Release v13.3 of the CUDA samples with CUDA 13.3 Toolkit (#435)

This is the release of the CUDA 13.3 samples, which include additions for CUDA Tile C++, and updated CCCL and Python samples.

python/2_CoreConcepts/memoryResources/README.md

@@ -60,15 +60,27 @@ verified on the host.
 ### Hardware
 
 - NVIDIA GPU with Compute Capability 7.0 or higher
-- Managed memory support (most discrete GPUs on Linux and Windows)
+- Managed memory support (most discrete GPUs)
 
 ### Software
 
 - CUDA Toolkit 13.0 or newer (matches `cuda-python` 13.x)
 - Python 3.10 or newer
 - `cuda-python` (>=13.0.0)
-- `cuda-core` (>=0.6.0)
-- `cupy-cuda13x` (>=13.0.0)
+- `cuda-core` (>=1.0.0)
+- `cupy-cuda13x` (>=14.0.0)
+
+### Platform Support
+
+The `ManagedMemoryResource` demo in this sample exercises **concurrent host
+access** to managed allocations while the GPU is active, which requires the
+device property `concurrent_managed_access=True`. This is only supported on
+Linux with HMM (Pascal and newer). On Windows (WDDM/MCDM/TCC) the property
+is `False`, so the sample exits early with a waive message and exit code
+`2`. The `DeviceMemoryResource` + `PinnedMemoryResource` demos in this
+sample would still work on Windows on their own, but to keep the sample
+self-contained the entire script waives when concurrent managed access is
+unavailable.
 
 ## Installation
 
@@ -82,8 +94,8 @@ pip install -r requirements.txt
 The `requirements.txt` installs:
 
 - `cuda-python` (>=13.0.0)
-- `cuda-core` (>=0.6.0)
-- `cupy-cuda13x` (>=13.0.0)
+- `cuda-core` (>=1.0.0)
+- `cupy-cuda13x` (>=14.0.0)
 
 ## How to Run
 

python/2_CoreConcepts/memoryResources/memoryResources.py

@@ -165,9 +165,9 @@ def demo_managed(device, stream, kernel, size):
         stream.sync()
 
         # No explicit copy: the same numpy view observes the GPU's writes.
-        assert np.allclose(managed_view, original * 0.5 + 10.0), (
-            "Managed memory result mismatch"
-        )
+        assert np.allclose(
+            managed_view, original * 0.5 + 10.0
+        ), "Managed memory result mismatch"
         print("  GPU writes observed directly through the host-visible mapping")
     finally:
         managed_buffer.close(stream)
@@ -222,6 +222,14 @@ def main():
     parser.add_argument("--device", type=int, default=0, help="CUDA device id")
     args = parser.parse_args()
 
+    if sys.platform == "win32":
+        print(
+            "This sample relies on ManagedMemoryResource with concurrent host "
+            "access, which is not supported on Windows "
+            "(concurrent_managed_access=False). Waiving this sample."
+        )
+        sys.exit(2)
+
     device = Device(args.device)
     device.set_current()
     print_gpu_info(device)

python/2_CoreConcepts/memoryResources/requirements.txt

@@ -1,5 +1,5 @@
 # Memory Resources Sample Requirements
 
 cuda-python>=13.0.0
-cuda-core>=0.6.0
-cupy-cuda13x>=13.0.0
+cuda-core>=1.0.0
+cupy-cuda13x>=14.0.0