H.264 VA-API Encode Comes To Gallium3D

[legume]

$ man fdatasync

LD_PRELOAD

Well I didn't write the apps I use - but anyway I was talking about default behavior, there's a tuneable that lets you change the % of ram allowed to fill before flush - that would help.

[legume]

Isn't that a simple color space conversion? I don't know much about video processing but I thought you should be able to do that in real time with no problems.

As you can see in my video, htop shows basically no significant CPU usage from any process.

Also I tried running an exact same pipeline on intel vaapi and radeonsi vaapi (including conversion to nv12) and the intel one was quite smooth while the radeonsi one was as choppy as you can see here.

I wouldn't say it's simple - for years the reverse has been offloaded to GPU to save CPU.
For me top, like cpufreq, doesn't seem to register SIMD usage very well.

Of course my CPU is old and I only have ffmpeg to easily bench with (gstreamer in theory possible to bench, but I couldn't get it to work).

I could do 25fps so your observation about intel vs radeon is quite valid.

If you are not running the latest patches, there's quite a big memory leak lurking to mess things up.

Also if you don't specify cbr you get cqp which currently seems to defaul to 0 = huge/slow.

Your prime setup is quite different from me - so I guess there will be difference there (I am nowhere near as slow as you see), but vaapi is still somewhat slower than omx for me - but it's still early days.

[legume]

I made another reply hours ago which still hasn't turned up - makes typing out stuff rather pointless on here :-(

[Gusar]

I made another reply hours ago which still hasn't turned up - makes typing out stuff rather pointless on here :-(

Blame the crappy vBulletin software. Indirectly, blame Phoronix for using said crappy vBulletin software. This vBulletin abomination is such immense crap, I can't believe people are paying money for it.

[haagch]

Of course my CPU is old and I only have ffmpeg to easily bench with (gstreamer in theory possible to bench, but I couldn't get it to work).

Well, you could do what I did and take http://www.sample-videos.com/video/m...y_720p_1mb.mp4 and run
LIBVA_DRIVER_NAME=radeonsi /usr/bin/time gst-launch-1.0 -e filesrc location=SampleVideo_1280x720_1mb.mp4 ! decodebin ! queue ! videoconvert ! queue ! video/x-raw,format=NV12 ! vaapih264enc ! h264parse ! matroskamux ! filesink location=output.mkv

If you are not running the latest patches, there's quite a big memory leak lurking to mess things up.

Well, this is the newest that is available as patchset on patchwork...

Also if you don't specify cbr you get cqp which currently seems to defaul to 0 = huge/slow.

Well, I had to look that one up. It's vaapih264enc rate-control=2.
No difference though, still takes 1:09 minutes.


So I revisited omx again.
/etc/xdg/gstomx.conf is the default location I believe. Mine looks a bit different:

[omxh264dec]
type-name=GstOMXH264Dec
core-name=/usr/lib/libomxil-bellagio.so.0
component-name=OMX.radeonsi.video_decoder.avc
rank=256
in-port-index=0
out-port-index=1

[omxmpeg2dec]
type-name=GstOMXMPEG2VideoDec
core-name=/usr/lib/libomxil-bellagio.so.0
component-name=OMX.radeonsi.video_decoder.mpeg2
rank=256
in-port-index=0
out-port-index=1

[omxh264enc]
type-name=GstOMXH264Enc
core-name=/usr/lib/libomxil-bellagio.so.0
component-name=OMX.radeonsi.video_encoder.avc
rank=256
in-port-index=0
out-port-index=1

Not sure where exactly I got. OMX.radeonsi.video_encoder.avc. I tried your component-name=OMX.mesa.video_encoder.avc but it doesn't make a difference.

omxregister-bellagio produces this file: http://haagch.frickel.club/files/.omxregister
I believe this is totally wrong, but I have no working one to compare.

[jones13]

Hi,
will this work with ffmpeg's vaapi encoder? I have tried to use VCE encoding on my kabini some time ago, but building these gstreamer pipelines and setting up omx bellagio was a pain...

[legume]

ffmpeg will work in the future - it was almost there last time I tried (needed patching).

