Merge branch 'main' into cuda_core_build_hooks_get_cuda_version_macro

Author: leofang

.github/actions/fetch_ctk/action.yml

@@ -140,7 +140,7 @@ runs:
         ls -l $CUDA_PATH
 
     - name: Upload CTK cache
-      if: ${{ always() &&
+      if: ${{ !cancelled() &&
               steps.ctk-get-cache.outputs.cache-hit != 'true' }}
       uses: actions/cache/save@5a3ec84eff668545956fd18022155c47e93e2684  # v4.2.3
       with:

.github/workflows/build-wheel.yml

@@ -230,8 +230,6 @@ jobs:
         with:
           # workaround for actions/runner-images#12377 (the cached 3.13.4 is buggy on Windows)
           python-version: ${{ matrix.python-version == '3.13' && '3.13.5' || matrix.python-version }}
-          # allow prereleases only fetches prereleases for unreleased versions of Python
-          allow-prereleases: true
 
       - name: verify free-threaded build
         if: endsWith(matrix.python-version, 't')

.github/workflows/ci.yml

@@ -193,6 +193,7 @@ jobs:
 
   checks:
     name: Check job status
+    if: always()
     runs-on: ubuntu-latest
     needs:
       - test-linux-64

.github/workflows/test-wheel-linux.yml

@@ -228,7 +228,6 @@ jobs:
         uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065  # v5.6.0
         with:
           python-version: ${{ matrix.PY_VER }}
-          allow-prereleases: true
         env:
           # we use self-hosted runners on which setup-python behaves weirdly...
           AGENT_TOOLSDIRECTORY: "/opt/hostedtoolcache"

.github/workflows/test-wheel-windows.yml

@@ -213,7 +213,6 @@ jobs:
         uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065  # v5.6.0
         with:
           python-version: ${{ matrix.PY_VER }}
-          allow-prereleases: true
 
       - name: Set up mini CTK
         if: ${{ matrix.LOCAL_CTK == '1' }}

cuda_bindings/docs/nv-versions.json

@@ -3,6 +3,10 @@
         "version": "latest",
         "url": "https://nvidia.github.io/cuda-python/cuda-bindings/latest/"
     },
+    {
+        "version": "13.0.2",
+        "url": "https://nvidia.github.io/cuda-python/cuda-bindings/13.0.2/"
+    },
     {
         "version": "13.0.1",
         "url": "https://nvidia.github.io/cuda-python/cuda-bindings/13.0.1/"

cuda_bindings/setup.py

@@ -329,7 +329,7 @@ def do_cythonize(extensions):
     return cythonize(
         extensions,
         nthreads=nthreads,
-        compiler_directives=dict(language_level=3, embedsignature=True, binding=True),
+        compiler_directives=dict(language_level=3, embedsignature=True, binding=True, freethreading_compatible=True),
         **extra_cythonize_kwargs,
     )
 

cuda_bindings/tests/cython/build_tests.bat

@@ -5,5 +5,5 @@ REM SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 
 setlocal
 	set CL=%CL% /I"%CUDA_HOME%\include"
-	cythonize -3 -i %~dp0test_*.pyx
+	cythonize -3 -i -Xfreethreading_compatible=True %~dp0test_*.pyx
 endlocal

cuda_bindings/tests/cython/build_tests.sh

@@ -14,4 +14,4 @@ else
   exit 1
 fi
 
-cythonize -3 -i ${SCRIPTPATH}/test_*.pyx
+cythonize -3 -i -Xfreethreading_compatible=True ${SCRIPTPATH}/test_*.pyx

cuda_core/build_hooks.py

@@ -6,8 +6,6 @@
 # - https://peps.python.org/pep-0517/
 # - https://setuptools.pypa.io/en/latest/build_meta.html#dynamic-build-dependencies-and-other-build-meta-tweaks
 # Specifically, there are 5 APIs required to create a proper build backend, see below.
-#
-# TODO: also implement PEP-660 API hooks
 
 import ctypes
 import functools
@@ -21,6 +19,7 @@
 from setuptools import Extension
 from setuptools import build_meta as _build_meta
 
+prepare_metadata_for_build_editable = _build_meta.prepare_metadata_for_build_editable
 prepare_metadata_for_build_wheel = _build_meta.prepare_metadata_for_build_wheel
 build_sdist = _build_meta.build_sdist
 get_requires_for_build_sdist = _build_meta.get_requires_for_build_sdist
@@ -165,10 +164,13 @@ def _get_proper_cuda_bindings_major_version() -> str:
 _extensions = None
 
 
-def build_wheel(wheel_directory, config_settings=None, metadata_directory=None):
-    # Customizing this hook is needed because we must defer cythonization until cuda-bindings,
+def _build_cuda_core():
+    # Customizing the build hooks is needed because we must defer cythonization until cuda-bindings,
     # now a required build-time dependency that's dynamically installed via the other hook below,
     # is installed. Otherwise, cimport any cuda.bindings modules would fail!
+    #
+    # This function populates "_extensions".
+    global _extensions
 
     # It seems setuptools' wildcard support has problems for namespace packages,
     # so we explicitly spell out all Extension instances.
@@ -193,21 +195,36 @@ def strip_prefix_suffix(filename):
 
     nthreads = int(os.environ.get("CUDA_PYTHON_PARALLEL_LEVEL", os.cpu_count() // 2))
     compile_time_env = {"CUDA_CORE_BUILD_MAJOR": _get_proper_cuda_bindings_major_version()}
-
-    global _extensions
     _extensions = cythonize(
         ext_modules,
         verbose=True,
         language_level=3,
         nthreads=nthreads,
-        compiler_directives={"embedsignature": True, "warn.deprecated.IF": False},
+        compiler_directives={"embedsignature": True, "warn.deprecated.IF": False, "freethreading_compatible": True},
         compile_time_env=compile_time_env,
     )
 
+    return
+
+
+def build_editable(wheel_directory, config_settings=None, metadata_directory=None):
+    _build_cuda_core()
+    return _build_meta.build_editable(wheel_directory, config_settings, metadata_directory)
+
+
+def build_wheel(wheel_directory, config_settings=None, metadata_directory=None):
+    _build_cuda_core()
     return _build_meta.build_wheel(wheel_directory, config_settings, metadata_directory)
 
 
-def get_requires_for_build_wheel(config_settings=None):
+def _get_cuda_bindings_require():
     cuda_major = _get_proper_cuda_bindings_major_version()
-    cuda_bindings_require = [f"cuda-bindings=={cuda_major}.*"]
-    return _build_meta.get_requires_for_build_wheel(config_settings) + cuda_bindings_require
+    return [f"cuda-bindings=={cuda_major}.*"]
+
+
+def get_requires_for_build_editable(config_settings=None):
+    return _build_meta.get_requires_for_build_editable(config_settings) + _get_cuda_bindings_require()
+
+
+def get_requires_for_build_wheel(config_settings=None):
+    return _build_meta.get_requires_for_build_wheel(config_settings) + _get_cuda_bindings_require()