On Apr 12, 2016 7:52 AM, "Eirik Byrkjeflot Anonsen" <<a href="mailto:eirik@eirikba.org">eirik@eirikba.org</a>> wrote: > > Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>> writes: > > > This tiny series converts nir_var_mode to be a bitfield instead of a > > regular enum and then converts two of the passes that currently take a mode > > to handle the new bitfield. The reason for this is that a number of > > different passes want to be able to act on some subset of the modes and can > > run more efficiently if they are called once with that set rather than once > > per mode. > > > > I'm not 100% sold on making it a bitfield and asserting only one bit is set > > in nir_validate. This isn't the standard pattern in the mesa compiler > > stack today so it may throw people off. It still looks kind of strage to > > me. On the other hand, it does have the potential to make things look > > nicer and gdb will pretty-print it for you. For what it's worth, the > > Khronos group settled on using this pattern for bitfield-capable enums in > > Vulkan. Rob seems to prefer this over a regular enum and using (1 << mode) > > everywhere so, unless there are major objections, we'll probably go with > > it. > > Would it help to name them *_bit to signify the fact that they are bits > rather than values? We could, but such a change would touch over half of NIR for, I think, marginal benefit. > eirik