aboutsummaryrefslogblamecommitdiffstats
path: root/plugins/WinVST/Biquad/BiquadProc.cpp
blob: 0dfbfe46fbdec37e7bf337dfecbb69a9e13532ba (plain) (tree)

































































































































































































































































































































                                                                                                              
/* ========================================
 *  Biquad - Biquad.h
 *  Copyright (c) 2016 airwindows, All rights reserved
 * ======================================== */

#ifndef __Biquad_H
#include "Biquad.h"
#endif

void Biquad::processReplacing(float **inputs, float **outputs, VstInt32 sampleFrames) 
{
    float* in1  =  inputs[0];
    float* in2  =  inputs[1];
    float* out1 = outputs[0];
    float* out2 = outputs[1];

	double overallscale = 1.0;
	overallscale /= 44100.0;
	overallscale *= getSampleRate();

	int type = ceil((A*3.999)+0.00001);
	
	biquad[0] = ((B*B*B*0.9999)+0.0001)*0.499;
	if (biquad[0] < 0.0001) biquad[0] = 0.0001;
	
    biquad[1] = (C*C*C*29.99)+0.01;
	if (biquad[1] < 0.0001) biquad[1] = 0.0001;
	
	double wet = (D*2.0)-1.0;
	
	//biquad contains these values:
	//[0] is frequency: 0.000001 to 0.499999 is near-zero to near-Nyquist
	//[1] is resonance, 0.7071 is Butterworth. Also can't be zero
	//[2] is a0 but you need distinct ones for additional biquad instances so it's here
	//[3] is a1 but you need distinct ones for additional biquad instances so it's here
	//[4] is a2 but you need distinct ones for additional biquad instances so it's here
	//[5] is b1 but you need distinct ones for additional biquad instances so it's here
	//[6] is b2 but you need distinct ones for additional biquad instances so it's here
	//[7] is LEFT stored delayed sample (freq and res are stored so you can move them sample by sample)
	//[8] is LEFT stored delayed sample (you have to include the coefficient making code if you do that)
	//[9] is RIGHT stored delayed sample (freq and res are stored so you can move them sample by sample)
	//[10] is RIGHT stored delayed sample (you have to include the coefficient making code if you do that)
	
	//to build a dedicated filter, rename 'biquad' to whatever the new filter is, then
	//put this code either within the sample buffer (for smoothly modulating freq or res)
	//or in this 'read the controls' area (for letting you change freq and res with controls)
	//or in 'reset' if the freq and res are absolutely fixed (use GetSampleRate to define freq)
	
	if (type == 1) { //lowpass
		double K = tan(M_PI * biquad[0]);
		double norm = 1.0 / (1.0 + K / biquad[1] + K * K);
		biquad[2] = K * K * norm;
		biquad[3] = 2.0 * biquad[2];
		biquad[4] = biquad[2];
		biquad[5] = 2.0 * (K * K - 1.0) * norm;
		biquad[6] = (1.0 - K / biquad[1] + K * K) * norm;
	}
	
	if (type == 2) { //highpass
		double K = tan(M_PI * biquad[0]);
		double norm = 1.0 / (1.0 + K / biquad[1] + K * K);
		biquad[2] = norm;
		biquad[3] = -2.0 * biquad[2];
		biquad[4] = biquad[2];
		biquad[5] = 2.0 * (K * K - 1.0) * norm;
		biquad[6] = (1.0 - K / biquad[1] + K * K) * norm;
	}
	
	if (type == 3) { //bandpass
		double K = tan(M_PI * biquad[0]);
		double norm = 1.0 / (1.0 + K / biquad[1] + K * K);
		biquad[2] = K / biquad[1] * norm;
		biquad[3] = 0.0; //bandpass can simplify the biquad kernel: leave out this multiply
		biquad[4] = -biquad[2];
		biquad[5] = 2.0 * (K * K - 1.0) * norm;
		biquad[6] = (1.0 - K / biquad[1] + K * K) * norm;
	}
	
