Moved OpenclMatrix to new file

This commit is contained in:
Tong Dong Qiu 2021-11-23 09:44:18 +01:00
parent a8b1d7c9dc
commit a71a5cd8fe
8 changed files with 133 additions and 66 deletions

View File

@ -103,6 +103,7 @@ if(OPENCL_FOUND)
list (APPEND MAIN_SOURCE_FILES opm/simulators/linalg/bda/CPR.cpp)
list (APPEND MAIN_SOURCE_FILES opm/simulators/linalg/bda/opencl.cpp)
list (APPEND MAIN_SOURCE_FILES opm/simulators/linalg/bda/openclKernels.cpp)
list (APPEND MAIN_SOURCE_FILES opm/simulators/linalg/bda/OpenclMatrix.cpp)
list (APPEND MAIN_SOURCE_FILES opm/simulators/linalg/bda/openclSolverBackend.cpp)
list (APPEND MAIN_SOURCE_FILES opm/simulators/linalg/bda/openclWellContributions.cpp)
endif()
@ -259,6 +260,7 @@ list (APPEND PUBLIC_HEADER_FILES
opm/simulators/linalg/bda/ILUReorder.hpp
opm/simulators/linalg/bda/opencl.hpp
opm/simulators/linalg/bda/openclKernels.hpp
opm/simulators/linalg/bda/OpenclMatrix.hpp
opm/simulators/linalg/bda/openclSolverBackend.hpp
opm/simulators/linalg/bda/openclWellContributions.hpp
opm/simulators/linalg/bda/Matrix.hpp

View File

@ -22,13 +22,7 @@
#if HAVE_FPGA
#include <vector>
namespace Opm
{
namespace Accelerator
{
class Matrix;
}
}
#include <opm/simulators/linalg/bda/Matrix.hpp>
#endif

View File

@ -30,7 +30,9 @@
#include <opm/simulators/linalg/PropertyTree.hpp>
#include <opm/simulators/linalg/bda/BdaBridge.hpp>
#include <opm/simulators/linalg/bda/BlockedMatrix.hpp>
#include <opm/simulators/linalg/bda/CPR.hpp>
#include <opm/simulators/linalg/bda/OpenclMatrix.hpp>
namespace Opm

View File

@ -24,11 +24,11 @@
#include <dune/istl/paamg/matrixhierarchy.hh>
#include <opm/simulators/linalg/bda/BlockedMatrix.hpp>
#include <opm/simulators/linalg/bda/opencl.hpp>
#include <opm/simulators/linalg/bda/Matrix.hpp>
#include <opm/simulators/linalg/bda/OpenclMatrix.hpp>
#include <opm/simulators/linalg/bda/ILUReorder.hpp>
#include <opm/simulators/linalg/bda/opencl.hpp>
#include <opm/simulators/linalg/bda/openclKernels.hpp>
#include <opm/simulators/linalg/bda/ChowPatelIlu.hpp>
#include <opm/simulators/linalg/bda/openclSolverBackend.hpp>
@ -41,6 +41,8 @@ namespace Accelerator
template <unsigned int block_size>
class openclSolverBackend;
class BlockedMatrix;
/// This class implements a Constrained Pressure Residual (CPR) preconditioner
template <unsigned int block_size>
class CPR

View File

@ -31,29 +31,6 @@ namespace Opm
namespace Accelerator
{
void OpenclMatrix::upload(cl::CommandQueue *queue, double *vals, int *cols, int *rows) {
std::vector<cl::Event> events(3);
cl_int err = queue->enqueueWriteBuffer(nnzValues, CL_FALSE, 0, sizeof(double) * block_size * block_size * nnzbs, vals, nullptr, &events[0]);
err |= queue->enqueueWriteBuffer(colIndices, CL_FALSE, 0, sizeof(int) * nnzbs, cols, nullptr, &events[1]);
err |= queue->enqueueWriteBuffer(rowPointers, CL_FALSE, 0, sizeof(int) * (Nb + 1), rows, nullptr, &events[2]);
cl::WaitForEvents(events);
events.clear();
if (err != CL_SUCCESS) {
// enqueueWriteBuffer is C and does not throw exceptions like C++ OpenCL
OPM_THROW(std::logic_error, "OpenclMatrix OpenCL enqueueWriteBuffer error");
}
}
void OpenclMatrix::upload(cl::CommandQueue *queue, Matrix *matrix) {
upload(queue, matrix->nnzValues.data(), matrix->colIndices.data(), matrix->rowPointers.data());
}
void OpenclMatrix::upload(cl::CommandQueue *queue, BlockedMatrix *matrix) {
upload(queue, matrix->nnzValues, matrix->colIndices, matrix->rowPointers);
}
/*Sort a row of matrix elements from a CSR-format.*/
void sortRow(int *colIndices, double *data, int left, int right) {
int l = left;

