On Sun, 22 Sep 2019 09:26:45 -0400 Alyssa Rosenzweig <alyssa.rosenzw...@collabora.com> wrote:
> > +your collabora address > > Thank you > > > > > > I think this batch_add_bo should probably dropped altogether? This > > > > > loop > > > > > is dealing with constant buffers; the shaders themselves were added > > > > > > > > > > > > > I'll double check. I couldn't find where BOs containing shader programs > > > > were added last time I looked. > > > > > > Masking a real bug :o > > > > > > It should probably happen in panfrost_patch_shader_state....? > > > > Ok, I'll add it there, I wasn't sure this function was called for all > > shaders, but looking at the code a second time it seems to be the case. > > I think so as well, yeah. > > > > As I stated before, I thought we should be adding the BO for > > > wallpapering when we actually wallpaper, since that's a slow path. Not > > > wallpapering is the default and ideally what most apps should do. > > > > Wallpapering happens too late (when we are flushing the batch) to have > > an impact on the dep graph, but we can probably know that wallpapering > > will be needed before that. My question remains though, are > > vertex/tiler supposed to touch the texture BO they're reading from, or > > should we only flag the BO for FRAGMENT use. > > Vertex/tiler should not touch the texture BO, unless you're texturing > from the vertex shader (which we're not for wallpapering). Okay, then adding the READ flag isn't really needed: a WRITE is more constraining than a READ, and the FRAGMENT job already writes the FBO BOs. _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
