On Tue, Apr 5, 2016 at 9:04 AM, Marek Olšák <mar...@gmail.com> wrote: > On Mon, Apr 4, 2016 at 2:43 PM, Rob Clark <robdcl...@gmail.com> wrote: >> Hmm, what are the requirements of eglClientWaitSyncKHR() about having >> current context bound? Not being thread-safe (ie. being associated w/ >> current ctx) is pretty important to allow apps that create fences at >> in-opportune times to not force a mid-frame flush for tilers.. :-( > > There are 2 kinds of fences already: > - owned by pipe_screen: pipe_fence_handle > - owned by pipe_context: PIPE_QUERY_GPU_FINISHED > > I think the latter is what you want.
yes, but very much no... query is a pretty horrible interface for this (for same/similar reason that timestamp queries are horrible for tilers) since it assumes fences are actually a single point in gpu rendering pipeline ;-) When I get some time for v2 of this patchset, I think adding an optional (might be null) pipe_context pointer to screen->fence_finish() seems like the most reasonable option. BR, -R > Marek _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
