[mohdfaour03]

What

One of the host waits #1471 lists as still affected: VideoOutDriver::presentLoop waits on flipQueue->changed (a std::condition_variable) with a 10 ms timeout while no flip is queued, and pumps SDL events and the pad after each wake. On Windows that wait goes through winpthreads, which ends timed waits on GetTickCount64 steps, so the poll ended 11 to 24 ms later instead of 10 ms. WaitForFlipRoom waits on the same condition with a 60 s bound and has the same granularity.

A plain high-resolution sleep is not a replacement here: a flip submitted while the present thread waits must wake it at once. libkernel already has the notify-capable precise wait the guest side uses, TimedWait::Condition (event plus high-resolution waitable timer per waiting thread, std::condition_variable_any under APS5_COARSE_TIMED_WAITS), and libSceVideoOut already links libkernel. The change makes FlipQueue::changed a TimedWait::Condition, turns the two timed waits into WaitUntil with the same bounds and predicates, and the eight notify_all into NotifyAll. On Linux TimedWait::Condition is the condition_variable_any path, so nothing changes there. This is the first use of TimedWait::Condition outside libkernel.

Tested

Windows 11 (10.0.26200), WinLibs GCC 15.2.0 (15.2.0posix-14.0.0-ucrt-r7), Release, BUILD_TESTING=ON.

• Standalone measurement against the built libkernel.prx, tick raised to 0.5 ms as Time.cpp does, 60 iterations each:

  wait	p50	min	max
  std::condition_variable::wait_for(10 ms), no notify	11.65 ms	10.86	24.43
  TimedWait::Condition::WaitUntil(10 ms), no notify	10.01 ms	10.00	11.25
  TimedWait::Condition::WaitUntil(10 ms), notified from another thread	0.13 ms	0.06	1.54
  std::condition_variable::wait_for(10 ms), notified from another thread	0.14 ms	0.06	2.36

  The notified rows show the wake-on-flip path is unchanged.

• cmake --build build --target libSceVideoOut video_out_flip_tests; video_out_flip_tests decode and video_out_flip_tests controls with SDL_VIDEODRIVER=dummy exit 0 (the modes video_out_flip_tests is not in ctest and its lifetime and reopen modes terminate on main #1040 reports as passing on main; the others need a param.json).

• python tools/check_conventions.py --base upstream/main and git diff --check pass.

• No title run; no Linux host here (the Linux path is the unchanged condition_variable_any fallback inside TimedWait::Condition).

Checklist

• Based on current main; no other open PR implements the same functions (fix(videoout): report GPU driver shutdown failure from the present thread at Shutdown #1225 and fix(avplayer): keep handed-out video frames until later frames are handed out #1601 do not touch presentLoop; fix(videoout): pace vblanks with a high-resolution timer #1472 and fix(agc): poll for completions with a high-resolution timer on Windows #1474 covered the other VideoOut and queue waits)
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed in TechnicalDebt — none added
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source — none added
• Generic behaviour, not specific to one title
• Depends on: none
• AI-assisted: yes

🤖 Generated with Claude Code

[oneandonlydean]

This breaks loading on Linux. With this PR on main (25cc6fa), libSceVideoOut.prx imports TimedWait::DeadlineNanos (_ZN9TimedWait13DeadlineNanosEm), which libkernel.prx contains but doesn't export, so a title stops at load with symbol lookup error ... undefined symbol and patched_libraries fails. nm -D on the PR's own build shows the same, so it isn't from main. Could you either export it from libkernel or keep the helper inside libSceVideoOut?

[adriadam10]

Thanks @oneandonlydean for catching this. Confirmed from the diff: VideoOutDriver.cpp now calls TimedWait::DeadlineNanos (declared in libkernel/Time/include/TimedWait.hpp:15), a plain function that libkernel.prx contains but doesn't export. libSceVideoOut.prx therefore imports an unresolved symbol and the title stops at load.

To fix, either:

• export the TimedWait API that libSceVideoOut needs from libkernel (and check patched_libraries passes), or
• keep the deadline computation inside libSceVideoOut (e.g. a local steady_clock-based deadline) and only use what libkernel already exports.

Please also run patched_libraries (and a load of a module importing it) before pushing; the PR CI no longer builds/runs ctest, so it won't show there.

[adriadam10]

Thanks @oneandonlydean for catching this. Confirmed: libkernel.prx doesn't export TimedWait::*, so libSceVideoOut.prx had an unresolved import.

Fixed on the branch:

• rebased onto current main (conflict in FlipRequest::Fail after ReleaseLocked; the new notify_all is now NotifyAll)
• TimedWait.cpp (self-contained) is compiled into libSceVideoOut (CMakeLists.txt) instead of being imported from libkernel

Checked: nm -D libSceVideoOut.prx has no TimedWait undefined symbol; patched_libraries and the VideoOut ctest group pass (flip modes needing a Vulkan window skip). The Windows timing numbers are the author's and weren't reproduced here. Merging on green CI.