Zeewolf - A1200 Frame Rate

[copse]

Originally Posted by alexh

There was an amazing project to optimise Zarch (virus) on the Archimedes and the developer was able to optimise the render so much (and he was using a very fast ARM) he was able to increase the viewport so much which greatly improved the game. It was quite incredible.

IIRC he had access to the source code, and couldn't release it. Was interesting to read the thread though and watch the videos on that channel.

[alexh]

Originally Posted by copse

IIRC he had access to the source code, and couldn't release it.

No he re-sourced the binary and reverse engineered it. The only reason I think he chose not to release it was that he wasn't happy with it (i.e. not finished). It was his project to do with as he liked afterall.

[copse]

Originally Posted by alexh

No he re-sourced the binary and reverse engineered it. The only reason I think he chose not to release it was that he wasn't happy with it (i.e. not finished). It was his project to do with as he liked afterall.

Haha, if I remember something it is generally spot on. You've forced me to check to make sure it is the case here, and to back it up. His very first words on the [ Show youtube player ] are:

soon after posting my last video on Zarch I was sent the original source code for it um so I thought I'd uh take the opportunity to demonstrate how I go about converting games or original source code to run on 32bit systems

[alexh]

This is the [ Show youtube player ] that we were discussing.

5:30 -> 7:00 - Source

He's had to go through, name all the labels by hand, comment all the asm. Lots of labels that still have non-human readable names. If he had the original source code when he started this then I don't believe the source code would look like this.

I think you're right he was sent the source code but he definitely started by re-sourcing the binary.

[paraj]

Had a closer look, and it's limited to around 4-5 frame because it spreads out some bitplane buffer clearing over multiple frames. Reworked that it does seem smoother on 060 (and much faster w/ JIT). Still haven't looked at rendering code.

Attached is WIP slave. Loading should also be quite a bit faster.

N.B. Passwords:

start 1
imago 5
tiburon 9
argus 13
marten 17
sockin 21
gannpay 25
escland 29
frampage 1 with plane

[rygar]

Originally Posted by paraj

Had a closer look, and it's limited to around 4-5 frame because it spreads out some bitplane buffer clearing over multiple frames. Reworked that it does seem smoother on 060 (and much faster w/ JIT). Still haven't looked at rendering code.

Attached is WIP slave. Loading should also be quite a bit faster.

N.B. Passwords:

start 1
imago 5
tiburon 9
argus 13
marten 17
sockin 21
gannpay 25
escland 29
frampage 1 with plane

Thank you. Some passwords missing on HOL

[tomcat666]

Smooth as butter now, Flies in 1 or 2 frames rendering time now, so pretty much 50 fps, better than Zeewolf 2, which flies in 2 or 3 frames.

[Karlos]

All it needs is a light touch from Mr FastMemFirst

[paraj]

Took closer look and rendering is done with CPU (Sort of like e.g. TFX). This code was already in extended (potentially fast) memory. Stack was not. That's why early, initial attempt did not bring much.

I've added frame time counter (number of scanlines / frame rendered) and there is indeed objective improvement on 060 (of course it's blown out the water by pistorm). Also added a bit of training (just for fun).

Since rendering is CPU based it might (on fast machines) benefit from TFX "trick" of moving render buffer to fast and copying later. Would also eliminate need for blitter clearing of buffer. However since there isn't as much overdrawing as in TFX benefit might not be huge, but maybe worth looking into.

[rsn8887]

With the WIP slave on MiSTer, I am now getting a dos error object not found loading from disk.1 after the WHDLoad Splash screen. Old slave works fine.

[paraj]

Originally Posted by rsn8887

With the WIP slave on MiSTer, I am now getting a dos error object not found loading from disk.1 after the WHDLoad Splash screen. Old slave works fine.

Old slave had disk.1 in "data" dir. Copy/move either thing around (disk.1 to new .slave dir or other way round) or use "DATA" option for whdload to tell it where disk is. (Maybe I'll change this to match old slave to avoid more confusion..)

[AmigaHope]

As was mentioned earlier in the thread, I'd take a close look at chip memory performance. Some games make their chip mem operations too close to vblank and wind up being some cycles too slow if the accelerator has a poor chip ram interface, causing the game to have to wait an entire extra frame doing nothing.

[paraj]

Originally Posted by AmigaHope

As was mentioned earlier in the thread, I'd take a close look at chip memory performance. Some games make their chip mem operations too close to vblank and wind up being some cycles too slow if the accelerator has a poor chip ram interface, causing the game to have to wait an entire extra frame doing nothing.

ChipMem speed is definitely important, but it also uses a blitter queue that only progresses once per frame to clear bitplane buffers. This locks the frame rate unnecessarily low (and wastes time). Combining that with CPU based rendering to chipmem means FPS is sub-10 even on 060. I've removed the first limitation, but it's still not particularly fast.


However earlier this morning I did a quick experiment, and "TFX makeover" definitely has potential.. stay tuned.

[alexh]

Thanks for all your hard work. This is a great game that will become even better and deserves to find a next generation audience.

[Seiya]

Zeewolf engine i think is otpmized for Fast RAM only.
in the past i bought a ram expansion card for my Amiga 500 (1,8 MB) and the game was slighty better in performance.

however i tried WHDLoad Zeewolf 1 and 2 in WinUAE (that emulate stock A1200 with 8 MB ZII): Very fast and fluid.

[paraj]

Well.. now it has other issues. Anyway, should be a tiny bit faster.

Comparison between original rendering code and new version on my B1260/50: https://i.imgur.com/JqW9UCf.mp4

[wiz12]

Originally Posted by paraj

Well.. now it has other issues. Anyway, should be a tiny bit faster.

Comparison between original rendering code and new version on my B1260/50: https://i.imgur.com/JqW9UCf.mp4

That's a much needed huge improvement in speed but dare I say it's now too fast as to be unplayable. Would probably need a frame limiter.

[Seiya]

Originally Posted by wiz12

That's a much needed huge improvement in speed but dare I say it's now too fast as to be unplayable. Would probably need a frame limiter.

yes, it is too fast. The good speed (whdload) should be like this:
[ Show youtube player ]

[paraj]

First it's too slow, then it's too fast, no pleasing you guys, eh?
Added frame limit (don't complain if it crashes when off and JIT is enabled).

Defaulting to 2, which seems nice, but don't know if 3 is more appropriate (that's what I get w/ original game in tactical view).

[alexh]

Originally Posted by paraj

First it's too slow, then it's too fast, no pleasing you guys, eh?

Increase the viewable area from 10x10 tiles (or whatever it is) to something bigger? That might slow it down a bit?

:-P

Thanks it looks incredibubble.