Merge branch 'main' into ipc-mempool

Author: leofang

.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

@@ -101,7 +101,7 @@ def get_cuda_paths():
         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,
     )
 

cuda_core/cuda/core/experimental/__init__.py

@@ -55,6 +55,8 @@
     DeviceMemoryResourceOptions,
     LegacyPinnedMemoryResource,
     MemoryResource,
+    VirtualMemoryResource,
+    VirtualMemoryResourceOptions,
 )
 from cuda.core.experimental._module import Kernel, ObjectCode  # noqa: E402
 from cuda.core.experimental._program import Program, ProgramOptions  # noqa: E402