Denise bugfixes and improvements

[Toni Wilen]

Originally Posted by PiCiJi

all IPL tests throw errors: e.g. AND.B: memory byte write: address ffff8197, expected 14 but got a0
What am I missing here?

Possibly wrong KS ROM? It must match ROM selected in cputestgen.ini. Or perhaps 32-bit address space was enabled in UAE? Or you can disable "high mem" tests by commenting out test_high_memory_xxx lines. ("high mem" = test that address space wrap around works and xxx.W addressing works when address is "negative"). test_low_memory_xx must be disabled if cycle counting or testing IPL. Fast RAM is only supported in this case.

But any test, not just IPL test would cause same errors.

edit: can not test it in my emu at the moment, because serial emulation is not complete enough

Yeah, it needs cycle-accurate serial port emulation. Originally I was planning to use external hardware (some arduino or PIC board) but it was annoying and later serial port emulation was improved to make this possible.

Are there some defaults to prevent tmem, hmem ... because of the limited ADF size

No, they are always needed. But you can reduce test mem space (test_memory_size) to 0x20000. 0x10000 possibly is too small but 0x20000 should work.

Using HD floppies should make it more manageable.

[amilo3438]

ECS only!
-----------
D (1993)(Essence).adf ... Glitch on the left window part during the rotation of the teddy bear!
(in "The Zone")(A500 1.3 ECS 512mb Chip + 512 Slow)

Note: This happens on OCS, but it shouldn't on ECS! (checked on winuae)
See also on: [ Show youtube player ] (@ 1:14)


A1000 only!
--------------
VF&Bey-CrCaptainFizz.adf ... Glitches on Denise A1000! (fine on winuae A1000)
(in "The Zone")(A1000 1.3 512mb Chip)


Glitches on a vector objects!
---------------------------------
Andromeda-Surrealism.adf
ANDROMEDA-Mindriot.adf
(in "The Zone")(A500 1.3 ECS 512mb Chip + 512 Slow)


Invalid refresh rates!
-------------------------
BRS-PrimusDemo.adf
Mexx-Intro92x.adf
Mexx-Plasma.adf
(in "The Zone")(A500 1.3 ECS 512mb Chip + 512 Slow)

[PiCiJi]

Originally Posted by Toni Wilen

Possibly wrong KS ROM? It must match ROM selected in cputestgen.ini. Or perhaps 32-bit address space was enabled in UAE? Or you can disable "high mem" tests by commenting out test_high_memory_xxx lines. ("high mem" = test that address space wrap around works and xxx.W addressing works when address is "negative"). test_low_memory_xx must be disabled if cycle counting or testing IPL. Fast RAM is only supported in this case.

thanks. yes forgot to update KS ROM path and generated IPL separated from others.
Now it works. test cases: DIVS, DIVU and AND passes in WinUAE.
will retest on my real Amiga and then try to complete serial emulation. I am curious if i get them working in Denise

@amilo3438 thanks for testing.

[amilo3438]

Glitches on a whole screen or partly!
-----------------------------------------
Quarterline-TrRickDangerous.adf
(in "The Zone")(A500 1.3 ECS 512mb Chip + 512 Slow)

[Toni Wilen]

Originally Posted by PiCiJi

complete serial emulation

cputester uses only static serial period so when SERDAT is written to, you only need to start a delay and then set INTREQ RBF bit.

If I calculated it correctly, it is (in CCKs): 1 + (number of bits to send) * SERPER + (SERPER - 1) / 2 + 3 + 9.

Nothing "normal" needs cycle-accurate serial emulation.

[PiCiJi]

nightly
@amilo3438 no need to retest. that are mostly Blitter/Copper bugs. I am almost there.

- improved serial emulation and add option for loopback tx<>rx
- fixed HD floppy emulation. kick 2.0 is minimum to handle HD disks.
- pass all cputester IPL tests, except: eorsr.w, mv2sr.w, orsr.w (didn't work in WinUAE either )
- pass all vAmigaTS IPL tests (more like WIPL tests)
- don't handle "possible" IPL delay in dependence of IPL change, e.g. 1->3 or 3->1

for stabilizing IPL, Paula has 2 CPU cycles (from 0.5 CCK to 0.5 of next CCK) Perhaps the 3/4 CCK delay when changing certain bits can never affect whether an interrupt is served after this or the next opcode.
IPL is pipelined 2 CCK, so Paula should have enough time to schedule the change.
just such a thought. Maybe somewhere else there is a mistake why IPL tests work for me even without this delay.

