TF330 Akiko C2P Compatibility (Turn Off Data Cache)

[nibiru]

Thanks everyone.

Should I try to flash the beta in the other thread then? I remember flashing the TF328 with my RaspPi, can I do the same?

There is a newer one AlenPPC can give you.

Got and flashed it right away!

Here are my findings with the V2 firmware:

• Sysinfo: speed boost from 1.63 vs A3000 to 1.97, this one seems nice!
• Fusion: fails to detect the Akiko at times, and when it runs the screen is mangled again, this time it doesn't matter whether the data cache is on or off
• Wing Commander: runs, but *a lot* slower than with the V1 firmware. I was really surprised by how fast it was going (compared to the CPU only version), meaning that Akiko is not super-crappy after all
• DoomAttack (aka Doom CD32) fails to initialise the Akiko C2P, resulting in an RTG version being used, that is terribly slow (my guess is that the OS fails with Akiko too, so it falls back to CPU for WritePixelArray8(). With the V1 firmware Doom was somewhat playable with the default screen size (and 1x1 pixels)

Hope this helps... I think I'll go back to V1 for now.

Cheers.

EDIT: I've also noticed that SysInfo reported the MMU as inactive earlier, while now it says: "In use". Not sure what are the implications, and if it would help to install a 68030.library...

[terriblefire]

I havent really done any C2P tests. It would be nice if there was a simple tool to exercise the Akiko C2P.

[alenppc]

I did some tests tonight and I can't get the akiko_c2p routines in DoomAttack to work at all, regardless of the firmware version or CPU type I am using. Actually they won't even work on the 328 for me. I just get a guru, consistently. I know I had them running at some point but I can't get them to work anymore. I know this is all academic since they were way slower than the 020 routines but I tested just for the sake of it.

EDIT: The behaviour is different on Akiko rev0. Instead of a guru, I get a "Chunky2Planar routine failed!" but no guru.

I am thinking this might be down to software setup. MULibs make no difference for me, whether they are installed or not.

I even extracted the archive twice to make sure the files weren't corrupt.

[terriblefire]

Im struggling to understand why the firmware would change the Akiko behaviour. Since there were no changes to how the CPU talks to the Motherboard. It was all SDRAM changes.

[alenppc]

I can't imagine many of this akiko functions were ever tested very extensively given the lack of options on the CD32. They are all probably very buggy to begin with.

[nibiru]

Im struggling to understand why the firmware would change the Akiko behaviour. Since there were no changes to how the CPU talks to the Motherboard. It was all SDRAM changes.

It seems in all cases (Fusion, WingCommander and DoomAttack) Akiko is not detected, even by the OS. DoomAttack gives ~5fps in RTG mode (and no corruption of gfx, even with the cache on. Meaning that the even the OS is falling back on the CPU version of WritePixelArray8()).

I agree with alenppc that Commodore just added the Akiko as they had some gates left on the chip. Even the OS fails to use it correctly on a 68030 (I could see gfx corruption with Doom played in a Wb window - v1 firmware).

Looking here it seems that Akiko is not improving much either on a SX-PRO32:

From what I recall by the few minutes I've been playing with the V1 firmware yesterday, WC was running slightly less better than the improved C2P patch.

So not really the end of the world if the Akiko C2P doesn't work anymore. It would probably still help a bit with games (or the Mac emulation) where the CPU spends 95% of its time in logic.

I guess fixing this now it's more for the sake of knowledge than for a good reason .

Also, I got ~5.3 meg/sec with the scsi.device from OS3.9, so I think I'll stick with the v2 firmware for now .

Although If anyone with a V1 Firmware wants to do some benchmarks, here are a couple of links:

WC Enhanced: http://eab.abime.net/showthread.php?t=92982
DoomAttack CD32: http://eab.abime.net/showthread.php?t=88837

Cheers.

[nibiru]

I know I had them running at some point but I can't get them to work anymore. I know this is all academic since they were way slower than the 020 routines but I tested just for the sake of it.

Actually I slightly disagree with this, on the TF328 Akiko almost doubles the FPS in WC. It would probably still be of great use on a 68030@25 Mhz, I need benchmark to prove it would help on a 68030@50. Apparently it does - it still seems to be faster than a 68040 according to this post:

http://eab.abime.net/showpost.php?p=131 ... stcount=63

I'd say that is probably the same to use Akiko or CPU+Blitter on a 68030@50Mhz (as someone else says there).

With the V2 firmware, I am getting ~9-16fps in DoomAttack.

[alenppc]

I have no clue why it won't work on any of my consoles anymore, but I am using the Aminet version of DoomAttack and I have Doom2 wad files.

The 020_optimized routine literally flies, ADoom is a bit slower.

[8 Bit Dreams]

is there any place to download Doom Attack from? can't find it on EAB...

[alenppc]

is there any place to download Doom Attack from? can't find it on EAB...

It's on Aminet The problem though is finding full version compatible wad files for DOOM1. If anyone can help let me know.

A few findings from tonight: installing MMULib v46 from aminet fixes the akiko caching issue. So effectively this forces the 68030.library to be loaded which fixes the 030 data cache bug.

I was finally able to initialize an akiko screenmode but it only worked once out of 10 times I tried to run it (firmware v1) haven't tried v2 yet.

Some stats
Doom shareware wad run as "doomattack.020 -timedemo demo1 -forcedemo" with 020_optimized c2p gives me 7.54 FPS (firmware v1) and 8.83 fps (v2)
Remember the time demo renders every frame so it's a lot slower than actual gameplay.