ScummVM/Falcon060 pre-release

[Eero Tamminen]

https://mikro.naprvyraz.sk/private/scum ... i-full.zip
https://mikro.naprvyraz.sk/private/scum ... i-lite.zip

Why these are not fastcall builds? Because fastcall is not ScummVM build default: https://github.com/mikrosk/scummvm/blob ... ase.sh#L10 ?

(Full version is 28MB instead of 26MB, and takes >9s to start, instead of <8s.)

[Eero Tamminen]

New version recognizes more games from ScummVM demos page!

M4 engine: https://wiki.scummvm.org/index.php?title=M4

Orion Burger: Seems to work!

Xeen engine: https://wiki.scummvm.org/index.php?title=Xeen

ScummVM invokes debugger, complain about missing "xeen.cc" (and in case of MM V, "dark.cc") file in these (non-interactive) demos:

• Might and Magic IV
• Might and Magic V
• Might and Magic: World of Xeen

Tinsel engine: https://wiki.scummvm.org/index.php?title=Tinsel

Discworld Noir: ScummVM crashes to NULL pointer write crash with 14+64 MB RAM:

Code: Select all

- 1. 0xe56176: Tinsel::decompressLZSS(Common::SeekableReadStream&, unsigned char*) -0x12bb1aa
- 2. 0x20f4cf4: Tinsel::Handle::LoadFile(Tinsel::MEMHANDLE*) +0x18c
- 3. 0x20f5362: Tinsel::Handle::LockMem(unsigned int) +0x238
- 4. 0x20b9c4c: Tinsel::StartNewScene(unsigned int, int) +0xc78
- 5. 0x20ce2ae: Tinsel::ChangeScene(bool) +0x7a
- 6. 0x20cefb8: Tinsel::TinselEngine::NextGameCycle() +0x14
- 7. 0x20cfbb6: Tinsel::TinselEngine::run() +0x36a
- 8. 0x101aa3a: runGame(Plugin const*, OSystem&, DetectedGame const&, void const*) +0x448

With enough memory (128MB TT-RAM), it exits slightly later:

Code: Select all

...
GEMDOS 0x3D Fopen("c:\scummvm\dw-noir-\cd1\english.scn", read-only) at PC=0x26780EC
GEMDOS 0x3E Fclose(65) at PC 0x26779FA
[00091945] WARNING: TODO: Complete scene loading logic for Noir!
[00091960] WARNING: TODO: Implement ADDCONV!
[00091975] WARNING: TODO: Implement ADDCONV!
GEMDOS 0x3D Fopen("c:\scummvm\dw-noir-\cd1\title.scn", read-only) at PC=0x26780EC
GEMDOS 0x3E Fclose(65) at PC 0x26779FA
[00140180] WARNING: TODO: Complete scene loading logic for Noir!
[00140200] WARNING: TODO: Implement SET3DTEXTURE(0x0A000014)!
GEMDOS 0x3D Fopen("c:\scummvm\dw-noir-\cd1\german.scn", read-only) at PC=0x26780EC
GEMDOS 0x3E Fclose(65) at PC 0x26779FA
[00156125] Background animation has finished!
[00156135] OSystem_Atari::quit()
[00156150] OSystem_Atari::~OSystem_Atari()
[00156160] ~AtariMixerManager()
[00156170] ~AtariGraphicsManager()
GEMDOS 0x3E Fclose(64) at PC 0x26779FA
GEMDOS 0x3E Fclose(2) at PC 0x26779FA
GEMDOS 0x3E Fclose(1) at PC 0x26779FA
GEMDOS 0x3E Fclose(0) at PC 0x26779FA
GEMDOS 0x4C Pterm(1) at PC 0x267AF52

Maybe it needs long file name support, as there are lot of warnings like this:

Code: Select all

[00038690] WARNING: FSDirectory::cacheDirectory: name clash when building cache, ignoring file 'lewton_u.h'!

[mikro]

Why these are not fastcall builds?

As mentioned elsewhere, it's not worth the effort (at least my effort). I don't use Thorsten's toolchain by default, therefore I don't test ScummVM with it and therefore it would be just another burden to maintain. I kept fastcall support in there as it costs me nothing.

