Add numba-cuda-mlir docs, cuda-core dependency, and toolkit extras (#2101)

* Add numba-cuda-mlir to docs, cuda-core/cuda-cccl deps to metapackage, cuda-toolkit extras

- Mention numba-cuda-mlir in README.md and DESCRIPTION.rst above the
  existing numba.cuda entry
- Add cuda-core~=1.0.0 and cuda-cccl~=1.0.0 as required dependencies
  of the cuda-python metapackage (closes #148, closes #691)
- Add bare cuda-toolkit==13.* to cuda-bindings[all] and remove cudla
  from the component-specific extras (closes #903)
- Add cuda-toolkit==12.*/13.* to cuda-core's cu12/cu13 extras

* Add numba-cuda-mlir to cuda-python docs index.rst

* Update install.rst

* Adjust PR 2101 dependencies for 13.3

Drop cuda-cccl from the metapackage until free-threaded Python wheels are available, and keep cuDLA in cuda-bindings[all] via a guarded direct dependency.

* Constrain cuDLA optional dependency to CUDA 13

Keep cuda-bindings[all] consistent with the CUDA 13 toolkit dependencies when installing cuDLA on Linux aarch64.

---------

Co-authored-by: Ralf W. Grosse-Kunstleve <rgrossekunst@nvidia.com>

Author: leofang

README.md

@@ -7,6 +7,7 @@ CUDA Python is the home for accessing NVIDIA’s CUDA platform from Python. It c
 * [cuda.pathfinder](https://nvidia.github.io/cuda-python/cuda-pathfinder/latest): Utilities for locating CUDA components installed in the user's Python environment
 * [cuda.coop](https://nvidia.github.io/cccl/unstable/python/coop.html): A Python module providing CCCL's reusable block-wide and warp-wide *device* primitives for use within Numba CUDA kernels
 * [cuda.compute](https://nvidia.github.io/cccl/unstable/python/compute/index.html): A Python module for easy access to CCCL's highly efficient and customizable parallel algorithms, like `sort`, `scan`, `reduce`, `transform`, etc. that are callable on the *host*
+* [numba-cuda-mlir](https://nvidia.github.io/numba-cuda-mlir/): An evolution of Numba CUDA that improves upon its technical foundation and performance to provide the future of CUDA Python JIT compilation. It currently supports developing CUDA **SIMT** kernels in Python, providing Python bindings for accelerated device libraries, and serving as a compiler for user-defined functions in accelerated libraries.
 * [numba.cuda](https://nvidia.github.io/numba-cuda/): A Python DSL that exposes CUDA **SIMT** programming model and compiles a restricted subset of Python code into CUDA kernels and device functions
 * [cuda.tile](https://docs.nvidia.com/cuda/cutile-python/): A new Python DSL that exposes CUDA **Tile** programming model and allows users to write NumPy-like code in CUDA kernels
 * [nvmath-python](https://docs.nvidia.com/cuda/nvmath-python/latest): Pythonic access to NVIDIA CPU & GPU Math Libraries, with [*host*](https://docs.nvidia.com/cuda/nvmath-python/latest/overview.html#host-apis), [*device*](https://docs.nvidia.com/cuda/nvmath-python/latest/overview.html#device-apis), and [*distributed*](https://docs.nvidia.com/cuda/nvmath-python/latest/distributed-apis/index.html) APIs. It also provides low-level Python bindings to host C APIs ([nvmath.bindings](https://docs.nvidia.com/cuda/nvmath-python/latest/bindings/index.html)).

cuda_bindings/docs/source/install.rst

@@ -12,7 +12,7 @@ Runtime Requirements
 * Linux (x86-64, arm64) and Windows (x86-64)
 * Python 3.10 - 3.14
 * Driver: Linux (580.65.06 or later) Windows (580.88 or later)
-* Optionally, NVRTC, nvJitLink, NVVM, and cuFile from CUDA Toolkit 13.x
+* Optionally, NVRTC, nvJitLink, nvFatBin, NVVM, cuFile, and cuDLA from CUDA Toolkit 13.x
 
