[Pixman] Pixman microbenchmarks (was: [ssse3]Optimization for fetch_scanline_x8r8g8b8)

[Siarhei Siamashka]

On Friday 03 September 2010 01:39:54 Soeren Sandmann wrote:
> Siarhei Siamashka <siarhei.siamashka at gmail.com> writes:
> > Looks like this data has been posted already:
> > http://lists.freedesktop.org/archives/pixman/2010-June/000231.html
> > 
> > Checking a few more things with microbenchmarks shows that the prefetch
> > distance of just 64 bytes ahead is way too small.
> 
> We should probably get these microbenchmarks into the tree.
>
> Microbenchmarks can be a little dangerous since they sometimes amplify
> irrelevant details and make them seem important, but they are clearly
> useful in many cases.

Microbenchmarks are especially useful when checking whether some operation is
memory performance limited or not. If it is, then no further optimizations are 
going to provide any measurable result. If not, then some potential for more
optimizations may still exist.

Additionally, when finetuning optimizations, it may be so that the effect of
each small change can be only measured by microbenchmarks. And the overall
cumulative effect of multiple small optimizations can be later confirmed by
running real applications or replays of their traces.

Regarding adding microbenchmarks into pixman tree, they still need a bit of
cleanup:
- Make sure that they compile and work fine on all the supported platforms
- Be able to measure time with sufficiently high precision on all platforms
- Ensure statistically relevant results (standard deviation calculation, etc.)
- Maybe add some way to automatically detect the sizes of caches because this 
is important if using memory throughput as a reference
- Present benchmark results in such a way that they are reasonably compact and
easy to interpret
- Maybe something else that I forgot to mention

This all may take some time and is a little bit boring. I guess I can do it 
eventually, but I surely would not mind if somebody else could step in
and take care of all the pixman microbenchmarks stuff :)

-- 
Best regards,
Siarhei Siamashka
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.freedesktop.org/archives/pixman/attachments/20100908/4151eed0/attachment.pgp>