Ffmpeg intel quick sync #3842

th0ma7 · 2020-01-02T03:14:17Z

Motivation: Migrate over to newer VAAPI driver implementation using iHD driver 19.4.0
Linked issues: #3830

Checklist

Build rule all-supported completed successfully
Package upgrade completed successfully
New installation of package completed successfully

New iHD driver and dependancies significantly increases compile time for x64 arch.

th0ma7 · 2020-01-02T03:24:25Z

New driver output on apollolake:

$ sudo /var/packages/ffmpeg/target/bin/vainfo 
Password: 
libva info: VA-API version 1.6.0
libva info: Trying to open /var/packages/ffmpeg/target/lib/iHD_drv_video.so
libva info: Found init function __vaDriverInit_1_6
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.6 (libva 2.6.0)
vainfo: Driver version: Intel iHD driver - 19.4.0
vainfo: Supported profile and entrypoints
      VAProfileNone                   :	VAEntrypointVideoProc
      VAProfileNone                   :	VAEntrypointStats
      VAProfileMPEG2Simple            :	VAEntrypointVLD
      VAProfileMPEG2Main              :	VAEntrypointVLD
      VAProfileH264Main               :	VAEntrypointVLD
      VAProfileH264Main               :	VAEntrypointEncSlice
      VAProfileH264Main               :	VAEntrypointFEI
      VAProfileH264Main               :	VAEntrypointEncSliceLP
      VAProfileH264High               :	VAEntrypointVLD
      VAProfileH264High               :	VAEntrypointEncSlice
      VAProfileH264High               :	VAEntrypointFEI
      VAProfileH264High               :	VAEntrypointEncSliceLP
      VAProfileVC1Simple              :	VAEntrypointVLD
      VAProfileVC1Main                :	VAEntrypointVLD
      VAProfileVC1Advanced            :	VAEntrypointVLD
      VAProfileJPEGBaseline           :	VAEntrypointVLD
      VAProfileJPEGBaseline           :	VAEntrypointEncPicture
      VAProfileH264ConstrainedBaseline:	VAEntrypointVLD
      VAProfileH264ConstrainedBaseline:	VAEntrypointEncSlice
      VAProfileH264ConstrainedBaseline:	VAEntrypointFEI
      VAProfileH264ConstrainedBaseline:	VAEntrypointEncSliceLP
      VAProfileVP8Version0_3          :	VAEntrypointVLD
      VAProfileHEVCMain               :	VAEntrypointVLD
      VAProfileHEVCMain               :	VAEntrypointEncSlice
      VAProfileHEVCMain               :	VAEntrypointFEI
      VAProfileHEVCMain10             :	VAEntrypointVLD
      VAProfileVP9Profile0            :	VAEntrypointVLD

th0ma7 · 2020-01-02T11:32:07Z

@kc6108 here you go:

ffmpeg -codecs 2>/dev/null | grep -Ei -e ' e?ac3' -e 'truehd' -e 'dts' -e 'h26[45]' -e 'hevc' -e 'qsv'

$ ./ffmpeg -codecs 2>/dev/null | grep -Ei -e ' e?ac3' -e 'truehd' -e 'dts' -e 'h26[45]' -e 'hevc' -e 'qsv'
 DEV.LS h264                 H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (decoders: h264 h264_v4l2m2m h264_qsv ) (encoders: libx264 libx264rgb h264_qsv h264_v4l2m2m h264_vaapi )
 DEV.L. hevc                 H.265 / HEVC (High Efficiency Video Coding) (decoders: hevc hevc_qsv ) (encoders: libx265 hevc_qsv hevc_vaapi )
 DEVIL. mjpeg                Motion JPEG (encoders: mjpeg mjpeg_qsv mjpeg_vaapi )
 DEV.L. mpeg2video           MPEG-2 video (decoders: mpeg2video mpegvideo mpeg2_v4l2m2m mpeg2_qsv ) (encoders: mpeg2video mpeg2_qsv mpeg2_vaapi )
 D.V.L. vc1                  SMPTE VC-1 (decoders: vc1 vc1_qsv vc1_v4l2m2m )
 DEV.L. vp8                  On2 VP8 (decoders: vp8 vp8_v4l2m2m libvpx vp8_qsv ) (encoders: libvpx vp8_v4l2m2m vp8_vaapi )
 DEA.L. ac3                  ATSC A/52A (AC-3) (decoders: ac3 ac3_fixed ) (encoders: ac3 ac3_fixed )
 DEA.LS dts                  DCA (DTS Coherent Acoustics) (decoders: dca ) (encoders: dca )
 DEA.L. eac3                 ATSC A/52B (AC-3, E-AC-3)
 DEA..S truehd               TrueHD

th0ma7 · 2020-01-02T13:01:26Z

My readings leans towards the following:
Option 1) _smd

I can port Synology ffmpeg evansport patches from 2.7.1 to 4.2.1. I think it's relatively easy.
I would need to compile against Intel® Media Processor SDK pre-built libraries that are provided through here:
https://sourceforge.net/projects/dsgpl/files/toolkit/DSM6.0/ds.evansport-6.0.dev.txz
By far the best would be to find the source code of the original Intel® Media Processor SDK (perhaps by asking Intel directly?)

Option 2) _qsv
I believe that the Intel® Media Processor SDK was superseeded by Intel® Media SDK. If this is true then it may be possible to get _qsv encoders/decoders working for evansport arch. I'll check if this is even feasible.

Option 3) _smd & _qsv
Get both options working so it is backward compatible.

th0ma7 · 2020-01-02T13:09:58Z

I have read some really good fixes for pixelation type issues are included with i965 v2.4 as well. Might pull in latest libva-utils as well. Just a thought.

https://github.com/intel/libva-utils/archive/2.6.0.tar.gz
https://github.com/intel/intel-vaapi-driver/archive/2.4.0.tar.gz

If I'm not mistaken, we can revert to the i965 driver by simply deleting/renaming the iHD driver? Is that correct?

I had pulled libva 2.6 but forgot to pull libva-utils as well. Will do that for next round of packages.

Also yes, it does fallback to "legacy" i965 driver when renaming iHD_drv_video.so:

$ sudo ./vainfo 
libva info: VA-API version 1.6.0
libva info: Trying to open /var/packages/ffmpeg/target/lib/iHD_drv_video.so
libva info: va_openDriver() returns -1
libva info: Trying to open /var/packages/ffmpeg/target/lib/i965_drv_video.so
libva info: Found init function __vaDriverInit_1_6
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.6 (libva 2.6.0)
vainfo: Driver version: Intel i965 driver for Intel(R) Broxton - 2.1.0
vainfo: Supported profile and entrypoints
      VAProfileMPEG2Simple            :	VAEntrypointVLD
      VAProfileMPEG2Main              :	VAEntrypointVLD
      VAProfileH264ConstrainedBaseline:	VAEntrypointVLD
      VAProfileH264ConstrainedBaseline:	VAEntrypointEncSlice
      VAProfileH264ConstrainedBaseline:	VAEntrypointEncSliceLP
      VAProfileH264Main               :	VAEntrypointVLD
      VAProfileH264Main               :	VAEntrypointEncSlice
      VAProfileH264Main               :	VAEntrypointEncSliceLP
      VAProfileH264High               :	VAEntrypointVLD
      VAProfileH264High               :	VAEntrypointEncSlice
      VAProfileH264High               :	VAEntrypointEncSliceLP
      VAProfileH264MultiviewHigh      :	VAEntrypointVLD
      VAProfileH264MultiviewHigh      :	VAEntrypointEncSlice
      VAProfileH264StereoHigh         :	VAEntrypointVLD
      VAProfileH264StereoHigh         :	VAEntrypointEncSlice
      VAProfileVC1Simple              :	VAEntrypointVLD
      VAProfileVC1Main                :	VAEntrypointVLD
      VAProfileVC1Advanced            :	VAEntrypointVLD
      VAProfileNone                   :	VAEntrypointVideoProc
      VAProfileJPEGBaseline           :	VAEntrypointVLD
      VAProfileJPEGBaseline           :	VAEntrypointEncPicture
      VAProfileVP8Version0_3          :	VAEntrypointVLD
      VAProfileVP8Version0_3          :	VAEntrypointEncSlice
      VAProfileHEVCMain               :	VAEntrypointVLD
      VAProfileHEVCMain               :	VAEntrypointEncSlice
      VAProfileHEVCMain10             :	VAEntrypointVLD
      VAProfileVP9Profile0            :	VAEntrypointVLD

Lastly, x64 package version -20 now available.

BenjaminPoncet · 2020-01-02T13:26:37Z

My readings leans towards the following:
Option 1) _smd

I can port Synology ffmpeg evansport patches from 2.7.1 to 4.2.1. I think it's relatively easy.

I would need to compile against Intel® Media Processor SDK pre-built libraries that are provided through here:
https://sourceforge.net/projects/dsgpl/files/toolkit/DSM6.0/ds.evansport-6.0.dev.txz
By far the best would be to find the source code of the original Intel® Media Processor SDK (perhaps by asking Intel directly?)

Option 2) _qsv
I believe that the Intel® Media Processor SDK was superseeded by Intel® Media SDK. If this is true then it may be possible to get _qsv encoders/decoders working for evansport arch. I'll check if this is even feasible.

Option 3) _smd & _qsv
Get both options working so it is backward compatible.

Such good news!
I'll test the -20 quickly (on Synology DS718+)
Just to understand, is it necessary to use the h264_qsv codec to use Quick Sync or does the iHD driver do it by itself through VAAPI?
Concretely EvanSport it would be great that it works with quick sync and option 3 would just be perfect.
Anyway, we have someone who I think will be happy to test both solutions on my gist.
Thanks again for your great work.

th0ma7 · 2020-01-02T14:09:40Z

I'll test the -20 quickly (on Synology DS718+)

Let me know how thing goes with it.

Just to understand, is it necessary to use the h264_qsv codec to use Quick Sync or does the iHD driver do it by itself through VAAPI?

I believe so.
Now question is, will videostation automatically call-up *_qsv instead of *_vaapi ?

Concretely EvanSport it would be great that it works with quick sync and option 3 would just be perfect.

I will need quite a few more cycles to get any of the options working for evansport ...
And I don't have a compatible NAS to test things up which further complicates things.

BenjaminPoncet · 2020-01-02T17:30:53Z

First test with the h265 10bit video I've been using since the beginning and without changing anything to the wrapper: It's much better, indeed there is no more pixelization in the first minutes but above all, the transcoding in medium quality (720p) and in low quality (480p) is much better.
So Quick Sync or not the iHD driver brings a real plus.
I will now test and compare using _qvs via the wrapper and a rewrite of the parameters.
I'm not sure I'll see a gain in quality but maybe a gain in CPU load.

Concerning EvanSport if it's easier to compile with Quick Sync at first, I can create a specific wrapper for EvanSport. Because if indeed Quick Sync works with EvanSport, SMD is only of interest for direct compatibility with VideoStation, not for ffmpeg use outside VideoStation...

ymartin59 · 2020-01-02T18:17:49Z

@th0ma7 I have evansport device (DS214play) so I may compile with specific options and try on my own with video samples. Let me know

BenjaminPoncet · 2020-01-02T23:46:24Z

@th0ma7,

I have finished my first QVS tests and the hevc_qsv driver seems not to work. Every time I use it in encoding, decoding or hevc/hevc transcoding I get the following error:

[hevc_qsv @ 0xd1d800] Could not load the requested plugin '33a61c0b4c27454ca8d85dde757c6f8e': specified object was not found (-9)

On the other hand, h264 transcoding works correctly.

Here are my logs for h264 transcoding (which works):

2020-01-02 23:59:22 - hYdCTjxZ - *** PROCESS START ***
2020-01-02 23:59:22 - hYdCTjxZ - MOVIE    = /volumeUSB2/usbshare/Films/VID_H264_AC3.mkv
2020-01-02 23:59:22 - hYdCTjxZ - FFMPEG   = /var/packages/ffmpeg/target/bin/ffmpeg
2020-01-02 23:59:22 - hYdCTjxZ - PARAMVS  =
2020-01-02 23:59:22 - hYdCTjxZ -          = -ss 0.000 -vaapi_device /dev/dri/renderD128 -hwaccel vaapi -hwaccel_output_format vaapi -noautorotate -i /volumeUSB2/usb
2020-01-02 23:59:22 - hYdCTjxZ -          = share/Films/VID_H264_AC3.mkv -vcodec h264_vaapi -vf format=nv12|vaapi,hwupload,setsar=sar=1,scale_vaapi=w=1920:h=1040 -v
2020-01-02 23:59:22 - hYdCTjxZ -          = sync 2 -bf 0 -vb 4500000 -acodec libmp3lame -ab 128K -ac 2 -f ssegment -segment_format mpegts -segment_list_type m3u8 -h
2020-01-02 23:59:22 - hYdCTjxZ -          = ls_seek_time 0 -segment_time 5 -segment_time_delta 0.000 -segment_start_number 00000 -avoid_negative_ts 0 -break_non_key
2020-01-02 23:59:22 - hYdCTjxZ -          = frames 1 -max_muxing_queue_size 1024 -map 0:0 -map 0:1 /tmp/VideoStation/HLS/e99c64d12acf192e0e7005d6c89bf96e_hYdCTjxZ/s
2020-01-02 23:59:22 - hYdCTjxZ -          = lice-%05d.ts
2020-01-02 23:59:22 - hYdCTjxZ - PARAMQVS =
2020-01-02 23:59:22 - hYdCTjxZ -          = -ss 0.000 -hwaccel qsv -vcodec h264_qsv -noautorotate -i /volumeUSB2/usbshare/Films/VID_H264_AC3.mkv -vcodec h264_qsv -v
2020-01-02 23:59:22 - hYdCTjxZ -          = f scale_qsv=w=1920:h=1040 -vsync 2 -bf 0 -vb 4500000 -acodec libmp3lame -ab 128K -ac 2 -f ssegment -segment_format mpegt
2020-01-02 23:59:22 - hYdCTjxZ -          = s -segment_list_type m3u8 -hls_seek_time 0 -segment_time 5 -segment_time_delta 0.000 -segment_start_number 00000 -avoid_
2020-01-02 23:59:22 - hYdCTjxZ -          = negative_ts 0 -break_non_keyframes 1 -max_muxing_queue_size 1024 -map 0:0 -map 0:1 /tmp/VideoStation/HLS/e99c64d12acf192
2020-01-02 23:59:22 - hYdCTjxZ -          = e0e7005d6c89bf96e_hYdCTjxZ/slice-%05d.ts
2020-01-02 23:59:22 - hYdCTjxZ - CHILDPID = 7102
2020-01-03 00:00:00 - hYdCTjxZ - *** KILLCHILD ***
2020-01-03 00:00:00 - hYdCTjxZ - *** CHILD END ***
2020-01-03 00:00:00 - hYdCTjxZ - *** PROCESS END ***

Here are my logs for h265 8bits -> h264 transcoding (which doesn't work):

2020-01-03 00:18:59 - vvT4Wijp - *** PROCESS START ***
2020-01-03 00:18:59 - vvT4Wijp - MOVIE    = /volumeUSB2/usbshare/Films/VID_H265_EAC3.mkv
2020-01-03 00:18:59 - vvT4Wijp - FFMPEG   = /var/packages/ffmpeg/target/bin/ffmpeg
2020-01-03 00:18:59 - vvT4Wijp - PARAMVS  =
2020-01-03 00:18:59 - vvT4Wijp -          = -ss 0.000 -vaapi_device /dev/dri/renderD128 -hwaccel vaapi -hwaccel_output_format vaapi -noautorotate -i /volumeUSB2/usb
2020-01-03 00:18:59 - vvT4Wijp -          = share/Films/VID_H265_EAC3.mkv -vcodec h264_vaapi -vf format=nv12|vaapi,hwupload,setsar=sar=1,scale_vaapi=w=1920:h=960 -v
2020-01-03 00:18:59 - vvT4Wijp -          = sync 2 -bf 0 -vb 3183492 -acodec libmp3lame -ab 128K -ac 2 -f ssegment -segment_format mpegts -segment_list_type m3u8 -h
2020-01-03 00:18:59 - vvT4Wijp -          = ls_seek_time 0 -segment_time 5 -segment_time_delta 0.000 -segment_start_number 00000 -avoid_negative_ts 0 -break_non_key
2020-01-03 00:18:59 - vvT4Wijp -          = frames 1 -max_muxing_queue_size 1024 -map 0:0 -map 0:1 /tmp/VideoStation/HLS/43fa6ce62f7fb6aef78e85ab5ae400ae_vvT4Wijp/s
2020-01-03 00:18:59 - vvT4Wijp -          = lice-%05d.ts
2020-01-03 00:18:59 - vvT4Wijp - PARAMQVS =
2020-01-03 00:18:59 - vvT4Wijp -          = -ss 0.000 -hwaccel qsv -vcodec hevc_qsv -noautorotate -i /volumeUSB2/usbshare/Films/VID_H265_EAC3.mkv -vcodec h264_qsv -
2020-01-03 00:18:59 - vvT4Wijp -          = vf scale_qsv=w=1920:h=960 -vsync 2 -bf 0 -vb 3183492 -acodec libmp3lame -ab 128K -ac 2 -f ssegment -segment_format mpegt
2020-01-03 00:18:59 - vvT4Wijp -          = s -segment_list_type m3u8 -hls_seek_time 0 -segment_time 5 -segment_time_delta 0.000 -segment_start_number 00000 -avoid_
2020-01-03 00:18:59 - vvT4Wijp -          = negative_ts 0 -break_non_keyframes 1 -max_muxing_queue_size 1024 -map 0:0 -map 0:1 /tmp/VideoStation/HLS/43fa6ce62f7fb6a
2020-01-03 00:18:59 - vvT4Wijp -          = ef78e85ab5ae400ae_vvT4Wijp/slice-%05d.ts
2020-01-03 00:18:59 - vvT4Wijp - CHILDPID = 20020
2020-01-03 00:19:04 - vvT4Wijp - STDERR   =
2020-01-03 00:19:04 - vvT4Wijp -          = [hevc_qsv @ 0xd1d800] Could not load the requested plugin '33a61c0b4c27454ca8d85dde757c6f8e': specified object was not f
2020-01-03 00:19:04 - vvT4Wijp -          = ound (-9)
2020-01-03 00:19:04 - vvT4Wijp -          = [hevc_qsv @ 0xd1d800] Error loading plugins
2020-01-03 00:19:04 - vvT4Wijp -          = [hevc_qsv @ 0xd1d800] Error initializing an MFX session
2020-01-03 00:19:04 - vvT4Wijp -          = Error while decoding stream #0:0: Function not implemented
2020-01-03 00:19:04 - vvT4Wijp -          = [hevc_qsv @ 0xd1d800] Could not load the requested plugin '33a61c0b4c27454ca8d85dde757c6f8e': specified object was not f
2020-01-03 00:19:04 - vvT4Wijp -          = ound (-9)
2020-01-03 00:19:04 - vvT4Wijp -          = [hevc_qsv @ 0xd1d800] Error loading plugins
2020-01-03 00:19:04 - vvT4Wijp -          = [hevc_qsv @ 0xd1d800] Error initializing an MFX session
2020-01-03 00:19:04 - vvT4Wijp -          = Error while decoding stream #0:0: Function not implemented
[hevc_qsv @ 0xd1d800] Could not load the requested plugin '33a61c0b4c27454ca8d85dde757c6f8e': specified object was not f/A    
2020-01-03 00:19:04 - vvT4Wijp -          = ound (-9)
2020-01-03 00:19:04 - vvT4Wijp -          = [hevc_qsv @ 0xd1d800] Error loading plugins
2020-01-03 00:19:04 - vvT4Wijp -          = [hevc_qsv @ 0xd1d800] Error initializing an MFX session
2020-01-03 00:19:04 - vvT4Wijp -          = Error while decoding stream #0:0: Function not implemented
2020-01-03 00:19:04 - vvT4Wijp -          = Too many packets buffered for output stream 0:1.
2020-01-03 00:19:04 - vvT4Wijp -          = [libmp3lame @ 0xd43ac0] 4 frames left in the queue on closing
2020-01-03 00:19:04 - vvT4Wijp -          = Conversion failed!
2020-01-03 00:19:04 - vvT4Wijp - *** CHILD END ***
2020-01-03 00:19:04 - vvT4Wijp - *** PROCESS END ***

Here is the wrapper used (Warning, you have to choose explicitly the encoder and decoder. By default it is in h265 decode mode and h264 encoding mode):
https://gist.github.com/BenjaminPoncet/4f7b635e993a8de1105a3f3bc361dea7

kc6108 · 2020-01-03T01:17:32Z

@BenjaminPoncet

You might try adding -load_plugin hevc_hw (or some derivative) as seen in this example:

ffmpeg -y -c:v hevc_qsv -load_plugin hevc_hw -i video-h265.mkv -an -c:v h264_qsv out.mp4

See here for more info:

https://emby.media/community/index.php?/topic/55876-quicksync-apollo-lake-windows-10-not-working-cpu-fallback/

15dd936825ad475ea34e35f3f54217a6 is the HEVC Software decoder

33a61c0b4c27454ca8d85dde757c6f8e is the HEVC hardware decoder

th0ma7 · 2020-01-03T01:31:18Z

Option 2) failed: Intel media driver doesn't support 32-bit:

th0ma7 · 2020-01-03T01:42:27Z

And libva-utils version 2.6.0 currently fails to build without x11 support.
(interesting that I had forgotten that one specifically)

with X11 support disabled, build fails intel/libva-utils#153

kc6108 · 2020-01-08T12:07:07Z

There isn’t a version 2.6. The latest is 2.4. We are at 2.1 right now. Your repo is the same:

https://github.com/th0ma7/spksrc/blob/ffmpeg-IntelQuickSync/cross/intel-vaapi-driver/digests

Your vainfo output above shows the same:

#3842 (comment)

vainfo: Driver version: Intel i965 driver for Intel(R) Broxton - 2.1.0

kc6108 · 2020-01-08T12:18:03Z

I thought about creating an issue to announce this to anyone wanting to get a jump on DSM 7.0:

https://event.synology.com/en-global/dsm_preview

I’ll let someone else do that if appropriate.

th0ma7 · 2020-01-08T12:54:25Z

I thought about creating an issue to announce this to anyone wanting to get a jump on DSM 7.0:

https://event.synology.com/en-global/dsm_preview

I’ll let someone else do that if appropriate.

Yup, I saw that last week. Thnx for sharing.

BenjaminPoncet · 2020-01-08T13:06:54Z

@kc6108,

Remind me again why you want to use the i965 driver?
Of all the tests I've done the iHD driver is much better : better quality, faster.

For info, to switch from one to the other just do the following commands:

export LIBVA_DRIVER_NAME=i965
OR
export LIBVA_DRIVER_NAME=iHD

You can see the change via the verbose mode:

/var/packages/ffmpeg/target/bin/ffmpeg -hwaccel vaapi -hwaccel_output_format vaapi -i "/volumeUSB2/usbshare/Test/VID_H265_EAC3.mkv" -c:v h264_vaapi -v verbose output.mp4

th0ma7 · 2020-01-08T14:58:42Z

There isn’t a version 2.6. The latest is 2.4. We are at 2.1 right now. Your repo is the same:

https://github.com/th0ma7/spksrc/blob/ffmpeg-IntelQuickSync/cross/intel-vaapi-driver/digests

You are totally right! Again good catch...
I mixed-it up with libva which was updated to 2.6...
Adding the update in the new packages...

th0ma7 · 2020-01-08T20:15:15Z

New x64 and apollolake packages ready.
Hopefully this will fix the apparmor issue.
Also note taht VAAPI driver got updated to 2.4.
As always, feedbacks greatly appreciated
PS: I think we are now getting closer for a fully functional iHD driver...

BenjaminPoncet · 2020-01-08T23:27:00Z

@th0ma7, You got it!

No more apparmor problems.

Regarding CPU usage and quality tests, I did not notice any significant difference between VAAPI and QSV on a h265 10bit to h264 8bit transcoding.
The only thing is that QSV is about 15% faster on a complete stream like in VideoStation.
I took the opportunity to test the new i965 driver and there are still the same pixelization problems in the first minute and a clearly worse quality in 480p but it's maybe a bit faster than iHD.

If you want to test in VideoStation, I created a special wrapper that replaces VAAPI by QSV. I tested it only with h264 and h265 but logically it should work with everything:

wget -O - https://gist.githubusercontent.com/BenjaminPoncet/4f7b635e993a8de1105a3f3bc361dea7/raw/ffmpeg-wrapper-qsv > /var/packages/VideoStation/target/bin/ffmpeg

And to test with i965, just add the following line at the beginning of the script:

export LIBVA_DRIVER_NAME=i965

In short, for me there's nothing to stop merge anymore.

Good job !!!

th0ma7 · 2020-01-09T00:05:43Z

Great job on the testing you too! woot!
@kc6108 pointed me exactly where needed so it was relatively easy to fix.

BTW, @BenjaminPoncet feel free to look & follow #3828
I'm thinking that your script could be something worth using in order to automatize the replacement of videostation ffmpeg with synocommunity ffmpeg + your wrapper.

@ymartin59 it's now ready for review and merge!

BenjaminPoncet · 2020-01-09T00:29:39Z

@th0ma7,

Maybe one more thing!

@kc6108, again, has found the explanation for a problem with VideoStation extracting .srt subtitles via the ffmpeg SynoCommunity.
It's still an Apparmor problem but I don't know if spk can do anything about it.
For this problem, I coded a workaround in the wapper.

See: https://gist.github.com/BenjaminPoncet/bbef9edc1d0800528813e75c1669e57e#gistcomment-3131604 and my answer below.

Sorry ;)

th0ma7 · 2020-01-09T00:58:03Z

Yeah I read the thread, nice finding from @kc6108 , again ;)
As it is totally independant from adding qsv I have opened up a new issue #3847.

th0ma7 · 2020-01-16T04:09:32Z

I just finished testing several audio and video codecs and found no regressions.

This awesome news, thnx for the feedback @kc6108
New packages soon to be released to the communuity up to this pull request.

ymartin59 · 2020-01-18T06:16:45Z

Well done to every contributor for this complex team effort, really impressive!

ymartin59 · 2020-01-18T06:57:46Z

@th0ma7 May you detail me the "x64" package split I have to apply to deliver architecture specific optimizations? What are architectures which are not specificly optimized with hardware acceleration? (I guess it is not a runtime decision but a compile-time switch, right?)
To avoid to build many packages for "non optimized", I will probably create a dedicated toolchain to produce a "x64" for ffmpeg binaries without hardware acceleration "x64-softffmpeg" for instance.
Am I correct with that idea?

th0ma7 · 2020-01-18T13:47:17Z

@ymartin59 this shouldn't matter... I may be wrong but my understanding is that at the end of day a call to vainfo validates all the available optimizations specific to your CPU (and there are a lot of subtilities for sub-optimization for every codecs).

Thus only what's available will work (e.g. nothing = no optimization) meaning that a non-optimized package is unecessary. Acceleration simply wont show up when calling ffmpeg -accels.

ymartin59 · 2020-01-18T17:01:14Z

@th0ma7 OK I see you have enabled all optimizations for any x64 architecture, so according to this compile directive, I may publish a single package x64. I will test in xpenology where hardware acceleration should not work.

* ffmpeg 4.2.1-22: add Intel Quick Sync support, VAAPI driver update, apparmor mfx driver fix * gmmlib 19.4.1: Intel Graphics Memory Management Library * libdrm 2.4.100: disable radeon, amdgpu and nouveau drivers * libva 2.6: update for Intel Media driver dependency * intel-media-driver 19.4.0r: build, install fixes, plugins.cfg * intel-mediasdk 19.4.0: initial package, path for plugins.cfg, move lib/mfx to lib * ffmpeg 4.2.1: fix #7030 from git.videolan.org for _qsv * intel-vaapi-driver: update from v2.1.0 to v2.4.0

th0ma7 added 12 commits December 30, 2019 22:22

gmmlib 19.4: Intel Graphics Memory Management Library

33c4e9f

gmmlib 19.4: Remove unecessary commented line

dd7ca11

libdrm 2.4.100: Disable radeon, amdgpu and nouveau drivers

749ac94

libva 2.6: Update for Intel Media driver dependency

43b9408

intel-media-driver 19.4.0r: initial draft package

dde42bb

intel-media-driver 19.4.0r: Build & install fixes

cc6dc4e

ffmpeg 4.2.1-20: Add Intel Quick Sync support

e1fd561

gmmlib 19.4.1: Remove invalid CAMKE variables

e6c62a3

intel-media-driver 19.4.0r: Fix install PATH

eb01d48

intel-mediasdk 19.4.0: Initial package

097e562

PLIST fixes for intel quick sync

1e46c2c

ffmpeg: add intel-media-sdk dependency for intel quick sync support

6acd6a6

th0ma7 requested a review from ymartin59 January 2, 2020 03:14

th0ma7 added enhancement request to enhance existing package status/needs-review labels Jan 2, 2020

th0ma7 self-assigned this Jan 2, 2020

gmmlib 19.4.1: Fix for x86 32-bit compilation for evansport arch

a4f7202

th0ma7 added 2 commits January 8, 2020 07:51

Reorder and remove dupicate mfx_hevcd

c14d5c6

intel-mediasdk 19.4.0: Apparmor fix to move lib/mfx under lib

d03bdc6

th0ma7 added 2 commits January 8, 2020 10:42

intel-vaapi-driver: Update from v2.1.0 to v2.4.0

bec485e

FFMPEG 4.2.1-22: VAAPI driver update and apparmor mfx driver fix

a700575

th0ma7 mentioned this pull request Jan 9, 2020

FFMPEG: add Intel Media SDK - Quick Sync support #3830

Closed

th0ma7 added status/ready-to-merge status/to-publish labels Jan 10, 2020

ymartin59 approved these changes Jan 18, 2020

View reviewed changes

ymartin59 removed the status/needs-review label Jan 18, 2020

ymartin59 merged commit 2f3aa0e into SynoCommunity:master Jan 18, 2020

th0ma7 deleted the ffmpeg-IntelQuickSync branch January 18, 2020 13:47

ymartin59 removed the status/to-publish label Jan 19, 2020

hgy59 removed the status/ready-to-merge label Nov 10, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Ffmpeg intel quick sync #3842

Ffmpeg intel quick sync #3842

th0ma7 commented Jan 2, 2020 •

edited

Loading

th0ma7 commented Jan 2, 2020

th0ma7 commented Jan 2, 2020

th0ma7 commented Jan 2, 2020

th0ma7 commented Jan 2, 2020

BenjaminPoncet commented Jan 2, 2020

th0ma7 commented Jan 2, 2020

BenjaminPoncet commented Jan 2, 2020

ymartin59 commented Jan 2, 2020

BenjaminPoncet commented Jan 2, 2020

kc6108 commented Jan 3, 2020

th0ma7 commented Jan 3, 2020

th0ma7 commented Jan 3, 2020

kc6108 commented Jan 8, 2020 •

edited

Loading

kc6108 commented Jan 8, 2020

th0ma7 commented Jan 8, 2020

BenjaminPoncet commented Jan 8, 2020 •

edited

Loading

th0ma7 commented Jan 8, 2020

th0ma7 commented Jan 8, 2020 •

edited

Loading

BenjaminPoncet commented Jan 8, 2020

th0ma7 commented Jan 9, 2020

BenjaminPoncet commented Jan 9, 2020

th0ma7 commented Jan 9, 2020

th0ma7 commented Jan 16, 2020 •

edited

Loading

ymartin59 commented Jan 18, 2020

ymartin59 commented Jan 18, 2020

th0ma7 commented Jan 18, 2020

ymartin59 commented Jan 18, 2020

Ffmpeg intel quick sync #3842

Ffmpeg intel quick sync #3842

Conversation

th0ma7 commented Jan 2, 2020 • edited Loading

Checklist

th0ma7 commented Jan 2, 2020

th0ma7 commented Jan 2, 2020

th0ma7 commented Jan 2, 2020

th0ma7 commented Jan 2, 2020

BenjaminPoncet commented Jan 2, 2020

th0ma7 commented Jan 2, 2020

BenjaminPoncet commented Jan 2, 2020

ymartin59 commented Jan 2, 2020

BenjaminPoncet commented Jan 2, 2020

kc6108 commented Jan 3, 2020

th0ma7 commented Jan 3, 2020

th0ma7 commented Jan 3, 2020

kc6108 commented Jan 8, 2020 • edited Loading

kc6108 commented Jan 8, 2020

th0ma7 commented Jan 8, 2020

BenjaminPoncet commented Jan 8, 2020 • edited Loading

th0ma7 commented Jan 8, 2020

th0ma7 commented Jan 8, 2020 • edited Loading

BenjaminPoncet commented Jan 8, 2020

th0ma7 commented Jan 9, 2020

BenjaminPoncet commented Jan 9, 2020

th0ma7 commented Jan 9, 2020

th0ma7 commented Jan 16, 2020 • edited Loading

ymartin59 commented Jan 18, 2020

ymartin59 commented Jan 18, 2020

th0ma7 commented Jan 18, 2020

ymartin59 commented Jan 18, 2020

th0ma7 commented Jan 2, 2020 •

edited

Loading

kc6108 commented Jan 8, 2020 •

edited

Loading

BenjaminPoncet commented Jan 8, 2020 •

edited

Loading

th0ma7 commented Jan 8, 2020 •

edited

Loading

th0ma7 commented Jan 16, 2020 •

edited

Loading