aco: implement 16-bit derivatives and improve fp16 support on gfx8
Both of these are probably useful for radeonsi, which uses 16-bit derivatives and can utilize 16-bit floating point on GFX8.
Due to an influx of spam, we have had to impose restrictions on new accounts. Please see this wiki page for instructions on how to get full permissions. Sorry for the inconvenience.
Both of these are probably useful for radeonsi, which uses 16-bit derivatives and can utilize 16-bit floating point on GFX8.