aboutsummaryrefslogtreecommitdiffstats
path: root/Alc/alcReverb.c
diff options
context:
space:
mode:
Diffstat (limited to 'Alc/alcReverb.c')
-rw-r--r--Alc/alcReverb.c549
1 files changed, 549 insertions, 0 deletions
diff --git a/Alc/alcReverb.c b/Alc/alcReverb.c
new file mode 100644
index 00000000..badbac34
--- /dev/null
+++ b/Alc/alcReverb.c
@@ -0,0 +1,549 @@
+// TODO: Look into the distance attenuation done by the statistical model,
+// and see if it matches 1 / sqrt (distance) or some variant thereof.
+// If not, try to map it so it can replace the current dry-path model.
+// Also see how it responds to the distance model. Then if necessary
+// update ALu.c to compensate.
+// TODO: Finalize all updates and add necessary comments. Merge changes
+// with latest GIT snapshot and test all parameters thoroughly. When
+// ready produce some .diff files for the changes and include with
+// alReverb.c for presentation.
+
+#include "config.h"
+
+#include <math.h>
+#include <stdlib.h>
+
+#include "AL/al.h"
+#include "AL/alc.h"
+#include "alMain.h"
+#include "alAuxEffectSlot.h"
+#include "alEffect.h"
+#include "alReverb.h"
+
+#ifdef HAVE_SQRTF
+#define aluSqrt(x) ((ALfloat)sqrtf((float)(x)))
+#else
+#define aluSqrt(x) ((ALfloat)sqrt((double)(x)))
+#endif
+
+// fixes for mingw32.
+#if defined(max) && !defined(__max)
+#define __max max
+#endif
+#if defined(min) && !defined(__min)
+#define __min min
+#endif
+
+typedef struct DelayLine
+{
+ // The delay lines use lengths that are powers of 2 to allow bitmasking
+ // instead of modulus wrapping.
+ ALuint Mask;
+ ALfloat *Line;
+} DelayLine;
+
+struct ALverbState
+{
+ // All delay lines are allocated as a single buffer to reduce memory
+ // fragmentation and teardown code.
+ ALfloat *SampleBuffer;
+ // Master reverb gain.
+ ALfloat Gain;
+ // Initial reverb delay.
+ DelayLine Delay;
+ // The tap points for the initial delay. First tap goes to early
+ // reflections, the second to late reverb.
+ ALuint Tap[2];
+ struct {
+ // Gain for early reflections.
+ ALfloat Gain;
+ // Early reflections are done with 4 delay lines.
+ ALfloat Coeff[4];
+ DelayLine Delay[4];
+ ALuint Offset[4];
+ } Early;
+ struct {
+ // Gain for late reverb.
+ ALfloat Gain;
+ // Diffusion of late reverb.
+ ALfloat Diffusion;
+ // Late reverb is done with 8 delay lines.
+ ALfloat Coeff[8];
+ DelayLine Delay[8];
+ ALuint Offset[8];
+ // The input and last 4 delay lines are low-pass filtered.
+ ALfloat LpCoeff[5];
+ ALfloat LpSample[5];
+ } Late;
+ ALuint Offset;
+};
+
+// All delay line lengths are specified in seconds.
+
+// The length of the initial delay line (a sum of the maximum delay before
+// early reflections and late reverb; 0.3 + 0.1).
+static const ALfloat MASTER_LINE_LENGTH = 0.4000f;
+
+// The lengths of the early delay lines.
+static const ALfloat EARLY_LINE_LENGTH[4] =
+{
+ 0.0015f, 0.0045f, 0.0135f, 0.0405f
+};
+
+// The lengths of the late delay lines.
+static const ALfloat LATE_LINE_LENGTH[8] =
+{
+ 0.0015f, 0.0037f, 0.0093f, 0.0234f,
+ 0.0100f, 0.0150f, 0.0225f, 0.0337f
+};
+
+// The last 4 late delay lines have a variable length dependent on the effect
+// density parameter and this multiplier.
