Merge branch 'main' into fix-docs-build-on-tag-trigger

Author: rwgk

cuda_bindings/cuda/bindings/runtime.pyx.in

@@ -21465,7 +21465,7 @@ def cudaDeviceRegisterAsyncNotification(int device, callbackFunc, userData):
 
     if err != cyruntime.cudaSuccess:
         return (_cudaError_t(err), None)
-    return (_cudaError_t(err), callback)
+    return (_cudaError_t_SUCCESS, callback)
 {{endif}}
 
 {{if 'cudaDeviceUnregisterAsyncNotification' in found_functions}}

cuda_core/cuda/core/_context.pyx

@@ -39,6 +39,10 @@ cdef class Context:
             return None
         return as_py(self._h_context)
 
+    @property
+    def _handle(self):
+        return self.handle
+
     def __eq__(self, other):
         if not isinstance(other, Context):
             return NotImplemented

cuda_core/cuda/core/_cpp/resource_handles.cpp

@@ -750,7 +750,8 @@ LibraryHandle create_library_handle_from_file(const char* path) {
         new LibraryBox{library},
         [](const LibraryBox* b) {
             GILReleaseGuard gil;
-            p_cuLibraryUnload(b->resource);
+            // TODO: re-enable once LibraryBox tracks its owning context
+            // p_cuLibraryUnload(b->resource);
             delete b;
         }
     );
@@ -768,7 +769,8 @@ LibraryHandle create_library_handle_from_data(const void* data) {
         new LibraryBox{library},
         [](const LibraryBox* b) {
             GILReleaseGuard gil;
-            p_cuLibraryUnload(b->resource);
+            // TODO: re-enable once LibraryBox tracks its owning context
+            // p_cuLibraryUnload(b->resource);
             delete b;
         }
     );

cuda_core/cuda/core/_event.pyx

@@ -295,9 +295,9 @@ cdef class IPCEventDescriptor:
     def __init__(self, *arg, **kwargs):
         raise RuntimeError("IPCEventDescriptor objects cannot be instantiated directly. Please use Event APIs.")
 
-    @classmethod
-    def _init(cls, reserved: bytes, busy_waited: cython.bint):
-        cdef IPCEventDescriptor self = IPCEventDescriptor.__new__(cls)
+    @staticmethod
+    def _init(reserved: bytes, busy_waited: cython.bint):
+        cdef IPCEventDescriptor self = IPCEventDescriptor.__new__(IPCEventDescriptor)
         self._reserved = reserved
         self._busy_waited = busy_waited
         return self
@@ -307,7 +307,7 @@ cdef class IPCEventDescriptor:
         return self._reserved == rhs._reserved
 
     def __reduce__(self):
-        return self._init, (self._reserved, self._busy_waited)
+        return IPCEventDescriptor._init, (self._reserved, self._busy_waited)
 
 
 def _reduce_event(event):

cuda_core/cuda/core/_memory/_buffer.pyx

@@ -123,9 +123,13 @@ cdef class Buffer:
         self._mem_attrs_inited = False
         return self
 
+    @staticmethod
+    def _reduce_helper(mr, ipc_descriptor):
+        return Buffer.from_ipc_descriptor(mr, ipc_descriptor)
+
     def __reduce__(self):
         # Must not serialize the parent's stream!
-        return Buffer.from_ipc_descriptor, (self.memory_resource, self.get_ipc_descriptor())
+        return Buffer._reduce_helper, (self.memory_resource, self.get_ipc_descriptor())
 
     @staticmethod
     def from_handle(

cuda_core/cuda/core/_memory/_ipc.pyx

@@ -81,15 +81,15 @@ cdef class IPCBufferDescriptor:
     def __init__(self, *arg, **kwargs):
         raise RuntimeError("IPCBufferDescriptor objects cannot be instantiated directly. Please use MemoryResource APIs.")
 
-    @classmethod
-    def _init(cls, reserved: bytes, size: int):
-        cdef IPCBufferDescriptor self = IPCBufferDescriptor.__new__(cls)
+    @staticmethod
+    def _init(reserved: bytes, size: int):
+        cdef IPCBufferDescriptor self = IPCBufferDescriptor.__new__(IPCBufferDescriptor)
         self._payload = reserved
         self._size = size
         return self
 
     def __reduce__(self):
-        return self._init, (self._payload, self._size)
+        return IPCBufferDescriptor._init, (self._payload, self._size)
 
     @property
     def size(self):

cuda_core/cuda/core/_module.pyx

@@ -561,6 +561,10 @@ cdef class Kernel:
         """
         return as_py(self._h_kernel)
 
+    @property
+    def _handle(self):
+        return self.handle
+
     @staticmethod
     def from_handle(handle, mod: ObjectCode = None) -> Kernel:
         """Creates a new :obj:`Kernel` object from a foreign kernel handle.
@@ -657,10 +661,9 @@ cdef class ObjectCode:
 
         return self
 
-    @classmethod
-    def _reduce_helper(cls, module, code_type, name, symbol_mapping):
-        # just for forwarding kwargs
-        return cls._init(module, code_type, name=name if name else "", symbol_mapping=symbol_mapping)
+    @staticmethod
+    def _reduce_helper(module, code_type, name, symbol_mapping):
+        return ObjectCode._init(module, code_type, name=name if name else "", symbol_mapping=symbol_mapping)
 
     def __reduce__(self):
         return ObjectCode._reduce_helper, (self._module, self._code_type, self._name, self._sym_map)

cuda_core/docs/nv-versions.json

@@ -3,6 +3,10 @@
         "version": "latest",
         "url": "https://nvidia.github.io/cuda-python/cuda-core/latest/"
     },
+    {
+        "version": "0.6.0",
+        "url": "https://nvidia.github.io/cuda-python/cuda-core/0.6.0/"
+    },
     {
         "version": "0.5.1",
         "url": "https://nvidia.github.io/cuda-python/cuda-core/0.5.1/"

cuda_core/tests/test_utils.py

@@ -18,7 +18,11 @@
 except ImportError:
     torch = None
 import cuda.core
-import ml_dtypes
+
+try:
+    import ml_dtypes
+except ImportError:
+    ml_dtypes = None
 import numpy as np
 import pytest
 from cuda.core import Device
@@ -545,6 +549,7 @@ def test_from_array_interface_unsupported_strides(init_cuda):
         param((slice(None, None, 2), slice(1, None, 2)), id="strided"),
     ],
 )
+@pytest.mark.skipif(ml_dtypes is None, reason="ml_dtypes is not installed")
 @pytest.mark.skipif(cp is None, reason="CuPy is not installed")
 @pytest.mark.skipif(cp is not None and _get_cupy_version_major() < 14, reason="CuPy version is less than 14.0.0")
 def test_ml_dtypes_bfloat16_dlpack(init_cuda, slices):
@@ -575,6 +580,7 @@ def test_ml_dtypes_bfloat16_dlpack(init_cuda, slices):
         param((slice(None, None, 2), slice(1, None, 2)), id="strided"),
     ],
 )
+@pytest.mark.skipif(ml_dtypes is None, reason="ml_dtypes is not installed")
 @pytest.mark.skipif(torch is None, reason="PyTorch is not installed")
 def test_ml_dtypes_bfloat16_torch_dlpack(init_cuda, slices):
     a = torch.tensor([1, 2, 3, 4, 5, 6], dtype=torch.bfloat16, device="cuda").reshape(2, 3)[slices]