path: root/plugins/LinuxVST/src/PurestEcho/PurestEchoProc.cpp

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

#ifndef __PurestEcho_H
#include "PurestEcho.h"
#endif

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

	int loopLimit = (int)(totalsamples * 0.499);
	//this is a double buffer so we will be splitting it in two
	
	double time = pow(A,2) * 0.999;
	double tap1 = B;
	double tap2 = C;
	double tap3 = D;
	double tap4 = E;
	
	double gainTrim = 1.0 / (1.0 + tap1 + tap2 + tap3 + tap4);
	//this becomes our equal-loudness mechanic. 0.2 to 1.0 gain on all things.
	double tapsTrim = gainTrim * 0.5;
	//the taps interpolate and require half that gain: 0.1 to 0.5 on all taps.
	
	int position1 = (int)(loopLimit * time * 0.25);
	int position2 = (int)(loopLimit * time * 0.5);
	int position3 = (int)(loopLimit * time * 0.75);
	int position4 = (int)(loopLimit * time);
	//basic echo information: we're taking four equally spaced echoes and setting their levels as desired.
	//position4 is what you'd have for 'just set a delay time'
	
	double volAfter1 = (loopLimit * time * 0.25) - position1;
	double volAfter2 = (loopLimit * time * 0.5) - position2;
	double volAfter3 = (loopLimit * time * 0.75) - position3;
	double volAfter4 = (loopLimit * time) - position4;
	//these are 0-1: casting to an (int) truncates fractional numbers towards zero (and is faster than floor() )
	//so, when we take the integer number (all above zero) and subtract it from the real value, we get 0-1
	double volBefore1 = (1.0 - volAfter1) * tap1;
	double volBefore2 = (1.0 - volAfter2) * tap2;
	double volBefore3 = (1.0 - volAfter3) * tap3;
	double volBefore4 = (1.0 - volAfter4) * tap4;
	//and if we are including a bit of the previous/next sample to interpolate, then if the sample position is 1.0001
	//we'll be leaning most heavily on the 'before' sample which is nearer to us, and the 'after' sample is almost not used.
	//if the sample position is 1.9999, the 'after' sample is strong and 'before' is almost not used.
	
	volAfter1 *= tap1;
	volAfter2 *= tap2;
	volAfter3 *= tap3;
	volAfter4 *= tap4;
	//and like with volBefore, we also want to scale this 'interpolate' to the loudness of this tap.
	//We do it here because we can do it only once per audio buffer, not on every sample. This assumes we're
	//not moving the tap every sample: if so we'd have to do this every sample as well.	
	
	int oneBefore1 = position1 - 1;
	int oneBefore2 = position2 - 1;
	int oneBefore3 = position3 - 1;
	int oneBefore4 = position4 - 1;
	if (oneBefore1 < 0) oneBefore1 = 0;
	if (oneBefore2 < 0) oneBefore2 = 0;
	if (oneBefore3 < 0) oneBefore3 = 0;
	if (oneBefore4 < 0) oneBefore4 = 0;
	int oneAfter1 = position1 + 1;
	int oneAfter2 = position2 + 1;
	int oneAfter3 = position3 + 1;
	int oneAfter4 = position4 + 1;
	//this is setting up the way we interpolate samples: we're doing an echo-darkening thing
	//to make it sound better. Pretty much no acoustic delay in human-breathable air will give
	//you zero attenuation at 22 kilohertz: forget this at your peril ;)
	
	double delaysBufferL;
	double delaysBufferR;
	

	long double inputSampleL;
	long double inputSampleR;
	    
    while (--sampleFrames >= 0)
    {
		inputSampleL = *in1;
		inputSampleR = *in2;
		if (inputSampleL<1.2e-38 && -inputSampleL<1.2e-38) {
			static int noisesource = 0;
			//this declares a variable before anything else is compiled. It won't keep assigning
			//it to 0 for every sample, it's as if the declaration doesn't exist in this context,
			//but it lets me add this denormalization fix in a single place rather than updating
			//it in three different locations. The variable isn't thread-safe but this is only
			//a random seed and we can share it with whatever.
			noisesource = noisesource % 1700021; noisesource++;
			int residue = noisesource * noisesource;
			residue = residue % 170003; residue *= residue;
			residue = residue % 17011; residue *= residue;
			residue = residue % 1709; residue *= residue;
			residue = residue % 173; residue *= residue;
			residue = residue % 17;
			double applyresidue = residue;
			applyresidue *= 0.00000001;
			applyresidue *= 0.00000001;
			inputSampleL = applyresidue;
		}
		if (inputSampleR<1.2e-38 && -inputSampleR<1.2e-38) {
			static int noisesource = 0;
			noisesource = noisesource % 1700021; noisesource++;
			int residue = noisesource * noisesource;
			residue = residue % 170003; residue *= residue;
			residue = residue % 17011; residue *= residue;
			residue = residue % 1709; residue *= residue;
			residue = residue % 173; residue *= residue;
			residue = residue % 17;
			double applyresidue = residue;
			applyresidue *= 0.00000001;
			applyresidue *= 0.00000001;
			inputSampleR = applyresidue;
			//this denormalization routine produces a white noise at -300 dB which the noise
			//shaping will interact with to produce a bipolar output, but the noise is actually
			//all positive. That should stop any variables from going denormal, and the routine
			//only kicks in if digital black is input. As a final touch, if you save to 24-bit
			//the silence will return to being digital black again.
		}

