Alles anzeigenMaybe you read the problem with the maps flickering.
I was able to record a log while driving sesion with carplay maps.
Please see attached logs and let us know if you can help.
I think is from IDR referesh.
I monitorised the native maps for flickering and isn t happening.
Hi Predator0 (and scheibandreas),
thanks a lot for sharing the logs — I had a look at your cl_probe.log and clusterd.log, and I think we just found the exact causes behind the map-only flicker!
First, the good news from your log: the incoming H.264 stream itself is completely clean (6,982 frames over ~440s, badnal=0, steady ~15.8 fps / ~63 ms per frame). Your log shows the older v19.3-based pipeline, and you were actually right to look closely at IDR refreshes — combined with a few rendering/buffer-sync bugs in the older cl_render (src/cl_render/cluster.c), which we just fixed in v21.0 / v21.0.1:
1. Mid-stream SPS/PPS + IDR refcount reset (cb_begin_sequence): Your log shows periodic idr 1 requests every ~30.7s, and each IDR arrives with an op1 (SPS/PPS avcC) header. In the older cluster.c, cb_begin_sequence() unconditionally reset g_dec.surf_refcount[i] = 0 on every sequence header, even for decoder surfaces still in use. In v21.0+ this is guarded with if (g_dec.num_surfaces == 0).
2. 50 ms heartbeat colliding with the ~15.8 fps (~63 ms/frame) stream: Older cluster.c re-posted the last buffer whenever no new frame arrived for >50 ms. Since the CarPlay cluster stream runs at ~15.8 fps (~63 ms per frame), that 50 ms heartbeat constantly collided with real frames. In v21.0+ this was raised to 250 ms.
3. Unsynchronized 2D blit before screen_post_window: Previously screen_post_window() was called right after screen_blit() without waiting for the 2D blit to finish (cl_flush_blits_idle() / SCREEN_WAIT_IDLE), so the compositor could scan out an unfinished buffer or the HW decoder could reuse the YV12 surface too early.
4. Front-buffer wrap-around on burst frames: Added cl_pick_next_pidx() (never blit into g_last_posted_pidx) plus a 20 ms burst throttle (g_batch_deferred) so queued frames after jitter/IDR can't lap the buffer ring and overwrite the live front buffer.
Since your car is on EU P2869 (Touarego fork), please don't flash my CN P2878 release package directly — but scheibandreas can port those exact fixes straight from src/cl_render/cluster.c in v21.0.1.
Also, quick update for everyone following the repo — I've just released v21.0.1 with a few more things that might be useful for your P2869 port:
- Wired CarPlay iAP2 fix (v21.0.1): Fixed a double 6-byte header skip in src/cluster-probe/wired/rg_core.inc (important for wired CarPlay HUD/lane guidance — I noticed Cinemo wired iAP2 is active in your dio_cluster.log too).
- Steering-wheel OK double-tap map switch (VW_HUD_4_OkMapSwitch): Double-press OK within 2s to toggle instantly between CarPlay map and stock VW map (handy for pACC / quick switch back without uninstalling).
- v21.0 visual & telemetry overhaul: Larger bold telemetry cards with sport bars, symmetrical 295px navigation card (with standby mode), and 43 real DSI channels (direct TPMS, pitch/roll, GNSS altitude, ZF 8HP65A D1–D8 gear ratio readout).
- Shared HMI popup filter (VWPopupFilter, Bundle 98) & ModKit clean-unit mount / touareg_netd security & stability fixes.
Full source and changelog are up at:
Would love to hear if porting those cluster.c sync/heartbeat fixes clears up the flicker on P2869!
Best regards,
a78c7