I use your filter, but the kernel can't pass the compile, you should consider the
"double" type in the kernel, some GPU card does not support double type
I add "#pragma OPENCL_EXTENSION cl_khr_fp64: enable " to the kernel, but it
does not works
I will check the error tomorrow
I tested this filter on Nvidia on Linux, using driver 355 and on the
Intel CPU-based OpenCL SDK so far.
Using floats potentially has an impact on the keying quality.
ffmpeg-devel mailing list