home *** CD-ROM | disk | FTP | other *** search
- Path: munnari.OZ.AU!metro!metro!news
- From: accolyte@wr.com.au (Accolyte)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: need help on A1200/030 board programming
- Date: 18 Feb 1996 03:51:31 GMT
- Organization: Information Services, The University of Sydney, NSW, Australia
- Distribution: inet
- Message-ID: <1596.6622T138T1200@wr.com.au>
- References: <4f4hiu$oum@ionews.io.org> <38232327@kone.fipnet.fi>
- <743.6612T966T1128@wr.com.au> <4fd491$k8r@sunsystem5.informatik.tu-muenchen.de>
- <2133.6613T901T923@wr.com.au> <4ft9ga$1fh@sunsystem5.informatik.tu-muenchen.de>
- NNTP-Posting-Host: dialup10.wr.com.au
- X-Newsreader: THOR 2.22 (Amiga;TCP/IP) *UNREGISTERED*
-
-
- >|> > That's true, seen from that point it is a problem. Don't want to sound
- >|> > like one of the OS-gurus ;) but if software behaves that way it mostly
- >|> > is done not so well.
- >|>
- >|> :) It's just a shame that you can't rely on the chip-speed. I mean, to
- >|> make your code compatible with accelerators you can't use 100% of the
- >|> stock 1200's available accesses to chip. I guess you can screw over peopl
- >
- > WHY ? you can do
- >
- > loop:
- > move.l d0,(a0)+
- > dbra d7,loop
- >
- > this will store 7mb/sec on A1200 and 4mb/sec on A4000.
- > the code is compatible, so what do you mean ?
-
- Sorry, I didn't mean compatible. I mean "to make your code work
- on computers with slower access times to chip".
-
- > Do not rely on things like "blitter is earlier ready than cpu".
- > just test for it. won't make anything slower.
-
- I don't, that'd be crazy. it's just that I'm using 95% of that 7mb/sec
- on a 1200 without fastmem, so I guess it'd drop into 2 frames on a
- 4000 without fast :\
-
-
- >|> > There are exceptions, for example cpu-scroll in chipmem. Will be just
- >|> > faster on synchronous cpus. What effect are you refering to ?
- >|>
- >|> No effect in particular, just a cpu-copy for example.
- >|> Try a cached copy loop from one bitmap to another (all in chip-ram).
- >
- > well, copy mostly loops don't benefit from caches, as they copy
- > mostly more than 200% of cache size.
- >
- > data cache is AFAIK turned off in chipmem.
-
- Still, even ignoring caches; A copy loop from one chip buffer to another
- chip buffer, is slower on my accelerated 1200 than it is without the card
- installed. That bites.. <sigh>
-
-
-
-