Retail is not a 3D audio engine. Every gameplay buffer is created with m_3D = 0 and the DirectSound 3D listener the client sets up is dead code; spatialization is two CPU scalars per voice, frozen at emission. This slice ports that math and demotes OpenAL to a voice bank. RetailSoundMixer (new, Core) carries the byte-decoded curve from SoundManager::GetAttenuation @0x00550020: g = dist < 5 ? vol : 25*vol/d2, clamped to 1 BEFORE the single master multiply, db = ceil(20*log10 g), with a hard -50 dB floor at which retail does not start the voice at all (audible radius ~94.2 m at unity). Pan is PlaySoundInternal @0x00550170's (int)(-15*sin(delta-bearing)) in whole decibels, truncating toward zero, forced to dead centre when (int)distance < 5, with no front/back and no elevation cue. Every AL source is now source-relative with rolloff 0 and the global distance model is None: AL's InverseDistanceClamped was first-power (2/d), quieter than retail up close and far louder at range with no cutoff whatsoever. That was the largest audible divergence in the subsystem (AP-28, retired here). RetailVoicePool (new, Core) ports the allocator at 0x0054FEC0: ring scan for a free or finished slot, then evict the first slot whose DAT priority is strictly lower, else drop. Eviction compared GAIN before, so a loud unimportant sound could silence a quiet important one. It lives in Core because the engine's play path talks to native AL handles and could not be tested; the pool now has 12 conformance tests. The listener keeps using the camera position, which the decode shows is retail-faithful (SmartBox::set_viewer @0x00452D36 hands the same collided camera Position to SoundManager) — only the heading extraction changes, since retail reads one compass bearing and never a forward/up basis. An earlier draft of the plan called this a defect; corrected in the plan so it is not fixed backwards. Opus review found and this commit fixes: a linear pan-to-azimuth mapping that saturated to full separation at 30 degrees (OpenAL Soft's own speaker angle) where retail gives 15 dB — now inverts the constant-power pan law, so full deflection reaches 0.776 of the arc and both channels stay live; the stale FUN_00550ad0 / gain-eviction class header, which contradicted the register row this commit writes; missing discriminating tests for clamp order and pan truncation; dead PlayingGain state whose comment invented a retail symbol; and a third in-tree copy of Position::heading, now delegating to MoveToMath.PositionHeading. MasterVolume folds into the mixer's one multiply instead of AL listener gain, so the cutoff, radius and dB quantisation move with the slider. Register: AP-28 retired; AP-173 (pan law), AP-174 (volume taxonomy), TS-64 (two unimplemented sound prefs), TS-65 (volume-squared quirk, applied on the ambient path only) filed. Research note corrected twice where its summary contradicted its own decode (30 m dB, floor vs trunc). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
263 lines
11 KiB
C#
263 lines
11 KiB
C#
using System;
|
||
using System.Numerics;
|
||
|
||
namespace AcDream.Core.Audio;
|
||
|
||
/// <summary>
|
||
/// One voice's mixing decision, as retail computes it at emission time.
|
||
/// </summary>
|
||
/// <param name="Play">
|
||
/// False when retail would not start the voice at all — the attenuation fell
|
||
/// below <see cref="RetailSoundMixer.VolMinDecibels"/>. Retail does not start a
|
||
/// quiet voice; it starts nothing, so no pool slot is consumed either.
|
||
/// </param>
|
||
/// <param name="Decibels">
|
||
/// Integer decibels, <c>ceil(20·log10 g)</c>, floored at
|
||
/// <see cref="RetailSoundMixer.VolMinDecibels"/>.
|
||
/// </param>
|
||
/// <param name="Pan">
|
||
/// Retail's DirectSound pan in whole decibels, <c>[-15, +15]</c>; negative is
|
||
/// left, positive right, 0 dead centre.
|
||
/// </param>
|
||
public readonly record struct RetailVoiceMix(bool Play, int Decibels, int Pan);
|
||
|
||
/// <summary>
|
||
/// Retail's sound mixing math — the whole of its spatialization.
