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
|
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2016
// Mehdi Goli Codeplay Software Ltd.
// Ralph Potter Codeplay Software Ltd.
// Luke Iwanski Codeplay Software Ltd.
// Contact: <eigen@codeplay.com>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
#define EIGEN_TEST_NO_LONGDOUBLE
#define EIGEN_TEST_NO_COMPLEX
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int64_t
#define EIGEN_USE_SYCL
#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>
using Eigen::array;
using Eigen::SyclDevice;
using Eigen::Tensor;
using Eigen::TensorMap;
namespace std {
template <typename T> T rsqrt(T x) { return 1 / std::sqrt(x); }
template <typename T> T square(T x) { return x * x; }
template <typename T> T cube(T x) { return x * x * x; }
template <typename T> T inverse(T x) { return 1 / x; }
}
#define TEST_UNARY_BUILTINS_FOR_SCALAR(FUNC, SCALAR, OPERATOR, Layout) \
{ \
/* out OPERATOR in.FUNC() */ \
Tensor<SCALAR, 3, Layout, int64_t> in(tensorRange); \
Tensor<SCALAR, 3, Layout, int64_t> out(tensorRange); \
in = in.random() + static_cast<SCALAR>(0.01); \
out = out.random() + static_cast<SCALAR>(0.01); \
Tensor<SCALAR, 3, Layout, int64_t> reference(out); \
SCALAR *gpu_data = static_cast<SCALAR *>( \
sycl_device.allocate(in.size() * sizeof(SCALAR))); \
SCALAR *gpu_data_out = static_cast<SCALAR *>( \
sycl_device.allocate(out.size() * sizeof(SCALAR))); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu(gpu_data, tensorRange); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_out(gpu_data_out, tensorRange); \
sycl_device.memcpyHostToDevice(gpu_data, in.data(), \
(in.size()) * sizeof(SCALAR)); \
sycl_device.memcpyHostToDevice(gpu_data_out, out.data(), \
(out.size()) * sizeof(SCALAR)); \
gpu_out.device(sycl_device) OPERATOR gpu.FUNC(); \
sycl_device.memcpyDeviceToHost(out.data(), gpu_data_out, \
(out.size()) * sizeof(SCALAR)); \
for (int64_t i = 0; i < out.size(); ++i) { \
SCALAR ver = reference(i); \
ver OPERATOR std::FUNC(in(i)); \
VERIFY_IS_APPROX(out(i), ver); \
} \
sycl_device.deallocate(gpu_data); \
sycl_device.deallocate(gpu_data_out); \
} \
{ \
/* out OPERATOR out.FUNC() */ \
Tensor<SCALAR, 3, Layout, int64_t> out(tensorRange); \
out = out.random() + static_cast<SCALAR>(0.01); \
Tensor<SCALAR, 3, Layout, int64_t> reference(out); \
SCALAR *gpu_data_out = static_cast<SCALAR *>( \
sycl_device.allocate(out.size() * sizeof(SCALAR))); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_out(gpu_data_out, tensorRange); \
sycl_device.memcpyHostToDevice(gpu_data_out, out.data(), \
(out.size()) * sizeof(SCALAR)); \
gpu_out.device(sycl_device) OPERATOR gpu_out.FUNC(); \
sycl_device.memcpyDeviceToHost(out.data(), gpu_data_out, \
(out.size()) * sizeof(SCALAR)); \
for (int64_t i = 0; i < out.size(); ++i) { \
SCALAR ver = reference(i); \
ver OPERATOR std::FUNC(reference(i)); \
VERIFY_IS_APPROX(out(i), ver); \
} \
sycl_device.deallocate(gpu_data_out); \
}
#define TEST_UNARY_BUILTINS_OPERATOR(SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(abs, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(sqrt, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(rsqrt, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(square, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(cube, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(inverse, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(tanh, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(exp, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(expm1, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(log, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(abs, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(ceil, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(floor, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(round, SCALAR, OPERATOR , Layout) \
