Fix device attribute handling for older drivers (NVIDIA#1437)

* Fix device attribute handling for older drivers

Add explicit default value handling for device attributes that may not
be supported by older CUDA drivers. When cuDeviceGetAttribute returns
CUDA_ERROR_INVALID_VALUE, return a sensible default instead of raising
an error.

- Add default parameter to _get_attribute() and _get_cached_attribute()
- Use default=0 for boolean/enablement attributes (returns False)
- Use default=1 for mem_sync_domain_count (single domain is traditional behavior)
- Use default=-1 for host_numa_id (indicates NUMA not supported)
- Document that gpu_pci_device_id/gpu_pci_subsystem_id return 0 if unsupported

Closes NVIDIA#1420

* Add exception specifiers and refine attribute defaults

- Add except? -2 to _get_attribute and _get_cached_attribute for proper
  exception propagation (-2 never clashes with valid return values)
- Keep default parameter untyped to allow None, cast to int when used
- Simplify gpu_pci_device_id/gpu_pci_subsystem_id to return 0 when
  unsupported (0 is never a valid PCI ID)

Author: Andy-Jost

cuda_core/cuda/core/_device.pyx

@@ -55,21 +55,21 @@ cdef class DeviceProperties:
         self._cache = {}
         return self
 
-    cdef inline _get_attribute(self, cydriver.CUdevice_attribute attr):
+    cdef inline int _get_attribute(self, cydriver.CUdevice_attribute attr, default=0) except? -2:
         """Retrieve the attribute value directly from the driver."""
         cdef int val
         cdef cydriver.CUresult err
         with nogil:
             err = cydriver.cuDeviceGetAttribute(&val, attr, self._handle)
-        if err == cydriver.CUresult.CUDA_ERROR_INVALID_VALUE:
-            return 0
+        if err == cydriver.CUresult.CUDA_ERROR_INVALID_VALUE and default is not None:
+            return <int>default
         HANDLE_RETURN(err)
         return val
 
-    cdef _get_cached_attribute(self, attr):
+    cdef inline int _get_cached_attribute(self, attr, default=0) except? -2:
         """Retrieve the attribute value, using cache if applicable."""
         if attr not in self._cache:
-            self._cache[attr] = self._get_attribute(attr)
+            self._cache[attr] = self._get_attribute(attr, default)
         return self._cache[attr]
 
     @property
@@ -787,6 +787,8 @@ cdef class DeviceProperties:
         """bool: Device supports buffer sharing with dma_buf mechanism."""
         return bool(self._get_cached_attribute(driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_DMA_BUF_SUPPORTED))
 
+    # Start of CUDA 12 device attributes
+
     @property
     def ipc_event_supported(self) -> bool:
         """bool: Device supports IPC Events."""
@@ -795,7 +797,7 @@ cdef class DeviceProperties:
     @property
     def mem_sync_domain_count(self) -> int:
         """int: Number of memory domains the device supports."""
-        return self._get_cached_attribute(driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_MEM_SYNC_DOMAIN_COUNT)
+        return self._get_cached_attribute(driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_MEM_SYNC_DOMAIN_COUNT, default=1)
 
     @property
     def tensor_map_access_supported(self) -> bool:
@@ -824,7 +826,7 @@ cdef class DeviceProperties:
     @property
     def host_numa_id(self) -> int:
         """int: NUMA ID of the host node closest to the device. Returns -1 when system does not support NUMA."""
-        return self._get_cached_attribute(driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_HOST_NUMA_ID)
+        return self._get_cached_attribute(driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_HOST_NUMA_ID, default=-1)
 
     @property
     def d3d12_cig_supported(self) -> bool:
@@ -848,12 +850,18 @@ cdef class DeviceProperties:
 
     @property
     def gpu_pci_device_id(self) -> int:
-        """int: The combined 16-bit PCI device ID and 16-bit PCI vendor ID."""
+        """int: The combined 16-bit PCI device ID and 16-bit PCI vendor ID.
+
+        Returns 0 if the driver does not support this query.
+        """
         return self._get_cached_attribute(driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_GPU_PCI_DEVICE_ID)
 
     @property
     def gpu_pci_subsystem_id(self) -> int:
-        """int: The combined 16-bit PCI subsystem ID and 16-bit PCI subsystem vendor ID."""
+        """int: The combined 16-bit PCI subsystem ID and 16-bit PCI subsystem vendor ID.
+
+        Returns 0 if the driver does not support this query.
+        """
         return self._get_cached_attribute(driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_GPU_PCI_SUBSYSTEM_ID)
 
     @property
@@ -872,6 +880,8 @@ cdef class DeviceProperties:
             self._get_cached_attribute(driver.CUdevice_attribute.CU_DEVICE_ATTRIBUTE_HOST_NUMA_MEMORY_POOLS_SUPPORTED)
         )
 
+    # Start of CUDA 13 device attributes
+
     @property
     def host_numa_multinode_ipc_supported(self) -> bool:
         """bool: Device supports HOST_NUMA location IPC between nodes in a multi-node system."""