|
||
///
|
||
/// <para>
|
||
/// <b>Retail is not a 3D audio engine.</b> Every gameplay sound buffer is
|
||
/// created with <c>m_3D = 0</c>; the DirectSound 3D listener the client sets up
|
||
/// (rolloff 0.01, front (−1,0,0), top (0,1,0)) is dead code that no buffer ever
|
||
/// consults. Spatialization is exactly two CPU-computed scalars per voice,
|
||
/// frozen at emission: a gain in whole decibels from distance, and a stereo pan
|
||
/// in whole decibels from bearing.
|
||
/// </para>
|
||
///
|
||
/// <para>
|
||
/// Sources: <c>SoundManager::GetAttenuation</c> @ <c>0x00550020</c> and
|
||
/// <c>SoundManager::PlaySoundInternal(SoundBufRef*, const Position*, float, int)</c>
|
||
/// @ <c>0x00550170</c>, both byte-decoded from the PDB-paired binary — Binary
|
||
/// Ninja elides the x87 memory constants in the first (its pseudo-C multiplies
|
||
/// by a literal <c>0f</c>, which would port as silence at every distance) and
|
||
/// misattributes a reused stack slot in the second (reporting the 5-metre
|
||
/// deadzone as an angle test rather than a distance test). Full decode:
|
||
/// <c>docs/research/2026-08-08-audio-retail-soundmanager-core.md</c> §1.
|
||
/// </para>
|
||
/// </summary>
|
||
public static class RetailSoundMixer
|
||
{
|
||
/// <summary>Distance below which gain is flat at the authored volume, metres.</summary>
|
||
public const float VolMinDistance = 5.0f;
|
||
|
||
/// <summary>
|
||
/// <see cref="VolMinDistance"/> squared — retail's numerator, so the curve
|
||
/// is continuous at the knee.
|
||
/// </summary>
|
||
public const float VolMinDistanceSq = 25.0f;
|
||
|
||
/// <summary>
|
||
/// Retail's audibility floor. A voice whose computed decibels fall below
|
||
/// this is never started.
|
||
/// </summary>
|
||
public const int VolMinDecibels = -50;
|
||
|
||
/// <summary>Retail's pan scale, applied to the sine of the reversed bearing.</summary>
|
||
public const float PanScale = -15.0f;
|
||
|
||
/// <summary>
|
||
/// Pan is forced to dead centre when <c>(int)distance</c> is below this.
|
||
/// Retail truncates the distance before comparing, so this is an integer
|
||
/// metre test, not a float one.
|
||
/// </summary>
|
||
public const int PanDeadzoneMetres = 5;
|
||
|
||
/// <summary>Retail's voice count (<c>playing_sounds_[0x10]</c>).</summary>
|
||
public const int VoiceCount = 16;
|
||
|
||
/// <summary>
|
||
/// Retail's own single-precision degrees-to-radians literal, as it appears
|
||
/// in the pan computation.
|
||
/// </summary>
|
||
private const float DegreesToRadians = 0.0174532924f;
|
||
|
||
/// <summary>
|
||
/// <c>SoundManager::GetAttenuation</c>. Returns whole decibels and whether
|
||
/// retail would start the voice.
|
||
///
|
||
/// <para>
|
||
/// <paramref name="masterVolume"/> is retail's ONE master multiply —
|
||
/// <c>effect_sound_volume</c> for ordinary sounds, <c>ambient_sound_volume</c>
|
||
/// when <c>ambient != 0</c>. Callers on the paths where retail ALSO
|
||
/// pre-multiplies by the same knob (its volume-squared quirk) must
|
||
/// pre-multiply <paramref name="volume"/> themselves; this function applies
|
||
/// the knob exactly once.
