Media Engine?

crazyc · Post by **crazyc** » Fri Jul 15, 2005 7:13 am

Some people might want to use the codecs in mebooter but I'd rather use the Media Engine as an extra CPU. To that end i've written a crappy sample that seems to work that runs a counter on the me. If anyone is interested i'll post it here.

ector · Post by **ector** » Fri Jul 15, 2005 7:31 am

Yes indeed, would be interesting. Does it have access to main memory?

crazyc · Post by **crazyc** » Fri Jul 15, 2005 8:02 am

ector wrote:Yes indeed, would be interesting. Does it have access to main memory?

I havn't investigated directly, but from looking at mebooter, it's local ram appears to be at 0x88300000 and visible from both CPU's. I'm not sure if it can see the main CPU's ram but I suspect it can and wouldn't be hard to check. Anyway, to the code.

Code: Select all

#include <psptypes.h>
#include <string.h>
#include <pspkernel.h>
#include <pspdebug.h>

void *me_run&#40;&#41;;
void *me_end&#40;&#41;;
void &#40;*sceSysregMeResetEnable&#41;&#40;&#41;;
void &#40;*sceSysregMeResetDisable&#41;&#40;&#41;;
void &#40;*sceSysreg_unk1&#41;&#40;&#41;;
void &#40;*sceSysregVmeResetDisable&#41;&#40;&#41;;

PSP_MODULE_INFO&#40;"ME_test", 0x1000, 1, 1&#41;;
PSP_MAIN_THREAD_ATTR&#40;0&#41;;

int main&#40;int param1, int param2&#41;
&#123;
	SceModuleInfo *sysreg_driver;
	pspDebugInstallErrorHandler&#40;NULL&#41;;
	pspDebugScreenInit&#40;&#41;;
	pspDebugScreenPrintf&#40;"start\n"&#41;;
	memcpy&#40;&#40;void *&#41;0xbfc00040, me_run, &#40;int&#41;&#40;me_end - me_run&#41;&#41;;
	sysreg_driver = pspDebugFindModule&#40;"sceSYSREG_Driver"&#41;;
	sceSysregMeResetEnable = pspDebugFindExportedFunction&#40;sysreg_driver, "sceSysreg_driver", 0xde59dacb&#41;;
	sceSysregMeResetDisable = pspDebugFindExportedFunction&#40;sysreg_driver, "sceSysreg_driver", 0x2db0eb28&#41;;
	sceSysreg_unk1 = pspDebugFindExportedFunction&#40;sysreg_driver, "sceSysreg_driver", 0x44f6cda7&#41;;
	sceSysregVmeResetDisable =  pspDebugFindExportedFunction&#40;sysreg_driver, "sceSysreg_driver",0x7558064a&#41;; 
	sceKernelDcacheWritebackInvalidateAll&#40;&#41;;
	&#40;*sceSysregMeResetEnable&#41;&#40;&#41;;
	&#40;*sceSysreg_unk1&#41;&#40;&#41;;
	&#40;*sceSysregMeResetDisable&#41;&#40;&#41;;
	&#40;*sceSysregVmeResetDisable&#41;&#40;&#41;;
	
	while&#40;1&#41;
	&#123;
		sceKernelDcacheWritebackInvalidateAll&#40;&#41;;
		pspDebugScreenPrintf&#40;"%08x", *&#40;unsigned int *&#41;0xbfc00060&#41;;	
	&#125;
	sceKernelExitGame&#40;&#41;;
	return 0;
&#125;

Code: Select all

#include <machine/regdef.h>
	.globl me_run
	.globl me_end

me_run&#58;

	li v0, 0xbfc00060
	li v1, 0
a&#58;
	addiu v1, v1, 1
	sw v1, 0&#40;v0&#41;
	b a

me_end&#58;

MDoggie · Post by **MDoggie** » Fri Jul 15, 2005 8:45 am

Nice work! I have a question about the Media Engine that I hope you'd answer for those of us who are still in the shallow waters of PSP programming :) How much, so far, has the Media Engine been implemented in the most popular homebrew programs? (Such as the many versions of Snes9x, NesterJ, etc).

