Graphics corruption in Devious Tools 019

[emoon]

http://janeway.exotica.org.uk/release.php?id=81142

http://ftp.amigascne.org/pub/amiga/UtilityDisks/DeviousTools/DeviousDezigns-DeviousTools19.adf

Expected result (screenshot taken from website)

And I get this in vAmiga 2.0b1

It seems to work with KS 2.0+ only so I use the KS from A500+ and I have set chipset Agnus: ECS
That being said I'm not 100% sure that this disk will work under A500+ settings.

Maybe same issue as #675

[mras0]

Extra info: It appears to work as expected with Kickstart-v2.05-rev37.299-1991-Commodore-A600.rom (and 3.1), but not Kickstart-v2.04-rev37.175-1991-Commodore-A500plus.rom

[emoon]

Ah that is interesting. I wonder if some bug was fixed in 2.05 that would cause this? 🤔

[mras0]

Just did a quick disassembly of both ROMs and the ECS (and AGA?) detection stuff changed between the two versions. In v2.04 it's done inline in the function starting at $fa857e while in v2.05 it's been moved to a function (that does something different) at $fb92e2.

I haven't bothered analyzing the issue, but changing Denise::peekDENISEID() to always return 0xFFFF (this is what WinUAE does for OCS) fixes it.

[dirkwhoffmann]

but changing Denise::peekDENISEID() to always return 0xFFFF

Great! It also fixes #675 (Simon the sorcerer).

[mras0]

BTW code in v2.04 works like this (clearly not hand written assembly):

UBYTE id = Custom->deniseid & 0xff;
for (int i = 0; i < 16; ++i) {
    if (id != (Custom->deniseid & 0xff)) {
        id = -1;
        break;
    }
}

And peekCustomFaulty16 returns $2000 and then $0000 16 times in a row, explaining why it detects something other than OCS. $00 -> ECS denise detected (would also have detected AGA if 2.04 supported it).

It seems like a good idea to always just return 0xffff for deniseid (even if this isn't 100% accurate).

[dirkwhoffmann]

BTW code in v2.04 works like this

Interesting! Based on the fact that they read the register 17 times, there seems to be some randomness involved on a real machine. In vAmiga, I always returned the last value on the data bus (if I remember correctly).

It seems like a good idea to always just return 0xffff

Yes, agreed. I'll do it this way from now on.

[mras0]

Interesting! Based on the fact that they read the register 17 times, there seems to be some randomness involved on a real machine. In vAmiga, I always returned the last value on the data bus (if I remember correctly).

Yes, the register isn't present on OCS, so it is reading bus noise there (http://amiga-dev.wikidot.com/hardware:deniseid). Given that they changed it (rather quickly), they must have found that it didn't always work reliably. The revised version isn't too different, but it disable interrupts, compares the full word value, mixes in some useless TST.W D5 instructions and does a read from intenar just before reading deniseid presumably to reduce the number of false positives.

Still doesn't seem very robust to me, but since it doesn't appear to have changed through 3.1 (I don't have 3.2 to check) it must have been good enough.

[dirkwhoffmann]

Fixed in v2.0b2 which is now online.