Atari-Forum

If you could do a PCB mounted version, that would be a smashing success because there's literally none of them available. :)

I think you got it with the Sysconf() call. When I get back from Forever party, I'll provide a version with some TOS defaults, I'm really curious about dmalloc performance.

Hmm it's true I tried it under MiNT only... I'll try later with TOS

There was one great idea a few years back - to reverse engineer the SLM protocol over ACSI and print using cups via CosmosEx. That would be really cool to have.

Not only I can't thank you enough for this analysis but I have one more request for you: as you have profiled the SCI games and you have a preconfigured setup... could you please try this binary and do *exactly the same* profiling towards malloc / free ? This is a 2.6.1 with Doug Lea's allocator (us...

Leisure Suit Larry - Love for Sail! Unlike LSL 2, LSL 7 is definitely too slow for emulated 32Mhz falcon. SCI VM is bound by alloc/free speed, and it does way too many get method calls: I've quickly glanced over this - not good, not good at all. There's a lot of list operations at least on 5 differ...

6. Get list of top symbols taking most time [1] in the profile: $ hatari_profile.py -r scummvm-symbols.sym -stp game-profile.txt Hmm, I tried this but didn't get very far. Latest Hatari master spits on me: (hundreds of warnings...) : WARNING: - 0x2ee8fe8 is also address for symbol '_init_clock' WAR...

Yeah, that was my immediate thought, i.e. it doesn't matter that malloc() is called not *so* much but how expensive it is (and it is expensive). So if there is a few NEWs / DELETEs every frame, it can easily contribute because it is fragmenting memory as hell. However I'm not so convinced that there...

Wow, it's a miracle that BS2 is playable even on CT60 with such heavy video decoding/depacking.

Ok, I have noted to investigate where the mallocs come from in the SCI engine. Thanks!

Frankly, I stopped watching/reading any Atari history "amazing facts" videos/articles a long time ago. People are just recycling the same stuff over and over again, citing websites which cite other websites and so on. The only reliable source of truth is Marty's & Curt's book and maybe...

Amiga "Dreamweb" crashed ScummVm to NULL pointer access. Amiga "The Big Red Adventure" demo complained "ERROR: Files 'fonts/sonya' not found!" (it worked on my (older) Linux ScummVm version). The latter shows me a warning in 2.8 that it is not completely supported and ...

"I have no mouth and I must scream" demo's problem is obvious - its engine emits fullscreen updates all the time. No help there except my vague idea to somehow detect / hardcode the engines which emit fullscreen updates, i.e. they don't need the intermediate buffer on Atari backend's side.

That implies Atlantis running only at ~10 FPS, although most of its time goes to delayMillis() / usleep() waiting. I.e. it's not limited by CPU perf, something is just forcing it to go slower. Btw are you sure about this? I've just tried to empirically (dis)prove it, i.e. I'm counting millis betwee...

Eero Tamminen wrote: ↑Sun Mar 12, 2023 12:41 amPS. Most annoying thing about the OPL CPU usage was that, there's actually no sound in the the full SOMI version (I haven't ripped the music from CD, and CD version apparently does not support MIDI music).

Luckily, the talkie version contains both CD and MIDI music so one can switch between them.

Thanks Eero, useful data! Maybe when you see the MAME OPL in your stats so high, you can take a note and then try with "No music" in audio options? That one handles exactly this, the MIDI music emulation (and not CD music and/or speech). It's nice to see that the rectangle-based C2P does i...

I think granularity shouldn't be a problem -- clock() (and therefore usleep()) still uses the 200 Hz timer internally. Only FreeMiNT makes it (perhaps?) better in case of usleep() but it's not worth it IMHO. usleep's main usage is here: https://github.com/scummvm/scummvm/blob/master/gui/gui-manager....

Accidentally, yesterday I created a list of things worth looking into from this thread: https://github.com/mikrosk/scummvm/wiki/Things-to-check So yes, I do have that ManagedSurface::setPalette() there as well as ScummEngine::testGfxUsageBit(). From your graphs I'd say that getting rid of clock() / ...

Out of curiosity, which cookie jar item ScummVm was querying thousands of times per second? I've taken a look. It's not cookies after all but https://github.com/scummvm/scummvm/blob/master/backends/platform/atari/osystem_atari.cpp#L157 So... is it really worth replacing with custom code? I mean, th...

You can always use a browser extension like "DownThemAll" or similar for mass-downloading. :)

Eero: not at all, v7 is for example The Dig or Full Throttle: https://wiki.scummvm.org/index.php/SCUMM/Versions Hmm, I thought I had a check for SuperVidel in Surface::copyFrom() but nope, it's only in Surface::create()... weird. I guess I'll have to take a look, there isn't *that* much platform-spe...

By default, you can't use blanks between operands. It is to be able to detect comments which, by default, can be separated by blanks.

You can change it in vasm by some option but I wouldn't recommend that.

Ah, right, that must be the cookie jar then. Okay, that's fixed in 2.8-git.

Btw that low latency audio mode is related only to scumm-v7 and v8 games and it is a 2.7+ feature.

Looking at scummvm documentation: https://docs.scummvm.org/en/v2.6.1/advanced_topics/understand_audio.html#audio-buffer-size At least "audio_buffer_size" config option value must be 256, 512, 1024, 2048, 4096, 8192, 16384, or 32768. Btw. mikro, why your scummvm version uses a differently ...

FLAC provides about 50% savings as well (lossless).

However it is strange that ADPCM WAV did not work, I can see it is supported by ScummVM. Did you try it on Windows/Linux version of ScummVM, too?

It must be optional, as TT support is still planned. Same as real mute, it's almost funny that ScummVM simply can't STFU and eats cycles even in complete silence.