Keep up the great work everyone, and now I will return to my mostly-lurker status, and continue to play catch-up from some of you guys about a half-mile back :)

Post by **mrbrown** » Fri Jul 15, 2005 8:49 am

Um this is the first publically available ME code anyone has seen.

Great work crazyc.

EdwardFMA · Post by **EdwardFMA** » Fri Jul 15, 2005 10:50 am

This is awesome now will this work like a second CPU can the Main Core and Media Core work together to get faster speeds on snes9x and other Games?

crazyc · Post by **crazyc** » Fri Jul 15, 2005 11:22 am

I just stuck this code at the beginning of snes9x to see if anything would break and as for as I can tell nothing did. So it seems we should be able to do just about anything with it.

This is awesome now will this work like a second CPU can the Main Core and Media Core work together to get faster speeds on snes9x and other Games?

Eventually. Some RPC system would have to be developed. Also, there may be a way to use the mebooter code to load homebrew code with the advantage of using the existing libraries. I looked, though, and it seems to me that the mebooter just contains a simple RPC system for just controlling the codecs and the VME.

EdwardFMA · Post by **EdwardFMA** » Fri Jul 15, 2005 1:05 pm

... *Crawls back into corner* ....

lebron23 · Post by **lebron23** » Sat Jul 16, 2005 11:08 am

[DELETED]

Post by **mrbrown** » Sat Jul 16, 2005 12:09 pm

lebron23 wrote:[DELETED]

No software requests.

crazyc · Post by **crazyc** » Mon Jul 18, 2005 2:39 pm

I've written a bootstrap and loader that will load an elf binary into the me address space and jump to the entry. There is a sample program included and a shell script to build it.

http://home.wi.rr.com/crazyz/meloader.tgz

Edit: better

yoyofr · Post by **yoyofr** » Tue Jul 19, 2005 7:51 pm

am I the only one testing this ? ;-)

it seems it cannot compile with latest svn (update a few hours ago).
the debugfindmodule,... functions cannot be found.
same for kprintf sample ....

anyway, if we can get this to work it will be a major boost for lots of
homebrew here (codec, emulator core like snes'apu, ....)...

crazyc · Post by **crazyc** » Wed Jul 20, 2005 1:53 am

yoyofr wrote:am I the only one testing this ? ;-)

At least someone else is interested in it. :)

yoyofr wrote: it seems it cannot compile with latest svn (update a few hours ago).
the debugfindmodule,... functions cannot be found.
same for kprintf sample ....

That is in modmgr.c which was removed for some reason.

Post by **TyRaNiD** » Wed Jul 20, 2005 2:13 am

modmgr.c was removed because you can just create a stub which defines the kernel calls and call the functions directly, no more hackery required. I was going to quickly port acrosss crazyc's small example to pspsdk for people to play with, that would obviously bring the stubs in as well.

yoyofr · Post by **yoyofr** » Wed Jul 20, 2005 2:14 am

ok, thx.
I'll play with it a bit now...

did you try to access main cpu memory from me ?

crazyc · Post by **crazyc** » Wed Jul 20, 2005 12:02 pm

yoyofr wrote:ok, thx.
I'll play with it a bit now...

did you try to access main cpu memory from me ?

I've had problems with this. I'm working on cache flush to hopefully make it easier. Maybe it would be a good idea to make two new libraries in svn, maybe libmeinit for cpu side bootstrap and libme for a utility library and me specific libc functions.

Post by **Warren** » Wed Jul 20, 2005 1:31 pm

I'm very interested in the ME and other processors/accelerators in the PSP but I'm busy with work and getting Net stuff into PSPSDK. I will however work on getting ME stuff into PSPSDK after net stuff is in. I think it would be nice to have some more things done/figured out for ME before we put it in though, especially a good sample.

yoyofr · Post by **yoyofr** » Thu Jul 21, 2005 2:35 am

good.
right now I'm quite busy with psp gu... I'll have a look at me when
it will be in pspsdk then

Shine · Post by **Shine** » Sat Jul 23, 2005 4:20 am

yoyofr wrote:good.
right now I'm quite busy with psp gu... I'll have a look at me when
it will be in pspsdk then

Thanks to TyRaNiD, a very basic test, which compiles with PSPSDK, is in svn now. I've fixed a bug, now the loop runs more than once :-) but more research is needed.

crazyc · Post by **crazyc** » Sat Jul 23, 2005 10:30 am

I've gotten cache flush working and as expected, system ram is mapped at 0x88000000 and is all accessable. Local ram is mapped at 0x80000000. Video ram appears to be inaccessable, at least at the usual address.

