ref: 8fa6f230775bf0c036441bf7d5d941e2312709dc
dir: /LEAF/Src/leaf-math.c/
/*============================================================================== leaf-math.c Created: 22 Jan 2017 7:02:56pm Author: Michael R Mulshine ==============================================================================*/ #if _WIN32 || _WIN64 #include "..\Inc\leaf-math.h" #include "..\Inc\leaf-tables.h" #else #include "../Inc/leaf-math.h" #include "../Inc/leaf-tables.h" #endif // The C-embedded Audio Library. #define TWO_TO_16 65536.f #define EXPONENTIAL_TABLE_SIZE 65536 float interpolate3max(float *buf, const int peakindex) { float a = buf[peakindex-1]; float b = buf[peakindex]; float c = buf[peakindex+1]; float realpeak; realpeak = b + (float)0.125 * (c - a) * (c - a) / ((float)2. * b - a - c); return(realpeak); } float interpolate3phase(float *buf, const int peakindex) { float a = buf[peakindex-1]; float b = buf[peakindex]; float c = buf[peakindex+1]; float fraction; fraction = ((float)0.5 * (c - a)) / ((float)2. * b - a - c); return(fraction); } // alternative implementation for abs() // REQUIRES: 32 bit integers int fastabs_int(int in){ unsigned int r; int const mask = in >> 31; r = (in ^ mask) - mask; return (r); } // alternative implementation for abs() // REQUIRES: 32 bit floats float fastabsf(float f) { union { float f; unsigned int ui; }alias; alias.f = f; alias.ui &= 0x7fffffff; return alias.f; } float fastexpf(float x) { x = 1.0f + (x * 0.0009765625f); x *= x; x *= x; x *= x; x *= x; x *= x; x *= x; x *= x; x *= x; x *= x; x *= x; return x; } float fasterexpf(float x) { x = 1.0 + (x * 0.00390625f); x *= x; x *= x; x *= x; x *= x; x *= x; x *= x; x *= x; x *= x; return x; } // fast floating-point exp2 function taken from Robert Bristow Johnson's // post in the music-dsp list on Date: Tue, 02 Sep 2014 16:50:11 -0400 float fastexp2f(float x) { if (x >= -127.0) { register float accumulator, xPower; register union {float f; int32_t i;} xBits; xBits.i = (int32_t)(x + 4096.0f) - 4096L; /* integer part */ x -= (float)(xBits.i); /* fractional part */ accumulator = 1.0f + 0.69303212081966f*x; xPower = x*x; accumulator += 0.24137976293709f*xPower; xPower *= x; accumulator += 0.05203236900844f*xPower; xPower *= x; accumulator += 0.01355574723481f*xPower; xBits.i += 127; /* bias integer part */ xBits.i<<= 23; /* move biased int part into exponent bits */ return accumulator * xBits.f; } else { return 0.0f; } } /* you pass in a float array to get back two indexes representing the volumes of the left (index 0) and right (index 1) channels when t is -1, volumes[0] = 0, volumes[1] = 1 when t = 0, volumes[0] = 0.707, volumes[1] = 0.707 (equal-power cross fade) when t = 1, volumes[0] = 1, volumes[1] = 0 */ void LEAF_crossfade(float fade, float* volumes) { volumes[0] = sqrtf(0.5f * (1.0f + fade)); volumes[1] = sqrtf(0.5f * (1.0f - fade)); } // dope af float LEAF_chebyshevT(float in, int n){ if (n == 0) return 1; else if (n == 1) return in; else return 2.0f * in * LEAF_chebyshevT(in, n-1) - LEAF_chebyshevT(in, n-2); } #if !(_WIN32 || _WIN64) float LEAF_CompoundChebyshevT(float in, int n, float* amps){ float T[n+1]; T[0] = 1.0f; T[1] = in; for (int i = 2; i <= n; ++i) T[i] = 2*in*T[i-1] - T[i-2]; float out = 0; float amp = 0; for (int i = 0; i < n; ++i){ out += amps[i]*T[i+1]; amp += amps[i]; } return out / amp ; } #endif float LEAF_frequencyToMidi(float f) { return (69.0f + 12.0f * log2(f * INV_440)); } // Jones shaper float LEAF_shaper(float input, float m_drive) { float fx = input * 2.0f; // prescale float w, c, xc, xc2, xc4; xc = LEAF_clip(-SQRT8, fx, SQRT8); xc2 = xc*xc; c = 0.5f*fx*(3.0f - (xc2)); xc4 = xc2 * xc2; w = (1.0f - xc2*0.25f + xc4*0.015625f) * WSCALE; float shaperOut = w*(c+ 0.05f*xc2)*(m_drive + 0.75f); shaperOut *= 0.5f; // post_scale return shaperOut; } // round input to nearest rnd float LEAF_round (float input, float rnd) { rnd = fabsf(rnd); if (rnd <= 0.0000001f) return input; float scale = 1.f / rnd; return roundf(input * scale) / scale; } union unholy_t unholy; float LEAF_bitwise_xor(float input, uint32_t op) { unholy.f = input; unholy.i = (unholy.i ^ op); return unholy.f; } float LEAF_reedTable(float input, float offset, float slope) { float output = offset + (slope * input); if ( output > 1.0f) output = 1.0f; if ( output < -1.0f) output = -1.0f; return output; } float LEAF_softClip(float val, float thresh) { float x; if(val > thresh) { x = thresh / val; return (1.0f - x) * (1.0f - thresh) + thresh; } else if(val < -thresh) { x = -thresh / val; return -((1.0f - x) * (1.0f - thresh) + thresh); } else { return val; } } float LEAF_clip(float min, float