I found the most useful, when the protocol spec contained also an example of input and output of their CRC calculation (either directly, or indirectly as an example packet). I then play with the mentioned CRC calculator page as long as I obtain the required result, and then reach for one of the solutions I already have pre-cooked and deform it in the required way.

Too often the "example implementation" misses "a few small details", e.g. the init value, or that one extra cycle has to be performed at the end - and only the "core routine" is given (preferrably in some arcane language or pseudolanguage, with constants given in octal...).

I found usually this in protocols covered by some sort of NDA - of course these are not subject to public scrutiny, and too often the author of the protocol is immune to rambling of the poor designer ...

JW