+static const ALfloat LATE_LINE_MULTIPLIER = 9.0f;
+
+static ALuint NextPowerOf2(ALuint value)
+{
+ ALuint powerOf2 = 1;
+
+ if(value)
+ {
+ value--;
+ while(value)
+ {
+ value >>= 1;
+ powerOf2 <<= 1;
+ }
+ }
+ return powerOf2;
+}
+
+// Basic delay line input/output routines.
+static __inline ALfloat DelayLineOut(DelayLine *Delay, ALuint offset)
+{
+ return Delay->Line[offset&Delay->Mask];
+}
+
+static __inline ALvoid DelayLineIn(DelayLine *Delay, ALuint offset, ALfloat in)
+{
+ Delay->Line[offset&Delay->Mask] = in;
+}
+
+// Delay line output routine for early reflections.
+static __inline ALfloat EarlyDelayLineOut(ALverbState *State, ALuint index)
+{
+ return State->Early.Coeff[index] *
+ DelayLineOut(&State->Early.Delay[index],
+ State->Offset - State->Early.Offset[index]);
+}
+
+// Given an input sample, this function produces a decorrelated stereo output
+// for early reflections.
+static __inline ALvoid EarlyReflection(ALverbState *State, ALfloat in, ALfloat *out)
+{
+ ALfloat d[4], v, f[4];
+
+ // Obtain the decayed results of each early delay line.
+ d[0] = EarlyDelayLineOut(State, 0);
+ d[1] = EarlyDelayLineOut(State, 1);
+ d[2] = EarlyDelayLineOut(State, 2);
+ d[3] = EarlyDelayLineOut(State, 3);
+
+ /* The following uses a lossless scattering junction from waveguide
+ * theory. It actually amounts to a householder mixing matrix, which
+ * will produce a maximally diffuse response, and means this can probably
+ * be considered a simple FDN.
+ * N
+ * ---
+ * \
+ * v = 2/N / di
+ * ---
+ * i=1
+ */
+ v = (d[0] + d[1] + d[2] + d[3]) * 0.5f;
+ // The junction is loaded with the input here.
+ v += in;
+
+ // Calculate the feed values for the delay lines.
+ f[0] = v - d[0];
+ f[1] = v - d[1];
+ f[2] = v - d[2];
+ f[3] = v - d[3];
+
+ // To increase reflection complexity (and help reduce coloration) the
+ // delay lines cyclicly refeed themselves (0 -> 1 -> 3 -> 2 -> 0...).
+ DelayLineIn(&State->Early.Delay[0], State->Offset, f[2]);
+ DelayLineIn(&State->Early.Delay[1], State->Offset, f[0]);
+ DelayLineIn(&State->Early.Delay[2], State->Offset, f[3]);
+ DelayLineIn(&State->Early.Delay[3], State->Offset, f[1]);
+
+ // To decorrelate the output for stereo separation, the cyclical nature
+ // of the feed path is exploited. The two outputs are obtained from the
+ // inner delay lines.
+ // Output is instant by using the inputs to them instead of taking the
+ // result of the two delay lines directly (f[0] and f[3] instead of d[1]
+ // and d[2]).
+ out[0] = State->Early.Gain * f[0];
+ out[1] = State->Early.Gain * f[3];
+}
+
+// Delay line output routine for late reverb.
+static __inline ALfloat LateDelayLineOut(ALverbState *State, ALuint index)
+{
+ return State->Late.Coeff[index] *
+ DelayLineOut(&State->Late.Delay[index],
+ State->Offset - State->Late.Offset[index]);
+}
+
+// Low-pass filter input/output routine for late reverb.
+static __inline ALfloat LateLowPassInOut(ALverbState *State, ALuint index, ALfloat in)
+{
+ State->Late.LpSample[index] = in + ((State->Late.LpSample[index] - in) *
+ State->Late.LpCoeff[index]);
+ return State->Late.LpSample[index];
+}
+
+// Given an input sample, this function produces a decorrelated stereo output
+// for late reverb.
