Merge remote-tracking branch 'upstream/main' into fix/check-cython-abi-anon-struct-renames

Author: mdboom

.github/dependabot.yml

@@ -1,9 +1,51 @@
-# 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: Apache-2.0
 
 version: 2
 updates:
+  # Python test-dependency updates for cuda_bindings
+  - package-ecosystem: pip
+    directory: /cuda_bindings
+    schedule:
+      interval: "monthly"
+      time: "09:00"
+      timezone: "America/Los_Angeles"
+    open-pull-requests-limit: 5
+    groups:
+      test-deps:
+        applies-to: version-updates
+        patterns: ["*"]
+        update-types: ["major", "minor", "patch"]
+
+  # Python test-dependency updates for cuda_core
+  - package-ecosystem: pip
+    directory: /cuda_core
+    schedule:
+      interval: "monthly"
+      time: "09:00"
+      timezone: "America/Los_Angeles"
+    open-pull-requests-limit: 5
+    groups:
+      test-deps:
+        applies-to: version-updates
+        patterns: ["*"]
+        update-types: ["major", "minor", "patch"]
+
+  # Python test-dependency updates for cuda_pathfinder
+  - package-ecosystem: pip
+    directory: /cuda_pathfinder
+    schedule:
+      interval: "monthly"
+      time: "09:00"
+      timezone: "America/Los_Angeles"
+    open-pull-requests-limit: 5
+    groups:
+      test-deps:
+        applies-to: version-updates
+        patterns: ["*"]
+        update-types: ["major", "minor", "patch"]
+
   # GitHub Actions updates targeting the default branch (main)
   - package-ecosystem: github-actions
     directory: /

cuda_bindings/pyproject.toml

@@ -46,15 +46,15 @@ all = [
 [dependency-groups]
 test = [
     "cython>=3.2,<3.3",
-    "setuptools>=77.0.0",
+    "setuptools>=80.0.0",
     # TODO: remove the Python 3.15 guard once 3.15 is officially supported
-    "matplotlib>=3.5.0; python_version < '3.15'",
-    "numpy>=1.21.1",
-    "pytest>=6.2.4",
-    "pytest-benchmark>=3.4.1",
-    "pytest-repeat",
-    "pytest-randomly",
-    "pyglet>=2.1.9",
+    "matplotlib>=3.5.0,<=3.10.9; python_version < '3.15'",
+    "numpy>=1.21.1,<=2.5.0",
+    "pytest==9.1.0",
+    "pytest-benchmark==5.2.3",
+    "pytest-repeat==0.9.4",
+    "pytest-randomly==4.1.0",
+    "pyglet==2.1.14",
 ]
 
 [project.urls]

cuda_core/pyproject.toml

@@ -60,11 +60,18 @@ cu13 = ["cuda-bindings[all]==13.*", "cuda-toolkit==13.*"]
 
 [dependency-groups]
 test = [
-    "cython>=3.2,<3.3", "setuptools", "pytest>=6.2.4", "pytest-benchmark",
-    "pytest-randomly", "pytest-repeat", "pytest-rerunfailures", "pytest-timeout",
-    "cloudpickle", "psutil",
+    "cython>=3.2,<3.3",
+    "setuptools>=80",
+    "pytest==9.1.0",
+    "pytest-benchmark==5.2.3",
+    "pytest-randomly==4.1.0",
+    "pytest-repeat==0.9.4",
+    "pytest-rerunfailures==16.3",
+    "pytest-timeout==2.4.0",
+    "cloudpickle==3.1.2",
+    "psutil==7.2.2",
     # TODO: remove the Python 3.15 guard once 3.15 is officially supported
-    "cffi; python_version < '3.15'",
+    "cffi==2.0.0; python_version < '3.15'",
 ]
 ml-dtypes = ["ml-dtypes>=0.5.4,<0.6.0"]
 test-cu12 = [ {include-group = "ml-dtypes" }, {include-group = "test" }, "cupy-cuda12x; python_version < '3.14'", "cuda-toolkit[cudart]==12.*"]  # runtime headers needed by CuPy

cuda_core/tests/conftest.py

@@ -290,6 +290,9 @@ def ipc_memory_resource(request, ipc_device):
     assert mr.is_ipc_enabled
     yield mr
     mr.close()
+    # TODO(seberg): Make sure the `mr` and it's buffers are fully torn down.
+    # May be unnecessary as `mr.close()` is not parallel with other work.
+    ipc_device.sync()
 