[legume]

Well, you could do what I did and take http://www.sample-videos.com/video/m...y_720p_1mb.mp4 and run
LIBVA_DRIVER_NAME=radeonsi /usr/bin/time gst-launch-1.0 -e filesrc location=SampleVideo_1280x720_1mb.mp4 ! decodebin ! queue ! videoconvert ! queue ! video/x-raw,format=NV12 ! vaapih264enc ! h264parse ! matroskamux ! filesink location=output.mkv

That's not really what I meant when talking about benching screencap + bgr0 -> nv12, but FWIW on that very very short sample using s/w decode (your exact command doesn't work for me and hwdec -> hwenc has issues curently) I get -

time gst-launch-1.0 -e filesrc location=~/big_buck_bunny_720p_1mb.mp4 ! qtdemux ! h264parse ! avdec_h264 ! queue ! videoconvert ! queue ! video/x-raw,format=NV12 ! vaapih264enc rate-control=cbr bitrate=2000 ! h264parse ! matroskamux ! filesink location=output.mkv
Setting pipeline to PAUSED ...
libva info: VA-API version 0.38.1
libva info: va_getDriverName() returns 0
libva info: Trying to open /usr/lib/dri/radeonsi_drv_video.so
libva info: Found init function __vaDriverInit_0_38
libva info: va_openDriver() returns 0
Pipeline is PREROLLING ...
Got context from element 'vaapiencodeh264-0': gst.vaapi.Display=context, gst.vaapi.Display=(GstVaapiDisplay)NULL;
Redistribute latency...
Pipeline is PREROLLED ...
Setting pipeline to PLAYING ...
New clock: GstSystemClock
Got EOS from element "pipeline0".
Execution ended after 0:00:00.797613942
Setting pipeline to PAUSED ...
Setting pipeline to READY ...
Setting pipeline to NULL ...
Freeing pipeline ...

real    0m1.236s
user    0m1.284s
sys     0m0.069s

The bench I was referring to not being able to get to work was the gstreamer equivalent of doing with ffmpeg -

ffmpeg -f x11grab -r 300 -s 1920x1080 -i :0.0 -f null -

and

ffmpeg -f x11grab -r 300 -s 1920x1080 -i :0.0 -pix_fmt nv12 -f null -

both of which output an fps result, I get -

100 and 22fps with cpufreq on_demand and
138 and 35fps with cpus full freq, maybe ffmpeg/my box is just exceptionally crap at this task!

omxregister-bellagio produces this file: http://haagch.frickel.club/files/.omxregister
I believe this is totally wrong, but I have no working one to compare.

Hmm, what does gstreamer say if you try to use it?
Mine looks like -

/usr/lib/bellagio/libomx_mesa.so
 ==> OMX.mesa.video_decoder ==> OMX.mesa.video_decoder.mpeg2:OMX.mesa.video_decoder.avc:
 ==> OMX.mesa.video_encoder ==> OMX.mesa.video_encoder.avc:
/usr/lib/bellagio/libomxvideosched.so
 ==> OMX.st.video.scheduler ==> OMX.st.video.scheduler: ==> 2 1,456192 1,304128
/usr/lib/bellagio/libomxaudio_effects.so
 ==> OMX.st.volume.component ==> OMX.st.volume.component: ==> 2 1,65536 1,32768
 ==> OMX.st.audio.mixer ==> OMX.st.audio.mixer: ==> 1 50,60000
/usr/lib/bellagio/libomxclocksrc.so
 ==> OMX.st.clocksrc ==> OMX.st.clocksrc:

[haagch]

Hmm, what does gstreamer say if you try to use it?

LC_ALL=C DRI_PRIME=1 LIBVA_DRIVER_NAME=radeonsi gst-launch-1.0 videotestsrc ! omxh264enc ! "video/x-h264,profile=high" ! h264parse ! matroskamux ! filesink location=output.mkv

Setting pipeline to PAUSED ...
OMX-could not load /usr/lib/bellagio/libomxvideosched.so ==> OMX.st.video.scheduler ==> OMX.st.video.scheduler: ==> 2 1,456192 1,304128: /usr/lib/bellagio/libomxvideosched.so ==> OMX.st.video.scheduler ==> OMX.st.video.scheduler: ==> 2 1,456192 1,304128: cannot open shared object file: No such file or directory
Pipeline is PREROLLING ...
OMX-In RM_getResource No resource to be handled
Pipeline is PREROLLED ...
Setting pipeline to PLAYING ...
New clock: GstSystemClock

As I said, it tries to open the whole "/usr/lib/bellagio/libomxvideosched.so ==> OMX.st.video.scheduler ==> OMX.st.video.scheduler: ==> 2 1,456192 1,304128" thing as a filename, see the first line in my .omxregister.

