GP2X Dangerous Linux Hacking Code

[Franxis]

how much of a speed improvement are we talking about?

I haven't measured it exactly but it is visually noticeable and easy to add to your code... It is only needed to call the MMU hacking table function after all mmap() done in your application. In Rlyeh minimal library it is needed to add the call at the end of the minimal library initialization function.

In MAME memory access to
- Video buffers.
- Extended memory (when loading biggest NeoGeo games).
changes from not cached to cached memory.

B)

 

[synkro]

is this already summed up in the wiki?

 

[pepone]

Damn, after getting it correctly working, it give me a boost of 10 to 20fps in gngeo :blink:

 

[dockthepod]

Damn, after getting it correctly working, it give me a boost of 10 to 20fps in gngeo :blink:

 

[pepone]

Damn, after getting it correctly working, it give me a boost of 10 to 20fps in gngeo :blink:

But...
It work very well when I launch gngeo from a telnet, but it crash very often when I use Selector or Rage2x

 

[Vimacs]

10-20fps? OMFG!
I wouldnt ever have guessed it would do so much of a difference.
I hope you get the issues sorted out.

 

[pepone]

10-20fps? OMFG!
I wouldnt ever have guessed it would do so much of a difference.

Yes, I'm really surprised too. it seems that blitting a software surface to the screen have big penalty.
But to be useful, I have to make it more stable.

 

[DaveC]

Damn, after getting it correctly working, it give me a boost of 10 to 20fps in gngeo :blink:

10-20 fps? holy shit!

Sounds awesome..

 

[yaustar]

Isnt that almost 30% speed increase?

 

[gaterooze]

You hackin' dudes are crazy-cool B)

 

[pepone]

I think I've got it working correctly, without crash.
What I've do is to put a 1sec pause at a specefic point:

lseek (gp2x_dev_mem, 0x6ec00, SEEK_SET);
write (gp2x_dev_mem, &newc1, 4);
write (gp2x_dev_mem, &newc2, 4);	

SDL_Delay(1000);

ttb = myuname(name);

I didn't get any crash with that

a beta is available here:
http://m.peponas.free.fr/gngeo/download/gngeo2x_0.6.12c.zip

PS: initial loading can be a little long (3/5 sec) and if the mmu patch fail, then you will no get no performance gain.

 

[Franxis]

I think I've got it working correctly, without crash.
What I've do is to put a 1sec pause at a specefic point:

lseek (gp2x_dev_mem, 0x6ec00, SEEK_SET);
write (gp2x_dev_mem, &newc1, 4);
write (gp2x_dev_mem, &newc2, 4);	

SDL_Delay(1000);

ttb = myuname(name);

I didn't get any crash with that

a beta is available here:
http://m.peponas.free.fr/gngeo/download/gngeo2x_0.6.12c.zip

PS: initial loading can be a little long (3/5 sec) and if the mmu patch fail, then you will no get no performance gain.

Great! The initial loading will be a lot faster if you comment the trace to the terminal (//printf()...).

 

[Franxis]

I'm having troubles patching the mmu table. Sometimes the read() in DecodeCoarse is ok and sometimes bad...

Still investigating... :unsure:

 

[pepone]

I'm having troubles patching the mmu table. Sometimes the read() in DecodeCoarse is ok and sometimes bad...

Still investigating... :unsure:

try the delay trick I use (just before ttb=myuname), it's seems that there is some sort of cache that prevent the uname routine to be correctly overwrited. I tested it with a delay of 200ms and it work great so far.

 

[Waninkoko]

Wow! 15 fps more with this hack!
CONGRATULATIONS SQUIDGE

I don't know if I have used this hack correctly. I have made a function that's executed only one time after SDL is initialised.

void mmuhack()
{
				initphys();
		volatile unsigned int *myBuf = trymmap((void *)0, 65536, 3, 1, memfd, 0x03000000);
		volatile unsigned int *secbuf = (unsigned int *)malloc(204800);

		printf("mmaped 0x03000000 buffer @ VA: %08X malloc'd buffer @ VA: %08X\n", myBuf, secbuf);

		hackpgtable();

It's this correct?
And yes, I can improve it a bit D

PD: I'm using MMap function written in ASM by Dzz.

 

[rlyeh]

What I've do is to put a 1sec pause at a specefic point:

lseek (gp2x_dev_mem, 0x6ec00, SEEK_SET);
write (gp2x_dev_mem, &newc1, 4);
write (gp2x_dev_mem, &newc2, 4);	

SDL_Delay(200);

ttb = myuname(name);

Will this work too? I cant test it right now, im not at home

volatile u32 temp1,temp2;

lseek (gp2x_dev_mem, 0x6ec00, SEEK_SET);
read (gp2x_dev_mem, &temp1, 4);
read (gp2x_dev_mem, &temp2, 4);

lseek (gp2x_dev_mem, 0x6ec00, SEEK_SET);
write (gp2x_dev_mem, &newc1, 4);
write (gp2x_dev_mem, &newc2, 4);	

while(temp1 != newc1 && temp2 != newc2);

ttb = myuname(name);

PS: Squidge, amazing work as usual Congrats!

 

[Squidge]

rlyeh: I don't think that'll work. I believe the problem to be something to do with caching, and Linux seems to do a task swap every 50ms, so you can either wait 100ms to be on the safe side, or flush the caches yourself using one of the special swi calls in the 0x9f000 range - check the kernel source for the exact number and the parameters. Of the two, I'd say waiting 100ms or so is the easiest

 

[Franxis]

I surrender... It runs randomly for me...

 

[pepone]

I surrender... It runs randomly for me...

Something else I did to get it to work: I've put the myuname function in a separate asm file. It seems that with -02 or -03, gcc does some weird thing with it, and then myuname doesn't return the correct ttb.

 

[Drag]

Good work, Squidge!

Unfortunately, I have pretty much 0 experience with anything outside of game programming with SDL, so I have no idea how to implement this into an SDL program. Would Waninkoko's solution work for me?