[FFmpeg-devel] npp scaling

[Timo Rothenpieler]

On 14/08/2025 15:17, Marco Oman via ffmpeg-devel wrote:
> Hi all,
> I am trying to compile FFmpeg with the --enable-libnpp flag configure on Ubuntu 24.04.
> 
> but I am getting an error in file vf_scale_npp.c
>> CC      libavfilter/vf_scale_npp.o
>> libavfilter/vf_scale_npp.c: In function 'nppscale_deinterleave':
>> libavfilter/vf_scale_npp.c:715:15: error: implicit declaration of function 'nppiYCbCr420_8u_P2P3R'; did you mean 'nppiYCbCr420_8u_P2P3R_Ctx'? [-Werror=implicit-function-declaration]
>>    715 |         err = nppiYCbCr420_8u_P2P3R(in->data[0], in->linesize[0],
> 
> After some investigation work I found that I have CUDA 13.0 on my machine and header file nppi_color_conversion.h installed by CUDA no have a 'double version' for each function, e.g. nppiYCbCr420_8u_P2P3R and nppiYCbCr420_8u_P2P3R_Ctx where the latter takes an extra argument of type NppStreamContext. It has only the latter, and so compilatioon breaks
> 
> So it looks like sooner or later there will be the need to update vf_scale_npp.c to make it compile under these new conditions (without breaking the old).
> 
> Anyone got the same problem? Anyone has plans for a fix?

I consider the npp based filters deprecated and have no plan to fix them.
I'll instead look into formalizing that, and making configure correctly 
disable them with incompatible versions of the libraries.

Use scale_cuda instead.