Re: .i368 vs .i586



Gregory Leblanc <GLeblanc@cu-portland.edu> writes:

> The "standard" for creating RPMs is to use a flag that says "optimize" for
> the 486 processor, but do NOT make it incompatible with other processors.  I
> haven't seen what kind of a difference change it from i486 to i586 processor
> would make, but perhaps somebody could try that.

Well, though this is possible it does not buy you much.  Things which
are influenced by this are alignment, preferences of certain
instructions etc.  If 586 is the highest version you want to make the
requirement then this is ok.

Where the current scheme really sucks is on i686 and up.  Here simple
tuning is not enough.  These processors are designed to run code fast
if the new possibilities are used (especially the coditional moves).
Due to the deep pipeline ode generated the old way (using jumps)
is much slower then CMs.

So, if you want to wat exactly one binary package then tune for i586.
It'll run everywhere.  Otherwise create another one which runs only on
i686 and up and whih will probably be substantially faster.

-- 
---------------.      drepper at gnu.org  ,-.   1325 Chesapeake Terrace
Ulrich Drepper  \    ,-------------------'   \  Sunnyvale, CA 94089 USA
Red Hat          `--' drepper at redhat.com   `------------------------



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]