[Mesa-dev] [PATCH 0/3] r600g removal of unnecessary cache flushes

Grazvydas Ignotas notasas at gmail.com
Mon May 30 23:58:44 UTC 2016

I've patched the CS validator and there seem to be some regressions:

arb_map_buffer_range 11/11 5/11
copybuffersubdata decrement-offset
copybuffersubdata increment-offset
copybuffersubdata offset=0
map_invalidate_range_bit decrement-offset
map_invalidate_range_bit increment-offset
map_invalidate_range_bit offset=0
arb_texture_buffer_object 77/78 76/78
arb_vertex_buffer_object 16/16 13/16
vbo-subdata-many drawarrays
vbo-subdata-many drawelements
vbo-subdata-many drawrangeelements

The "good" run was done on newer git master than what your
r600-opt-flushes branch is based on, don't know if it has any effect.
Should I attempt to bisect?


On Tue, May 31, 2016 at 1:30 AM, Marek Olšák <maraeo at gmail.com> wrote:
> On Tue, May 31, 2016 at 12:17 AM, Grazvydas Ignotas <notasas at gmail.com> wrote:
>> I've tried it on AMD JUNIPER (DRM 2.43.0 / 4.6.0) and it does not pass
>> the kernel CS validator with
>> [drm:evergreen_packet3_check.isra.14 [radeon]] *ERROR* Packet3 opcode
>> 42 not supported
>> printed many times.
>> I guess it needs some kernel patch? I haven't found anything relevant
>> in Alex's kernel git.
> Yeah, it needs a kernel patch which doesn't exist yet I think.
> Marek

More information about the mesa-dev mailing list