|
||
/// </para>
|
||
/// </summary>
|
||
public static bool TryGetAttenuation(
|
||
float distanceMetres,
|
||
float volume,
|
||
float masterVolume,
|
||
out int decibels)
|
||
{
|
||
float g = distanceMetres < VolMinDistance
|
||
? volume
|
||
: (VolMinDistanceSq * volume) / (distanceMetres * distanceMetres);
|
||
|
||
if (g > 1.0f) g = 1.0f;
|
||
g *= masterVolume;
|
||
|
||
if (g <= 0.0f || float.IsNaN(g))
|
||
{
|
||
decibels = VolMinDecibels;
|
||
return false;
|
||
}
|
||
|
||
decibels = (int)MathF.Ceiling(20.0f * MathF.Log10(g));
|
||
if (decibels >= VolMinDecibels)
|
||
return true;
|
||
|
||
decibels = VolMinDecibels;
|
||
return false;
|
||
}
|
||
|
||
/// <summary>
|
||
/// Retail's compass heading convention (<c>Position::heading</c> @
|
||
/// <c>0x005A9520</c>): degrees clockwise from +Y (north), +X (east) = 90°.
|
||
///
|
||
/// <para>
|
||
/// Delegates to <see cref="Physics.Motion.MoveToMath.PositionHeading"/> —
|
||
/// the pinned port of the same retail function, with golden-cardinal
|
||
/// coverage — rather than carrying a third copy of the formula. Retail's own
|
||
/// constant is the double <c>57.29577951308232</c>; the single-precision
|
||
/// narrowing there is at most ~1.5e-6 degrees, far inside the integer
|
||
/// truncation the pan applies afterwards.
|
||
/// </para>
|
||
/// </summary>
|
||
public static float CompassHeadingDegrees(Vector3 from, Vector3 to) =>
|
||
Physics.Motion.MoveToMath.PositionHeading(from, to);
|
||
|
||
/// <summary>
|
||
/// Retail's heading-difference normalisation, verbatim: <c>fmod(delta, 360)</c>
|
||
/// then <c>if (!(delta <= 180)) delta -= 360</c>.
|
||
///
|
||
/// <para>
|
||
/// The output range is retail's <c>(-360, 180]</c>, NOT <c>(-180, 180]</c> —
|
||
/// retail does not fold negatives back up, so an input of −270 stays −270.
|
||
/// That is harmless because the only consumer is <c>sin</c>, which has
|
||
/// period 360, and reproducing it exactly keeps this function comparable to
|
||
/// the decode.
|
||
/// </para>
|
||
/// </summary>
|
||
public static float NormalizeSignedDegrees(float degrees)
|
||
{
|
||
float delta = degrees % 360.0f;
|
||
if (!(delta <= 180.0f)) delta -= 360.0f;
|
||
return delta;
|
||
}
|
||
|
||
/// <summary>
|
||
/// Retail's pan, in whole decibels. <paramref name="bearingSourceToListener"/>
|
||
/// is <c>Position::heading(soundPos, listenerPos)</c> — the REVERSED
|
||
/// bearing; combined with the negative <see cref="PanScale"/> that yields
|
||
/// the correct handedness (a source due east of a north-facing listener
|
||
/// pans right).
|
||
///
|
||
/// <para>
|
||
/// There is no front/back and no elevation cue: a source dead ahead and one
|
||
/// directly behind both pan to 0, and Z reaches the mix only through
|
||
/// distance.
|
||
/// </para>
|
||
/// </summary>
|
||
public static int GetPan(
|
||
float bearingSourceToListener,
|
||
float listenerHeadingDegrees,
|
||
float distanceMetres,
|
||
bool panningEnabled = true)
|
||
{
|
||
if (!panningEnabled)
|
||
return 0;
|
||
|
||
// Retail truncates distance to an int before the deadzone compare.
|
||
if (Math.Abs((int)distanceMetres) < PanDeadzoneMetres)
|
||
return 0;
|
||
|
||
float delta = NormalizeSignedDegrees(bearingSourceToListener - listenerHeadingDegrees);
|
||
int pan = (int)(MathF.Sin(delta * DegreesToRadians) * PanScale);
|
||
return Math.Clamp(pan, (int)PanScale, (int)-PanScale);
|
||
}
|
||
|
||
/// <summary>
|
||
/// The whole per-play decision for a world sound: distance gain plus pan,
|
||
/// composed the way <c>PlaySoundInternal</c> composes them.
