Page 1 of 1
DMA failure on different consoles.
Posted: Mon Oct 31, 2016 8:12 pm
by Jpg3D
Hi, after months without being able to work on my project, I'm programming my rom again. I'll try to explain the best i can the problem i have.
I hope any of you can help me to fix this error.
The pistol's tiles are DMAed to VRAM. The text tiles are filled to the VRAM by writting them using data port using a MOVE loop.
I tried to transfer the pistol's tiles using DMA and tested the rom on the consoles that i got access.
This is a pic testing the rom on my own console. It works fine and i dont get any glitches. Tested on MegaEverdrive.
This is what i get on my friend's console. Tested on Everdrive.
This is the fragment of my code that i use to fill the VRAM using DMA
Code: Select all
DMATOVRAM equ $40000080
******** Entrada: D0 number of bytes/words
******** D1 = 3 bytes DMA source
******** D2 = 2 bytes DMA destination
DMAGEN:
MOVE.W #$8F02,$00C00004 * increment 2 bytes
MOVE.W #$8174,$00C00004 activate dma
MOVE.W #$9300,D4 * Low Counter
OR.B D0,D4
MOVE.W D4,$00C00004
ASR.W #8,D0
MOVE.W #$9400,D4 * High Counter
OR.B D0,D4
MOVE.W D4,$00C00004
LSR.L #1,D1
MOVE.W #$9500,D4 * Low source adress
OR.B D1,D4
LSR.L #8,D1
MOVE.W D4,$00C00004
MOVE.W #$9600,D4 * MID source adress
OR.B D1,D4
LSR.L #8,D1
MOVE.W D4,$00C00004
MOVE.W #$9700,D4 * High source adress
AND.B #%01111111,D1
OR.B D1,D4
MOVE.W D4,$00C00004
ASL.L #2,D2 * Destination address and operation
LSR.W #2,D2
SWAP D2
OR.L #DMATOVRAM,D2
MOVE.L D2,$00C00004
RTS
I dont have any idea why on different consoles the same code results on a different behaviour.
What can be causing this DMA failure?
Really thank you for your help.
Re: DMA failure on different consoles.
Posted: Mon Oct 31, 2016 10:08 pm
by Mask of Destiny
What model and revision do you and your friend have?
Re: DMA failure on different consoles.
Posted: Mon Oct 31, 2016 10:34 pm
by Jpg3D
Mine is :Megadrive 1 PAL-B(VE) 1601-07 <-- Works fine
Friend's :Megadrive 1 PAL-G 1601-18 <-- Glitchy DMA
Re: DMA failure on different consoles.
Posted: Tue Nov 01, 2016 11:14 am
by Stef
It looks like a memory timing issue... Sometime small voltage difference can make the everdrive memory behaves differently. For instance the everdrive flash speed cannot be the same depending on which MD system i'm using it, i have to test on my "slowest" MD to ensure it works everywhere.
Still for me it only happened during flashing and never during in game memory access condition.
Also you used MegaEverdrive on your system while using simple Everdrive on your friend system. MegaEverdrive has some extra hardware allowing extended ROM area... maybe it also affects positively your code in some way.
Do you have a sample test rom so we can eventually test on our own system ? I've a PAL and JAP MD1 with an everdrive...
Re: DMA failure on different consoles.
Posted: Tue Nov 01, 2016 12:33 pm
by Jpg3D
Here's the rom
https://ufile.io/56a7
I'm pretty sure that is my fault. Probably i'm missing something on my DMA function that causes this failure on different cartidges, because other homebrew works fine on myfriends MegaDrive.
Also i tested it on another system.
MegaDrive 2 -> MK-1631-50 <-- Works Perfect with my MegaEverdrive
Re: DMA failure on different consoles.
Posted: Tue Nov 01, 2016 2:46 pm
by Natsumi
Sanity check; disable ints. Always disable ints when writing to VDP unless you can be sure there is no possibility of ints happening (like during VInt with no HInts). Or better yet, write better DMA code. This one here is terribly slow, and increases likelihood of ints happening during it. The best case is just using raw values with move.l's and one move.w, to (aN). This is generally what actual games do too.
Re: DMA failure on different consoles.
Posted: Tue Nov 01, 2016 6:55 pm
by Sik
Actually the best case is to store the commands in a queue and then send the entire queue to the VDP when vblank starts (not talking about the vblank interrupt, just the main loop), this gives you the most optimal bandwidth usage and ensures DMAs won't clash with interrupts unless you make too many of them. It also means you can safely spend more time building the DMA command since the only critical part is when the queue is being sent to the VDP.
On the flipside, it also means you need to keep any data to be transferred around until the next frame. So don't use this for data put in local variables since they stop existing once you return from the function.
Re: DMA failure on different consoles.
Posted: Tue Nov 01, 2016 8:46 pm
by Stef
The rom size is 21 KB, i strongly suggest you to align its size to 128 KB boundary... That may explain the difference you observe between the MegaEverdrive and the Everedrive.
Re: DMA failure on different consoles.
Posted: Wed Nov 02, 2016 4:31 pm
by TmEE co.(TM)
ED requires 128KB alignment to work properly, or at least used to require the alignment.
Re: DMA failure on different consoles.
Posted: Wed Nov 02, 2016 6:48 pm
by HardWareMan
Tested with my system. Pistol appears correctly but very short time. About 0.5 sec. Everything else like on emulator. MD1 and my own flash cart.
Re: DMA failure on different consoles.
Posted: Wed Nov 02, 2016 6:51 pm
by Jpg3D
I aligned the size to exactly 131.072 bytes (128kbs), blocked interrupts, and rewritted the DMA function so now it's much faster. It's almost a copy from the DMA function that you can find at this website.
http://www.hacking-cult.org/?r/18/25
Unfortunately the problem persists. 100% the same behaviour as i menctioned before. Perfect functionality on my megadrive and exactly the same glitchy stuff on my friend's one. The code runs right as expected when i remove the calls to DMA functions so im absolutely sure the problem is that im doing something wrong when implementing the DMA function.
Code: Select all
DMA2VM MACRO
MOVE.W #$9400+(((LENGTH)>>9)&$FF),($C00004).l
MOVE.W #$9300+(((LENGTH)>>1)&$FF),($C00004).l
MOVE.W #$9600+(((SOURCE>>1)&$FF00)>>8),($C00004).l
MOVE.W #$9500+((SOURCE>>1)&$FF),($C00004).l
MOVE.W #$9700+((((SOURCE>>1)&$FF0000)>>16)&$7F),($C00004).l
MOVE.L #($80+((DESTINATION&$3FFF)<<16)+((DESTINATION&$4000)>>14)+((%001&$4)<<2)+((%001&$3)<<30)),($C00004).l
ENDM
At this point i dont have any idea what can be causing this bug.
Re: DMA failure on different consoles.
Posted: Fri Nov 04, 2016 2:01 am
by Natsumi
Now, this may be completely crazy suggestion, but try clearing the RAM after init but before any other DMA. Don't bother making it efficient, you just want to waste time. This fixed major issues with a homebrew I wrote earlier. Without this, the graphics wouldn't look right, or just nothing loaded. Adding it and I never had issues since.
Re: DMA failure on different consoles.
Posted: Fri Nov 11, 2016 6:26 pm
by Jpg3D
Thank you all for you help. It really helped me to think back to my first code and fix it.
Finally the error was fixed after rewritten A LOT of ugly n00bish code. Probably there were something wrong at my very beginning lines because i had not really good idea of how to program this machine. I fixed many bugs so i dont know really well where my code was fixed. After a lot of tests it's working 100% as i want.
<3