On Thursday 08 February 2001  2:06 am, Reinoud Zandijk wrote:
> Hi hi Chris!,
>
> On Thu, 8 Feb 2001, Chris Gilbert wrote:
> > I've just been playing with my kinetic card, I've made the kernel not use
> > any of the on board ram (note it corrupts the console cursor)
> >
> > Anyway the test was how long it takes to do make configure for gmake
> > (configure scripts hammer the pmap rather badly on rpc)
> >
> > pkgsrc and obj dir's mounted off remote machine.  reboot done between,
> > and make clean also done so both worked off freshly extracted and patched
> > source. Test conditions were near identical as I can do at near midnight
> > :)
> >
> :(( that's not a good sign ... the main bottle neck is now the networkcard
>
> ..... have you tried a memory disc? that would REALLY show :)) I know you
> cant use the harddisc with this hack which is pretty bad :(

Hard disc works fine.  We don't do DMA on the Risc-PC, so the only thing that 
won't work is sound (which doesn't anyway) and the cursor, which is 
presumably held in DRAM somewhere (not sure why it's not at the top of the 
VRAM though)  Oh that and anyone without VRAM (but that's broken already on 
the kinetic as we try to use the top of memory for screen memory, which the 
vidc can't see as it's on the kinetic)

> > Anyway figures are:
> > 	with onboard ram	kinetic only		%
> > real		293.22		204.40			70
> > user		29.99		21.37			71.3
> > sys		225.60		151.14			67
> >
> > So it looks like there's a boost to be had from the kinetic :)
>
> Hehe... yeah well its less than i anticipated i must admit but looking at
> the remote network stuff it doenst suprise me though ... it cripples these
> figures.

Don't know, configure scripts have always been really bad on arm platform, 
it's all the fast forking and exiting of the test processes.  I might try 
shifting it onto a local disk as you suggest, but I don't expect a huge gain.

> > If anyone wants a copy of the kernel shout up, and I'll upload one, oh
> > it's based of -current (note that the diff isn't ready for the real
> > world, it's a HACK, if looks for any memory blocks that aren't kinetic
> > ram and shuffles the dram blocks down in the bootconfig)
>
> *grin* well the diffs would be interesting to see *grin* ... i'll have a
> look at the move tomorrow for even if its a 140% speed its good :)

It feels faster to use as well, however a full impementation needs us to be 
able to tag up the memory blocks as DMA capable, and Fast v slow.

> > Tommorrow I plan to compare this to the new bootloader (it makes things
> > feel slower, but I want something more substantial than just a feeling ;)
>
> Wierd.... i for some reason notice a speed increase! ... esp. on the
> screen .... hmmm.... wierd.

Grin, well as I said I'll benchmark it this evening.  It's just a feeling, 
without figures it's hard to really judge.

Cheers,
Chris