68882 presumably working oddly

[rockersuke]

Hi!

I thought my 50 Mhz 68882 was working fine and giving some extra speed to FPU-ready soft in my 68030-50 Mhz Blizzard MKIV when I noticed that Vistapro renders were totally wrong. There goes some pics of the same DEM file in both low and high quality with the FPU and the no-FPU versions in my A1200:

I tried some other FPU soft as some fractal exploration program or POV FPU version. Images were erratic, as this POV example: the funny pic is from my Amiga and the "right" one was rendered with WinUAE emulating a 68030+68882 setup:

CPU command, showconfig or Sysinfo detect the 68882 correctly. Vista renders were around 50% faster with the FPU, so It's supossedly doing its job... but with very weird results.

After many testing around (wich included replacing all the fast RAM) lack of stack, overheating, or defective memory have been discarded as causes of this misbehaviour.

Anything I forgot to check? Any suggestion (apart from being on the hunt for another copro ) ?

--

[NovaCoder]

That's very strange, an FPU that gets it sums wrong is not much use

I would say that it was very unlikely that the FPU was 'broken', it should either work or not.

You can't really compare the WINUAE image with the real thing either (apples and pears).

I suppose you could just use a simple program that did some basic floating point arithmetic to test your FPU. I could write one for you pretty easily but there's prob something similar on AmiNet already.

[rkauer]

Also GVP install disk for the old A3001 have a very good FPU test. You can find the image on Installer's Heaven.

[Loedown]

Originally Posted by NovaCoder

*SNIP* I would say that it was very unlikely that the FPU was 'broken', it should either work or not.

Unfortunately, it's not so cut and dried as that. Often chips can fail in strange ways and if the 68882 has partially failed, which to me is what it appears, then it can certainly give some of these strange results. I also think that WinUAE and original hardware should give identical results, unless there's a coding fault in WinUAE.

I have the original GVP install disk, which does have a FPU test, which I can easily up to anywhere it's required, it's the Faaaaassstttttprep disk, used with the GVP '030 Combo board.

Paul

[rockersuke]

Thanks everyone for the suggestions!

I cannot check now (I'm at job) but I'm recollecting some "floating point" stuff from aminet wich could give some hints and also taking a look at Installers Heaven for those installers with FPU tests. I'll tell you how It goes!

--

[rockersuke]

Well, there it goes!

I tried several tests, including system-info utilities from the GVP disks. All of them were quite efficient on telling me the FPU was there and how damn fast It is, but couldn't give any clue about why all those MFLOPS were not giving the right results. Benchmark utils offered me a lot of numbering wich, due to my ignorance on low level FPU workings, I couldn't interpret in any meaningful way ^_^'

Then I found something closer to my level: a very simple FPU Celsius to Farenheit converter. As I was expecting results were wrong, consistently wrong, I mean, no random output. It insisted on 0 Celsius corresponding to 36.4444 Fº, while WinUAE was offering the right number, a plain 32.0.

Guess It seems that definitively the FPU is somehow broken, I think!

--

[NovaCoder]

Originally Posted by rockersuke

Well, there it goes!

I tried several tests, including system-info utilities from the GVP disks. All of them were quite efficient on telling me the FPU was there and how damn fast It is, but couldn't give any clue about why all those MFLOPS were not giving the right results. Benchmark utils offered me a lot of numbering wich, due to my ignorance on low level FPU workings, I couldn't interpret in any meaningful way ^_^'

Then I found something closer to my level: a very simple FPU Celsius to Farenheit converter. As I was expecting results were wrong, consistently wrong, I mean, no random output. It insisted on 0 Celsius corresponding to 36.4444 Fº, while WinUAE was offering the right number, a plain 32.0.

Guess It seems that definitively the FPU is somehow broken, I think!

--

Yes it looks like it's forgotten how to add up

Are you sure it ever worked correctly BTW?

[musashi5150]

Now just to put it on eBay as "untested"

[rockersuke]

Hello!

Just a quick note to let you know the happy ending: new FPU arrived today and both renders and basic temperature converting are working right! So now I know the old one was definitively defective.

--

[prowler]

Thanks for the update, rockersuke!