Basically my fork could work too for S905 but first some project specific args have to be passed to several packages. For example Retroarch has specific build args for generic & RPi and IMHO S905 will then need some specific stuff too. Same goes for emulation cores, most of them have specific make flags according to the project. If an emulator needs optimizations for S905 this has to be passed too.
I believe I implemented a few changes for s905 but didn't spend much time optimizing. It all seems to work at reasonable speeds so I haven't spent anymore time on it.