allanger/paulxstretch
1
0
Fork 0
Code Issues 1 Pull Requests 1 Actions Packages Projects Releases 2 Activity
b5492f2f60
paulxstretch/Source/PS_Source/Stretch.cpp
essej c37aab3dcc fixed scaling issue with VDSP fft impl
2022-04-18 11:46:04 -04:00

510 lines
14 KiB
C++
Raw Blame History

/*
Copyright (C) 2006-2011 Nasca Octavian Paul
Author: Nasca Octavian Paul
This program is free software; you can redistribute it and/or modify
it under the terms of version 2 of the GNU General Public License
as published by the Free Software Foundation.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License (version 2) for more details.
You should have received a copy of the GNU General Public License (version 2)
along with this program; if not, write to the Free Software Foundation,
Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#if PS_USE_VDSP_FFT
#define VIMAGE_H // crazy hack needed
#include <Accelerate/Accelerate.h>
//#include <vForce.h>
#endif
#include "Stretch.h"
#include <stdlib.h>
#include <math.h>
FFT::FFT(int nsamples_, bool no_inverse)
{
nsamples=nsamples_;
if (nsamples%2!=0) {
nsamples+=1;
Logger::writeToLog("WARNING: Odd sample size on FFT::FFT() "+String(nsamples));
};
smp.resize(nsamples);
for (int i = 0; i < nsamples; i++)
smp[i] = 0.0;
freq.resize(nsamples/2+1);
for (int i=0;i<nsamples/2+1;i++)
freq[i]=0.0;
window.data.resize(nsamples);
for (int i=0;i<nsamples;i++)
window.data[i]=0.707f;
window.type=W_RECTANGULAR;
data.resize(nsamples,true);
bool allow_long_planning = false; // g_propsfile->getBoolValue("fftw_allow_long_planning", false);
//double t0 = Time::getMillisecondCounterHiRes();
#if PS_USE_VDSP_FFT
int maxlog2N = 1;
while ((1 << maxlog2N) < nsamples)
++maxlog2N;
log2N = maxlog2N;
planfft = vDSP_create_fftsetup(maxlog2N, kFFTRadix2);
m_workReal.resize(nsamples,false);
m_workImag.resize(nsamples,false);
//Logger::writeToLog("fftsize: " + String(nsamples) + " log2N: " + String(log2N));
#else
if (allow_long_planning)
{
//fftwf_plan_with_nthreads(2);
planfftw=fftwf_plan_r2r_1d(nsamples,data.data(),data.data(),FFTW_R2HC,FFTW_MEASURE);
if (no_inverse == false)
planifftw=fftwf_plan_r2r_1d(nsamples,data.data(),data.data(),FFTW_HC2R,FFTW_MEASURE);
} else
{
//fftwf_plan_with_nthreads(2);
planfftw=fftwf_plan_r2r_1d(nsamples,data.data(),data.data(),FFTW_R2HC,FFTW_ESTIMATE);
//fftwf_plan_with_nthreads(2);
if (no_inverse == false)
planifftw=fftwf_plan_r2r_1d(nsamples,data.data(),data.data(),FFTW_HC2R,FFTW_ESTIMATE);
}
#endif
//double t1 = Time::getMillisecondCounterHiRes();
//Logger::writeToLog("Creating FFTW3 plans took "+String(t1-t0)+ "ms");
static int seed = 0;
m_randgen = std::mt19937(seed);
++seed;
};
FFT::~FFT()
{
#if PS_USE_VDSP_FFT
vDSP_destroy_fftsetup((FFTSetup)planfft);
#else
fftwf_destroy_plan(planfftw);
if (planifftw!=nullptr)
fftwf_destroy_plan(planifftw);
#endif
};
void FFT::smp2freq()
{
#if PS_USE_VDSP_FFT
const int halfsamples = nsamples / 2;
COMPLEX_SPLIT A;
A.realp = m_workReal.data();
A.imagp = m_workImag.data();
// apply window
//vDSP_vmul(gInFIFO, 1, mWindow, 1, gFFTworksp, 1, fftFrameSize);
//convert real input to even-odd
vDSP_ctoz((COMPLEX*)smp.data(), 2, &A, 1, halfsamples);
//memset(ioData->mBuffers[0].mData, 0, ioData->mBuffers[0].mDataByteSize);
// forward fft
vDSP_fft_zrip((FFTSetup)planfft, &A, 1, log2N, FFT_FORWARD);
// result is in split packed complex A.realp[0] is DC, A.imagp[0] is NY, so we zero NY before doing mag^2
A.imagp[0] = 0.0f;