val, float max) { if (val < min) { return min; } else if (val > max) { return max; } else { return val; } } int LEAF_clipInt(int min, int val, int max) { if (val < min) { return min; } else if (val > max) { return max; } else { return val; } } oBool LEAF_isPrime(uint64_t number ) { if ( number == 2 ) return OTRUE; if ( number & 1 ) { for ( int i=3; i<(int)sqrt((double)number)+1; i+=2 ) if ( (number % i) == 0 ) return OFALSE; return OTRUE; // prime } else return OFALSE; // even } // Adapted from MusicDSP: http://www.musicdsp.org/showone.php?id=238 float LEAF_tanh(float x) { if( x < -3.0f ) return -1.0f; else if( x > 3.0f ) return 1.0f; else return x * ( 27.0f + x * x ) / ( 27.0f + 9.0f * x * x ); } void LEAF_generate_sine(float* buffer, int size) { float phase; for (int i = 0; i < size; i++) { phase = (float) i / (float) size; buffer[i] = sinf(phase * TWO_PI); } } void LEAF_generate_sawtooth(float* buffer, float basefreq, int size) { int harmonic = 1; float phase = 0.0f; float freq = harmonic * basefreq; float amp; while (freq < (leaf.sampleRate * 0.5)) { amp = 1.0f / harmonic; for (int i = 0; i < size; i++) { phase = (float) i / (float) size; buffer[i] += (amp * sinf(harmonic * phase * TWO_PI)); } harmonic++; freq = harmonic * basefreq; } } void LEAF_generate_triangle(float* buffer, float basefreq, int size) { int harmonic = 1; float phase = 0.0f; float freq = harmonic * basefreq; float amp = 1.0f; int count = 0; float mult = 1.0f; while (freq < (leaf.sampleRate * 0.5)) { amp = 1.0f / (float)(harmonic * harmonic); if (count % 2) mult = -1.0f; else mult = 1.0f; for (int i = 0; i < size; i++) { phase = (float) i / (float) size; buffer[i] += (mult * amp * sinf(harmonic * phase * TWO_PI)); } count++; harmonic += 2; freq = harmonic * basefreq; } } void LEAF_generate_square(float* buffer, float basefreq, int size) { int harmonic = 1; float phase = 0.0f; float freq = harmonic * basefreq; float amp = 1.0f; while (freq < (leaf.sampleRate * 0.5)) { amp = 1.0f / (float)(harmonic); for (int i = 0; i < size; i++) { phase = (float) i / (float) size; buffer[i] += (amp * sinf(harmonic * phase * TWO_PI)); } harmonic += 2; freq = harmonic * basefreq; } } //----------------------------------------------------------------------------- // name: mtof() // desc: midi to freq, from PD source //----------------------------------------------------------------------------- float LEAF_midiToFrequency(float f) { if( f <= -1500.0f ) return (0); else if( f > 1499.0f ) return (LEAF_midiToFrequency(1499.0f)); else return ( powf(2.0f, (f - 69.0f) * 0.083333333333333f) * 440.0f ); } // alpha, [0.0, 1.0] float LEAF_interpolate_hermite (float A, float B, float C, float D, float alpha) { alpha = LEAF_clip(0.0f, alpha, 1.0f); float a = -A*0.5f + (3.0f*B)*0.5f - (3.0f*C)*0.5f + D*0.5f; float b = A - (5.0f*B)*0.5f + 2.0f*C - D * 0.5f; float c = -A*0.5f + C*0.5f; float d = B; return a*alpha*alpha*alpha + b*alpha*alpha + c*alpha + d; } // from http://www.musicdsp.org/archive.php?classid=5#93 //xx is alpha (fractional part of sample value) //grabbed this from Tom Erbe's Delay pd code float LEAF_interpolate_hermite_x(float yy0, float yy1, float yy2, float yy3, float xx) { // 4-point, 3rd-order Hermite (x-form) float c0 = yy1; float c1 = 0.5f * (yy2 - yy0); float y0my1 = yy0 - yy1; float c3 = (yy1 - yy2) + 0.5f * (yy3 - y0my1 - yy2); float c2 = y0my1 + c1 - c3; return ((c3 * xx + c2) * xx + c1) * xx + c0; } // alpha, [0.0, 1.0] float LEAF_interpolation_linear (float A, float B, float alpha) { alpha = LEAF_clip(0.0f, alpha, 1.0f); float omAlpha = 1.0f - alpha; // First 1/2 of interpolation float out = A * omAlpha; out += B * alpha; return out; } #define LOGTEN 2.302585092994 float mtof(float f) { if (f <= -1500.0f) return(0); else if (f > 1499.0f) return(mtof(1499.0f)); else return (8.17579891564f * expf(0.0577622650f * f)); } float fast_mtof(float f) { return (8.17579891564f * fastexpf(0.0577622650f * f)); } float faster_mtof(float f) { return (8.17579891564f * fastexpf(0.0577622650f * f)); } float ftom(float f) { return (f > 0 ? 17.3123405046f * logf(.12231220585f * f) : -1500.0f); } float powtodb(float f) { if (f <= 0) return (0); else { float val = 100 + 10.f/LOGTEN * logf(f); return (val < 0 ? 0 : val); } } float rmstodb(float f) { if (f <= 0) return (0); else { float val = 100 + 20.f/LOGTEN * log(f); return (val < 0 ? 0 : val); } } float dbtopow(float f) { if (f <= 0) return(0); else { if (f > 870.0f) f = 870.0f; return (expf((LOGTEN * 0.1f) * (f-100.0f))); } } float dbtorms(float f) { if (f <= 0) return(0); else { if (f > 485.0f) f = 485.0f; } return (expf((LOGTEN * 0.05f) * (f-100.0f))); }