[CUDA graphs] moves memory sharing intro paragraph (#64996)

Michael Carilli · facebook-github-bot · commit 36cac2be4db9 · 2021-09-14T17:53:43.000-07:00
Summary: Puts memory sharing intro under Sharing memory... header, where it should have been all along. Pull Request resolved: #64996 Reviewed By: mruberry Differential Revision: D30948619 Pulled By: ngimel fbshipit-source-id: 5d9dd267b34e9d3fc499d4738377b58a22da1dc2
diff --git a/docs/source/notes/cuda.rst b/docs/source/notes/cuda.rst
@@ -918,16 +918,16 @@ separate private pool for each capture. If you capture multiple graphs,
 this conservative approach ensures graph replays never corrupt each other's values,
 but sometimes needlessly wastes memory.
 
+Sharing memory across captures
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
 To economize the memory stashed in private pools, :class:`torch.cuda.graph`
 and :func:`torch.cuda.make_graphed_callables` optionally allow different
 captures to share the same private pool.
 It's safe for a set of graphs to share a private pool if you know they'll always
 be replayed in the same order they were captured,
 and never be replayed concurrently.
 
-Sharing memory across captures with torch.cuda.graph
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
 :class:`torch.cuda.graph`'s ``pool`` argument is a hint to use a particular private pool,
 and can be used to share memory across graphs as shown::
 
@@ -949,9 +949,6 @@ and can be used to share memory across graphs as shown::
     g1.replay()
     g2.replay()
 
-Sharing memory across captures with torch.cuda.make_graphed_callables
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
 With :func:`torch.cuda.make_graphed_callables`, if you want to graph several
 callables and you know they'll always run in the same order (and never concurrently)
 pass them as a tuple in the same order they'll run in the live workload, and
