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
|
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.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_USE_GPU
#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>
using Eigen::Tensor;
void test_cuda_nullary() {
Tensor<std::complex<float>, 1, 0, int> in1(2);
Tensor<std::complex<float>, 1, 0, int> in2(2);
in1.setRandom();
in2.setRandom();
std::size_t float_bytes = in1.size() * sizeof(float);
std::size_t complex_bytes = in1.size() * sizeof(std::complex<float>);
std::complex<float>* d_in1;
std::complex<float>* d_in2;
float* d_out2;
cudaMalloc((void**)(&d_in1), complex_bytes);
cudaMalloc((void**)(&d_in2), complex_bytes);
cudaMalloc((void**)(&d_out2), float_bytes);
cudaMemcpy(d_in1, in1.data(), complex_bytes, cudaMemcpyHostToDevice);
cudaMemcpy(d_in2, in2.data(), complex_bytes, cudaMemcpyHostToDevice);
Eigen::GpuStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream);
Eigen::TensorMap<Eigen::Tensor<std::complex<float>, 1, 0, int>, Eigen::Aligned> gpu_in1(
d_in1, 2);
Eigen::TensorMap<Eigen::Tensor<std::complex<float>, 1, 0, int>, Eigen::Aligned> gpu_in2(
d_in2, 2);
Eigen::TensorMap<Eigen::Tensor<float, 1, 0, int>, Eigen::Aligned> gpu_out2(
d_out2, 2);
gpu_in1.device(gpu_device) = gpu_in1.constant(std::complex<float>(3.14f, 2.7f));
gpu_out2.device(gpu_device) = gpu_in2.abs();
Tensor<std::complex<float>, 1, 0, int> new1(2);
Tensor<float, 1, 0, int> new2(2);
assert(cudaMemcpyAsync(new1.data(), d_in1, complex_bytes, cudaMemcpyDeviceToHost,
gpu_device.stream()) == cudaSuccess);
assert(cudaMemcpyAsync(new2.data(), d_out2, float_bytes, cudaMemcpyDeviceToHost,
gpu_device.stream()) == cudaSuccess);
assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess);
for (int i = 0; i < 2; ++i) {
VERIFY_IS_APPROX(new1(i), std::complex<float>(3.14f, 2.7f));
VERIFY_IS_APPROX(new2(i), std::abs(in2(i)));
}
cudaFree(d_in1);
cudaFree(d_in2);
cudaFree(d_out2);
}
static void test_cuda_sum_reductions() {
Eigen::GpuStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream);
const int num_rows = internal::random<int>(1024, 5*1024);
const int num_cols = internal::random<int>(1024, 5*1024);
Tensor<std::complex<float>, 2> in(num_rows, num_cols);
in.setRandom();
Tensor<std::complex<float>, 0> full_redux;
full_redux = in.sum();
std::size_t in_bytes = in.size() * sizeof(std::complex<float>);
std::size_t out_bytes = full_redux.size() * sizeof(std::complex<float>);
std::complex<float>* gpu_in_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(in_bytes));
std::complex<float>* gpu_out_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(out_bytes));
gpu_device.memcpyHostToDevice(gpu_in_ptr, in.data(), in_bytes);
TensorMap<Tensor<std::complex<float>, 2> > in_gpu(gpu_in_ptr, num_rows, num_cols);
TensorMap<Tensor<std::complex<float>, 0> > out_gpu(gpu_out_ptr);
out_gpu.device(gpu_device) = in_gpu.sum();
Tensor<std::complex<float>, 0> full_redux_gpu;
gpu_device.memcpyDeviceToHost(full_redux_gpu.data(), gpu_out_ptr, out_bytes);
gpu_device.synchronize();
// Check that the CPU and GPU reductions return the same result.
VERIFY_IS_APPROX(full_redux(), full_redux_gpu());
gpu_device.deallocate(gpu_in_ptr);
gpu_device.deallocate(gpu_out_ptr);
}
static void test_cuda_mean_reductions() {
Eigen::GpuStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream);
const int num_rows = internal::random<int>(1024, 5*1024);
const int num_cols = internal::random<int>(1024, 5*1024);
Tensor<std::complex<float>, 2> in(num_rows, num_cols);
in.setRandom();
Tensor<std::complex<float>, 0> full_redux;
full_redux = in.mean();
std::size_t in_bytes = in.size() * sizeof(std::complex<float>);
std::size_t out_bytes = full_redux.size() * sizeof(std::complex<float>);
std::complex<float>* gpu_in_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(in_bytes));
std::complex<float>* gpu_out_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(out_bytes));
gpu_device.memcpyHostToDevice(gpu_in_ptr, in.data(), in_bytes);
TensorMap<Tensor<std::complex<float>, 2> > in_gpu(gpu_in_ptr, num_rows, num_cols);
TensorMap<Tensor<std::complex<float>, 0> > out_gpu(gpu_out_ptr);
out_gpu.device(gpu_device) = in_gpu.mean();
Tensor<std::complex<float>, 0> full_redux_gpu;
gpu_device.memcpyDeviceToHost(full_redux_gpu.data(), gpu_out_ptr, out_bytes);
gpu_device.synchronize();
// Check that the CPU and GPU reductions return the same result.
VERIFY_IS_APPROX(full_redux(), full_redux_gpu());
gpu_device.deallocate(gpu_in_ptr);
gpu_device.deallocate(gpu_out_ptr);
}
static void test_cuda_product_reductions() {
Eigen::GpuStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream);
const int num_rows = internal::random<int>(1024, 5*1024);
const int num_cols = internal::random<int>(1024, 5*1024);
Tensor<std::complex<float>, 2> in(num_rows, num_cols);
in.setRandom();
Tensor<std::complex<float>, 0> full_redux;
full_redux = in.prod();
std::size_t in_bytes = in.size() * sizeof(std::complex<float>);
std::size_t out_bytes = full_redux.size() * sizeof(std::complex<float>);
std::complex<float>* gpu_in_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(in_bytes));
std::complex<float>* gpu_out_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(out_bytes));
gpu_device.memcpyHostToDevice(gpu_in_ptr, in.data(), in_bytes);
TensorMap<Tensor<std::complex<float>, 2> > in_gpu(gpu_in_ptr, num_rows, num_cols);
TensorMap<Tensor<std::complex<float>, 0> > out_gpu(gpu_out_ptr);
out_gpu.device(gpu_device) = in_gpu.prod();
Tensor<std::complex<float>, 0> full_redux_gpu;
gpu_device.memcpyDeviceToHost(full_redux_gpu.data(), gpu_out_ptr, out_bytes);
gpu_device.synchronize();
// Check that the CPU and GPU reductions return the same result.
VERIFY_IS_APPROX(full_redux(), full_redux_gpu());
gpu_device.deallocate(gpu_in_ptr);
gpu_device.deallocate(gpu_out_ptr);
}
EIGEN_DECLARE_TEST(test_cxx11_tensor_complex)
{
CALL_SUBTEST(test_cuda_nullary());
CALL_SUBTEST(test_cuda_sum_reductions());
CALL_SUBTEST(test_cuda_mean_reductions());
CALL_SUBTEST(test_cuda_product_reductions());
}
|
