Compiling LinuxCNC using a graphic card

Those GPU's have relative simple processors but a load of them in parallel. It will only be feasible if you have a compiler which uses the GPU (does not exist AFAIK) and a lot of classes which are very similar (so you can benefit from the parallelism).

Compiling is something you do once and use the result often. Therefore it is more beneficial to invest in algorithms which unleash the power of the GPU and not in the tooling to compile those algorithms.