Richard Freeman posted <[EMAIL PROTECTED]>, excerpted below, on Tue, 28 Jun 2005 11:11:35 -0400:
> On Tue, June 28, 2005 10:47 am, Mauro Venanzi said: >> why have you the "Os" flag? waht is it? >> the same dubt for mtune >> try to write -O2 not Os > > -Os optimizes for size and works just fine for me. In systems with > limited RAM it is likely to give better performance than -O2 since you are > less likely to swap. Most of the optimizations are the same, except that > you aren't going to unroll loops or anything like that. Many of the > size-sacrificing optimizations don't add much performance-wise in the > first place. > > I believe that somebody posted a lengthy discussion regarding compiler > optimizations in the forums complete with benchmarks. This largely > influenced my decision to switch to -Os since at the time I was running > with 512MB RAM. Now I'm at 1GB, but even so I wouldn't say have RAM to > burn, and disk cache is probably going to do more for performance than > having loops unrolled... Exactly, and I'm an -Os person myself. However, if something goes haywire, I do at times try recompiling with "-march=k8 -O2 -pipe" (or even -O1), just to be sure. The last time I had a similar problem, it was an issue with 32-bit only, and due to the glibc I was using (the masked version for use with an also masked gcc4). The 64-bit glibc was working fine, but further updates to gcc or sandbox were failing, when they tried to do the config for their 32-bit compile sections. The problem was solved by emerge --packageonly a previous (and unmasked) glibc version. Since I use FEATURES=buildpkg, that was easily done. Note that emerging a binpkg doesn't require a working gcc... -- Duncan - List replies preferred. No HTML msgs. "Every nonfree program has a lord, a master -- and if you use the program, he is your master." Richard Stallman in http://www.linuxdevcenter.com/pub/a/linux/2004/12/22/rms_interview.html -- [email protected] mailing list
