mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-21 07:19:46 +08:00
Code Issues Packages Projects Releases Wiki Activity
6aec83263d
eigen/unsupported/test/cxx11_tensor_reverse_sycl.cpp

254 lines
9.1 KiB
C++
Raw Normal View History

Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2015
// Mehdi Goli Codeplay Software Ltd.
// Ralph Potter Codeplay Software Ltd.
// Luke Iwanski Codeplay Software Ltd.
// Contact: <eigen@codeplay.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 Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
#define EIGEN_USE_SYCL
#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>
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_reverse(const Eigen::SyclDevice& sycl_device) {
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
IndexType dim1 = 2;
IndexType dim2 = 3;
IndexType dim3 = 5;
IndexType dim4 = 7;
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
array<IndexType, 4> tensorRange = {{dim1, dim2, dim3, dim4}};
Tensor<DataType, 4, DataLayout, IndexType> tensor(tensorRange);
Tensor<DataType, 4, DataLayout, IndexType> reversed_tensor(tensorRange);
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
tensor.setRandom();
array<bool, 4> dim_rev;
dim_rev[0] = false;
dim_rev[1] = true;
dim_rev[2] = true;
dim_rev[3] = false;
[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_in_data = static_cast<DataType*>(
sycl_device.allocate(tensor.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data = static_cast<DataType*>(sycl_device.allocate(
reversed_tensor.dimensions().TotalSize() * sizeof(DataType)));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +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> > in_gpu(gpu_in_data,
tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu(gpu_out_data,
tensorRange);
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +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
sycl_device.memcpyHostToDevice(
gpu_in_data, tensor.data(),
(tensor.dimensions().TotalSize()) * sizeof(DataType));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
[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
sycl_device.memcpyDeviceToHost(
reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
// Check that the CPU and GPU reductions return the same result.
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < 2; ++i) {
for (IndexType j = 0; j < 3; ++j) {
for (IndexType k = 0; k < 5; ++k) {
for (IndexType l = 0; l < 7; ++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),
reversed_tensor(i, 2 - j, 4 - k, l));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
}
}
}
}
dim_rev[0] = true;
dim_rev[1] = false;
dim_rev[2] = false;
dim_rev[3] = false;
out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
[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
sycl_device.memcpyDeviceToHost(
reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < 2; ++i) {
for (IndexType j = 0; j < 3; ++j) {
for (IndexType k = 0; k < 5; ++k) {
for (IndexType l = 0; l < 7; ++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), reversed_tensor(1 - i, j, k, l));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
}
}
}
}
dim_rev[0] = true;
dim_rev[1] = false;
dim_rev[2] = false;
dim_rev[3] = true;
out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
[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
sycl_device.memcpyDeviceToHost(
reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < 2; ++i) {
for (IndexType j = 0; j < 3; ++j) {
for (IndexType k = 0; k < 5; ++k) {
for (IndexType l = 0; l < 7; ++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),
reversed_tensor(1 - i, j, k, 6 - l));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
}
}
}
}
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
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_expr_reverse(const Eigen::SyclDevice& sycl_device,
bool LValue) {
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
IndexType dim1 = 2;
IndexType dim2 = 3;
IndexType dim3 = 5;
IndexType dim4 = 7;
array<IndexType, 4> tensorRange = {{dim1, dim2, dim3, dim4}};
Tensor<DataType, 4, DataLayout, IndexType> tensor(tensorRange);
Tensor<DataType, 4, DataLayout, IndexType> expected(tensorRange);
Tensor<DataType, 4, DataLayout, IndexType> result(tensorRange);
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
tensor.setRandom();
array<bool, 4> dim_rev;
dim_rev[0] = false;
dim_rev[1] = true;
dim_rev[2] = false;
dim_rev[3] = true;
[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_in_data = static_cast<DataType*>(
sycl_device.allocate(tensor.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data_expected = static_cast<DataType*>(sycl_device.allocate(
expected.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data_result = static_cast<DataType*>(
sycl_device.allocate(result.dimensions().TotalSize() * sizeof(DataType)));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +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> > in_gpu(gpu_in_data,
tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu_expected(
gpu_out_data_expected, tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu_result(
gpu_out_data_result, tensorRange);
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +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
