Merge branch 'main' into fix-slow-patterngen-test

Author: Andy-Jost

cuda_core/tests/helpers/__init__.py

@@ -1,9 +1,13 @@
 # SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+import functools
 import os
 import pathlib
 import sys
+from typing import Union
+
+from cuda.core._utils.cuda_utils import handle_return
 
 CUDA_PATH = os.environ.get("CUDA_PATH")
 CUDA_INCLUDE_PATH = None
@@ -24,3 +28,38 @@
     # Import shared platform helpers for tests across repos
     sys.path.insert(0, str(pathlib.Path(__file__).resolve().parents[3] / "cuda_python_test_helpers"))
     from cuda_python_test_helpers import *  # noqa: F403
+
+
+@functools.cache
+def supports_ipc_mempool(device_id: Union[int, object]) -> bool:
+    """Return True if mempool IPC via POSIX file descriptor is supported.
+
+    Uses cuDeviceGetAttribute(CU_DEVICE_ATTRIBUTE_MEMPOOL_SUPPORTED_HANDLE_TYPES)
+    to check for CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR support. Does not
+    require an active CUDA context.
+    """
+    if IS_WSL:  # noqa: F405
+        return False
+
+    try:
+        # Lazy import to avoid hard dependency when not running GPU tests
+        try:
+            from cuda.bindings import driver  # type: ignore
+        except Exception:
+            from cuda import cuda as driver  # type: ignore
+
+        # Initialize CUDA
+        handle_return(driver.cuInit(0))
+
+        # Resolve device id from int or Device-like object
+        dev_id = int(getattr(device_id, "device_id", device_id))
+
+        # Query supported mempool handle types bitmask
+        attr = driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_MEMPOOL_SUPPORTED_HANDLE_TYPES
+        mask = handle_return(driver.cuDeviceGetAttribute(attr, dev_id))
+
+        # Check POSIX FD handle type support via bitmask
+        posix_fd = driver.CUmemAllocationHandleType.CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR
+        return (int(mask) & int(posix_fd)) != 0
+    except Exception:
+        return False

cuda_core/tests/test_memory.py

@@ -38,15 +38,14 @@
 from cuda.core._memory import IPCBufferDescriptor
 from cuda.core._utils.cuda_utils import CUDAError, handle_return
 from cuda.core.utils import StridedMemoryView
-from helpers import IS_WINDOWS
+from helpers import IS_WINDOWS, supports_ipc_mempool
 from helpers.buffers import DummyUnifiedMemoryResource
 
 from conftest import (
     create_managed_memory_resource_or_skip,
     skip_if_managed_memory_unsupported,
     skip_if_pinned_memory_unsupported,
 )
-from cuda_python_test_helpers import supports_ipc_mempool
 
 POOL_SIZE = 2097152  # 2MB size
 

cuda_pathfinder/cuda/pathfinder/__init__.py

@@ -3,8 +3,6 @@
 
 """cuda.pathfinder public APIs"""
 
-from cuda.pathfinder._version import __version__  # noqa: F401
-
 from cuda.pathfinder._dynamic_libs.load_dl_common import DynamicLibNotFoundError as DynamicLibNotFoundError
 from cuda.pathfinder._dynamic_libs.load_dl_common import LoadedDL as LoadedDL
 from cuda.pathfinder._dynamic_libs.load_nvidia_dynamic_lib import load_nvidia_dynamic_lib as load_nvidia_dynamic_lib
@@ -14,6 +12,8 @@
 from cuda.pathfinder._headers.find_nvidia_headers import find_nvidia_header_directory as find_nvidia_header_directory
 from cuda.pathfinder._headers.supported_nvidia_headers import SUPPORTED_HEADERS_CTK as _SUPPORTED_HEADERS_CTK
 
+from cuda.pathfinder._version import __version__  # isort: skip  # noqa: F401
+
 # Indirections to help Sphinx find the docstrings.
 #: Mapping from short CUDA Toolkit (CTK) library names to their canonical
 #: header basenames (used to validate a discovered include directory).

cuda_python_test_helpers/cuda_python_test_helpers/__init__.py

@@ -2,20 +2,15 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import ctypes
-import functools
 import os
 import platform
 import sys
 from contextlib import suppress
-from typing import Union
-
-from cuda.core._utils.cuda_utils import handle_return
 
 __all__ = [
     "IS_WINDOWS",
     "IS_WSL",
     "libc",
-    "supports_ipc_mempool",
     "under_compute_sanitizer",
 ]
 
@@ -68,38 +63,3 @@ def under_compute_sanitizer() -> bool:
     # Another common indicator: sanitizer injectors are configured via env vars.
     inj = os.environ.get("CUDA_INJECTION64_PATH", "")
     return "compute-sanitizer" in inj or "cuda-memcheck" in inj
-
-
-@functools.cache
-def supports_ipc_mempool(device_id: Union[int, object]) -> bool:
-    """Return True if mempool IPC via POSIX file descriptor is supported.
-
-    Uses cuDeviceGetAttribute(CU_DEVICE_ATTRIBUTE_MEMPOOL_SUPPORTED_HANDLE_TYPES)
-    to check for CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR support. Does not
-    require an active CUDA context.
-    """
-    if _detect_wsl():
-        return False
-
-    try:
-        # Lazy import to avoid hard dependency when not running GPU tests
-        try:
-            from cuda.bindings import driver  # type: ignore
-        except Exception:
-            from cuda import cuda as driver  # type: ignore
-
-        # Initialize CUDA
-        handle_return(driver.cuInit(0))
-
-        # Resolve device id from int or Device-like object
-        dev_id = int(getattr(device_id, "device_id", device_id))
-
-        # Query supported mempool handle types bitmask
-        attr = driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_MEMPOOL_SUPPORTED_HANDLE_TYPES
-        mask = handle_return(driver.cuDeviceGetAttribute(attr, dev_id))
-
-        # Check POSIX FD handle type support via bitmask
-        posix_fd = driver.CUmemAllocationHandleType.CU_MEM_HANDLE_TYPE_POSIX_FILE_DESCRIPTOR
-        return (int(mask) & int(posix_fd)) != 0
-    except Exception:
-        return False