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
|
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2015
// 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_reduction_sycl
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_SYCL
#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>
static void test_full_reductions_sycl(const Eigen::SyclDevice& sycl_device) {
const int num_rows = 452;
const int num_cols = 765;
array<int, 2> tensorRange = {{num_rows, num_cols}};
Tensor<float, 2> in(tensorRange);
Tensor<float, 0> full_redux;
Tensor<float, 0> full_redux_gpu;
in.setRandom();
full_redux = in.sum();
float* gpu_in_data = static_cast<float*>(sycl_device.allocate(in.dimensions().TotalSize()*sizeof(float)));
float* gpu_out_data =(float*)sycl_device.allocate(sizeof(float));
TensorMap<Tensor<float, 2> > in_gpu(gpu_in_data, tensorRange);
TensorMap<Tensor<float, 0> > out_gpu(gpu_out_data);
sycl_device.memcpyHostToDevice(gpu_in_data, in.data(),(in.dimensions().TotalSize())*sizeof(float));
out_gpu.device(sycl_device) = in_gpu.sum();
sycl_device.memcpyDeviceToHost(full_redux_gpu.data(), gpu_out_data, sizeof(float));
// Check that the CPU and GPU reductions return the same result.
VERIFY_IS_APPROX(full_redux_gpu(), full_redux());
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
static void test_first_dim_reductions_sycl(const Eigen::SyclDevice& sycl_device) {
int dim_x = 145;
int dim_y = 1;
int dim_z = 67;
array<int, 3> tensorRange = {{dim_x, dim_y, dim_z}};
Eigen::array<int, 1> red_axis;
red_axis[0] = 0;
array<int, 2> reduced_tensorRange = {{dim_y, dim_z}};
Tensor<float, 3> in(tensorRange);
Tensor<float, 2> redux(reduced_tensorRange);
Tensor<float, 2> redux_gpu(reduced_tensorRange);
in.setRandom();
redux= in.sum(red_axis);
float* gpu_in_data = static_cast<float*>(sycl_device.allocate(in.dimensions().TotalSize()*sizeof(float)));
float* gpu_out_data = static_cast<float*>(sycl_device.allocate(redux_gpu.dimensions().TotalSize()*sizeof(float)));
TensorMap<Tensor<float, 3> > in_gpu(gpu_in_data, tensorRange);
TensorMap<Tensor<float, 2> > out_gpu(gpu_out_data, reduced_tensorRange);
sycl_device.memcpyHostToDevice(gpu_in_data, in.data(),(in.dimensions().TotalSize())*sizeof(float));
out_gpu.device(sycl_device) = in_gpu.sum(red_axis);
sycl_device.memcpyDeviceToHost(redux_gpu.data(), gpu_out_data, redux_gpu.dimensions().TotalSize()*sizeof(float));
// Check that the CPU and GPU reductions return the same result.
for(int j=0; j<reduced_tensorRange[0]; j++ )
for(int k=0; k<reduced_tensorRange[1]; k++ )
VERIFY_IS_APPROX(redux_gpu(j,k), redux(j,k));
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
static void test_last_dim_reductions_sycl(const Eigen::SyclDevice &sycl_device) {
int dim_x = 567;
int dim_y = 1;
int dim_z = 47;
array<int, 3> tensorRange = {{dim_x, dim_y, dim_z}};
Eigen::array<int, 1> red_axis;
red_axis[0] = 2;
array<int, 2> reduced_tensorRange = {{dim_x, dim_y}};
Tensor<float, 3> in(tensorRange);
Tensor<float, 2> redux(reduced_tensorRange);
Tensor<float, 2> redux_gpu(reduced_tensorRange);
in.setRandom();
redux= in.sum(red_axis);
float* gpu_in_data = static_cast<float*>(sycl_device.allocate(in.dimensions().TotalSize()*sizeof(float)));
float* gpu_out_data = static_cast<float*>(sycl_device.allocate(redux_gpu.dimensions().TotalSize()*sizeof(float)));
TensorMap<Tensor<float, 3> > in_gpu(gpu_in_data, tensorRange);
TensorMap<Tensor<float, 2> > out_gpu(gpu_out_data, reduced_tensorRange);
sycl_device.memcpyHostToDevice(gpu_in_data, in.data(),(in.dimensions().TotalSize())*sizeof(float));
out_gpu.device(sycl_device) = in_gpu.sum(red_axis);
sycl_device.memcpyDeviceToHost(redux_gpu.data(), gpu_out_data, redux_gpu.dimensions().TotalSize()*sizeof(float));
// Check that the CPU and GPU reductions return the same result.
for(int j=0; j<reduced_tensorRange[0]; j++ )
for(int k=0; k<reduced_tensorRange[1]; k++ )
VERIFY_IS_APPROX(redux_gpu(j,k), redux(j,k));
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
void test_cxx11_tensor_reduction_sycl() {
cl::sycl::gpu_selector s;
Eigen::SyclDevice sycl_device(s);
CALL_SUBTEST((test_full_reductions_sycl(sycl_device)));
CALL_SUBTEST((test_first_dim_reductions_sycl(sycl_device)));
CALL_SUBTEST((test_last_dim_reductions_sycl(sycl_device)));
}
|
