You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Mac: the display link adds ~20 ms of input lag on fixed-rate screens, and fullscreen latency 2 applies to every screen #822
Controls feel slightly unresponsive, in fullscreen especially. The game holds 60 fps.
Measurements
All runs are fullscreen, along the profile's smoke route from a new save, with DM_NVN_TSV. Rows are counted from 40 s on.
Display link, latency 2 (fullscreen default)
Display link, DM_PRESENT_LATENCY=1
DM_PRESENT_LINK=0
Game present call → on screen, p50
50 ms
50 ms
29 ms
p5 / p95
37 / 53 ms
39 / 57 ms
19 / 45 ms
@link_lead_ms p50
32.8 ms
32.8 ms
—
On-screen intervals off 16.7 ± 2 ms
0.3%
0.3%
0.3%
Frame step changes
none
none
none
The 50 ms figure matches the log's own game enqueue to target display p50 50.2 ms.
Dropped frames after loading: 27 in about 7,200 with the link at latency 2, and 20 with the link off.
At latency 1, 70 and 184 presents were shown a refresh late in two runs. Almost all of them came in one stretch of about 10 s while the game loaded from its menu. After that the rate matched latency 2.
The choice follows the current screen mode's refresh interval range and updates on display, mode and fullscreen changes. DM_PRESENT_LATENCY still overrides it.
This is a general correction in the shared presentation path. Confidence in the policy fix is high; the existing ProMotion workaround remains a workaround for the system's missed presents.
Validation: the full graphics suite passed, including 64 display-link checks with the Metal lifecycle test; all 185 iOS sources compiled; C/Objective-C lint had no new findings; all 11 documentation tests passed. The focused display-link and documentation checks also passed after rebasing onto current main. Physical fixed-rate gameplay validation is still pending.
Regression risk was judged moderate because shared Mac presentation state changes. The selected fullscreen comparisons were TotK (the original workaround), BotW (another affected frame-step profile), and Pokemon Scarlet (a different engine), all named by regress.py impact. They were stopped at the maintainer's request during shader/host preparation, before gameplay comparisons; no game regression result is claimed.
The remaining roughly 20 ms of presentation delay is unresolved. Your measurements show the same 50 ms median with latency 1 and 2, so narrowing the policy does not establish a fix for that delay. Its cause and fix remain explicitly open in the reference finding. Closing this issue for the policy correction does not mark that timing investigation as resolved.
Thanks for the measurements and diagnosis; they are credited in the commit with a co-author trailer.
Activity
Display link: limit fullscreen latency 2 to variable-refresh screens (#…
seedds commentedon Nov 1, 2026
The fullscreen latency policy is corrected in cadd4be6.
DM_PRESENT_LATENCYstill overrides it.This is a general correction in the shared presentation path. Confidence in the policy fix is high; the existing ProMotion workaround remains a workaround for the system's missed presents.
Validation: the full graphics suite passed, including 64 display-link checks with the Metal lifecycle test; all 185 iOS sources compiled; C/Objective-C lint had no new findings; all 11 documentation tests passed. The focused display-link and documentation checks also passed after rebasing onto current main. Physical fixed-rate gameplay validation is still pending.
Regression risk was judged moderate because shared Mac presentation state changes. The selected fullscreen comparisons were TotK (the original workaround), BotW (another affected frame-step profile), and Pokemon Scarlet (a different engine), all named by
regress.py impact. They were stopped at the maintainer's request during shader/host preparation, before gameplay comparisons; no game regression result is claimed.The remaining roughly 20 ms of presentation delay is unresolved. Your measurements show the same 50 ms median with latency 1 and 2, so narrowing the policy does not establish a fix for that delay. Its cause and fix remain explicitly open in the reference finding. Closing this issue for the policy correction does not mark that timing investigation as resolved.
Thanks for the measurements and diagnosis; they are credited in the commit with a co-author trailer.
Display link: limit fullscreen latency 2 to variable-refresh screens (s…