Skip to content

[rocsparse] Fix csric0 dispatch when rows have many nonzeros #5360

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

Open
jsandham wants to merge 6 commits into
base: develop
Choose a base branch
from
+35 −30
Open

[rocsparse] Fix csric0 dispatch when rows have many nonzeros #5360

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
fc30d9e
Add missing fence to csrsv compute kernel
jsandham Mar 11, 2026
2792208
Fix csric0 by adding missing fences
jsandham Mar 11, 2026
fa34062
Add extra csric0 dense matrix test
jsandham Mar 11, 2026
e5e5934
Remove unneeded WF_SIZE from csrsm kernel template parameters
jsandham Mar 11, 2026
d36db75
Revert fence additions as unneeded and update changelog and version n…
jsandham Mar 18, 2026
a6b66e8
Update projects/rocsparse/CHANGELOG.md
jsandham Mar 18, 2026
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
5 changes: 5 additions & 0 deletions projects/rocsparse/CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,11 @@
Documentation for rocSPARSE is available at
[https://rocm.docs.amd.com/projects/rocSPARSE/en/latest/](https://rocm.docs.amd.com/projects/rocSPARSE/en/latest/).

## rocSPARSE 4.5.1 for ROCm 7.13.0

### Resolved issues
* Fix `rocsparse_[s|d|c|z]csric0` where `rocsparse_status_invalid_value` was being returned when the maximum number of non-zeros in any row is between 513 and 1024.

## rocSPARSE 4.5.0 for ROCm 7.12.0

### Added
Expand Down
2 changes: 1 addition & 1 deletion projects/rocsparse/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -193,7 +193,7 @@ endif()
message(STATUS "GPU_TARGETS: ${GPU_TARGETS}")

# Setup version
set(VERSION_STRING "4.5.0")
set(VERSION_STRING "4.5.1")
set(SOVERSION_STRING "1.0")
set(MIN_ROCPRIM_VERSION "4.0.0" CACHE STRING "Minimum version of rocPRIM.")
set(MIN_ROCBLAS_VERSION "4.1.0" CACHE STRING "Minimum version of rocBLAS.")
Expand Down
1 change: 1 addition & 0 deletions projects/rocsparse/clients/tests/test_csric0.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,7 @@ Definitions:
- &M_N_range_extra
- { M: 10, N: 10 }
- { M: 235, N: 235 }
- { M: 600, N: 600 }
- { M: 1200, N: 1200 }

Tests:
Expand Down
2 changes: 1 addition & 1 deletion projects/rocsparse/library/src/level3/csrsm_device.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,7 @@

namespace rocsparse
{
template <uint32_t BLOCKSIZE, uint32_t WF_SIZE, bool SLEEP, typename I, typename J, typename T>
template <uint32_t BLOCKSIZE, bool SLEEP, typename I, typename J, typename T>
ROCSPARSE_DEVICE_ILF void csrsm_device(rocsparse_operation transB,
J m,
J nrhs,
Expand Down
53 changes: 26 additions & 27 deletions projects/rocsparse/library/src/level3/rocsparse_csrsm_solve_impl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@
#include "csrsm_device.h"
namespace rocsparse
{
template <uint32_t BLOCKSIZE, uint32_t WFSIZE, bool SLEEP, typename I, typename J, typename T>
template <uint32_t BLOCKSIZE, bool SLEEP, typename I, typename J, typename T>
ROCSPARSE_KERNEL(BLOCKSIZE)
void csrsm(rocsparse_operation transB,
J m,
Expand All @@ -57,21 +57,21 @@ namespace rocsparse
{
ROCSPARSE_DEVICE_HOST_SCALAR_GET(alpha);

rocsparse::csrsm_device<BLOCKSIZE, WFSIZE, SLEEP>(transB,
m,
nrhs,
alpha,
csr_row_ptr,
csr_col_ind,
csr_val,
B,
ldb,
done_array,
map,
zero_pivot,
idx_base,
fill_mode,
diag_type);
rocsparse::csrsm_device<BLOCKSIZE, SLEEP>(transB,
m,
nrhs,
alpha,
csr_row_ptr,
csr_col_ind,
csr_val,
B,
ldb,
done_array,
map,
zero_pivot,
idx_base,
fill_mode,
diag_type);
}

template <typename I, typename J, typename T>
Expand Down Expand Up @@ -211,11 +211,10 @@ namespace rocsparse

if(blockdim == 64)
{

if(gcn_arch_name == rocpsarse_arch_names::gfx908 && asicRev < 2)
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<64, 64, true>),
(rocsparse::csrsm<64, true>),
csrsm_blocks,
csrsm_threads,
0,
Expand All @@ -240,7 +239,7 @@ namespace rocsparse
else
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<64, 64, false>),
(rocsparse::csrsm<64, false>),
csrsm_blocks,
csrsm_threads,
0,
Expand Down Expand Up @@ -268,7 +267,7 @@ namespace rocsparse
if(gcn_arch_name == rocpsarse_arch_names::gfx908 && asicRev < 2)
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<128, 64, true>),
(rocsparse::csrsm<128, true>),
csrsm_blocks,
csrsm_threads,
0,
Expand All @@ -293,7 +292,7 @@ namespace rocsparse
else
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<128, 64, false>),
(rocsparse::csrsm<128, false>),
csrsm_blocks,
csrsm_threads,
0,
Expand Down Expand Up @@ -321,7 +320,7 @@ namespace rocsparse
if(gcn_arch_name == rocpsarse_arch_names::gfx908 && asicRev < 2)
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<256, 64, true>),
(rocsparse::csrsm<256, true>),
csrsm_blocks,
csrsm_threads,
0,
Expand All @@ -346,7 +345,7 @@ namespace rocsparse
else
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<256, 64, false>),
(rocsparse::csrsm<256, false>),
csrsm_blocks,
csrsm_threads,
0,
Expand Down Expand Up @@ -374,7 +373,7 @@ namespace rocsparse
if(gcn_arch_name == rocpsarse_arch_names::gfx908 && asicRev < 2)
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<512, 64, true>),
(rocsparse::csrsm<512, true>),
csrsm_blocks,
csrsm_threads,
0,
Expand All @@ -399,7 +398,7 @@ namespace rocsparse
else
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<512, 64, false>),
(rocsparse::csrsm<512, false>),
csrsm_blocks,
csrsm_threads,
0,
Expand Down Expand Up @@ -427,7 +426,7 @@ namespace rocsparse
if(gcn_arch_name == rocpsarse_arch_names::gfx908 && asicRev < 2)
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<1024, 64, true>),
(rocsparse::csrsm<1024, true>),
csrsm_blocks,
csrsm_threads,
0,
Expand All @@ -452,7 +451,7 @@ namespace rocsparse
else
{
RETURN_IF_HIPLAUNCHKERNELGGL_ERROR(
(rocsparse::csrsm<1024, 64, false>),
(rocsparse::csrsm<1024, false>),
csrsm_blocks,
csrsm_threads,
0,
Expand Down
2 changes: 1 addition & 1 deletion projects/rocsparse/library/src/precond/csric0/rocsparse_csric0_kernel_launch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,7 @@ rocsparse_status rocsparse::csric0_kernel_launch(rocsparse_handle handle,

rocsparse::csric0_kernel_launch_t launch{};

if(sleep || (trm_info->get_max_nnz() > 1024))
if(sleep || (trm_info->get_max_nnz() > 512))
{
launch = rocsparse::find_csric0_kernel_binsearch_launch(handle, csric0_info, A);
}
Expand Down
Loading