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

Add chebyshev Iteration #1289

Open
wants to merge 18 commits into
base: develop
Choose a base branch
from
Open

Add chebyshev Iteration #1289

Show file tree
Hide file tree
Changes from 17 commits
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
75f83c3
init chebyshev
yhmtsai Feb 22, 2023
53d075d
add the check for kept data and remove lend
yhmtsai Mar 21, 2023
2d05dc0
core test
yhmtsai Mar 21, 2023
1a289e3
change to foci
yhmtsai Mar 21, 2023
cb64b4c
use iteration from stop criterion and update doc
yhmtsai Aug 3, 2023
188236e
extract the split residual update and update test
yhmtsai Aug 8, 2023
2412d54
use preconditioner not inner_solver for chebyshev
yhmtsai Aug 8, 2023
5a5439e
add chebyshev test and fix the number of generated
yhmtsai Aug 9, 2023
aa2e704
use scalar from ws and move some local variables
yhmtsai Sep 12, 2023
e70647f
review update
yhmtsai Jan 11, 2024
494a1f9
update with new format
yhmtsai Jan 11, 2024
d0e9b35
update format
yhmtsai Jun 28, 2024
9309aea
add config for chebyshev
yhmtsai Jun 28, 2024
03d1399
use mixed_type in test properly and update documentation
yhmtsai Jan 3, 2025
2e8e25b
add fused kernel for chebyshev
yhmtsai Jan 3, 2025
61b6655
pass alpha/beta by value to kernel and add test
yhmtsai Jan 7, 2025
385d784
format update
yhmtsai Feb 13, 2025
37d6ecc
use double/complex<double> for fosi
yhmtsai Feb 14, 2025
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
2 changes: 2 additions & 0 deletions common/unified/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,8 @@ set(UNIFIED_SOURCES
solver/bicgstab_kernels.cpp
solver/cg_kernels.cpp
solver/cgs_kernels.cpp
solver/cgs_kernels.cpp
solver/chebyshev_kernels.cpp
solver/common_gmres_kernels.cpp
solver/fcg_kernels.cpp
solver/gcr_kernels.cpp
Expand Down
65 changes: 65 additions & 0 deletions common/unified/solver/chebyshev_kernels.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,65 @@
// SPDX-FileCopyrightText: 2017 - 2025 The Ginkgo authors
//
// SPDX-License-Identifier: BSD-3-Clause

#include "core/solver/chebyshev_kernels.hpp"

#include <ginkgo/core/matrix/dense.hpp>

#include "common/unified/base/kernel_launch.hpp"


namespace gko {
namespace kernels {
namespace GKO_DEVICE_NAMESPACE {
namespace chebyshev {


template <typename ValueType, typename ScalarType>
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this actually need a second precision, or can we always use the highest precision available? This is memory-bound after all.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have the same thought and thinking maybe we just use double or complex directly?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If there is a numerical benefit to using a higher-precision multiplication every time, I would go for it

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

after rethink about it, I do not have a good way to use double or complex double without any information from the vector or matrix. Maybe still keep the same way as how we handle IR/richardson now. also I need the ScalarType, which is the ValueType from Chebyshev class

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

currently keep the value type of class but change fosi, alpha, beta type to double/complex
when dpcpp only support float not double, it will cast to float/complex internally to keep the same interface among all backends

void init_update(std::shared_ptr<const DefaultExecutor> exec,
const ScalarType alpha,
const matrix::Dense<ValueType>* inner_sol,
matrix::Dense<ValueType>* update_sol,
matrix::Dense<ValueType>* output)
{
run_kernel(
exec,
[] GKO_KERNEL(auto row, auto col, auto alpha, auto inner_sol,
auto update_sol, auto output) {
const auto inner_val = inner_sol(row, col);
update_sol(row, col) = inner_val;
output(row, col) += alpha * inner_val;
},
output->get_size(), alpha, inner_sol, update_sol, output);
}

GKO_INSTANTIATE_FOR_EACH_VALUE_AND_SCALAR_TYPE(
GKO_DECLARE_CHEBYSHEV_INIT_UPDATE_KERNEL);


template <typename ValueType, typename ScalarType>
void update(std::shared_ptr<const DefaultExecutor> exec, const ScalarType alpha,
const ScalarType beta, matrix::Dense<ValueType>* inner_sol,
matrix::Dense<ValueType>* update_sol,
matrix::Dense<ValueType>* output)
{
run_kernel(
exec,
[] GKO_KERNEL(auto row, auto col, auto alpha, auto beta, auto inner_sol,
auto update_sol, auto output) {
const auto val = inner_sol(row, col) + beta * update_sol(row, col);
inner_sol(row, col) = val;
update_sol(row, col) = val;
output(row, col) += alpha * val;
},
output->get_size(), alpha, beta, inner_sol, update_sol, output);
}

GKO_INSTANTIATE_FOR_EACH_VALUE_AND_SCALAR_TYPE(
GKO_DECLARE_CHEBYSHEV_UPDATE_KERNEL);


