Crazy idea to improve speed on games for AGA machines

[sonninnos]

Oh man, if only Stunt Car Racer could be as smooth as the C64 SuperCPU version, instead of just faster..

[DanyPPC]

Well done jotd !

I played SuperCars 2 AGA 0.89 on A1200/BlizzIV and A1200/ACA1233 and the framerate/speed is better than previous 0.88 version.

On WinUAE this improvement isn't visible.

[turrican3]

You know exactly what i will propose : shadow warriors.

[jotd]

Originally Posted by Toni Wilen

It should be relatively easy if game does not use horizontal scrolling.

Horizontal scrolling can make it very complex task. Bitplane pointers must be 8 bytes divisible, different BPLCON1 range (0-15 -> 0-63). Horizontal scrolling also requires smaller DDFSTRT (than OCS/ECS) = lost sprites.

horizontal scrollers (not using the blitter) are not an easy task...

ECS: bitplanes aligned on 2 bytes, BPLCON1 scroller shift 0-15 bits
AGA FMODE=3: bitplanes aligned on 8 bytes, BPLCON1 scroller shift upgraded to 0-63 bits.

Means that a lot of reverse engineering must be done to adapt those games.

[vulture]

@jotd

I tried Chase HQ II on WinUAE and saw no real difference compared to the real thing, so, I guess the ICE racers will remain crappy as they currently are.

@turrican3

Great idea, I just hope it's one of those that can be improved as I recall it also plays the same way under WinUAE as on the real system.

[Aardvark]

Top Gear 2 (OCS) Road updates every 1/3 frames in cycle-exact mode (68020), 1/2 frames in non-cycle-exact.

Top Gear 2 (AGA) 1/5 frames in cycle-exact, 1/2 in non-cycle-exact

Power Drift: Framerate fluctuates between 1/5 and 1/10 in cycle-exact, 1/3 to 1/5 in non-cycle-exact.

FIFA International Soccer: 1/2 to 1/3 frames scrolling in cycle-exact, 1/1 in non-cycle-exact.

[jotd]

Originally Posted by DanyPPC

Well done jotd !

I played SuperCars 2 AGA 0.89 on A1200/BlizzIV and A1200/ACA1233 and the framerate/speed is better than previous 0.88 version.

On WinUAE this improvement isn't visible.

Coincidence? I just sold more than 10 copies of the game today!

As Toni reminded, for a lot of games it will be simply too difficult because of horizontal scrolling. Supercars 2 doesn't really scroll, everything is blitted on screen each time. Why it runs so fast, maybe it's because the background is not made of tiles, so there are "only" 7 calls to the blitter. Redrawing the same surface but with small tiles takes more time, and most tiled games use hardware scrolling. Blitting the whole background each time is costly, but avoids to erase the objects prior to redrawing. Good tradeoff, very well executed by Shaun Southern, an inspiring programmer.

Not using hardware scrolling allowed Supercars 2 to have a 2 player mode with vertical split screen. Not sure it's possible using hardware scrolling.

So once again this is happening with a bit of luck for Supercars 2:

- source code available and which can be easily rebuilt (not sure anyone can rebuild Jim Power or Turrican 3)
- 256 color pics available to replace old 64 color EHB ones
- FMODE possible thanks to blitter scrolling
- more color for cars: easy way to enhance the visuals without redrawing all graphics
- module player that can play modules & sounds at the same time (and reusing Lotus tunes, and getting original J.M.D/saimon69 original tunes too)
- programmer (me) who loves that game

[dlfrsilver]

Originally Posted by jotd

Coincidence? I just sold more than 10 copies of the game today!

As Toni reminded, for a lot of games it will be simply too difficult because of horizontal scrolling. Supercars 2 doesn't really scroll, everything is blitted on screen each time. Why it runs so fast, maybe it's because the background is not made of tiles, so there are "only" 7 calls to the blitter. Redrawing the same surface but with small tiles takes more time, and most tiled games use hardware scrolling. Blitting the whole background each time is costly, but avoids to erase the objects prior to redrawing. Good tradeoff, very well executed by Shaun Southern, an inspiring programmer.

Not using hardware scrolling allowed Supercars 2 to have a 2 player mode with vertical split screen. Not sure it's possible using hardware scrolling.

So once again this is happening with a bit of luck for Supercars 2:

- source code available and which can be easily rebuilt (not sure anyone can rebuild Jim Power or Turrican 3)
- 256 color pics available to replace old 64 color EHB ones
- FMODE possible thanks to blitter scrolling
- more color for cars: easy way to enhance the visuals without redrawing all graphics
- module player that can play modules & sounds at the same time (and reusing Lotus tunes, and getting original J.M.D/saimon69 original tunes too)
- programmer (me) who loves that game

Turrican 3 can be. You need to prepare an extended ADF and activate Winuae Longtrack option in the config file you use.

Then you use the script and you write the master disk.