	if (type == 4) { //notch
		double K = tan(M_PI * biquad[0]);
		double norm = 1.0 / (1.0 + K / biquad[1] + K * K);
		biquad[2] = (1.0 + K * K) * norm;
		biquad[3] = 2.0 * (K * K - 1) * norm;
		biquad[4] = biquad[2];
		biquad[5] = biquad[3];
		biquad[6] = (1.0 - K / biquad[1] + K * K) * norm;
	}
	    
    while (--sampleFrames >= 0)
    {
		long double inputSampleL = *in1;
		long double inputSampleR = *in2;
		if (fabs(inputSampleL)<1.18e-37) inputSampleL = fpd * 1.18e-37;
		if (fabs(inputSampleR)<1.18e-37) inputSampleR = fpd * 1.18e-37;
		long double drySampleL = inputSampleL;
		long double drySampleR = inputSampleR;
		
		
		inputSampleL = sin(inputSampleL);
		inputSampleR = sin(inputSampleR);
		//encode Console5: good cleanness
		
		/*
		 long double mid = inputSampleL + inputSampleR;
		 long double side = inputSampleL - inputSampleR;
		 //assign mid and side.Between these sections, you can do mid/side processing
		 
		 long double tempSampleM = (mid * biquad[2]) + biquad[7];
		 biquad[7] = (mid * biquad[3]) - (tempSampleM * biquad[5]) + biquad[8];
		 biquad[8] = (mid * biquad[4]) - (tempSampleM * biquad[6]);
		 mid = tempSampleM; //like mono AU, 7 and 8 store mid channel
		 
		 long double tempSampleS = (side * biquad[2]) + biquad[9];
		 biquad[9] = (side * biquad[3]) - (tempSampleS * biquad[5]) + biquad[10];
		 biquad[10] = (side * biquad[4]) - (tempSampleS * biquad[6]);
		 inputSampleR = tempSampleS; //note: 9 and 10 store the side channel		
		 
		 inputSampleL = (mid+side)/2.0;
		 inputSampleR = (mid-side)/2.0;
		 //unassign mid and side
		 */
		
		long double tempSampleL = (inputSampleL * biquad[2]) + biquad[7];
		biquad[7] = (inputSampleL * biquad[3]) - (tempSampleL * biquad[5]) + biquad[8];
		biquad[8] = (inputSampleL * biquad[4]) - (tempSampleL * biquad[6]);
		inputSampleL = tempSampleL; //like mono AU, 7 and 8 store L channel
		
		long double tempSampleR = (inputSampleR * biquad[2]) + biquad[9];
		biquad[9] = (inputSampleR * biquad[3]) - (tempSampleR * biquad[5]) + biquad[10];
		biquad[10] = (inputSampleR * biquad[4]) - (tempSampleR * biquad[6]);
		inputSampleR = tempSampleR; //note: 9 and 10 store the R channel
		
		if (inputSampleL > 1.0) inputSampleL = 1.0;
		if (inputSampleL < -1.0) inputSampleL = -1.0;
		if (inputSampleR > 1.0) inputSampleR = 1.0;
		if (inputSampleR < -1.0) inputSampleR = -1.0;
		//without this, you can get a NaN condition where it spits out DC offset at full blast!
		inputSampleL = asin(inputSampleL);
		inputSampleR = asin(inputSampleR);
		//amplitude aspect
		
		if (wet < 1.0) {
			inputSampleL = (inputSampleL*wet) + (drySampleL*(1.0-fabs(wet)));
			inputSampleR = (inputSampleR*wet) + (drySampleR*(1.0-fabs(wet)));
			//inv/dry/wet lets us turn LP into HP and band into notch
		}
		
		//begin 32 bit stereo floating point dither
		int expon; frexpf((float)inputSampleL, &expon);
		fpd ^= fpd << 13; fpd ^= fpd >> 17; fpd ^= fpd << 5;
		inputSampleL += ((double(fpd)-uint32_t(0x7fffffff)) * 5.5e-36l * pow(2,expon+62));
		frexpf((float)inputSampleR, &expon);
		fpd ^= fpd << 13; fpd ^= fpd >> 17; fpd ^= fpd << 5;
		inputSampleR += ((double(fpd)-uint32_t(0x7fffffff)) * 5.5e-36l * pow(2,expon+62));
		//end 32 bit stereo floating point dither
		
