1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
5 //
6 // This Source Code Form is subject to the terms of the Mozilla
7 // Public License v. 2.0. If a copy of the MPL was not distributed
8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
9 
10 #define EIGEN_TEST_NO_LONGDOUBLE
11 #define EIGEN_TEST_NO_COMPLEX
12 #define EIGEN_TEST_FUNC cxx11_tensor_random_cuda
13 #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
14 #define EIGEN_USE_GPU
15 
16 #include "main.h"
17 #include <Eigen/CXX11/Tensor>
18 
19 
test_cuda_random_uniform()20 void test_cuda_random_uniform()
21 {
22   Tensor<float, 2> out(72,97);
23   out.setZero();
24 
25   std::size_t out_bytes = out.size() * sizeof(float);
26 
27   float* d_out;
28   cudaMalloc((void**)(&d_out), out_bytes);
29 
30   Eigen::CudaStreamDevice stream;
31   Eigen::GpuDevice gpu_device(&stream);
32 
33   Eigen::TensorMap<Eigen::Tensor<float, 2> > gpu_out(d_out, 72,97);
34 
35   gpu_out.device(gpu_device) = gpu_out.random();
36 
37   assert(cudaMemcpyAsync(out.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, gpu_device.stream()) == cudaSuccess);
38   assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess);
39 
40   // For now we just check thes code doesn't crash.
41   // TODO: come up with a valid test of randomness
42 }
43 
44 
test_cuda_random_normal()45 void test_cuda_random_normal()
46 {
47   Tensor<float, 2> out(72,97);
48   out.setZero();
49 
50   std::size_t out_bytes = out.size() * sizeof(float);
51 
52   float* d_out;
53   cudaMalloc((void**)(&d_out), out_bytes);
54 
55   Eigen::CudaStreamDevice stream;
56   Eigen::GpuDevice gpu_device(&stream);
57 
58   Eigen::TensorMap<Eigen::Tensor<float, 2> > gpu_out(d_out, 72,97);
59 
60   Eigen::internal::NormalRandomGenerator<float> gen(true);
61   gpu_out.device(gpu_device) = gpu_out.random(gen);
62 
63   assert(cudaMemcpyAsync(out.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, gpu_device.stream()) == cudaSuccess);
64   assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess);
65 }
66 
test_complex()67 static void test_complex()
68 {
69   Tensor<std::complex<float>, 1> vec(6);
70   vec.setRandom();
71 
72   // Fixme: we should check that the generated numbers follow a uniform
73   // distribution instead.
74   for (int i = 1; i < 6; ++i) {
75     VERIFY_IS_NOT_EQUAL(vec(i), vec(i-1));
76   }
77 }
78 
79 
test_cxx11_tensor_random_cuda()80 void test_cxx11_tensor_random_cuda()
81 {
82   CALL_SUBTEST(test_cuda_random_uniform());
83   CALL_SUBTEST(test_cuda_random_normal());
84   CALL_SUBTEST(test_complex());
85 }
86