diff options
author | Archit Taneja <architt@codeaurora.org> | 2017-03-23 13:28:15 +0300 |
---|---|---|
committer | Rob Clark <robdclark@gmail.com> | 2017-04-08 13:59:36 +0300 |
commit | 8480adacfd58c79d884aa17af77a7fa7efae7257 (patch) | |
tree | ad1578323806b266f0f6e010271b75cc61c0bba8 /tools/perf/scripts/python/call-graph-from-postgresql.py | |
parent | 359ae86248a5f92dd475e2d8012364a97e57026a (diff) | |
download | linux-8480adacfd58c79d884aa17af77a7fa7efae7257.tar.xz |
drm/msm/mdp5: Assign 'right' mixer to CRTC state
Dynamically assign a right mixer to mdp5_crtc_state in the CRTC's
atomic_check path. Assigning the right mixer has some constraints,
i.e, only a few LMs can be paired together. Update mdp5_mixer_assign
to handle these constraints.
Firstly, we need to identify whether we need a right mixer or not.
At the moment, there are 2 scenarios where a right mixer might be
needed:
- If any of the planes connected to this CRTC is too wide (i.e, is
comprised of 2 hwpipes).
- If the CRTC's mode itself is too wide (i.e, a 4K mode on HDMI).
We implement both these checks in the mdp5_crtc_atomic_check(), and
pass 'need_right_mixer' to mdp5_setup_pipeline.
If a CRTC is already assigned a single mixer, and a new atomic commit
brings in a drm_plane that needs 2 hwpipes, we can successfully commit
this mode without requiring a full modeset, provided that we still use
the previously assigned mixer as the left mixer. If such an assignment
isn't possible, we'd need to do a full modeset. This scenario has been
ignored for now.
The mixer assignment code is a bit messy, considering we have at most
4 LM instances in hardware. This can probably be re-visited later with
simplified logic.
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Diffstat (limited to 'tools/perf/scripts/python/call-graph-from-postgresql.py')
0 files changed, 0 insertions, 0 deletions