sycl_device.memcpyHostToDevice(
gpu_in_data, tensor.data(),
(tensor.dimensions().TotalSize()) * sizeof(DataType));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
if (LValue) {
out_gpu_expected.reverse(dim_rev).device(sycl_device) = in_gpu;
} else {
out_gpu_expected.device(sycl_device) = in_gpu.reverse(dim_rev);
}
[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
sycl_device.memcpyDeviceToHost(
expected.data(), gpu_out_data_expected,
expected.dimensions().TotalSize() * sizeof(DataType));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
array<IndexType, 4> src_slice_dim;
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
src_slice_dim[0] = 2;
src_slice_dim[1] = 3;
src_slice_dim[2] = 1;
src_slice_dim[3] = 7;
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
array<IndexType, 4> src_slice_start;
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
src_slice_start[0] = 0;
src_slice_start[1] = 0;
src_slice_start[2] = 0;
src_slice_start[3] = 0;
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
array<IndexType, 4> dst_slice_dim = src_slice_dim;
array<IndexType, 4> dst_slice_start = src_slice_start;
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < 5; ++i) {
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
if (LValue) {
[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
out_gpu_result.slice(dst_slice_start, dst_slice_dim)
.reverse(dim_rev)
.device(sycl_device) = in_gpu.slice(src_slice_start, src_slice_dim);
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
} else {
out_gpu_result.slice(dst_slice_start, dst_slice_dim).device(sycl_device) =
in_gpu.slice(src_slice_start, src_slice_dim).reverse(dim_rev);
}
src_slice_start[2] += 1;
dst_slice_start[2] += 1;
}
[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
sycl_device.memcpyDeviceToHost(
result.data(), gpu_out_data_result,
result.dimensions().TotalSize() * sizeof(DataType));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < expected.dimension(0); ++i) {
for (IndexType j = 0; j < expected.dimension(1); ++j) {
for (IndexType k = 0; k < expected.dimension(2); ++k) {
for (IndexType l = 0; l < expected.dimension(3); ++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(result(i, j, k, l), expected(i, j, k, l));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
}
}
}
}
dst_slice_start[2] = 0;
result.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
sycl_device.memcpyHostToDevice(
gpu_out_data_result, result.data(),
(result.dimensions().TotalSize()) * sizeof(DataType));
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < 5; ++i) {
[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
if (LValue) {
out_gpu_result.slice(dst_slice_start, dst_slice_dim)
.reverse(dim_rev)
.device(sycl_device) = in_gpu.slice(dst_slice_start, dst_slice_dim);
} else {
out_gpu_result.slice(dst_slice_start, dst_slice_dim).device(sycl_device) =
in_gpu.reverse(dim_rev).slice(dst_slice_start, dst_slice_dim);
}
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
dst_slice_start[2] += 1;
}
[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
sycl_device.memcpyDeviceToHost(
result.data(), gpu_out_data_result,
result.dimensions().TotalSize() * sizeof(DataType));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
for (IndexType i = 0; i < expected.dimension(0); ++i) {
for (IndexType j = 0; j < expected.dimension(1); ++j) {
for (IndexType k = 0; k < expected.dimension(2); ++k) {
for (IndexType l = 0; l < expected.dimension(3); ++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(result(i, j, k, l), expected(i, j, k, l));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +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>
void sycl_reverse_test_per_device(const cl::sycl::device& d) {
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
QueueInterface queueInterface(d);
auto sycl_device = Eigen::SyclDevice(&queueInterface);
Reducing warnings in Sycl backend.
2017-02-01 23:29:53 +08:00
test_simple_reverse<DataType, RowMajor, int64_t>(sycl_device);
test_simple_reverse<DataType, ColMajor, int64_t>(sycl_device);
test_expr_reverse<DataType, RowMajor, int64_t>(sycl_device, false);
test_expr_reverse<DataType, ColMajor, int64_t>(sycl_device, false);
test_expr_reverse<DataType, RowMajor, int64_t>(sycl_device, true);
test_expr_reverse<DataType, ColMajor, int64_t>(sycl_device, true);
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
}
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
EIGEN_DECLARE_TEST(cxx11_tensor_reverse_sycl) {
[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
for (const auto& device : Eigen::get_sycl_supported_devices()) {
std::cout << "Running on "
<< device.get_info<cl::sycl::info::device::name>() << std::endl;
CALL_SUBTEST_1(sycl_reverse_test_per_device<short>(device));
CALL_SUBTEST_2(sycl_reverse_test_per_device<int>(device));
CALL_SUBTEST_3(sycl_reverse_test_per_device<unsigned int>(device));
#ifdef EIGEN_SYCL_DOUBLE_SUPPORT
CALL_SUBTEST_4(sycl_reverse_test_per_device<double>(device));
#endif
CALL_SUBTEST_5(sycl_reverse_test_per_device<float>(device));
Adding Tensor ReverseOp; TensorStriding; TensorConversionOp; Modifying Tensor Contractsycl to be located in any place in the expression tree.
2017-01-16 21:58:49 +08:00
}
}
Reference in New Issue Copy Permalink