TEST_UNARY_BUILTINS_FOR_SCALAR(log1p, SCALAR, OPERATOR , Layout)
#define TEST_IS_THAT_RETURNS_BOOL(SCALAR, FUNC, Layout) \
{ \
/* out = in.FUNC() */ \
Tensor<SCALAR, 3, Layout, int64_t> in(tensorRange); \
Tensor<bool, 3, Layout, int64_t> out(tensorRange); \
in = in.random() + static_cast<SCALAR>(0.01); \
SCALAR *gpu_data = static_cast<SCALAR *>( \
sycl_device.allocate(in.size() * sizeof(SCALAR))); \
bool *gpu_data_out = \
static_cast<bool *>(sycl_device.allocate(out.size() * sizeof(bool))); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu(gpu_data, tensorRange); \
TensorMap<Tensor<bool, 3, Layout, int64_t>> gpu_out(gpu_data_out, tensorRange); \
sycl_device.memcpyHostToDevice(gpu_data, in.data(), \
(in.size()) * sizeof(SCALAR)); \
gpu_out.device(sycl_device) = gpu.FUNC(); \
sycl_device.memcpyDeviceToHost(out.data(), gpu_data_out, \
(out.size()) * sizeof(bool)); \
for (int64_t i = 0; i < out.size(); ++i) { \
VERIFY_IS_EQUAL(out(i), std::FUNC(in(i))); \
} \
sycl_device.deallocate(gpu_data); \
sycl_device.deallocate(gpu_data_out); \
}
#define TEST_UNARY_BUILTINS(SCALAR, Layout) \
TEST_UNARY_BUILTINS_OPERATOR(SCALAR, +=, Layout) \
TEST_UNARY_BUILTINS_OPERATOR(SCALAR, =, Layout) \
TEST_IS_THAT_RETURNS_BOOL(SCALAR, isnan, Layout) \
TEST_IS_THAT_RETURNS_BOOL(SCALAR, isfinite, Layout) \
TEST_IS_THAT_RETURNS_BOOL(SCALAR, isinf, Layout)
static void test_builtin_unary_sycl(const Eigen::SyclDevice &sycl_device) {
int64_t sizeDim1 = 10;
int64_t sizeDim2 = 10;
int64_t sizeDim3 = 10;
array<int64_t, 3> tensorRange = {{sizeDim1, sizeDim2, sizeDim3}};
TEST_UNARY_BUILTINS(float, RowMajor)
TEST_UNARY_BUILTINS(float, ColMajor)
}
namespace std {
template <typename T> T cwiseMax(T x, T y) { return std::max(x, y); }
template <typename T> T cwiseMin(T x, T y) { return std::min(x, y); }
}
#define TEST_BINARY_BUILTINS_FUNC(SCALAR, FUNC, Layout) \
{ \
/* out = in_1.FUNC(in_2) */ \
Tensor<SCALAR, 3, Layout, int64_t> in_1(tensorRange); \
Tensor<SCALAR, 3, Layout, int64_t> in_2(tensorRange); \
Tensor<SCALAR, 3, Layout, int64_t> out(tensorRange); \
in_1 = in_1.random() + static_cast<SCALAR>(0.01); \
in_2 = in_2.random() + static_cast<SCALAR>(0.01); \
Tensor<SCALAR, 3, Layout, int64_t> reference(out); \
SCALAR *gpu_data_1 = static_cast<SCALAR *>( \
sycl_device.allocate(in_1.size() * sizeof(SCALAR))); \
SCALAR *gpu_data_2 = static_cast<SCALAR *>( \
sycl_device.allocate(in_2.size() * sizeof(SCALAR))); \
SCALAR *gpu_data_out = static_cast<SCALAR *>( \
sycl_device.allocate(out.size() * sizeof(SCALAR))); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_1(gpu_data_1, tensorRange); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_2(gpu_data_2, tensorRange); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_out(gpu_data_out, tensorRange); \
sycl_device.memcpyHostToDevice(gpu_data_1, in_1.data(), \
(in_1.size()) * sizeof(SCALAR)); \
sycl_device.memcpyHostToDevice(gpu_data_2, in_2.data(), \
(in_2.size()) * sizeof(SCALAR)); \
gpu_out.device(sycl_device) = gpu_1.FUNC(gpu_2); \
sycl_device.memcpyDeviceToHost(out.data(), gpu_data_out, \
(out.size()) * sizeof(SCALAR)); \
for (int64_t i = 0; i < out.size(); ++i) { \
SCALAR ver = reference(i); \
ver = std::FUNC(in_1(i), in_2(i)); \
VERIFY_IS_APPROX(out(i), ver); \
} \
sycl_device.deallocate(gpu_data_1); \
sycl_device.deallocate(gpu_data_2); \
sycl_device.deallocate(gpu_data_out); \
}
#define TEST_BINARY_BUILTINS_OPERATORS(SCALAR, OPERATOR, Layout) \
{ \
/* out = in_1 OPERATOR in_2 */ \
