Ben Widawsky <benjamin.widaw...@intel.com> writes: > Here is one proposal to fix the issue. I noticed that only formats > without alpha were failing. This sucks for RGBX formats (which > technically aren't fast clearable based on the surface format). The > hunk for moving the format should happen regardless of this patch.
If we do this we could limit it to just multisample buffers. I think the problem only happens when sampling the cleared texels but we effectively never do that for single-sampled surfaces because in that case we always do a resolve before using them as textures. Being able to fast clear a single-sampled RGBX window system buffer is probably one of the more useful use cases so it's probably worth allowing that. We could also remove the explicit check the those two FLOAT16 formats in that case because neither of them have alpha bits. > Neil has another patch which attempts to not disable formats, and > workaround the hardware issue that we cannot explain. I just wanted to > put this on the list. If anyone's interested, the patch is here: https://github.com/bpeel/mesa/commit/2c7b2ddc8dba388665c258a1182 Obviously I'm not suggesting we merge that patch because it's a massive kludge. The test case which we're using to find these issues is here: http://patchwork.freedesktop.org/patch/64578/ - Neil _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
