Shifting long vectors between AVX registers (on pre AVX512VL CPUs)

Question

I in the process of making some 2D/planar image data resampler and current highest performance approach of 2D convolution need to perform shift of long float32 vector between AV