Update Python wrapper for C repo PR #8 + CUDA 12/13 support #4
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Replace osqp_update_rho() with osqp_update_step_sizes(rho, sigma) - Rename settings: rho -> rho_init, sigma -> sigma_init - Remove obsolete settings: rho_is_vec, adaptive_rho_tolerance_greater/less - Remove rho_estimate from OSQPInfo - Add update_step_sizes(rho_init, sigma_init) method to Python interface - Add deprecation warnings for old parameter names - Update tests to use new API
- Switch from b/diagonal_step_sizes to b/cuda-13-blackwell-support which includes both the diagonal step sizes API changes AND CUDA 13/ Blackwell architecture support - Update CLAUDE.md with clearer build instructions: - Emphasize activating venv before uv pip commands - Add uv build instructions for backend wheels - Document proper wheel installation workflow
- Restructure backend directories: cuda -> cuda12, add cuda13 - CUDA 12: Update from 12.6 to 12.9 - CUDA 13: Add new backend package (osqp-cu13) for CUDA 13.1 - Consolidate workflows: build_cuda12.yml and build_cuda13.yml each handle Linux + Windows builds and PyPI publishing - Both packages produce same osqp_cuda module (mutually exclusive) - Update CLAUDE.md with build instructions for both backends
bstellato
changed the title
- Skip Python 3.8 for CUDA 13 (requires C++17, manylinux2014 uses older compilers) - Add missing Windows CUDA 13 components (cuda_runtime, cublas, curand, cusparse)
- Add CMAKE_CUDA_STANDARD=17 and CMAKE_CXX_STANDARD=17 for CUDA 13 - Install full CUDA 13 toolkit on Windows (component names changed) - Remove CMAKE_GENERATOR_TOOLSET to let CMake auto-detect VS integration - Fix Windows CUDA cache path
- Switch Windows runner from windows-2022 to windows-2025 - Add CMAKE_CUDA_STANDARD=17 and CMAKE_CXX_STANDARD=17 to pyproject.toml (environment variables alone don't get passed to CMake via scikit-build-core)
- CUDA 12: Fix cache path from Intel OneAPI to NVIDIA CUDA directory - CUDA 13: Use explicit component list with -s flag instead of undefined CUDATOOLKIT_INSTALL_ARGS variable
Collaborator
|
Thanks @bstellato! Let me know when you think you're "done" with this PR and I can take a further look at any failing CI scenarios - looks like its only Windows right now so that should be fixable. |
Collaborator
Author
|
Thanks @vineetbansal ! Could you please take care of the remaining failing CUDA 13 builds on windows? I may have specified CXX 17 in a couple of places where it is not strictly needed. Otherwise, this PR should also automatically build wheels for both CUDA 12 and 13. |
vineetbansal
force-pushed
the
feature/pr8-compat
branch
2 times, most recently
from
ac966ec to
c4fc7e2
Compare
components for cuda toolkit on windows
vineetbansal
force-pushed
the
feature/pr8-compat
branch
from
8ee4b0e to
cd7aea4
Compare
Collaborator
|
@bstellato - should be good now. |
Collaborator
Author
|
Amazing! Thank you so much @vineetbansal! Before merging this PR, I propose to:
Then we should be good to go! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
This PR updates the Python wrapper to be compatible with the C repo PR #8 (diagonal step sizes) and adds support for both CUDA 12 and CUDA 13 backends.
API Changes (from C PR #8)
rho→rho_init(initial step size)sigma→sigma_init(initial step size)update_rho()→update_step_sizes(rho_init=..., sigma_init=...)CUDA Backend Support
osqp_cudamodule (mutually exclusive)Files Changed
src/bindings.cpp.in: Updated bindings for new APIsrc/osqp/interface.py: Updated Python interfacebackend/cuda12/: CUDA 12 backend configurationbackend/cuda13/: CUDA 13 backend configuration.github/workflows/build_cuda12.yml: Build + publish CUDA 12 wheels.github/workflows/build_cuda13.yml: Build + publish CUDA 13 wheelsCLAUDE.md: Updated build documentationTest plan