|
||
/// </summary>
|
||
public static RetailVoiceMix Mix(
|
||
Vector3 listenerPosition,
|
||
float listenerHeadingDegrees,
|
||
Vector3 sourcePosition,
|
||
float volume,
|
||
float masterVolume,
|
||
bool panningEnabled = true)
|
||
{
|
||
float distance = Vector3.Distance(listenerPosition, sourcePosition);
|
||
|
||
// Pan is computed from the reversed bearing: source -> listener.
|
||
float bearing = CompassHeadingDegrees(sourcePosition, listenerPosition);
|
||
int pan = GetPan(bearing, listenerHeadingDegrees, distance, panningEnabled);
|
||
|
||
bool play = TryGetAttenuation(distance, volume, masterVolume, out int decibels);
|
||
return new RetailVoiceMix(play, decibels, pan);
|
||
}
|
||
|
||
/// <summary>
|
||
/// Linear amplitude for whole decibels: <c>10^(db/20)</c>. Retail hands the
|
||
/// decibel value straight to <c>IDirectSoundBuffer::SetVolume</c>, which is
|
||
/// hundredths of a decibel; OpenAL wants linear gain, so the conversion
|
||
/// happens here rather than changing the quantisation.
|
||
/// </summary>
|
||
public static float LinearGain(int decibels) =>
|
||
MathF.Pow(10.0f, decibels / 20.0f);
|
||
|
||
/// <summary>
|
||
/// Converts retail's pan (whole decibels of inter-channel difference, the
|
||
/// quantity <c>IDirectSoundBuffer::SetPan</c> expresses directly) into a
|
||
/// normalised stereo position in <c>[-1, 1]</c> for a constant-power panner
|
||
/// such as OpenAL's.
|
||
///
|
||
/// <para>
|
||
/// A constant-power panpot at position <c>p</c> over a speaker pair puts
|
||
/// <c>cos((p+1)·π/4)</c> in the left channel and <c>sin((p+1)·π/4)</c> in
|
||
/// the right, so the inter-channel difference is
|
||
/// <c>20·log10(tan((p+1)·π/4))</c>. Inverting that for a target difference
|
||
/// gives <c>p = (4/π)·atan(10^(pan/20)) − 1</c>, which reaches only ±0.775
|
||
/// at retail's ±15 dB — both channels stay live, exactly as DirectSound's
|
||
/// one-channel attenuation keeps them. Mapping pan linearly onto the
|
||
/// panner's full range instead would saturate to infinite separation at the
|
||
/// edges, which retail never does.
|
||
/// </para>
|
||
/// </summary>
|
||
public static float StereoPositionFromPan(int pan)
|
||
{
|
||
float difference = MathF.Pow(10.0f, pan / 20.0f);
|
||
float position = (4.0f / MathF.PI) * MathF.Atan(difference) - 1.0f;
|
||
return Math.Clamp(position, -1.0f, 1.0f);
|
||
}
|
||
|
||
/// <summary>
|
||
/// The maximum distance at which a sound of this volume is audible at all,
|
||
/// in metres — the radius where <c>ceil(20·log10(25·vol·master/d²))</c>
|
||
/// last reaches <see cref="VolMinDecibels"/>. About 94.2 m at
|
||
/// <c>vol·master == 1</c>, 66.6 m at 0.5, 29.8 m at 0.1. Diagnostic and
|
||
/// test use; the live path gets the same answer from
|
||
/// <see cref="TryGetAttenuation"/>.
|
||
/// </summary>
|
||
public static float AudibleRadius(float volume, float masterVolume)
|
||
{
|
||
float scale = volume * masterVolume;
|
||
if (scale <= 0f) return 0f;
|
||
|
||
// Audible while db >= -50, i.e. ceil(20·log10 g) >= -50, i.e.
|
||
// 20·log10 g > -51 (ceil of anything above -51 is at least -50).
|
||
float minGain = MathF.Pow(10.0f, -51.0f / 20.0f);
|
||
return MathF.Sqrt(VolMinDistanceSq * scale / minGain);
|
||
}
|
||
}
|