home *** CD-ROM | disk | FTP | other *** search
- Path: cup.portal.com!CygnusSoft
- From: CygnusSoft@cup.portal.com (Bruce - Dawson)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Mand2000 and '060s
- Date: 25 Feb 1996 22:00:05 -0800
- Organization: The Portal System (TM)
- Sender: pccop@unix.portal.com
- Message-ID: <150630@cup.portal.com>
- NNTP-Posting-Host: news1.unix.portal.com
-
- I have had some reports of poor performance from Mand2000 when running
- with an '060. These reports are sketchy, but basically describe slower
- than expected calc speed ('020 range) and an extremely jerky mouse.
-
- This sounds exactly like the symptoms encountered by some 68040 users
- when programs made frequent use of the 68881 instructions that were
- emulated on the '040, typically the float to int conversion instructions.
- Therefore, I am assuming that this jerky mouse/poor performance problem
- is because of instructions that exist on the 68040, but are emulated on
- the 68060. The obvious culprit, which Mand2000 makes extensive use of
- when available, would be the 32-bit by 32-bit multiply instructions.
-
- I don't have an '060, so I need somebody who does have one to run some
- tests for me. These tests can be done with the demo or the release
- version. Any version will do. This is what I need:
-
- a) Run Mand2000 and zoom in and out, without going in more than two zooms,
- and observe the responsiveness of the mouse. It should be excellent.
- b) Open a status window and observe which calculations are being used.
- It should be the sixteen-bit calcs. Zoom in until it changes
- (approximately four-five zooms). It _should_ change to FPU calcs. If
- not, that's the problem.
- c) Whatever it changes to, observe the responsiveness of the mouse. If
- it's at FPU calcs it should be excellent, if 32-bit calcs it will
- probably be poor.
- d) Finally, zoom in until the status window reports that 64-bit calcs
- are being used. At this point I expect the mouse to become very jerky
- when Mand2000 is calculating. Further zooming, into the 96-bit calcs
- and beyond would also be interesting, but less important.
-
- For some reason the emulation of unimplemented instructions on the
- Amiga destroys the real time behaviour of the interrupt queue. Also,
- while Motorola claims that the 32-bit instructions are emulated as
- fast as they would execute on an '040, this evidence suggests otherwise.
- So:
-
- 1) Can the problems with interrupts getting delayed for long periods of
- time be fixed, or is it a fundamental flaw in the Amiga. In the case of
- the Mand2000 calcs, the 32-bit multiply is getting used as between 25%
- and 90% of the instructions, thus apparently overloading things.
- 2) Can the emulation be fixed so that it runs as quickly as promised - ie;
- are these problems due to a bad installation (060.library missing or some
- such) or are these instructions doomed to be pokey.
- 3) Finally, if these instructions are hopelessly crippled on the '060,
- how many Mand2000 '060 users are there. Enough that I should test for the
- presence of an '060 and revert to 16-bit multiplies (the code is there,
- for 68000 chips)? Any beta test volunteers?
-
- Thanks for any research/answers you can give me.
-
- .Bruce Dawson, Cygnus Software.
-