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
|
// 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_broadcast_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;
template <typename DataType, int DataLayout>
static void test_broadcast_sycl_fixed(const Eigen::SyclDevice &sycl_device){
// BROADCAST test:
int inDim1=2;
int inDim2=3;
int inDim3=5;
int inDim4=7;
int bDim1=2;
int bDim2=3;
int bDim3=1;
int bDim4=4;
array<int, 4> in_range = {{inDim1, inDim2, inDim3, inDim4}};
array<int, 4> broadcasts = {{bDim1, bDim2, bDim3, bDim4}};
array<int, 4> out_range; // = in_range * broadcasts
for (size_t i = 0; i < out_range.size(); ++i)
out_range[i] = in_range[i] * broadcasts[i];
Tensor<DataType, 4, DataLayout> input(in_range);
Tensor<DataType, 4, DataLayout> out(out_range);
for (size_t i = 0; i < in_range.size(); ++i)
VERIFY_IS_EQUAL(out.dimension(i), out_range[i]);
for (int i = 0; i < input.size(); ++i)
input(i) = static_cast<DataType>(i);
DataType * gpu_in_data = static_cast<DataType*>(sycl_device.allocate(input.dimensions().TotalSize()*sizeof(DataType)));
DataType * gpu_out_data = static_cast<DataType*>(sycl_device.allocate(out.dimensions().TotalSize()*sizeof(DataType)));
TensorMap<TensorFixedSize<DataType, Sizes<2, 3, 5, 7>, DataLayout>> gpu_in(gpu_in_data, in_range);
TensorMap<Tensor<DataType, 4, DataLayout>> gpu_out(gpu_out_data, out_range);
sycl_device.memcpyHostToDevice(gpu_in_data, input.data(),(input.dimensions().TotalSize())*sizeof(DataType));
gpu_out.device(sycl_device) = gpu_in.broadcast(broadcasts);
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(DataType));
for (int i = 0; i < inDim1*bDim1; ++i) {
for (int j = 0; j < inDim2*bDim2; ++j) {
for (int k = 0; k < inDim3*bDim3; ++k) {
for (int l = 0; l < inDim4*bDim4; ++l) {
VERIFY_IS_APPROX(input(i%2,j%3,k%5,l%7), out(i,j,k,l));
}
}
}
}
printf("Broadcast Test with fixed size Passed\n");
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
template <typename DataType, int DataLayout>
static void test_broadcast_sycl(const Eigen::SyclDevice &sycl_device){
// BROADCAST test:
int inDim1=2;
int inDim2=3;
int inDim3=5;
int inDim4=7;
int bDim1=2;
int bDim2=3;
int bDim3=1;
int bDim4=4;
array<int, 4> in_range = {{inDim1, inDim2, inDim3, inDim4}};
array<int, 4> broadcasts = {{bDim1, bDim2, bDim3, bDim4}};
array<int, 4> out_range; // = in_range * broadcasts
for (size_t i = 0; i < out_range.size(); ++i)
out_range[i] = in_range[i] * broadcasts[i];
Tensor<DataType, 4, DataLayout> input(in_range);
Tensor<DataType, 4, DataLayout> out(out_range);
for (size_t i = 0; i < in_range.size(); ++i)
VERIFY_IS_EQUAL(out.dimension(i), out_range[i]);
for (int i = 0; i < input.size(); ++i)
input(i) = static_cast<DataType>(i);
DataType * gpu_in_data = static_cast<DataType*>(sycl_device.allocate(input.dimensions().TotalSize()*sizeof(DataType)));
DataType * gpu_out_data = static_cast<DataType*>(sycl_device.allocate(out.dimensions().TotalSize()*sizeof(DataType)));
TensorMap<Tensor<DataType, 4, DataLayout>> gpu_in(gpu_in_data, in_range);
TensorMap<Tensor<DataType, 4, DataLayout>> gpu_out(gpu_out_data, out_range);
sycl_device.memcpyHostToDevice(gpu_in_data, input.data(),(input.dimensions().TotalSize())*sizeof(DataType));
gpu_out.device(sycl_device) = gpu_in.broadcast(broadcasts);
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(DataType));
for (int i = 0; i < inDim1*bDim1; ++i) {
for (int j = 0; j < inDim2*bDim2; ++j) {
for (int k = 0; k < inDim3*bDim3; ++k) {
for (int l = 0; l < inDim4*bDim4; ++l) {
VERIFY_IS_APPROX(input(i%inDim1,j%inDim2,k%inDim3,l%inDim4), out(i,j,k,l));
}
}
}
}
printf("Broadcast Test Passed\n");
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
template<typename DataType, typename dev_Selector> void sycl_broadcast_test_per_device(dev_Selector s){
QueueInterface queueInterface(s);
auto sycl_device = Eigen::SyclDevice(&queueInterface);
test_broadcast_sycl_fixed<DataType, RowMajor>(sycl_device);
test_broadcast_sycl<DataType, RowMajor>(sycl_device);
test_broadcast_sycl_fixed<DataType, ColMajor>(sycl_device);
test_broadcast_sycl<DataType, ColMajor>(sycl_device);
}
void test_cxx11_tensor_broadcast_sycl() {
printf("Test on GPU: OpenCL\n");
CALL_SUBTEST(sycl_broadcast_test_per_device<float>((cl::sycl::gpu_selector())));
printf("repeating the test on CPU: OpenCL\n");
CALL_SUBTEST(sycl_broadcast_test_per_device<float>((cl::sycl::cpu_selector())));
printf("repeating the test on CPU: HOST\n");
CALL_SUBTEST(sycl_broadcast_test_per_device<float>((cl::sycl::host_selector())));
printf("Test Passed******************\n" );
}
|