The file size cut is nice but otherwise... I don't see a point, on CT60/accelerated TT the startup time is just a couple of seconds (and this is going to improve as soon as I finally implement the dynamic engine loading).

[Eero Tamminen]

While ScummVM startup showed largest measurable benefit, I think that's just because it's straight code. Games use timed events for many updates, which hides the performance differences to an extent, which can explain why measurable improvement in them is not more than few percents (at best). Main thing is that there were no regressions observed from fastcall usage, so it does seem to be both size and performance benefit.

(Single digit perf improvements are not large enough to be user noticeable alone, but coupled with other improvements they can...)

[mikro]

Let's wait until / when / whether I manage to finally start my delay(xx); optimisations which are indeed the biggest bottleneck. Of course, this doesn't apply to every single game, IIRC SCUMM doesn't have any of such hacks.

It's great we have such a large database of your measurements, we can immediately tell how much it helped.

[Eero Tamminen]

Tested few game demos also with the new lite version, on emulated TT, and checked that e.g. screen shake works fine.

Noticed rendering regression though:

wetlands-bug.png

with demo: https://downloads.scummvm.org/frs/demos ... mo2-en.zip

of Wetlands: https://wiki.scummvm.org/index.php?title=Wetlands


As to positive things, "Return to Zork" works surprisingly well, considering that it has (good looking) rendered camera runs, speech and MIDI music: https://wiki.scummvm.org/index.php?title=Return_to_Zork

While speech at start of demo intro was rather janky, one at the end of intro had no hitches in Hatari (with buffer=1024, channels=2, rate=12517, mode=single, vsync=true ScummVM options).

IMHO definitely worth checking, if you haven't tried it yet (136MB): https://downloads.scummvm.org/frs/demos ... mo2-en.zip

According to Wikipedia, game was rather successful despite competing against "Gabriel Knight: Sins of the Fathers" and "Day of the Tentacle".

[Eero Tamminen]

Took a look at another Hypno engine game, and cursor in "Spiderman: The Sinister Six" demo also does not look right:

grab0001.png

In "Wetlands" the issue was not with the (red circle) cursor though, but rendering of something else. Maybe Hypno engine has multiple rendering issues?

[mikro]

Eeeew. I knew I should have tested TT before building it. :-) Default settings, i.e. not direct rendering?

hypno had issues with the cursor because the game was setting custom cursor colours but the engine didn't support it (expected 16bpp backend): https://bugs.scummvm.org/ticket/14524 but that has been fixed so maybe there was something else introduced in the meantime.

[Eero Tamminen]

Ah, sorry, I forgot to mention Spiderman demo is 640x480, so it requires full version, and I ran it under Falcon emulation.

However, I tested now also "Wetlands" demo with full ScummVM version + under Falcon emulation, and the rendering issue happens there too. I.e. issue is specific neither to TT, nor lite version, but a generic one.

Oddly, the green targeting rectangle in "Wetlands" demo looks fine now and then (e.g. in the beginning). It's possible that the sheared rendering happens only when it's partly outside of the screen. I'm not sure what the sheared textured object is, maybe enemy ship that's also partly outside the screen?

[mikro]

Hypno engine (and all others requiring unaligned surfaces) has been fixed, it was a regression. However I think I found also a hidden bug which was there before - the unaligned flag was set only once. So it's possible that if you launched a game which needed the flag (e.g. Wetlands) and then another one which needed it as well (but wasn't in my 'whitelist'), it would work while it shouldn't (because it needs to be added to the 'whitelist'). So you may notice some glitches like this in games where it worked before.

https://mikro.naprvyraz.sk/private/scum ... i-full.zip
https://mikro.naprvyraz.sk/private/scum ... i-lite.zip

[Eero Tamminen]

Hypno engine (and all others requiring unaligned surfaces) has been fixed, it was a regression.

Verified, glitches in both of the Hypno engine games are gone with the new version!

