Problematic Demos.

[amilo3438]

Originally Posted by Toni Wilen

EDIT: MadeInCroatia for some reason does not immediately disable CIA-B interrupts when it takes over the system but only later and if TOD interrupt triggers during that period: crash.

So it will crash on WinUAE, but on a real A500 it will not crash, right? (this is what confuses me)

p.s.
I found some general info about the bug here (probably you know for it) in "Time-of-Day (TOD) Clock" and "Errata" parts.
http://en.wikipedia.org/wiki/MOS_Technology_CIA

Due to a bug in many 6526s (see also errata below), the alarm IRQ would not always occur when the seconds component of the alarm time is exactly zero. The workaround is to set the alarm's tenths value to 0.1 seconds.

The 8520 revision of the CIA, used in the Amiga, modified the time-of-day clock to be a 24-bit binary counter, replacing the BCD format of the 6526. Other behavior was similar, however.

In addition to the aforementioned alarm clock interrupt bug, many CIAs exhibited a defect in which the part would fail to generate a timer B hardware interrupt if the interrupt control register (ICR) was read one or two clock cycles before the time when the interrupt should have actually occurred.

EDIT: In addition, found some more infos here:
http://forum.6502.org/viewtopic.php?...e12c0&start=30

the article: George Hug, Toward 2400 (Transactor Vol. 9, Issue 3 - Feb. 1989 p.62)
http://archive.org/stream/transactor...e/n63/mode/1up (see in RS-232 bugs)

[amilo3438]

Key West BBS Intro 2 (Intro) -> http://janeway.exotica.org.uk/release.php?id=73893

A500 ECS Agnus 1MB ... some glitch shows on left-below side on screen after a while!

[Toni Wilen]

Originally Posted by amilo3438

Key West BBS Intro 2 (Intro) -> http://janeway.exotica.org.uk/release.php?id=73893

A500 ECS Agnus 1MB ... some glitch shows on left-below side on screen after a while!

A500 confirmed and does not seem to be ECS specific.

Originally Posted by Toni Wilen

Worst ever report but untick "CIA TOD Bug" in Advanced chipset as a workaround.

Thanks for your answer!
In this mode, this demo works.

For me demo "madeinCroatia" has always been a kind of test for compatibility. It does not always run the first time, even on the original Amiga1200/600.
I think the question can be closed.
Thanks again for your answer and great WinUAE emulator. I look forward to the release of new versions.

[amilo3438]

Originally Posted by Sergey

For me demo "madeinCroatia" has always been a kind of test for compatibility. It does not always run the first time, even on the original Amiga1200/600.

Dont know does it works on a real A1200, but in WinUAE Quickstart A1200 base config. it will show a guru message after a while.

[Toni Wilen]

Unreliable startup most likely confirms it is disk timing sensitive (takes too long -> crash due to TOD interrupt). Disk is practically the only random part in hardware. (initial disk position, how long it takes for disk to rotate at full speed)

[vagrant]

Delight - Overdose -- http://janeway.exotica.org.uk/release.php?id=7257

It's been so long since I have seen this demo, I can't remember if this is correct behavior or not.. but I assume it's not. (sorry if a bit vague)

During certain plasma effects the framerate drops significantly.
1st instance is about half way through the demo.
Here's a short 50fps video capture of what I mean: [ Show youtube player ]

The capture is from standard A500 quickstart 1.3
Again, apologies if it's correct behavior

[Toni Wilen]

Originally Posted by ED-209

Delight - Overdose -- http://janeway.exotica.org.uk/release.php?id=7257

It's been so long since I have seen this demo, I can't remember if this is correct behavior or not.. but I assume it's not. (sorry if a bit vague)

During certain plasma effects the framerate drops significantly.
1st instance is about half way through the demo.
Here's a short 50fps video capture of what I mean: [ Show youtube player ]

The capture is from standard A500 quickstart 1.3
Again, apologies if it's correct behavior

It is correct behavior. A500 confirmed. Some plasmas are slower than others.

[vagrant]

Thanks Toni

[Toni Wilen]

MadeInCroatia

Properly fixed.

68000 interrupt line (IPL) sampling seems to be microcode controlled after all because for example some MOVE.L variants fetch interrupt line state before write and some after write. (even when cycle sequence is exact same between variants after writes and only difference is effective address calculations before writes)

[Toni Wilen]

Made In Croatia problem better explanation (before everyone forget what was the real problem..)

