mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-27 07:29:52 +08:00
Code Issues Packages Projects Releases Wiki Activity
31f796ebef
eigen/unsupported/test/cxx11_tensor_shuffling_sycl.cpp

118 lines
4.2 KiB
C++
Raw Normal View History

Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
// 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
Get rid of EIGEN_TEST_FUNC, unit tests must now be declared with EIGEN_DECLARE_TEST(mytest) { /* code */ }. This provide several advantages: - more flexibility in designing unit tests - unit tests can be glued to speed up compilation - unit tests are compiled with same predefined macros, which is a requirement for zapcc
2018-07-17 20:46:15 +08:00
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int64_t
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
#define EIGEN_USE_SYCL
#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>
using Eigen::array;
using Eigen::SyclDevice;
using Eigen::Tensor;
using Eigen::TensorMap;
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
template <typename DataType, int DataLayout, typename IndexType>
[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
2019-11-28 18:08:54 +08:00
static void test_simple_shuffling_sycl(const Eigen::SyclDevice& sycl_device) {
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
IndexType sizeDim1 = 2;
IndexType sizeDim2 = 3;
IndexType sizeDim3 = 5;
IndexType sizeDim4 = 7;
array<IndexType, 4> tensorRange = {{sizeDim1, sizeDim2, sizeDim3, sizeDim4}};
[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
2019-11-28 18:08:54 +08:00
Tensor<DataType, 4, DataLayout, IndexType> tensor(tensorRange);
Tensor<DataType, 4, DataLayout, IndexType> no_shuffle(tensorRange);
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
tensor.setRandom();
[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
2019-11-28 18:08:54 +08:00
const size_t buffSize = tensor.size() * sizeof(DataType);
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
array<IndexType, 4> shuffles;
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
shuffles[0] = 0;
shuffles[1] = 1;
shuffles[2] = 2;
shuffles[3] = 3;
[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
2019-11-28 18:08:54 +08:00
DataType* gpu_data1 = static_cast<DataType*>(sycl_device.allocate(buffSize));
DataType* gpu_data2 = static_cast<DataType*>(sycl_device.allocate(buffSize));
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
[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
2019-11-28 18:08:54 +08:00
TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu1(gpu_data1,
tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu2(gpu_data2,
tensorRange);
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
sycl_device.memcpyHostToDevice(gpu_data1, tensor.data(), buffSize);
[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
2019-11-28 18:08:54 +08:00
gpu2.device(sycl_device) = gpu1.shuffle(shuffles);
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
sycl_device.memcpyDeviceToHost(no_shuffle.data(), gpu_data2, buffSize);
Fixed race condition in the tensor_shuffling_sycl test
2016-12-20 07:34:42 +08:00
sycl_device.synchronize();
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
VERIFY_IS_EQUAL(no_shuffle.dimension(0), sizeDim1);
VERIFY_IS_EQUAL(no_shuffle.dimension(1), sizeDim2);
VERIFY_IS_EQUAL(no_shuffle.dimension(2), sizeDim3);
VERIFY_IS_EQUAL(no_shuffle.dimension(3), sizeDim4);
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < sizeDim1; ++i) {
for (IndexType j = 0; j < sizeDim2; ++j) {
for (IndexType k = 0; k < sizeDim3; ++k) {
for (IndexType l = 0; l < sizeDim4; ++l) {
[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
2019-11-28 18:08:54 +08:00
VERIFY_IS_EQUAL(tensor(i, j, k, l), no_shuffle(i, j, k, l));
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
}
}
}
}
shuffles[0] = 2;
shuffles[1] = 3;
shuffles[2] = 1;
shuffles[3] = 0;
[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
2019-11-28 18:08:54 +08:00
array<IndexType, 4> tensorrangeShuffle = {
{sizeDim3, sizeDim4, sizeDim2, sizeDim1}};
Tensor<DataType, 4, DataLayout, IndexType> shuffle(tensorrangeShuffle);
DataType* gpu_data3 = static_cast<DataType*>(sycl_device.allocate(buffSize));
TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu3(
gpu_data3, tensorrangeShuffle);
gpu3.device(sycl_device) = gpu1.shuffle(shuffles);
Fixed race condition in the tensor_shuffling_sycl test
2016-12-20 07:34:42 +08:00
sycl_device.memcpyDeviceToHost(shuffle.data(), gpu_data3, buffSize);
sycl_device.synchronize();
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
VERIFY_IS_EQUAL(shuffle.dimension(0), sizeDim3);
VERIFY_IS_EQUAL(shuffle.dimension(1), sizeDim4);
VERIFY_IS_EQUAL(shuffle.dimension(2), sizeDim2);
VERIFY_IS_EQUAL(shuffle.dimension(3), sizeDim1);
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < sizeDim1; ++i) {
for (IndexType j = 0; j < sizeDim2; ++j) {
for (IndexType k = 0; k < sizeDim3; ++k) {
for (IndexType l = 0; l < sizeDim4; ++l) {
[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
2019-11-28 18:08:54 +08:00
VERIFY_IS_EQUAL(tensor(i, j, k, l), shuffle(k, l, j, i));
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
}
}
}
}
}
[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
2019-11-28 18:08:54 +08:00
template <typename DataType, typename dev_Selector>
void sycl_shuffling_test_per_device(dev_Selector s) {
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
QueueInterface queueInterface(s);
auto sycl_device = Eigen::SyclDevice(&queueInterface);
test_simple_shuffling_sycl<DataType, RowMajor, int64_t>(sycl_device);
test_simple_shuffling_sycl<DataType, ColMajor, int64_t>(sycl_device);
}
[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
2019-11-28 18:08:54 +08:00
EIGEN_DECLARE_TEST(cxx11_tensor_shuffling_sycl) {
for (const auto& device : Eigen::get_sycl_supported_devices()) {
Adding TensorShuffling backend for sycl; adding TensorReshaping backend for sycl; cleaning up the sycl backend.
2016-11-29 23:30:42 +08:00
CALL_SUBTEST(sycl_shuffling_test_per_device<float>(device));
}
}
Reference in New Issue Copy Permalink