warp.set_cuda_allocator#

warp.set_cuda_allocator(allocator)[source]#

Set the memory allocator for all CUDA devices.

Any object with allocate(size_in_bytes) -> int and deallocate(ptr, size_in_bytes) methods can be used.

Pass None to restore the built-in allocator.

Parameters:

allocator (Allocator | None) – An Allocator-compatible object, or None.

Return type:

None