It has following code:

000002AC 47f9 00bf d000           LEA.L $00bfd000,A3
000002B2 177c 007f 0d00           MOVE.B #$7f,(A3, $0d00) == $00bfdd00
000002B8 422b 0e00                CLR.B (A3, $0e00) == $00bfde00
000002BC 2b7c a000 2000 ffca      MOVE.L #$a0002000,(A5, -$0036) == $00dff09a
000002C4 21fc 0000 05b0 0078      MOVE.L #$000005b0,$00000078

Because of TOD bug, CIA-B ALARM interrupt is active at this point (and has been active since the initial vector part started, confirmed on my real A500 with oscilloscope connected to CIA-B interrupt pin)

Move to $bfdd00 does not clear the interrupt, only reads clear interrupts, disabling CIA interrupt by writing to $bfdd00 won't disable already active interrupts.

Then comes move.l #$a0002000,-$36(a5). Enable CIA-B interrupt ($a000->INTENA) and attempts to clear currently active CIA-B interrupt ($2000->INTREQ).

INTREQ write does nothing because it is external (CIA) interrupt. (Logic analyzer confirmed, CPU IPL lines won't change)

Finally interrupt vector is set, same vector that previous line enabled. If interrupt starts before this write: crash.

Reason: some MOVE.L variants read interrupt state before writes and some after writes. For example MOVE.L #$a0002000,$dff09a will read it after both writes (and would crash) but MOVE.L #$a0002000,(an) or x(an) read it before writes (and work). All 3 variants have 100% identical write,write,prefetch cycle sequence. (Most/all MOVE.W's seem to read if after writes)

This is yet another undocumented 68000 behavior, it seems interrupt line sampling is microcode controlled and different instructions sample it in different phases during instruction execution... (Fortunately in Amiga only MOVEs matter because custom registers are read-only or write-only)

[mc6809e]

Originally Posted by Toni Wilen

This is yet another undocumented 68000 behavior, it seems interrupt line sampling is microcode controlled and different instructions sample it in different phases during instruction execution... (Fortunately in Amiga only MOVEs matter because custom registers are read-only or write-only)

Ah, this explains why MOVEM.L + DIVS has a longer latency than a pair of MOVEM.L or DIVS.

MOVEM.L must sample very early and DIVS very late. Together they have potential 300+ cycle latency.

[amilo3438]

Captain Fizz Meets The Blaster-Trons (Crack Intro) by Beyonders & Vision Factory
http://janeway.exotica.org.uk/release.php?id=73946

tested on A500 1.2/1.3KS OCS 0.5MB configuration

winuae_2700 (2013.12.05) (OK)
test winuae_2710b0 (2013.12.24 09:56) (OK)
test winuae_2710b0 (2013.12.24 21:07) (broken)
winuae_2710b1 (2014.01.04) (broken)

[Toni Wilen]

Originally Posted by mc6809e

Ah, this explains why MOVEM.L + DIVS has a longer latency than a pair of MOVEM.L or DIVS.

MOVEM.L must sample very early and DIVS very late. Together they have potential 300+ cycle latency.

Which MOVEM.L? Used addressing mode can make a difference.

Originally Posted by amilo3438

Captain Fizz Meets The Blaster-Trons (Crack Intro) by Beyonders & Vision Factory
http://janeway.exotica.org.uk/release.php?id=73946

tested on A500 1.2/1.3KS OCS 0.5MB configuration

winuae_2700 (2013.12.05) (OK)
test winuae_2710b0 (2013.12.24 09:56) (OK)
test winuae_2710b0 (2013.12.24 21:07) (broken)
winuae_2710b1 (2014.01.04) (broken)

Real A500: exact same behavior as in 3.0.0. A1000 (real or emulation): works!

[mc6809e]

Originally Posted by Toni Wilen

Which MOVEM.L? Used addressing mode can make a difference.

MOVEM.L x.L, D0-D7/A0-A7 followed by DIVS x.L, Dn

[vagrant]

Fluffy Bears - Masturbation

I got a request to capture this old polish trackmo.. there is some gfx glitch during the opening erm.. 'animation'.

Glitch appears on standard quickstart 1.3, (all other memory/ks configs cause guru.)

The demo is not mentioned on any database, so this disk comes direct from source. Not sure if the disk itself is faulty..

[Toni Wilen]

Use ECS. Same OCS sprite problem as in some earlier posts.

[vagrant]

Gah, ECS agnus with 512k chip must have been only combination I didn't try.. working now, thanks toni

[amilo3438]

Deep Core - CD32/AGA (Crack Intro) by Epix -> http://janeway.exotica.org.uk/release.php?id=74075

A500 OCS/ECS (most common) ... not sure, but maybe it should be checked on a real Amiga also!

[Toni Wilen]

Originally Posted by amilo3438

Deep Core - CD32/AGA (Crack Intro) by Epix -> http://janeway.exotica.org.uk/release.php?id=74075

A500 OCS/ECS (most common) ... not sure, but maybe it should be checked on a real Amiga also!

Yeah, same weird behavior on real A500 too.