+static __inline ALvoid LateReverb(ALverbState *State, ALfloat in, ALfloat *out)
+{
+ ALfloat din, d[8], v, dv, f[8];
+
+ // Since the input will be sent directly to the output as in the early
+ // reflections function, it needs to take into account some immediate
+ // absorption.
+ in = LateLowPassInOut(State, 0, in);
+
+ // When diffusion is full, no input is directly passed to the variable-
+ // length delay lines (the last 4).
+ din = (1.0f - State->Late.Diffusion) * in;
+
+ // Obtain the decayed results of the fixed-length delay lines.
+ d[0] = LateDelayLineOut(State, 0);
+ d[1] = LateDelayLineOut(State, 1);
+ d[2] = LateDelayLineOut(State, 2);
+ d[3] = LateDelayLineOut(State, 3);
+ // Obtain the decayed and low-pass filtered results of the variable-
+ // length delay lines.
+ d[4] = LateLowPassInOut(State, 1, LateDelayLineOut(State, 4));
+ d[5] = LateLowPassInOut(State, 2, LateDelayLineOut(State, 5));
+ d[6] = LateLowPassInOut(State, 3, LateDelayLineOut(State, 6));
+ d[7] = LateLowPassInOut(State, 4, LateDelayLineOut(State, 7));
+
+ // The waveguide formula used in the early reflections function works
+ // great for high diffusion, but it is not obviously paramerized to allow
+ // a variable diffusion. With only limited time and resources, what
+ // follows is the best variation of that formula I could come up with.
+ // First, there are 8 delay lines used. The first 4 are fixed-length and
+ // generate the highest density of the diffuse response. The last 4 are
+ // variable-length, and are used to smooth out the diffuse response. The
+ // density effect parameter alters their length. The inner two delay
+ // lines of each group have their signs reversed (more about this later).
+ v = (d[0] - d[1] - d[2] + d[3] +
+ d[4] - d[5] - d[6] + d[7]) * 0.25f;
+ // Diffusion is applied as a reduction of the junction pressure for all
+ // branches. This presents two problems. When the diffusion factor (0
+ // to 1) reaches 0.5, the average feed value is reduced (the junction
+ // becomes lossy). Thus, at 0.5 the signal decays almost twice as fast
+ // as it should. The second problem is the introduction of some
+ // resonant frequencies (coloration). The reversed signs above are used
+ // to help combat some of the coloration by adding variations along the
+ // feed cycle.
+ v *= State->Late.Diffusion;
+ // Load the junction with the input. To reduce the noticeable echo of
+ // the longer delay lines (the variable-length ones) the input is loaded
+ // with the inverse of the effect diffusion. So at full diffusion, the
+ // input is not applied to the last 4 delay lines. Input signs reversed
+ // to balance the equation.
+ dv = v + din;
+ v += in;
+
+ // As with the reversed signs above, to balance the equation the signs
+ // need to be reversed here, too.
+ f[0] = d[0] - v;
+ f[1] = d[1] + v;
+ f[2] = d[2] + v;
+ f[3] = d[3] - v;
+ f[4] = d[4] - dv;
+ f[5] = d[5] + dv;
+ f[6] = d[6] + dv;
+ f[7] = d[7] - dv;
+
+ // Feed the fixed-length delay lines with their own cycle (0 -> 1 -> 3 ->
+ // 2 -> 0...).
+ DelayLineIn(&State->Late.Delay[0], State->Offset, f[2]);
+ DelayLineIn(&State->Late.Delay[1], State->Offset, f[0]);
+ DelayLineIn(&State->Late.Delay[2], State->Offset, f[3]);
+ DelayLineIn(&State->Late.Delay[3], State->Offset, f[1]);
+ // Feed the variable-length delay lines with their cycle (4 -> 6 -> 7 ->
+ // 5 -> 4...).
