[cairo] Cairo 1.9.2 performance

[Zack Weinberg]

Clemens Eisserer <linuxhippy at gmail.com> wrote:

> > This is particularly true
> > for gcc which is not known to produce blazingly fast code anyway
> Any results to back that claim, otherwise please don't spread fud.

It is certainly true of the code generated at -O0, which doesn't even
try to keep things in registers.

-O2 and -Os have been competitive with proprietary compilers on a wide
range of benchmarks for years now.  It's worth trying both and seeing
which works better.

-O3 is meant to be used on specific files that are known to contain hot
innermost loops -- if applied to an entire program it is likely to bulk
up the code size to the point where important things start falling out
of I-cache.

(I used to be a gcc developer.)

zw