28 --weights 192 32 3 3 " + psd0},
        TestCase{env, precision + flags +   "--input 64 32 14 14 --weights 160 32 3 3 " + psd0},
        TestCase{env, precision + flags +   "--input 64 32 7 7 --weights 128 32 3 3 " + psd0},
        TestCase{env, precision + flags +   "--input 64 32 55 55 --weights 96 32 1 7 " + psd0},
        TestCase{env, precision + flags +   "--input 64 32 28 28 --weights 64 32 1 7 " + psd0},
        TestCase{env, precision + flags +   "--input 64 32 14 14 --weights 32 32 1 7 " + psd0},

        TestCase{env, precision + flags +   "--input 64 64 56 56 --weights 256 64 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 64 56 56 --weights 64 64 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 64 73 73 --weights 80 64 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 64 56 56 --weights 64 64 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 128 55 55 --weights 16 128 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 128 28 28 --weights 16 128 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 128 14 14 --weights 16 128 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 128 7 7 --weights 16 128 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 16 64 56 56 --weights 256 64 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 16 64 56 56 --weights 64 64 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 16 64 73 73 --weights 80 64 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 16 64 56 56 --weights 64 64 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 16 128 55 55 --weights 16 128 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 16 128 28 28 --weights 16 128 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 16 128 7 7 --weights 16 128 1 1 " + psd1},

        TestCase{env, precision + flags +   "--input 64 64 55 55 --weights 16 128 1 1 " + psd0},
        TestCase{env, precision + flags +   "--input 64 128 28 28 --weights 16 128 1 1 " + psd0},
        TestCase{env, precision + flags +   "--input 64 128 14 14 --weights 16 128 1 1 " + psd0},
        TestCase{env, precision + flags +   "--input 64 128 7 7 --weights 16 128 1 1 " + psd0},

        TestCase{env, precision + flags +   "--input 64 128 28 28 --weights 512 128 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 160 73 73 --weights 64 160 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 192 35 35 --weights 32 192 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 192 35 35 --weights 48 192 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 192 35 35 --weights 64 192 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 192 28 28 --weights 16 192 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 192 28 28 --weights 32 192 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 192 28 28 --weights 64 192 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 192 28 28 --weights 96 192 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 256 35 35 --weights 48 256 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 256 35 35 --weights 64 256 1 1 " + psd1},

        TestCase{env, precision + flags +   "--input 64 256 56 56 --weights 128 256 1 1 " + psd0},
        TestCase{env, precision + flags +   "--input 64 256 56 56 --weights 512 256 1 1 " + psd0},


        TestCase{env, precision + flags +   "--input 64 256 56 56 --weights 64 256 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 256 28 28 --weights 128 256 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 256 28 28 --weights 32 256 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 256 28 28 --weights 64 256 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 288 35 35 --weights 48 288 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 288 35 35 --weights 64 288 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 384 35 35 --weights 192 384 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 384 35 35 --weights 64 384 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 384 35 35 --weights 96 384 1 1 " + psd1},

        TestCase{env, precision + flags +   "--input 64 480 14 14 --weights 16 480 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 480 14 14 --weights 192 480 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 480 14 14 --weights 64 480 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 480 14 14 --weights 96 480 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 512 28 28 --weights 128 512 1 1 " + psd1},

        TestCase{env, precision + flags +   "--input 64 512 28 28 --weights 256 512 1 1 " + psd0},

        TestCase{env, precision + flags +   "--input 64 512 14 14 --weights 112 512 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 512 14 14 --weights 128 512 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 512 14 14 --weights 144 512 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 512 14 14 --weights 160 512 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 512 14 14 --weights 24 512 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 512 14 14 --weights 32 512 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 64 512 14 14 --weights 64 512 1 1 " + psd1},

        TestCase{env, precision + flags +   "--input 128 832 7 7 --weights 32 832 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 128 832 7 7 --weights 192 832 1 1 " + psd1},
        TestCase{env, precision + flags +   "--input 128 832 7 7 --weights 128 832 1 1 " + psd1},

        TestCase{env, precision + flags +   "--input 128 832 7 7 --weights 32 832 1 1 " + psd3},
        TestCase{env, precision + flags +   "--input 128 832 7 7 --weights 192 832 1 1 " + psd3},
        TestCase{env, precision + flags +   "--input 128 832 7 7 --weights 128 832 1 1 " + psd3},
        TestCase{env, precision + flags +   "--input 16 2048 7 7 --weights 192 2048 1 1 " + psd3},

        TestCase{env, precision + flags +   "--input 64 32 28 28 --weights 192 32 3 3 " + psd2},
        TestCase{env, precision + flags +   "--input 8 16 14 14 --weights 32 16 1 1 " + psd4},
        TestCase{env, precision + flags +   "--input 64 32 14 14 --weights 192 32 3 3 " + psd2},
        TestCase{env, precision + flags +   "--input 64 32 7 7 --weights 192 32 3 3 " + psd2},
        TestCase{env, precision + flags +   "--input 64 32 28 28 --weights 192 32 3 3 " + psd5},
        TestCase{env, precision + flags +   "--input 64 32 14 14 --weights 192 32 3 3 " + psd5},
        TestCase{env, precision + flags +   "--input 64 32 7 7 --weights 192 32 3 3 " + psd5}
    };

    return test_cases ;
}

} //namespace test_conv_for_implicit_gemm

using namespace test_conv_for_implicit_gemm;

TEST_P(GPU_ConvImplicitGemm_BFP16, Test_conv_for_implicit_gemm_bf16)
{
    const auto& handle = get_handle();
    if(IsTestSupportedForDevice(handle) && !SkipTest())
    {
        Run2dDriver(miopenBFloat16);
    }
    else
    {
        GTEST_SKIP();
    }
};

TEST_P(GPU_ConvImplicitGemm_FP16, Test_conv_for_implicit_gemm_half)
{
    const auto& handle = get_handle();
    if(IsTestSupportedForDevice(handle) && !SkipTest())
    {
        Run2dDriver(miopenHalf);
    }
    else
    {
        GTEST_SKIP();
    }
};


INSTANTIATE_TEST_SUITE_P(Full,
                             GPU_ConvImplicitGemm_BFP16,
                             testing::Values(GetTestCases("--bfloat16")));

INSTANTIATE_TEST_SUITE_P(Full,
                             GPU_ConvImplicitGemm_FP16,
                             testing::Values(GetTestCases("--half")));