[Teki]

yesterday i tested the new version and it seems already a bit faster but as eero said at
**** feeble files *** intro there are 3 or 4 big loadings when it begin to load its a littlebit more stuttering and less frames seen and the same at the end of loading in the middle of this animations you see much more frames then on the older ports but therefore at end and beginning of this animation less ... this flight to the asteroid field from shooting the asteroids you see nearly nothing but the game itself is playable then
after a while i tested little big advendture again and what should i say ... its at a playable speed now (very happy) its not super fast now also intro sound stuttering a littlebit but it runs at playable speed now ... more coming ...

[Teki]

orion burger : works well
backyard sport games: intros littlebit slow and sound stuttering, in game itself all seems ok
last dynasty : working but sound often stuttering (not always)
hades challenge: makes stripes in graphics
might and magic 4 and 5 : worlds of xeen. clouds of xeen working fine
little big big adventure 2 (NOT 1) : not fully supported, intro now works with 2fps (instead of 0.5 before), menu ios ok then new game and it crashes

[Eero Tamminen]

hades challenge: makes stripes in graphics

According to ScummVM docs, it has minor issues: https://www.scummvm.org/compatibility/D ... h:hadesch/
But should be completable with 3 minigames skipped: https://wiki.scummvm.org/index.php?title=Hadesch
However, there's no demo available/supported: https://www.scummvm.org/demos/

Could you provide a screenshot of the problem?

little big big adventure 2 (NOT 1) : not fully supported, intro now works with 2fps (instead of 0.5 before), menu ios ok then new game and it crashes

ScummVM pages do not list LBA2, only LBA:
https://www.scummvm.org/compatibility/
https://wiki.scummvm.org/index.php?title=Engines
?

[Teki]

hades challenge looks like this...
i testet single buffering and direct rendering fom netusbee and from ssd
060 with 90 mhz and 14+512mb ram

[Teki]

edited : the pictures begin from down
LBA2 have a lot of options, not sure but i think it was less options in the scumm ports before...
intro starts with around 1 fps then after 15 minutes of intro the main menu comes and runs well
clicked new game then another menu enter you name ... entered name and press return then intro starts again...
after intro it goes imidetly to desktop/mint ...
the loading seems same then the fps ... the led just blinks all second for a very short ms .... i think itcan load all the time
this loading seems while intro is slow, i think it can be much better
btw. you can also skip intro

[Eero Tamminen]

hades challenge looks like this...
i testet single buffering and direct rendering fom netusbee and from ssd

Every screen line is offset wrong (too much), not just some parts of the screen.

I wonder whether Hades Challenge does some buffering of its own?

[mikro]

While regretably still no progress / time on anything ScummVM-related, today I treated myself with the la(te)st member of the Ryzen 7 AM4 family:

time ./backends/platform/atari/build-release.sh before (2700):

real 3m3,188s
user 38m4,779s
sys 2m16,974s

after (5800XT):

real 1m57,064s
user 23m11,193s
sys 1m26,870s

Not bad for a CPU with the same amount of cores/threads and "only" upgraded from 3.2 GHz to 3.8 GHz.

[viking272]

Nice improvements! The rush to AM5 has left some good deals on AM4.

[mikro]

With the upcoming 2.9.0 release I thought it would be a good idea to provide some snapshot to get some testing / feedback. I have noticed one issue which I'm not able to replicate, palette was restored incorrectly on Falcon, when returning to a 640x480@16c. But I have an idea what's wrong.

https://mikro.naprvyraz.sk/private/scum ... i-lite.zip
https://mikro.naprvyraz.sk/private/scum ... i-full.zip

As usual, this is latest master with latest versions of everything (mintlib etc). As you might have noticed, the executable is a bit bigger, most likely due to more games supported but I haven't checked in more detail.

[marss]

Thanks

[Eero Tamminen]

With the upcoming 2.9.0 release I thought it would be a good idea to provide some snapshot to get some testing / feedback. I have noticed one issue which I'm not able to replicate, palette was restored incorrectly on Falcon, when returning to a 640x480@16c. But I have an idea what's wrong.

https://mikro.naprvyraz.sk/private/scum ... i-lite.zip
https://mikro.naprvyraz.sk/private/scum ... i-full.zip

I tried few games with the full version under 32Mhz Falcon emulation + EmuTOS v1.3. And there are indeed some regressions, see below.

Btw. Is this "fastcall" build or plain MiNTELF one?


