Add support for commit-timing-v1. no wlroots #1230
Conversation
src/wlserver.cpp
Outdated
|
|
||
| #define NANOS_IN_SEC 1000000000 | ||
| static uint64_t timer_get_target_present_nsec(uint64_t timestamp_nsec, uint32_t rounding_mode) { | ||
| uint64_t refresh_nsec = NANOS_IN_SEC/(GetVBlankTimer().GetRefresh()/1000); |
There was a problem hiding this comment.
-
We have utilities to do refresh -> cycle already
mHzToRefreshCycle -
Doesn't all this rounding logic need to actually be where we check the timestamp in steamcompmgr?
Right now this problem will only manifest on hotplug, which isn't a big deal -- but some day we will support multiple displays, so we should move it there.
There was a problem hiding this comment.
(I know that adding stuff to the ResListEntry whatever crap is painful, it is high on my to-do to rework that garbage soon and have wlserver create the commits directly).
src/wlserver.cpp
Outdated
| target_nsec -= (target_nsec % refresh_nsec); | ||
| } | ||
| else if (rounding_mode == TIMER_ROUNDING_NOT_BEFORE) { | ||
| target_nsec = ((target_nsec+(refresh_nsec-1))/refresh_nsec)*refresh_nsec; |
There was a problem hiding this comment.
I don't get why we are touching anything for TIMER_ROUNDING_NOT_BEFORE.
There is no world where we need to normalize it to a refresh cycle, and we probably shouldn't because multiple displays.
There was a problem hiding this comment.
Moved this logic to steamcompmgr, leaving out this case.
src/wlserver.cpp
Outdated
|
|
||
| for (const auto& commit : *commits) | ||
| { | ||
| if (commit.desired_present_time > target_present_nsec) |
There was a problem hiding this comment.
This logic doesn't work because some pending commits have already probably moved to steamcompmgr thread already.
Unfortunately Gamescope is not set up to really allow you to check something like this given it's architecture.
I am happy to just drop the implementation of the protocol error for now.
There was a problem hiding this comment.
Actually, you could just store the highest used target_present_nsec that gets updated at commit time on the wlserver_wl_surface_info instead of checking every commit.
There was a problem hiding this comment.
You should also probably do that for your actual wlroots implementation instead of checking over every commit ;)
There was a problem hiding this comment.
Huh, nice...
From now on whenever I find myself mindlessly proliferating cpu cycles, I'll ask myself if I could do any better :D
However I think I'll just leave a very trivial check for now, since it appears to be controversial what to actually consider an error in this case.
Signed-off-by: Sergio Gómez <sergio.g.delreal@gmail.com>
There might be other done commits which would be lost otherwise. Signed-off-by: Sergio Gómez <sergio.g.delreal@gmail.com>
sergio-gdr
force-pushed
the
commit_timing_internal
branch
from
652327a
to
c405f4a
Compare
Second WIP attempt at bringing support for commit-timing-v1 to gamescope.
Previous attempt uses wlroots' internal infrastructure.
The protocol offers 2 target stages: presentation and latching. I've ignored the distinction for now and just take the provided timestamp to be intended for "presentation".
I've also not tried to work through how a timestamp request would interact with the current implementation of fifo-v1 (something for a future iteration).