I didn't get IPL MOVE.B and MOVE.L on DD disk. no problem, works good on HD disk

couldn't confirm the cputester tests on real Amiga
my real Amiga: A500 Rev6A 0.5 chip, 0.5 slow (switchable), M-Tec 68020i turbo with 4 MB fast ram (switchable), Gotek floppy replacement, harddisk with 2 MB fastram
set serial jumper TX<>RX
I can't disable my turbo card without disabling fastram too. ok rebuild some IPL tests for 2 MB fastram (confirmed in emulation) and tried my harddisk fastram.

There are always cycle issues in the tests
1. could try to remove my turbo card completely. (maybe switching off is not enough ... unlikely)
(need a second 68k first. Last time I cursed violently to put on the 68k. I'll never take it down again from card)
2. could try to remove my slowram (maybe switching off is not enough ... unlikely)
3. my harddisk is showing some boot screen before the floppy takes over (F10)
(problem for these tests? should i format the hd ... only need the fastram)

[amilo3438]

Originally Posted by PiCiJi

@amilo3438 no need to retest. that are mostly Blitter/Copper bugs. I am almost there.

That's great news. Of course, I will wait to retest the demos again until you're finish with a current phase!

Originally Posted by PiCiJi

- pass all cputester IPL tests, except: eorsr.w, mv2sr.w, orsr.w (didn't work in WinUAE either )

I guess it must be correct then! (if it is same as on winuae)

[PiCiJi]

new nightly:

I am now also testing Denise on a new computer with Windows 11. Wasapi Audio driver causes problems.
At first I thought it was Windows 11. In fact, it is due to a changed audio hardware on the mainboard. This is fixed.

some Blitter / Copper fixes.
OLD_BULLS-Fugazi: Bug 3/3 fixed (black lines in rotating animation. there is a single glitch after a few seconds ... confirmed on real hardware)

Maybe other problematic demos will work.

[amilo3438]

Originally Posted by PiCiJi

new nightly:

Maybe other problematic demos will work.

New nightly seems to have also fixed (according to post #20):

03) Glitches on a vector objects! (all except one)

Except: FLD-Demodisk050.adf ... A: Cementery / Possessed

Check on winuae problematic demos on link (last one): https://eab.abime.net/showpost.php?p...&postcount=989

and also have fixed:

05) Problem at horizontal and diagonal bars mixes! (RamJam - Cooper Slave)

[PiCiJi]

some infos about Mission Elevator Sound bug

When fixing errors using test programs, I noticed that the audio interrupt delay was one cycle too long.
I increased this by 2 a few weeks ago due to Mission Elevator Sound bug. Although the increase of 1 cycle has improved it, but not as good as 2 cycles.
That was incorrect. The game ran better than the original.
I tested on my real Amiga.
In 2 out of 3 attempts, this bug occurred. Once after 5 minutes and once after more than 10 minutes and once not at all within about 20 minutes.
Something like this behaves in emulation.
So the slight deterioration is correct in my opinion.

[amilo3438]

Originally Posted by PiCiJi

some infos about Mission Elevator Sound bug

Look what is in Docs\winuaechangelog.txt file written (quote):

- Audio interrupts are delayed by 2 cycles, logic analyzer confirmed. Emulated only in cycle-exact mode. (Mission Elevator)

[PiCiJi]

Originally Posted by amilo3438

Look what is in Docs\winuaechangelog.txt file written (quote):

- Audio interrupts are delayed by 2 cycles, logic analyzer confirmed. Emulated only in cycle-exact mode. (Mission Elevator)

yes, I previously had 3 cycles delay.

[amilo3438]

Originally Posted by PiCiJi

some infos about Mission Elevator Sound bug

I tested on my real Amiga.
In 2 out of 3 attempts, this bug occurred. Once after 5 minutes and once after more than 10 minutes and once not at all within about 20 minutes.
Something like this behaves in emulation.
So the slight deterioration is correct in my opinion.

Can confirm the bug in winuae5100 beta0 (2023.07.02) occurred after cca 5 min and in denise nightly build_6367f10 (2023.07.03) occurred after 10min. (so it is random)
(did only one test on each emu) EDIT: Emu config was A500 1.2 OCS 512 Chip ram!

