Merge remote-tracking branch 'origin/main' into cythonize-linker

Co-authored-by: Cursor <cursoragent@cursor.com>

# Conflicts:
#	cuda_core/cuda/core/_cpp/resource_handles.cpp
#	cuda_core/cuda/core/_cpp/resource_handles.hpp
#	cuda_core/cuda/core/_program.pxd
#	cuda_core/cuda/core/_resource_handles.pxd
#	cuda_core/cuda/core/_resource_handles.pyx
#	cuda_core/cuda/core/_utils/cuda_utils.pxd
#	cuda_core/cuda/core/_utils/cuda_utils.pyx
#	cuda_core/docs/source/release/0.6.x-notes.rst

Author: Andy-Jost

cuda_bindings/MANIFEST.in

@@ -1,7 +1,7 @@
 # SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 
-recursive-include cuda/ *.pyx *.pxd *.pxi
+recursive-include cuda/ *.pyx *.pxd *.pxi *.pyx.in *.pxd.in *.pxi.in *.h
 # at least with setuptools 75.0.0 this folder was added erroneously
 # to the payload, causing file copying to the build environment failed
 exclude cuda/bindings cuda?bindings

cuda_bindings/cuda/bindings/_internal/utils.pyx

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 
@@ -8,6 +8,28 @@ from libcpp.utility cimport move
 from cython.operator cimport dereference as deref
 
 
+cdef extern from *:
+    """
+    #if defined(__clang__)
+        #define _COMPILER_VERSION ("Clang " __clang_version__)
+    #elif defined(__GNUC__) || defined(__GNUG__)
+        #define _COMPILER_VERSION ("GCC " __VERSION__)
+    #elif defined(_MSC_VER)
+        #define _COMPILER_VERSION ("MSVC " Py_STRINGIFY(_MSC_VER))
+    #else
+        #define _COMPILER_VERSION ("Unknown Compiler")
+    #endif
+    """
+    cdef char *_COMPILER_VERSION
+
+
+cpdef str get_c_compiler():
+    """
+    Returns a string describing the C compiler used to build cuda.bindings
+    """
+    return _COMPILER_VERSION.decode()
+
+
 cdef bint is_nested_sequence(data):
     if not cpython.PySequence_Check(data):
         return False

cuda_bindings/docs/source/release/13.1.2-notes.rst

@@ -1,4 +1,4 @@
-.. SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+.. SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 .. SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 
 .. module:: cuda.bindings
@@ -22,6 +22,13 @@ Bugfixes
   (`PR #1336 <https://github.com/NVIDIA/cuda-python/pull/1336>`_)
 * Fixed ABI incompatibility bugs in cuFILE bindings introduced in v13.1.0.
 
+Miscellaneous
+-------------
+
+* Wheel and installed package sizes significantly reduced (e.g., on a typical Linux x86_64
+  build, wheel from ~16.6 MB to ~5.7 MB and installed from ~152 MB to ~23 MB) by excluding
+  Cython source files, generated C++ files, and template files from distribution packages.
+
 Known issues
 ------------
 

cuda_bindings/pyproject.toml

@@ -56,8 +56,13 @@ Documentation = "https://nvidia.github.io/cuda-python/"
 [tool.setuptools.packages.find]
 include = ["cuda*"]
 
-[tool.setuptools.exclude-package-data]
-"*" = ["*.cpp"]
+[tool.setuptools]
+include-package-data = false
+
+[tool.setuptools.package-data]
+"*" = ["*.pxd", "*.pxi"]
+"cuda.bindings._bindings" = ["*.h"]
+"cuda.bindings._lib" = ["*.h"]
 