3 Skulls of the Toltecs (demo)

Demo crash at startup to NULL pointer access when one moves mouse for the first time:

Code: Select all

- 1. 0x2568fee: AtariVidelManager::drawMaskedSprite(Graphics::Surface&, int, Graphics::Surface const&, Graphics::Surface const&, int, int, Common::Rect const&) +0x154f34a (sprite8_xloop +0x2)
- 2. 0x256d360: Cursor::draw(bool, bool) +0x1c0
- 3. 0x2567ee4: AtariGraphicsManager::updateScreenInternal(Graphics::Surface const&) +0xea
- 4. 0x25689fe: virtual thunk to ModularGraphicsBackend::updateScreen() +0x14326 (AtariGraphicsManager::updateScreen() +0x4c2)
- 5. 0x2219dac: Toltecs::ToltecsEngine::drawScreen() +0x8e
- 6. 0x221aaf4: Toltecs::ScriptInterpreter::execScriptFunction(unsigned short) -0x7864 (Toltecs::ToltecsEngine::updateScreen() +0xa4)
- 7. 0x2222bac: Toltecs::ScriptInterpreter::execOpcode(unsigned char) +0x7cc
- 8. 0x2222d6c: Toltecs::ScriptInterpreter::runScript() +0xaa
- 9. 0x221949e: Toltecs::ToltecsEngine::run() +0x4e6
- 10. 0x101f648: runGame(Plugin const*, OSystem&, DetectedGame const&, void const*) +0x448
...
WARN : Bus Error reading at address $0, PC=$2568fee addr_e3=2568fee op_e3=3018

Panic: Bus Error
fw=b008 ir=0000 ssr=0061
addr=00000000 sr=0318 pc=02568fee

D0-3: ffffffff 00000020 02eb000e 000002a0
D4-7: 000002a0 02ebffff 0000ffff 00000007
A0-3: 00000000 00000000 000284c0 02eb6548
A4-7: 00000000 0000011e 02c8f5e8 00007f70

Wetlands (demo)

Colors are all wrong during the interactive part of the demo:

wetlands-colors.png

Note: they reverted to correct ones when I quit the game in middle of that, with Ctrl-Q...


Eco Quest (demo)

Freezes at start constantly outputting these to Hatari console:

Code: Select all

[00289830] fillScreen: 1x1 0

It's possible all that printing makes it really slow, but it did not react to anything within few minutes.


Touche (demo)

The arrow cursor in the start of game has some garbage below it.

touche-arrow-bug.png

Note: crosshair cursor after that did not have the same issue.


AGI demo (for Apple IIgs)

Pressing "<Shift>-0" in AGI demo does not show next page of Sierra game demos.

It's also really slow. Instead of using STMIDI (global default option), it creates music itself.

"Police Quest" subdemo (3) seems to freeze at start.


Demos for Leasure Suit Larry in the land of lounge lizards and Magic Tales demo: Baba Yaga, Samurai, Imo

In first (Polish) demo, startup fails to "Error running game: reading data failed" dialog.

In second (Hebrew) demo, ScummVM quits during startup:

Code: Select all

[00876055] WARNING: Invalid section name: ~:\demo\splash!
[00876085] failed to find key '100' in section '~:\demo\splash' of book config!
[00876120] OSystem_Atari::quit()

=> Has support for non-English languages regressed?


Playtoons: Uncle Archibald (demo)

Does not find engine any more: "Couldn't identify game 'playtoons1' for the engine 'gob".


Demos for The Big Red Adventure and Kingdom: Far Reaches

DOS demos crash to NULL pointer access, Amiga demo for former invokes ScummVM debugger after "File 'fonts/sonya' not found!".

They are not fully supported yet though => I think these engine can be disabled for release, and enabled again after it.


Misc notes

There were weird message on start of RAMA demo:

Code: Select all

[01272640] Patching supernov.dat failed - resource type mismatch
[01272675] Patching teenagen.dat failed - resource type mismatch

There are few demos where Ctrl-Q does not work to return to ScummVM menu:

• Discworld
• Discworld II

And I saw these kind of warnings during couple of Sierra demos:

Code: Select all

[00364020] WARNING: sci/christmas1988 calls update() from timer!
[00225825] WARNING: sci/laurabow calls update() from timer!

