zoran

trident tm5600的linux驱动

	Buz tolerable, LML3 almost perfect (occasional single frame drops)
SiS735
	LML33 perfect, Buz tolerable.
VIA KT133(*)
	LML33 starting to get annoying, Buz poor enough that I have up.

Both 440BX boards were dual CPU versions.
--
Bernhard Praschinger later added:
--
AMD 751
	Buz perfect-tolerable
AMD 760
	Buz perfect-tolerable
--
In general, people on the user mailinglist won't give you much of a chance
if you have a VIA-based motherboard. They may be cheap, but sometimes, you'd
rather want to spend some more money on better boards. In general, VIA
mainboard's IDE/PCI performance will also suck badly compared to others.
You'll noticed the DC10+/DC30+ aren't mentioned anywhere in the overview.
Basically, you can assume that if the Buz works, the LML33 will work too. If
the LML33 works, the DC10+/DC30+ will work too. They're most tolerant to
different mainboard chipsets from all of the supported cards.

If you experience timeouts during capture, buy a better mainboard or lower
the quality/buffersize during capture (see 'Concerning buffer sizes, quality,
output size etc.'). If it hangs, there's little we can do as of now. Check
your IRQs and make sure the card has its own interrupts.

===========================

4. Programming interface

This driver conforms to video4linux and video4linux2, both can be used to
use the driver. Since video4linux didn't provide adequate calls to fully
use the cards' features, we've introduced several programming extensions,
which are currently officially accepted in the 2.4.x branch of the kernel.
These extensions are known as the v4l/mjpeg extensions. See zoran.h for
details (structs/ioctls).

Information - video4linux:
http://roadrunner.swansea.linux.org.uk/v4lapi.shtml
Documentation/video4linux/API.html
/usr/include/linux/videodev.h

Information - video4linux/mjpeg extensions:
./zoran.h
(also see below)

Information - video4linux2:
http://linuxtv.org
http://v4l2spec.bytesex.org/
/usr/include/linux/videodev2.h

More information on the video4linux/mjpeg extensions, by Serguei
Miridonovi and Rainer Johanni:
--
The ioctls for that interface are as follows:

BUZIOC_G_PARAMS
BUZIOC_S_PARAMS

Get and set the parameters of the buz. The user should always do a
BUZIOC_G_PARAMS (with a struct buz_params) to obtain the default
settings, change what he likes and then make a BUZIOC_S_PARAMS call.

BUZIOC_REQBUFS

Before being able to capture/playback, the user has to request
the buffers he is wanting to use. Fill the structure
zoran_requestbuffers with the size (recommended: 256*1024) and
the number (recommended 32 up to 256). There are no such restrictions
as for the Video for Linux buffers, you should LEAVE SUFFICIENT
MEMORY for your system however, else strange things will happen ....
On return, the zoran_requestbuffers structure contains number and
size of the actually allocated buffers.
You should use these numbers for doing a mmap of the buffers
into the user space.
The BUZIOC_REQBUFS ioctl also makes it happen, that the next mmap
maps the MJPEG buffer instead of the V4L buffers.

BUZIOC_QBUF_CAPT
BUZIOC_QBUF_PLAY

Queue a buffer for capture or playback. The first call also starts
streaming capture. When streaming capture is going on, you may
only queue further buffers or issue syncs until streaming
capture is switched off again with a argument of -1 to
a BUZIOC_QBUF_CAPT/BUZIOC_QBUF_PLAY ioctl.

BUZIOC_SYNC

Issue this ioctl when all buffers are queued. This ioctl will
block until the first buffer becomes free for saving its
data to disk (after BUZIOC_QBUF_CAPT) or for reuse (after BUZIOC_QBUF_PLAY).

BUZIOC_G_STATUS

Get the status of the input lines (video source connected/norm).

For programming example, please, look at lavrec.c and lavplay.c code in
lavtools-1.2p2 package (URL: http://www.cicese.mx/~mirsev/DC10plus/)
and the 'examples' directory in the original Buz driver distribution.

Additional notes for software developers:

   The driver returns maxwidth and maxheight parameters according to
   the current TV standard (norm). Therefore, the software which
   communicates with the driver and "asks" for these parameters should
   first set the correct norm. Well, it seems logically correct: TV
   standard is "more constant" for current country than geometry
   settings of a variety of TV capture cards which may work in ITU or
   square pixel format. Remember that users now can lock the norm to
   avoid any ambiguity.
--
Please note that lavplay/lavrec are also included in the MJPEG-tools
(http://mjpeg.sf.net/).

===========================

5. Applications

Applications known to work with this driver:

TV viewing:
* xawtv
* kwintv
* probably any TV application that supports video4linux or video4linux2.

MJPEG capture/playback:
* mjpegtools/lavtools (or Linux Video Studio)
* gstreamer
* mplayer

General raw capture:
* xawtv
* gstreamer
* probably any application that supports video4linux or video4linux2

Video editing:
* Cinelerra
* MainActor
* mjpegtools (or Linux Video Studio)

===========================

6. Concerning buffer sizes, quality, output size etc.

The zr36060 can do 1:2 JPEG compression. This is really the theoretical
maximum that the chipset can reach. The driver can, however, limit compression
to a maximum (size) of 1:4. The reason for this is that some cards (e.g. Buz)
can't handle 1:2 compression without stopping capture after only a few minutes.
With 1:4, it'll mostly work. If you have a Buz, use 'low_bitrate=1' to go into
1:4 max. compression mode.

