Question for Toni: What was Marvin Marvellous interrupt fix?

[rsn8887]

Hello Toni,
I am trying to back port a fix you made over 15 years ago (!) to fix Marvin's Marvelous Adventure. This would allow me to make the game work on UAE4All2 Vita/Switch.

You mentioned the fix here:
http://eab.abime.net/showthread.php?t=4589

You mentioned an interrupt for the music routine was ignored, and that you fixed that.

I tried the GitHub repo, but it doesn't go back quite far enough for me to do a diff. Any chance you remember what you fixed in the code?

[Toni Wilen]

I think it was intlev() (custom.cpp) related, check that level 6 interrupt mask uses both 0x4000 and 0x2000.

[rsn8887]

Awesome, thanks for the hint I will check this out and see if I can back port this fix.

[rsn8887]

Spot on, interrupt 6 mask was using “0x2000” instead of “0x6000”. This single number change fixed it! Amazing, thanks!

Are you on Twitter so I might put a shout out to thank for your help in the next uae4all2 release tweet?

While I have your attention, any chance you remember how you fixed IK+ sound fx? I think I remember the sound fix worked at some point in WinUAE even without cycle exact...

[Toni Wilen]

Originally Posted by rsn8887

Spot on, interrupt 6 mask was using “0x2000” instead of “0x6000”. This single number change fixed it! Amazing, thanks!

Are you on Twitter so I might put a shout out to thank for your help in the next uae4all2 release tweet?

I still don't do social media..

While I have your attention, any chance you remember how you fixed IK+ sound fx? I think I remember the sound fix worked at some point in WinUAE even without cycle exact...

It wasn't a fix. It is either IK+ or some other game breaking unless in CE mode. (I think Warhead was one). Audio interrupt must not come too late or too early..

[rsn8887]

Thanks that makes sense. I remembered that CE fixed it for good, but I thought there was some earlier fix. I don’t think I can back port CE mode into UAE4ALL2. It requires too many changes.

Do you maybe remember a fix for Celtic Legends? I get no keyboard response during initial language select screen :/ not sure it might be one of those games that require CE, too.

[Toni Wilen]

Originally Posted by rsn8887

Do you maybe remember a fix for Celtic Legends? I get no keyboard response during initial language select screen :/ not sure it might be one of those games that require CE, too.

Keyboard interrupt is weird but I can't see anything that would require CE.

It reads keycode, sets keyboard handshake, then polls CIAICR for another keyboard interrupt (!?), then does handshake wait (vpos check), disables handshake and exits.

My guess is that it hangs in interrupt poll loop. Check if it keeps reading CIAICR in tight loop with CPU PC around $5148. (hundreds of CIAICR reads per frame)

ReadCIAA() case 13.

[rsn8887]

Awesome, thanks! I will look into that ASAP.

[Foebane]

I don't blame anyone for not being on social media, I'm only "slightly" on Facebook. The whole thing will be blown to pieces by Article 13 when it's passed, anyway. Say goodbye to YouTube et al!

[Toni Wilen]

If it gets stuck there, it most likely is CIA related update. There shouldn't be any major updates and most changes should be relatively easy to merge.

[rsn8887]

Hello Toni, my method of testing is very crude: make change to source, compile, run on Switch, see if it hangs or not. Repeat.

I did a bunch of this type of debugging, focussing on READCIAA case 13 and related code. I also compared code to newer UAE versions and tried a bunch of changes, but to no avail.

The code in Uae4All2 is here, note it is code from a very old UAE version (0.8.22):
https://github.com/rsn8887/uae4all2/...c/cia.cpp#L487

Note the RethinkICRA() during case 13 does nothing, since ciaaicr is set to zero just before it is called. But that is similar to what happens in more modern UAE. I tried all kinds of changes there, no success.

One place an interrupt on keypress is set is during a certain keyboard hack routine. It is some kind of delay with a state machine. Supposedly it was implemented to allow the user to type ahead many characters without losing any keypresses:
https://github.com/rsn8887/uae4all2/...c/cia.cpp#L348

This keyboard hack is not present in more modern UAE.

It was my immediate suspect. It should be noted it seems to work with all other games. As far as I understand, RethinkICRA() called during this hack should cause the interrupt 2, but only every once in a while. I have now tried to extensively debug this function, but with no success for Celtic Legends.

Another funny observation: The WHDLoad quit key still works during the language select screen.

[Toni Wilen]

Originally Posted by rsn8887

One place an interrupt on keypress is set is during a certain keyboard hack routine. It is some kind of delay with a state machine. Supposedly it was implemented to allow the user to type ahead many characters without losing any keypresses

It still does the same but now it works like real hardware, wait for handshake pulse before sending next (buffered) character. Originally (I don't know why) it didn't bother with keyboard handshake and sent the next key code after it was read.

EDIT: Actually, I think it may be related to resync, keyboard starts sending resync bits after longish delay if it does not get handshake quickly enough and this may be what the weird keyboard code in game is waiting. It probably isn't designed to work that way, someone probably got confused with keyboard operation..

Check and implement kblostsynccnt

[rsn8887]

Thanks for the hint!

I found out something very interesting: UAE4All2 is based on (Win?)UAE 0.8.22. But in WinUAE 0.8.22, the keyboard works just fine with Celtic Legends.

So it looks like the problem is one of the changes that was made by the Pandora/Android developers later.

I found the Winuae 0.8.22 source code on sourceforge (thank god for that). So now I can simply compare the two sources, and hopefully find the bug!!!

[rsn8887]

OK I fixed it! The ciaasdr_cnt handling was commented out in places and missing in others.

A src comparison with 0.8.22 R9 showed the problem clearly.

Here's my commit that fixes Celtic Legends:
https://github.com/rsn8887/uae4all2/...2665d6396987ad

I hope it doesn't break anything else.

I will test it and include the fix in a new release of UAE4All2 Vita/Switch.

Thanks, Toni!