CUDA Python 12.9.1

Release notes

• https://nvidia.github.io/cuda-python/latest/release/12.9.1-notes.html

Documentation

• https://nvidia.github.io/cuda-python/12.9.1/

cuda-pathfinder v1.1.0

• CTK 13.0.0 compatibility
• Bug fix: load libnvJitLink.so.12 from conda, not /usr/local/cuda PR #767

cuda-pathfinder v1.0.0

First release of cuda-pathfinder as a stand-alone module.

cuda.pathfinder replaces cuda.bindings.path_finder, which was released with cuda-bindings 12.9.0 and is now deprecated.

Note that cuda-pathfinder is a noarch package and has no dependencies (other than a Python 3.9+ interpreter).

Please see cuda/pathfinder/README for more information.

cuda.core v0.3.1

cuda.core v0.3.1 release announcement

Release note

• https://nvidia.github.io/cuda-python/cuda-core/latest/release/0.3.1-notes.html

All functionalities are currently hosted under the cuda.core.experimental namespace. Once the features become stable they will be moved out of experimental.

Documentation

• https://nvidia.github.io/cuda-python/cuda-core/0.3.1/

Sample codes

• https://github.com/NVIDIA/cuda-python/tree/cuda-core-v0.3.1/cuda_core/examples

What's Changed

• Bump github/codeql-action from 3.28.19 to 3.29.0 by @dependabot in #710
• Fix Windows build CI by @leofang in #713
• Bump pypa/cibuildwheel from 2.23.3 to 3.0.0 by @dependabot in #711
• Ensure correct handling of buffers allocated with LegacyPinnedMemoryResource.allocate as kernel parameters by @shwina in #717
• Fix nvbugpro 5348750 by @oleksandr-pavlyk in #725
• Add a "Getting Started" page to the documentation by @shwina in #720
• Bump korthout/backport-action from 3.2.0 to 3.2.1 by @dependabot in #738
• Bump github/codeql-action from 3.29.0 to 3.29.2 by @dependabot in #737
• cuda_core/tests/test_event.py::test_timing_success WSL compatibility by @rwgk in #740
• Restore option to run testing without cupy installed. by @rwgk in #741
• Cythonize away some perf hot spots by @leofang in #709
• cuda_core forward compatibility changes. by @rwgk in #722
• Update docs for v0.3.1 release by @leofang in #695

Full Changelog: cuda-core-v0.3.0...cuda-core-v0.3.1

cuda.core v0.3.0

cuda.core v0.3.0 release announcement

Release note

• https://nvidia.github.io/cuda-python/cuda-core/latest/release/0.3.0-notes.html

All functionalities are currently hosted under the cuda.core.experimental namespace. Once the features become stable they will be moved out of experimental.