View File

@ -22,45 +22,11 @@
#include <vector>
#include <opm/simulators/linalg/bda/opencl.hpp>
#include <opm/simulators/linalg/bda/BlockedMatrix.hpp>
namespace Opm
{
namespace Accelerator
{
class Matrix;
/// This struct resembles a csr matrix, only doubles are supported
/// The matrix data is stored in OpenCL Buffers
class OpenclMatrix {
public:
OpenclMatrix(cl::Context *context, int Nb_, int Mb_, int nnzbs_, unsigned int block_size_)
: Nb(Nb_),
Mb(Mb_),
nnzbs(nnzbs_),
block_size(block_size_)
{
nnzValues = cl::Buffer(*context, CL_MEM_READ_WRITE, sizeof(double) * block_size * block_size * nnzbs);
colIndices = cl::Buffer(*context, CL_MEM_READ_WRITE, sizeof(int) * nnzbs);
rowPointers = cl::Buffer(*context, CL_MEM_READ_WRITE, sizeof(int) * (Nb + 1));
}
void upload(cl::CommandQueue *queue, double *vals, int *cols, int *rows);
void upload(cl::CommandQueue *queue, Matrix *matrix);
void upload(cl::CommandQueue *queue, BlockedMatrix *matrix);
cl::Buffer nnzValues;
cl::Buffer colIndices;
cl::Buffer rowPointers;
int Nb, Mb;
int nnzbs;
unsigned int block_size;
};
/// This struct resembles a csr matrix, only doubles are supported
/// The data is stored in contiguous memory, such that they can be copied to a device in one transfer.
class Matrix {

View File

@ -0,0 +1,58 @@
/*
Copyright 2021 Equinor ASA
This file is part of the Open Porous Media project (OPM).
OPM is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
OPM is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with OPM. If not, see <http://www.gnu.org/licenses/>.
*/
#include <config.h>
#include <opm/common/OpmLog/OpmLog.hpp>
#include <opm/common/ErrorMacros.hpp>
#include <opm/simulators/linalg/bda/OpenclMatrix.hpp>
#include <opm/simulators/linalg/bda/BlockedMatrix.hpp>
#include <opm/simulators/linalg/bda/Matrix.hpp>
namespace Opm
{
namespace Accelerator
{
void OpenclMatrix::upload(cl::CommandQueue *queue, double *vals, int *cols, int *rows) {
std::vector<cl::Event> events(3);
cl_int err = queue->enqueueWriteBuffer(nnzValues, CL_FALSE, 0, sizeof(double) * block_size * block_size * nnzbs, vals, nullptr, &events[0]);
err |= queue->enqueueWriteBuffer(colIndices, CL_FALSE, 0, sizeof(int) * nnzbs, cols, nullptr, &events[1]);
err |= queue->enqueueWriteBuffer(rowPointers, CL_FALSE, 0, sizeof(int) * (Nb + 1), rows, nullptr, &events[2]);
cl::WaitForEvents(events);
events.clear();
if (err != CL_SUCCESS) {
// enqueueWriteBuffer is C and does not throw exceptions like C++ OpenCL
OPM_THROW(std::logic_error, "OpenclMatrix OpenCL enqueueWriteBuffer error");
}
}
void OpenclMatrix::upload(cl::CommandQueue *queue, Matrix *matrix) {
upload(queue, matrix->nnzValues.data(), matrix->colIndices.data(), matrix->rowPointers.data());
}
void OpenclMatrix::upload(cl::CommandQueue *queue, BlockedMatrix *matrix) {
upload(queue, matrix->nnzValues, matrix->colIndices, matrix->rowPointers);
}
} // namespace Accelerator
} // namespace Opm

View File

@ -0,0 +1,66 @@
/*
Copyright 2021 Equinor ASA
This file is part of the Open Porous Media project (OPM).
OPM is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
OPM is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with OPM. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef OPM_OPENCLMATRIX_HEADER_INCLUDED
#define OPM_OPENCLMATRIX_HEADER_INCLUDED
#include <vector>
#include <opm/simulators/linalg/bda/opencl.hpp>
namespace Opm
{
namespace Accelerator
{
class Matrix;
class BlockedMatrix;
/// This struct resembles a csr matrix, only doubles are supported
/// The matrix data is stored in OpenCL Buffers
class OpenclMatrix {
public:
OpenclMatrix(cl::Context *context, int Nb_, int Mb_, int nnzbs_, unsigned int block_size_)
: Nb(Nb_),
Mb(Mb_),
nnzbs(nnzbs_),
block_size(block_size_)
{
nnzValues = cl::Buffer(*context, CL_MEM_READ_WRITE, sizeof(double) * block_size * block_size * nnzbs);
colIndices = cl::Buffer(*context, CL_MEM_READ_WRITE, sizeof(int) * nnzbs);
rowPointers = cl::Buffer(*context, CL_MEM_READ_WRITE, sizeof(int) * (Nb + 1));
}
void upload(cl::CommandQueue *queue, double *vals, int *cols, int *rows);
void upload(cl::CommandQueue *queue, Matrix *matrix);
void upload(cl::CommandQueue *queue, BlockedMatrix *matrix);
cl::Buffer nnzValues;
cl::Buffer colIndices;
cl::Buffer rowPointers;
int Nb, Mb;
int nnzbs;
unsigned int block_size;
};
} // namespace Accelerator
} // namespace Opm
#endif // OPM_OPENCLMATRIX_HEADER_INCLUDED