[cairo] Performance work update

[Carl Worth]

On Thu, 18 Aug 2005 13:00:29 -0500, Billy Biggs wrote:
>   However, this doesn't preclude them going into CVS, I just think I'd
> feel really bad if people used their results to claim "my system is
> faster than your system", "cairo is slower than X", or even just "cairo
> is too slow", which sometimes happens when benchmark apps are posted.

Yeah, I wasn't looking for a system-characterization suite. Just that
it would be nice to easily pick out cases like "image-backend is
actually faster than xlib-backend" thing. And being able to generate a
table like that on one system with various versions of the X server
would be very useful for helping us plan some fallback strategies I
think.

I definitely think any combined-test run should be something layered
above. You certainly don't to lose the ability to run individual
micro-benchmarks like you have now.

>   Not yet, I've been trying to ensure these sort of tweaks have the OK
> from various people to ensure I'm not doing something which would hurt
> xserver performance, and that I'm not breaking anything, although maybe
> I'm beeing too cautious.

I can't comment on xserver. But for cairo, "make check" is the
authority when it comes to rendering results, and if it misses
something then that's a bug in the test suite.

I think most of the performance work you're doing falls into the
localized get-the-same-answer-a-different-way category which is easier
for the test suite to cover. Things like my hash rewrite that change a
bunch of logic and introduce memory leaks and crashes are harder for
make check to cover and definitely need more careful review.

-Carl
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.freedesktop.org/archives/cairo/attachments/20050818/406eef53/attachment-0001.pgp