head: MIPS: A workaround for the R4000 divide/shift errata

Eric Christopher echristo@redhat.com
Sat Feb 21 13:45:00 GMT 2004


And since Chris has let me know that I'm not very clear:

i'd like the option so that people can either a) turn it off, or b) turn
it on, and c) yes, default for r4000

Chris's reasons are good. Performance reasons are good too if you don't
happen to have a processor with those bugs.

Basically everytime we add an option to work around a processor bug it
should be controlled by a -mfix-* option so that it can be
enabled/disabled at will.

-eric


On Thu, 2004-02-19 at 15:47, cgd@broadcom.com wrote:
> At 19 Feb 2004 15:44:38 -0800, Chris G. Demetriou wrote:
> > I.e., the binary may be a little bit slower than optimal on any
> > processor, but it should would run portably on all, without need for
> > special versions w/ errata fixes.
> 
> In case it's not clear, the theory here is that most users people
> don't *care* if 'bc' or 'top' or xterm are optimal, they just care
> that the binary that they have works.
> 
> And it should be possible for creators of binary distributions --
> Linux distros, NetBSD, whoever -- should be able to create 'generic'
> binaries that work around all bugs that GCC can work around.
> 
> It probably shouldn't be the default for most uses of those systems,
> but it should at least be an option.
> 
> 
> 
> cgd
-- 
Eric Christopher <echristo@redhat.com>



More information about the Gcc-patches mailing list