[FFmpeg-devel] [PATCH 5/5] lavd: Add KMS frame grabber

[Mark Thompson]

On 19/09/17 22:21, Andy Furniss wrote:
> Mark Thompson wrote:
>> On 15/09/17 00:15, Andy Furniss wrote:
>>> Andy Furniss wrote:
>>>> Mark Thompson wrote:
>>>>> ---
>>>>> Now sets the trusted packet flag; otherwise unchanged.
>>>>>
>>>>>
>>>>>    configure                |   1 +
>>>>>    libavdevice/Makefile     |   1 +
>>>>>    libavdevice/alldevices.c |   1 +
>>>>>    libavdevice/kmsgrab.c    | 455 +++++++++++++++++++++++++++++++++++++++++++++++
>>>>>    4 files changed, 458 insertions(+)
>>>>>    create mode 100644 libavdevice/kmsgrab.c
>>>>>
>>>>> diff --git a/configure b/configure
>>>>> index 6581c53c1a..76a7591ceb 100755
>>>>> --- a/configure
>>>>> +++ b/configure
>>>>> @@ -3040,6 +3040,7 @@ gdigrab_indev_select="bmp_decoder"
>>>>>    iec61883_indev_deps="libiec61883"
>>>>>    jack_indev_deps="jack"
>>>>>    jack_indev_deps_any="sem_timedwait dispatch_dispatch_h"
>>>>> +kmsgrab_indev_deps="libdrm"
>>>>
>>>> Doesn't get built for me = doesn't show up as indev after configure anything special needed?
>>>
>>> Never mind I found --enable-libdrm (had tried --enable-kmsgrab)
>>
>> I assume you're going to try AMD + Mesa + VAAPI.
>>
>> VAAPI_DISABLE_INTERLACE=1 ./ffmpeg_g -y -format bgr0 -device /dev/dri/card1 -f kmsgrab -i - -vsync 0 -init_hw_device vaapi=v:/dev/dri/renderD129 -filter_hw_device v -vf 'hwmap,scale_vaapi=w=1920:h=1080:format=nv12' -c:v h264_vaapi -profile 578 -bf 0 out.mp4
>>
>> Three Mesa issues I had to get around:
>>
>> * Device derivation doesn't work because the Mesa driver doesn't want to initialise on the DRM master device for some reason; making the matching device separately does work.
>> * Against current git, you need to reapply the VAAPI_DISABLE_INTERLACE patch and use it - if not, the colour conversion just barfs because it wants interlaced surfaces.
>> * The postproc scaler seems to only write the luma plane when converting from RGB - this is also visible when uploading normal RGB images, so just a bug somewhere.
>>
>> With that, it works to record the screen in greyscale...
>>
>> I have some other Mesa stuff to do queued up (libva2 with VAAPI export for EGL import on AMD), so I'll pursue these further soonish.
> 
> Leo just posted a patchset on the mesa list that makes vaapi work better with this and postproc generally.
> 
> You don't need the env with those and kmsgrab is now working for me - up to a point ...

Yep, that works for me now too - with colour!  (The export to GL for playback is still messed up by interlacing without patching it out, but that looked orthogonal anyway.)

> That point being around 7k frames it will run out of something.
> 
> [AVHWFramesContext @ 0x31ed880] Failed to create surface from DRM object: 2 (resource allocation failed).
> [Parsed_hwmap_0 @ 0x3114c40] Failed to map frame: -5.
> 
> I see that memory is reducing before this although I still have spare - is this the same issue you explained on users WRT leaking on decode?

Yeah, I also run out of ... something ... at around 7200 frames.  It's not fds or memory.  I don't think it's the buffer problem (which, incidentally, should finally be fixable sensibly in libva2 soon), because that ended up manifesting as leaking memory.  It's also not a problem for Intel (I've already been running that for a long time to test).  Maybe some other sort of handle on the Mesa side?  I'll investigate further tomorrow.

Thanks,

- Mark