Homebrew Homebrew Development

[nic0lette]

Dumb question -- is there anything like the NetLoader/3dslink pair from the Homebrew Launcher but runs as a CIA (from EmuNAND or A9LHax...NAND)

 

[daxtsu]

Dumb question -- is there anything like the NetLoader/3dslink pair from the Homebrew Launcher but runs as a CIA (from EmuNAND or A9LHax...NAND)

You mean for developing homebrew? Probably the closest is this: https://github.com/Cpasjuste/libctrshell

 

[nic0lette]

You mean for developing homebrew? Probably the closest is this: https://github.com/Cpasjuste/libctrshell

For developing homebrew, yes, but basically something to load the code over the network like one can do with Netloader/3dslink. Network printfs sound neat, but loading code without yanking my sdcard out constantly is more interesting to me XD

 

[daxtsu]

For developing homebrew, yes, but basically something to load the code over the network like one can do with Netloader/3dslink. Network printfs sound neat, but loading code without yanking my sdcard out constantly is more interesting to me XD

If I'm not mistaken, that library has code to push new CIAs over the currently running one for testing purposes.

 

[nic0lette]

If I'm not mistaken, that library has code to push new CIAs over the currently running one for testing purposes.

Interesting. Thanks for the tip!

 

[TarableCode]

I just did a quick test and it's possible to send network printfs over udp broadcast messages and read them using netcat. (at least on windows)
That would be great since you wouldn't need to hardcode an IP to listen to debug info.

Weird thing is that netcat required the -w 1 parameter or no more messages would be received until netcat was reopened.
I'll have to fiddle with my socket code more and see what's going on.

Beyond that I could hook it into devoptab so it could be hooked into stdio and written to using regular printfs.


edit:
It totally works shame the code is a messy pile though.

Simple proof of concept if anyone needs, it should probably be rewritten at some point.

#include <3ds.h>
#include <stdio.h>
#include <string.h>
#include <malloc.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <sys/iosupport.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <netdb.h>
#include <fcntl.h>

/* Correct buffer size? */
#define SOCBUF_SIZE 128 * 1024

/*
 * If set, this will also write to the default console device as set up by libctru.
*/
#define ALSO_PRINT_CONSOLE

/*
 * If set, sets the socket to non-blocking I/O mode.
 * I have no idea if this matters or not.
 */
#define DONT_BLOCK

/*
 * con_write From libctru.
 * Are we even supposed to be able to access this?
 */
extern ssize_t con_write(struct _reent *r,int fd,const char *ptr, size_t len);

static int DebugMsgSocket = -1;
static void* SOCBuffer = NULL;
static struct sockaddr_in ServerAddr;

/*
 * Our device descriptor. 
 * Maybe one day implement stdin over UDP?
 */
static const devoptab_t MyDevOpTab = {
	"con",
	0,
	NULL,
	NULL,
	Netcon_Write,
	NULL,
	NULL,
	NULL
};

static ssize_t Netcon_Write( struct _reent* r, int fd, const char* ptr, size_t len ) {
	ssize_t Count = 0;

#ifdef ALSO_PRINT_CONSOLE
	/*
	 * Only care about the number of characters written to the console.
	 */
	Count = con_write( r, fd, ptr, len );
	sendto( DebugMsgSocket, ptr, len, 0, ( const struct sockaddr* ) &ServerAddr, sizeof( struct sockaddr_in ) );
#else
	Count = sendto( DebugMsgSocket, ptr, len, 0, ( const struct sockaddr* ) &ServerAddr, sizeof( struct sockaddr_in ) );
#endif

	return Count;
}

/*
 * TODO:
 * This is awful.
 * Just awful.
 */
int netprintf_Init( int Port ) {
	int Flags = 0;
	u32 Status = 0;

	if ( acInit( ) == 0 ) {
		ACU_GetWifiStatus( &Status );
		
		if ( Status > 0 ) {
			SOCBuffer = memalign( 0x1000, SOCBUF_SIZE );

			if ( SOCBuffer ) {
				if ( socInit( ( u32* ) SOCBuffer, SOCBUF_SIZE ) == 0 ) {
					DebugMsgSocket = socket( AF_INET, SOCK_DGRAM, 0 );

					if ( DebugMsgSocket != -1 ) {
						memset( &ServerAddr, 0, sizeof( ServerAddr ) );

	#ifdef DONT_BLOCK
						Flags = fcntl( DebugMsgSocket, F_GETFL, 0 );
						Flags|= O_NONBLOCK;

						fcntl( DebugMsgSocket, F_SETFL, Flags );
	#else
						/* Keep compiler from bitchin' */
						Flags++;
	#endif

						ServerAddr.sin_family = AF_INET;
						ServerAddr.sin_addr.s_addr = INADDR_BROADCAST;
						ServerAddr.sin_port = htons( Port );

						devoptab_list[ STD_OUT ] = &MyDevOpTab;
						devoptab_list[ STD_ERR ] = &MyDevOpTab;

						setvbuf( stdout, NULL, _IONBF, 0 );
						setvbuf( stderr, NULL, _IONBF, 0 );

						return 1;
					}
				} 

				linearFree( SOCBuffer );
			}
		}
	}

	return 0;
}

/*
 * same
 */
void netprintf_Close( void ) {
	if ( DebugMsgSocket )
		closesocket( DebugMsgSocket );

	socExit( );
	acExit( );

	if ( SOCBuffer )
		free( SOCBuffer );
}

 

[YugamiSekai]

Where would future homebrew devs start (after learning C/C++)?

