While an input is locked, single-shot deviations of the frontend
measurements (line count, field period beyond PCNT_TOLERANCE, hsync width
beyond HSYNC_WIDTH_TOLERANCE, activity flag or the validity check) make
get_status() re-program the mode or drop sync. A NeXT SoG source produced
such glitches (868 -> 814 -> 868 lines) every few minutes, each one a
visible blink.
With this option on, a deviation is acted upon only once it has been seen
on three consecutive status reads; the first two deviating reads are
replaced by the previous stable values. That is 30 ms in the normal case
and up to about 105 ms if the vsync wait in get_status() times out on every
read. Counting deviating reads (rather than requiring identical readings)
guarantees that a real mode change or sync loss is always accepted on the
third read. The counter is cleared whenever the input is not locked or the
option is off. Default Off (stock behaviour). Stored as profile item id 69.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WiXHy7Jhh5yTdsBaenucgw
tvp_source_setup() disables the TVP7002 internal H-PLL coast for every
PC-type preset. That is right for RGBHV, but with sync-on-green or
composite sync the serration pulses during vsync disturb the H-PLL every
frame; on a NeXTstation Color (1120x832, SoG) this caused continuous
horizontal jitter, jitter in the top third of the picture and a periodic
HDMI re-sync. With the option on, coast is re-enabled after setup when the
sync source is SOG or CSYNC; RGBHV is untouched. Default Off, so existing
behaviour is unchanged. Stored as profile item id 68.
On the NeXT, this option together with the existing settings Analog sync
LPF = 10 MHz, ADC PLL BW = Low and H-PLL Pre/Post-Coast = 4/4 ran 19
minutes without a sync loss, mode re-program or IT6613 video-stable
change. SPARCstation 20 framebuffers (ZX, TurboGX) over composite sync are
stable without it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WiXHy7Jhh5yTdsBaenucgw
All entries go into GROUP_FIXED_VHZ at the end of the table, so they are
considered only for sources with the matching refresh rate; every other
source is matched exactly as before.
- 1120x832@72 (NeXTstation Color, ADB configuration): 868 total lines,
62.50 kHz, 90.000 MHz pixel clock (h_total 1440), 1.22 us hsync (110 px),
3-line vsync; back porches 185/30 and a 45 degree sampling phase tuned on
hardware. Without it the signal matched 1024x768 and was cropped.
- Sun, timings from the cg6 (GX/TGX) and cg14 (SX) framebuffer PROMs:
1152x900@66 (94.5 MHz, 1528x937), 1152x900@76 (108 MHz, 1504 total;
cg6 937 lines @ 76.64 Hz as "1152x900H", cg14 943 lines @ 76.15 Hz as
"1152x900HS"), 1280x1024@67 (118.125 MHz, 1648x1075, "1280x1024L") and
1280x1024@76 (135 MHz, 1664x1066, "1280x1024H"). Previously these matched
1280x960 or the VESA-timed 1280x1024 preset (h_total 1688).
1152x900@66 was verified on a SPARCstation 20 with ZX (Leo) and TurboGX
boards; the other Sun entries are not yet verified on hardware. Mode names
are limited to 10 characters, hence the H/HS/L suffixes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WiXHy7Jhh5yTdsBaenucgw
get_pure_lm_mode() picks the preset with the nearest total line count.
Several workstation display modes share line counts with each other or with
VESA modes while having one fixed refresh rate (e.g. Sun 1152x900 is 937
lines at both 66 Hz and 76.6 Hz on cg6 boards, and Sun 1280x1024@76 has the
1066 lines of VESA 1280x1024).
Add GROUP_FIXED_VHZ. Presets in this group are candidates only when the
measured refresh is within +-0.3 Hz of the preset (the measured refresh is
exact to about 0.01 Hz). They are appended at the end of the table, so the
early exit of the search loop is replaced by a flag that skips the
remaining regular presets exactly as before but still lets fixed-refresh
presets take part, including the existing equal-line refresh tie-break
(group >= GROUP_720P). Group-indexed arrays in video_modes.c and
controls.c get the extra entry.
No preset uses the group yet, so behaviour is unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WiXHy7Jhh5yTdsBaenucgw
marqs