This commit introduces bopti for fxcg50 with the first basic 16-bit
formats. The performance is rather slow, especially for large images,
and will need refinements and/or overclock to be really efficient in
full-screen real-time applications.
This commit introduces bopti for fx-CG 50. Currently the only
interfaces are the bopti_render_{clip,noclip} functions, and the
only supported formats are r5g6b5 and r5g6b5a.
The algorithm for r5g6b5 is optimized to perform longword accesses
using movua.l, whereas the algorithm for r5g6b5a uses plain word
accesses because transparency checks feel more difficult than one
more loop iteration.
These algorithms are still slow for large surfaces and struggle to
keep up 25 FPS in full-screen, so possible improvements with the
DMA should definitely be tested before restorting to overclock.
lephe