 .. note::
 
@@ -52,10 +52,12 @@ Where the optional dependencies include:
 
 * ``nvidia-cuda-nvrtc`` (NVRTC runtime compilation library)
 * ``nvidia-nvjitlink`` (nvJitLink library)
+* ``nvidia-nvfatbin`` (nvFatBin library)
 * ``nvidia-nvvm`` (NVVM library)
 * ``nvidia-cufile`` (cuFile library, Linux only)
+* ``nvidia-cudla`` (cuDLA library, Linux aarch64 only)
 
-These are now installed through the ``cuda-toolkit`` metapackage for improved dependency resolution.
+These are now installed through the ``cuda-toolkit`` metapackage, where available, for improved dependency resolution.
 
 Installing from Conda
 ---------------------
@@ -74,6 +76,10 @@ For example:
 
    $ conda install -c conda-forge cuda-python cuda-version=13
 
+.. note::
+
+   Tegra users can install the cuDLA conda package from conda-forge through ``conda install -c conda-forge libcudla cuda-version=13``, if it does not already exist on the system.
+
 Installing from Source
 ----------------------
 

cuda_bindings/pyproject.toml

@@ -36,8 +36,10 @@ dependencies = ["cuda-pathfinder >=1.4.2"]
 
 [project.optional-dependencies]
 all = [
-    "cuda-toolkit[nvrtc,nvjitlink,nvvm,nvfatbin,cudla]==13.*",
+    "cuda-toolkit[nvrtc,nvjitlink,nvvm,nvfatbin]==13.*",
     "cuda-toolkit[cufile]==13.*; sys_platform == 'linux'",
+    "cuda-toolkit==13.*",
+    "nvidia-cudla==13.*; platform_system == 'Linux' and platform_machine == 'aarch64'",
 ]
 
 [dependency-groups]

cuda_core/pyproject.toml

@@ -54,8 +54,8 @@ dependencies = [
 ]
 
 [project.optional-dependencies]
-cu12 = ["cuda-bindings[all]==12.*"]
-cu13 = ["cuda-bindings[all]==13.*"]
+cu12 = ["cuda-bindings[all]==12.*", "cuda-toolkit==12.*"]
+cu13 = ["cuda-bindings[all]==13.*", "cuda-toolkit==13.*"]
 
 [dependency-groups]
 test = ["cython>=3.2,<3.3", "setuptools", "pytest>=6.2.4", "pytest-benchmark", "pytest-randomly", "pytest-repeat", "pytest-rerunfailures", "pytest-timeout", "cloudpickle", "psutil", "cffi"]

cuda_python/DESCRIPTION.rst

@@ -12,6 +12,7 @@ CUDA Python is the home for accessing NVIDIA's CUDA platform from Python. It con
 * `cuda.pathfinder <https://nvidia.github.io/cuda-python/cuda-pathfinder/latest>`_: Utilities for locating CUDA components installed in the user's Python environment
 * `cuda.coop <https://nvidia.github.io/cccl/unstable/python/coop.html>`_: A Python module providing CCCL's reusable block-wide and warp-wide *device* primitives for use within Numba CUDA kernels
 * `cuda.compute <https://nvidia.github.io/cccl/unstable/python/compute/index.html>`_: A Python module for easy access to CCCL's highly efficient and customizable parallel algorithms, like ``sort``, ``scan``, ``reduce``, ``transform``, etc. that are callable on the *host*
+* `numba-cuda-mlir <https://nvidia.github.io/numba-cuda-mlir/>`_: An evolution of Numba CUDA that improves upon its technical foundation and performance to provide the future of CUDA Python JIT compilation. It currently supports developing CUDA **SIMT** kernels in Python, providing Python bindings for accelerated device libraries, and serving as a compiler for user-defined functions in accelerated libraries.
 * `numba.cuda <https://nvidia.github.io/numba-cuda/>`_: A Python DSL that exposes CUDA **SIMT** programming model and compiles a restricted subset of Python code into CUDA kernels and device functions
 * `cuda.tile <https://docs.nvidia.com/cuda/cutile-python/>`_: A new Python DSL that exposes CUDA **Tile** programming model and allows users to write NumPy-like code in CUDA kernels
 * `nvmath-python <https://docs.nvidia.com/cuda/nvmath-python/latest>`_: Pythonic access to NVIDIA CPU & GPU Math Libraries, with `host <https://docs.nvidia.com/cuda/nvmath-python/latest/overview.html#host-apis>`_, `device <https://docs.nvidia.com/cuda/nvmath-python/latest/overview.html#device-apis>`_, and `distributed <https://docs.nvidia.com/cuda/nvmath-python/latest/distributed-apis/index.html>`_ APIs. It also provides low-level Python bindings to host C APIs (`nvmath.bindings <https://docs.nvidia.com/cuda/nvmath-python/latest/bindings/index.html>`_).

