Reminds me of ye olden days when kernel transforms were merely weighted multiplicative and/or additive matrixes applied to every point in the source arriving at

Reminds me of ye olden days when kernel transforms were merely weighted multiplicative and/or additive matrixes applied to every point in the source arriving at pixel data in the target. Blur, sharpen, color channel filter, color swap, invert, etc. An extremely diagonalizable problem suitable for massive parallelism and concurrent calculation because there is little/no dependency on prior calculations.