Fast Red Zone and other fastmem-relocated projects

[Aardvark]

Originally Posted by jotd

yes!! found a fake reloc that made the tunnel crash. Zoned the new version. Works now.

Works perfect now, as far as i can tell.

[jotd]

Originally Posted by lmimmfn

That makes sense but wouldn't it make relocation simpler if you had a map of addresses visited by the program counter to seperate code execution vs gfx/sound etc. so you're only focusing on code jmp instructions/and/or interrupts?

You would need a specialized emulator for that. Besides there are also data accesses. If they're read only, it doesn't matter if you're reading the unrelocated copy or not, but if they're read/write it can be a problem. And anything that generates false alarms must be fixed anyway.


I'm glad that Starglider 2 works ! I have another version to support not sure I'm going to do it, there are so many manual steps for that game...

[tolkien]

Disk access in Hunter works again. Fantastic! But I have two new issues reported in Mantis.
(Sorry for the resume name of one of them but cant edit the entry).

[lmimmfn]

Originally Posted by jotd

You would need a specialized emulator for that. Besides there are also data accesses. If they're read only, it doesn't matter if you're reading the unrelocated copy or not, but if they're read/write it can be a problem. And anything that generates false alarms must be fixed anyway.

No, I understand it would need support in the emulator and like you say if addresses visited by the program counter included both read and write access then it would be easy(via correlation) to identify self modifying code.

I'm sure Toni could magic it up

[jotd]

Self-modifying code is already supported (with limitations). My problem is more specialized than that.

more news:

Small update on blitter emulation, thanks to paraj and his C++ blitter emulation code, I converted to C then asm and tested on one of my games, it seems to work, so later I'm going to try to plug it in Killing Cloud and later on FighterBomber

I've tested Infestation (beta, unstable) on my 060 and it runs much smoother than the original. Which is good news.
I've also tested Starglider 2 and on my 060 it makes no difference (in chipmem it's already very smooth). But maybe it does make a difference on 020 or 030 + fast. Testers welcome.

[Angus]

This is awesome work, Jean-François. :-)

[jotd]

Small update & news on the global project:

- someone reported bugs in Hunter relocated. I have fixed them, but not released the fix yet (3 versions to fix)
- paraj helped me with the blitter code. There are still issues but it more or less works.So I have plugged the blitter emulation into "the killing cloud" and it kind of worked. Gfx glitches because blitter emulation isn't perfect (yet) but it already looks faster than with the blitter on fast cpus. Maybe a mixed mode (half blitter / half cpu depending on the bitplanes) would rock.

[Reynolds]

Cool, can't wait to see it on my screen!

[Reynolds]

4D Sports Driving just came to my mind... it also looks horrible in terms of frames / sec...

[jotd]

4D Sports Driving is probably coded in C with old compilers. Time to use my profiler on that one (that one and Hard Driving 2).

News: paraj is helping me with Killing Cloud. The issue here is more than fastmem but heavy use of blitter.
Blitter emulation works but is too slow so he rewrote the drawing routine more globally. Seems to speed it up. More to come...

Started work on relocating Fighter Bomber too.

[Reynolds]

Originally Posted by jotd

4D Sports Driving is probably coded in C with old compilers. Time to use my profiler on that one (that one and Hard Driving 2).

News: paraj is helping me with Killing Cloud. The issue here is more than fastmem but heavy use of blitter.
Blitter emulation works but is too slow so he rewrote the drawing routine more globally. Seems to speed it up. More to come...

Started work on relocating Fighter Bomber too.

Sir, with other fellas here you're definitely deserving a star on the Amiga Walk of Fame.

If I can help with testing, please let me know.

[jotd]

I'll zone beta versions as soon as they're more or less running

I profiled 4D Sports Driving and the program spends most of the time ... waiting for the blitter to finish... Paraj replaced blitter routines by cpu routines but that was some extra work to rewrite drawing routines (not just plugging blitter emulation). That speeds up the game (Killing Cloud). But that's a lot of specific work for each game, without guarantee that it will provide significant speed up.

[paraj]

I've spent way too much time on this, but here's a Killing Cloud slave for testing. It's still a shit^H^H^Hlideslow when clouds are off, but hopefully it's faster than previously.

Added a FPS counter so speed can be objective compared. The other 3 options disable stuff that I think are logical optimizations but may have bugs (I've only tested the first level for a couple of minutes, not actually played the game) or not be improvements on other machines.

On my A1200 (Blizzard 1260@50MHz) I get 2-3 FPS with clouds off (with most of the city in view) and around 10-16 with them on.

Comments on whether the various options improve speed for different configurations (or if they don't work of course) would be great.

[jotd]

Some good news about the project

- paraj worked wonders on speeding up Killing Cloud, and the result is in the zone for you to test (I have fixed some other remaining issues too).

- I have profiled Hard Driving 2 and I don't think the game is slow or taxing for a fast machine. It's just that whdload slave introduced "speed regulation" (because it was maybe too fast at times) but in a wrong way (adding cpu delays before each blitter wait), so it may be slower that it should. I have to re-do the speed regulation so it's steady 25 fps and not too fast either hopefully. The game is coded in C, but it seems that critical parts were done in asm, and it's natively running in fastmem (standard amiga exe). Good effort for a Tengen game!

- Still struggling with Fighter Bomber. But making progress. Paraj had a look at the drawing routines and it appears that they're similiar to Killing Cloud (same dev) so game can probably be sped up a great deal. Let's hope.

Send in your speedup ideas for 3D or even 2D games. I'll study them. It seems that Vampire struggles with some games (with code running in chip, can someone confirm?) even more than A1260 phase 5 board, reading the number of mantis bugs reported for that configuration. I'm not sure I can fix every game like that, it's a lot of work. Maybe they should fix the Vampire instead.

[Stevo2k]

Aquaventura Could do with an fps injection...

[saimon69]

Aquaventura is a whole different can of worms - especially the 1989 press demo -_-

[jotd]

Just tested Aquaventura. Not the fastest, even on 060. Will check on WinUAE.

I tested last paraj build of Killing Cloud and wow! it's almost playable with "cloud off". We get 5FPS instead of 2... So in normal play it's very smooth.

[Stevo2k]

Would this work on sprite scaled games like: Line of Fire, Galaxy Force, Power Drift?

[Retroplay]

Pressing F2 ingame in Killing Cloud v2.0 (fast relocated slave) makes it quit back to Workbench.

If I do the same with old v1.0 or v2.0 NoReloc slave, it switches to rear view.

[jotd]

Originally Posted by Stevo2k

Would this work on sprite scaled games like: Line of Fire, Galaxy Force, Power Drift?

If a game crawls on real machine but runs smooth on WinUAE (A1200 speed, or fastest possible), it means that it could benefit from a speedup. If the game runs from chip memory, relocating to fast memory gives some speedup, unless it's relying heavily on the blitter.


I'm targetting 3D games mostly because they're also CPU intensive because of all the math. It may not be the case for sprite scaling games. Power Drift looks not very good I don't see how it could be better.