[PATCH 0/2] Composite window resize performance tweaks

[Adam Jackson]

We're doing the backfill for initial window contents more than we need to.
Resize of bg=None windows will just leave trails of the old window
decorations, which can't be what any app wants.  Resize and map of
backgrounded windows will paint the background in the exposed areas.

It's a little awkward to test performance on this, since you end up
measuring the performance of the compositor, the drm memory manager, and
however tightly the CM and X server race in processing intermediate
states.  But, with something like:

    http://people.freedesktop.org/~ajax/resize-bench.c

and:

    Xvfb -ac -terminate -screen 0 1366x768x24+32 :1

on a 2.6GHz Core 2 Duo, I get:

Uncomposited:		1.8 sec
Composited, unpatched:	12.4 sec
Composited, patched:	9.3 sec

Which is decent.  Note that the uncomposited number is actually kind of
depressing, 9ms for a synchronous resize is rather a lot of a frame time.
Someone should work on that.

- ajax