This code is a C++ port of John Costella's 'Magic Kernel', which he wrote in C#. It is a 1:3:3:1 convolution filter in one dimension.
It was originally implemented as part of a simple bitmap class and simple file format. The bitmap class handles 16-bit linear data primarily, in 4 channels. I am using this for stereoscopic computer graphics for digital cinema.
The file format is defined in simple.h. The bitmap object is defined in linear16.h.
Power-of-two upsampling and downscaling is straight forward through the respective methods declared in linear16.h (Downscale/Upscale). Although these are 2D scalers, it is possible to scale in only one dimension: e.g., UpscaleX() and UpscaleY().There are also some other image processing goodies there, which can safely be ignored. The main scaling work in done in C functions in filter16.cpp
I am sure there are several optimizations that could be done to the core filter code. But it works.
Lastly, the linear16 bitmap class has can read and write 24-bit BMP files, converting back and forth between 8- and 16-bit primaries, but ignoring sRGB-to-lin conversion, and vice versa (I'm lazy). It would be relatively easy to add read/write methods for PNG & jpeg.
There is a basic demo program with two sample images. It builds and runs on Linux (Ubuntu).