[mikro]

Btw. Is this "fastcall" build or plain MiNTELF one?

Plain mintelf, if I were building the fastcall flavour, my first thought on this report would be "it must be certainly fastcall's fault". :-)

3 Skulls of the Toltecs (demo)

Demo crash at startup to NULL pointer access when one moves mouse for the first time:

Confirmed. Please provide the exact versions (ideally links) so I don't have to download multiple files: this looks like the German version as English one ends up in the debugger with some file missing.

Wetlands (demo)

Colors are all wrong during the interactive part of the demo:

Not sure which demo but the first one (wetlands-dos-demo1-en.zip) crashes hard, so that looks like even a bigger problem.

Eco Quest (demo)

Freezes at start constantly outputting these to Hatari console:

Confirmed. I hate that engine, it's the same one which was requesting of adding thousands of dirty rectangles during that bubble intro. This is similar, there hasn't been any change on my side. Perhaps it is the text output... I guess I'll remove it & add some special handling for retarded cases like this.

Touche (demo)

The arrow cursor in the start of game has some garbage below it.
touche-arrow-bug.png

Note: crosshair cursor after that did not have the same issue.

touche-dos-demo-en.zip crashes hard on me as soon as I move the cursor. So again smells like different version. Anyway => trouble.

AGI demo (for Apple IIgs)

Pressing "<Shift>-0" in AGI demo does not show next page of Sierra game demos.

Same on Linux.

It's also really slow. Instead of using STMIDI (global default option), it creates music itself.

Which one exactly? I ran King's Quest, sounded OK in Aranym.

"Police Quest" subdemo (3) seems to freeze at start.

Confirmed. I vaguely remember something with Police Quest but nothing concrete.

Demos for Leasure Suit Larry in the land of lounge lizards and Magic Tales demo: Baba Yaga, Samurai, Imo

In first (Polish) demo, startup fails to "Error running game: reading data failed" dialog.

Same on PC.

In second (Hebrew) demo, ScummVM quits during startup:

Code: Select all

[00876055] WARNING: Invalid section name: ~:\demo\splash!
[00876085] failed to find key '100' in section '~:\demo\splash' of book config!
[00876120] OSystem_Atari::quit()

Huh? Hebrew what demo?

The other t wo (Yaga, Samurai) seem to be OK but again, I can't be sure of the version. I tested the Mac one.

What is "lmo" ?

Playtoons: Uncle Archibald (demo)

Does not find engine any more: "Couldn't identify game 'playtoons1' for the engine 'gob".

Tried the two demos, all good in Aranym. Most likely playtoons1 = 10 characters.

Demos for The Big Red Adventure and Kingdom: Far Reaches

DOS demos crash to NULL pointer access, Amiga demo for former invokes ScummVM debugger after "File 'fonts/sonya' not found!".

They are not fully supported yet though => I think these engine can be disabled for release, and enabled again after it.

Same on PC, I wonder why they are enabled by default.

There were weird message on start of RAMA demo:

Code: Select all

[01272640] Patching supernov.dat failed - resource type mismatch
[01272675] Patching teenagen.dat failed - resource type mismatch

Both OK on Aranym.

There are few demos where Ctrl-Q does not work to return to ScummVM menu:

• Discworld
• Discworld II

Yeah, same on PC.

And I saw these kind of warnings during couple of Sierra demos:

Code: Select all

[00364020] WARNING: sci/christmas1988 calls update() from timer!
[00225825] WARNING: sci/laurabow calls update() from timer!

Yes, that's an indication that a workaround is in place and it shouldn't be happening but oh well.

[mikro]

AGI demo (for Apple IIgs)

Pressing "<Shift>-0" in AGI demo does not show next page of Sierra game demos.

It's also really slow. Instead of using STMIDI (global default option), it creates music itself.

"Police Quest" subdemo (3) seems to freeze at start.