100% JPEG quality is thus 1:2 compression in practice. So for a full PAL frame
(size 720x576). The JPEG fields are stored in YUY2 format, so the size of the
fields are 720x288x16/2 bits/field (2 fields/frame) = 207360 bytes/field x 2 =
414720 bytes/frame (add some more bytes for headers and DHT (huffman)/DQT
(quantization) tables, and you'll get to something like 512kB per frame for
1:2 compression. For 1:4 compression, you'd have frames of half this size.

Some additional explanation by Martin Samuelsson, which also explains the
importance of buffer sizes:
--
> Hmm, I do not think it is really that way. With the current (downloaded
> at 18:00 Monday) driver I get that output sizes for 10 sec:
> -q 50 -b 128 : 24.283.332 Bytes
> -q 50 -b 256 : 48.442.368
> -q 25 -b 128 : 24.655.992
> -q 25 -b 256 : 25.859.820

I woke up, and can't go to sleep again. I'll kill some time explaining why
this doesn't look strange to me.

Let's do some math using a width of 704 pixels. I'm not sure whether the Buz
actually use that number or not, but that's not too important right now.

704x288 pixels, one field, is 202752 pixels. Divided by 64 pixels per block;
3168 blocks per field. Each pixel consist of two bytes; 128 bytes per block;
1024 bits per block. 100% in the new driver mean 1:2 compression; the maximum
output becomes 512 bits per block. Actually 510, but 512 is simpler to use
for calculations.

Let's say that we specify d1q50. We thus want 256 bits per block; times 3168
becomes 811008 bits; 101376 bytes per field. We're talking raw bits and bytes
here, so we don't need to do any fancy corrections for bits-per-pixel or such
things. 101376 bytes per field.

d1 video contains two fields per frame. Those sum up to 202752 bytes per
frame, and one of those frames goes into each buffer.

But wait a second! -b128 gives 128kB buffers! It's not possible to cram
202752 bytes of JPEG data into 128kB!

This is what the driver notice and automatically compensate for in your
examples. Let's do some math using this information:

128kB is 131072 bytes. In this buffer, we want to store two fields, which
leaves 65536 bytes for each field. Using 3168 blocks per field, we get
20.68686868... available bytes per block; 165 bits. We can't allow the
request for 256 bits per block when there's only 165 bits available! The -q50
option is silently overridden, and the -b128 option takes precedence, leaving
us with the equivalence of -q32.

This gives us a data rate of 165 bits per block, which, times 3168, sums up
to 65340 bytes per field, out of the allowed 65536. The current driver has
another level of rate limiting; it won't accept -q values that fill more than
6/8 of the specified buffers. (I'm not sure why. "Playing it safe" seem to be
a safe bet. Personally, I think I would have lowered requested-bits-per-block
by one, or something like that.) We can't use 165 bits per block, but have to
lower it again, to 6/8 of the available buffer space: We end up with 124 bits
per block, the equivalence of -q24. With 128kB buffers, you can't use greater
than -q24 at -d1. (And PAL, and 704 pixels width...)

The third example is limited to -q24 through the same process. The second
example, using very similar calculations, is limited to -q48. The only
example that actually grab at the specified -q value is the last one, which
is clearly visible, looking at the file size.
--

Conclusion: the quality of the resulting movie depends on buffer size, quality,
whether or not you use 'low_bitrate=1' as insmod option for the zr36060.c
module to do 1:4 instead of 1:2 compression, etc.

If you experience timeouts, lowering the quality/buffersize or using
'low_bitrate=1 as insmod option for zr36060.o might actually help, as is
proven by the Buz.

===========================

7. It hangs/crashes/fails/whatevers! Help!

Make sure that the card has its own interrupts (see /proc/interrupts), check
the output of dmesg at high verbosity (load zr36067.o with debug=2,
load all other modules with debug=1). Check that your mainboard is favorable
(see question 2) and if not, test the card in another computer. Also see the
notes given in question 3 and try lowering quality/buffersize/capturesize
if recording fails after a period of time.

If all this doesn't help, give a clear description of the problem including
detailed hardware information (memory+brand, mainboard+chipset+brand, which
MJPEG card, processor, other PCI cards that might be of interest), give the
system PnP information (/proc/interrupts, /proc/dma, /proc/devices), and give
the kernel version, driver version, glibc version, gcc version and any other
information that might possibly be of interest. Also provide the dmesg output
at high verbosity. See 'Contacting' on how to contact the developers.

===========================

8. Maintainers/Contacting

The driver is currently maintained by Laurent Pinchart and Ronald Bultje
(<[e-mail]> and <[e-mail]>). For bug
reports or questions, please contact the mailinglist instead of the developers
individually. For user questions (i.e. bug reports or how-to questions), send
an email to <[e-mail]>, for developers (i.e. if you want to
help programming), send an email to <[e-mail]>. See
http://www.sf.net/projects/mjpeg/ for subscription information.

For bug reports, be sure to include all the information as described in
the section 'It hangs/crashes/fails/whatevers! Help!'. Please make sure
you're using the latest version (http://mjpeg.sf.net/driver-zoran/).

Previous maintainers/developers of this driver include Serguei Miridonov
<[e-mail]>, Wolfgang Scherr <[e-mail]>, Dave Perks
<[e-mail]> and Rainer Johanni <[e-mail]>.

===========================

9. License

This driver is distributed under the terms of the General Public License.

    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation; either version 2 of the License, or
    (at your option) any later version.

    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program; if not, write to the Free Software
    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

See http://www.gnu.org/ for more information.