Port-i386 archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: [PATCH] i386 copy.S routine optimization?

Le 10/06/13 22:24, Greg Troxel a écrit :
Have you run a full test suite (atf-run) with and without the patch.
Have you done a full build.sh release run (no objdirs, just after boot
so clear caches) both ways?  results?


I didn't run any test suite for now because I only sent the patch knowing
that it's not to be applied - a priori - because of some subtlety about
x86 pipeline and cache that I actually am looking forward to understand
if someone has the time to explain it on the mailing list.

I actually showed this patch to Radoslaw Kujawa who then showed it to David Holland, the latter said, I quote:

"My guess is that it's faster the way it is due to pipelining and probable cache stalls;"

My intention, by sending my previous e-mail, was to seek some precisions about the pipeline/cache subtlety that would make my patch (which seems "good" from a newbie point of view) slower than the current code.

I am sorry if I mislead you in thinking this patch was tested and good for commit.

Best regards,

Yann Sionneau

Home | Main Index | Thread Index | Old Index