 
 @pytest.fixture

cuda_core/tests/memory_ipc/test_peer_access.py

@@ -92,6 +92,8 @@ def test_main(self, ipc_mempool_device_x2, grant_access_in_parent):
         assert process.exitcode == 0
 
         buffer.close()
+        # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+        dev1.sync()
         mr.close()
 
     def child_main(self, mr, buffer):
@@ -129,4 +131,6 @@ def child_main(self, mr, buffer):
             PatternGen(dev0, NBYTES).verify_buffer(buffer, seed=False)
 
         buffer.close()
+        # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+        dev1.sync()
         mr.close()

cuda_core/tests/memory_ipc/test_send_buffers.py

@@ -29,6 +29,7 @@ def test_main(self, ipc_device, nmrs):
         device = ipc_device
         options = DeviceMemoryResourceOptions(max_size=POOL_SIZE, ipc_enabled=True)
         mrs = [DeviceMemoryResource(device, options=options) for _ in range(nmrs)]
+        buffers = []
 
         try:
             # Allocate and fill memory.
@@ -54,6 +55,10 @@ def test_main(self, ipc_device, nmrs):
                 pgen.verify_buffer(buffer, seed=True)
                 buffer.close()
         finally:
+            for buffer in buffers:
+                buffer.close()
+            # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+            device.sync()
             for mr in mrs:
                 mr.close()
 

cuda_core/tests/memory_ipc/test_workerpool.py

@@ -35,6 +35,7 @@ def test_main(self, ipc_device, nmrs):
         device = ipc_device
         options = DeviceMemoryResourceOptions(max_size=POOL_SIZE, ipc_enabled=True)
         mrs = [DeviceMemoryResource(device, options=options) for _ in range(nmrs)]
+        buffers = []
 
         try:
             buffers = [mr.allocate(NBYTES, stream=device.default_stream) for mr, _ in zip(cycle(mrs), range(NTASKS))]
@@ -45,8 +46,11 @@ def test_main(self, ipc_device, nmrs):
             pgen = PatternGen(device, NBYTES)
             for buffer in buffers:
                 pgen.verify_buffer(buffer, seed=True)
-                buffer.close()
         finally:
+            for buffer in buffers:
+                buffer.close()
+            # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+            device.sync()
             for mr in mrs:
                 mr.close()
 
@@ -77,6 +81,7 @@ def test_main(self, ipc_device, nmrs):
         device = ipc_device
         options = DeviceMemoryResourceOptions(max_size=POOL_SIZE, ipc_enabled=True)
         mrs = [DeviceMemoryResource(device, options=options) for _ in range(nmrs)]
+        buffers = []
 
         try:
             buffers = [mr.allocate(NBYTES, stream=device.default_stream) for mr, _ in zip(cycle(mrs), range(NTASKS))]
@@ -90,8 +95,11 @@ def test_main(self, ipc_device, nmrs):
             pgen = PatternGen(device, NBYTES)
             for buffer in buffers:
                 pgen.verify_buffer(buffer, seed=True)
-                buffer.close()
         finally:
+            for buffer in buffers:
+                buffer.close()
+            # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+            device.sync()
             for mr in mrs:
                 mr.close()
 
