First version of cuda.bindings.path_finder
#578
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -65,6 +65,8 @@ jobs: | |
| } | ||
| } | ||
|
|
||
| "CUDA_BINDINGS_PATH_FINDER_TEST_ALL_LIBNAMES=1" >> $env:GITHUB_ENV | ||
|
|
||
| # Make outputs from the previous job as env vars | ||
| $CUDA_CORE_ARTIFACT_BASENAME = "cuda-core-python${PYTHON_VERSION_FORMATTED}-${{ inputs.host-platform }}" | ||
| "PYTHON_VERSION_FORMATTED=${PYTHON_VERSION_FORMATTED}" >> $env:GITHUB_ENV | ||
|
|
@@ -165,8 +167,7 @@ jobs: | |
| uses: Jimver/[email protected] | ||
| with: | ||
| cuda: ${{ inputs.cuda-version }} | ||
| method: 'local' | ||
|
|
||
| - name: Update PATH | ||
| if: ${{ inputs.local-ctk == '1' }} | ||
|
|
@@ -190,7 +191,7 @@ jobs: | |
|
|
||
| Push-Location ./cuda_bindings | ||
| pip install -r requirements.txt | ||
| pytest -ra -s -v tests/ | ||
| # skip Cython tests for now (NVIDIA/cuda-python#466) | ||
| Pop-Location | ||
|
|
||
|
|
@@ -214,7 +215,7 @@ jobs: | |
|
|
||
| Push-Location ./cuda_core | ||
| pip install -r "tests/requirements-cu${TEST_CUDA_MAJOR}.txt" | ||
| pytest -ra -s -v tests/ | ||
| Pop-Location | ||
|
|
||
| - name: Ensure cuda-python installable | ||
|
|
||
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,61 @@ | ||
| # `cuda.bindings.path_finder` Module | ||
|
|
||
| ## Public API (Work in Progress) | ||
|
|
||
| Currently exposes two primary interfaces: | ||
|
|
||
| ``` | ||
| cuda.bindings.path_finder._SUPPORTED_LIBNAMES # ('nvJitLink', 'nvrtc', 'nvvm') | ||
| cuda.bindings.path_finder._load_nvidia_dynamic_library(libname: str) -> LoadedDL | ||
| ``` | ||
|
|
||
| **Note:** | ||
| These APIs are prefixed with an underscore because they are considered | ||
| experimental while undergoing active development, although already | ||
| reasonably well-tested through CI pipelines. | ||
|
|
||
| ## Library Loading Search Priority | ||
|
|
||
| The `load_nvidia_dynamic_library()` function implements a hierarchical search | ||
| strategy for locating NVIDIA shared libraries: | ||
|
Comment on lines
+19
to
+20
There was a problem hiding this comment. I think we should clarify that the current behavior is that the search is done on a per library basis and that there is no centralized mechanism that ensures all libraries are found from the same location / installation method. I.E. if someone has a In the future we should probably not allow this behavior to occur, but for now we should probably clearly document that this is not behavior we intend to support in the future and is essentially undefined behavior as of now. There was a problem hiding this comment. FWIW in nvmath-python docs we simplified what Keith said into: "Don't mix-n-match pip, conda, and system (installations of CUDA)" 😄 https://docs.nvidia.com/cuda/nvmath-python/latest/installation.html#troubleshooting |
||
|
|
||
|
There was a problem hiding this comment. I think we're missing using the There was a problem hiding this comment. Also, if someone has already loaded libraries via some other mechanism I think we end up finding those before going down any of these other search paths? |
||
| 1. **Python Package Ecosystem** | ||
| - Scans `sys.path` to find libraries installed via NVIDIA Python wheels | ||
|
|
||
| 2. **Conda Environments** | ||
| - Leverages Conda-specific paths through our fork of `get_cuda_paths()` from Numba | ||
|
|
||
| 3. **System Installations** | ||
| - Checks traditional system locations via the same `get_cuda_paths()` implementation | ||
|
Comment on lines
+28
to
+29
There was a problem hiding this comment. Can you make this more explicit? Does this check |
||
|
|
||
| 4. **OS Default Mechanisms** | ||
| - Falls back to native loader: | ||
| - `dlopen()` on Linux | ||
| - `LoadLibraryW()` on Windows | ||
|
|
||
| ## Implementation Philosophy | ||
|
|
||
| The current implementation balances stability and evolution: | ||
|
|
||
| - **Baseline Foundation:** Uses a fork of Numba's `cuda_paths.py` that has been | ||
| battle-tested in production environments | ||
|
|
||
| - **Validation Infrastructure:** Comprehensive CI testing matrix being developed to cover: | ||
| - Various Linux/Windows environments | ||
| - Python packaging formats (wheels, conda) | ||
| - CUDA Toolkit versions | ||
|
|
||
| - **Roadmap:** Planned refactoring to: | ||
| - Unify library discovery logic | ||
| - Improve maintainability | ||
| - Better enforce search priority | ||
| - Expand platform support | ||
|
|
||
| ## Maintenance Requirements | ||
|
|
||
| These key components must be updated for new CUDA Toolkit releases: | ||
|
|
||
| - `supported_libs.SUPPORTED_LIBNAMES` | ||
| - `supported_libs.SUPPORTED_WINDOWS_DLLS` | ||
| - `supported_libs.SUPPORTED_LINUX_SONAMES` | ||
| - `supported_libs.EXPECTED_LIB_SYMBOLS` | ||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think the path is correct