[AmigaHope]

I know Armour Geddon II uses blits for fills (among other things to get the sand dune texture effect), probably AG1 does as well. Maybe these could benefit.

[tomcat666]

Originally Posted by jotd

Fighter Bomber was already heavily processed by paraj & me: relocation to fastmem and usage of fmode=3 if AGA chipset. Nothing to optimize further here. Game is just too ambitious or too badly written.

Maybe you didn't test the latest whdload version ? It "flies" at around 10 fps on my A1200/060. Not great but better than the previous slideshow of a game that it is originally.

Just tested it on my A1200 with pistorm32 ... it goes as high as 25 fps, the lowest is at 12 fps, Basically it is either 25, 16 or 12 fps, no other numbers pop up But even at 12 it feels very smooth.
Edit: Ok, on Mt.Rushmore it goes as low as 8

[vroom6sri]

@tomcat666 - did you notice any bugs with Fighter Bomber?

I only tried it quickly back then and it was a bit glitchy but I had forgotten all about it until now and I'm AFA (Away From Amiga) for a while now so can't retry it.

[tomcat666]

Originally Posted by vroom6sri

@tomcat666 - did you notice any bugs with Fighter Bomber?

I only tried it quickly back then and it was a bit glitchy but I had forgotten all about it until now and I'm AFA (Away From Amiga) for a while now so can't retry it.

Was not playing it a lot, just a few minutes, not a big fan. But I didn't notice any bugs really or that it was glitchy...

[Amigajay]

I used the fmode command for tude for games like Combat Air Patrol on my CD32 discs that gave a nice speed-up.

[daver]

Lion king AGA and Zool 2 AGA two games nearly unplayable due to extrem slowdowns.
my dream is that some day i get the option to play amiga games on Graphics Card instead of the cutsom chipset may never happen but that would be great.

[arcam]

Originally Posted by jotd

For some games that are slow, because lots of blitting / 32 colors / EHB and all, those could be sped up on AGA machines by changing FMODE = 0 to FMODE = 3 (and also adjusting modulo as possibly bitplane addresses)

I way trying to add FMODE=3 to my supercars 2 AGA hack (I lazily retained old FMODE=0 setting), and while testing on real hardware, I had the impression that the game was much smoother.

Paraj already did such work in Killing Cloud (besides rewriting the part of the 3D rendering engine ).

AGA games probably already use that mode (well, we could have surprises) but if you have ideas about games that are slow and could be made faster, please do tell.

Hi Jotd,
What do you mean when you explain to add FMODE=3 ? In the icon via TOOLTYPE ? I tried but Snoopdos dont show any call of FMODE parameter. Thanks a lot in advance

[ReadOnlyCat]

Originally Posted by jotd

[...] Supercars 2 doesn't really scroll, everything is blitted on screen each time. Why it runs so fast, maybe it's because the background is not made of tiles, so there are "only" 7 calls to the blitter.

Wow, I am mindblown. I would never have guessed the whole screen was blitted rather than hardware scrolled.
I have been wondering about the (relatively) low frame rate of the game for decades but now that explains it. I guess this should be visible in the UAE DMA debugger but I never fired it for that game.

Originally Posted by jotd

Redrawing the same surface but with small tiles takes more time, and most tiled games use hardware scrolling. Blitting the whole background each time is costly, but avoids to erase the objects prior to redrawing. Good tradeoff, very well executed by Shaun Southern, an inspiring programmer.

Not using hardware scrolling allowed Supercars 2 to have a 2 player mode with vertical split screen. Not sure it's possible using hardware scrolling.

I agree that Shaun Southern is a good programmer but I am wondering whether this tradeoff is really worth it.

Sure, this allows to use the same code for both sides of the vertical split screen but if you think about it he could simply have used hardware scrolling for the single player and left part of the split screen, then used full-blit only for the right part of the split screen.

In total, only half of the screen would have needed to be blit, thus requiring about half the DMA bandwidth.
The resulting code would have been slightly more complex but not by much (tile based hardware scrolling is easy enough).

Interesting choice in any case.
Thanks for sharing jotd! ?

[turrican3]

you stopped this jf ?

[alexh]

It's being discussed in the Lotus 2 thread active right now.

[jotd]

In total, only half of the screen would have needed to be blit, thus requiring about half the DMA bandwidth.

yes and no. Blitting the whole background each time avoids having to restore the backgrounds locally. There are no sprites so everything must be re-blitted & computed.

Blitting the whole background each time simplifies a lot of computations involved when you want to erase previous BOB positions. It's impressive how he managed to do so many blits and so many clipping. The secret is: he's preparing his blits in a blit list, then performs them in a chain. There are almost no blitwaits, as he uses the blitter interrupt to trigged the next blit. In the meantime, the available CPU time is used to compute a lot of shit. You also need to know exactly how many blits you're going to do in advance and kind of "duplicate" the code (between parameter computation and actual blit).

Genius! (but also tricky to maintain as if the parameters are shifted mayhem ensues). Same method used on the Lotus series.