On Sat, 19 Feb 2000, Richard Henderson wrote:
> On Sat, Feb 19, 2000 at 12:49:30PM +0100, Gerard Roudier wrote:
> > It is fine to be impatient. I have been impatient for about 6 months about
> > a fix for the alpha __delay()/__udelay() stuff that can be 200% wrong due
> > to inlining and that notably breaks the SCSI clock speed measurement by
> > ncr/sym53c8xx at initialisation. This stuff does not seem to have been
> > fixed in all still alive linux branches (I mean 2.0/2.2/2.3).
>
> No, it has in fact been fixed.
>
> ".align 4\n"
> "1: subq %0,1,%0\n"
> " bge %0,1b\n"
> " nop"
>
> That last nop is magic. The type of instruction that follows the
> branch affects slotting on ev6, which affects the speed of the loop.
Thanks for the info.
Hmmm... Let this `slotting' be a bit smarter in some later chip and issue
will come again. The change seems to appear in linux-2.2.14 which has been
available in January IIRC. This explains that still most actually used
distributions haven't it yet.
Btw, how does this trick victimize the BOGOMIPS result? I don't care that
much about this number, but it seems numerous people are very sensitive to
it and it may happen customers they will get or not depending on this
bogus number being comfortably high or not. ;-)
Gérard.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Feb 23 2000 - 21:00:24 EST