+ DelayLineIn(&State->Late.Delay[4], State->Offset, f[5]);
+ DelayLineIn(&State->Late.Delay[5], State->Offset, f[7]);
+ DelayLineIn(&State->Late.Delay[6], State->Offset, f[4]);
+ DelayLineIn(&State->Late.Delay[7], State->Offset, f[6]);
+
+ // Output is derived from the values fed to the inner two variable-length
+ // delay lines (5 and 6).
+ out[0] = State->Late.Gain * f[7];
+ out[1] = State->Late.Gain * f[4];
+}
+
+// This creates the reverb state. It should be called only when the reverb
+// effect is loaded into a slot that doesn't already have a reverb effect.
+ALverbState *VerbCreate(ALCcontext *Context)
+{
+ ALverbState *State = NULL;
+ ALuint length[13], totalLength, index;
+
+ State = malloc(sizeof(ALverbState));
+ if(!State)
+ return NULL;
+
+ // All line lengths are powers of 2, calculated from the line timings and
+ // the addition of an extra sample (for safety).
+ length[0] = NextPowerOf2((ALuint)(MASTER_LINE_LENGTH*Context->Frequency) + 1);
+ totalLength = length[0];
+ for(index = 0;index < 4;index++)
+ {
+ length[1+index] = NextPowerOf2((ALuint)(EARLY_LINE_LENGTH[index]*Context->Frequency) + 1);
+ totalLength += length[1+index];
+ }
+ for(index = 0;index < 4;index++)
+ {
+ length[5+index] = NextPowerOf2((ALuint)(LATE_LINE_LENGTH[index]*Context->Frequency) + 1);
+ totalLength += length[5+index];
+ }
+ for(index = 4;index < 8;index++)
+ {
+ length[5+index] = NextPowerOf2((ALuint)(LATE_LINE_LENGTH[index]*(1.0f + LATE_LINE_MULTIPLIER)*Context->Frequency) + 1);
+ totalLength += length[5+index];
+ }
+
+ // They all share a single sample buffer.
+ State->SampleBuffer = malloc(totalLength * sizeof(ALfloat));
+ if(!State->SampleBuffer)
+ {
+ free(State);
+ return NULL;
+ }
+ for(index = 0; index < totalLength;index++)
+ State->SampleBuffer[index] = 0.0f;
+
+ // Each one has its mask and start address calculated one time.
+ State->Gain = 0.0f;
+ State->Delay.Mask = length[0] - 1;
+ State->Delay.Line = &State->SampleBuffer[0];
+ totalLength = length[0];
+
+ State->Tap[0] = 0;
+ State->Tap[1] = 0;
+
+ State->Early.Gain = 0.0f;
+ // All fixed-length delay lines have their read-write offsets calculated
+ // one time.
+ for(index = 0;index < 4;index++)
+ {
+ State->Early.Coeff[index] = 0.0f;
+ State->Early.Delay[index].Mask = length[1 + index] - 1;
+ State->Early.Delay[index].Line = &State->SampleBuffer[totalLength];
+ totalLength += length[1 + index];
+
+ State->Early.Offset[index] = (ALuint)(EARLY_LINE_LENGTH[index] * Context->Frequency);
+ }
+
+ State->Late.Gain = 0.0f;
+ State->Late.Diffusion = 0.0f;
+ for(index = 0;index < 8;index++)
+ {
+ State->Late.Coeff[index] = 0.0f;
+ State->Late.Delay[index].Mask = length[5 + index] - 1;
+ State->Late.Delay[index].Line = &State->SampleBuffer[totalLength];
+ totalLength += length[5 + index];
+
+ State->Late.Offset[index] = 0;
+ if(index < 4)
+ {
+ State->Late.Offset[index] = (ALuint)(LATE_LINE_LENGTH[index] * Context->Frequency);
+ State->Late.LpCoeff[index] = 0.0f;
+ State->Late.LpSample[index] = 0.0f;
+ }
+ else if(index == 4)
+ {
+ State->Late.LpCoeff[index] = 0.0f;
+ State->Late.LpSample[index] = 0.0f;
+ }
+ }
+
+ State->Offset = 0;
+ return State;
+}
+
+// This destroys the reverb state. It should be called only when the effect
+// slot has a different (or no) effect loaded over the reverb effect.
