Uae4arm: Amiga emulator with JIT and DispmanX

MiDWaN I've a small question.

Is your SDL2 port running. Or is it a type of concept. What is the current state?

Also are you plannung to bump your sources to the latest release of TomB?

Regards

Amiga reset via : LAMIGA RAMIGA + CTRL not working. When entry in GUI then RIGHT CONTROL reset Amiga and ESC key. When Amiga run is impossible to reset Amiga from keyboard ( PC keyboard or Amiga 1200 keyboard via Keyrah V2)

Can You check it or where I can find solution to fix it ? I checked sources but is different from other old sources. Regards Paul

After some debugging, it works under X11 but not under console.
New code use keystate to get a table of SDLK key but the left windows key get never indicated as pressed in console as neither SDLK_LMETA nor SDLK_LSUPER whereas I get both indicated as pressed under X11 for left windows key.
So i need to find a fix for console mode.
EDIT: left windows is seen as ... left alt in console mode
EDIT2: just pushed a workaround for this.

I merged the latest version from pandora in github.
The audio/graphics dis synchronization disappears but I feel like audio get a little bit worst
EDIT: looks like it's link to the configuration I was loading which has cpu set to more compatible at least it's not activated by default. Let me know if you feel the same...

I merged the latest version from pandora in github.
The audio/graphics dis synchronization disappears but I feel like audio get a little bit worst
EDIT: looks like it's link to the configuration I was loading which has cpu set to more compatible at least it's not activated by default. Let me know if you feel the same...

Hi Chips, with your last build i have the following issues:
Audio plays very slowly and out of synch (tested with Starstruck by TBL and Switchback by Rebels)
The screen in pal mode is not entirely filled (bottom quarter is black), no problem with Workbench using Picasso mode (800x600 or 1024x768)

Tested on Raspberry Pi 3, 68040/JIT/Internal FPU/WB 3.9 (problems persist with other cpu/jit settings too)

Hi Chips, I ran this latest build of UAE4ARM and custom joystick controls seem to not be working any more.

Previously I was able to assign a button on the joypad to translate to pushing up on the joypad (which is mostly used for jumping on amiga games)

I merged the latest version from pandora in github.
The audio/graphics dis synchronization disappears but I feel like audio get a little bit worst
EDIT: looks like it's link to the configuration I was loading which has cpu set to more compatible at least it's not activated by default. Let me know if you feel the same...

Hi Chips, with your last build i have the following issues:
Audio plays very slowly and out of synch (tested with Starstruck by TBL and Switchback by Rebels)
The screen in pal mode is not entirely filled (bottom quarter is black), no problem with Workbench using Picasso mode (800x600 or 1024x768)

I know, since two pandora delivery startruck is worst than what is was... this problem is more on internal emulation core which is more known by TomB than myself (indeed I never try to understand emulation core)...

Hi Chips, I ran this latest build of UAE4ARM and custom joystick controls seem to not be working any more.

Previously I was able to assign a button on the joypad to translate to pushing up on the joypad (which is mostly used for jumping on amiga games)

How did you enable this ? you activate custom control and set up for <B> for ex ? which version was working for this ?

How did you enable this ? you activate custom control and set up for <B> for ex ? which version was working for this ?

My Pi isn't with me at the moment, but from memory:

I ticked the option for Custom control, then mapped B button to act as Joypad up.

The one thing I noticed different about this version is the name of my controller is mentioned in the joystick menu.

Sorry I don't know what the version number was for the working version with custom controls was. It was within the last 1-2 months via an updated RetroPie.

I also tested the latest build on my rpi 3.
I've altered the makefile to optimize the built for the rpi3-cpu.
Currently I've just tested sysspeed under Workbench an so I get 7 till 8 times the speed of an A4000. That's pretty good.
I also tested Starstruck. The Video ist a little bit stuttering. But I need to switch to a 50hz Video mode.
Also is it not possible to reset the Emulator using Crl-Amiga-Amiga and the drive leds on Keyboard are not working.

I just installed this emulator again after a long time and noticed the performance has gone right down. It used to run great on a Pi 2 but now there are audio stutters on most WHDLoad demos/games and generally seems to struggle. The scrolling is a bit jerky as well. PS3 wireless controllers now seem to work with the latest commit but CD32 controls don't. Is this all because of JIT2 and TomB's move to the new weird input method ? Maybe it should go back to the old JIT as the Pi doesn't really have the power anyway.

And it still makes this repeating noise when quiting a WHDLoad demo/game with JIT on. What's that ?

I noticed something, when using WHDLoad demos/games and CPU is set to Fastest or 25mhz with JIT ON, I get stuttering audio that cuts out every few seconds and the scrolling stutters. If I set it to 25mhz with JIT OFF then the audio plays normally and I get smooth scrolling !!! There's either something wrong with JIT or the CPU settings.

I'm using a Pi 2.

I also tested the latest build on my rpi 3.
I've altered the makefile to optimize the built for the rpi3-cpu.
Currently I've just tested sysspeed under Workbench an so I get 7 till 8 times the speed of an A4000. That's pretty good.
I also tested Starstruck. The Video ist a little bit stuttering. But I need to switch to a 50hz Video mode.
Also is it not possible to reset the Emulator using Crl-Amiga-Amiga and the drive leds on Keyboard are not working.

