You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
cuda.core: VirtualMemoryResourceOptions accepts values the driver rejects, and modify_allocation mishandles edge inputs #2910
Validation gaps found while analyzing #2882, each a small change: (1) addr_align larger than the granularity makes the adjacent-reservation probe fail with a misaligned hint, which is treated as fatal instead of falling back to the slow path; (2) new_size < buf.size is silently treated as an access-only update; (3) allocation_type=MANAGED is exposed but current drivers reject it in cuMemCreate; (4) modify_allocation guards a user-reachable condition with assert; (5) allocate(0) reaches the driver and fails while sibling resources return an empty Buffer; (6) the default handle_type=POSIX_FD fails on native Windows WDDM devices and on location_type="host"; (7) for host locations device_id is validated and ignored, HOST_NUMA_CURRENT can never allocate, and a host resource with peers reports is_device_accessible=False. Related: #2694.
I'd like to pick up items (4) and (5) from this list: raise a proper exception instead of the assert in modify_allocation, and have allocate(0) return an empty Buffer like the sibling resources. Happy to keep those to one small PR, if that split works for you. Could I be assigned?
Thanks for the offer, @bishal0922. #2917 has since merged and took care of items 1, 2, 4 and 5 and the host half of item 6, so the assert and allocate(0) are already handled.
Activity
bishal0922 commentedon Sep 30, 2026
I'd like to pick up items (4) and (5) from this list: raise a proper exception instead of the
assertinmodify_allocation, and haveallocate(0)return an emptyBufferlike the sibling resources. Happy to keep those to one small PR, if that split works for you. Could I be assigned?fix(cuda.core): move VirtualMemoryResource onto the _rt handle layer (#…
Andy-Jost commentedon Oct 2, 2026
Thanks for the offer, @bishal0922. #2917 has since merged and took care of items 1, 2, 4 and 5 and the host half of item 6, so the
assertandallocate(0)are already handled.