Merge branch 'main' into rparolin/greptile_config

Author: rparolin

cuda_bindings/docs/source/conf.py

@@ -9,13 +9,11 @@
 
 # -- Path setup --------------------------------------------------------------
 
-# If extensions (or modules to document with autodoc) are in another directory,
-# add these directories to sys.path here. If the directory is relative to the
-# documentation root, use os.path.abspath to make it absolute, like shown here.
 import os
+import sys
+from pathlib import Path
 
-# import sys
-# sys.path.insert(0, os.path.abspath('.'))
+sys.path.insert(0, str((Path(__file__).parents[3] / "cuda_python" / "docs" / "exts").absolute()))
 
 
 # -- Project information -----------------------------------------------------
@@ -41,6 +39,7 @@
     "myst_nb",
     "enum_tools.autoenum",
     "sphinx_copybutton",
+    "release_toc",
 ]
 
 nb_execution_mode = "off"

cuda_bindings/docs/source/release.rst

@@ -6,38 +6,6 @@ Release Notes
 
 .. toctree::
    :maxdepth: 3
+   :glob:
 
-   13.0.3 <release/13.0.3-notes.rst>
-   13.0.2 <release/13.0.2-notes.rst>
-   13.0.1 <release/13.0.1-notes.rst>
-   13.0.0 <release/13.0.0-notes.rst>
-   12.9.4 <release/12.9.4-notes.rst>
-   12.9.3 <release/12.9.3-notes.rst>
-   12.9.2 <release/12.9.2-notes.rst>
-   12.9.1 <release/12.9.1-notes.rst>
-   12.9.0 <release/12.9.0-notes.rst>
-   12.8.0 <release/12.8.0-notes.rst>
-   12.6.2 <release/12.6.2-notes.rst>
-   12.6.1 <release/12.6.1-notes.rst>
-   12.6.0 <release/12.6.0-notes.rst>
-   12.5.0 <release/12.5.0-notes.rst>
-   12.4.0 <release/12.4.0-notes.rst>
-   12.3.0 <release/12.3.0-notes.rst>
-   12.2.1 <release/12.2.1-notes.rst>
-   12.2.0 <release/12.2.0-notes.rst>
-   12.1.0 <release/12.1.0-notes.rst>
-   12.0.0 <release/12.0.0-notes.rst>
-   11.8.7 <release/11.8.7-notes.rst>
-   11.8.6 <release/11.8.6-notes.rst>
-   11.8.5 <release/11.8.5-notes.rst>
-   11.8.4 <release/11.8.4-notes.rst>
-   11.8.3 <release/11.8.3-notes.rst>
-   11.8.2 <release/11.8.2-notes.rst>
-   11.8.1 <release/11.8.1-notes.rst>
-   11.8.0 <release/11.8.0-notes.rst>
-   11.7.1 <release/11.7.1-notes.rst>
-   11.7.0 <release/11.7.0-notes.rst>
-   11.6.1 <release/11.6.1-notes.rst>
-   11.6.0 <release/11.6.0-notes.rst>
-   11.5.0 <release/11.5.0-notes.rst>
-   11.4.0 <release/11.4.0-notes.rst>
+   release/*[0-9]-notes

cuda_core/cuda/core/experimental/_memory.pyx

@@ -1132,7 +1132,7 @@ class VirtualMemoryResourceOptions:
     location_type: VirtualMemoryLocationTypeT = "device"
     handle_type: VirtualMemoryHandleTypeT = "posix_fd"
     granularity: VirtualMemoryGranularityT = "recommended"
-    gpu_direct_rdma: bool = True
+    gpu_direct_rdma: bool = False
     addr_hint: Optional[int] = 0
     addr_align: Optional[int] = None
     peers: Iterable[int] = field(default_factory=tuple)
@@ -1211,6 +1211,11 @@ class VirtualMemoryResource(MemoryResource):
         if platform.system() == "Windows":
             raise NotImplementedError("VirtualMemoryResource is not supported on Windows")
 
