Commit 2877cf8
committed
docs: note the AE-VMAX-clamp follow-up — wire-rate cap isn't VEDU
After this PR landed, follow-up investigation found the 140-fps
wire-rate cap at 480×352 isn't VEDU encoder throughput as we
suspected in §6.2. Tracing /proc/umap/isp showed MaxLine=2060
during a 140-fps run: at HMAX=0x100 and VMAX=2060, sensor fps =
72.8 MHz / (256 × 2060) = 138 — the wire rate IS the sensor rate,
just at much-larger-than-needed VMAX.
The AE library calls cmos_slow_framerate_set every frame with its
chosen VMAX. For M7 mode it picks values ~2.5× the datasheet
vmax_min, leaving 2× speed on the table. Clamping VMAX to
vmax_min in that function more than doubles wire-rate fps:
480×352 @240 went from 140 fps to 319 fps (+128%) on hi3516ev300.
The clamp isn't landed here — it shrinks AE max-integration-time
from ~28ms to ~3ms, which is fine for FPV daylight but unusable
indoors. A follow-up PR adds it behind an opt-in sensor-INI
dispatch (`Isp_SnsMode=5` = M7 + FPS-priority AE), with the
trade-off documented and the gk init failure at 480×352 noted as
a known issue.
§8.5 added to the doc with the head-to-head table and pointer to
the kaeru reference (`v4-ae-vmax-clamp-doubles-deliverable-fps`)
and the companion PR.1 parent 9f087e8 commit 2877cf8
1 file changed
Lines changed: 29 additions & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
897 | 897 | | |
898 | 898 | | |
899 | 899 | | |
| 900 | + | |
| 901 | + | |
| 902 | + | |
| 903 | + | |
| 904 | + | |
| 905 | + | |
| 906 | + | |
| 907 | + | |
| 908 | + | |
| 909 | + | |
| 910 | + | |
| 911 | + | |
| 912 | + | |
| 913 | + | |
| 914 | + | |
| 915 | + | |
| 916 | + | |
| 917 | + | |
| 918 | + | |
| 919 | + | |
| 920 | + | |
| 921 | + | |
| 922 | + | |
| 923 | + | |
| 924 | + | |
| 925 | + | |
| 926 | + | |
| 927 | + | |
| 928 | + | |
900 | 929 | | |
901 | 930 | | |
902 | 931 | | |
| |||
0 commit comments