GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-05-01 00:04:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Added regression test for float16

Browse Source
This commit is contained in:
Benoit Steiner 2016-02-19 06:23:28 +00:00
parent 7151bd8768
commit f36c0c2c65
1 changed files with 60 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
unsupported/test/cxx11_tensor_of_float16_cuda.cu Normal file
View File

@ -0,0 +1,60 @@
// 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_TEST_NO_COMPLEX
#define EIGEN_TEST_FUNC cxx11_tensor_of_float16_cuda
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_GPU
#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>
using Eigen::Tensor;
void test_cuda_conversion() {
Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream);
int num_elem = 101;
float* d_float = (float*)gpu_device.allocate(num_elem * sizeof(float));
half* d_half = (half*)gpu_device.allocate(num_elem * sizeof(half));
float* d_conv = (float*)gpu_device.allocate(num_elem * sizeof(float));
Eigen::TensorMap<Eigen::Tensor<float, 1>, Eigen::Aligned> gpu_float(
d_float, num_elem);
Eigen::TensorMap<Eigen::Tensor<half, 1>, Eigen::Aligned> gpu_half(
d_half, num_elem);
Eigen::TensorMap<Eigen::Tensor<float, 1>, Eigen::Aligned> gpu_conv(
d_conv, num_elem);
gpu_float.device(gpu_device) = gpu_float.random();
gpu_half.device(gpu_device) = gpu_float.cast<half>();
gpu_conv.device(gpu_device) = gpu_half.cast<float>();
Tensor<float, 1> initial(num_elem);
Tensor<float, 1> final(num_elem);
gpu_device.memcpyDeviceToHost(initial.data(), d_float, num_elem*sizeof(float));
gpu_device.memcpyDeviceToHost(final.data(), d_conv, num_elem*sizeof(float));
for (int i = 0; i < num_elem; ++i) {
VERIFY_IS_APPROX(initial(i), final(i));
}
gpu_device.deallocate(d_float);
gpu_device.deallocate(d_half);
gpu_device.deallocate(d_conv);
}
void test_cxx11_tensor_of_float16_cuda()
{
CALL_SUBTEST_1(test_cuda_conversion());
}