 [tool.setuptools.dynamic]
 readme = { file = ["DESCRIPTION.rst"], content-type = "text/x-rst" }

cuda_bindings/tests/nvml/test_device.py

@@ -71,7 +71,7 @@ def test_get_nv_link_supported_bw_modes(all_devices):
     for device in all_devices:
         with unsupported_before(device, None):
             modes = nvml.device_get_nvlink_supported_bw_modes(device)
-        assert isinstance(modes, nvml.NvLinkSupportedBWModes)
+        assert isinstance(modes, nvml.NvLinkSupportedBWModes_v1)
         # #define NVML_NVLINK_TOTAL_SUPPORTED_BW_MODES 23
         assert len(modes.bw_modes) <= 23
         assert not hasattr(modes, "total_bw_modes")

cuda_bindings/tests/test_utils.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 
 import importlib
@@ -9,6 +9,7 @@
 
 import pytest
 from cuda.bindings import driver, runtime
+from cuda.bindings._internal.utils import get_c_compiler
 from cuda.bindings.utils import get_cuda_native_handle, get_minimal_required_cuda_ver_from_ptx_ver, get_ptx_ver
 
 have_cufile = importlib.util.find_spec("cuda.bindings.cufile") is not None
@@ -110,3 +111,9 @@ def test_cyclical_imports(module):
     subprocess.check_call(  # noqa: S603
         [sys.executable, Path(__file__).parent / "utils" / "check_cyclical_import.py", f"cuda.bindings.{module}"],
     )
+
+
+def test_get_c_compiler():
+    c_compiler = get_c_compiler()
+    prefix = ("GCC", "Clang", "MSVC", "Unknown")
+    assert sum(c_compiler.startswith(p) for p in prefix) == 1

cuda_core/MANIFEST.in

@@ -4,3 +4,4 @@
 
 recursive-include cuda/core *.pyx *.pxd *.pxi
 recursive-include cuda/core/_cpp *.cpp *.hpp
+recursive-include cuda/core/_include *.h *.hpp

cuda_core/docs/source/release/0.6.x-notes.rst

@@ -17,8 +17,8 @@ Breaking Changes
 ----------------
 
 - Building ``cuda.core`` from source now requires ``cuda-bindings`` >= 12.9.0, due to Cython-level
-  dependencies on the NVVM bindings (``cynvvm``). Pre-built wheels are unaffected. The previous
-  minimum was 12.8.0.
+  dependencies on the NVVM and nvJitLink bindings (``cynvvm``, ``cynvjitlink``). Pre-built wheels
+  are unaffected. The previous minimum was 12.8.0.
 
 
 New features
@@ -36,4 +36,11 @@ None.
 Fixes and enhancements
 ----------------------
 
-None.
+- Reduced Python overhead in :class:`Program` and :class:`Linker` by moving compilation and
+  linking operations to the C level and releasing the GIL during backend calls. This benefits
+  workloads that create many programs or linkers, and enables concurrent compilation in
+  multithreaded applications.
+- Wheel and installed package sizes significantly reduced (e.g., on a typical Linux x86_64
+  build, wheel from ~4.6 MB to ~1.6 MB and installed from ~26 MB to ~4.4 MB) by excluding
+  Cython source files, generated C++ files, and other build artifacts from distribution
+  packages.

cuda_core/pyproject.toml

@@ -72,9 +72,13 @@ issues = "https://github.com/NVIDIA/cuda-python/issues/"
 [tool.setuptools.packages.find]
 include = ["cuda.core*"]
 
+[tool.setuptools]
+include-package-data = false
+
 [tool.setuptools.package-data]
-"cuda.core._include" = ["*.h", "*.hpp", "*.cuh"]
-"cuda.core._cpp" = ["*.cpp", "*.hpp"]
+"*" = ["*.pxd"]
+"cuda.core._include" = ["*.h", "*.hpp"]
+"cuda.core._cpp" = ["*.hpp"]
 
 [tool.setuptools.dynamic]
 readme = { file = ["DESCRIPTION.rst"], content-type = "text/x-rst" }

cuda_pathfinder/cuda/pathfinder/_utils/find_sub_dirs.py

@@ -50,4 +50,9 @@ def find_sub_dirs_sys_path(sub_dirs: Sequence[str]) -> list[str]:
 
 
 def find_sub_dirs_all_sitepackages(sub_dirs: Sequence[str]) -> list[str]:
-    return find_sub_dirs((site.getusersitepackages(), *site.getsitepackages()), sub_dirs)
+    parent_dirs = list(site.getsitepackages())
+    if site.ENABLE_USER_SITE:
+        user_site = site.getusersitepackages()
+        if user_site:
+            parent_dirs.insert(0, user_site)
+    return find_sub_dirs(parent_dirs, sub_dirs)