Age | Commit message (Collapse) | Author | Files | Lines |
|
Signed-off-by: Matěj Cepl <mcepl@cepl.eu>
|
|
An example written to demonstrate an issue with scaling performance in
firefox, in particular how it was causing stuttering through an
animation.
Reported-by: Maksim Lin <maks@manichord.com>
|
|
cairo-traces evidently used to be included in cairo proper, and the
documentation was referencing it as if it still was, which is a *bit*
confusing. Instead, extract the macro-benchmark specific directions
and move it here.
Signed-off-by: Bryce Harrington <b.harrington@samsung.com>
Signed-off-by: Uli Schlachter <psychon@znc.in>
|
|
Is this the first instance of a MULTIPLY operation in the wild?
|
|
|
|
|
|
This issue was highlighted in a bug report
https://bugzilla.mozilla.org/show_bug.cgi?id=424529 and demonstrated at
http://home.arcor.de/bazonbloch/scrolltest/whyisscrollingonlinuxslowerthanonwindows.htm
|
|
Presuming that the off-by-one bug in firefox will be fixed and so
prevent the allocation of huge intermediate surfaces and so reduce the
problem to that of lots and lots of curves
|
|
This highlights a deficiency in firefox's canvas implementation that
does not cull the drawn region to the visible area. Hence it generates
up to 20,000 x 20,000 pixmaps and completely renders them only to show a
1366x768 region. This represents quite a challenge!
http://ie.microsoft.com/testdrive/Performance/Chalkboard
|
|
Provided by Jindrich Makovicka
(https://bugs.freedesktop.org/show_bug.cgi?id=49534) to exercise a bug
he was encountering on his machine where the xlib backend was over 200x
slower than the image baseline.
|
|
|
|
From https://bugs.freedesktop.org/show_bug.cgi?id=38927 this poppler
replay is only mean to take a second!
|
|
This is a mixture of invisible gradients and a lot of unaligned alpha
composites
|
|
|
|
|
|
|
|
What novelty. Again multiple tests within one trace. Grr.
|
|
A pair of highly artifical benchmarks that exercise the basics. The
utility of having such synthetic benchmarks combined in a single trace
is a little awkward, so I may remove it later if I find little value in
the results.
|
|
Another HTML5 canvas demo,
http://ie.microsoft.com/testdrive/Performance/Paintball/Default.html
|
|
Another HTML5 canvas demo,
http://ie.microsoft.com/testdrive/performance/fishbowl/default.html
|
|
|
|
So that everybody is clear on what the licence actually is and so can
redistribute/contribute without undue concern or confusion.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
A typical run of GrADS, an open-source program for the analysis and
display of meteorological data provided by Jennifer Adams.
|
|
When reviewing the results from running cairo-perf-trace over the whole
set, similarly named full and shortened benchmark traces are coalesced
and so generate very misleading results.
|
|
And the README entry for full/firefox-woodtv
|
|
A trace that exercises a bug in the RENDER drivers triggering fallbacks
handling a 1299x15000 background image.
|
|
Siarhei Siamashka provided a couple of traces that are more typical from
the world of mobile devices: using bitmap [a1] fonts and zooming in on
web pages. As these profiles are strikingly different from our current
set of benchmarks, these make a worthwhile addition.
|
|
Traces were created with some host endian values.
Tools now expect all values to be big endian, thus old
traces must be updated to work.
|
|
I uploaded a bound trace which is not sufficiently future proof - i.e.
it breaks as soon as we try to improve the format and bytecodes.
|
|
This trace record evolution/gtk trying to create an illegal Window (one
with a larger size than 32k) and so we cannot replay it.
|
|
|
|
|
|
|
|
Based on following the INSTALL instructions from
http://git.savannah.gnu.org/cgit/maposmatic/ocitysmap.git/tree/INSTALL
and tracing the simple example "ocitysmap-render -f png -c Sanguinet".
|
|
|
|
|
|
$(pkg-config) is not a GNU Make extension.
|
|
|
|
|
|
This is to cut down the tarball to a manageable size as the intent is to
use the snapshot as means of distributing benchmark traces (for use
with the Phoronix Test Suite).
|
|
|
|
|
|
A simple target based on cairo/build/Makefile.am.releasing, to
autogenerate a tarball of the traces and upload to
cairographics.org/snapshots.
|
|
Add the first 10 seconds from the full versions of evolution,
gnome-system-monitor, poppler, swfdec-giant-steps and swfdec-youtube.
(My main fear is the initial sequence of operations may be similar and so
make the benchmark overly responsive to a single factor. Time will tell.)
|
|
Some applications, like poppler-blitz, use a single surface for the
entire run and so never trigger the timeout in surface_create(). So move
the check to context_create(), as contexts are meant to be transient!
|
|
Scrolling within a UTF-8 document, we draw lots of text, from a variety of
fonts, at fairly scattered positions.
|
|
Scroll the full height of http://planet.gnome.org/, chosen as being my
read visited site, so its performance is crucial to my sanity!
|
|
These are the untrimmed traces from the synthetic talos firefox tests, as
these provide wide coverage of firefox.
|
|
More the currents trace to a full/ directory and introduce an empty
benchmark/ directory. I plan to add selected short traces to the
benchmark/ directory that will be useful for the casual benchmarker (or
for quick runs over large number of revisions). In addition to the
reorganisation, I've added a csi-trace utility that has the ability to
trim a trace after a certain period of time has elapsed (since it has to
draw the trace to an image surface in order to satisfy some internal
requirements, this is a reasonable method of trimming replay times).
Note csi-trace probably requires some patches currently in wip queues,
but that should not affect normal operations.
|
|
Watching the graphing widgets whilst tracing itself (obviously!) and using
epiphany.
|