From 31b864380f59827a6e4d67d201b22b176fc1c8e3 Mon Sep 17 00:00:00 2001 From: Valiantsin Dzerakh Date: Mon, 26 Oct 2026 08:23:27 +0300 Subject: [PATCH] Keep the replay on the submitting thread unless TOTK_REPLAY_ASYNC=1, since no timing showed a gain yet, and list async_preempt and replay_async in totk.ini.example Co-Authored-By: Claude Opus 5.5 --- docs/backlog.md | 19 +++++++++++++------ runtime/vk_backend.c | 2 +- totk.ini.example | 5 +++++ 3 files changed, 19 insertions(+), 7 deletions(-) diff --git a/docs/backlog.md b/docs/backlog.md index cf6afd28..e16eca77 100644 --- a/docs/backlog.md +++ b/docs/backlog.md @@ -26,8 +26,13 @@ Measure before building: every item states the number that decides it. copies, locks) before choosing a worker-thread upload, GPU deswizzle, image pooling or a per-frame budget. 3. Virtual-core preemption as Horizon does (MacNativeGames 8bf0893 / e4b4601, PR #488: asynchronous preemption of an owner in game code, priority inheritance, the SDK's preemption-disable protocol; Mac worst-1% frame 18.4 to 10.6 - ms): `task/preempt`, `build/v121-preempt/HANDOFF.md`. Port only if the measured waits behind an owner in game code - are significant; our checkpoints at indirect calls may already cover most. + ms): `task/preempt`, `build/v121-preempt/NOTES.md`. Ported (`TOTK_ASYNC_PREEMPT`, on by default; 0 restores the old + scheduling, 2 keeps inheritance and the disable protocol without interrupting). It cuts the waits it targets + (`TOTK_CORE_DIAG`: game-class waits of 1 ms or more 53 / 41 / 27 to 4 / 10 / 18 in field / fight / monsters2), but + the A/B timings show no mean-fps gain: fight >25 ms frames base 62, port 96 over four pairs (fps 58.7 / 58.2, + inside run-to-run noise); one of three base field runs froze 7 s at frame 2388 with the render thread blocked on + nothing, no port run did. Next: user play with it on, `TOTK_ASYNC_PREEMPT=0` first on a new hang or stutter; if + fight stays worse, try a shorter first spin (`AP_FIRST_NS`, 250 us) or mode 2. 4. Start-up re-warm when the driver cache does not match the exe path or device (5a.3): `task/rewarm` (main merged in, 0b3bc73), untested in game. Needed before the laptop package. 5. Shared shader list: the newer MacNativeGames lists collapse variants that differ only in texture kind into `"*"` @@ -118,10 +123,12 @@ present cadence by back-edge pacing (fight intervals outside 16.7 +- 1 ms 24.5% 52.9% of frames over 16.7 ms, 6.3% over 20 ms; frames over 20 ms are 87% replay (19 ms) with no pipeline or shader waits. The user's play added 0 shader translations and 29 full pipelines. 1. Uneven frames: replay off the critical path. Translate command buffers at `EndRecording` or on a queued replay - thread, so `QueueSubmitCommands` only queues; `TOTK_REPLAY_ASYNC=0` restores the synchronous path. Merge with - the new path off by default if the play test leaves any doubt. Code and probes done in `build/v121-rq` - (0 validation errors); no valid timings yet (its runs fell in the 21:00 file-cache corruption). Next: rebase on - runtime/, measure `fight.txt` warm with `--quiet`. + thread, so `QueueSubmitCommands` only queues. Merged 2026-10-26 (eb17360) off by default: + `TOTK_REPLAY_ASYNC=1` turns it on. Reads wait for the texture's newest writer, releases are queued, drain + waits give the virtual core up; checks and validation clean (`build/archive/v121-replayq/NOTES.md`). On the + RX 9070 no fps gain: fight and monsters2 are bound by the game threads, field render busy 10 to 1.3 ms. With + the replay itself longer than the frame (the 1050 Ti case) there is nothing to overlap: the game's main thread + waits each frame for the previous frame's GPU work. Next: a user test on the laptops with `replay_async = 1`. 2. Hitches when something new appears: first texture uploads decode ASTC on the CPU inside the frame (367 ms at frame 219 of `fight.txt` on the RX 9070; 402-714 ms load frames on the GTX 1050 Ti, `build/mem-plan.md`). Desktop AMD and NVIDIA have no ASTC; Intel Gen9-11 iGPUs do (check that `vk_texture.c` uses it when the diff --git a/runtime/vk_backend.c b/runtime/vk_backend.c index 03515864..4e7f97c9 100644 --- a/runtime/vk_backend.c +++ b/runtime/vk_backend.c @@ -170,7 +170,7 @@ int vk_replay_async(void) { static int on = -1; if (on < 0) { const char *e = cfg_get("TOTK_REPLAY_ASYNC"); - on = !(e && !strcmp(e, "0")); + on = e && !strcmp(e, "1"); // off by default: no measured gain yet, one more frame of latency rt_log("vk: replay %s (TOTK_REPLAY_ASYNC=%d)", on ? "on its own thread, at most one frame behind the game" : "on the submitting thread", on); } diff --git a/totk.ini.example b/totk.ini.example index 3f5de663..e558e8fb 100644 --- a/totk.ini.example +++ b/totk.ini.example @@ -12,6 +12,11 @@ # Fix for physics that bounce at high frame rates. # highfps_fix = 1 +[performance] +# A thread waiting for a virtual core interrupts the owner when it runs game code (0 = off, the old scheduling). +# async_preempt = 1 +# Replay NVN submissions on their own thread, at most one frame behind the game (1 = on). +# replay_async = 0 [display] # Presentation mode: mailbox (default when the driver offers it, vsync otherwise), fifo (vsync) or immediate. # vk_present = mailbox