Documentation

• https://nvidia.github.io/cuda-python/cuda-core/0.3.0/

Sample codes

• https://github.com/NVIDIA/cuda-python/tree/cuda-core-v0.3.0/cuda_core/examples

What's Changed

• cuda.core: CUResult, cudaError explanations by @rwgk in #503
• DOC: Add admonition to docstrings for cuda.core handle properties by @carterbox in #573
• NEW: Make event timing error messages more specific and actionable by @carterbox in #559
• Change cuda.core license to Apache-2.0 & make contributing guides clear by @leofang in #583
• Add lint instructions by @msaroufim in #581
• fix indexing bug in saxpy.py by @msaroufim in #582
• Consolidate shared info between README.md and DESCRIPTION.rst by @vzhurba01 in #590
• PyTorch example by @msaroufim in #579
• Address remaining OSRB requests + document known installation issues by @leofang in #626
• Initial version of pre-commit "Check SPDX-License-Identifier" by @rwgk in #625
• Implement Kernel.num_arguments, and Kernel.arguments_info by @oleksandr-pavlyk in #612
• Move dependencies from requirements.txt to an optional packaging extra by @kkraus14 in #638
• Always build and run Cython tests + other CI improvements by @leofang in #640
• Implement device and context properties for Event by @NaderAlAwar in #618
• Enable serialization/deserialization of ObjectCode instances by @brandon-b-miller in #660
• Add more ObjectCode constructors by @brandon-b-miller in #652
• Unify Common CI Code for Windows and Linux by @cryos in #645
• MNT: Bump DLPack header to 1.1 by @leofang in #667
• Add tests to cover scalar handling in launch() + Fix fp16 bug by @leofang in #669
• Feature/occupancy by @oleksandr-pavlyk in #648
• Repair Windows wheels by @leofang in #673
• Migrate to windows-2022 for Windows CI builds by @cryos in #672
• Bugfix/multiple ptxas options values by @oleksandr-pavlyk in #678
• Clean up cffi resources in file by @oleksandr-pavlyk in #679
• Support cooperative launch by @leofang in #676
• Allow ObjectCode to have a name by @leofang in #682
• Make compute-sanitizer not report API errors as errors by @leofang in #687
• Update the notes on the CCCL and nvmath-python projects by @leofang in #688
• Switch to use CUDA driver APIs in Device constructor by @leofang in #460
• Bump github/codeql-action from 3.28.18 to 3.28.19 by @dependabot in #700
• Bump conda-incubator/setup-miniconda from 3.1.1 to 3.2.0 by @dependabot in #701
• Add phase 1 of CUDA Graphs support by @vzhurba01 in #455
• Make a few memory management objects public + Miscellaneous doc updates by @leofang in #693
• Bump cuda.core to v0.3.0 by @leofang in #703

New Contributors

• @gmarkall made their first contribution in #522
• @kkraus14 made their first contribution in #549
• @pre-commit-ci made their first contribution in #552
• @msaroufim made their first contribution in #581
• @cryos made their first contribution in #555
• @dependabot made their first contribution in #700

Full Changelog: cuda-core-v0.2.0...cuda-core-v0.3.0

CUDA Python 12.9.0

Release notes

• https://nvidia.github.io/cuda-python/12.9.0/release/12.9.0-notes.html

Documentation

• https://nvidia.github.io/cuda-python/12.9.0/

CUDA Python 11.8.7

Release notes

• https://nvidia.github.io/cuda-python/12.9.0/release/11.8.7-notes.html

cuda.core v0.2.0

cuda.core v0.2.0 release announcement

Release note

• https://nvidia.github.io/cuda-python/cuda-core/latest/release/0.2.0-notes.html

All functionalities are currently hosted under the cuda.core.experimental namespace. Once the features become stable they will be moved out of experimental.

Key Features and Enhancements

• Add ProgramOptions to facilitate the passing of runtime compile options to Program.
• Add pythonic access to Device and Kernel attributes.

For full details please refer to the release note above.

Breaking Changes

• The stream attribute is removed from LaunchConfig. Instead, the Stream object should now be directly passed to launch as an argument.
• The signature for launch is changed by swapping positional arguments, the new signature is now (stream, config, kernel, *kernel_args)
• Change __cuda_stream__ from attribute to method.
• The Program.compile method no longer accepts the options argument. Instead, you can optionally pass an instance of ProgramOptions to the constructor of Program.
• Device.properties now provides attribute getters instead of a dictionary interface.
• The .handle attribute of various cuda.core objects now returns the underlying Python object instead of a (type-erased) Python integer.

