To be sure, 64-bit alignment is not required on x86_64, and the hit from
the extra insn cycles from the transparent fix-up is negligible (< 0.05%
on x86_64). And modern RISC hardware that "we" would be likely to care
about, e.g. PPC and aarch64 (i.e. ARM), also does transparent fix-up, so
it's not an issue, per se, for those platforms; 'though we don't know
what the hit is for the fix-ups on them.
However---
gcc -fsanitize=undefined (and clang, where -fsanitize originated) do
care about alignment, and trying to run gluster built with -fsanitize is
currently a mild exercise in futility.
-fsanitize is still fairly new. It first appeared in gcc-4.8, which in
Fedora first appeared in F20 AFAIK.
It's not clear to me yet how much value there is, or could be, in using
-fsanitize during our SDLC. So far it has only uncovered the mem-pool
alignment problem. ;-)
Anyway, this is just an FYI. Feel free to share any thoughts you have
here in gluster-devel about the merits of fixing (or not fixing) the
mem-pool alignment.
--
Kaleb
_______________________________________________
Gluster-devel mailing list
Gluster-devel@gluster.org
http://supercolony.gluster.org/mailman/listinfo/gluster-devel
