OilfieldToolsNet/opm-simulators
4
0
Fork 0
mirror of https://github.com/OPM/opm-simulators.git synced 2025-02-25 18:55:30 -06:00
Code Issues Packages Projects Releases Wiki Activity

clang-format ILUKernels.cu

Browse Source
This commit is contained in:
Tobias Meyer Andersen 2024-10-02 14:59:29 +02:00
parent e82a9fa56c
commit 89df54e379
1 changed files with 61 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

61
opm/simulators/linalg/gpuistl/detail/preconditionerKernels/ILU0Kernels.cu
View File

@ -161,8 +161,9 @@ namespace
mmOverlap<InputScalar, blocksize>(&srcReorderedLowerMat[ij * scalarsInBlock],
&srcDiagonal[j * scalarsInBlock],
&srcReorderedLowerMat[ij * scalarsInBlock]);
if (copyResultToOtherMatrix){
moveBlock<blocksize, InputScalar, OutputScalar>(&srcReorderedLowerMat[ij * scalarsInBlock], &dstReorderedLowerMat[ij * scalarsInBlock]);
if (copyResultToOtherMatrix) {
moveBlock<blocksize, InputScalar, OutputScalar>(&srcReorderedLowerMat[ij * scalarsInBlock],
&dstReorderedLowerMat[ij * scalarsInBlock]);
}
// we have now accounted for an element under the diagonal and the diagonal element above it
@ -202,8 +203,9 @@ namespace
// A_jk = A_ij * A_jk
InputScalar tmp[scalarsInBlock] = {0};
mmNoOverlap<InputScalar, blocksize>(
&srcReorderedLowerMat[ij * scalarsInBlock], &srcReorderedUpperMat[jk * scalarsInBlock], tmp);
mmNoOverlap<InputScalar, blocksize>(&srcReorderedLowerMat[ij * scalarsInBlock],
&srcReorderedUpperMat[jk * scalarsInBlock],
tmp);
matrixSubtraction<InputScalar, blocksize>(ikBlockPtr, tmp);
incrementAcrossSplitStructure(ik, ikState, endOfRowILower, startOfRowIUpper);
++jk;
@ -217,12 +219,14 @@ namespace
}
}
invBlockInPlace<InputScalar, blocksize>(&srcDiagonal[reorderedIdx * scalarsInBlock]);
if (copyResultToOtherMatrix){
moveBlock<blocksize, InputScalar, OutputScalar>(&srcDiagonal[reorderedIdx * scalarsInBlock], &dstDiagonal[reorderedIdx * scalarsInBlock]);
if (copyResultToOtherMatrix) {
moveBlock<blocksize, InputScalar, OutputScalar>(&srcDiagonal[reorderedIdx * scalarsInBlock],
&dstDiagonal[reorderedIdx * scalarsInBlock]);
// also move all values above the diagonal on this row
for (int block = startOfRowIUpper; block < endOfRowIUpper; ++block) {
moveBlock<blocksize, InputScalar, OutputScalar>(&srcReorderedUpperMat[block * scalarsInBlock], &dstReorderedUpperMat[block * scalarsInBlock]);
moveBlock<blocksize, InputScalar, OutputScalar>(&srcReorderedUpperMat[block * scalarsInBlock],
&dstReorderedUpperMat[block * scalarsInBlock]);
}
}
}
@ -340,10 +344,12 @@ namespace
for (int block = nnzIdx; block < nnzIdxLim; ++block) {
const int col = colIndices[block];
mmvMixedGeneral<blocksize, MatrixScalar, LinearSolverScalar, LinearSolverScalar, LinearSolverScalar>(&mat[block * blocksize * blocksize], &v[col * blocksize], rhs);
mmvMixedGeneral<blocksize, MatrixScalar, LinearSolverScalar, LinearSolverScalar, LinearSolverScalar>(
&mat[block * blocksize * blocksize], &v[col * blocksize], rhs);
}
mvMixedGeneral<blocksize, MatrixScalar, LinearSolverScalar, LinearSolverScalar, LinearSolverScalar>(&dInv[reorderedIdx * blocksize * blocksize], rhs, &v[naturalRowIdx * blocksize]);
mvMixedGeneral<blocksize, MatrixScalar, LinearSolverScalar, LinearSolverScalar, LinearSolverScalar>(
&dInv[reorderedIdx * blocksize * blocksize], rhs, &v[naturalRowIdx * blocksize]);
