[Cin] ffmpeg - cuda

sergiogomez at tostado.com.ar sergiogomez at tostado.com.ar
Fri Apr 10 18:29:28 CEST 2020


Thanks for reply! I could compile CinGG with cuda, I used debian's
package nvidia-cuda-sdk to compile ffmpeg and CinGG too, but cuda
samples is needed too (basically, helper_cuda.h and few more). Debian
use /usr/lib/cuda for libs and /usr/bin for nvcc, so I must to create a
symbolic to /usr/lib/cuda/bin/nvcc (export variable path was not
sufficient, compilation procces search nvcc un CUDA_PATH/bin/nvcc)
that's all. 

My original question was: Can I pass some imput parameters to ffmpeg in
render process? And I ask this again. Why? Becase ffmpeg -hwaccel cuvid
can enconder x264 at 13,7x and without this parameter, just 2.5x. And at
the moment, x264 is sufficient for me. 

Thanks a lot! Sincerelly. 

Sergio 

El 2020-04-10 01:21, Phyllis Smith escribió:

> GG writes: 
> There are 3 classes of hw accel in cinelerra via ffmpeg.  vaapi, vdpau, and cuda.  The generic hw interface of ffmpeg is use to select hw decoding using the selected preference in the settings->preferences->performance->use hw device popup menu.  The decode_activate operation attempts to decode the first hw frame, and if the initialization fails then the sw decoder is used as a fallback.  The normal 3rd party build will configure and build a current version of ffmpeg with a few patches that provide improved operation.  I would recommend using it.   The standard cmdline versions of ffmpeg and ffprobe are built into the thirdparty build directory for ffmpeg with the normal static library builds. 
> 
> To use the hw for transcodes is not exactly straight forward from the command line.  For example, vdpau is decode only. 
> There are free and non-free versions, and the non-free versions are not in the 3rd party build.  There is a vaapi transcode src code example in ffmpeg-4.2/doc/examples/vaapi_transcode.c.  It probably only works with a limited set of codec formats. 
> It is a good idea to load, vainfo, vdpauinfo.  See "hardware acceleration" in the "Performance andTips" chapter 19 of CinelerraGG_Manual.pdf on the website. 
> 
> Various nvidia codecs require non free api library interfaces.  The implementation may require at least a certain version of nvidia software or library baselines before it can work properly.  I would have to study the exact problem you are trying to solve, and the code support before it can be determined if the library can do the problem you wish to solve. 
> 
> Please feel free to ask specific questions.  gg 
> 
> On Thu, Apr 9, 2020 at 7:14 PM <sergiogomez at tostado.com.ar> wrote: 
> 
>> Hi all. I'm a debian user, but today I compiled ffmpeg (why?) with hardware acceleration (based on https://developer.nvidia.com/ffmpeg ) 
>> 
>> I want to use this ffmpeg version to encode (rendering) in CinGG. It's possible without --with-cuda option in CinGG compilation?  
>> 
>> I can use "ffmpeg -hwaccel cuvid " in console to transcode, but those are input parameters, not just output options. 
>> 
>> Thaks all 
>> 
>> Sergio -- 
>> Cin mailing list
>> Cin at lists.cinelerra-gg.org
>> https://lists.cinelerra-gg.org/mailman/listinfo/cin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20200410/dc01cf95/attachment.html>


More information about the Cin mailing list