[Mesa-dev] [PATCH 01/12] vl: add parameters for VAAPI encode

[Andy Furniss]

Christian König wrote:
> Am 05.07.2016 um 13:17 schrieb Christian König:
>> Am 01.07.2016 um 18:18 schrieb Andy Furniss:
>>> Christian König wrote:
>>>> Am 01.07.2016 um 18:02 schrieb Andy Furniss:
>>>>> Christian König wrote:
>>>>>> Am 01.07.2016 um 16:42 schrieb Andy Furniss:
>>>>>>> Boyuan Zhang wrote:
>>>>>>>> Signed-off-by: Boyuan Zhang <boyuan.zhang at amd.com>
>>>>>>>
>>>>>>> Is this supposed to be the same functionally as the first version?
>>>>>>>
>>>>>>> I notice on Tonga that I previously got cabac, but don't now.
>>>>>>>
>>>>>>> I see the new options are now in radeon_vce_52.c whereas before
>>>>>>> radeon_vce_40_2_2.c was changed - is this why?
>>>>>>
>>>>>> We wanted to keep the code for the old firmware versions as it is,
>>>>>> because we otherwise would need to test them again.
>>>>>>
>>>>>> Because of this we moved all modifications into radeon_vce_52.c.
>>>>>
>>>>> Oh OK, I guess it's early days but cabac was one thing that seemed OK.
>>>>>
>>>>> There are many issued on Tonga using vaapi - but then I suppose
>>>>> gstreamer and ffmpeg are not really set up for it yet.
>>>>>
>>>>> eg. worse first = gpu vm fault/lock with gstreamer but not ffmpeg.
>>>>>
>>>>> to provoke do
>>>>>
>>>>> gst-launch-1.0 videotestsrc ! video/x-raw, format=NV12, width=1920,
>>>>> height=1080, framerate=60/1 ! vaapih264enc | fakesink
>>>>>
>>>>> then flip away/back to another desktop soon will lock.
>>>>>
>>>>> Maybe I am too early eg. other patches needed? or does what you say
>>>>> above mean that vaapienc shouldn't even be enabled for older chips?
>>>>
>>>> Tonga should be perfectly supported, you just need recent enough
>>>> firmware.
>>>>
>>>> What does "dmesg | grep 'Found VCE firmware Version'" give you?
>>>
>>> Found VCE firmware Version: 50.17 Binary ID: 3
>>
>> Mhm, well looks like we haven't released the right firmware for Tonga
>> yet :(
>>
>> Going to leave you a note when we have everything together.
>
> The new firmware landed on Monday and the final patchset was released
> yesterday night.
>
> Have fun testing it,

Found VCE firmware Version: 52.4 Binary ID: 3

No luck I'm afraid. The above test will still hard lock me after about 
30 seconds.

I also notice it's eating memory 2.6 gig by the time I lock.

Replace vaapih264enc with omxh264enc and all is OK, will run for ages
no increasing memory usage. In fact I think this firmware & or kernel
has fixed a powerplay omx issue - will update my bug for that when I
have tested more.

FWIW the above test would make cqp with a qp of 0 - testing cbr does
also leak/lock, though specifying bitrates does sort of work, the result
is much higher than requested (maybe some framerate assumption - I am 
testing
high rates.