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
Copy file name to clipboardExpand all lines: docs/docfiles/dpctl.memory_api.rst
-19Lines changed: 0 additions & 19 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -28,22 +28,3 @@ Functions
28
28
---------
29
29
30
30
.. autofunction:: dpctl.memory.as_usm_memory
31
-
32
-
Comparison with Rapids Memory Manager (RMM)
33
-
-------------------------------------------
34
-
35
-
RMM implements DeviceBuffer which is Cython native class wrapping around something similar to ``std::vector<unsigned char, custom_cuda_allocator (calls resource manager)>`` which is called device_buffer.
36
-
37
-
DeviceBuffer stores a unique pointer to an instance of this class. DeviceBuffer implements ``__cuda_array_interface__``. Direct constructors always allocate
38
-
new memory and copy provided inputs into the newly allocated array.
39
-
40
-
Zero-copy construction is possible from a ``unique_ptr<device_ buffer>``, with
41
-
the ownership being moved to the Cython extension instance.
42
-
43
-
DeviceBuffer provides ``__reduce__`` method to support pickling (which works by copying content of the device buffer to host) and provides the following set of routines, among others:
44
-
45
-
- ``copy_to_host(host_buf_obj)`` to copy content of the underlying device_buffer to a host buffer
46
-
- ``copy_from_host(host_buf_obf)`` to copy content of the host buffer into memory of underlying device_buffer
47
-
- ``copy_from_device(cuda_ary_obj)`` to copy device memory underlying cuda_ary_obj Python object implementing ``__cuda_array_interface__`` to the memory underlying DeviceBuffer instance.
0 commit comments