.. SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
.. SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE

.. module:: cuda.bindings

``cuda-bindings`` 13.3.0 Release notes
======================================

Highlights
----------

* Support for new APIs introduced in CUDA 13.3, including driver logical
  endpoint APIs, graph recapture APIs, NVRTC Tile IR and bundled-header APIs,
  and related runtime graph/event APIs.
  (`PR #2139 <https://github.com/NVIDIA/cuda-python/pull/2139>`_)

* Add ``cuda.bindings.cudla`` bindings.
  (`PR #2034 <https://github.com/NVIDIA/cuda-python/pull/2034>`_)

* Add the ``nvvmLLVMVersion`` binding.
  (`PR #1774 <https://github.com/NVIDIA/cuda-python/pull/1774>`_)

* Add additional NVML APIs introduced in CUDA 13.2.
  (`PR #1830 <https://github.com/NVIDIA/cuda-python/pull/1830>`_)

Bugfixes
--------

* Fixed the ``cuDevSmResourceSplit`` and ``cudaDevSmResourceSplit`` binding
  signatures so ``groupParams`` is accepted as a sequence matching the CUDA API.
  (`PR #1766 <https://github.com/NVIDIA/cuda-python/pull/1766>`_)

* Fixed nested resource pointer handling to accept both ``str`` and ``bytes``
  inputs.
  (`PR #1698 <https://github.com/NVIDIA/cuda-python/pull/1698>`_)

* Fixed ``nvmlDeviceGetFieldValues`` and ``nvmlDeviceClearFieldValues`` handling
  of empty field lists so they return empty results instead of raising
  ``NVML_ERROR_INVALID_ARGUMENT``.
  (`PR #1982 <https://github.com/NVIDIA/cuda-python/pull/1982>`_)

* Fixed ``CUDA_PYTHON_CUDA_PER_THREAD_DEFAULT_STREAM=0`` incorrectly enabling
  per-thread default stream mode.
  (`PR #2076 <https://github.com/NVIDIA/cuda-python/pull/2076>`_)

* Fixed a use-after-free in ``cudaGraphGetEdges``, ``cudaGraphNodeGetDependencies``,
  ``cudaGraphNodeGetDependentNodes``, ``cudaStreamGetCaptureInfo``, and their
  driver-API counterparts (``cuGraphGetEdges``, ``cuGraphNodeGetDependencies``,
  ``cuGraphNodeGetDependentNodes``, ``cuStreamGetCaptureInfo``). The returned
  ``cudaGraphEdgeData``/``CUgraphEdgeData`` wrappers were backed by a scratch
  buffer that was freed before the call returned, leaving every wrapper holding
  a dangling pointer. The returned wrappers now own deep copies of the edge
  data.
  (`Issue #1804 <https://github.com/NVIDIA/cuda-python/issues/1804>`_,
  `PR #2083 <https://github.com/NVIDIA/cuda-python/pull/2083>`_)

* Fixed a double-free in the generated setters for list-valued struct members
  (e.g. ``CUlaunchConfig.attrs``, ``CUDA_MEM_ALLOC_NODE_PARAMS.accessDescs``,
  external-semaphore and batch-mem-op node parameter arrays, and their runtime
  counterparts). Assigning an empty list freed the internal buffer but left
  the cached pointer non-NULL, so a subsequent assignment or ``__dealloc__``
  would call ``free()`` again on the dangling pointer.
  (`PR #2112 <https://github.com/NVIDIA/cuda-python/pull/2112>`_)

Miscellaneous
-------------

* Add ``cuda.bindings.utils.check_nvvm_compiler_options()`` to check whether a
  set of NVVM compiler options is supported by the installed NVVM library.
  (`PR #1837 <https://github.com/NVIDIA/cuda-python/pull/1837>`_)

* NVRTC bindings now use pre-generated Cython files and no longer require
  pyclibrary header parsing at build time.
  (`PR #1900 <https://github.com/NVIDIA/cuda-python/pull/1900>`_)

* Improved generated documentation and argument names, including the ``ind_ex``
  argument naming bug.
  (`PR #1927 <https://github.com/NVIDIA/cuda-python/pull/1927>`_,
  `PR #2082 <https://github.com/NVIDIA/cuda-python/pull/2082>`_)

* Fixed ``cuda-bindings`` debug builds.
  (`PR #1890 <https://github.com/NVIDIA/cuda-python/pull/1890>`_)

* Declare ``cuda-pathfinder`` as a host dependency for pixi path-dependency
  builds of ``cuda-bindings``.
  (`PR #1926 <https://github.com/NVIDIA/cuda-python/pull/1926>`_)

Known issues
------------

* Updating from older versions (v12.6.2.post1 and below) via ``pip install -U cuda-python`` might not work. Please do a clean re-installation by uninstalling ``pip uninstall -y cuda-python`` followed by installing ``pip install cuda-python``.
* ``nvml.system_get_process_name`` on WSL can return incorrect values.  To work around this, set the locale to "C" before calling ``nvml.device_get_compute_running_processes_v3`` (which sets the process names) and before calling ``nvml.system_get_process_name``. ``cuda_core`` does this automatically, but users of the raw NVML API will need to do this manually.