diff options
author | Miao Wang <miaowang@google.com> | 2017-03-06 13:45:08 -0800 |
---|---|---|
committer | Miao Wang <miaowang@google.com> | 2017-03-07 16:30:11 -0800 |
commit | 2b8756b6f1de65d3f8bffab45be6c44ceb7411fc (patch) | |
tree | 0488797fc544fe977bec6418c73445759f052482 /unsupported/test/cxx11_tensor_random_cuda.cu | |
parent | 353bba589de58014a35f8f3666b7b96353c300f8 (diff) | |
download | eigen-2b8756b6f1de65d3f8bffab45be6c44ceb7411fc.tar.gz |
Rebase Eigen to 3.3.3.
Bug: 34161771
Test: mm and RenderScript BLAS tests pass on bullhead.
Change-Id: Ia448b3202708e395fed9c783ea4323289d69dbef
Diffstat (limited to 'unsupported/test/cxx11_tensor_random_cuda.cu')
-rw-r--r-- | unsupported/test/cxx11_tensor_random_cuda.cu | 88 |
1 files changed, 88 insertions, 0 deletions
diff --git a/unsupported/test/cxx11_tensor_random_cuda.cu b/unsupported/test/cxx11_tensor_random_cuda.cu new file mode 100644 index 000000000..b3be199e1 --- /dev/null +++ b/unsupported/test/cxx11_tensor_random_cuda.cu @@ -0,0 +1,88 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2014 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_TEST_NO_COMPLEX +#define EIGEN_TEST_FUNC cxx11_tensor_random_cuda +#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int +#define EIGEN_USE_GPU + +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 +#include <cuda_fp16.h> +#endif +#include "main.h" +#include <Eigen/CXX11/Tensor> + + +void test_cuda_random_uniform() +{ + Tensor<float, 2> out(72,97); + out.setZero(); + + std::size_t out_bytes = out.size() * sizeof(float); + + float* d_out; + cudaMalloc((void**)(&d_out), out_bytes); + + Eigen::CudaStreamDevice stream; + Eigen::GpuDevice gpu_device(&stream); + + Eigen::TensorMap<Eigen::Tensor<float, 2> > gpu_out(d_out, 72,97); + + gpu_out.device(gpu_device) = gpu_out.random(); + + assert(cudaMemcpyAsync(out.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, gpu_device.stream()) == cudaSuccess); + assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess); + + // For now we just check thes code doesn't crash. + // TODO: come up with a valid test of randomness +} + + +void test_cuda_random_normal() +{ + Tensor<float, 2> out(72,97); + out.setZero(); + + std::size_t out_bytes = out.size() * sizeof(float); + + float* d_out; + cudaMalloc((void**)(&d_out), out_bytes); + + Eigen::CudaStreamDevice stream; + Eigen::GpuDevice gpu_device(&stream); + + Eigen::TensorMap<Eigen::Tensor<float, 2> > gpu_out(d_out, 72,97); + + Eigen::internal::NormalRandomGenerator<float> gen(true); + gpu_out.device(gpu_device) = gpu_out.random(gen); + + assert(cudaMemcpyAsync(out.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, gpu_device.stream()) == cudaSuccess); + assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess); +} + +static void test_complex() +{ + Tensor<std::complex<float>, 1> vec(6); + vec.setRandom(); + + // Fixme: we should check that the generated numbers follow a uniform + // distribution instead. + for (int i = 1; i < 6; ++i) { + VERIFY_IS_NOT_EQUAL(vec(i), vec(i-1)); + } +} + + +void test_cxx11_tensor_random_cuda() +{ + CALL_SUBTEST(test_cuda_random_uniform()); + CALL_SUBTEST(test_cuda_random_normal()); + CALL_SUBTEST(test_complex()); +} |