Skip to content

v0.16.5 - Click-jolt filter for pads with microswitches and performance improvements

Choose a tag to compare

@x-iso x-iso released this 31 Oct 00:44
· 46 commits to main since this release

Added

  • Click-jolt filter for gyro. On pads whose microswitches sit on the same
    board as the motion sensor, every click knocks the sensor, and the gyro reads
    it as a sharp flick. In the Calibration window, turn on Click-jolt filter
    and open buttons…, then press each button a few times to see how big and
    how long each one's jolt is. Buttons that jolt the gyro tick themselves, with
    the time they need, unless you turn off Apply measurements automatically,
    and you can tick or adjust any of them by hand. The gyro and accelerometer then
    hold their value through the jolt of those buttons' presses and releases, then
    slope back to the live signal over the Release time, and nothing else is
    touched or delayed. If a pad reports the jolt before the click, the window says
    so and offers Look-ahead of up to 5 ms, which catches that part too at the
    cost of delaying the motion by that much. A small graph on each button's row
    shows its last click: the gyro in red and, on ticked buttons, what the filter
    leaves of it in blue. Change a time and blue redraws on that same click, so you
    can see how much hold and release it needs. Hover a graph to enlarge it.

  • Save and load calibration. The Calibration window's title bar now has
    Save… and Load…, which keep all of its settings in a .fxcal file: gyro,
    orientation, stick and trigger calibration, deadzones and filters. Use them to
    apply the same settings on another patch or PC, or when a pad connects through
    another dongle or channel. Loading replaces all of that pad's calibration
    settings, including any the file doesn't have, which go back to their defaults.

  • Lower processing rates. The processing sample rate can now go down to the
    polling rate, as low as 125 Hz, and can follow the output rate instead: Same
    as output
    uses the least CPU, 2x output costs twice that and makes timers,
    gestures and PWM twice as fine. Set by hand stays the default, and your rate is
    unchanged. On the patches we measured, even 1x barely changed the output.

  • Freeze and dim unused windows. A new option beside Background repaint rate
    in Settings, off by default. FlexInput windows without focus stop drawing and
    stay dimmed on their last frame, including when another FlexInput window has
    focus. Mappings and overlays keep running. Your first click on a dimmed main window
    works straight away, and passing the pointer over it costs nothing. A sub-patch
    editor wakes when the pointer moves over it. A pinned or see-through window
    stays live.

Changed

  • Mappings use about half the CPU. The engine no longer looks up and copies
    every pin by name on every tick, which was most of its work, so a patch runs
    about twice as fast with exactly the same outputs. Nothing to set up, and nothing
    is locked: every edit still applies at once.
  • Easy mode draws faster. Every frame re-read and re-hashed all presets and
    serialized the whole sub-patch to JSON, about a third of a frame even without a
    presets folder. A new or edited preset now shows up within 2 seconds.
  • A minimized FlexInput uses less CPU. Minimized windows no longer draw
    anything. Mappings, overlays and shortcuts carry on as before. On Linux this
    needs a desktop that reports minimized windows, which Wayland doesn't; the new
    option below covers it there, since a minimized window has no focus.
  • Tidier gyro section in the Calibration window. Axis alignment is now
    folded away under a chevron, and says "applied" when an alignment is in
    effect. Its Roll, Pitch and Yaw buttons fit on one line, with the movement for
    each in a tooltip.
  • The spike filter is now off by default. Now that it works everywhere (see
    below), it delays the gyro by at least one report, so only a pad with spikes
    should pay for it. Calibrating turns it on when it finds spikes, with the least
    setting that removes them, and leaves it alone when it finds none. A pad you
    have already changed its settings on keeps them.

Fixed

  • Average's window is now set in milliseconds. It counted processing ticks,
    so the same patch averaged over a different length of time at every sample
    rate. Existing patches load with the length they had at the default 2000 Hz: a
    window of 10 becomes 5 ms. If you run a different sample rate, your window was
    already a different length of time and now becomes that one, so check it.

  • The spike filter now works on SDL and Joy-Con 2 pads. It counted polls
    rather than reports from the pad, so once polling ran a few times faster than
    the pad reports, spikes went straight through at every setting. Its window now
    counts real reports whatever the polling rate. On these pads it now really
    does delay the gyro while on: by one report at window 3 (1 ms at 1000 Hz, 4 ms
    at 250 Hz) and more at wider windows, so check your window if you raised it
    while it wasn't doing anything. The spike check in calibration tunes from real
    reports too.

  • Gyro aim on the virtual DualSense and DualShock 4 no longer goes
    diagonal.
    The virtual pad reported its tilt on the wrong axes, forward
    and sideways swapped, and its roll reversed. A game that aims relative to
    gravity (player or world space gyro) then read a pad held nose-up as
    tipped to one side, so raising a pad held at an angle moved the aim
    diagonally. The virtual DualSense Edge and the Audio & Haptics DualSense
    had the same fault. They now send motion as a real pad in the same pose
    would. Local-space gyro only had its roll reversed.

Full Changelog: v0.16.4...v0.16.5