Yup, I was right, we did discuss this before. There's also answer to your SHIFT+0 issue:

AGI demo (AGI engine): https://wiki.scummvm.org/index.php?title=AGI_Demo

Issues:

• Switching to 2nd set of demos (in title screen) using Shift-0 key, does not work
• Police Guest demo freezes at start [1]

Should have read the ScummWM wiki closer. One needs to use F10 (instead of the Shift-0 stated in the demo itself) to switch to 2nd games list.

All other 4 demos under "AGI demo" work fine, only Police Quest one freezes (so that e.g. Ctrl-Q does not quit ScummVM).

I seems that I let the freeze go in favour of full PQ game (which works) but yeah, I should take a look. The other issues are either solved or I need to think about a proper solution, so next build is coming soon.

[Eero Tamminen]

Hm. There are now more demo variants on ScummVM demos page...

3 Skulls of the Toltecs (demo)

Demo crash at startup to NULL pointer access when one moves mouse for the first time:

Confirmed. Please provide the exact versions (ideally links) so I don't have to download multiple files: this looks like the German version as English one ends up in the debugger with some file missing.

I was using English version: https://downloads.scummvm.org/frs/demos ... s-demo.zip

md5sum: 587904bfbe486bcd9faeb73ae1105590

Wetlands (demo)

Colors are all wrong during the interactive part of the demo:

Not sure which demo but the first one (wetlands-dos-demo1-en.zip) crashes hard, so that looks like even a bigger problem.

Second one: https://downloads.scummvm.org/frs/demos ... mo2-en.zip

Eco Quest (demo)

Freezes at start constantly outputting these to Hatari console:

Confirmed. I hate that engine, it's the same one which was requesting of adding thousands of dirty rectangles during that bubble intro. This is similar, there hasn't been any change on my side. Perhaps it is the text output... I guess I'll remove it & add some special handling for retarded cases like this.

I remember that. Now it was outputting those logs as fast as Linux console could print them, so printing is definitely slowing it down.

Touche (demo)

The arrow cursor in the start of game has some garbage below it.
touche-arrow-bug.png

Note: crosshair cursor after that did not have the same issue.

touche-dos-demo-en.zip crashes hard on me as soon as I move the cursor. So again smells like different version. Anyway => trouble.

That's the one. I guess you're using Aranym?

Note that getting backtraces under Hatari is nowadays ease:

• Use "--debug-except autostart,bus" to invoke debugger on application bus errors
• Enable profiling to get backtrace from it when debugger is invoked

("autostart" option enabled exception catching only when program is autostarted from GEMDOS HD, so that TOS boot-time HW detection does not invoke debugger.)

I.e. you need to start ScummVM under Hatari:

Code: Select all

hatari --trace os_base --debug-except autostart,bus \
  --machine falcon --dsp off  --fpu 68882 --cpuclock 32 \
  -s 14 --addr24 off --ttram 64 --midi on ./scummvm.ttp

And once ScummVM is up, invoke debugger (which loads program debug symbols) with AltGr+SysRQ, type "profile on" and "c" (to continue).

Then click on the game that bus errors, and you'll automatically get a backtrace when that happens, in Hatari console.

AGI demo (for Apple IIgs)
...

It's also really slow. Instead of using STMIDI (global default option), it creates music itself.

Which one exactly? I ran King's Quest, sounded OK in Aranym.

KQ. I doubt ScummVM synthetizing sounds instead of using (ST) MIDI would be a problem for these old games under Aranym. With Hatari the performance is really bad...

In second (Hebrew) demo, ScummVM quits during startup:

Code: Select all

[00876055] WARNING: Invalid section name: ~:\demo\splash!
[00876085] failed to find key '100' in section '~:\demo\splash' of book config!
[00876120] OSystem_Atari::quit()

Huh? Hebrew what demo?

"Magic Tales demo: Baba Yaga, Samurai, Imo" demo.

ScummVM demos web-page calls it just "Magic Tales demo", and has only Hebrew version of it:
https://www.scummvm.org/f/rs/demos/comp ... mac-he.zip

There were weird message on start of RAMA demo:

Code: Select all

[01272640] Patching supernov.dat failed - resource type mismatch
[01272675] Patching teenagen.dat failed - resource type mismatch

Both OK on Aranym.

RAMA game demo does indeed work OK, but why SCI engine would try to patch TeenAgent or SuperNova engine .dat files on startup??