07:30 Lynne: can compilers optimize imageLoad()+imageStore() if only a single component was updated?
08:54 glehmann: Lynne: at least for amd no, imageStore always writes all components
08:56 Lynne: I'm guessing there won't be that bad of a perf penalty to this, since cache lines are surely wider than 16-bits
08:57 Lynne: wait, imageStore/Loads don't bypass cache, do they? I know they bypass texel cache on most implementations, I think
11:59 mareko: they don't
12:00 mareko: it's complicated
12:04 Lynne: they don't bypass texel cache? I thought storage images were handled as nothing more than buffers when it came to accessing them
12:05 mareko: the cache hierarchy is mostly general purpose now
12:09 Lynne: nice