Skip to content

[BLAS][SPARSE][DPC++] Add CUDA_TARGETS option to pass for AoT compila… #710

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

Merged
sknepper merged 2 commits into from
Oct 9, 2025
Merged

[BLAS][SPARSE][DPC++] Add CUDA_TARGETS option to pass for AoT compila… #710

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ce59887
[BLAS][SPARSE][DPC++] Add CUDA_TARGETS option to pass for AoT compila…
abagusetty Sep 17, 2025
c0e7148
[SYCL] Fall back to the logic when CUDA_TARGETS is not provided
abagusetty Sep 23, 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
1 change: 1 addition & 0 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,6 +70,7 @@ option(ENABLE_CUSPARSE_BACKEND "Enable the cuSPARSE backend for the SPARSE_BLAS
option(ENABLE_ROCSPARSE_BACKEND "Enable the rocSPARSE backend for the SPARSE_BLAS interface" OFF)

set(ONEMATH_SYCL_IMPLEMENTATION "dpc++" CACHE STRING "Name of the SYCL compiler")
set(CUDA_TARGETS "" CACHE STRING "Target CUDA architectures")
set(HIP_TARGETS "" CACHE STRING "Target HIP architectures")

## Testing
Expand Down
18 changes: 15 additions & 3 deletions src/blas/backends/cublas/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,12 +41,24 @@ target_include_directories(${LIB_OBJ}
)
target_compile_options(${LIB_OBJ} PRIVATE ${ONEMATH_BUILD_COPT})

if(NOT ${ONEMATH_SYCL_IMPLEMENTATION} STREQUAL "adaptivecpp")
if (NOT "${ONEMATH_SYCL_IMPLEMENTATION}" STREQUAL "adaptivecpp")
target_compile_options(ONEMATH::SYCL::SYCL INTERFACE
-fsycl-targets=nvptx64-nvidia-cuda -fsycl-unnamed-lambda
)
target_link_options(ONEMATH::SYCL::SYCL INTERFACE
-fsycl-targets=nvptx64-nvidia-cuda
)

if (DEFINED CUDA_TARGETS AND NOT "${CUDA_TARGETS}" STREQUAL "")
target_compile_options(ONEMATH::SYCL::SYCL INTERFACE
-fsycl-targets=nvptx64-nvidia-cuda -fsycl-unnamed-lambda)
-Xsycl-target-backend --cuda-gpu-arch=${CUDA_TARGETS}
)
target_link_options(ONEMATH::SYCL::SYCL INTERFACE
-fsycl-targets=nvptx64-nvidia-cuda)
-Xsycl-target-backend --cuda-gpu-arch=${CUDA_TARGETS}
)
endif()
endif()

target_link_libraries(${LIB_OBJ} PUBLIC ONEMATH::SYCL::SYCL ONEMATH::cuBLAS::cuBLAS)
target_compile_features(${LIB_OBJ} PUBLIC cxx_std_11)
set_target_properties(${LIB_OBJ} PROPERTIES
Expand Down
6 changes: 3 additions & 3 deletions src/blas/backends/generic/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,11 +82,11 @@ if(GENERIC_BLAS_TUNING_TARGET)
-fsycl-targets=nvptx64-nvidia-cuda -fsycl-unnamed-lambda)
target_link_options(ONEMATH::SYCL::SYCL INTERFACE
-fsycl-targets=nvptx64-nvidia-cuda)
if(DEFINED CUDA_TARGET)
if(DEFINED CUDA_TARGETS)
target_compile_options(ONEMATH::SYCL::SYCL INTERFACE
-Xsycl-target-backend --cuda-gpu-arch=${CUDA_TARGET})
-Xsycl-target-backend --cuda-gpu-arch=${CUDA_TARGETS})
target_link_options(ONEMATH::SYCL::SYCL INTERFACE
-Xsycl-target-backend --cuda-gpu-arch=${CUDA_TARGET})
-Xsycl-target-backend --cuda-gpu-arch=${CUDA_TARGETS})
endif()
else()
message(WARNING "Compiler is not supported."
Expand Down
18 changes: 18 additions & 0 deletions src/sparse_blas/backends/cusparse/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,24 @@ target_include_directories(${LIB_OBJ}

target_compile_options(${LIB_OBJ} PRIVATE ${ONEMATH_BUILD_COPT})

if (NOT "${ONEMATH_SYCL_IMPLEMENTATION}" STREQUAL "adaptivecpp")
target_compile_options(ONEMATH::SYCL::SYCL INTERFACE
-fsycl-targets=nvptx64-nvidia-cuda -fsycl-unnamed-lambda
)
target_link_options(ONEMATH::SYCL::SYCL INTERFACE
-fsycl-targets=nvptx64-nvidia-cuda
)

if (DEFINED CUDA_TARGETS AND NOT "${CUDA_TARGETS}" STREQUAL "")
target_compile_options(ONEMATH::SYCL::SYCL INTERFACE
-Xsycl-target-backend --cuda-gpu-arch=${CUDA_TARGETS}
)
target_link_options(ONEMATH::SYCL::SYCL INTERFACE
-Xsycl-target-backend --cuda-gpu-arch=${CUDA_TARGETS}
)
endif()
endif()

Comment on lines +46 to +63
Copy link

@spencerpatty spencerpatty Sep 29, 2025

Choose a reason for hiding this comment

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

this is fine, it is same update as for BLAS, for CUDA + SYCL linking, so no reason to object

Copy link
Contributor Author

@abagusetty abagusetty Sep 30, 2025

Choose a reason for hiding this comment

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

Correct, was just adding it for consistency between the domains. Having a way to add AoT support too and to avoid issues with newer CUDA versions
Context: CUDA 13 dropped support for several older NVIDIA GPU architectures, specifically Maxwell (compute capability 5.x), Pascal (compute capability 6.x), and Volta (compute capability 7.0, 7.2) architectures.

if (${CMAKE_VERSION} VERSION_LESS "3.17.0")
find_package(CUDA 12.2 REQUIRED)
target_include_directories(${LIB_OBJ} PRIVATE ${CUDA_INCLUDE_DIRS})
Expand Down
Loading