You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* add ltoir test support
* add options for multi-modules
* add tests
* add bitcode test
* [pre-commit.ci] auto code formatting
* fix format
* [pre-commit.ci] auto code formatting
* refresh
* apply bitcode file from cupy_test helpers
* use 2 tuples
* refresh
* format
* [pre-commit.ci] auto code formatting
* fix from upstream
* refresh from upstream
* fix tests
* take path_finder from PR 447
* add builder files
* use python lists/tuples
* libdevice integration
* refresh
* refresh
* refresh
* use cuda_pathfinder module for libdevice
* rebase
* rebase
* tests
* Address reviews
* put libdevice stuff under _static_libs
* refresh reviews
* change program to cython per PR 1565
* fix import
* fix tests
* fix ruff check
* ruff fix find_libdevice
* [pre-commit.ci] auto code formatting
* add spdx and copyright
* rm redundant include and fix test
* [pre-commit.ci] auto code formatting
* refresh tests
* add correct libdevice for CTK> 13
* [pre-commit.ci] auto code formatting
* revamp design of pathfinder as LocatedHeaderDir
* refresh
* [pre-commit.ci] auto code formatting
* fix mypy errirs
* fix base var declaration
* format changes
* format changes
* rename to bitcodelib
* refresh
* refresh
* revert
* refresh test_nvvm to use cupy test helpers
* refresh test_nvvm
* [pre-commit.ci] auto code formatting
* refresh with actual test
* rm try_common_paths
* refresh 1
* [pre-commit.ci] auto code formatting
* refresh
* refresh
* [pre-commit.ci] auto code formatting
* refresh
* [pre-commit.ci] auto code formatting
* refresh pathfinder test after rebae
* Uniform handling of cuda_python_test_helpers dependency in cuda_bindings, cuda_core
* pre-commit cleanup
* Prefer installed cuda-python-test-helpers in test bootstrap.
Avoid silently overriding site-packages with checkout paths during test setup by only injecting the repo helper path when the helper distribution is not installed, and fail fast with a clear error when neither source is available.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Remove cuda_core Pixi PYTHONPATH override for test helpers.
Keep helper resolution consistent with test bootstrap logic so Pixi runs do not silently force checkout helpers ahead of installed packages.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Remove helper package install from coverage workflow.
Coverage jobs now use the same conftest bootstrap path as other test flows, reducing redundant setup and keeping helper resolution behavior consistent across CI.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Use per-module pytest plugin registration for NVVM helper fixtures.
Keep fixture registration next to the tests that consume it and drop direct fixture imports that required lint suppressions.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Remove "Renamed" comments
* Make bitcode locate/find consistently raise not-found errors.
Raise BitcodeLibNotFoundError from locate_bitcode_lib instead of returning None, simplify find_bitcode_lib to rely on that behavior, and update bitcode tests to handle expected not-found cases outside all_must_work mode.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Refocus bitcode-lib tests on real coverage and narrow mocks.
Keep the environment-backed locate/find smoke test as the primary happy-path check, and replace broad mock scenarios with targeted branch tests for search order and not-found diagnostics.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Undo accidental change in cuda_pathfinder/tests/test_find_nvidia_binaries.py
* Use direct platform_aware.IS_WINDOWS import (rather than the detour through supported_nvidia_libs)
* Add found_via metadata for bitcode library discovery.
Record whether locate_bitcode_lib resolved via site-packages, conda, or CUDA_HOME to mirror pathfinder discovery APIs, and update focused tests to validate the reported source.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Export SUPPORTED_BITCODE_LIBS in cuda.pathfinder public API.
Expose supported bitcode names through __init__ using the existing Sphinx indirection pattern so users can discover the API surface consistently with other supported-* constants.
Co-authored-by: Cursor <cursoragent@cursor.com>
* refresh
* [pre-commit.ci] auto code formatting
* refresh
* [pre-commit.ci] auto code formatting
* Serialize NVVM compile/libdevice mutation per Program instance.
Add a per-instance threading lock around the NVVM verify/libdevice-add/compile path to prevent concurrent compile races from double-adding libdevice. The lock is taken under the GIL and held across nogil sections, and we verified this ordering avoids introducing a lock/GIL deadlock cycle.
Co-authored-by: Cursor <cursoragent@cursor.com>
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: brandon-b-miller <brmiller@nvidia.com>
Co-authored-by: Ralf W. Grosse-Kunstleve <rgrossekunst@nvidia.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
0 commit comments