Created attachment 134777 [details] Screenshot from commit ezBench Vulkan trace render validation caught a change in DOTA2 rendering from: ------------------------------------------------------------------- commit a62a97933578a813beb0d27cc8e404850f7fd302 Author: Lionel Landwerlin <lionel.g.landwerlin@intel.com> AuthorDate: Wed Jul 19 12:14:19 2017 +0100 Commit: Lionel Landwerlin <lionel.g.landwerlin@intel.com> CommitDate: Fri Oct 6 16:32:20 2017 +0100 anv: enable multiple planes per image/imageView This change introduce the concept of planes for image & views. It matches the planes available in new formats. We also refactor depth & stencil support through the usage of planes for the sake of uniformity. In the backend (genX_cmd_buffer.c) we have to take some care though with regard to auxilliary surfaces. Multiplanar color buffers can have multiple auxilliary surfaces but depth & stencil share the same HiZ one (only store in the depth plane). v2: by Jason Remove unused aspect parameters from anv_blorp.c Assert when attempting to resolve YUV images Drop redundant logic for plane offset in make_surface() Rework anv_foreach_plane_aspect_bit() Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Reviewed-by: Jason Ekstrand <jason@jlekstrand.net> ------------------------------------------------------------------- Change is in the "Talent tree" rendering in the game bottom bar. I wasn't able to see the faulty rendering with latest DOTA2, nor a trace from the latest DOTA2 (taken using latest vktrace). It's only visible with a trace from slightly (1-2 month) older DOTA2 (taken with vktrace form that time). If you think bug is somewhere else than Mesa, just mark this as NOTOURBUG (or INVALID).
Created attachment 134778 [details] Screenshot before commit
Fixed by a07f7b26198ce0f5c8799481a673754968ac5daf anv/cmd_buffer: Take bo_offset into account in fast clear state addresses.
Because this regression affected a popular commercial software title, and because it was NOT detected by automated tests, I think it should remain open until a test exists which prevents future regressions. Lionel, can you write a crucible test to exercise your code path and demonstrates this regression?
I've written a test that uses the fast clear but with a07f7b26198ce0f5c8799481a673754968ac5daf reverted, I can't get this test to fail... The bug must be a bit more complicated than what I understood.
Anyway the old trace can be provided?
https://patchwork.freedesktop.org/patch/189648/
Anybody fancy reviewing the crucible patch in the previous comment? Or should I just land it?
Please land the test with or without review.
Done.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.