		if (gcount < 0 || gcount > loopLimit) gcount = loopLimit;
		dL[gcount+loopLimit] = dL[gcount] = inputSampleL * tapsTrim;
		dR[gcount+loopLimit] = dR[gcount] = inputSampleR * tapsTrim; //this is how the double buffer works:
		//we can look for delay taps without ever having to 'wrap around' within our calculation.
		//As long as the delay tap is less than our loop limit we can always just add it to where we're
		//at, and get a valid sample back right away, no matter where we are in the buffer.
		//The 0.5 is taking into account the interpolation, by padding down the whole buffer.
		
		delaysBufferL = (dL[gcount+oneBefore4]*volBefore4);
		delaysBufferL += (dL[gcount+oneAfter4]*volAfter4);
		delaysBufferL += (dL[gcount+oneBefore3]*volBefore3);
		delaysBufferL += (dL[gcount+oneAfter3]*volAfter3);
		delaysBufferL += (dL[gcount+oneBefore2]*volBefore2);
		delaysBufferL += (dL[gcount+oneAfter2]*volAfter2);
		delaysBufferL += (dL[gcount+oneBefore1]*volBefore1);
		delaysBufferL += (dL[gcount+oneAfter1]*volAfter1);

		delaysBufferR = (dR[gcount+oneBefore4]*volBefore4);
		delaysBufferR += (dR[gcount+oneAfter4]*volAfter4);
		delaysBufferR += (dR[gcount+oneBefore3]*volBefore3);
		delaysBufferR += (dR[gcount+oneAfter3]*volAfter3);
		delaysBufferR += (dR[gcount+oneBefore2]*volBefore2);
		delaysBufferR += (dR[gcount+oneAfter2]*volAfter2);
		delaysBufferR += (dR[gcount+oneBefore1]*volBefore1);
		delaysBufferR += (dR[gcount+oneAfter1]*volAfter1);
		//These are the interpolated samples. We're adding them first, because we know they're smaller
		//and while the value of delaysBuffer is small we'll add similarly small values to it. Note the order.
		
		delaysBufferL += (dL[gcount+position4]*tap4);
		delaysBufferL += (dL[gcount+position3]*tap3);
		delaysBufferL += (dL[gcount+position2]*tap2);
		delaysBufferL += (dL[gcount+position1]*tap1);

		delaysBufferR += (dR[gcount+position4]*tap4);
		delaysBufferR += (dR[gcount+position3]*tap3);
		delaysBufferR += (dR[gcount+position2]*tap2);
		delaysBufferR += (dR[gcount+position1]*tap1);
		//These are the primary samples for the echo, and we're adding them last. As before we're starting with the
		//most delayed echoes, and ending with what we think might be the loudest echo. We're building this delaybuffer
		//from the faintest noises to the loudest, to avoid adding a bunch of teeny values at the end.
		//You can of course put the last echo as loudest, but with diminishing echo volumes this is optimal.
		//This technique is also present in other plugins such as Iron Oxide.
		
		inputSampleL = (inputSampleL * gainTrim) + delaysBufferL;
		inputSampleR = (inputSampleR * gainTrim) + delaysBufferR;
		//this could be just inputSample += d[gcount+position1];
		//for literally a single, full volume echo combined with dry.
		//What I'm doing is making the echoes more interesting.
		
		gcount--;
		
		//stereo 32 bit dither, made small and tidy.
		int expon; frexpf((float)inputSampleL, &expon);
		long double dither = (rand()/(RAND_MAX*7.737125245533627e+25))*pow(2,expon+62);
		inputSampleL += (dither-fpNShapeL); fpNShapeL = dither;
		frexpf((float)inputSampleR, &expon);
		dither = (rand()/(RAND_MAX*7.737125245533627e+25))*pow(2,expon+62);
		inputSampleR += (dither-fpNShapeR); fpNShapeR = dither;
		//end 32 bit dither

		*out1 = inputSampleL;
		*out2 = inputSampleR;