But I noticed it DOES produce an output file. It plays very slowly for some reason and mpv prints
Invalid video timestamp: 1.633000 -> 1.567000
every frame, but it seems to work otherwise.

I suppose that it didn't really work earlier whenever I tried, because the color format was wrong. With I420 I really get a data flow error with omxh264enc.
I just tried the equivalent:
DRI_PRIME=1 gst-launch-1.0 ximagesrc display-name=:0 use-damage=0 startx=0 starty=0 endx=1919 endy=1079 ! queue ! videoconvert ! queue ! video/x-raw,format=NV12,framerate=30/1 ! omxh264enc ! "video/x-h264,profile=high" ! h264parse ! matroskamux ! filesink location=output.mkv
and it produces this video:

https://youtu.be/mbGDPtVIt9I

25-30% CPU load and heavy artifacting, but it works.

[haagch]

Ah, misread the question. Are you sure it copies right into the phoronix wysiwyg editor?
With yours I get

LC_ALL=C DRI_PRIME=1 gst-launch-1.0 ximagesrc display-name=:0 use-damage=0 startx=0 starty=0 endx=1919 endy=1079 ! queue ! videoconvert ! queue ! video/x-raw,format=NV12,frame
rate=30/1 ! omxh264enc ! "video/x-h264,profile=high" ! h264parse ! matroskamux ! filesink location=output.mkv
Setting pipeline to PAUSED ...
OMX-could not load ==> OMX.mesa.video_decoder ==> OMX.mesa.video_decoder.mpeg2:OMX.mesa.video_decoder.avc:: ==> OMX.mesa.video_decoder ==> OMX.mesa.video_decoder.mpeg2:OMX.mesa.video_decode
r.avc:: cannot open shared object file: No such file or directory
OMX-could not load ==> OMX.mesa.video_encoder ==> OMX.mesa.video_encoder.avc:: ==> OMX.mesa.video_encoder ==> OMX.mesa.video_encoder.avc:: cannot open shared object file: No such file or di
rectory
OMX-could not load ==> OMX.st.video.scheduler ==> OMX.st.video.scheduler: ==> 2 1,456192 1,304128: ==> OMX.st.video.scheduler ==> OMX.st.video.scheduler: ==> 2 1,456192 1,304128: cannot ope
n shared object file: No such file or directory
OMX-could not load ==> OMX.st.volume.component ==> OMX.st.volume.component: ==> 2 1,65536 1,32768: ==> OMX.st.volume.component ==> OMX.st.volume.component: ==> 2 1,65536 1,32768: cannot ope
n shared object file: No such file or directory
OMX-could not load ==> OMX.st.audio.mixer ==> OMX.st.audio.mixer: ==> 1 50,60000: ==> OMX.st.audio.mixer ==> OMX.st.audio.mixer: ==> 1 50,60000: cannot open shared object file: No such file
or directory
OMX-could not load ==> OMX.st.clocksrc ==> OMX.st.clocksrc:: ==> OMX.st.clocksrc ==> OMX.st.clocksrc:: cannot open shared object file: No such file or directory
Pipeline is live and does not need PREROLL ...
Setting pipeline to PLAYING ...
New clock: GstSystemClock
OMX-In RM_getResource No resource to be handled

Also I think my previous post was not unapproved, but completely lost.
Tl;dr was: I noticed it does indeed produce an output file and I think on my previous tries maybe it didn't because of the color format. I420 produces a data flow error, but converted to NV12 works. Here's a video: https://www.youtube.com/watch?v=mbGDPtVIt9I. 20-25% CPU usage and very artifact-y, but works.