New examples

• jit_lto_fractal.py — Demonstrates just-in-time link-time optimization for fractal generation. (Device, LaunchConfig, Linker, LinkerOptions, Program, ProgramOptions) (#475)
• simple_multi_gpu_example.py — Example of using multiple GPUs. (Device, Program, LaunchConfig) (#304)
• show_device_properties.py — Displays detailed device properties. (Device) (#474)

Documentation

• https://nvidia.github.io/cuda-python/cuda-core/0.2.0/

Sample codes

• https://github.com/NVIDIA/cuda-python/tree/cuda-core-v0.2.0/cuda_core/examples

Test fixes

• Clean up device initialization in some tests. (#507)

What's Changed

• Add back CuPy as an optional test dependency + Fix an example bug by @leofang in #334
• add warning to the nvjitlink ctor when falling back to cuLink by @ksimpson-work in #315
• Set up a preliminary doc build/publish pipeline by @leofang in #325
• Fix doc ci permissions by @leofang in #338
• Add conda installation instructions by @leofang in #321
• Add a dummy email address to the doc bot by @leofang in #343
• multi gpu example by @amwi04 in #304
• Change __cuda_stream__ from attribute to method by @NaderAlAwar in #389
• Ensure deprecation warnings from cuda.bindings are swallowed by @ksimpson-work in #404
• Add the options data class to program by @ksimpson-work in #237
• Update to use the new NVKS runners by @leofang in #426
• Disable notebook execution for cuda.bindings by @vzhurba01 in #424
• Stop tracking cached files and Jupyter notebook hashes in doc builds by @vzhurba01 in #425
• Documentation remove gpu dependency by @ksimpson-work in #398
• handle CTK version specific options in the linker test by @ksimpson-work in #371
• support ptx code type for program by @ksimpson-work in #317
• Update release checklist to focus on subpackages by @vzhurba01 in #427
• Kernel attributes by @ksimpson-work in #360
• Use cpu runner to build docs by @leofang in #434
• device properties by @ksimpson-work in #409
• Update linker options sequence handling by @ksimpson-work in #436
• Expose ObjectCode as public API + prune unnecessary input arguments by @ksimpson-work in #435
• Pin Sphinx <8.2.0 to fix doc build by @leofang in #456
• CI: Add Windows GPU runner for tests by @leofang in #444
• Improve program checks by @ksimpson-work in #394
• Various handle-related changes and improvements by @leofang in #463
• Improve perf of accessing dev.compute_capability by @leofang in #459
• Device properties example by @samaid in #474
• Add ObjectCode ptx constructor by @brandon-b-miller in #470
• Add a Linker example by @vzhurba01 in #475
• Add error log producing test by @ksimpson-work in #423
• Apply __new__ approach to disabling __init__ by @rwgk in #484
• switch the launch argument order by @ksimpson-work in #316
• NEW: Create an Event without recording to Stream by @carterbox in #487
• Clearer error messages (cuda.core) by @rwgk in #458
• Add event timing by @leofang in #481
• Fix merge error by @leofang in #495
• add public handle to object code by @ksimpson-work in #492
• Bump cuda-core ver to v0.2.0 by @leofang in #494
• Increase tolerance in test_timing() to avoid flaky tests. by @rwgk in #498
• Fix test_timing flakiness under Windows by @rwgk in #508
• NEW: Add Event to public API by @carterbox in #501
• cuda.core: Change selected .decode() calls to .decode("utf-8", errors="backslashreplace") by @rwgk in #510
• clean up device initialization in test by @ksimpson-work in #507
• Add @functools.lru_cache decorator for get_binding_version() by @rwgk in #512
• Fix dangling pointer problem in _linker.py by @rwgk in #516
• cuda.core: release notes update by @rwgk in #519
• Set release date by @vzhurba01 in #523

New Contributors

• @amwi04 made their first contribution in #304
• @NaderAlAwar made their first contribution in #389
• @samaid made their first contribution in #474
• @brandon-b-miller made their first contribution in #470
• @carterbox made their first contribution in #487

Full Changelog: cuda-core-v0.1.1...cuda-core-v0.2.0

CUDA Python 12.8.0

Release notes

• https://nvidia.github.io/cuda-python/12.8.0/release/12.8.0-notes.html

Documentation

• https://nvidia.github.io/cuda-python/12.8.0/

CUDA Python 11.8.6

Release notes

• https://nvidia.github.io/cuda-python/12.8.0/release/11.8.6-notes.html