		*out1 = inputSampleL;
		*out2 = inputSampleR;

		*in1++;
		*in2++;
		*out1++;
		*out2++;
    }
}

void Biquad::processDoubleReplacing(double **inputs, double **outputs, VstInt32 sampleFrames) 
{
    double* in1  =  inputs[0];
    double* in2  =  inputs[1];
    double* out1 = outputs[0];
    double* out2 = outputs[1];

	double overallscale = 1.0;
	overallscale /= 44100.0;
	overallscale *= getSampleRate();
	
	int type = ceil((A*3.999)+0.00001);
	
	biquad[0] = ((B*B*B*0.9999)+0.0001)*0.499;
	if (biquad[0] < 0.0001) biquad[0] = 0.0001;
	
    biquad[1] = (C*C*C*29.99)+0.01;
	if (biquad[1] < 0.0001) biquad[1] = 0.0001;
	
	double wet = (D*2.0)-1.0;
	
	//biquad contains these values:
	//[0] is frequency: 0.000001 to 0.499999 is near-zero to near-Nyquist
	//[1] is resonance, 0.7071 is Butterworth. Also can't be zero
	//[2] is a0 but you need distinct ones for additional biquad instances so it's here
	//[3] is a1 but you need distinct ones for additional biquad instances so it's here
	//[4] is a2 but you need distinct ones for additional biquad instances so it's here
	//[5] is b1 but you need distinct ones for additional biquad instances so it's here
	//[6] is b2 but you need distinct ones for additional biquad instances so it's here
	//[7] is LEFT stored delayed sample (freq and res are stored so you can move them sample by sample)
	//[8] is LEFT stored delayed sample (you have to include the coefficient making code if you do that)
	//[9] is RIGHT stored delayed sample (freq and res are stored so you can move them sample by sample)
	//[10] is RIGHT stored delayed sample (you have to include the coefficient making code if you do that)
	
	//to build a dedicated filter, rename 'biquad' to whatever the new filter is, then
	//put this code either within the sample buffer (for smoothly modulating freq or res)
	//or in this 'read the controls' area (for letting you change freq and res with controls)
	//or in 'reset' if the freq and res are absolutely fixed (use GetSampleRate to define freq)
	
	if (type == 1) { //lowpass
		double K = tan(M_PI * biquad[0]);
		double norm = 1.0 / (1.0 + K / biquad[1] + K * K);
		biquad[2] = K * K * norm;
		biquad[3] = 2.0 * biquad[2];
		biquad[4] = biquad[2];
		biquad[5] = 2.0 * (K * K - 1.0) * norm;
		biquad[6] = (1.0 - K / biquad[1] + K * K) * norm;
	}
	
	if (type == 2) { //highpass
		double K = tan(M_PI * biquad[0]);
		double norm = 1.0 / (1.0 + K / biquad[1] + K * K);
		biquad[2] = norm;
		biquad[3] = -2.0 * biquad[2];
		biquad[4] = biquad[2];
		biquad[5] = 2.0 * (K * K - 1.0) * norm;
		biquad[6] = (1.0 - K / biquad[1] + K * K) * norm;
	}
	
	if (type == 3) { //bandpass
		double K = tan(M_PI * biquad[0]);
		double norm = 1.0 / (1.0 + K / biquad[1] + K * K);
		biquad[2] = K / biquad[1] * norm;
		biquad[3] = 0.0; //bandpass can simplify the biquad kernel: leave out this multiply
		biquad[4] = -biquad[2];
		biquad[5] = 2.0 * (K * K - 1.0) * norm;
		biquad[6] = (1.0 - K / biquad[1] + K * K) * norm;
	}
	
	if (type == 4) { //notch
		double K = tan(M_PI * biquad[0]);
		double norm = 1.0 / (1.0 + K / biquad[1] + K * K);
		biquad[2] = (1.0 + K * K) * norm;
		biquad[3] = 2.0 * (K * K - 1) * norm;
		biquad[4] = biquad[2];
		biquad[5] = biquad[3];
		biquad[6] = (1.0 - K / biquad[1] + K * K) * norm;
	}
	