+ALvoid VerbDestroy(ALverbState *State)
+{
+ if(State)
+ {
+ free(State->SampleBuffer);
+ State->SampleBuffer = NULL;
+ free(State);
+ }
+}
+
+// This updates the reverb state. This is called any time the reverb effect
+// is loaded into a slot.
+ALvoid VerbUpdate(ALCcontext *Context, ALeffectslot *Slot, ALeffect *Effect)
+{
+ ALverbState *State = Slot->ReverbState;
+ ALuint index, index2;
+ ALfloat length, lpcoeff, cw, g;
+ ALfloat hfRatio = Effect->Reverb.DecayHFRatio;
+
+ // Calculate the master gain (from the slot and master reverb gain).
+ State->Gain = Slot->Gain * Effect->Reverb.Gain;
+
+ // Calculate the initial delay taps.
+ length = Effect->Reverb.ReflectionsDelay;
+ State->Tap[0] = (ALuint)(length * Context->Frequency);
+ length += Effect->Reverb.LateReverbDelay;
+ State->Tap[1] = (ALuint)(length * Context->Frequency);
+
+ // Calculate the early reflections gain. Right now this uses a gain of
+ // 0.75 to compensate for the increase in density. It should probably
+ // use a power (RMS) based measurement from the resulting distribution of
+ // early delay lines.
+ State->Early.Gain = Effect->Reverb.ReflectionsGain * 0.75f;
+
+ // Calculate the gain (coefficient) for each early delay line.
+ for(index = 0;index < 4;index++)
+ State->Early.Coeff[index] = pow(10.0f, EARLY_LINE_LENGTH[index] /
+ Effect->Reverb.LateReverbDelay *
+ -60.0f / 20.0f);
+
+ // Calculate the late reverb gain, adjusted by density, diffusion, and
+ // decay time. To be accurate, the adjustments should probably use power
+ // measurements for each contribution, but they are not too bad as they
+ // are.
+ State->Late.Gain = Effect->Reverb.LateReverbGain *
+ (0.45f + (0.55f * Effect->Reverb.Density)) *
+ (1.0f - (0.25f * Effect->Reverb.Diffusion)) *
+ (1.0f - (0.025f * Effect->Reverb.DecayTime));
+ State->Late.Diffusion = Effect->Reverb.Diffusion;
+
+ // The EFX specification does not make it clear whether the air
+ // absorption parameter should always take effect. Both Generic Software
+ // and Generic Hardware only apply it when HF limit is flagged, so that's
+ // what is done here.
+ // If the HF limit parameter is flagged, calculate an appropriate limit
+ // based on the air absorption parameter.
+ if(Effect->Reverb.DecayHFLimit)
+ {
+ ALfloat limitRatio;
+
+ // The following is my best guess at how to limit the HF ratio by the
+ // air absorption parameter.
+ // For each of the last 4 delays, find the attenuation due to air
+ // absorption in dB (converting delay time to meters using the speed
+ // of sound). Then reversing the decay equation, solve for HF ratio.
+ // The delay length is cancelled out of the equation, so it can be
+ // calculated once for all lines.
+ limitRatio = 1.0f / (log10(Effect->Reverb.AirAbsorptionGainHF) *
+ SPEEDOFSOUNDMETRESPERSEC*Effect->Reverb.DecayTime/
+ -60.0f * 20.0f);
+ // Need to limit the result to a minimum of 0.1, just like the HF
+ // ratio parameter.
+ limitRatio = __max(limitRatio, 0.1f);
+
+ // Using the limit calculated above, apply the upper bound to the
+ // HF ratio.
+ hfRatio = __min(hfRatio, limitRatio);
+ }
+
+ cw = cos(2.0f*3.141592654f * LOWPASSFREQCUTOFF / Context->Frequency);
+
+ for(index = 0;index < 8;index++)
+ {
+ // Calculate the length (in seconds) of each delay line.
