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
|
// 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_TEST_FUNC cxx11_tensor_builtins_sycl
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#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) \
{ \
/* out OPERATOR in.FUNC() */ \
Tensor<SCALAR, 3> in(tensorRange); \
Tensor<SCALAR, 3> out(tensorRange); \
in = in.random() + static_cast<SCALAR>(0.01); \
out = out.random() + static_cast<SCALAR>(0.01); \
Tensor<SCALAR, 3> 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>> gpu(gpu_data, tensorRange); \
TensorMap<Tensor<SCALAR, 3>> 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 (int 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> out(tensorRange); \
out = out.random() + static_cast<SCALAR>(0.01); \
Tensor<SCALAR, 3> reference(out); \
SCALAR *gpu_data_out = static_cast<SCALAR *>( \
sycl_device.allocate(out.size() * sizeof(SCALAR))); \
TensorMap<Tensor<SCALAR, 3>> 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 (int 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) \
TEST_UNARY_BUILTINS_FOR_SCALAR(abs, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(sqrt, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(rsqrt, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(square, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(cube, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(inverse, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(tanh, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(exp, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(log, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(abs, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(ceil, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(floor, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(round, SCALAR, OPERATOR) \
TEST_UNARY_BUILTINS_FOR_SCALAR(log1p, SCALAR, OPERATOR)
#define TEST_IS_THAT_RETURNS_BOOL(SCALAR, FUNC) \
{ \
/* out = in.FUNC() */ \
Tensor<SCALAR, 3> in(tensorRange); \
Tensor<bool, 3> 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>> gpu(gpu_data, tensorRange); \
TensorMap<Tensor<bool, 3>> 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 (int 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) \
TEST_UNARY_BUILTINS_OPERATOR(SCALAR, += ) \
TEST_UNARY_BUILTINS_OPERATOR(SCALAR, = ) \
TEST_IS_THAT_RETURNS_BOOL(SCALAR, isnan) \
TEST_IS_THAT_RETURNS_BOOL(SCALAR, isfinite) \
TEST_IS_THAT_RETURNS_BOOL(SCALAR, isinf)
static void test_builtin_unary_sycl(const Eigen::SyclDevice &sycl_device) {
int sizeDim1 = 10;
int sizeDim2 = 10;
int sizeDim3 = 10;
array<int, 3> tensorRange = {{sizeDim1, sizeDim2, sizeDim3}};
TEST_UNARY_BUILTINS(float)
/// your GPU must support double. Otherwise, disable the double test.
TEST_UNARY_BUILTINS(double)
}
void test_cxx11_tensor_builtins_sycl() {
cl::sycl::gpu_selector s;
QueueInterface queueInterface(s);
Eigen::SyclDevice sycl_device(&queueInterface);
CALL_SUBTEST(test_builtin_unary_sycl(sycl_device));
}
|
