how to cross compile ffmpeg with nvidia cuda_sdk and libnpp #249

CC	libavfilter/vf_scale_cuda.o
NVCC	libavfilter/vf_scale_cuda.ptx
ERROR: No supported gcc/g++ host compiler found, but clang-3.8 is available.
       Use 'nvcc -ccbin clang-3.8' to use that instead.
ffbuild/common.mak:95: recipe for target 'libavfilter/vf_scale_cuda.ptx' failed
make: *** [libavfilter/vf_scale_cuda.ptx] Error 1
Any suggestions on where to from here to cross-compile it ?
Roughie Process so far:
Install this package under an ubuntu 17.04 VM, to get access to the nvcc compiler. It isn't the latest version, but should still work.

sudo apt-get install -y nvidia-cuda-toolkit
This isn't enough, we also need the Windows libs and includes, so download the relevant cuda toolkit exe under Windows 10

from the nvidia developer site, https://developer.nvidia.com/cuda-downloads

cuda_8.0.61_win10.exe
Then run cuda_8.0.61_win10.exe in Windows 10.

It'll extract the files to a temporary folder before it then tries to install the stuff.

After it's finished extracting and then whilst it's figuring out how to install, copy the full extracted folder tree to another place so we have our own copy if it.

After our own folder copies have finished, cancel the installation.
Locate folders cuda/include and cuda/lib/x64 in the newly copied folder tree and copy them to somewhere else alongside one another.

Then also locate folders /npp/include and /npp/lib/x64 and also copy them to somewhere else alongside one another.
Back under Linux Ubuntu, use sudo to create folder cuda_8.0.61 under /usr and then set permissions on it sort of like this

[code]sudo mkdir /usr/cuda_8.0.61

chmod +777 -R /usr/cuda_8.0.61[/code]
Copy those 4 folder trees we copied earlier, from Windows to the Ubuntu box to become like these

[code]/usr/cuda_8.0.61/cuda/include

/usr/cuda_8.0.61/cuda/lib/x64

/usr/cuda_8.0.61/npp/include

/usr/cuda_8.0.61/npp/lib/x64[/code]
Now we should be able to type nvcc  and see it try to compile something and fail which is OK

By now the necessary folders and whatnot exist (well, nearly) for ffmpeg to be built (cross-compiled) with cuda sdk.
Edit the ffmpeg cross-compile build script and add these flags to the ffmpeg configure, to cross-compile stuff with the cuda sdk

[code]--enable-cuda-sdk --enable-filter=scale_cuda --extra-cflags=-I/usr/cuda_8.0.61/cuda/include --extra-ldflags=-L/usr/cuda_8.0.61/cuda/lib/x64 [/code]
Then if you're game, also add these flags to the ffmpeg cross-compile configure line

[code]--enable-libnpp --extra-cflags=-I/usr/cuda_8.0.61/npp/include --extra-ldflags=-L/usr/cuda_8.0.61/npp/lib/x64[/code]
Try to cross-compile ffmpeg and see how the luck holds. [i](edit: it blows up, hence this post)[/i]
edit: interesting info here.  i think i'm stuck

https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=862522