Fix #702: Update cyruntime.getLocalRuntimeVersion to use pathfinder (#929)

* Fix #702: Update cyruntime.getLocalRuntimeVersion to use pathfinder

* Try to fix Windows build

* Fix Windows

* More realistic minimum version

* Simplify gil handling

* Unload module when done with it

* Don't need try/except on Windows

* Don't need gil

* Fix message

* Fix test

* DynamicLibrary -> DynamicLib

* Update to not use pywin32

* Add cast

* Try to fix cast again

* Try to fix cast again

* Address comments in PR

* Fix lib -> loaded_dl

* Add changelog

Author: mdboom

cuda_bindings/cuda/bindings/_lib/windll.pxd

@@ -12,6 +12,7 @@ cdef extern from "windows.h" nogil:
     ctypedef unsigned long DWORD
     ctypedef const wchar_t *LPCWSTR
     ctypedef const char *LPCSTR
+    ctypedef int BOOL
 
     cdef DWORD LOAD_LIBRARY_SEARCH_SYSTEM32 = 0x00000800
 
@@ -23,6 +24,8 @@ cdef extern from "windows.h" nogil:
 
     FARPROC _GetProcAddress "GetProcAddress"(HMODULE hModule, LPCSTR lpProcName)
 
+    BOOL _FreeLibrary "FreeLibrary"(HMODULE hLibModule)
+
 cdef inline uintptr_t LoadLibraryExW(str path, HANDLE hFile, DWORD dwFlags):
     cdef uintptr_t result
     cdef wchar_t* wpath = PyUnicode_AsWideCharString(path, NULL)
@@ -37,3 +40,6 @@ cdef inline uintptr_t LoadLibraryExW(str path, HANDLE hFile, DWORD dwFlags):
 
 cdef inline FARPROC GetProcAddress(uintptr_t hModule, const char* lpProcName) nogil:
     return _GetProcAddress(<HMODULE>hModule, lpProcName)
+
+cdef inline BOOL FreeLibrary(uintptr_t hLibModule) nogil:
+    return _FreeLibrary(<HMODULE>hLibModule)

cuda_bindings/cuda/bindings/cyruntime.pyx.in

@@ -1885,35 +1885,46 @@ cdef cudaError_t cudaGraphicsVDPAURegisterOutputSurface(cudaGraphicsResource** r
 
 {{if True}}
 
-{{if 'Windows' != platform.system()}}
+from libc.stdint cimport uintptr_t
+from cuda.pathfinder import load_nvidia_dynamic_lib
+{{if 'Windows' == platform.system()}}
+cimport cuda.bindings._lib.windll as windll
+{{else}}
 cimport cuda.bindings._lib.dlfcn as dlfcn
 {{endif}}
 
 cdef cudaError_t getLocalRuntimeVersion(int* runtimeVersion) except ?cudaErrorCallRequiresNewerDriver nogil:
-    {{if 'Windows' == platform.system()}}
-    with gil:
-        raise NotImplementedError('"getLocalRuntimeVersion" is unsupported on Windows')
-    {{else}}
+    
     # Load
-    handle = dlfcn.dlopen('libcudart.so.13', dlfcn.RTLD_NOW)
-    if handle == NULL:
-        with gil:
-            raise RuntimeError(f'Failed to dlopen libcudart.so.13')
+    with gil:
+        loaded_dl = load_nvidia_dynamic_lib("cudart")
+        {{if 'Windows' == platform.system()}}
+        handle = <uintptr_t>loaded_dl._handle_uint
+        {{else}}
+        handle = <void *><uintptr_t>loaded_dl._handle_uint
+        {{endif}}
 
+    {{if 'Windows' == platform.system()}}
+    __cudaRuntimeGetVersion = windll.GetProcAddress(handle, b'cudaRuntimeGetVersion')
+    {{else}}
     __cudaRuntimeGetVersion = dlfcn.dlsym(handle, 'cudaRuntimeGetVersion')
+    {{endif}}
 
     if __cudaRuntimeGetVersion == NULL:
         with gil:
-            raise RuntimeError(f'Function "cudaRuntimeGetVersion" not found in libcudart.so.13')
+            raise RuntimeError(f'Function "cudaRuntimeGetVersion" not found in {loaded_dl.abs_path}')
 
     # Call
     cdef cudaError_t err = cudaSuccess
     err = (<cudaError_t (*)(int*) except ?cudaErrorCallRequiresNewerDriver nogil> __cudaRuntimeGetVersion)(runtimeVersion)
 
     # Unload
+    {{if 'Windows' == platform.system()}}
+    windll.FreeLibrary(handle) 
+    {{else}}
     dlfcn.dlclose(handle)
+    {{endif}}
 
     # Return
     return err
-    {{endif}}
 {{endif}}

cuda_bindings/docs/source/release/12.9.X-notes.rst

@@ -15,6 +15,7 @@ Highlights
 * Automatic CUDA library path detection based on ``CUDA_HOME``, eliminating the need to manually set ``LIBRARY_PATH`` environment variables for installation.
 * The Python overhead of calling functions in CUDA bindings in ``driver``, ``runtime`` and ``nvrtc`` has been reduced by approximately 30%.
 * Updated the ``cuda.bindings.runtime`` module to statically link against the CUDA Runtime library from CUDA Toolkit 12.9.1.
+* ``cyruntime.getLocalRuntimeVersion`` now uses pathfinder to find the CUDA runtime.
 
 
 Known issues

cuda_bindings/docs/source/release/13.X.Y-notes.rst

@@ -18,6 +18,7 @@ Highlights
 * The Python overhead of calling functions in CUDA bindings in ``driver``, ``runtime`` and ``nvrtc`` has been reduced by approximately 30%.
 * On Windows, the ``pywin32`` dependency has been removed. The necessary Windows API functions are now accessed directly.
 * Updated the ``cuda.bindings.runtime`` module to statically link against the CUDA Runtime library from CUDA Toolkit 13.0.1.
+* ``cyruntime.getLocalRuntimeVersion`` now uses pathfinder to find the CUDA runtime.
 
 
 Known issues

cuda_bindings/tests/test_cudart.py

@@ -9,6 +9,7 @@
 
 import cuda.bindings.driver as cuda
 import cuda.bindings.runtime as cudart
+from cuda import pathfinder
 from cuda.bindings import runtime
 
 
@@ -1400,3 +1401,13 @@ def test_struct_pointer_comparison(target):
     c = target(456)
     assert a != c
     assert hash(a) != hash(c)
+
+
+def test_getLocalRuntimeVersion():
+    try:
+        err, version = cudart.getLocalRuntimeVersion()
+    except pathfinder.DynamicLibNotFoundError:
+        pytest.skip("cudart dynamic lib not available")
+    else:
+        assertSuccess(err)
+        assert version >= 12000  # CUDA 12.0