Page 2 of 2

Re: Low Performance Falcon

Posted: Wed Jul 10, 2019 10:32 am
by dhedberg
Wietze, I forced the i-cache off in our JOY demo and it then behaves exactly like the video you sent me of your computer running it. The problems starts on the same frame, so it seems like your i-cache is disabled (by hardware) or malfunctioning. It's not due to software control as in the released version of JOY I enable the i-cache by setting cacr.

Re: Low Performance Falcon

Posted: Wed Jul 10, 2019 11:43 am
by tin
Just some quick info:
That's exactly the reason for the test/test2.tos thingies - to make sure that there is no Ram access I and D cache are frozen, and SMC was used (now removed) to check for complete cache coverage of the loop in question (an illegal is written into the code after freeze, which the cpu won't see due to frozeen caches). The SMC check produces an illegal on wietzes machine, but not on mine. Also his timing is like non-cached code.
This let me to believe the cache isn't enabled. This can happen if the trace to pin 71 is broken or sth like that. To double check I did a 2nd version test2 with cache off, and both behave the same on wietzes machine and as intended on mine.

Re: Low Performance Falcon

Posted: Wed Jul 10, 2019 9:06 pm
by wietze
Ok, today we ran some more tests.

Benchmark results are the same results when testing with icache on/off. This confirms the story of Daniel.

I've checked the cpu speed; in both 8mhz and 16mhz modes, and indeed the results are different from each mode; this in order to validate that I ran my initial benchmarks in 16mhz, which I have (lower results obtained when running at 8mhz).

So this concludes that it must be an icache issue.

We have checked the CDIS (pin 71 of cpu), and if it is connected to pin 20 of j20, which it was. We checked the nearby pins (pin 70 and 72) for short circuits; which there werent.

Im kind of running out of options in order to verify if its something related to the board; or if its actually the CPU thats borked.

I can obtain a replacement 030, which Im of course willing to risk; Im just wondering if we can do more to narrow down the cause, and determine if its IN the CPU or at the board level.

Input welcome.

Re: Low Performance Falcon

Posted: Wed Jul 10, 2019 9:28 pm
by exxos
What voltage do you get on pin 122 ?