vidix.txt

linux下的MPEG1

		VIDIX - VIDeo Interface for *niX
		~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
		
This interface was designed and introduced as interface to userspace drivers
to provide DGA everywhere where it's possible (unline X11).
I hope that these drivers will be portable same as X11 (not only on *nix).

What is it:
- It's portable successor of mga_vid technology which is located in user-space.
- Unlikely X11 it's provides DGA everywhere where it's possible.
- Unlikely v4l it provides interface for video playback
- Unlikely linux's drivers it uses mathematics library.

Why it was developed:
As said Vladimir Dergachev 
(http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/gatos/km/km.rfc.txt):
"0) Motivation
    v4l, v4l2 and Xv are all suffering from the same problem: attempt to fit 
    existing multimedia devices into a fixed scheme."
Well - I tried to implement something similar by motivation.

How it works:
~~~~~~~~~~~~~

This interface is almost finished. But I guess it can be expanded by developer's
requests.
So any suggestions, reports, criticism are gladly accepted.

1) APP calls vixGetVersion to check age of driver ;)
2) APP calls vixProbe. Driver should return 0 if it can handle something in PC.
3) APP calls vixGetCapability. Driver should return filled
    vidix_capability_t.type field at least.
4) If above calls were succesful then APP calls vixInit function
   (Driver can have not exported this function in this case call will be
    skiped).
5) After initializing of driver APP calls vixGetCapability again
   (In this case driver must fill every field of struct)
6) APP calls vixQueryFourcc. Driver should answer - can it configure
   video memory for given fourcc or not.
7) APP calls vixConfigPlayback. Driver should prepare BES on this call.
   APP pass to driver following info:
   ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   vidix_playback_t.fourcc	- contains fourcc of movie
   vidix_playback_t.capability	- currently contsinas copy of vidix_capability_t.flags
   vidix_playback_t.blend_factor- currently unused   
   vidix_playback_t.src		- x,y,w,h fields contain original movie size
				  (in pixels) x and y often are nulls.
   vidix_playback_t.src.pitch.y   These fields contain source pitches
   vidix_playback_t.src.pitch.u - for each Y,U,V plane in bytes.
   vidix_playback_t.src.pitch.v   (For packed fourcc only Y value is used)
				  They are hints for driver to use same destinition
				  pitches as in source memory (to speed up
				  memcpy process).
				  Note: when source pitches are unknown or
				  variable these field will be filled into 0.
   vidix_playback_t.dest	- x,y,w,h fields contains destinition rectange
				  on the screen in pixels.
   vidix_playback_t.num_frames  - maximal # of frames which can be used by APP.
				  (Currently 10).
   Driver should fill following fields:
   ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   vidix_playback_t.num_frames  - real # of frames which will be used by driver.
				  (Should be less or equal to app's num_frames).
				  
   vidix_playback_t.dest.pitch.y   These fields should contain alignment
   vidix_playback_t.dest.pitch.u - for each Y,U,V plane in bytes.
   vidix_playback_t.dest.pitch.v   (For packed fourcc only Y value is used)
   
   vidix_playback_t.frame_size	- Driver should tell to app which size of
				      source frame (src.w and src.h) should
				      use APP (according to pitches and offsets)

   vidix_playback_t.offsets	- offsets from begin of BES memory for each frame
   
   vidix_playback_t.offset.y	  These field should contain offset
   vidix_playback_t.offset.u	- for each Y,U,V plane within frame.
   vidix_playback_t.offset.v	  (For packed fourcc only Y value is used)

   vidix_playback_t.dga_addr	- Address of BES memory.
   
Also see this picture:

VIDEO MEMORY layout:
 +-----------  It's begin of video memory     End of video memory--------------+
 |                                                                             |
 v                                                                             v
 [      RGB memory                         |         YUV memory    |  UNDEF    ]
					   ^
					   |
					   +---- begin of BES memory

BES MEMORY layout:
 +-------- begin of BES memory
 |
 v
 [ | |                      |       |       |
   ^ ^                      ^       ^       ^
   | |                      |       |       + BEGIN of second frame
   | |                      |       + BEGIN of V plane
   | |                      + BEGIN of U plane
   | +------- BEGIN of Y plane
   |
   +--------- BEGIN of first frame

This means that in general case:
offset of frame != offset of BES
offset of Y plane != offset of first frame

But often: vidix_playback_t.offsets[0] = vidix_playback_t.offset.y = 0;

Formula: (For Y plane) copy source to:
		  vidix_playback_t.dga_addr +
		  vidix_playback_t.offsets[i] +
		  vidix_playback_t.offset.y

8) APP calls vixPlaybackOn. Driver should activate BES on this call.
9) PLAYBACK. Driver should sleep here ;)
   But during playback can be called:
     vixFrameSelect (if this function is exported)
	Driver should prepare and activate corresponded frame.
	This function is used only for double and trilpe buffering and
	never used for single buffering playback.
     vixGet(Set)GrKeys (if this function is exported)
	This interface should be tuned but intriduced for overlapped playback
	and video effects (TYPE_FX)
     vixPlaybackGet(Set)Eq (if this function is exported)
	For color correction.
10) APP calls vixPlaybackOff. Driver should deactivate BES on this call.
11) If vixDestroy is defined APP calls this function before unloading driver
    from memory.
    
    
What functions are mandatory:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
vixGetVersion
vixProbe
vixGetCapability
vixQueryFourcc
vixConfigPlayback
vixPlaybackOn
vixPlaybackOff

All other functions are optionaly.

BUSMASTERING
************

Busmastering is technique to implement data transfer through DMA.
This technique is intended to free CPU for other useful work to
speedup movie playback. The speedup will be different on different
CPUs OSes and videocards. Only thing which SHOULD be implemented
it's the fact that frame should be transfered faster than 1/fps.
(I.e. faster than 33ms for 30 fps or faster than 40ms for 25 fps)
VIDIX implementation of BM (busmastering) is slightly specific.
During driver development you should keep in mind the next rules:
1. BM is implemented as parallel process which should work
   simultaneously with frame decoding.
2. To have possibility to use busmastering by non-ROOT users
   driver should rather call functions from libdha than from libc.
   (Example: driver should call bm_lock_mem instead of mlock)
3. To speedup data transfer player will pass pointer to the DMA buffer
   which will have the same structure (planes and strides) as video memory
   (In this connexion driver should allocate frames in video memory
   same as if BM would not be implemented).

Interface:
~~~~~~~~~~

