From 1a9f92e7812135017ef46ca2d29963f77ada8e55 Mon Sep 17 00:00:00 2001 From: Benoit Steiner Date: Mon, 27 Jun 2016 16:02:52 -0700 Subject: Added a test to validate the tensor scan evaluation on GPU. The test is currently disabled since the code segfaults. --- unsupported/test/cxx11_tensor_scan_cuda.cu | 77 ++++++++++++++++++++++++++++++ 1 file changed, 77 insertions(+) create mode 100644 unsupported/test/cxx11_tensor_scan_cuda.cu (limited to 'unsupported/test/cxx11_tensor_scan_cuda.cu') diff --git a/unsupported/test/cxx11_tensor_scan_cuda.cu b/unsupported/test/cxx11_tensor_scan_cuda.cu new file mode 100644 index 000000000..35e19e51c --- /dev/null +++ b/unsupported/test/cxx11_tensor_scan_cuda.cu @@ -0,0 +1,77 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2016 Benoit Steiner +// +// 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_scan_cuda +#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int +#define EIGEN_USE_GPU + + +#include "main.h" +#include + +using Eigen::Tensor; +typedef Tensor::DimensionPair DimPair; + +template +void test_cuda_cumsum(int m_size, int k_size, int n_size) +{ + std::cout << "Testing for (" << m_size << "," << k_size << "," << n_size << ")" << std::endl; + Tensor t_input(m_size, k_size, n_size); + Tensor t_result(m_size, k_size, n_size); + Tensor t_result_gpu(m_size, k_size, n_size); + + t_input.setRandom(); + + std::size_t t_input_bytes = t_input.size() * sizeof(float); + std::size_t t_result_bytes = t_result.size() * sizeof(float); + + float* d_t_input; + float* d_t_result; + + cudaMalloc((void**)(&d_t_input), t_input_bytes); + cudaMalloc((void**)(&d_t_result), t_result_bytes); + + cudaMemcpy(d_t_input, t_input.data(), t_input_bytes, cudaMemcpyHostToDevice); + + Eigen::CudaStreamDevice stream; + Eigen::GpuDevice gpu_device(&stream); + + Eigen::TensorMap > + gpu_t_input(d_t_input, Eigen::array(m_size, k_size, n_size)); + Eigen::TensorMap > + gpu_t_result(d_t_result, Eigen::array(m_size, k_size, n_size)); + + gpu_t_result.device(gpu_device) = gpu_t_input.cumsum(1); + t_result = t_input.cumsum(1); + + cudaMemcpy(t_result_gpu.data(), d_t_result, t_result_bytes, cudaMemcpyDeviceToHost); + for (size_t i = 0; i < t_result.size(); i++) { + if (fabs(t_result(i) - t_result_gpu(i)) < 1e-4f) { + continue; + } + if (Eigen::internal::isApprox(t_result(i), t_result_gpu(i), 1e-4f)) { + continue; + } + std::cout << "mismatch detected at index " << i << ": " << t_result(i) + << " vs " << t_result_gpu(i) << std::endl; + assert(false); + } + + cudaFree((void*)d_t_input); + cudaFree((void*)d_t_result); +} + + +void test_cxx11_tensor_scan_cuda() +{ + CALL_SUBTEST_1(test_cuda_cumsum(128, 128, 128)); + CALL_SUBTEST_2(test_cuda_cumsum(128, 128, 128)); +} -- cgit v1.2.3 From 72a4d49315ee25c64e8e457735c494e77863a3e0 Mon Sep 17 00:00:00 2001 From: Gael Guennebaud Date: Thu, 1 Sep 2016 13:39:33 +0200 Subject: Fix compilation with CUDA 8 --- unsupported/test/cxx11_tensor_argmax_cuda.cu | 1 + unsupported/test/cxx11_tensor_cast_float16_cuda.cu | 2 +- unsupported/test/cxx11_tensor_contract_cuda.cu | 2 +- unsupported/test/cxx11_tensor_cuda.cu | 2 +- unsupported/test/cxx11_tensor_device.cu | 2 +- unsupported/test/cxx11_tensor_random_cuda.cu | 1 + unsupported/test/cxx11_tensor_reduction_cuda.cu | 1 + unsupported/test/cxx11_tensor_scan_cuda.cu | 2 +- 8 files