[Bug 449356] Refactor gfxPangoFontGroup for user fonts

[bugzilla-daemon at mozilla.org]

Do not reply to this email.  You can add comments to this bug at
https://bugzilla.mozilla.org/show_bug.cgi?id=449356


Karl Tomlinson (:karlt) <mozbugz at karlt.net> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Attachment #341049|0                           |1
        is obsolete|                            |




--- Comment #6 from Karl Tomlinson (:karlt) <mozbugz at karlt.net>  2008-10-12 21:42:06 PDT ---
Created an attachment (id=342841)
 --> (https://bugzilla.mozilla.org/attachment.cgi?id=342841)
font selection in Mozilla's PangoFcFontMap 0.4

This is a proof of concept patch.  It still needs a bit of tidying up.
Results from Talos on the try server:

            ref w/patch change
Ts (ms)    1435  1463     +2%
Tp (ms)     446   441     -1%
Tp_RSS (MB) 110    80    -27%

Reference log:
http://tinderbox.mozilla.org/showlog.cgi?log=MozillaTry/1223866852.1223870369.23733.gz
w/patch:
http://tinderbox.mozilla.org/showlog.cgi?log=MozillaTry/1223860252.1223863695.10237.gz

The Ts change may be due to gfxFontconfigUtils being initialized earlier,
reverting some of the savings from the patch in bug 404857 comment 13.  If so,
doing some of the initialization lazily should recover that.  (Some of it may
never be used.)

There is more happening in the Tp test than is indicated in the overall result
here.  The Tp test runs through 400 pages then re-runs through the same set of
400 pages another 9 times.  The reported result is an average of some sort
over all 10 runs of 400 pages (possibly with some values discarded or other
magic thrown in).

Without the patch, the mean page load time for the first run is 584 ms, the
mean for the second run is 444 ms, and the mean over the remaining runs
is 460 ms.  The increase after the second run is a bit concerning but it is
what would be expected from bug 453200.

With this patch, the mean for the first run is 446 ms (-24%), the mean
for the second run is 442 ms (~0%), and the mean over the remaining runs
is 459 ms (~0%).

The time taken for zooming is most related (but not equivalent) to the time
for the first run, so IMO having a fast first run is more important than the
weight attributed to it in Talos Tp.

-- 
Configure bugmail: https://bugzilla.mozilla.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.