Tensor<SCALAR, 3, Layout, int64_t> in_1(tensorRange); \
Tensor<SCALAR, 3, Layout, int64_t> in_2(tensorRange); \
Tensor<SCALAR, 3, Layout, int64_t> out(tensorRange); \
in_1 = in_1.random() + static_cast<SCALAR>(0.01); \
in_2 = in_2.random() + static_cast<SCALAR>(0.01); \
Tensor<SCALAR, 3, Layout, int64_t> reference(out); \
SCALAR *gpu_data_1 = static_cast<SCALAR *>( \
sycl_device.allocate(in_1.size() * sizeof(SCALAR))); \
SCALAR *gpu_data_2 = static_cast<SCALAR *>( \
sycl_device.allocate(in_2.size() * sizeof(SCALAR))); \
SCALAR *gpu_data_out = static_cast<SCALAR *>( \
sycl_device.allocate(out.size() * sizeof(SCALAR))); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_1(gpu_data_1, tensorRange); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_2(gpu_data_2, tensorRange); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_out(gpu_data_out, tensorRange); \
sycl_device.memcpyHostToDevice(gpu_data_1, in_1.data(), \
(in_1.size()) * sizeof(SCALAR)); \
sycl_device.memcpyHostToDevice(gpu_data_2, in_2.data(), \
(in_2.size()) * sizeof(SCALAR)); \
gpu_out.device(sycl_device) = gpu_1 OPERATOR gpu_2; \
sycl_device.memcpyDeviceToHost(out.data(), gpu_data_out, \
(out.size()) * sizeof(SCALAR)); \
for (int64_t i = 0; i < out.size(); ++i) { \
VERIFY_IS_APPROX(out(i), in_1(i) OPERATOR in_2(i)); \
} \
sycl_device.deallocate(gpu_data_1); \
sycl_device.deallocate(gpu_data_2); \
sycl_device.deallocate(gpu_data_out); \
}
#define TEST_BINARY_BUILTINS_OPERATORS_THAT_TAKES_SCALAR(SCALAR, OPERATOR, Layout) \
{ \
/* out = in_1 OPERATOR 2 */ \
Tensor<SCALAR, 3, Layout, int64_t> in_1(tensorRange); \
Tensor<SCALAR, 3, Layout, int64_t> out(tensorRange); \
in_1 = in_1.random() + static_cast<SCALAR>(0.01); \
Tensor<SCALAR, 3, Layout, int64_t> reference(out); \
SCALAR *gpu_data_1 = static_cast<SCALAR *>( \
sycl_device.allocate(in_1.size() * sizeof(SCALAR))); \
SCALAR *gpu_data_out = static_cast<SCALAR *>( \
sycl_device.allocate(out.size() * sizeof(SCALAR))); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_1(gpu_data_1, tensorRange); \
TensorMap<Tensor<SCALAR, 3, Layout, int64_t>> gpu_out(gpu_data_out, tensorRange); \
sycl_device.memcpyHostToDevice(gpu_data_1, in_1.data(), \
(in_1.size()) * sizeof(SCALAR)); \
gpu_out.device(sycl_device) = gpu_1 OPERATOR 2; \
sycl_device.memcpyDeviceToHost(out.data(), gpu_data_out, \
(out.size()) * sizeof(SCALAR)); \
for (int64_t i = 0; i < out.size(); ++i) { \
VERIFY_IS_APPROX(out(i), in_1(i) OPERATOR 2); \
} \
sycl_device.deallocate(gpu_data_1); \
sycl_device.deallocate(gpu_data_out); \
}
#define TEST_BINARY_BUILTINS(SCALAR, Layout) \
TEST_BINARY_BUILTINS_FUNC(SCALAR, cwiseMax , Layout) \
TEST_BINARY_BUILTINS_FUNC(SCALAR, cwiseMin , Layout) \
TEST_BINARY_BUILTINS_OPERATORS(SCALAR, + , Layout) \
TEST_BINARY_BUILTINS_OPERATORS(SCALAR, - , Layout) \
TEST_BINARY_BUILTINS_OPERATORS(SCALAR, * , Layout) \
TEST_BINARY_BUILTINS_OPERATORS(SCALAR, / , Layout)
static void test_builtin_binary_sycl(const Eigen::SyclDevice &sycl_device) {
int64_t sizeDim1 = 10;
int64_t sizeDim2 = 10;
int64_t sizeDim3 = 10;
array<int64_t, 3> tensorRange = {{sizeDim1, sizeDim2, sizeDim3}};
TEST_BINARY_BUILTINS(float, RowMajor)
TEST_BINARY_BUILTINS_OPERATORS_THAT_TAKES_SCALAR(int, %, RowMajor)
TEST_BINARY_BUILTINS(float, ColMajor)
TEST_BINARY_BUILTINS_OPERATORS_THAT_TAKES_SCALAR(int, %, ColMajor)
}
EIGEN_DECLARE_TEST(cxx11_tensor_builtins_sycl) {
for (const auto& device :Eigen::get_sycl_supported_devices()) {
QueueInterface queueInterface(device);
Eigen::SyclDevice sycl_device(&queueInterface);
CALL_SUBTEST(test_builtin_unary_sycl(sycl_device));
CALL_SUBTEST(test_builtin_binary_sycl(sycl_device));
}
}
|
