Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Cuda kernels for Lower triangular solve #336

Merged
merged 19 commits into from
Sep 6, 2019
Merged

Cuda kernels for Lower triangular solve #336

Show file tree
Hide file tree
Changes from 16 commits
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
8a1255b
Add the cuda kernels and tests.
pratikvn Aug 1, 2019
5016b64
Update the tests.
pratikvn Aug 2, 2019
22eb4b7
Fix the const qualif prob and cusparse handle issue.
pratikvn Aug 5, 2019
5ca8d60
Fix a bug in matrix_gen, fix bugs in trs kernels for CUSPARSE.
pratikvn Aug 6, 2019
3de1f63
Add if defines to use csrsmX for solves.
pratikvn Aug 10, 2019
dd4a27c
Rename the CUDA Trs to LowerTrs.
pratikvn Aug 12, 2019
f6167c8
Update with the new tr gen for cuda test.
pratikvn Aug 14, 2019
55e9b02
Fix the include typo and multi-rhs cusparse thanks to @yhmtsai.
pratikvn Aug 19, 2019
f956c3d
Remove the rhs from the generate kernel and add num rhs as a param.
pratikvn Aug 22, 2019
f9ee985
Remove the now redundant TRS_FACTORY gen and use LIN_OP_FACTORY.
pratikvn Aug 23, 2019
c371d9b
Minor random lmat usage in the tests.
pratikvn Aug 23, 2019
d89c1a0
Add a clear kernel to clear the analysis data.
pratikvn Aug 23, 2019
b03c57f
Fix the multiple rhs problem.
pratikvn Aug 27, 2019
02d2787
Review update.
pratikvn Aug 27, 2019
d01e275
Fix the cusparse csrsm_solve issue.
pratikvn Aug 29, 2019
d8d1bb8
Move the cusparse struct to LowerTrs class.
pratikvn Sep 2, 2019
ec12e2f
Move the SolveStruct creation and destr to the class constr, destr.
pratikvn Sep 4, 2019
90b6f0a
Move the LowerTrs core tests back to `core/test`. Review updates.
pratikvn Sep 5, 2019
050c727
Review updates.
pratikvn Sep 6, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions core/device_hooks/common_kernels.inc.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,6 +185,12 @@ GKO_INSTANTIATE_FOR_EACH_VALUE_TYPE(GKO_DECLARE_CG_STEP_2_KERNEL);
namespace lower_trs {


GKO_DECLARE_LOWER_TRS_CHECK_TRANSPOSABILITY_KERNEL()
GKO_NOT_COMPILED(GKO_HOOK_MODULE);

GKO_DECLARE_LOWER_TRS_INIT_STRUCT_KERNEL()
GKO_NOT_COMPILED(GKO_HOOK_MODULE);

template <typename ValueType, typename IndexType>
GKO_DECLARE_LOWER_TRS_GENERATE_KERNEL(ValueType, IndexType)
GKO_NOT_COMPILED(GKO_HOOK_MODULE);
Expand Down
34 changes: 27 additions & 7 deletions core/solver/lower_trs.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,23 +48,31 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

namespace gko {
namespace solver {


namespace lower_trs {


GKO_REGISTER_OPERATION(generate, lower_trs::generate);
GKO_REGISTER_OPERATION(init_struct, lower_trs::init_struct);
GKO_REGISTER_OPERATION(perform_transpose, lower_trs::perform_transpose);
GKO_REGISTER_OPERATION(solve, lower_trs::solve);


} // namespace lower_trs


template <typename ValueType, typename IndexType>
void LowerTrs<ValueType, IndexType>::init_trs_solve_struct()
{
this->get_executor()->run(lower_trs::make_init_struct(this->solve_struct_));
}


template <typename ValueType, typename IndexType>
void LowerTrs<ValueType, IndexType>::generate()
{
this->get_executor()->run(
lower_trs::make_generate(gko::lend(system_matrix_), gko::lend(b_)));
this->get_executor()->run(lower_trs::make_generate(
gko::lend(system_matrix_), this->solve_struct_.get(),
parameters_.num_rhs));
}


