Skip to content

Enable architecture selection for DPCTL_TARGET_CUDA #2096

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
wants to merge 10 commits into
base: master
Choose a base branch
from
Open

Enable architecture selection for DPCTL_TARGET_CUDA #2096

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
bae5f3d
Clean up DPCTL_TARGET_CUDA handling
vlad-perevezentsev Jun 4, 2025
de83f29
Ver 1: Add sm_* offload arch support to DPCTL_TARGET_CUDA
vlad-perevezentsev Jun 5, 2025
9751a71
Ver 2: Use nvidia_gpu_sm_* alias in -fsycl-targets for CUDA
vlad-perevezentsev Jun 5, 2025
4a3ecf8
Update CUDA build docs
vlad-perevezentsev Jun 5, 2025
58710ed
Merge master into update_cuda_build
vlad-perevezentsev Jun 5, 2025
80f7bc7
Improve robustness of DPCTL_TARGET_CUDA handling
vlad-perevezentsev Jun 5, 2025
d395910
Update DPCTL_TARGET_CUDA option description
vlad-perevezentsev Jun 5, 2025
758e00f
Apply remarks
vlad-perevezentsev Jun 6, 2025
155ca52
Merge master into update_cuda_build
vlad-perevezentsev Jun 6, 2025
feee948
Use string(CONCAT) for multi-line DPCTL_TARGET_CUDA description
vlad-perevezentsev Jun 6, 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
31 changes: 23 additions & 8 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,9 +25,15 @@ option(DPCTL_GENERATE_COVERAGE_FOR_PYBIND11_EXTENSIONS
"Build dpctl pybind11 offloading extensions with coverage instrumentation"
OFF
)
option(DPCTL_TARGET_CUDA
"Build DPCTL to target CUDA devices"
OFF
string(CONCAT _desc_target_cuda
"Build DPCTL to target CUDA device. "
"Set to a truthy value (e.g., ON, TRUE) to use default architecture (sm_50), "
"or to a specific architecture like sm_80."
)
set(DPCTL_TARGET_CUDA
""
CACHE STRING
"${_desc_target_cuda}"
)
set(DPCTL_TARGET_HIP
""
Expand All @@ -51,15 +57,24 @@ set(_dpctl_sycl_target_compile_options)
set(_dpctl_sycl_target_link_options)

set(_dpctl_sycl_targets)
set(_dpctl_cuda_arch)
set(_dpctl_amd_targets)

if ("x${DPCTL_SYCL_TARGETS}" STREQUAL "x")
if (DPCTL_TARGET_CUDA)
set(_dpctl_sycl_targets "nvptx64-nvidia-cuda,spir64-unknown-unknown")
else()
if (DEFINED ENV{DPCTL_TARGET_CUDA})
set(_dpctl_sycl_targets "nvptx64-nvidia-cuda,spir64-unknown-unknown")
if (NOT "x${DPCTL_TARGET_CUDA}" STREQUAL "x")
Copy link
Collaborator

Choose a reason for hiding this comment

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

Is it fair to validate DPCTL_TARGET_CUDA only in case when empty DPCTL_SYCL_TARGETS?

Copy link
Collaborator

Choose a reason for hiding this comment

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

it was how we were doing it before—but it looks like current logical flow will add HIP targets even when DPCTL_SYCL_TARGETS is not none, but not CUDA

so that should probably be changed, either make DPCTL_SYCL_TARGETS exclusive from both or check DPCTL_TARGET_CUDA as well

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

In my understanding, when the user passes DPCTL_SYCL_TARGETS he is responsible for the correctness of the flags.

The logic of checking if (NOT “x${DPCTL_TARGET_HIP}” STREQUAL “x”) when DPCTL_SYCL_TARGETS is not none was added to pass the correct compile and link options.

if(_dpctl_amd_targets)
      list(APPEND _dpctl_sycl_target_compile_options -Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=${_dpctl_amd_targets})
      list(APPEND _dpctl_sycl_target_link_options -Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=${_dpctl_amd_targets})
   endif()

I am already working on PR that will refresh the logic for AMD build using aliases to remove if(_dpctl_amd_targets) branch.

if(DPCTL_TARGET_CUDA MATCHES "^sm_")
set(_dpctl_cuda_arch ${DPCTL_TARGET_CUDA})
elseif(DPCTL_TARGET_CUDA MATCHES "^(ON|TRUE|YES|Y|1)$")
set(_dpctl_cuda_arch "sm_50")
else()
message(FATAL_ERROR
"Invalid value for DPCTL_TARGET_CUDA: \"${DPCTL_TARGET_CUDA}\". "
"Expected 'ON', 'TRUE', 'YES', 'Y', '1', or a CUDA architecture like 'sm_80'."
)
endif()
set(_dpctl_sycl_targets "nvidia_gpu_${_dpctl_cuda_arch},spir64-unknown-unknown")
endif()

if (NOT "x${DPCTL_TARGET_HIP}" STREQUAL "x")
set(_dpctl_amd_targets ${DPCTL_TARGET_HIP})
if(_dpctl_sycl_targets)
Expand Down
24 changes: 22 additions & 2 deletions docs/doc_sources/beginners_guides/installation.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,13 +159,33 @@ The following plugins from CodePlay are supported:
.. _codeplay_nv_plugin: https://developer.codeplay.com/products/oneapi/nvidia/
.. _codeplay_amd_plugin: https://developer.codeplay.com/products/oneapi/amd/

``dpctl`` can be built for CUDA devices as follows:
``dpctl`` can be built for CUDA devices using the ``DPCTL_TARGET_CUDA`` CMake option,
which accepts a specific compute architecture string:

.. code-block:: bash

python scripts/build_locally.py --verbose --cmake-opts="-DDPCTL_TARGET_CUDA=sm_80"

To use the default architecture (``sm_50``),
set ``DPCTL_TARGET_CUDA`` to a value such as ``ON``, ``TRUE``, ``YES``, ``Y``, or ``1``:

.. code-block:: bash

python scripts/build_locally.py --verbose --cmake-opts="-DDPCTL_TARGET_CUDA=ON"

And for AMD devices
Note that kernels are built for ``sm_50`` by default, allowing them to work on a wider
range of architectures, but limiting the usage of more recent CUDA features.

For reference, compute architecture strings like ``sm_80`` correspond to specific
CUDA Compute Capabilities (e.g., Compute Capability 8.0 corresponds to ``sm_80``).
A complete mapping between NVIDIA GPU models and their respective
Compute Capabilities can be found in the official
`CUDA GPU Compute Capability <https://developer.nvidia.com/cuda-gpus>`_ documentation.

A full list of available SYCL alias targets is available in the
`DPC++ Compiler User Manual <https://intel.github.io/llvm/UsersManual.html>`_.

To build for AMD devices, use:

.. code-block:: bash

Expand Down
Loading