Edit: and despite what sony's block diagram shows, the me has an fpu.

Saotome · Post by **Saotome** » Sat Jul 23, 2005 10:22 pm

crazyc wrote:...
Edit: and despite what sony's block diagram shows, the me has an fpu.

could it be possible that both cores share one FPU. or did you check already (write to FPU register from main core, read the same FPU register from ME)?

Post by **mrbrown** » Sat Jul 23, 2005 11:51 pm

Saotome wrote:
crazyc wrote:...
Edit: and despite what sony's block diagram shows, the me has an fpu.
could it be possible that both cores share one FPU. or did you check already (write to FPU register from main core, read the same FPU register from ME)?

You can't share the FPU, it's a coprocessor attached to each CPU individually.

Saotome · Post by **Saotome** » Sun Jul 24, 2005 3:01 am

Ok. I was just wondering, because of the block diagram. But maybe they just wanted to keep the diagram simple ;)

Post by **Warren** » Sun Jul 24, 2005 7:17 am

crazyc wrote:I've gotten cache flush working and as expected, system ram is mapped at 0x88000000 and is all accessable. Local ram is mapped at 0x80000000. Video ram appears to be inaccessable, at least at the usual address.

Edit: and despite what sony's block diagram shows, the me has an fpu.

Please keep the memory map in the wiki up to date as you discover new things.

http://wiki.ps2dev.org/psp:memory_map

crazyc · Post by **crazyc** » Sun Jul 24, 2005 8:17 am

Here's a tarball with a sample to dump the chunks of address space. It also includes the start of libraries for me init and C code running on the me. No psp libraries can be used on the me, due to the lack of kernel functions, except psplibc (not right now though, becuase sbrk is incompatible; until that is resolved, use newlib).

http://home.wi.rr.com/crazyz/melib.tgz

Klimru · Post by **Klimru** » Thu Jul 28, 2005 1:04 am

is this processor fast enough to say, play an mp3? i'm not aware of the specs on it

ector · Post by **ector** » Thu Jul 28, 2005 1:04 am

Easily. It runs the same speed as the main CPU, unless I have misunderstood something.

Klimru · Post by **Klimru** » Thu Jul 28, 2005 1:19 am

maybe i'll try and get libmad to work on it

crazyc · Post by **crazyc** » Thu Jul 28, 2005 1:33 am

It's well suited to things like background music because of the lack of cache coherency. I haven't been able to tell if the audio hardware is directly accessable from it though. Note the sbrk in tarball is broken, here is a correction.

Code: Select all

#include <psptypes.h>
#include <sys/unistd.h>
#include <errno.h>

#define ME_HEAP_BOTTOM &#40;void *&#41;0x80000000
#define ME_HEAP_TOP &#40;void *&#41;0x80200000

extern int _me_stack_size;
static void *_me_heap_ptr = ME_HEAP_BOTTOM;

void *_sbrk&#40;ptrdiff_t incr&#41;
&#123;
	void *new_brk, *prev_brk = _me_heap_ptr;
	if&#40;incr&#41;
	&#123;
		new_brk = _me_heap_ptr + incr;
		if&#40;&#40;new_brk + _me_stack_size&#41; > ME_HEAP_TOP&#41;
		&#123;
			errno = ENOMEM;
			return &#40;void *&#41;-1;
		&#125;
		if&#40;new_brk < ME_HEAP_BOTTOM&#41;
			new_brk = ME_HEAP_BOTTOM;
		_me_heap_ptr = new_brk;
	&#125;
	return prev_brk;
&#125;

Edit: it looks like the exception handler location is set by loading cop0 register 25 (usually perfcnt) with the address of your handler.

Klimru · Post by **Klimru** » Thu Jul 28, 2005 3:27 am

thats pretty much all i want it for is background music... hopefully an mp3. The decoding of the mp3 just kills the processing in the game i'm working on, so, running it on a seperate processor all together would be perfect assuming its fast enough.