Path: utzoo!attcan!uunet!husc6!think!ames!killer!tness7!tness1!nuchat!sugar!ficc!peter
From: peter@ficc.UUCP (Peter da Silva)
Newsgroups: comp.arch
Subject: Re: Self-modifying code (and bitblt)
Message-ID: <1151@ficc.UUCP>
Date: 25 Jul 88 16:30:42 GMT
References: <5254@june.cs.washington.edu> <76700032@p.cs.uiuc.edu> <302@laic.UUCP> <7439@ico.ISC.COM>
Organization: SCADA
Lines: 49

In article <7439@ico.ISC.COM>, rcd@ico.ISC.COM (Dick Dunn) writes:
      Me:
> > > >You can save an even greater factor by building the BitBlt into the
> > > >hardware....
> > > No.  You can save a large factor over *poor* software implementations of
> > > BitBlt...
  Me:
> > However, a hardware BitBlt using DMA can be operating while the CPU is
> > servicing other processes, with no slowdown in CPU speed...

> First, that's only true if either the bitblt is being done to *and* from
> memory not accessed by the CPU, or the memory bandwidth is high enough to
> handle both the CPU and the bitblt DMA cycles.  I'd guess that's rare; we
> seem to be afflicted with memory-bound CPUs in a lot of systems.

Well, for the Amiga the memory bandwidth is high enough, *and* the blitter
only affects display memory. Memory requests for the CPU to memory outside
the display memory area (the first 512K as it turns out) is on another bus.

> But more importantly, in many cases you can't do anything else because
> either (a) there's only one process running or (b) you can't afford the
> context-switch time.

The Amiga context switch time is on the order of a procedure call... the
Amiga Exec is a realtime operating system... all processes are "lightweight"
in UNIX jargon.

> Unless you get a bitblt operation that takes longer
> than *two* context switches (out and back), it costs more.  For the case
> where you're bitblting characters (probably the most common case), you will
> lose big.

When scrolling a window you just blit the whole thing instead of doing it
a character at a time, and you win big (80 by 24 by 8 by 8 bits == 15,360
bytes). As for text, see below.

> [dynamic compilation]
> Ummm...but what does this prove?  Not much, I think, until someone actually
> tries it and finds out whether it's slower.  Given that the Amiga already
> has a useful bitblt (of whatever form), there's no great incentive to write
> another.

Well, there are already programs out that SetFunction the Text() routine to
speed up text drawing of fixed 8-by-8 characters (the common case, as you
say) and there is apparently lots of dynamically compile BitBlt code out
there waiting to be nabbed.
-- 
Peter da Silva  `-_-'  Ferranti International Controls Corporation.
"Have you hugged  U  your wolf today?" (uunet,tness1)!sugar!ficc!peter.