home *** CD-ROM | disk | FTP | other *** search
- Path: hydra.zrz.TU-Berlin.DE!rawneiha
- From: rawneiha@hydra.zrz.TU-Berlin.DE (Philipp Boerker)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: Free Cycles
- Date: 20 Feb 1996 16:17:23 GMT
- Organization: Technical University Berlin, Germany
- Message-ID: <4gcs6j$d6c@brachio.zrz.TU-Berlin.DE>
- References: <xDPTwMD0aRz1@azure.people-x.people.de> <xIIBkMD0aez2@azure.people-x.people.de>
- NNTP-Posting-Host: hydra.zrz.tu-berlin.de
-
- azure@people-x.people.de (tim bscke) writes:
-
- >TB> I tried to measure the amount of free cycles I have after a chip-write
- >TB> (MOVE.L d0,(chip)+) on my computer , a A1200/30/28 Mhz.
- >TB> But not matter, which resolution an depth I took (different bandwidths)
- >TB> I got alwas 12 free cycles.
-
- If you have timed this with different workbench modes I would suspect that
- the WB does 16bit fmode for lores, 32bit fmode for hires and 32bit double CAS
- for Shires! The mem times would stay the same.
-
-
- > Free cycles between two:
-
- > move.l (fast)+,(chip)+ ;18 cylces
- > clr.l (chip)+ ;24(!!!) cycles
-
- > Timing for A1200/68030/28 Mhz
-
- > Why ?? This seems to be really strange to me.. esp. the clr.l..
-
- I guess that clr waits for the mem because of cache/mem-coherence in
- multiprocessor environments. Move.l is executed before the mem is
- ready so it's a few cycles faster. Still weired behavior!
-
- Greets,
- Phil.
- grond/matrix
-
-