+        # Validate RDMA support if requested
+        if self.config.gpu_direct_rdma and self.device is not None:
+            if not self.device.properties.gpu_direct_rdma_supported:
+                raise RuntimeError("GPU Direct RDMA is not supported on this device")
+
     @staticmethod
     def _align_up(size: int, gran: int) -> int:
         """

cuda_core/docs/source/conf.py

@@ -9,12 +9,11 @@
 
 # -- Path setup --------------------------------------------------------------
 
-# If extensions (or modules to document with autodoc) are in another directory,
-# add these directories to sys.path here. If the directory is relative to the
-# documentation root, use os.path.abspath to make it absolute, like shown here.
 import os
+import sys
+from pathlib import Path
 
-# sys.path.insert(0, os.path.abspath('.'))
+sys.path.insert(0, str((Path(__file__).parents[3] / "cuda_python" / "docs" / "exts").absolute()))
 
 
 # -- Project information -----------------------------------------------------
@@ -41,6 +40,7 @@
     "enum_tools.autoenum",
     "sphinx_copybutton",
     "sphinx_toolbox.more_autodoc.autoprotocol",
+    "release_toc",
 ]
 
 # Add any paths that contain templates here, relative to this directory.

cuda_core/docs/source/release.rst

@@ -6,11 +6,6 @@ Release Notes
 
 .. toctree::
    :maxdepth: 3
+   :glob:
 
-   0.4.0 <release/0.4.0-notes>
-   0.3.2 <release/0.3.2-notes>
-   0.3.1 <release/0.3.1-notes>
-   0.3.0 <release/0.3.0-notes>
-   0.2.0 <release/0.2.0-notes>
-   0.1.1 <release/0.1.1-notes>
-   0.1.0 <release/0.1.0-notes>
+   release/*[0-9]-notes

cuda_core/examples/thread_block_cluster.py

@@ -22,6 +22,10 @@
     launch,
 )
 
+if np.lib.NumpyVersion(np.__version__) < "2.2.5":
+    print("This example requires NumPy 2.2.5 or later", file=sys.stderr)
+    sys.exit(0)
+
 # prepare include
 cuda_path = os.environ.get("CUDA_PATH", os.environ.get("CUDA_HOME"))
 if cuda_path is None:

cuda_core/tests/test_memory.py

@@ -28,7 +28,7 @@
 from cuda.core.experimental._utils.cuda_utils import handle_return
 from cuda.core.experimental.utils import StridedMemoryView
 
-from cuda_python_test_helpers import IS_WSL, supports_ipc_mempool
+from cuda_python_test_helpers import supports_ipc_mempool
 
 POOL_SIZE = 2097152  # 2MB size
 
@@ -322,13 +322,13 @@ def test_vmm_allocator_basic_allocation():
     This test verifies that VirtualMemoryResource can allocate memory
     using CUDA VMM APIs with default configuration.
     """
-    if platform.system() == "Windows":
-        pytest.skip("VirtualMemoryResource is not supported on Windows TCC")
-    if IS_WSL:
-        pytest.skip("VirtualMemoryResource is not supported on WSL")
-
     device = Device()
     device.set_current()
+
+    # Skip if virtual memory management is not supported
+    if not device.properties.virtual_memory_management_supported:
+        pytest.skip("Virtual memory management is not supported on this device")
+
     options = VirtualMemoryResourceOptions()
     # Create VMM allocator with default config
     vmm_mr = VirtualMemoryResource(device, config=options)
@@ -361,13 +361,17 @@ def test_vmm_allocator_policy_configuration():
     with different allocation policies and that the configuration affects
     the allocation behavior.
     """
-    if platform.system() == "Windows":
-        pytest.skip("VirtualMemoryResource is not supported on Windows TCC")
-    if IS_WSL:
-        pytest.skip("VirtualMemoryResource is not supported on WSL")
     device = Device()
     device.set_current()
 
+    # Skip if virtual memory management is not supported
+    if not device.properties.virtual_memory_management_supported:
+        pytest.skip("Virtual memory management is not supported on this device")
+
+    # Skip if GPU Direct RDMA is supported (we want to test the unsupported case)
+    if not device.properties.gpu_direct_rdma_supported:
+        pytest.skip("This test requires a device that doesn't support GPU Direct RDMA")
+
     # Test with custom VMM config
     custom_config = VirtualMemoryResourceOptions(
         allocation_type="pinned",
@@ -420,13 +424,13 @@ def test_vmm_allocator_grow_allocation():
     This test verifies that VirtualMemoryResource can grow existing
     allocations while preserving the base pointer when possible.
     """
-    if platform.system() == "Windows":
-        pytest.skip("VirtualMemoryResource is not supported on Windows TCC")
-    if IS_WSL:
-        pytest.skip("VirtualMemoryResource is not supported on WSL")
     device = Device()
     device.set_current()
 
