Newsgroups: comp.arch
Path: utzoo!henry
From: henry@utzoo.uucp (Henry Spencer)
Subject: Re: Self-modifying code
Message-ID: <1988Jul21.202410.20049@utzoo.uucp>
Organization: U of Toronto Zoology
References: <5254@june.cs.washington.edu> <76700032@p.cs.uiuc.edu> <1090@ficc.UUCP> <1988Jul18.231158.19500@utzoo.uucp> <1404@daisy.UUCP>
Date: Thu, 21 Jul 88 20:24:10 GMT

In article <1404@daisy.UUCP> david@daisy.UUCP (David Schachter) writes:
>In article <1988Jul18.231158.19500@utzoo.uucp> henry@utzoo.uucp (Henry Spencer) writes:
>>The dynamic-compilation implementations on
>>the Blit et al run the memory at nearly full speed for the usual cases
>>of BitBlt; it is fundamentally impossible to do better...
>
>I disagree.  It is not necessarily true that good software blit code can
>run as fast as hardware supported blit operations...

On the Blit hardware?  I doubt it.  In any case, if the implementation
uses the full memory bandwidth (a more precise form of my earlier statement),
then it *is* fundamentally impossible to do better.  The ways around this
generally involve structuring the hardware in such a way that the hardware
has access to memory bandwidth that cannot be used by the software.  Other
things being equal (which they often aren't, of course), one may question
whether this is really such a smart idea.

>...  What about a CPU with a 16 bit bus competing with a blit
>chip with 64 bit access to the frame buffer?

One wonders whether it might be a better investment to put in a wider CPU,
which should speed up *everything*, not just BitBlt.

>What about a CPU that must
>fetch instructions on the same bus used for data?

What about it?  Most modern CPUs have ways around this on at least a small
scale.  The Blit CPU is a lousy old 68000, not even a 68010; by using fun
instructions like move-multiple, one can build code (for the usual BitBlt
cases) that needs instruction fetches relatively rarely.

I agree that it's possible to break the hardware in such a way that a
software BitBlt is inherently slow.  I prefer unbroken hardware myself.
-- 
Anyone who buys Wisconsin cheese is|  Henry Spencer at U of Toronto Zoology
a traitor to mankind.  --Pournelle |uunet!mnetor!utzoo! henry @zoo.toronto.edu