cuda_python/docs/source/index.rst

@@ -12,6 +12,7 @@ multiple components:
 - `cuda.pathfinder`_: Utilities for locating CUDA components installed in the user's Python environment
 - `cuda.coop`_: A Python module providing CCCL's reusable block-wide and warp-wide *device* primitives for use within Numba CUDA kernels
 - `cuda.compute`_: A Python module for easy access to CCCL's highly efficient and customizable parallel algorithms, like ``sort``, ``scan``, ``reduce``, ``transform``, etc. that are callable on the *host*
+- `numba-cuda-mlir`_: An evolution of Numba CUDA that improves upon its technical foundation and performance to provide the future of CUDA Python JIT compilation. It currently supports developing CUDA **SIMT** kernels in Python, providing Python bindings for accelerated device libraries, and serving as a compiler for user-defined functions in accelerated libraries.
 - `numba.cuda`_: A Python DSL that exposes CUDA **SIMT** programming model and compiles a restricted subset of Python code into CUDA kernels and device functions
 - `cuda.tile`_: A new Python DSL that exposes CUDA **Tile** programming model and allows users to write NumPy-like code in CUDA kernels
 - `nvmath-python`_: Pythonic access to NVIDIA CPU & GPU Math Libraries, with `host`_, `device`_, and `distributed`_ APIs. It also provides low-level Python bindings to host C APIs (`nvmath.bindings`_).
@@ -22,6 +23,7 @@ multiple components:
 
 .. _cuda.coop: https://nvidia.github.io/cccl/unstable/python/coop.html
 .. _cuda.compute: https://nvidia.github.io/cccl/unstable/python/compute/index.html
+.. _numba-cuda-mlir: https://nvidia.github.io/numba-cuda-mlir/
 .. _numba.cuda: https://nvidia.github.io/numba-cuda/
 .. _cuda.tile: https://docs.nvidia.com/cuda/cutile-python/
 .. _nvmath-python: https://docs.nvidia.com/cuda/nvmath-python/latest
@@ -52,6 +54,7 @@ be available, please refer to the `cuda.bindings`_ documentation for installatio
    cuda.pathfinder <https://nvidia.github.io/cuda-python/cuda-pathfinder/latest>
    cuda.coop <https://nvidia.github.io/cccl/unstable/python/coop.html>
    cuda.compute <https://nvidia.github.io/cccl/unstable/python/compute/index.html>
+   numba-cuda-mlir <https://nvidia.github.io/numba-cuda-mlir/>
    numba.cuda <https://nvidia.github.io/numba-cuda/>
    cuda.tile <https://docs.nvidia.com/cuda/cutile-python/>
    nvmath-python <https://docs.nvidia.com/cuda/nvmath-python/>

cuda_python/setup.py

@@ -32,6 +32,7 @@
     version=version,
     install_requires=[
         f"cuda-bindings{matcher}{version}",
+        "cuda-core~=1.0.0",
         "cuda-pathfinder~=1.1",
     ],
     extras_require={