diff options
author | Mehdi Goli <mehdi.goli@codeplay.com> | 2019-11-28 10:08:54 +0000 |
---|---|---|
committer | Mehdi Goli <mehdi.goli@codeplay.com> | 2019-11-28 10:08:54 +0000 |
commit | 00f32752f7d0b193c6788691c3cf0b76457a044d (patch) | |
tree | 792e46110f0751ea8802fa9d403d1472d5977ac3 /unsupported/test/cxx11_tensor_image_op_sycl.cpp | |
parent | ea51a9eace7e4f0ea839e61eb2df85ccfb94aee8 (diff) |
[SYCL] Rebasing the SYCL support branch on top of the Einge upstream master branch.
* Unifying all loadLocalTile from lhs and rhs to an extract_block function.
* Adding get_tensor operation which was missing in TensorContractionMapper.
* Adding the -D method missing from cmake for Disable_Skinny Contraction operation.
* Wrapping all the indices in TensorScanSycl into Scan parameter struct.
* Fixing typo in Device SYCL
* Unifying load to private register for tall/skinny no shared
* Unifying load to vector tile for tensor-vector/vector-tensor operation
* Removing all the LHS/RHS class for extracting data from global
* Removing Outputfunction from TensorContractionSkinnyNoshared.
* Combining the local memory version of tall/skinny and normal tensor contraction into one kernel.
* Combining the no-local memory version of tall/skinny and normal tensor contraction into one kernel.
* Combining General Tensor-Vector and VectorTensor contraction into one kernel.
* Making double buffering optional for Tensor contraction when local memory is version is used.
* Modifying benchmark to accept custom Reduction Sizes
* Disabling AVX optimization for SYCL backend on the host to allow SSE optimization to the host
* Adding Test for SYCL
* Modifying SYCL CMake
Diffstat (limited to 'unsupported/test/cxx11_tensor_image_op_sycl.cpp')
-rw-r--r-- | unsupported/test/cxx11_tensor_image_op_sycl.cpp | 103 |
1 files changed, 103 insertions, 0 deletions
diff --git a/unsupported/test/cxx11_tensor_image_op_sycl.cpp b/unsupported/test/cxx11_tensor_image_op_sycl.cpp new file mode 100644 index 000000000..db1c0206e --- /dev/null +++ b/unsupported/test/cxx11_tensor_image_op_sycl.cpp @@ -0,0 +1,103 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2016 +// Mehdi Goli Codeplay Software Ltd. +// Ralph Potter Codeplay Software Ltd. +// Luke Iwanski Codeplay Software Ltd. +// Contact: <eigen@codeplay.com> +// 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_DEFAULT_DENSE_INDEX_TYPE int64_t +#define EIGEN_USE_SYCL + +#include "main.h" +#include <unsupported/Eigen/CXX11/Tensor> + +using Eigen::array; +using Eigen::SyclDevice; +using Eigen::Tensor; +using Eigen::TensorMap; + +using Eigen::Tensor; +using Eigen::RowMajor; +template <typename DataType, int DataLayout, typename IndexType> +static void test_image_op_sycl(const Eigen::SyclDevice &sycl_device) +{ + IndexType sizeDim1 = 245; + IndexType sizeDim2 = 343; + IndexType sizeDim3 = 577; + + array<IndexType, 3> input_range ={{sizeDim1, sizeDim2, sizeDim3}}; + array<IndexType, 3> slice_range ={{sizeDim1-1, sizeDim2, sizeDim3}}; + + Tensor<DataType, 3,DataLayout, IndexType> tensor1(input_range); + Tensor<DataType, 3,DataLayout, IndexType> tensor2(input_range); + Tensor<DataType, 3, DataLayout, IndexType> tensor3(slice_range); + Tensor<DataType, 3, DataLayout, IndexType> tensor3_cpu(slice_range); + + + + typedef Eigen::DSizes<IndexType, 3> Index3; + Index3 strides1(1L,1L, 1L); + Index3 indicesStart1(1L, 0L, 0L); + Index3 indicesStop1(sizeDim1, sizeDim2, sizeDim3); + + Index3 strides2(1L,1L, 1L); + Index3 indicesStart2(0L, 0L, 0L); + Index3 indicesStop2(sizeDim1-1, sizeDim2, sizeDim3); + Eigen::DSizes<IndexType, 3> sizes(sizeDim1-1,sizeDim2,sizeDim3); + + tensor1.setRandom(); + tensor2.setRandom(); + + + DataType* gpu_data1 = static_cast<DataType*>(sycl_device.allocate(tensor1.size()*sizeof(DataType))); + DataType* gpu_data2 = static_cast<DataType*>(sycl_device.allocate(tensor2.size()*sizeof(DataType))); + DataType* gpu_data3 = static_cast<DataType*>(sycl_device.allocate(tensor3.size()*sizeof(DataType))); + + TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu1(gpu_data1, input_range); + TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu2(gpu_data2, input_range); + TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu3(gpu_data3, slice_range); + + sycl_device.memcpyHostToDevice(gpu_data1, tensor1.data(),(tensor1.size())*sizeof(DataType)); + sycl_device.memcpyHostToDevice(gpu_data2, tensor2.data(),(tensor2.size())*sizeof(DataType)); + gpu3.device(sycl_device)= gpu1.slice(indicesStart1, sizes) - gpu2.slice(indicesStart2, sizes); + sycl_device.memcpyDeviceToHost(tensor3.data(), gpu_data3,(tensor3.size())*sizeof(DataType)); + + tensor3_cpu = tensor1.stridedSlice(indicesStart1,indicesStop1,strides1) - tensor2.stridedSlice(indicesStart2,indicesStop2,strides2); + + + for (IndexType i = 0; i <slice_range[0] ; ++i) { + for (IndexType j = 0; j < slice_range[1]; ++j) { + for (IndexType k = 0; k < slice_range[2]; ++k) { + VERIFY_IS_EQUAL(tensor3_cpu(i,j,k), tensor3(i,j,k)); + } + } + } + sycl_device.deallocate(gpu_data1); + sycl_device.deallocate(gpu_data2); + sycl_device.deallocate(gpu_data3); +} + + +template<typename DataType, typename dev_Selector> void sycl_computing_test_per_device(dev_Selector s){ + QueueInterface queueInterface(s); + auto sycl_device = Eigen::SyclDevice(&queueInterface); + test_image_op_sycl<DataType, RowMajor, int64_t>(sycl_device); +} + +EIGEN_DECLARE_TEST(cxx11_tensor_image_op_sycl) { + for (const auto& device :Eigen::get_sycl_supported_devices()) { + CALL_SUBTEST(sycl_computing_test_per_device<float>(device)); +#ifdef EIGEN_SYCL_DOUBLE_SUPPORT + CALL_SUBTEST(sycl_computing_test_per_device<double>(device)); +#endif + } +} |