+ length = LATE_LINE_LENGTH[index];
+ if(index >= 4)
+ {
+ index2 = index - 3;
+
+ length *= 1.0f + (Effect->Reverb.Density * LATE_LINE_MULTIPLIER);
+
+ // Calculate the delay offset for the variable-length delay
+ // lines.
+ State->Late.Offset[index] = (ALuint)(length * Context->Frequency);
+
+ // Calculate the decay equation for each low-pass filter.
+ g = pow(10.0f, length / (Effect->Reverb.DecayTime * hfRatio) *
+ -60.0f / 20.0f);
+ g = __max(g, 0.1f);
+ g *= g;
+ // Calculate the gain (coefficient) for each low-pass filter.
+ lpcoeff = 0.0f;
+ if(g < 0.9999f) // 1-epsilon
+ lpcoeff = (1 - g*cw - aluSqrt(2*g*(1-cw) - g*g*(1 - cw*cw))) / (1 - g);
+
+ // Very low decay times will produce minimal output, so apply an
+ // upper bound to the coefficient.
+ State->Late.LpCoeff[index2] = __min(lpcoeff, 0.98f);
+ }
+
+ // Calculate the gain (coefficient) for each line.
+ State->Late.Coeff[index] = pow(10.0f, length / Effect->Reverb.DecayTime *
+ -60.0f / 20.0f);
+ }
+
+ // This just calculates the coefficient for the late reverb input low-
+ // pass filter. It is calculated based the average (hence -30 instead
+ // of -60) length of the inner two variable-length delay lines.
+ length = LATE_LINE_LENGTH[5] * (1.0f + Effect->Reverb.Density * LATE_LINE_MULTIPLIER) +
+ LATE_LINE_LENGTH[6] * (1.0f + Effect->Reverb.Density * LATE_LINE_MULTIPLIER);
+
+ g = pow(10.0f, length / (Effect->Reverb.DecayTime * hfRatio) * -30.0f / 20.0f);
+ g = __max(g, 0.1f);
+ g *= g;
+
+ lpcoeff = 0.0f;
+ if(g < 0.9999f) // 1-epsilon
+ lpcoeff = (1 - g*cw - aluSqrt(2*g*(1-cw) - g*g*(1 - cw*cw))) / (1 - g);
+
+ State->Late.LpCoeff[0] = __min(lpcoeff, 0.98f);
+}
+
+// This processes the reverb state, given the input samples and an output
+// buffer.
+ALvoid VerbProcess(ALverbState *State, ALuint SamplesToDo, const ALfloat *SamplesIn, ALfloat (*SamplesOut)[OUTPUTCHANNELS])
+{
+ ALuint index;
+ ALfloat in, early[2], late[2], out[2];
+
+ for(index = 0;index < SamplesToDo;index++)
+ {
+ // Feed the initial delay line.
+ DelayLineIn(&State->Delay, State->Offset, SamplesIn[index]);
+
+ // Calculate the early reflection from the first delay tap.
+ in = DelayLineOut(&State->Delay, State->Offset - State->Tap[0]);
+ EarlyReflection(State, in, early);
+
+ // Calculate the late reverb from the second delay tap.
+ in = DelayLineOut(&State->Delay, State->Offset - State->Tap[1]);
+ LateReverb(State, in, late);
+
+ // Mix early reflections and late reverb.
+ out[0] = State->Gain * (early[0] + late[0]);
+ out[1] = State->Gain * (early[1] + late[1]);
+
+ // Step all delays forward one sample.
+ State->Offset++;
+
+ // Output the results.
+ SamplesOut[index][FRONT_LEFT] += out[0];
+ SamplesOut[index][FRONT_RIGHT] += out[1];
+ SamplesOut[index][SIDE_LEFT] += out[0];
+ SamplesOut[index][SIDE_RIGHT] += out[1];
+ SamplesOut[index][BACK_LEFT] += out[0];
+ SamplesOut[index][BACK_RIGHT] += out[1];
+ }
+}