@Dominus
I have exactly the same issues with the same games when I use SDL1 build. So it probably should be DOSBox's endian issues. Maybe Jmarsh was in the impression that I have everywhere such issues, but nope, just "hugi" diskmag and those 3 games. And the common from all of them that they indeed use more colors than usual 256.

If you have some free minutes, can you plz check at least that "HUGI" mag on your PPC hardware, with SDL1 or SDL2 does not matter: https://files.scene.org/get:nl-ftp/mags/hugi/hugi17.zip

Thanks in advance!

This thread is for discussing the dynamic recompiler. If you get the same issue with core=normal it shouldn't be in this thread.

@jmarsh
Right, moved all there: Wrong colors on BIG ENDIAN in games/apps which use 15,16 or 32bit video modes

Hi,

First of all, big thx for writing PPC dynarec. I've made a port for MorphOS operating system and it works pretty well. (only //#define USE_SDA_BASE was needed). MorphOS runs on PowerMacs mostly and i'm testing DosBox on my PMac G5 2.3 PCIe. There is one strange thing regarding speed. I've checked Quake1 demo (timedemo demo1) and for some runs i got 24 fps (320x200), cycles set to max or auto and for some runs i got 16fps. It is totally random. Other users had similar results.

Check if the same thing occurs with the frameskip set to 1 or 2.

Could also be the host OS doing something odd with cpu scheduling / power saving / modes. (Although I have nothing to suggest in terms how to check or force it into the highest-performance mode).

jmarsh wrote on 2020-01-26, 02:27:

Check if the same thing occurs with the frameskip set to 1 or 2.

Hi, so i got only ~16 fps values with frameskip 1, 2, 3.

MorphOS has DFS2 turned on on PowerBooks and ImacsG5 so on PowerMacG5 it always runs full speed.

btw. What about playing with cache size defines, is it worth playing with them?

@rzookol
I not sure if we can rely on those benchmarks in end. For example, for now, I set "cycles=100000", and in quake1 have 53fps. Same for PCPBench: more cycles I set, the more FPS I have _writen_, but visually all the same.

@Jmarsh
I think there indeed something wrong with "fade-in/fadeouts" happens when JIT is used as ClassicHasClass found before. I found some games which with dynamic core enabled suffer from slow fade in/fade out, but when using "normal" it's more or less ok:

1). Mortal Kombat 3: issue happens not all the time, but most of the time when finish round and things should fade down. Sometimes it abnormally slow.
2). Fallout: all fade in/fade-outs in the whole game is slow. Cpu usage has not risen much, it is like something else make be slow.

With Fallout, it was very hard to check with core=normal, as without jit it pains on loading, but with Mortal Kombat 3 it was easy reproducible (because without JIT on 2ghz CPU its more or less fast as well). And so, when i didn't use JIT after round end, i always have fast fade out. When i use just, then sometime it ok, but most of time its like with micros-pauses. Like pause, then a few frames ok, then again pause. And that pause can be as fast as well as long enough. For example in the fallout, its probably a 1-second pause, and happens 4-5 pauses during fade in/fade out with Jit. With mortal kombat 3 it the same, but sometimes it can even be stuck and nothing happens. In very rare cases I even need to restart DOSBox.

And I can see that on a moment of those slow fade out/fade-ins, CPU loading are 100% (while in whole game about 50-60 while you play).

If you get pausing you have cycles set too high.

I use cycles=auto, and while in-game it never reaches 100% of CPU loading, but when fade-out happens at the end of a round, sometimes it reaches 100% CPU and hits those pauses. Only with fadings.

Hi, guys! Sorry for the thread necromancy, but I was browsing around the SVN and I didn't see this patch merged. Are there any plans for upstreaming it, eventually? I'd love to give this a spin on my powerpc and ppc64 Debian machines. 😉

It's missing all the required autotools/configure changes because I don't have a powerpc machine that uses them.

Is anybody else working on it? I have the machines, but I'm not familiar with autotools scripts (aside from compiling existing projects from source) or SVN…

jmarsh wrote on 2020-05-08, 14:38:

It's missing all the required autotools/configure changes because I don't have a powerpc machine that uses them.

Have you tried requesting access to the GCC compile farm? They have BE POWER machines available for free software developers.

Hi,

Did anyone merge 3dfx emulation to PowerPC version? I did it on MorphOS and it seems the emulated card is not working correctly.

@Jmarsh

Hello mate,
Don't know if you are still working on the PowerPC optimization, but I've opened a ticket on dosbox-staging site asking some questions to the team and it would be nice to have your point of view 😀

https://github.com/dosbox-staging/dosbox-staging/issues/689

Was a ticket opened on the official dosbox site?
What games?
What kind of performance are you looking for, hopefully something realistic?

DosFreak wrote on 2020-10-28, 18:27:

Was a ticket opened on the official dosbox site?
What games?
What kind of performance are you looking for, hopefully something realistic?

It wasn't a question related to a single specific game, it was merely a possible enanched request asking Jmarsh if it could be possible to improve a bit more (even just a little bit more) the performance of the dynamic core for PowerPC.. this may helps a lot low specs PPC based hardware

To do something there needs to be a reason.
Are you looking for 1fps, 2fps, 60fps?
How much faster does it need to be for what you want to do for which you are being vague for some reason?