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

void PurestEcho::processDoubleReplacing(double **inputs, double **outputs, VstInt32 sampleFrames) 
{
    double* in1  =  inputs[0];
    double* in2  =  inputs[1];
    double* out1 = outputs[0];
    double* out2 = outputs[1];
	
	int loopLimit = (int)(totalsamples * 0.499);
	//this is a double buffer so we will be splitting it in two
	
	double time = pow(A,2) * 0.999;
	double tap1 = B;
	double tap2 = C;
	double tap3 = D;
	double tap4 = E;
	
	double gainTrim = 1.0 / (1.0 + tap1 + tap2 + tap3 + tap4);
	//this becomes our equal-loudness mechanic. 0.2 to 1.0 gain on all things.
	double tapsTrim = gainTrim * 0.5;
	//the taps interpolate and require half that gain: 0.1 to 0.5 on all taps.
	
	int position1 = (int)(loopLimit * time * 0.25);
	int position2 = (int)(loopLimit * time * 0.5);
	int position3 = (int)(loopLimit * time * 0.75);
	int position4 = (int)(loopLimit * time);
	//basic echo information: we're taking four equally spaced echoes and setting their levels as desired.
	//position4 is what you'd have for 'just set a delay time'
	
	double volAfter1 = (loopLimit * time * 0.25) - position1;
	double volAfter2 = (loopLimit * time * 0.5) - position2;
	double volAfter3 = (loopLimit * time * 0.75) - position3;
	double volAfter4 = (loopLimit * time) - position4;
	//these are 0-1: casting to an (int) truncates fractional numbers towards zero (and is faster than floor() )
	//so, when we take the integer number (all above zero) and subtract it from the real value, we get 0-1
	double volBefore1 = (1.0 - volAfter1) * tap1;
	double volBefore2 = (1.0 - volAfter2) * tap2;
	double volBefore3 = (1.0 - volAfter3) * tap3;
	double volBefore4 = (1.0 - volAfter4) * tap4;
	//and if we are including a bit of the previous/next sample to interpolate, then if the sample position is 1.0001
	//we'll be leaning most heavily on the 'before' sample which is nearer to us, and the 'after' sample is almost not used.
	//if the sample position is 1.9999, the 'after' sample is strong and 'before' is almost not used.
	
	volAfter1 *= tap1;
	volAfter2 *= tap2;
	volAfter3 *= tap3;
	volAfter4 *= tap4;
	//and like with volBefore, we also want to scale this 'interpolate' to the loudness of this tap.
	//We do it here because we can do it only once per audio buffer, not on every sample. This assumes we're
	//not moving the tap every sample: if so we'd have to do this every sample as well.	
	
	int oneBefore1 = position1 - 1;
	int oneBefore2 = position2 - 1;
	int oneBefore3 = position3 - 1;
	int oneBefore4 = position4 - 1;
	if (oneBefore1 < 0) oneBefore1 = 0;
	if (oneBefore2 < 0) oneBefore2 = 0;
	if (oneBefore3 < 0) oneBefore3 = 0;
	if (oneBefore4 < 0) oneBefore4 = 0;
	int oneAfter1 = position1 + 1;
	int oneAfter2 = position2 + 1;
	int oneAfter3 = position3 + 1;
	int oneAfter4 = position4 + 1;
	//this is setting up the way we interpolate samples: we're doing an echo-darkening thing
	//to make it sound better. Pretty much no acoustic delay in human-breathable air will give
	//you zero attenuation at 22 kilohertz: forget this at your peril ;)
	
	double delaysBufferL;
	double delaysBufferR;
	

	long double inputSampleL;
	long double inputSampleR;

    while (--sampleFrames >= 0)
    {
		inputSampleL = *in1;
		inputSampleR = *in2;
		if (inputSampleL<1.2e-38 && -inputSampleL<1.2e-38) {
			static int noisesource = 0;
			//this declares a variable before anything else is compiled. It won't keep assigning
			//it to 0 for every sample, it's as if the declaration doesn't exist in this context,
			//but it lets me add this denormalization fix in a single place rather than updating
			//it in three different locations. The variable isn't thread-safe but this is only
			//a random seed and we can share it with whatever.
			noisesource = noisesource % 1700021; noisesource++;
			int residue = noisesource * noisesource;
			residue = residue % 170003; residue *= residue;
			residue = residue % 17011; residue *= residue;
			residue = residue % 1709; residue *= residue;
			residue = residue % 173; residue *= residue;
			residue = residue % 17;
			double applyresidue = residue;
			applyresidue *= 0.00000001;
			applyresidue *= 0.00000001;
			inputSampleL = applyresidue;
		}
		if (inputSampleR<1.2e-38 && -inputSampleR<1.2e-38) {
			static int noisesource = 0;
			noisesource = noisesource % 1700021; noisesource++;
			int residue = noisesource * noisesource;
			residue = residue % 170003; residue *= residue;
			residue = residue % 17011; residue *= residue;
			residue = residue % 1709; residue *= residue;
			residue = residue % 173; residue *= residue;
			residue = residue % 17;
			double applyresidue = residue;
			applyresidue *= 0.00000001;
			applyresidue *= 0.00000001;
			inputSampleR = applyresidue;
			//this denormalization routine produces a white noise at -300 dB which the noise
			//shaping will interact with to produce a bipolar output, but the noise is actually
			//all positive. That should stop any variables from going denormal, and the routine
			//only kicks in if digital black is input. As a final touch, if you save to 24-bit
			//the silence will return to being digital black again.
		}
		
