On 2013-09-28 09:03:53 +0200, Marc Glisse wrote: > On Fri, 27 Sep 2013, Steve M. Robbins wrote: > > >For readers of gmp-discuss and Daniel Schepler: this concerns a bug > >reported to Debian regarding GMP on the x32 architecture. In a > >nutshell: x32 selects a limb size of 8, which means it is larger > >than a pointer and some software (gap, pari) fail to build. The > >question is whether it makes sense to change the limb size to 4 on > >x32. I'd appreciate your thoughts. > > > >See full thread at http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=724320
There is/was a similar problem on PowerPC with the mode32 ABI (except that this was worse as we got wrong results): http://trac.macports.org/ticket/18780 but AFAIK, this wasn't the default. > the whole point of creating x32 was to benefit from the speed advantages of > amd64 without having a pointer size of 64 bits. Those speed advantages > include using more registers, but also using hardware 64 bit long long. The > speed difference between 32 and 64 bits is very large (a factor >2 for > multiplication IIRC, not even counting the fact that we have super-optimized > asm for amd64 and may not be spending so much time tweaking the x86 asm for > new platforms these days) so it doesn't make sense to me to penalize x32 I agree. PARI is not the only software that uses GMP. > that way. The assumption that mp_limb_t and void* have the same size is the > same kind of problem that we had when people assumed things about long and > void*, and most of those got fixed with time (though I know it was very > painful to do so in some cases). Including PARI for long and void*? If no, I doubt that changing GMP would solve the problem. If yes, can't a similar fix be done in PARI? Otherwise couldn't PARI use its own version of GMP with a limb size of 4 bytes? -- Vincent Lefèvre <vinc...@vinc17.net> - Web: <http://www.vinc17.net/> 100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/> Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon) -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
