Figure out how to avoid redundant resource synchronization for read-only operations
e.g. a LUT will be read from 1000 times in a row, no need to have synchronization between each usage. How to fix?
- Hard-code disable synchronization for all textures which don't support writing (easy)
- Figure out how to somehow get
tex_signal
/tex_barrier
to skip over redundant operations (hard/impossible?)