Expand All @@ -76,9 +84,21 @@ void LowerTrs<ValueType, IndexType>::apply_impl(const LinOp *b, LinOp *x) const

auto dense_b = as<const Vector>(b);
auto dense_x = as<Vector>(x);

exec->run(
lower_trs::make_solve(gko::lend(system_matrix_), dense_b, dense_x));
bool transposability = false;
std::shared_ptr<Vector> trans_b;
std::shared_ptr<Vector> trans_x;
this->get_executor()->run(
lower_trs::make_perform_transpose(transposability));
if (transposability) {
trans_b = Vector::create(exec, gko::transpose(dense_b->get_size()));
trans_x = Vector::create(exec, gko::transpose(dense_x->get_size()));
} else {
trans_b = Vector::create(exec);
trans_x = Vector::create(exec);
}
exec->run(lower_trs::make_solve(
gko::lend(system_matrix_), this->solve_struct_.get(),
gko::lend(trans_b), gko::lend(trans_x), dense_b, dense_x));
thoasm marked this conversation as resolved.
Show resolved Hide resolved
}


Expand Down
24 changes: 20 additions & 4 deletions core/solver/lower_trs_kernels.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,26 +40,42 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <ginkgo/core/base/executor.hpp>
#include <ginkgo/core/matrix/csr.hpp>
#include <ginkgo/core/matrix/dense.hpp>
#include <ginkgo/core/solver/lower_trs.hpp>


namespace gko {
namespace kernels {
namespace lower_trs {


#define GKO_DECLARE_LOWER_TRS_CHECK_TRANSPOSABILITY_KERNEL() \
void perform_transpose(std::shared_ptr<const DefaultExecutor> exec, \
pratikvn marked this conversation as resolved.
Show resolved Hide resolved
bool &transposability)


#define GKO_DECLARE_LOWER_TRS_INIT_STRUCT_KERNEL() \
void init_struct(std::shared_ptr<const DefaultExecutor> exec, \
std::shared_ptr<gko::solver::SolveStruct> &solve_struct)
thoasm marked this conversation as resolved.
Show resolved Hide resolved


#define GKO_DECLARE_LOWER_TRS_GENERATE_KERNEL(_vtype, _itype) \
void generate(std::shared_ptr<const DefaultExecutor> exec, \
const matrix::Csr<_vtype, _itype> *matrix, \
const matrix::Dense<_vtype> *b)
gko::solver::SolveStruct *solve_struct, \
const gko::size_type num_rhs)


#define GKO_DECLARE_LOWER_TRS_SOLVE_KERNEL(_vtype, _itype) \
void solve(std::shared_ptr<const DefaultExecutor> exec, \
const matrix::Csr<_vtype, _itype> *matrix, \
#define GKO_DECLARE_LOWER_TRS_SOLVE_KERNEL(_vtype, _itype) \
void solve(std::shared_ptr<const DefaultExecutor> exec, \
const matrix::Csr<_vtype, _itype> *matrix, \
gko::solver::SolveStruct *solve_struct, \
pratikvn marked this conversation as resolved.
Show resolved Hide resolved
matrix::Dense<_vtype> *trans_b, matrix::Dense<_vtype> *trans_x, \
const matrix::Dense<_vtype> *b, matrix::Dense<_vtype> *x)


#define GKO_DECLARE_ALL_AS_TEMPLATES \
GKO_DECLARE_LOWER_TRS_CHECK_TRANSPOSABILITY_KERNEL(); \
GKO_DECLARE_LOWER_TRS_INIT_STRUCT_KERNEL(); \
template <typename ValueType, typename IndexType> \
GKO_DECLARE_LOWER_TRS_SOLVE_KERNEL(ValueType, IndexType); \
template <typename ValueType, typename IndexType> \
Expand Down
Loading