The interface of BM is implemented through 2 functions:
  vixPlaybackCopyFrame
  vixQueryDMAStatus


vixPlaybackCopyFrame 

should prepare engine to copy frame from
system memory into video framebuffer. After that driver should
send command into engine to start data transfer and return
control immediatedly.

The structure vidix_dma_s in details:

typedef struct vidix_dma_s
{
	/* 
	app -> driver. 
	Virtual address of source.
	Note: source buffer is allocated by using malloc
	or memalign();
	*/
	void *		src;
	/* 
	app -> driver. 
	Destinition offset within of video memory.
	It will point offset within of YUV memory where
	destinition data should be stored.
	*/
	unsigned 	dest_offset;
	/* app -> driver. Size of data to be transfered in bytes. */
	unsigned 	size;
	/*
	can accept ORed values of BM_DMA* definitions
	BM_DMA_ASYNC -	default value which indicates that transactiion
			should work asynchronously.
	BM_DMA_SYNC - 	may be ignored due speedup reasons
	BM_DMA_FIXED_BUFFS - indicates that player was started by ROOT
			and source DMA buffers were already locked in memory
			through mlock().
	/* app -> driver: idx of src buffer.
		    if BM_DMA_FIXED_BUFFS flags is set then this field
		    indicates which from buffers currently is passed
		    into driver. This field maybe ignored by driver but
		    it would be better to use that for minor speedup
		    of engine preparing. */
	unsigned 	idx;
	/* for internal use by driver.
	   Driver may use them on its opinion  */
	void *		internal[VID_PLAY_MAXFRAMES];
}vidix_dma_t;


vixQueryDMAStatus

should check out DMA status and return 1 if BM is busy
and 0 otherwise. Note: this function shouldn't wait any
changes in DMA state.

A few words about of non-linux systems
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Well, there is only one problem which stops us to use BM on
nono-linux systems: it's lacking of possibility to perform
convertion from virtual to physical address in user-space.
This problem is sloved by so-called dhahelper driver for
linux. What about of other OSes then this driver requires
to be ported first. (Of course, except of DOS and DOS32
where these convertions are unnecessary).

Useful links:
~~~~~~~~~~~~~
Guide to DTV		http://www.digitaltelevision.com/dtvbook/toc.shtml
Fourcc			http://www.webartz.com/fourcc/
MPEG			http://www.mpeg.org/MPEG/index.html
Analog colors		http://www.miranda.com/en/app_notes/TN/TN-05/TN-05.htm

Please send your suggestions, reports, feedback to [e-mail]

Best regards! Nick Kurshev.