On Tue, Jun 7, 2016 at 11:02 AM, Marek Olšák <mar...@gmail.com> wrote: > On Tue, Jun 7, 2016 at 9:14 AM, Michel Dänzer <mic...@daenzer.net> wrote: >> On 02.06.2016 21:20, Michel Dänzer wrote: >>> On 02.06.2016 00:35, Marek =?UNKNOWN?B?T2zFocOhaw==?= wrote: >>>> Module: Mesa >>>> Branch: master >>>> Commit: fc1479a95432f291623fa5bafe524701e77af3ca >>>> URL: >>>> http://cgit.freedesktop.org/mesa/mesa/commit/?id=fc1479a95432f291623fa5bafe524701e77af3ca >>>> >>>> Author: Marek Olšák <marek.ol...@amd.com> >>>> Date: Thu May 12 13:33:06 2016 +0200 >>>> >>>> gallium/radeon: degrade tiled textures mapped often to linear >>>> >>>> Reviewed-by: Nicolai Hähnle <nicolai.haeh...@amd.com> >>>> Reviewed-by: Bas Nieuwenhuizen <b...@basnieuwenhuizen.nl> >>> >>> This commit broke the piglit test spec@!opengl 1.2@tex3d-maxsize on my >>> Kaveri (radeon driver, 1GB VRAM, 2GB GTT). It prints lots of lines >>> >>> radeon: Not enough memory for command submission. >>> >>> to stderr, with corresponding lines >>> >>> [drm:radeon_cs_ioctl [radeon]] *ERROR* Failed to parse relocation -12! >>> >>> in dmesg. >> >> Marek, any ideas for this regression? Do you need more information about it? > > Sorry, I have yet to test it. Tonga & Fiji (amdgpu) aren't affected.
I think the problem is that textures are added to the buffer list and then the draw call checks memory usage and flushes accordingly. The problem is the textures are not removed from the buffer list at that point. Example: - Limit: 2 GB GTT - Draw call 1: 1 GB GTT usage - Draw call 2: +1.5 GB GTT usage - Flushes the IB without draw call 2 but with 2.5GB GTT usage. Marek _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
