On Mon, Aug 29, 2005 at 04:13:57PM -0500, Billy Biggs wrote: > Steve Langasek ([EMAIL PROTECTED]):
> > > The fix is to delete the MMX registers from the clobbers list, or > > > compile with -mmmx/-msse. I prefer removing them, as using > > > -mmmx/-msse is scary and opens you up to more gcc bugs. > > While it's true that using more registers introduces more > > possibilities of running into gcc bugs, I think mmx is used often > > enough that the risk is minimal (excepting deliberate behavior > > changes, that is). However, your packages should also not fail to run > > on MMX-less processors, which from what I can tell is the case here -- > > so dropping this does indeed seem the best option. > The -msse and -mmmx options are supposed to only allow using the mmx > registers in inline assembly, or using the intrinsics, and not to > authorize gcc to use these instructions. (Note that this is > inconsistent with -msse2 which apparently does cause gcc to emit > SSE2 instructions) So, I think in the best of all possible worlds, we > would use -mmmx and -msse to announce that we have code which uses them, > but still expect gcc to otherwise compile pre-P3 usable code. The question is, what prevents the inline assembly in question from being executed on 486 or non-MMX 586 systems? I didn't see any arch checks in the code that would prevent this, but maybe I missed something. -- Steve Langasek Give me a lever long enough and a Free OS Debian Developer to set it on, and I can move the world. [EMAIL PROTECTED] http://www.debian.org/
signature.asc
Description: Digital signature