    while (--sampleFrames >= 0)
    {
		long double inputSampleL = *in1;
		long double inputSampleR = *in2;
		if (fabs(inputSampleL)<1.18e-43) inputSampleL = fpd * 1.18e-43;
		if (fabs(inputSampleR)<1.18e-43) inputSampleR = fpd * 1.18e-43;
		long double drySampleL = inputSampleL;
		long double drySampleR = inputSampleR;
		
		
		inputSampleL = sin(inputSampleL);
		inputSampleR = sin(inputSampleR);
		//encode Console5: good cleanness
		
		/*
		 long double mid = inputSampleL + inputSampleR;
		 long double side = inputSampleL - inputSampleR;
		 //assign mid and side.Between these sections, you can do mid/side processing
		 
		 long double tempSampleM = (mid * biquad[2]) + biquad[7];
		 biquad[7] = (mid * biquad[3]) - (tempSampleM * biquad[5]) + biquad[8];
		 biquad[8] = (mid * biquad[4]) - (tempSampleM * biquad[6]);
		 mid = tempSampleM; //like mono AU, 7 and 8 store mid channel
		 
		 long double tempSampleS = (side * biquad[2]) + biquad[9];
		 biquad[9] = (side * biquad[3]) - (tempSampleS * biquad[5]) + biquad[10];
		 biquad[10] = (side * biquad[4]) - (tempSampleS * biquad[6]);
		 inputSampleR = tempSampleS; //note: 9 and 10 store the side channel		
		 
		 inputSampleL = (mid+side)/2.0;
		 inputSampleR = (mid-side)/2.0;
		 //unassign mid and side
		 */
		
		long double tempSampleL = (inputSampleL * biquad[2]) + biquad[7];
		biquad[7] = (inputSampleL * biquad[3]) - (tempSampleL * biquad[5]) + biquad[8];
		biquad[8] = (inputSampleL * biquad[4]) - (tempSampleL * biquad[6]);
		inputSampleL = tempSampleL; //like mono AU, 7 and 8 store L channel
		
		long double tempSampleR = (inputSampleR * biquad[2]) + biquad[9];
		biquad[9] = (inputSampleR * biquad[3]) - (tempSampleR * biquad[5]) + biquad[10];
		biquad[10] = (inputSampleR * biquad[4]) - (tempSampleR * biquad[6]);
		inputSampleR = tempSampleR; //note: 9 and 10 store the R channel
		
		if (inputSampleL > 1.0) inputSampleL = 1.0;
		if (inputSampleL < -1.0) inputSampleL = -1.0;
		if (inputSampleR > 1.0) inputSampleR = 1.0;
		if (inputSampleR < -1.0) inputSampleR = -1.0;
		//without this, you can get a NaN condition where it spits out DC offset at full blast!
		inputSampleL = asin(inputSampleL);
		inputSampleR = asin(inputSampleR);
		//amplitude aspect
		
		if (wet < 1.0) {
			inputSampleL = (inputSampleL*wet) + (drySampleL*(1.0-fabs(wet)));
			inputSampleR = (inputSampleR*wet) + (drySampleR*(1.0-fabs(wet)));
			//inv/dry/wet lets us turn LP into HP and band into notch
		}
		
		//begin 64 bit stereo floating point dither
		int expon; frexp((double)inputSampleL, &expon);
		fpd ^= fpd << 13; fpd ^= fpd >> 17; fpd ^= fpd << 5;
		inputSampleL += ((double(fpd)-uint32_t(0x7fffffff)) * 1.1e-44l * pow(2,expon+62));
		frexp((double)inputSampleR, &expon);
		fpd ^= fpd << 13; fpd ^= fpd >> 17; fpd ^= fpd << 5;
		inputSampleR += ((double(fpd)-uint32_t(0x7fffffff)) * 1.1e-44l * pow(2,expon+62));
		//end 64 bit stereo floating point dither
		
		*out1 = inputSampleL;
		*out2 = inputSampleR;

		*in1++;
		*in2++;
		*out1++;
		*out2++;
    }
}