[Bug 105458] [CI] igt at kms_ccs@pipe-[a|b]-crc-sprite-planes-basic - fail - Test assertion failure function try_config, Failed assertion: ret == 0

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=105458

Matt Roper <matthew.d.roper at intel.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |matthew.d.roper at intel.com
           Assignee|juhapekka.heikkila at gmail.co |matthew.d.roper at intel.com
                   |m                           |

--- Comment #10 from Matt Roper <matthew.d.roper at intel.com> ---
The problem here is the DDB algorithm that was initially proposed by the bspec.
 A pipe's DDB allocation is divided up between its active planes according to
the % of the total data rate that they contribute.  That's usually a reasonable
heuristic, but breaks down when you have a large difference in the size (data
rate) of your planes.  If you divide the DDB proportionally, the very small
plane will get such a small allocation that it will fail to even achieve it's
level 0 watermark requirements.  This problems pops up most often on APL (due
to its smaller DDB size in general) and/or on systems with high display
resolutions (due to the greater size disparity between "full-screen plane" and
"256x256 plane".

I've posted a patch series here:

   https://patchwork.freedesktop.org/series/53682/

which switches to an alternate DDB algorithm aimed at avoiding these kinds of
problems.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-gfx-bugs/attachments/20181206/e41d2f6c/attachment.html>