+    # Skip if virtual memory management is not supported (we need it for VMM)
+    if not device.properties.virtual_memory_management_supported:
+        pytest.skip("Virtual memory management is not supported on this device")
+
     options = VirtualMemoryResourceOptions()
 
     vmm_mr = VirtualMemoryResource(device, config=options)
@@ -458,6 +462,29 @@ def test_vmm_allocator_grow_allocation():
     grown_buffer.close()
 
 
+def test_vmm_allocator_rdma_unsupported_exception():
+    """Test that VirtualMemoryResource throws an exception when RDMA is requested but device doesn't support it.
+
+    This test verifies that the VirtualMemoryResource constructor throws a RuntimeError
+    when gpu_direct_rdma=True is requested but the device doesn't support virtual memory management.
+    """
+    device = Device()
+    device.set_current()
+
+    # Skip if virtual memory management is not supported (we need it for VMM)
+    if not device.properties.virtual_memory_management_supported:
+        pytest.skip("Virtual memory management is not supported on this device")
+
+    # Skip if GPU Direct RDMA is supported (we want to test the unsupported case)
+    if device.properties.gpu_direct_rdma_supported:
+        pytest.skip("This test requires a device that doesn't support GPU Direct RDMA")
+
+    # Test that requesting RDMA on an unsupported device throws an exception
+    options = VirtualMemoryResourceOptions(gpu_direct_rdma=True)
+    with pytest.raises(RuntimeError, match="GPU Direct RDMA is not supported on this device"):
+        VirtualMemoryResource(device, config=options)
+
+
 def test_mempool(mempool_device):
     device = mempool_device
 

cuda_pathfinder/docs/source/conf.py

@@ -9,13 +9,11 @@
 
 # -- Path setup --------------------------------------------------------------
 
-# If extensions (or modules to document with autodoc) are in another directory,
-# add these directories to sys.path here. If the directory is relative to the
-# documentation root, use os.path.abspath to make it absolute, like shown here.
 import os
+import sys
+from pathlib import Path
 
-# import sys
-# sys.path.insert(0, os.path.abspath('.'))
+sys.path.insert(0, str((Path(__file__).parents[3] / "cuda_python" / "docs" / "exts").absolute()))
 
 
 # -- Project information -----------------------------------------------------
@@ -41,6 +39,7 @@
     "myst_nb",
     "enum_tools.autoenum",
     "sphinx_copybutton",
+    "release_toc",
 ]
 
 nb_execution_mode = "off"

cuda_pathfinder/docs/source/release.rst

@@ -6,12 +6,6 @@ Release Notes
 
 .. toctree::
    :maxdepth: 3
+   :glob:
 
-   1.3.1 <release/1.3.1-notes>
-   1.3.0 <release/1.3.0-notes>
-   1.2.3 <release/1.2.3-notes>
-   1.2.2 <release/1.2.2-notes>
-   1.2.1 <release/1.2.1-notes>
-   1.2.0 <release/1.2.0-notes>
-   1.1.0 <release/1.1.0-notes>
-   1.0.0 <release/1.0.0-notes>
+   release/*[0-9]-notes

cuda_python/docs/exts/release_toc.py

@@ -0,0 +1,30 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
+
+from pathlib import Path
+
+from packaging.version import Version
+from sphinx.directives.other import TocTree
+
+
+class TocTreeSorted(TocTree):
+    """A toctree directive that sorts entries by version."""
+
+    def parse_content(self, toctree):
+        super().parse_content(toctree)
+
+        if not toctree["glob"]:
+            return
+
+        entries = toctree.get("entries", [])
+        if not entries:
+            return
+
+        entries = [(Version(Path(x[1]).name.removesuffix("-notes")), x[1]) for x in entries]
+        entries.sort(key=lambda x: x[0], reverse=True)
+        entries = [(str(x[0]), x[1]) for x in entries]
+        toctree["entries"] = entries
+
+
+def setup(app):
+    app.add_directive("toctree", TocTreeSorted, override=True)