[PiCiJi]

Originally Posted by amilo3438

Can confirm the bug in winuae5100 beta0 (2023.07.02) occurred after cca 5 min and in denise nightly build_6367f10 (2023.07.03) occurred after 10min. (so it is random)
(did only one test on each emu)

Yes, it's purely coincidental. (not quite) However, that doesn't mean you can't compare it. To make the randomness more comparable, you would have to play several times (5-10) for 20 or 30 minutes on the original and repeat the whole thing in the emulation. So you have an approximate comparison.

I didn't feel like it anymore after 3 attempts. But I was basically only interested in whether it happened on the original or not. And it happens on the original in 2 out of 3 attempts. It was clear to me that the programmer of the game could hardly have overlooked. Maybe it doesn't happen on an Agnus OCS. (unlikely) I can't test that at the moment.

[PiCiJi]

new nightly:
fix: MWB-DemoCollectionDisk16.adf ... MF-New Intro
2 bugs found

1. (upper part) demo changes Blitter channel usage while running. (fixed)
2. (lower part OCS) demo sets SHRES and Agnus OCS accidentally switched to SHRES BPL fetching (fixed)

(lower part ECS) demo from 89 use SHRES with BPUx = 3.

This seems to be illegal and there is no BPL fetching taking place. On a real Agnus ECS + OCS Denise, the lower part is not visible. (fixed)
In WinUAE 2023.05.31, there is no difference between ECS and OCS Agnus. This seems like a bug to me.

[amilo3438]

Originally Posted by PiCiJi

new nightly:
fix: MWB-DemoCollectionDisk16.adf ... MF-New Intro

It also fixed 2 more problems:

12) Glitches on a scrolling text!
FLD-Demodisk056.adf ... F: Molecules / IT (at 1st it seems ok, but then errors!)

14) Stops after a while!
FLD-Demodisk022.adf ... D: Party-Intro Defcon 1

[amilo3438]

Originally Posted by PiCiJi

MF-New Intro
2 bugs found

1. (upper part) demo changes Blitter channel usage while running. (fixed)
2. (lower part OCS) demo sets SHRES and Agnus OCS accidentally switched to SHRES BPL fetching (fixed)

(lower part ECS) demo from 89 use SHRES with BPUx = 3.

This seems to be illegal and there is no BPL fetching taking place. On a real Agnus ECS + OCS Denise, the lower part is not visible. (fixed)
In WinUAE 2023.05.31, there is no difference between ECS and OCS Agnus. This seems like a bug to me.

I checked in Denise:
In OCS "MAGNETIC FILEDS" text is visiable and in ECS is not! (as you confirmed on a real ECS + OCS Denise machine)

Checked in the latest WinUAE:
In OCS and ECS the "MAGNETIC FILEDS" is visiable !? (hope Toni read this) http://janeway.exotica.org.uk/release.php?id=21844

[PiCiJi]

Originally Posted by amilo3438

It also fixed 2 more problems:

nice

nightly:
BplCon1 changes are not only taken into account during the next processing of Bpl1Dat, but directly after a cycle delay.

fix Anathema-NeSouthEast65.adf ... Turnips intro

[amilo3438]

Originally Posted by PiCiJi

nice

nightly:
BplCon1 changes are not only taken into account during the next processing of Bpl1Dat, but directly after a cycle delay.

fix Anathema-NeSouthEast65.adf ... Turnips intro

And this fixed also 2 another similar gfx problems:

RAZOR1911-WeShaveAss.adf ... some problem with scroller text on the end!
VD-AbsoluteInebriation-A.adf ... some gfx problems on part after the title screen

[Toni Wilen]

Originally Posted by PiCiJi

This seems to be illegal and there is no BPL fetching taking place. On a real Agnus ECS + OCS Denise, the lower part is not visible. (fixed)
In WinUAE 2023.05.31, there is no difference between ECS and OCS Agnus. This seems like a bug to me.

Yeah, for some reason Agnus BPLCON0 masked SHRES bit if no ECS Denise, probably was forgotten when Agnus and Denise BPLCON0 was separated long time ago.

Display is also completely blank if ECS Denise because it uses invalid HDIW start value (Denise hcounter starts from 2)