NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/sys/amiga/programmer/5048

Internet Message Format | 1996-08-05 | 1.9 KB

Path: ar.ar.com.au!not-for-mail
From: [e-mail] (Storm)
Newsgroups: comp.sys.amiga.programmer
Subject: Re: One hardware-basher's manifesto
Date: 8 Mar 1996 16:19:52 +1100
Organization: I need to put my ORGANIZATION here.
Message-ID: <4hog1o$[e-mail]>
References: <04000205714080640503@BIRDLAND> <4ge8na$[e-mail]> <[e-mail]> <4gmlg9$[e-mail]>
NNTP-Posting-Host: ar.ar.com.au
X-Newsreader: TIN [UNIX 1.3 941216BETA PL0]

Michael van Elst ([e-mail]) wrote:

: .. and the CPU. A 040 does always use bursts for cachable data, the only
: way to turn bursts off is to disable the cache and _that's_ slowing down
: everything significantly. Disabling bursts almost never helps.

I have a question about an 040's data cache! Well, not so much a question,
I just want to make sure my understanding is correct..

With the copyback mode, any writes to a cached memory location are only
written to the cache, right? And then, when that bit gets pushed out of
the cache to make room for something else, they get written to 'real'
memory, right?

So, if you flush the cache (after self-modifying code for instance), it
will possibly be forced to write a large amount of stuff to main memory,
right?!

I want to know, because I've been using self-modifying code taking into
account that the exec routine to flush caches take ~1 raster line to
execute on a standard A1200. But then it occurred to me that due to this
copyback cache on the 040, it could take a _seriously_ longer time -
long enough that self-modifying code could almost never provide any
speedup?

Am I right in all of this? Or have I missed something?

--                  ______________________________
                    \_/ "\/\/\__"\/ "\/ "\/\__"\_/
Storm / Cydonia      / / / / / / / / / / / / ' /       Packing class
                    / /\/>  / / / / / / / / / /__     & kicking arse!
    (coder)         \__/ \_/\__/\__/\/\/\/\/\/ \/