[Openchrome-users] New to the list

[Xavier Bachelot]

Derek Adams wrote:
> Xavier Bachelot wrote:
>> Derek Adams wrote:
>>> Coleman Kane wrote:
>>>> On Wed, 2008-06-04 at 08:14 -0700, Derek Adams wrote:
>>>>  
>>>>> Hello, I'm new to this list and I'm interested in development of 
>>>>> the CX700M2 xvmc support.
>>>>> I was wondering if people could clue me in with a little more 
>>>>> detail than the site contains.
>>>>>
>>>>> 1. Is anyone working on CX700M2 xvmc support?
>>>>> 2. Is anything preventing that development beyond a lack of workforce?
>>>>>
>> Lack of documentation.
>>
>>>>> I'd like to contribute to the effort of implementing xvmc on this 
>>>>> chipset.
>>>>> Thanks!
>>>>>
>>>>> -Derek
>>>>>     
>>>> I am attempting to do the same. I've been talking to the rep that runs
>>>> http://linux.via.com, however the news looks grim. It seems that VIA
>>>> won't release the XvMC docs for the newer hardware because they are
>>>> worried about getting sued by the MPEG LA cartel. I've not been able to
>>>> get much traction on convincing them to release it myself. VIA did 
>>>> admit
>>>> that Intel seems to be violating this "agreement" by releasing the same
>>>> information for their IGP chipsets.
>>>>
>>>> If anyone knows has any ideas on pushing this, I'd like to hear. It
>>>> would be really valuable for me to just get the docs, then we could
>>>> write this part ourselves.
>>>>
>>>>   
>>>
>>> Has anyone made any attempt at reverse engineering the hardware?
>>
>> The current XvMC code was done thru RE.
>>
>>> Perhaps some of the registers are similar to previous chipsets that 
>>> we have existing xvmc code for?
>>
>> Perhaps. The current hardware decoder engine is known as Unified Video 
>> Decoder and has a lot of added capacities than the original engine 
>> found on the Unichrome and Unichrome Pro.
>>
>>> Does VIA release source of their Linux driver?
>>
>> Only some part of the sources and the hardware decoder is just a 
>> binary blob.
>>
>>> Do they have a driver that support xvmc on the CX700M2?
>>>
>> Yes, they do. They also have some patches to xine and mplayer to allow 
>> them to take advantage of their binary blob.
>>
>>> Sorry for so many questions!
>>>
>> No worries. Feel free to ask if you have more.
>>
>> Regards,
>> Xavier
>>
>>
> Thanks very much for the info. I'll try to see what I can get out of 
> inspecting the binaries and the mplayer patches. I potentially have 
> access to a PCI bus analyzer which I have used in the past to do some 
> reverse engineering. Hopefully, the secrets of that binary blob won't be 
> secret for long!
> 
> *evil laugh*
> 
> A general question about MPEG decoders  though:
> Do they usually work as a standalone function in the video chipset? If I 
> send units of encoded data to them, do I get decoded units back that I 
> have to piece together myself? Or does the decoded info get sent down 
> some internal pipeline and is only then seen on the screen?
> 
This is becoming a bit too technical for me now, sorry.
What I know is the decoder is not fed with raw MPEG data, the datas need 
to be prepared first. What is currently implemented is XvMC VLD aka 
slice decoding. There is currently no support for mocomp or iDCT, this 
would be needed for mpeg4. Obviously, getting only mpeg2 to work on the 
CX700 (K8M890 and P4M900 too) would be a nice starting point :-)

This should answer at least some of your questions :
http://www.xfree86.org/~mvojkovi/XvMC_API.txt

Regards,
Xavier