//DebugLogC("post fft: %g %g\n", A.realp[fftFrameSize/4], A.imagp[fftFrameSize/4 + 1]);
// forward scale
const float scale = 0.5f;
vDSP_vsmul(A.realp, 1, &scale, A.realp, 1, halfsamples);
vDSP_vsmul(A.imagp, 1, &scale, A.imagp, 1, halfsamples);
// Absolute square (equivalent to mag^2)
vDSP_zvmags(&A, 1, freq.data(), 1, halfsamples);
// take square root
//vvsqrtf(freq.data(), freq.data(), &halfsamples);
for (int i=1; i < halfsamples;i++)
{
freq[i]=sqrt(freq[i]);
}
freq[0] = 0.0;
#else
for (int i=0;i<nsamples;i++)
data[i]=smp[i];
fftwf_execute(planfftw);
for (int i=1;i<nsamples/2;i++)
{
REALTYPE c=data[i];
REALTYPE s=data[nsamples-i];
freq[i]=sqrt(c*c+s*s);
};
freq[0]=0.0;
#endif
};
void FFT::freq2smp()
{
const REALTYPE inv_2p15_2pi=1.0f/16384.0f*(float)c_PI;
#if PS_USE_VDSP_FFT
const int halfsamples = nsamples / 2;
COMPLEX_SPLIT A;
A.realp = m_workReal.data();
A.imagp = m_workImag.data();
for (int i=1;i<nsamples/2;i++)
{
unsigned int rand = m_randdist(m_randgen);
REALTYPE phase = rand*inv_2p15_2pi;
m_workReal[i] = freq[i]*cos(phase);
m_workImag[i] = freq[i]*sin(phase);
};
m_workReal[0] = m_workImag[0] = 0.0;
// inverse fft
vDSP_fft_zrip((FFTSetup)planfft, &A, 1, log2N, FFT_INVERSE);
// unpack
vDSP_ztoc(&A, 1, (COMPLEX*)data.data(), 2, halfsamples);
// scale
float scale = 1.f / nsamples;
vDSP_vsmul(data.data(), 1, &scale, smp.data(), 1, nsamples);
#else
for (int i=1;i<nsamples/2;i++)
{
unsigned int rand = m_randdist(m_randgen);
REALTYPE phase=rand*inv_2p15_2pi;
data[i]=freq[i]*cos(phase);
data[nsamples-i]=freq[i]*sin(phase);
};
data[0]=data[nsamples/2+1]=data[nsamples/2]=0.0;
fftwf_execute(planifftw);
for (int i=0;i<nsamples;i++)
smp[i]=data[i]/nsamples;
#endif
};
void FFT::applywindow(FFTWindow type)
{
if (window.type!=type){
window.type=type;
switch (type){
case W_RECTANGULAR:
for (int i=0;i<nsamples;i++) window.data[i]=0.707f;
break;
case W_HAMMING:
for (int i=0;i<nsamples;i++) window.data[i]=(float)(0.53836-0.46164*cos(2.0*c_PI*i/(nsamples+1.0)));
break;
case W_HANN:
for (int i=0;i<nsamples;i++) window.data[i]=(float)(0.5*(1.0-cos(2*c_PI*i/(nsamples-1.0))));
break;
case W_BLACKMAN:
for (int i=0;i<nsamples;i++) window.data[i]=(float)(0.42-0.5*cos(2*c_PI*i/(nsamples-1.0))+0.08*cos(4*c_PI*i/(nsamples-1.0)));
break;
case W_BLACKMAN_HARRIS:
for (int i=0;i<nsamples;i++) window.data[i]=(float)(0.35875-0.48829*cos(2*c_PI*i/(nsamples-1.0))+0.14128*cos(4*c_PI*i/(nsamples-1.0))-0.01168*cos(6*c_PI*i/(nsamples-1.0)));
break;
};
};
//for (int i=0;i<nsamples;i++) smp[i]*=window.data[i];
FloatVectorOperations::multiply(smp.data(), window.data.data(), nsamples);
}
Stretch::Stretch(REALTYPE rap_,int /*bufsize_*/,FFTWindow w,bool bypass_,REALTYPE samplerate_,int /*stereo_mode_*/)
{
freezing=false;
onset_detection_sensitivity=0.0;
samplerate=samplerate_;
rap=rap_;
bypass = bypass_;
remained_samples=0.0;
window_type=w;
require_new_buffer=false;
c_pos_percents=0.0;
extra_onset_time_credit=0.0;
skip_samples=0;
};
Stretch::~Stretch()
{
};
void Stretch::set_rap(REALTYPE newrap){
//if ((rap>=1.0)&&(newrap>=1.0))
rap=newrap;
};
void Stretch::do_analyse_inbuf(REALTYPE *smps){
//get the frequencies
FloatVectorOperations::copy(infft->smp.data(), old_smps.data(), bufsize);
FloatVectorOperations::copy(infft->smp.data()+bufsize, smps, bufsize);
FloatVectorOperations::copy(old_freq.data(), infft->freq.data(), bufsize);
/*
for (int i=0;i<bufsize;i++) {
infft->smp[i]=old_smps[i];
infft->smp[i+bufsize]=smps[i];
