Re: cc optimisation levels ?

I did some benchmarks a good while back with a few different
optimization levels and types (loop-unrolling, i586 and i686 specific
optimizations, O, 02, 03, etc.) and you're quite right. The difference
was marginal...Not unmeasurable, but not enough to be willing to trade
stability for the added performance. Although I didn't find any
stability problems with the GCC of the time (2.93, maybe?) at any
optimization level. That was Squid 2.2STABLE5+hno0202, maybe things
have changed in Squid and GCC by now.

Anyway...the difference was along the lines of 2-5%. Unfortunately, I
wasn't keeping terribly good benchmark data at the time (no graphs, no
configuration details, nothing but the polygraph benchmark logs
remain...and even some of those have disappeared with system
reinstalls). I'll probably get back to doing lots of benchmarking
before we begin shipping with 2.4, so I'll give these kinds of tweaks
another run, and see what happens.

Alex Rousskov wrote:

> Hi,
>
> Is there any Squid-specific data that shows the speed
> advantage of, say, -O2 over -O or over -O0? I would not be surprised
> if the advantage is marginal, at best.
>
> Thanks,
>
> Alex.
Received on Sat Jan 20 2001 - 14:09:00 MST