Skip to content

Commit

Permalink
Browse files Browse the repository at this point in the history
Doc tweaks for Stan's managed memory patch
  • Loading branch information
inducer committed Mar 9, 2014
1 parent 8690016 commit 22eca6d
Show file tree
Hide file tree
Showing 3 changed files with 13 additions and 2 deletions.
2 changes: 1 addition & 1 deletion doc/source/_static/akdoc.css
Expand Up @@ -11,7 +11,7 @@ body > div.container {
}

dd {
margin-left: 40px;
margin-left: 40px !important;
}

tt.descname {
Expand Down
12 changes: 11 additions & 1 deletion doc/source/driver.rst
Expand Up @@ -1112,6 +1112,12 @@ to and from the device.
operating system, and host compiler target architecture. Check the CUDA
C Programming Guide and CUDA release notes for details.

.. warning::

This interface to managed memory should be considered experimental. It is
provided as a preview, but for now the same interface stability guarantees
as for the rest of PyCUDA do not apply.

Managed Memory Allocation
~~~~~~~~~~~~~~~~~~~~~~~~~

Expand Down Expand Up @@ -1212,7 +1218,7 @@ an explicit copy::
median = np.median(a) # Computed on host!

.. warning::

The CUDA Unified Memory model has very specific rules regarding concurrent
access of managed memory allocations. Host access to any managed array
is not allowed while the GPU is executing a kernel, regardless of whether
Expand All @@ -1224,6 +1230,10 @@ an explicit copy::
appendix of the CUDA C Programming Guide for further details on the
concurrency restrictions.

If you are encountering interpreter terminations due to concurrency issues,
the `faulthandler <http://pypi.python.org/pypi/faulthandler>` module may be
helpful in locating the location in your Python program where the faulty
access is occurring.

Arrays and Textures
^^^^^^^^^^^^^^^^^^^
Expand Down
1 change: 1 addition & 0 deletions doc/source/misc.rst
Expand Up @@ -11,6 +11,7 @@ Version 2014.1

* Add :meth:`PointerHolderBase.as_buffer` and :meth:`DeviceAllocation.as_buffer`.
* Support for :class:`device_attribute` values added in CUDA 5.0, 5.5, and 6.0.
* Support for :ref:`managed_memory`. (contributed by Stan Seibert)

Version 2013.1.1
----------------
Expand Down

0 comments on commit 22eca6d

Please sign in to comment.