Fast Red Zone and other fastmem-relocated projects

[lmimmfn]

Originally Posted by jotd

I have described the process in one answer here https://retrocomputing.stackexchange.com/a/24613/3316

Basically, disassembly gets you the easy ones (LEA, MOVE direct... in obvious code sections). The problems arise with move.l # and mostly dc.ls (pointers in data)

There are a lot of "little secrets" (which aren't secrets) to prepare the source file (like renaming the labels with the offset in it, or splitting ORIB #0,d0 i 2 DC.W 0...), plus the script which catches undefined pointer table labels you just inserted and declares them at the proper location in the source file. README in my github for the projects try to list how to proceed.

With kickstart files it's easier since addresses range from 0xF8000/0xFC000 to 0x100000.

On a game, you have to use memory protection (winuae or whdload) so if the game tries to access the old memory you catch it. But it can be very tedious so I have developed a lot of scripts to try to detect pointer tables.


The most annoying case (happens in Starglider) is a list of structures containing pointers and also data, but not periodically (kind of varying structures I dunno...). In that case if you declare a relocation when it's data then you introduce a bug which is super hard to track. Then you have to doubt of all the relocations, remove them, run with memory protection on, retry.... Time consuming even with the right tools. Without any tools it's just impossible.



Ah, that makes sense. So ATM play with the old install with CACHES on. But it hasn't speed regulation so for instance in WinUAE the game is unplayable because of speed (a bit like freescape games)
But I think I'm very close to making it work from fastmem so stay tuned Working on infestation ATM.

Yesterday I tried Hunter on my 060 with fastmem or chipmem and I thought that even with chipmem the FPS were good. But clearly with fastmem it was smoother.

Thanks, that makes it clear, I was curious as I disassembled Buggy Boy on the CPC and found it frustrating to determine which addresses were data and which were executable code,

I wonder if UAE supported dumping all addresses(and counts) the PC visits would speed it up? I.e. play a game from start to finish, disassemble the code and enrich with the dumped PC addresses to remove dummy labels in data segments of the code?

[jotd]

accessing data in words or longwords would make a difference. If access as word, then it's not part of an address, if access as longword, then probably an address (if in addressable range), but not always.

[Reynolds]

Haven't checked but saw a YT vid of Flight of the intruder, wonder if it also got speed-up with the Fast Ram reallocation?

[jotd]

either it's on an emulator, or real amiga. Game naturally uses fastmem, even in whdload. No big gain to expect.

Some news:

- good progress made on infestation, will zone a playable beta in a few
- still have the project to adapt paraj CPU blitter code for Fighter Bomber and Killing Cloud
- Starglider 2: standby but will finish it

[Predseda]

Killing Cloud really deserves to be faster game.

[jotd]

Infestation fastmem / no caches / no sound to test in the zone

[Galahad/FLT]

Originally Posted by jotd

So all the various sections are different programs? that complicates the work all right (encrpytion: not that much as whdload slave already decrypts them)

Yup, every vector game is separate code entirely, majority is in lower 512k chip ram, but it does have some code parts that load into slow ram/upper chip ram.

Its easy enough to track what code is what, simply latch onto the games loader (there is only one), and once it finishes loading, game is unpacked and decrypted, and it also passes in registers where the data was loaded and its size, so at least you can have a compare table knowing that those two elements are enough to identify what has just been loaded.

[jotd]

yes, with whdload slave already existing it makes life easier

[tolkien]

Hunter now runs like light in my a1260 but there is an issue with disk access.
I have written the problem in mantis.

[jotd]

good catch. There were hidden relocs like

move.l  #$0000233c,d0

how to decide if it's an address or a value?? by using watchpoints and trying to cover as much code as possible. Otherwise, it's hard to tell. If you relocate all moves immediate to data register by default there's a greater risk that you introduce a hard to find game bug.


So fixed, and update sent to whdload site. Careful: slave itself isn't important to update. The .reloc files have been updated to take those new relocs into account.

[tolkien]

Thanks mate. Im really enjoying this game. You are the master!

[jotd]

Starglider 2 fastmem update in the zone. Now with crosshairs and sound.

Fixed a few crashes. I have a doubt with the tunnel from the planet. I flew through it and exited it: nothing there. Only an empty planet. Well. Have a try.

[lmimmfn]

Originally Posted by jotd

accessing data in words or longwords would make a difference. If access as word, then it's not part of an address, if access as longword, then probably an address (if in addressable range), but not always.

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?

[Aardvark]

Originally Posted by jotd

Starglider 2 fastmem update in the zone. Now with crosshairs and sound.

Fixed a few crashes. I have a doubt with the tunnel from the planet. I flew through it and exited it: nothing there. Only an empty planet. Well. Have a try.

Still crashes in the tunnel, also saving doesn't work.

[jotd]

Originally Posted by Aardvark

Still crashes in the tunnel, also saving doesn't work.

Ok watching your first video again. There are several tunnels I may have taken another one. Will try to play exactly like you did.


About Red Zone (just re-read your message about the crash on 060/ 100MHz), I was wondering if the game should not be slowed down like other games. I don't have a 68060/100 but "just" a 68060/50 and it runs fine, but for instance with WinUAE if you set "fastest possible" the game is super-slow (which is not logical). So maybe speed regulation (like I did in Starglider 2) could be interesting to add.


Speed regulation is necessary on games that expect/are targetting no more than 25 fps on any machine because there are a lot of operations for a 68000. If the cpu is fast enough, the cycle can be completed in 1 frame and the game runs at 50 fps but twice as fast. Happened to F/18 interceptor, and a lot of other games (Populous 2, Syndicate, a lot of 2D games as well like Magic Pockets).

[Aardvark]

Originally Posted by jotd

About Red Zone (just re-read your message about the crash on 060/ 100MHz), I was wondering if the game should not be slowed down like other games. I don't have a 68060/100 but "just" a 68060/50 and it runs fine, but for instance with WinUAE if you set "fastest possible" the game is super-slow (which is not logical). So maybe speed regulation (like I did in Starglider 2) could be interesting to add.

Crashes and buggy behavior in Red Zone with TF1260 was due to cache related bug in it's beta/alpha firmware. I also have Warp1260 now, and Red Zone behaves normally with it in high clock speed.

Stephen Leary explains this bug in detail here at 32:33 mark
[ Show youtube player ]

[jotd]

Originally Posted by Aardvark

Still crashes in the tunnel, also saving doesn't work.

ok I zoned a new version, fast memory. Save now works. I've been in the tunnel a lot of times and it never crashes for me. Can you try with NOCACHE just in case?

[Aardvark]

Originally Posted by jotd

ok I zoned a new version, fast memory. Save now works. I've been in the tunnel a lot of times and it never crashes for me. Can you try with NOCACHE just in case?

No changes, it crashes even in Winuae with basic 020+8mb config
[ Show youtube player ]

[jotd]

more interesting. I don't have a crash with WinUAE. But with 020 + 8MB Z2 I get one ! so time to investigate more thanks!!

[jotd]

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