Geoff Wing wrote:
> Havard Eidnes <he@NetBSD.org> typed:
> :Geoff wrote:
> :> Multiprocessor corruption under i386 - been around ~forever (well,
> :> as long as we've had multiprocessor for i386 branch).
> :> I've had it on PentiumII-233MHz, Pentium4-3000, C2D-E6850's.
> : OK.  Maybe I should test with that program as well.  However,
> : the name service for mersenne.org appears to be inoperational at
> : the moment -- ns1.sbsi-net.com doesn't respond, and ns2.* replies
> : with SERVFAIL.  Bah!
> 
> Also see the thread in port-i386, "fftw_test failing on Core 2 Duo/GENERIC.MP"
> which seems to be exhibiting the same symptoms.
> I should probably try an unoptimised kernel one of these days to see if it's
> some mis-optimisation or whether it's some other missing barrier or flushing
> problem or some such.
> 
I've seen something similar when attempting to run 2 mencoder transcode 
sessions on my Athlon MP Dual system. In all cases I've been transcoding 
from MPEG2 Transport Streams to XViD.

One mencoder process is perfectly stable. Running a second (transcoding 
a different stream) will cause one of the 2 encodes to crash eventually. 
Rerunning the transcode once the first completes works successfully so 
it can't be blamed on the data stream being processes. In my experience 
this is pretty repeatable.

Mike