changed, 8 insertions(+), 5 deletions(-) (limited to 'unsupported/test/cxx11_tensor_scan_cuda.cu') diff --git a/unsupported/test/cxx11_tensor_argmax_cuda.cu b/unsupported/test/cxx11_tensor_argmax_cuda.cu index 41ccbe974..c3f4bfa53 100644 --- a/unsupported/test/cxx11_tensor_argmax_cuda.cu +++ b/unsupported/test/cxx11_tensor_argmax_cuda.cu @@ -12,6 +12,7 @@ #define EIGEN_TEST_FUNC cxx11_tensor_cuda #define EIGEN_USE_GPU +#include #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_cast_float16_cuda.cu b/unsupported/test/cxx11_tensor_cast_float16_cuda.cu index f22b99de8..559aae269 100644 --- a/unsupported/test/cxx11_tensor_cast_float16_cuda.cu +++ b/unsupported/test/cxx11_tensor_cast_float16_cuda.cu @@ -13,7 +13,7 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU - +#include #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_contract_cuda.cu b/unsupported/test/cxx11_tensor_contract_cuda.cu index 98ac180ef..5282dc3b9 100644 --- a/unsupported/test/cxx11_tensor_contract_cuda.cu +++ b/unsupported/test/cxx11_tensor_contract_cuda.cu @@ -14,7 +14,7 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU - +#include #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_cuda.cu b/unsupported/test/cxx11_tensor_cuda.cu index 284b46803..29ea76857 100644 --- a/unsupported/test/cxx11_tensor_cuda.cu +++ b/unsupported/test/cxx11_tensor_cuda.cu @@ -13,7 +13,7 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU - +#include #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_device.cu b/unsupported/test/cxx11_tensor_device.cu index b6ca54d93..bbda95906 100644 --- a/unsupported/test/cxx11_tensor_device.cu +++ b/unsupported/test/cxx11_tensor_device.cu @@ -13,7 +13,7 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU - +#include #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_random_cuda.cu b/unsupported/test/cxx11_tensor_random_cuda.cu index fa1a46732..32f71e808 100644 --- a/unsupported/test/cxx11_tensor_random_cuda.cu +++ b/unsupported/test/cxx11_tensor_random_cuda.cu @@ -13,6 +13,7 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU +#include #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_reduction_cuda.cu b/unsupported/test/cxx11_tensor_reduction_cuda.cu index 6d8f01c02..9b5e60e99 100644 --- a/unsupported/test/cxx11_tensor_reduction_cuda.cu +++ b/unsupported/test/cxx11_tensor_reduction_cuda.cu @@ -12,6 +12,7 @@ #define EIGEN_TEST_FUNC cxx11_tensor_reduction_cuda #define EIGEN_USE_GPU +#include #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_scan_cuda.cu b/unsupported/test/cxx11_tensor_scan_cuda.cu index 35e19e51c..c8a99f62b 100644 --- a/unsupported/test/cxx11_tensor_scan_cuda.cu +++ b/unsupported/test/cxx11_tensor_scan_cuda.cu @@ -13,7 +13,7 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU - +#include #include "main.h" #include -- cgit v1.2.3 From dabc81751fe2a1eed16967e140e65558e52b5a5f Mon Sep 17 00:00:00 2001 From: Gael Guennebaud Date: Mon, 5 Sep 2016 17:14:20 +0200 Subject: Fix compilation when cuda_fp16.h does not exist. --- test/cuda_basic.cu | 2 ++ unsupported/test/cxx11_tensor_argmax_cuda.cu | 2 ++ unsupported/test/cxx11_tensor_cast_float16_cuda.cu | 2 ++ unsupported/test/cxx11_tensor_contract_cuda.cu | 2 ++ unsupported/test/cxx11_tensor_cuda.cu | 2 ++ unsupported/test/cxx11_tensor_device.cu | 2 ++ unsupported/test/cxx11_tensor_of_float16_cuda.cu | 2 ++ unsupported/test/cxx11_tensor_random_cuda.cu | 2 ++ unsupported/test/cxx11_tensor_reduction_cuda.cu | 2 ++ unsupported/test/cxx11_tensor_scan_cuda.cu | 2 ++ 10 files changed, 20 insertions(+) (limited to 'unsupported/test/cxx11_tensor_scan_cuda.cu') diff --git a/test/cuda_basic.cu b/test/cuda_basic.cu index 3cf37d221..0a515ce7b 100644 --- a/test/cuda_basic.cu +++ b/test/cuda_basic.cu @@ -20,7 +20,9 @@ #include #include +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include "cuda_common.h" diff --git a/unsupported/test/cxx11_tensor_argmax_cuda.cu b/unsupported/test/cxx11_tensor_argmax_cuda.cu index c3f4bfa53..6fe8982f2 100644 --- a/unsupported/test/cxx11_tensor_argmax_cuda.cu +++ b/unsupported/test/cxx11_tensor_argmax_cuda.cu @@ -12,7 +12,9 @@ #define EIGEN_TEST_FUNC cxx11_tensor_cuda #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_cast_float16_cuda.cu b/unsupported/test/cxx11_tensor_cast_float16_cuda.cu index 559aae269..88c233994 100644 --- a/unsupported/test/cxx11_tensor_cast_float16_cuda.cu +++ b/unsupported/test/cxx11_tensor_cast_float16_cuda.cu @@ -13,7 +13,9 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_contract_cuda.cu b/unsupported/test/cxx11_tensor_contract_cuda.cu index 5282dc3b9..767e9c678 100644 --- a/unsupported/test/cxx11_tensor_contract_cuda.cu +++ b/unsupported/test/cxx11_tensor_contract_cuda.cu @@ -14,7 +14,9 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_cuda.cu b/unsupported/test/cxx11_tensor_cuda.cu index f1e91a879..bf216587a 100644 --- a/unsupported/test/cxx11_tensor_cuda.cu +++ b/unsupported/test/cxx11_tensor_cuda.cu @@ -12,7 +12,9 @@ #define EIGEN_TEST_FUNC cxx11_tensor_cuda #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_device.cu b/unsupported/test/cxx11_tensor_device.cu index bbda95906..fde20ddf2 100644 --- a/unsupported/test/cxx11_tensor_device.cu +++ b/unsupported/test/cxx11_tensor_device.cu @@ -13,7 +13,9 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_of_float16_cuda.cu b/unsupported/test/cxx11_tensor_of_float16_cuda.cu index a6375d34a..cbf401c86 100644 --- a/unsupported/test/cxx11_tensor_of_float16_cuda.cu +++ b/unsupported/test/cxx11_tensor_of_float16_cuda.cu @@ -13,7 +13,9 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_random_cuda.cu b/unsupported/test/cxx11_tensor_random_cuda.cu index 32f71e808..b3be199e1 100644 --- a/unsupported/test/cxx11_tensor_random_cuda.cu +++ b/unsupported/test/cxx11_tensor_random_cuda.cu @@ -13,7 +13,9 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_reduction_cuda.cu b/unsupported/test/cxx11_tensor_reduction_cuda.cu index 9b5e60e99..80e151d43 100644 --- a/unsupported/test/cxx11_tensor_reduction_cuda.cu +++ b/unsupported/test/cxx11_tensor_reduction_cuda.cu @@ -12,7 +12,9 @@ #define EIGEN_TEST_FUNC cxx11_tensor_reduction_cuda #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include diff --git a/unsupported/test/cxx11_tensor_scan_cuda.cu b/unsupported/test/cxx11_tensor_scan_cuda.cu index c8a99f62b..761d11fd1 100644 --- a/unsupported/test/cxx11_tensor_scan_cuda.cu +++ b/unsupported/test/cxx11_tensor_scan_cuda.cu @@ -13,7 +13,9 @@ #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int #define EIGEN_USE_GPU +#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 #include +#endif #include "main.h" #include -- cgit v1.2.3