Cache

cvcuda.cache_size(scope: cvcuda.ThreadScope = <ThreadScope.GLOBAL: 0>) → int

Returns the quantity of items in the NVCV Python cache

Parameters:

scope (nvcv.ThreadScope) – Thread scope that must be either nvcv.ThreadScope.GLOBAL or nvcv.ThreadScope.LOCAL.

cvcuda.clear_cache(scope: cvcuda.ThreadScope = <ThreadScope.GLOBAL: 0>) → None

Clears the NVCV Python cache

Parameters:

scope (nvcv.ThreadScope) – Thread scope that must be either nvcv.ThreadScope.GLOBAL or nvcv.ThreadScope.LOCAL.

cvcuda.current_cache_size_inbytes() → int

Returns indexed cache storage [in bytes] for the current CUDA device, including pending entries. This is not total device memory usage; user-owned uncached and wrapped external storage is excluded.

cvcuda.get_cache_limit_inbytes() → int

Returns the cache limit [in bytes] for the current CUDA device.

cvcuda.set_cache_limit_inbytes(arg: SupportsInt | SupportsIndex, /) → None

Sets the retained-cache budget [in bytes] for the current CUDA device, shared across threads. Pending GPU resources remain charged until retirement and may temporarily exceed this budget. Zero disables retention of completed allocations, not safe reuse of pending tensors.