Date   : Wed, 12 Nov 1997 17:08:13 +0000
From   : Dave Gilbert <gro.gilbert@...>
Subject: Re: Speeding up BeebEm

Hi,
  My belief is that the performance loss in Beebem is mainly due to the
added complexity due to emulating hardware devices. I reckon this
significantly reduces the cachability of the code and the benefits of
branch buffers etc.

  The timing accuracy of Beebem is damn good - the only program I know
of which has timing problems is Revs; I'm not sure how to acheive that
type of accuracy with threads. Threads would also be difficult to
implement portably.

  On the Unix version of beebem with sound support (0.6 was the last one
I released - I'm not sure about hte one modified my MikeBUK@...) I think
I allowed it to regulate its speed based on the sound buffer; so it
should run at about the right speed.

  It was originally developed on a P90 with 16MB of RAM running Linux
and ran at about beeb speed without sound support; with sound support it
gets slower - but I would have thought a P200 would have done fine.

  It doesn't use much RAM.

Dave

-- 
 David Alan Gilbert - gro.gilbert @ treblig.org ---- by bread alone. He