		if (gcount < 0 || gcount > loopLimit) gcount = loopLimit;
		dL[gcount+loopLimit] = dL[gcount] = inputSampleL * tapsTrim;
		dR[gcount+loopLimit] = dR[gcount] = inputSampleR * tapsTrim; //this is how the double buffer works:
		//we can look for delay taps without ever having to 'wrap around' within our calculation.
		//As long as the delay tap is less than our loop limit we can always just add it to where we're
		//at, and get a valid sample back right away, no matter where we are in the buffer.
		//The 0.5 is taking into account the interpolation, by padding down the whole buffer.
		
		delaysBufferL = (dL[gcount+oneBefore4]*volBefore4);
		delaysBufferL += (dL[gcount+oneAfter4]*volAfter4);
		delaysBufferL += (dL[gcount+oneBefore3]*volBefore3);
		delaysBufferL += (dL[gcount+oneAfter3]*volAfter3);
		delaysBufferL += (dL[gcount+oneBefore2]*volBefore2);
		delaysBufferL += (dL[gcount+oneAfter2]*volAfter2);
		delaysBufferL += (dL[gcount+oneBefore1]*volBefore1);
		delaysBufferL += (dL[gcount+oneAfter1]*volAfter1);
		
		delaysBufferR = (dR[gcount+oneBefore4]*volBefore4);
		delaysBufferR += (dR[gcount+oneAfter4]*volAfter4);
		delaysBufferR += (dR[gcount+oneBefore3]*volBefore3);
		delaysBufferR += (dR[gcount+oneAfter3]*volAfter3);
		delaysBufferR += (dR[gcount+oneBefore2]*volBefore2);
		delaysBufferR += (dR[gcount+oneAfter2]*volAfter2);
		delaysBufferR += (dR[gcount+oneBefore1]*volBefore1);
		delaysBufferR += (dR[gcount+oneAfter1]*volAfter1);
		//These are the interpolated samples. We're adding them first, because we know they're smaller
		//and while the value of delaysBuffer is small we'll add similarly small values to it. Note the order.
		
		delaysBufferL += (dL[gcount+position4]*tap4);
		delaysBufferL += (dL[gcount+position3]*tap3);
		delaysBufferL += (dL[gcount+position2]*tap2);
		delaysBufferL += (dL[gcount+position1]*tap1);
		
		delaysBufferR += (dR[gcount+position4]*tap4);
		delaysBufferR += (dR[gcount+position3]*tap3);
		delaysBufferR += (dR[gcount+position2]*tap2);
		delaysBufferR += (dR[gcount+position1]*tap1);
		//These are the primary samples for the echo, and we're adding them last. As before we're starting with the
		//most delayed echoes, and ending with what we think might be the loudest echo. We're building this delaybuffer
		//from the faintest noises to the loudest, to avoid adding a bunch of teeny values at the end.
		//You can of course put the last echo as loudest, but with diminishing echo volumes this is optimal.
		//This technique is also present in other plugins such as Iron Oxide.
		
		inputSampleL = (inputSampleL * gainTrim) + delaysBufferL;
		inputSampleR = (inputSampleR * gainTrim) + delaysBufferR;
		//this could be just inputSample += d[gcount+position1];
		//for literally a single, full volume echo combined with dry.
		//What I'm doing is making the echoes more interesting.
		
		gcount--;
		
		//stereo 64 bit dither, made small and tidy.
		int expon; frexp((double)inputSampleL, &expon);
		long double dither = (rand()/(RAND_MAX*7.737125245533627e+25))*pow(2,expon+62);
		dither /= 536870912.0; //needs this to scale to 64 bit zone
		inputSampleL += (dither-fpNShapeL); fpNShapeL = dither;
		frexp((double)inputSampleR, &expon);
		dither = (rand()/(RAND_MAX*7.737125245533627e+25))*pow(2,expon+62);
		dither /= 536870912.0; //needs this to scale to 64 bit zone
		inputSampleR += (dither-fpNShapeR); fpNShapeR = dither;
		//end 64 bit dither

		*out1 = inputSampleL;
		*out2 = inputSampleR;

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