While the math seems to have been left as an exercise for the reader, the mask table is basically the same data as the bitmap - just with whole bytes filled in instead of individual bits. Or rather, with whole pixels filled in - whatever that pixel size may be.

A brief explanation of the idea is that you have your bitmap representation of a glyph, made up of a byte per row, with bits representing columns, and you take this bitmap and expand it out so that each bit is now 8 bits. Note that for an 8-bit bitmap row, this means a mask row is 8 bytes - the example code seems to assume 32-bit values, so you need to two of them, but I'm lazy and will use 64-bit values. So a row of 01101100b becomes 0x00FFFF00_FFFF0000. Except... we're probably on a little-endian machine, so this is actually backwards. It should be byte reversed to 0x0000FFFF_00FFFF00. We do this for every row in our bitmap and then the masking can be used to write multiple pixels at a time, which is probably faster than individual calls to a set-pixel function.