simplysanatoriandClaude Fable 5.1 b2aea99bef perf: session-h20 results (after-lock clean but one refresh late; before unreachable; the clean probe ends at the default phase, so phase alone is not the cause) ...

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

2026-10-16 02:13:23 -04:00

..

mockdump

Lane S: the race eye images are clean in the mock (zigzag 0 % per eye, standing and driving); session-h13 xrprobe replay plan

2026-10-15 06:45:51 -04:00

panels

perf/panels: cadence summary and one-command analysis; xrmock: compositor model (spec/lazyrel/lazyacq), end-after-wake histogram

2026-10-14 01:48:07 -04:00

session-h8

perf/session-h8: the eighth session's two queued launches (old ending, sets ahead), recorded with the frame log

2026-10-14 04:36:15 -04:00

session-h9

perf/session-h9: the ninth session's queued launches (located with XRE eye-offset logging, fixed eye offsets, optional fixed + one refresh ahead), what to look for, and the mock numbers

2026-10-14 07:02:14 -04:00

session-h10

H9.md: call sequence, Vulkan hand-over, switches, lane N verdict; session 10 launch files

2026-10-14 12:44:37 -04:00

session-h11

perf/session-h11: lane Q's plan (race quad launch, first-write barrier launch) and mock results

2026-10-14 16:11:19 -04:00

session-h12

perf/session-h12: headset results (pose check, one-ahead lateness, fresh poses, panel zigzag per eye and over time)

2026-10-15 06:12:08 -04:00

session-h13

perf/session-h13: headset results (all six replay cases clean)

2026-10-15 13:08:31 -04:00

session-h14

perf: session-h14 headset results (desktop present and release backlog ruled out; drained queue cleared the left eye only); session-h15 env files (90 Hz, 1200 px eyes); wrapper KARTFRAME_EYE_PX

2026-10-15 15:26:44 -04:00

session-h15

perf: session-h15 results (90 Hz and 1200 px both doubled: GPU load size and refresh rate ruled out); session-h16 env (thread priority boost and uclamp off)

2026-10-15 16:49:43 -04:00

session-h16

perf: session-h16 results (ending the frame 4 ms after the wake-up removes the doubling); session-h17 delay sweep env files

2026-10-15 19:58:14 -04:00

session-h17

perf: session-h17 results (2.5 ms ends were bimodal, 6 ms bad for the right eye); session-h18 band-mapping env files (4000, 4500, 3500)

2026-10-16 00:10:36 -04:00

session-h18

perf: session-h18 results (4.0 ms not clean on repeat; the band is not fixed to the app's wake-up)

2026-10-16 00:19:50 -04:00

session-h19

perf: session-h19 results (clean vs doubled 4 ms launches separated by the end's phase against the compositor's update window)

2026-10-16 01:31:41 -04:00

session-h20

perf: session-h20 results (after-lock clean but one refresh late; before unreachable; the clean probe ends at the default phase, so phase alone is not the cause)

2026-10-16 02:13:23 -04:00

abgp.sh

RACE120.md: the race's GPU profile before and after, A/B pairs, soaks at 120 and 90 Hz; frame/perf/abgp.sh

2026-10-13 22:12:59 -04:00

bgload.sh

Raised priority for the threads a headset frame waits on (RealtimeKit nice -15); timing runs hold both projects' locks

2026-10-13 14:26:09 -04:00

gpuprof.py

WIP lane I: per-segment GPU profile, direct eye targets (measured choice), native race render clip, exact display-time stamp

2026-10-13 19:11:43 -04:00

H1.md

H1 session analysis (frame/perf/H1.md); DEV.md: KARTFRAME_HUD_CLIP

2026-10-13 04:22:08 -04:00

H2.md

H2.md: weight range as measured

2026-10-13 12:42:49 -04:00

H3.md

H3.md: third headset session analysis, lane H fixes and measurements

2026-10-13 17:55:24 -04:00

H4.md

H4.md: fix branch hash

2026-10-13 19:33:16 -04:00

H7.md

H7.md: the mock's scoring, the A/B with SteamVR's compositor running (loaded race: ends p99 7.8 -> 0.47 ms, 111 -> 120 sets/s, 7.8 -> 0.03 % late) and the next session's plan

2026-10-14 04:36:15 -04:00

H9.md

H9.md: call sequence, Vulkan hand-over, switches, lane N verdict; session 10 launch files

2026-10-14 12:44:37 -04:00

hitches.py

Lane E results (frame/perf/RESULTS.md); perf tools split 1p and split-screen demo races; mock MOCK_XR_RATES

2026-10-13 04:56:09 -04:00

hudrace.sh

Menus: the environment layer goes over the screen quad by default, so the pointer shows on the screen

2026-10-13 17:50:47 -04:00

iosample.py

Lane E instrumentation: game-frame hitch lines, per-set pacing samples, perf harness; log fixes

2026-10-13 01:55:44 -04:00

LANE_S.md

Lane S: the race eye images are clean in the mock (zigzag 0 % per eye, standing and driving); session-h13 xrprobe replay plan

2026-10-15 06:45:51 -04:00

LANE_U.md

perf: LANE_U.md (release ordered correctly, verdict B; ranked candidates) and session-h14 (no desktop swapchain; release idle)

2026-10-15 13:46:22 -04:00

LANE_W.md

perf: LANE_W.md (clock matching, the end lock design and risks); session-h19 env files and README

2026-10-16 00:34:30 -04:00

LANE_X.md

perf: session-h20 (end phase after / before / 4000 control), DEV.md lane X, mock results

2026-10-16 01:59:44 -04:00

pacing-parts.py

Lane E results (frame/perf/RESULTS.md); perf tools split 1p and split-screen demo races; mock MOCK_XR_RATES

2026-10-13 04:56:09 -04:00

PANELS.md

PANELS.md: the panel recording's findings, the mechanism, the mock proof and the next session's A/B; session-report: eye pairs and end-after-wake rows

2026-10-14 02:18:31 -04:00

RACE120.md

RACE120.md: the race's GPU profile before and after, A/B pairs, soaks at 120 and 90 Hz; frame/perf/abgp.sh

2026-10-13 22:12:59 -04:00

RESULTS.md

Lane E results (frame/perf/RESULTS.md); perf tools split 1p and split-screen demo races; mock MOCK_XR_RATES

2026-10-13 04:56:09 -04:00

run.sh

Direct eye output: measured in Aurora with GPU timestamps on the race's refresh sets; run.sh samples the GPU clock

2026-10-13 20:03:13 -04:00