[cairo] Cairo performance testing with X11 forwarding
chris at chris-wilson.co.uk
Sun Jun 17 05:02:30 PDT 2012
On Sat, 16 Jun 2012 17:36:19 -0400, Patrick Wilbur <patrick.wilbur at gmail.com> wrote:
> I have been manually (as in, not using * on the command-line) testing
> traces in the benchmark/ directory, one-by-one, to see what I can observe.
> For some of the tests (including swfdec-youtube.trace,
> firefox-scrolling.trace, gnome-system-monitor.trace, and
> chromium-tabs.trace) everything works like a charm when run over ssh -X.
> These I can use as several benchmarks, which is great.
> For others, I see a spike in CPU usage followed by CPU usage dropping out
> (back to my normal idle usage). At this point, the test hangs and
> cairo-perf-trace displays no further output (than up to the test name just
> before printing values).
> Seems quite strange, no? While I feel the traces that are working make
> good benchmarks, it would be nice to be able to get all traces working.
Just sounds like you are hitting a bug where communication stalls. If
you use something like xtrace (http://xtrace.alioth.debian.org/), or
xscope, hopefully they will give a clue as to what either end is waiting
upon. One of the tricky parts of running cairo-perf-trace is that it
emits a lot of requests and expects no replies (at least outside of
startup and synchronisation) and it will quite easily overflow the
16-bit counter in the request protocol which needs careful handling
inside your libX11/libxcb. I mention that as I have seen that bug in the
dim and distant past and it has similar symptoms.
Chris Wilson, Intel Open Source Technology Centre
More information about the cairo