warp.CudaManagedAllocator

class warp.CudaManagedAllocator

Allocator that creates CUDA managed-memory arrays.

The allocator object is not bound to one CUDA device and can be reused across devices. Each allocation still happens under a specific CUDA context, and that context’s device must support CUDA managed memory. Warp pushes the target CUDA context before invoking allocate(). Direct calls to allocate() require an active CUDA context.

Managed allocation uses cudaMallocManaged(). Allocate managed arrays before CUDA graph capture begins and reuse the existing arrays inside captured work for maximum CUDA compatibility.

__init__()

Methods

__init__()
allocate(size_in_bytes)
deallocate(ptr, size_in_bytes)

Attributes

deallocate_requires_context_guard
memory_kind

deallocate_requires_context_guard = False

memory_kind = 5

allocate(size_in_bytes)

deallocate(ptr, size_in_bytes)