 

[elhobbs]

Where would future homebrew devs start (after learning C/C++)?

Setup toolchains and libraries. Compile/run examples. Edit examples. Then either contribute to an existing project or start something new.

 

[YugamiSekai]

Anyone know if @xem will update this: https://github.com/xem/3DShomebrew/wiki

 

[MRJPGames]

Anyone know if @xem will update this: https://github.com/xem/3DShomebrew/wiki

Most likely not. I did at the time contribute to making it and have learned lot's more since if you have any questions just ask them here, I'm sure I or someone else will be willing to help you out!

 

[DiscostewSM]

Is it possible to use texture combiners to cull out a particular color?

 

[elhobbs]

Is it possible to use texture combiners to cull out a particular color?

You could use the constant texenv color to set one channel to 0 and interpolate or modulate ops. That would remove all red for instance - maybe not what you where looking for though.

 

[TarableCode]

Indexed->RGB conversion is slow as balls but I wondered if anyone had figured out some kind of witchcraft to speed it up or hax around it?
Only updating dirty regions is the best solution so far but a full screen update on the o3ds is still not great.

 

[MasterFeizz]

Indexed->RGB conversion is slow as balls but I wondered if anyone had figured out some kind of witchcraft to speed it up or hax around it?
Only updating dirty regions is the best solution so far but a full screen update on the o3ds is still not great.

Double buffer and offload it to the system core. From doing that I saved 5ms but increased complexity, so I reverted back to the original code.

 

[DiscostewSM]

You could use the constant texenv color to set one channel to 0 and interpolate or modulate ops. That would remove all red for instance - maybe not what you where looking for though.

Nah, not a channel, but a particular color of my choosing. We're coming up with a solution for possible 8-bit paletted textures using the GPU_L8 format to get 255 adjustable colors with reflective color LUTs. It works from tests, but the problem is the remaining entry. From what we understand, it's basically an invalid entry, and seems to always display pure black no matter what is set in the LUTs. We'd like to use it as a transparency index. Dunno if internally when it's processed that it registers as the color set in the LUTs, but if it is, and we can match that up with some color to discard it, then that would help. We'd basically limit the LUTs to 15-bit color combinations with the exception of this one LUT entry, so we can cull it out and not cull anything else.

We do have the option of using GPU_LA8, and force the alpha to full on everything except for that entry, but it needs to be tested, and it may not even work if the invalid entry defaults to a solid black color. I'd rather not use it (if it works) because it requires double the texture space vs GPU_L8.

 

[TarableCode]

Double buffer and offload it to the system core. From doing that I saved 5ms but increased complexity, so I reverted back to the original code.

What were your times like on o3ds on the syscore?
From what I read we only have ~80MHz worth of CPU time there and I found on my end a 512x384 framebuffer took ~10ms to complete on the main thread.

Curse that damn pica200 for not having any indexed modes.
-_-

 

[MasterFeizz]

What were your times like on o3ds on the syscore?
From what I read we only have ~80MHz worth of CPU time there and I found on my end a 512x384 framebuffer took ~10ms to complete on the main thread.

Curse that damn pica200 for not having any indexed modes.
-_-

I didn't really time it, but it was enough for 60fps , but there was no scaling involved. Are you using the GPU for scaling?

 

[TarableCode]

Yeah, I'll take free linear scaling any day
The texture upload is nothing, it's the indexed->rgb that seems to be the slowest part, unless I've been timing it wrong?

I've been using osGetTime before and after the framebuffer conversions and getting the difference.

 

[MasterFeizz]

Yeah, I'll take free linear scaling any day
The texture upload is nothing, it's the indexed->rgb that seems to be the slowest part, unless I've been timing it wrong?

I've been using osGetTime before and after the framebuffer conversions and getting the difference.

Thats how I timed it, simple and accurate enough.

 

[TheCruel]

Indexed->RGB conversion is slow as balls but I wondered if anyone had figured out some kind of witchcraft to speed it up or hax around it?
Only updating dirty regions is the best solution so far but a full screen update on the o3ds is still not great.

The GPU supports procedural textures, which includes a color LUT, but it's not completely documented and I don't know of anyone who has successfully achieved it. I will be toying with it this week though for usage in ScummVM.

Some data on it:
https://www.3dbrew.org/wiki/GPU/Procedural_Texture_Generation
https://www.3dbrew.org/wiki/GPU/Internal_Registers#GPUREG_PROCTEX_LUT