First Amiga 600 FPGA Accelerator - Vampire 600

[Toni Wilen]

Originally Posted by majsta

Do you want to say that everything is fine and that is somehow connected to my problems running sysinfo on WinUAE where sysifo crashes when I add more FastRam?

Sysinfo is unstable in some configurations. It is normal. It usually works if booted without startup-sequence.

[majsta]

So maybe just maybe we can consider that this is working and I should conclude again that I have lost so much time about nothing because I had 64MB autoconfigured like this month ago

[robinsonb5]

Originally Posted by majsta

So maybe just maybe we can consider that this is working and I should conclude again that I have lost so much time about nothing because I had 64MB autoconfigured like this month ago

Well... there *is* still a problem - at least my end, with that memory test failure.

The lesson here, I guess, is that you can safely ignore misreporting from SysInfo provided everything else works.

Disabling the cache didn't fix the SysInfo error - but the question is does it fix the memory check error?

[matthey]

@majsta
Yes, you should check your memory. Robinsonb5's problem may be specific to his hardware but it may not. It would be very helpful for diagnosing at the very least. I have PM'ed you a small and easy to use memory checker. I didn't recognize what robinsonb5 was using.

[majsta]

Hey this is amazing tool!!! I have never realize that such thing exist. Let's try it. Thank you! I stopped this test after half an hour and didn't report anything wrong. I know that those memtests need to be running complete day but my experience with memory testings on PC says that if there are serious problem it will be detected in first half an hour rest of the time will just show you how huge problem is.

[robinsonb5]

Originally Posted by matthey

@majsta
Yes, you should check your memory. Robinsonb5's problem may be specific to his hardware but it may not. It would be very helpful for diagnosing at the very least. I have PM'ed you a small and easy to use memory checker. I didn't recognize what robinsonb5 was using.

I was using this: http://www.logikstate.com/MemCheck.adf
I don't know where it came from originally, but DJ_KARTA on the Chameleon mailing list gave that to me when I was debugging memory issues with the Chameleon 64's Minimig core.

I use other testers as well, but that one's useful because it freezes the system while testing so it can test memory that's in use. My gut feeling is that the errors I'm seeing happen when the tester hits a region of memory that's aliased to its own data segment. The fact that the first run gives a different result from subsequent runs, and that subsequent runs are absolutely consistent is what makes me think it's cache related - but it could simply be that on the first run the tester loads to a different address from subsequent runs. I'd say there's a very good chance there's a bug in either the SDRAM controller or cache which results in the second 32 meg being aliased to the first.

[robinsonb5]

Originally Posted by majsta

Hey this is amazing tool!!! I have never realize that such thing exist. Let's try it. Thank you! I stopped this test after half an hour and didn't report anything wrong. I know that those memtests need to be running complete day but my experience with memory testings on PC says that if there are serious problem it will be detected in first half an hour rest of the time will just show you how huge problem is.

We're actually testing for an unusual situation here. We're not testing for *faulty* memory. We're testing for either a bug in address decoding, caching, or SDRAM controller row/column mapping - or maybe even a bad solder joint - none of which are situations a "normal" memory tester is designed to catch. So the conventional wisdom about running the tester for a day doesn't really apply here.

Is that test running with or without cache, by the way?

[majsta]

I don't know but I know for a fact that I tested this sdram controller without adding it into FastRam and didn't find any problems I had before on my 4bit SDRAM due to long clk trace. Also LFSR test is ok.
I m running any game or program now and everything works. Just to notice all of this is with cache, 6.27Mips and 64MB of FastRam.
Also I have run just now this test you suggested and didn't report any error. Here is the picture.

Now I can only think of bad solder joint AGAIN, or we did something different in the codes.

[matthey]

Originally Posted by majsta

Hey this is amazing tool!!! I have never realize that such thing exist. Let's try it. Thank you! I stopped this test after half an hour and didn't report anything wrong. I know that those memtests need to be running complete day but my experience with memory testings on PC says that if there are serious problem it will be detected in first half an hour rest of the time will just show you how huge problem is.

It's actually part of a hardware checker that connects to the Amiga ports so some of the tests aren't functional without it. It's still a nice little program without the hardware. Running the test for 1 pass checks all the memory but it can only check what is not allocated. Robinsonb5's memory checker sounds like it has some advantages.

@robinsonb5
Thanks for the link. That does look like a very good memory tester.

[majsta]

Yes I remember now. Actually I think I have seen that specific part on Ebay for sale few years ago. I wanted to buy that but since I didn't have Amiga there was no point :P
Anyway I think that this is working :P

[robinsonb5]

Originally Posted by majsta

I m running any game or program now and everything works. Just to notice all of this is with cache, 6.27Mips and 64MB of FastRam.
Also I have run just now this test you suggested and didn't report any error. Here is the picture.

Excellent news - then it looks like both the SDRAM controller and Cache are fine - and there's some other problem my end.

Now I can only think of bad solder joint AGAIN

It's possible, but I don't *think* so - because the board passes the ZPU-based memory tester I put together a couple of weeks ago specifically to test for that! (An Amiga with dual-core CPUs So if there *is* a bad solder joint I have a bug in my memory tester!

I think it's time to write an ADF of AmosPro to disk and do some poking...

[robinsonb5]

OK I'm an idiot!

As soon as I started poking in AMOSPro I could see that there were RAM aliases, then I remembered that I'd mapped the 8 meg of Zorro II RAM to the upper 16 meg of the SDRAM like so:

  sdram_addr(25)<= sel_zii_fast;
  sdram_addr(24)<= sel_zii_fast or cpu_addr(24);
  sdram_addr(23 downto 1)<=cpu_addr(23 downto 1);

(My plan was to map the lower 48 meg directly and use the upper 8 meg free for kickstart images, $C00000 RAM or whatever else)
When I tried out the Zorro III autoconfig code, I forgot to undo that mapping!
Having changed it to

sdram_addr(25 downto 1)<=cpu_addr(25 downto 1);

All is well, and it runs just fine with 64 meg of RAM.

[matthey]

Good to hear there is nothing wrong with the hardware. The human brain can always be reprogrammed with some difficulty .

[lukassid]

Great progress guys.

[majsta]

@robinsonb5 hehe yes that was the first thing I solved simply but I have done like this.

sdram_addr(31 downto 1)<=cpu_addr(31 downto 1);

That's complete idea about Vampire boards. You can see how this could give us major progress because now actually first time in Amiga history we have hardware where all of us can participate and experiment to the final result. All of us like team

[Lord Aga]

Majsta, which version of Sysinfo are you running ?

[majsta]

I think that 3.24 version. Also some time ago I tried new version but that one had bigger problems so I contacted author and he said that those problems were related to using different compiler than on original version?

[Lord Aga]

Hm, yes, version 4 does give slightly different speed results. But I reckon it may be better at recognizing larger memory add-ons.

[majsta]

Yes version 4 detects 64MB. Thank you Also I went to http://sysinfo.d0.se/ support page for new SysInfo and here is what I have found.

Updated memory size calculation in Memory Info to handle 64MB+ sizes correctly.
Updated board size calculation to handle 64MB+ correctly, added MB suffix for sizes greater than 2MB.

Also speed is the same, and CPU also detected. All seems fine now.

[Lord Aga]

Woohoo, I'm helpful