[L0FKA]

Optional native 44.1 kHz engine rate

Some interfaces cannot run at 48 kHz (hardware locked to 44.1 kHz, or a driver that refuses 48000).
This adds an opt-in GameSampleRate=44100 under [Config] that makes the game's audio engine run
natively at 44.1 kHz, so a 44.1 kHz interface needs no external sample-rate conversion.

What it does

• Patches the game's 48000 Hz constants on the September 2022 (d1b38fcb) and December 2024 Learn & Play
  (6ea6d1ba) builds:
  • the approved-rate check;
  • the output rate;
  • the capture rate, which is a double;
  • six input-side integer sites.
    All 9 sites are byte-verified before anything is written. If any site doesn't match, nothing is patched
    and the game keeps 48 kHz. A memory-protection or write failure after that check is logged and stops further
    patching; earlier sites, and possibly the failing one, may already have changed. There is no rollback.
    The third supported build logs "not supported for this game version" and is left alone.
• At 44.1 kHz the convolution amp cabinet would reject its 48 kHz impulse responses, which silently drops
  cabinet coloring. A hook in the convolution engine's init resamples each IR to 44.1 kHz with a
  windowed-sinc filter inside the engine's init when the cabinet media loads (about 40 ms per cabinet).
  It runs once per cabinet load, not on every audio block. The hook's stolen bytes are
  verified first; on a mismatch the cabinet is left unhooked and the rate patch still applies.
  The media header is bounds-checked against the media size before any allocation or read (partition count,
  table entries of 1..128 bins, 64-bit payload total, plausible authored rate); anything else is left untouched.
• Blank, 0 or 48000 keeps the existing 48 kHz audio behaviour; only startup log lines are added. Other values
  are rejected with a log line.
• If a driver refuses 48000 Hz but accepts 44100 Hz, the log now suggests the option once.

Why 44.1 kHz only: the note detector's analysis window is a fixed number of samples. At 96 kHz the
window is half as long, which raises the lowest detectable pitch to about 47 Hz, so a bass low E
(41.2 Hz) could no longer be detected.

Testing

September 2022 build, MOTU M Series, Int32LSB, 128 frames. A 110 Hz test tone was fed into the input,
and the in-game tuner was read:

Setting	Log	Tuner
default (48 kHz)	rate patch skipped	A, +6 cents
GameSampleRate=44100	9 sites patched, driver switched to 44100, both clients at 44100	A, +5 cents

If the capture rate did not match the engine rate, the tone would read about 147 cents sharp, so the two
readings agreeing within 1 cent shows the input path runs at the engine rate. The cabinet log showed the
IR resampled 48000 -> 44100 (31 partitions, about 40 ms, at load), and the cabinet effect was active.
In-game pitch and tone checked by ear.

An offline test fed the cabinet parser 11 malformed and valid headers, inside buffers with canary bytes after the
media. Every malformed header was refused, nothing was read or written past the media, and no exception escaped;
both valid cases behaved correctly. With the new partition-count bound removed, a 4 GiB partition-count case escapes
as vector too long, so the test does catch the bug.

Not tested:

• The Learn & Play build has been checked statically Reports from Learn & Play users would be very welcome.

[mdias]

This looks pretty good, but I would change one small thing.
Instead of:

[Config]
GameSampleRate=44100

I would rather have:

[Config]
GameSampleRate44100=true

Since if I understand correctly that's the only real alternative sample rate we support.
Do you agree?

[L0FKA]

Yes, going up in 96 kHz territory is moot, going 44.1 kHz improves bass detection range. I'll drop the rename in a sec.