@@ -127,6 +135,7 @@ def test_main(self, ipc_device, nmrs):
         device = ipc_device
         options = DeviceMemoryResourceOptions(max_size=POOL_SIZE, ipc_enabled=True)
         mrs = [DeviceMemoryResource(device, options=options) for _ in range(nmrs)]
+        buffers = []
 
         try:
             buffers = [mr.allocate(NBYTES, stream=device.default_stream) for mr, _ in zip(cycle(mrs), range(NTASKS))]
@@ -137,8 +146,11 @@ def test_main(self, ipc_device, nmrs):
             pgen = PatternGen(device, NBYTES)
             for buffer in buffers:
                 pgen.verify_buffer(buffer, seed=True)
-                buffer.close()
         finally:
+            for buffer in buffers:
+                buffer.close()
+            # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+            device.sync()
             for mr in mrs:
                 mr.close()
 

cuda_core/tests/test_memory.py

@@ -1103,6 +1103,8 @@ def test_device_memory_resource_with_options(init_cuda):
     device.sync()
     dst_buffer.close()
     src_buffer.close()
+    # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+    device.sync()
 
 
 def test_pinned_memory_resource_with_options(init_cuda):
@@ -1149,6 +1151,8 @@ def test_pinned_memory_resource_with_options(init_cuda):
     device.sync()
     dst_buffer.close()
     src_buffer.close()
+    # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+    device.sync()
 
 
 def test_managed_memory_resource_with_options(init_cuda):
@@ -1365,6 +1369,8 @@ def test_mempool_ipc_errors(mempool_device):
         Buffer.from_ipc_descriptor(mr, handle, stream=device.default_stream)
 
     buffer.close()
+    # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+    device.sync()
 
 
 def test_pinned_mempool_ipc_basic():
@@ -1405,6 +1411,8 @@ def test_pinned_mempool_ipc_basic():
     assert ipc_desc.size == 1024
 
     buffer.close()
+    # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+    device.sync()
     mr.close()
 
 
@@ -1436,6 +1444,8 @@ def test_pinned_mempool_ipc_errors():
         Buffer.from_ipc_descriptor(mr, handle, stream=device.default_stream)
 
     buffer.close()
+    # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+    device.sync()
     mr.close()
 
 

cuda_core/tests/test_object_protocols.py

@@ -233,7 +233,11 @@ def sample_ipc_buffer_descriptor(ipc_device):
     options = DeviceMemoryResourceOptions(max_size=POOL_SIZE, ipc_enabled=True)
     mr = DeviceMemoryResource(ipc_device, options=options)
     buf = mr.allocate(64, stream=ipc_device.default_stream)
-    return buf.ipc_descriptor
+    descriptor = buf.ipc_descriptor
+    buf.close()
+    # TODO(seberg): 2026-06: mr close may be unsafe with incomplete `buf.close()`
+    ipc_device.sync()
+    return descriptor
 
 
 @pytest.fixture

cuda_core/tests/test_tensor_map.py

@@ -308,6 +308,22 @@ def test_invalid_data_type(self, dev, skip_if_no_tma):
                 data_type=42,
             )
 
+    def test_as_tensor_map_host_view_rejected_without_tma(self):
+        """``as_tensor_map`` rejects a non-device-accessible (host) view with a
+        clear error, exercising the ``as_tensor_map`` -> ``_from_tiled`` path
+        without needing TMA-capable hardware."""
+        host = np.zeros((64, 64), dtype=np.float32)
+        view = StridedMemoryView.from_any_interface(host, stream_ptr=-1)
+        with pytest.raises(ValueError, match="device-accessible"):
+            view.as_tensor_map(
+                box_dim=(32, 32),
+                data_type=TensorMapDataType.FLOAT32,
+                element_strides=(1, 1),
+                swizzle=TensorMapSwizzle.SWIZZLE_128B,
+                l2_promotion=TensorMapL2Promotion.L2_128B,
+                oob_fill=TensorMapOOBFill.NAN_REQUEST_ZERO_FMA,
+            )
+
 
 class TestTensorMapDtypeMapping:
     """Test automatic dtype inference from numpy dtypes."""