This might be responsible for the bad r300g MSAA performance results on Phoronix. I have no other explanation. There is an optimization in r300g which forces the VRAM domain for non-staging CB and DB only. Other than that, VRAM|GTT is the default for all textures and GTT is the default for all buffers, so it's pretty conservative. For the future buffer-eviction heuristics, we should take into account the actual resource usage - MSAA resources (and presumably scanout resources as well) should stay in VRAM and shouldn't be evicted by lesser resources unless the MSAA ones are idle for a very long time. MSAA resources are usually pretty big (128 MB is needed for Full HD 8x MSAA (CB+DB)), but that shouldn't stop the kernel from evicting as many less-important resources as is necessary to reserve enough space for MSAA. Marek On Thu, Jan 17, 2013 at 7:30 PM, Alex Deucher <alexdeucher@xxxxxxxxx> wrote: > One more fix on top. Just a revert of the bo placement patch that has > been causing corruption for a number of people. > > Revert "drm/radeon: do not move bo to different placement at each cs" > > This reverts commit d025e9e2b890db679f1246037bf65bd4be512627. > > This causes corruption for a number of users and needs further > investigation in the next cycle. > https://bugzilla.kernel.org/show_bug.cgi?id=52491 > https://bugs.freedesktop.org/show_bug.cgi?id=58659 > http://lists.freedesktop.org/archives/dri-devel/2013-January/032961.html > > Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx> > > Thanks! > > Alex > > > On Tue, Jan 15, 2013 at 9:21 AM, <alexdeucher@xxxxxxxxx> wrote: >> From: Alex Deucher <alexander.deucher@xxxxxxx> >> >> Hi Dave, >> >> Just a few small fixes. >> >> The following changes since commit 7b4cf994e4c6ba48872bb25253cc393b7fb74c82: >> >> udldrmfb: udl_get_edid: drop unneeded i-- (2013-01-14 08:45:27 +1000) >> >> are available in the git repository at: >> git://people.freedesktop.org/~agd5f/linux drm-fixes-3.8 >> >> Alex Deucher (1): >> drm/radeon: clear reset flags if engines are idle >> >> Jerome Glisse (1): >> drm/radeon: improve semaphore debugging on lockup >> >> Marek Olšák (1): >> drm/radeon: allow FP16 color clear registers on r500 >> >> drivers/gpu/drm/radeon/evergreen.c | 6 ++++++ >> drivers/gpu/drm/radeon/ni.c | 6 ++++++ >> drivers/gpu/drm/radeon/r600.c | 6 ++++++ >> drivers/gpu/drm/radeon/radeon.h | 2 ++ >> drivers/gpu/drm/radeon/radeon_drv.c | 3 ++- >> drivers/gpu/drm/radeon/radeon_ring.c | 2 ++ >> drivers/gpu/drm/radeon/radeon_semaphore.c | 4 ++++ >> drivers/gpu/drm/radeon/reg_srcs/rv515 | 2 ++ >> drivers/gpu/drm/radeon/si.c | 6 ++++++ >> 9 files changed, 36 insertions(+), 1 deletions(-) > _______________________________________________ > dri-devel mailing list > dri-devel@xxxxxxxxxxxxxxxxxxxxx > http://lists.freedesktop.org/mailman/listinfo/dri-devel _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/dri-devel