And which optimizations was thant?
Normally "-march=native -mtune=native" does the trick, but that seems to create a error in gcc on my rpi3:
*** Error in `gcc': double free or corruption (top): 0x00624c40 ***

It seems like you can't run the installer for OS3.5 nor OS3.9 wit JIT enabled. If it is the installer script(?) fails to compile or it gets out of memory errors... (Running the latest UAE4ARM from git)

Just a note on some of the "available" options in UAE4Arm. Taken from the readme file :-

JIT enables the Just-in-time compiler. This may breaks compatibility in some games. You will not notice a big performance
improvement as long as you didn't select "Fastest" in "CPU Speed".

Note: In current version, you will not see a difference in the performance for 68020, 68030 and 68040. The cpu cycles for
the opcodes are based on 68020. The different cycles for 68030 and 68040 will come in a later version.

So JIT is pretty much useless unless you have CPU set to Fastest, and when you do use it you get sound drop outs and un-smooth scrolling because the Pi doesn't have enough power. (And 68030 and 68040 don't make any difference).

I'm using Amibian with one of the latest builds of UAE4ARM.
My DEL and HOME keys are not working.

But when I select the old UAE4ARM they are working. (but it keeps crashing on my so I want to use the latest version)

Do you guys know if this is a fault in the latest versions of UAE4ARM?
Or how to fix it?

Thanks!

@deepx

Sounds like the same issue I had: https://github.com/Chips-fr/uae4arm-rpi/issues/28

The latest commit with working function keys (for me) was a33d65d, before one of the TomB merges.

Midwan's SDL2 port would probably have solved this but it seems he went MIA as well same as every other attempt to fix up Amiga ARM emulation for some reason

Ouch, I was waiting for the SDL2 version too...

My DEL and HOME keys are not working.

But when I select the old UAE4ARM they are working. (but it keeps crashing on my so I want to use the latest version)

I just added support for DEL, HELP and parenthesis amiga key. I choose the same mapping as of fsuae (HOME is amiga left parenthesis)...

My DEL and HOME keys are not working.

But when I select the old UAE4ARM they are working. (but it keeps crashing on my so I want to use the latest version)

I just added support for DEL, HELP and parenthesis amiga key. I choose the same mapping as of fsuae (HOME is amiga left parenthesis)...

Thanks!!!

I just tried to understand how fastest cpu mode was working and found that some piece of it wasn't working well moreover the default value for pandora seems quite specific.
You have to understand how fastest cpu mode works: it try to emulate the maximum amiga cycle each frame. But by default it doesn't know how fast you host is (raspberry pi here). So it measure how much host cpu is free for each frame and so if host seems too slow or too fast for the current frame it will remove or add amiga cycles to be emulated for the next frame. This is done for every frame.

But you can't be sure that current frame reflect the cpu load of the next one, so you can underestimate. So you need have some margin and to make a choice of how much margin you take:
- either you're aggressive with a low margin and amiga emulation will be fast but if next frame needs more treatment than planned you'll have frame drop
-either you put a high margin so that you ensure few frame drop will occurs but emulated amiga will be globally slower.

I feel like pandora setting was quite agressive.

I increase the margin in latest push, moreover you can play with it too: just edit src/drawing.cpp and look for following defines:

#define OFFSET_FREE_CPU_CYCLES 1500
#define STEP_CYCLES 250

First one (OFFSET_FREE_CPU_CYCLES) increase the margin compare with pandora. It's in number of cycles. There is 20000 cycles per frames.
Second one should be less important: it is the number of cycle that can be change between each frame if raspberry is too slow or too fast... but 250 seems better than original value of 500...

So for people who get quite some frame drop/audio issue in fastest mode, try to use latest push, and moreover try to increase first define. And let me know if it improve and with which value.

The bad news is that sysinfo give slightly slower benckmark value (as expected ) and it break the double buffering mechanism.
The good news is that first number in LED is now working: this is the percentage of Rpi cpu free. So in fastest you can see that the higher first define is the higher the raspberry cpu free is (before it was around 7% now more like 12% since i increased the margin).

Interesting to see your finds on this. I knew there was something wrong with it. This also means it's wrong on the Android version. Did you check the other speeds (25, 14, 7mhz) to see if TomB had specific timings for the Pandora only ?

I would not say it's pandora specific but more a choice to give priority to benchmark and amiga cpu speed...hence there is no perfect value.
I just improved the fps counter, it was wrongly saying frame drop from time to time despite frames where ok. Now value could be tune again... it's difficult since I have some frame drop from time to time but could be other process in raspbian...
I've taken improvements from yzi: his audio management without semaphore but with replay mecanism. There is gl shader available in gles version but it looks like quite early to use it...

Latest version is a lot slower, I think something needs to be reverted. On a Pi 3, the game 'Jim Power' I used to get about 41fps on the title screen (68020, 25mhz, Fast Copper On). Now I get 25fps !!!!!!!!

This is explained by the fact that double buffering is deactivated currently. If rpi is too slow for every frame then you drop directly to 25fps (half of the framerate) instead of being able to retrieve frames from time to time...

This is explained by the fact that double buffering is deactivated currently. If rpi is too slow for every frame then you drop directly to 25fps (half of the framerate) instead of being able to retrieve frames from time to time...

So this is because of your 'Fastest CPU' speed fix. It breaks double buffering and also affects all other speed settings and not just 'Fastest' ? I hope this can be fixed because that's a huge frame rate drop. So for instance, if a game can only reach 49 fps and not 50 it automatically drops to 25 fps ?

I don't think you would achieve 49fps if every frame is too heavy, and if its not the case the drop will not be the same... 49fps looks like more a sporadic overloading.
Jim power is the worse case: every frame is overloading the rpi, only few time it succeed to be smooth. It could be an interesting game for searching optimizations ...