} // namespace chebyshev
} // namespace GKO_DEVICE_NAMESPACE
} // namespace kernels
} // namespace gko
3 changes: 2 additions & 1 deletion core/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,8 +92,8 @@ target_sources(${ginkgo_core}
matrix/scaled_permutation.cpp
matrix/sellp.cpp
matrix/sparsity_csr.cpp
multigrid/pgm.cpp
multigrid/fixed_coarsening.cpp
multigrid/pgm.cpp
preconditioner/batch_jacobi.cpp
preconditioner/gauss_seidel.cpp
preconditioner/sor.cpp
Expand All @@ -112,6 +112,7 @@ target_sources(${ginkgo_core}
solver/cb_gmres.cpp
solver/cg.cpp
solver/cgs.cpp
solver/chebyshev.cpp
solver/direct.cpp
solver/fcg.cpp
solver/gcr.cpp
Expand Down
10 changes: 5 additions & 5 deletions core/config/config_helper.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: 2017 - 2024 The Ginkgo authors
// SPDX-FileCopyrightText: 2017 - 2025 The Ginkgo authors
//
// SPDX-License-Identifier: BSD-3-Clause

Expand All @@ -24,10 +24,9 @@ namespace gko {
namespace config {


#define GKO_INVALID_CONFIG_VALUE(_entry, _value) \
GKO_INVALID_STATE(std::string("The value >" + _value + \
"< is invalid for the entry >" + _entry + \
"<"))
#define GKO_INVALID_CONFIG_VALUE(_entry, _value) \
GKO_INVALID_STATE(std::string("The value >") + _value + \
"< is invalid for the entry >" + _entry + "<")


#define GKO_MISSING_CONFIG_ENTRY(_entry) \
Expand All @@ -52,6 +51,7 @@ enum class LinOpFactoryType : int {
Direct,
LowerTrs,
UpperTrs,
Chebyshev,
Factorization_Ic,
Factorization_Ilu,
Cholesky,
Expand Down
3 changes: 2 additions & 1 deletion core/config/registry.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: 2017 - 2024 The Ginkgo authors
// SPDX-FileCopyrightText: 2017 - 2025 The Ginkgo authors
//
// SPDX-License-Identifier: BSD-3-Clause

Expand Down Expand Up @@ -32,6 +32,7 @@ configuration_map generate_config_map()
{"solver::Direct", parse<LinOpFactoryType::Direct>},
{"solver::LowerTrs", parse<LinOpFactoryType::LowerTrs>},
{"solver::UpperTrs", parse<LinOpFactoryType::UpperTrs>},
{"solver::Chebyshev", parse<LinOpFactoryType::Chebyshev>},
{"factorization::Ic", parse<LinOpFactoryType::Factorization_Ic>},
{"factorization::Ilu", parse<LinOpFactoryType::Factorization_Ilu>},
{"factorization::Cholesky", parse<LinOpFactoryType::Cholesky>},
Expand Down
4 changes: 3 additions & 1 deletion core/config/solver_config.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: 2017 - 2024 The Ginkgo authors
// SPDX-FileCopyrightText: 2017 - 2025 The Ginkgo authors
//
// SPDX-License-Identifier: BSD-3-Clause

Expand All @@ -12,6 +12,7 @@
#include <ginkgo/core/solver/cb_gmres.hpp>
#include <ginkgo/core/solver/cg.hpp>
#include <ginkgo/core/solver/cgs.hpp>
#include <ginkgo/core/solver/chebyshev.hpp>
#include <ginkgo/core/solver/direct.hpp>
#include <ginkgo/core/solver/fcg.hpp>
#include <ginkgo/core/solver/gcr.hpp>
Expand Down Expand Up @@ -43,6 +44,7 @@ GKO_PARSE_VALUE_TYPE_BASE(CbGmres, gko::solver::CbGmres);
GKO_PARSE_VALUE_AND_INDEX_TYPE(Direct, gko::experimental::solver::Direct);
GKO_PARSE_VALUE_AND_INDEX_TYPE(LowerTrs, gko::solver::LowerTrs);
GKO_PARSE_VALUE_AND_INDEX_TYPE(UpperTrs, gko::solver::UpperTrs);
GKO_PARSE_VALUE_TYPE(Chebyshev, gko::solver::Chebyshev);


template <>
Expand Down
11 changes: 11 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 @@ -63,6 +63,7 @@
#include "core/solver/cb_gmres_kernels.hpp"
#include "core/solver/cg_kernels.hpp"
#include "core/solver/cgs_kernels.hpp"
#include "core/solver/chebyshev_kernels.hpp"
#include "core/solver/common_gmres_kernels.hpp"
#include "core/solver/fcg_kernels.hpp"
#include "core/solver/gcr_kernels.hpp"
Expand Down Expand Up @@ -674,6 +675,16 @@ GKO_STUB_CB_GMRES_CONST(GKO_DECLARE_CB_GMRES_SOLVE_KRYLOV_KERNEL);
} // namespace cb_gmres


namespace chebyshev {


GKO_STUB_VALUE_AND_SCALAR_TYPE(GKO_DECLARE_CHEBYSHEV_INIT_UPDATE_KERNEL);
GKO_STUB_VALUE_AND_SCALAR_TYPE(GKO_DECLARE_CHEBYSHEV_UPDATE_KERNEL);


} // namespace chebyshev


namespace ir {


Expand Down
Loading
Loading