TFX on a PiStorm32

[jotd]

Vroom is already too fast for some machines. I have some requests to slow it down

it's by no means a 3D game, but a "classical" 2.5D racer. Amiga has no trouble with those (when done properly, see Chase HQ )

[Aardvark]

As far as I can tell, the plain 020 version is just as fast as the 040 on 100mhz TF1260.
[ Show youtube player ]

[DDNI]

This progress is one of those totally unexpected developments that really delights! Thank you

[tomcat666]

Originally Posted by dlfrsilver

Vroom is already hyper optimised for Amiga (from the programmer itself!).

Doesn't work on most modern accelerators because of SMC (same goes for one of the most popular racers - Stunt Car Racer)

[Angus]

Originally Posted by paraj

@Angus: Try the plain 020 version instead. If that works, other issue can be debugged later (what 060 libraries are you using? mmulib?)

Plain 020 version worked well! :-)


In NTSC (Help key) I got 8-10 fps, and a bit less in PAL. This is on fairly low detail.



I'm afraid I'm a bit vague about my system software. In the startup-sequence I have Cyberpatcher, rom2fast, Copymem060 - and I run Fastexec which speeds up some 3D games like Star Crusader. I don't seem to have mmulib installed.

[Aardvark]

Originally Posted by tomcat666

Doesn't work on most modern accelerators because of SMC (same goes for one of the most popular racers - Stunt Car Racer)

CPU caches can't be disabled with PiStorm? Vroom and SCR runs without issues on my modern accelerator.

[tomcat666]

Originally Posted by Angus

In NTSC (Help key) I got 8-10 fps, and a bit less in PAL. This is on fairly low detail.

8-10 fps or the number displayed is between 8 and 10 ?

FPS is 50/number (I think )

[tomcat666]

Originally Posted by Aardvark

CPU caches can't be disabled with PiStorm? Vroom and SCR runs without issues on my modern accelerator.

Not completely, it is a pure JIT cpu, you can only set how many instructions are in the cache, which doesn't help in those cases.
When all bugs and features are done Michal will work on an interpretive version, but when that happens, no idea

[Angus]

Originally Posted by tomcat666

8-10 fps or the number displayed is between 8 and 10 ?

FPS is 50/number (I think )

Oh, jeepers!


I was referring to the number displayed, which I thought seemed approx fps?

[alexh]

which would make it 60/number in NTSC mode

[tomcat666]

Originally Posted by alexh

which would make it 60/number in NTSC mode

On pistorm32 here it flickers between 1 and 2 on NTSC while on PAL it mostly stays on 2

[jotd]

I've watched the videos, it's really excellent! I'm considering PiStorm a "next gen amiga", using the original custom chips & chipmem, so it's really worth the effort.

I'm going to test that on my A1200/060 where chipmem access is really slow. I'm confident that speed can be gained there too.

Don't hesitate to propose new ideas of programs that could be accelerated in your opinion.

[paraj]

Yes, the displayed number is number of vblanks/update, so for PAL it's 50/x and NTSC 60/x.

Quick comparison from my 060 (only have shitty composite capture available):
https://imgur.com/a/xZgONHk

Plain 020 version (PAL).

[wiz12]

Originally Posted by jotd

I've watched the videos, it's really excellent! I'm considering PiStorm a "next gen amiga", using the original custom chips & chipmem, so it's really worth the effort.

I'm going to test that on my A1200/060 where chipmem access is really slow. I'm confident that speed can be gained there too.

Don't hesitate to propose new ideas of programs that could be accelerated in your opinion.

Could Zeewolf or Zeewolf 2 benefit from this?

[alexh]

Zeewolf and Zeewolf 2 are a bit odd. The engine doesn't scale with a faster CPU and the addition of FastRAM but it does improve.

A500 16-bit ChipRAM + 7MHz 68000 vs A1200 32-bit ChipRAM + 14MHz 020 (no FastRAM)
[ Show youtube player ]

A1200 vs A1200 + 8MB FastRAM
[ Show youtube player ]

A1200 TF1260 (11m 30s)
[ Show youtube player ]

I suspect, like TFX, Zeewolf renders in planar mode directly into ChipRAM and uses the Blitter. So the CPU spends most of it's time waiting for ChipRAM accesses or Blitter to finish.

[jotd]

superb work paraj! (as always)

[lmimmfn]

Amazing work paraj, thanks

Originally Posted by tomcat666

Thank you. the Plain 020 version now works really nicely too and the improvement in the FPS is very much visible (as you said) also on pistorm32... it never goes to 4 now, it is mostly on 2. A little video of the beta 4 and plain 020 version:

[ Show youtube player ]

Wow, this looks absolutely stunning

[Angus]

Originally Posted by jotd

Don't hesitate to propose new ideas of programs that could be accelerated in your opinion.

Forgive me for banging on...... :-)


Could any of this be applicable to Tornado and whatever it's doing with it's AGA display chip memory issue, or whatever it is that makes the 060 version run at a similar speed to the 030 version?





I'll shut up now.

[Aardvark]

Originally Posted by jotd

Don't hesitate to propose new ideas of programs that could be accelerated in your opinion.

Red Baron also has some chipset bottleneck. The Mac version is runs smooth on same machine with MuEVD video drivers.
[ Show youtube player ]

[tomcat666]

Originally Posted by Angus

Could any of this be applicable to Tornado and whatever it's doing with it's AGA display chip memory issue, or whatever it is that makes the 060 version run at a similar speed to the 030 version?

Tornado AGA definetly seems a candidate, a quality game with quite sub-quality FPS even on accelerators (not as bad as TFX but similar).