On 6 July 2017 at 13:46, Deucher, Alexander <Alexander.Deucher at amd.com> wrote: >> Attach it to analogous primitive? > > Radeon libdrm is much different than amdgpu. There is no analog. > Upon a closer look, indeed there isn't. Must have gotten confused earlier. >> >> > I think the current radeon API is simpler. Maybe a follow up change can >> change amdgpu's API similar to radeon. >> > >> Exposing 3 entry points instead of 1 is _not_simpler. Also you cannot >> change the existing API, since it also breaks the ABI. >> Leading to crash/cause memory corruption when using existing binaries. >> >> >> - is adding yet another header really justified? >> > radeon_asic_id.h? That is going to be used by ddx/mesa. >> > >> Where it's used is orthogonal. You don't need a separate _public_ >> header for nearly every entry point ;-) > > Actually having a separate header makes sense for radeon. We currently expose a separate header for each set of functionality (one for buffer management, one for command submission, one for surface management). Adding the asic names to any of the existing ones doesnâ??t really make sense from a functional standpoint. > Seems a bit unfortunate, but you have a point. Having three extra entry points seems a bit much, but it seems like the best one can do atm. Thanks Emil
