Re: bzip2 unaligned trap
On Fri, 26 Aug 2005, Steve Langasek wrote:
Unless they're running a fully ppc64 system (which is
contraindicated by kernel and glibc upstream), they're not going to
get the same unaligned traps on ppc as on alpha.
See:
arch/ppc/traps.c AlignmentException -> fix_alignment()
arch/ppc/align.c -> fix_alignment
Seems at least several models of PPC (all?) do trap on unaligned
access, the kernel just doesn't complain as it does on Alpha.
Anyway, I've never seen trap-related errors from ppc; the other
arch that tends to generate these a lot is ia64, which rivals sparc
for least-efficient fix-ups.
Nah, AIUI, it's (at best) inefficient on /all/ the common linux
architectures, including x86 (trap and fix is done in microcode now,
one presumes). On other systems which refuse to fixup, it's fatal.
So for code not intended to be portable beyond Linux, or beyond x86
OSes, it's a performance problem.
For portable code, properly aligned access is *required*.
It's a user-space software problem whichever way you look at it.
regards,
--
Paul Jakma paul@clubi.ie paul@jakma.org Key ID: 64A2FF6A
Fortune:
It was pity stayed his hand.
"Pity I don't have any more bullets," thought Frito.
-- _Bored_of_the_Rings_, a Harvard Lampoon parody of Tolkein
Reply to: