GPU Acceleration

[sub]

There was a bug fix on a couple transcoder options a couple of weeks back. I thought it was before the stable release, but maybe it was after it. It does seem to be working fine for INTEL and D3D11VA here.

Maybe try the latest NShared.dll as a next step.

[mvallevand]

Better for qsv but directx still doesn't support resolution profile

Code:

2020-04-10 21:01:15.619    [DEBUG][22]    Using directx encoder
2020-04-10 21:01:15.619    [DEBUG][22]    Default transcoder args: -y [SEEK] -hwaccel:v:0 d3d11va -hwaccel_device:v:0 0 -i [SOURCE] -map_metadata -1 -threads 0 -ignore_unknown -map 0:v:0? -map 0:a:[AUDIO_STREAM] -map -0:s -codec:v:0 h264_qsv -b:v [BITRATE] -maxrate [BITRATE] -bufsize [BUFSIZE] -codec:a:0 [AUDIO_CODEC] -strict experimental -ac [AUDIO_CHANNELS] -ab [AUDIO_BITRATE] -hls_time [SEGMENT_DURATION] -start_number 0 -hls_list_size [SEGMENT_COUNT] -y [TARGET]

I would love to see the decoder parameter passed one day too

Martin

[aderlopas]

There was a bug fix on a couple transcoder options a couple of weeks back. I thought it was before the stable release, but maybe it was after it. It does seem to be working fine for INTEL and D3D11VA here.

Maybe try the latest NShared.dll as a next step.

Thanks Sub, I have tried both with the downloaded DLL and they still both fail.
Attaching the logs.

[aderlopas]

If this is working on both your sides then could it be the type of stream that it comes from my aerial in Greece?
That made me thinking also why with the below parameters is working and what these parameters are...

-hwaccel_device:v:0 0 -init_hw_device qsv=hw -filter_hw_device hw  and -vf hwupload=extra_hw_frames=64,format=qsv

Just reading some more, you guys tell me if these are relevant or not...

From: http://git.videolan.org/?p=ffmpeg.git;a=...e76da44758

extra_hw_frames is needed for some cases.
(eg: qsv decoding + qsv vpp + qsv look_ahead encoding, transcoding
failed if no extra_hw_frames supported:

From http://www.ffmpeg-archive.org/Issue-with...91919.html

(a). Unset -hwaccel_output_format option. When present, this disallows
downloading textures to system memory.

Do we want the textures in system memory or not here?

Also the below:

Where possible, stick to either full H/W decode+encode, OR skip H/W decode
entirely if the input format is likely to be unsupported by the underlying
device.
In the latter case, the use of the hwupload or hwupload_cuda filter(s) can
then be used to create the required CUDA contexts for filters that depend
on them, such as yadif_cuda, overlay_cuda, etc.
Note that with hwupload filter, a device reference is required to upload
hwframes to, typically via the -init_hw_device

Finally from:

https://github.com/Intel-Media-SDK/Media...h-MediaSDK

Also note on: 

Code:

-vf hwupload=extra_hw_frames=64,format=qsv

, this ensure that MFX runtime receives supported pixel format. 

The extra argument 

Code:

extra_hw_frames=64

 passed to 

Code:

hwupload

 has to do with the MFX runtime requiring a fixed frame pool size to be per-allocated.

[sub]

Sorry, I don't know. Seems weird you'd need these extra parameters but mine works absolutely fine without it. Might come down to source material or something.

[aderlopas]

Sorry, I don't know. Seems weird you'd need these extra parameters but mine works absolutely fine without it.  Might come down to source material or something.

Thanks Sub, do you want me to send you any feed to try?

Also if you add those to the command you build in nshared.dll does this have any adverse impact?

[sub]

Original command also works if I add:

-hwaccel_device:v:0 0 -init_hw_device qsv=hw -filter_hw_device hw  and -vf hwupload=extra_hw_frames=64,format=qsv

"C:\Program Files\NPVR\Other\ffmpeg.exe" -y  -hwaccel:v:0 d3d11va -hwaccel_device:v:0 0 -init_hw_device qsv=hw -filter_hw_device hw -i "http://127.0.0.1:8866/live?channeloid=7562&transcoder=3f149498-ce76-492c-a394-c0b2ab103364&client=5c53419a-db74-43eb-b730-b60c7506a7e8" -map_metadata -1 -threads 0 -ignore_unknown -map 0:v:0? -map 0:a:0? -map -0 -codec:v:0 h264_qsv -b:v 1500000 -maxrate 1500000 -bufsize 3000000 -codec:a:0 aac -strict experimental -ac 2 -ab 128000 -vf hwupload=extra_hw_frames=64,format=qsv -f null NUL

and it uses 19% CPU and 26% GPU

I just tried a build with these extra parameters, but ffmpeg was now failing for me, so I've rolled that change back.

[sub]

I have both nvidia and intel video devices in my machine. Those command line options might be assuming that d3d11va device 0 is the intel device, and it might not be on my machine.

I would just focus on the intel qsv command line for now (not the d3d11va), so that any commands a specific to intel. I'm pretty much finished my day now, so I'll see what you've posted in the morning.

[aderlopas]

Original command also works if I add:

-hwaccel_device:v:0 0 -init_hw_device qsv=hw -filter_hw_device hw  and -vf hwupload=extra_hw_frames=64,format=qsv

"C:\Program Files\NPVR\Other\ffmpeg.exe" -y  -hwaccel:v:0 d3d11va -hwaccel_device:v:0 0 -init_hw_device qsv=hw -filter_hw_device hw -i "http://127.0.0.1:8866/live?channeloid=7562&transcoder=3f149498-ce76-492c-a394-c0b2ab103364&client=5c53419a-db74-43eb-b730-b60c7506a7e8" -map_metadata -1 -threads 0 -ignore_unknown -map 0:v:0? -map 0:a:0? -map -0 -codec:v:0 h264_qsv -b:v 1500000 -maxrate 1500000 -bufsize 3000000 -codec:a:0 aac -strict experimental -ac 2 -ab 128000 -vf hwupload=extra_hw_frames=64,format=qsv -f null NUL

and it uses 19% CPU and 26% GPU

I just tried a build with these extra parameters, but ffmpeg was now failing for me, so I've rolled that change back.

Sorry to hear that Sub, but thanks for trying.

Maybe another approach would help here?

Can we possibly add in settings a field where we can add switches to ffmpeg to be added to the command as passthrough when ffmpeg command executes?
If that can be done then I would just need to add the below and it would work for me:

-init_hw_device qsv=hw -filter_hw_device hw -vf hwupload=extra_hw_frames=64,format=qsv

[sub]

You can override the string in the <TranscodeHLS> setting of config.xml.

Look at the "Using transcoder args" log message in nrecord.log to give you the starting point for replacing 'default' with.