[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Default ISA/tuning flags for GCC, --enable-kernel= level for glibc



On Mon, 2009-01-26 at 12:30 +0100, Jakub Jelinek wrote:
> On Mon, Jan 26, 2009 at 12:06:49PM +0100, Dominik 'Rathann' Mierzejewski wrote:
> > I don't know about non-x86 arches (it's probably less of a problem), but
> > for x86 such change can (as you probably know) affect the overall speed
> > of the generated code and must be benchmarked. I would be in favour of
> > changing the default cflags to include -march=i586 if it gives measurable
> > benefits, i.e. faster binaries and/or smaller code size, but slower binaries
> > are not acceptable.
> 
> Performance on i?86 is mostly the matter of -mtune, which should stay the
> same (-mtune=generic, tuning for contemporary Intel and AMD CPUs).
> The difference between -march=i386 -mtune=generic and -march=i486 -mtune=generic
> is just that xadd[bwl], bswap and cmpxchg[bwl] insns can be generated for the
> latter, for -march=i586 -mtune=generic also cmpxchg8b insn.  Except for
> bswap, that's mainly for __sync_* builtins, OpenMP #pragma omp atomic, and
> other users of atomic insns.

AFAIK everyone's dumped i386 support already, for reasons I can't find a
good link for right now. (Something to do with the lack of atomic
instructions :) There's no reason to not move up to -march=i486.

Attachment: signature.asc
Description: This is a digitally signed message part


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]