}
}
} // namespace
@ -459,10 +465,11 @@ LUFactorizationSplit(InputScalar* srcReorderedLowerMat,
int rowsInLevelSet,
int thrBlockSize)
{
int threadBlockSize
= ::Opm::gpuistl::detail::getCudaRecomendedThreadBlockSize(cuLUFactorizationSplit<blocksize, InputScalar, OutputScalar, copyResultToOtherMatrix>, thrBlockSize);
int threadBlockSize = ::Opm::gpuistl::detail::getCudaRecomendedThreadBlockSize(
cuLUFactorizationSplit<blocksize, InputScalar, OutputScalar, copyResultToOtherMatrix>, thrBlockSize);
int nThreadBlocks = ::Opm::gpuistl::detail::getNumberOfBlocks(rowsInLevelSet, threadBlockSize);
cuLUFactorizationSplit<blocksize, InputScalar, OutputScalar, copyResultToOtherMatrix><<<nThreadBlocks, threadBlockSize>>>(srcReorderedLowerMat,
cuLUFactorizationSplit<blocksize, InputScalar, OutputScalar, copyResultToOtherMatrix>
<<<nThreadBlocks, threadBlockSize>>>(srcReorderedLowerMat,
lowerRowIndices,
lowerColIndices,
reorderedUpperMat,
@ -505,19 +512,25 @@ INSTANTIATE_KERNEL_WRAPPERS(double, 5);
INSTANTIATE_KERNEL_WRAPPERS(double, 6);
#define INSTANTIATE_MIXED_PRECISION_KERNEL_WRAPPERS(blocksize) \
/* double preconditioner */\
template void solveLowerLevelSetSplit<blocksize, double, double>(double*, int*, int*, int*, int, int, const double*, double*, int); \
/* float matrix, double compute preconditioner */\
template void solveLowerLevelSetSplit<blocksize, double, float>(float*, int*, int*, int*, int, int, const double*, double*, int); \
/* float preconditioner */\
template void solveLowerLevelSetSplit<blocksize, float, float>(float*, int*, int*, int*, int, int, const float*, float*, int); \
/* double preconditioner */ \
template void solveLowerLevelSetSplit<blocksize, double, double>( \
double*, int*, int*, int*, int, int, const double*, double*, int); \
/* float matrix, double compute preconditioner */ \
template void solveLowerLevelSetSplit<blocksize, double, float>( \
float*, int*, int*, int*, int, int, const double*, double*, int); \
/* float preconditioner */ \
template void solveLowerLevelSetSplit<blocksize, float, float>( \
float*, int*, int*, int*, int, int, const float*, float*, int); \
\
/* double preconditioner */\
template void solveUpperLevelSetSplit<blocksize, double, double>(double*, int*, int*, int*, int, int, const double*, double*, int); \
/* float matrix, double compute preconditioner */\
template void solveUpperLevelSetSplit<blocksize, double, float>(float*, int*, int*, int*, int, int, const float*, double*, int); \
/* float preconditioner */\
template void solveUpperLevelSetSplit<blocksize, float, float>(float*, int*, int*, int*, int, int, const float*, float*, int); \
/* double preconditioner */ \
template void solveUpperLevelSetSplit<blocksize, double, double>( \
double*, int*, int*, int*, int, int, const double*, double*, int); \
/* float matrix, double compute preconditioner */ \
template void solveUpperLevelSetSplit<blocksize, double, float>( \
float*, int*, int*, int*, int, int, const float*, double*, int); \
/* float preconditioner */ \
template void solveUpperLevelSetSplit<blocksize, float, float>( \
float*, int*, int*, int*, int, int, const float*, float*, int);
INSTANTIATE_MIXED_PRECISION_KERNEL_WRAPPERS(1);
INSTANTIATE_MIXED_PRECISION_KERNEL_WRAPPERS(2);