old_freq[i]=infft->freq[i];
};
*/
infft->applywindow(window_type);
infft->smp2freq();
};
void Stretch::do_next_inbuf_smps(REALTYPE *smps){
FloatVectorOperations::copy(very_old_smps.data(), old_smps.data(), bufsize);
FloatVectorOperations::copy(old_smps.data(), new_smps.data(), bufsize);
FloatVectorOperations::copy(new_smps.data(), smps, bufsize);
/*
for (int i=0;i<bufsize;i++) {
very_old_smps[i]=old_smps[i];
old_smps[i]=new_smps[i];
new_smps[i]=smps[i];
};
*/
};
REALTYPE Stretch::do_detect_onset(){
REALTYPE result=0.0;
if (onset_detection_sensitivity>1e-3){
REALTYPE os=0.0,osinc=0.0;
REALTYPE osincold=1e-5f;
int maxk=1+(int)(bufsize*500.0/(samplerate*0.5));
int k=0;
for (int i=0;i<bufsize;i++) {
osinc+=infft->freq[i]-old_freq[i];
osincold+=old_freq[i];
if (k>=maxk) {
k=0;
os+=osinc/osincold;
osinc=0;
};
k++;
};
os+=osinc;
if (os<0.0) os=0.0;
//if (os>1.0) os=1.0;
REALTYPE os_strength=(float)(pow(20.0,1.0-onset_detection_sensitivity)-1.0);
REALTYPE os_strength_h=os_strength*0.75f;
if (os>os_strength_h){
result=(os-os_strength_h)/(os_strength-os_strength_h);
if (result>1.0f) result=1.0f;
};
if (result>1.0f) result=1.0f;
};
return result;
};
void Stretch::setBufferSize(int bufsize_)
{
if (bufsize == 0 || bufsize_ != bufsize)
{
bufsize = bufsize_;
if (bufsize < 8) bufsize = 8;
out_buf = floatvector(bufsize);
old_freq = floatvector(bufsize);
very_old_smps = floatvector(bufsize);
new_smps = floatvector(bufsize);
old_smps = floatvector(bufsize);
old_out_smps = floatvector(bufsize * 2);
infft = std::make_unique<FFT>(bufsize * 2);
fft = std::make_unique<FFT>(bufsize * 2);
outfft = std::make_unique<FFT>(bufsize * 2);
}
jassert(infft != nullptr && fft != nullptr && outfft != nullptr);
fill_container(outfft->smp, 0.0f);
for (int i = 0; i<bufsize * 2; i++) {
old_out_smps[i] = 0.0;
};
for (int i = 0; i<bufsize; i++) {
old_freq[i] = 0.0f;
new_smps[i] = 0.0f;
old_smps[i] = 0.0f;
very_old_smps[i] = 0.0f;
};
}
REALTYPE Stretch::process(REALTYPE *smps,int nsmps)
{
jassert(bufsize > 0);
REALTYPE onset=0.0;
if (bypass){
//for (int i=0;i<bufsize;i++) out_buf[i]=smps[i];
FloatVectorOperations::copy(out_buf.data(), smps, bufsize);
return 0.0;
};
if (smps!=NULL){
if ((nsmps!=0)&&(nsmps!=bufsize)&&(nsmps!=get_max_bufsize())){
printf("Warning wrong nsmps on Stretch::process() %d,%d\n",nsmps,bufsize);
return 0.0;
};
if (nsmps!=0){//new data arrived: update the frequency components
do_analyse_inbuf(smps);
if (nsmps==get_max_bufsize()) {
for (int k=bufsize;k<get_max_bufsize();k+=bufsize) do_analyse_inbuf(smps+k);
};
if (onset_detection_sensitivity>1e-3) onset=do_detect_onset();
};
//move the buffers
if (nsmps!=0){//new data arrived: update the frequency components
do_next_inbuf_smps(smps);
if (nsmps==get_max_bufsize()) {
for (int k=bufsize;k<get_max_bufsize();k+=bufsize) do_next_inbuf_smps(smps+k);
};
};
//construct the input fft
int start_pos=(int)(floor(remained_samples*bufsize));
if (start_pos>=bufsize) start_pos=bufsize-1;
FloatVectorOperations::copy(fft->smp.data(), very_old_smps.data() + start_pos, bufsize-start_pos);
FloatVectorOperations::copy(fft->smp.data() + (bufsize - start_pos), old_smps.data() , bufsize);
FloatVectorOperations::copy(fft->smp.data() + (2*bufsize - start_pos), new_smps.data() , start_pos);
/*
for (int i=0;i<bufsize-start_pos;i++) fft->smp[i]=very_old_smps[i+start_pos];
for (int i=0;i<bufsize;i++) fft->smp[i+bufsize-start_pos]=old_smps[i];
for (int i=0;i<start_pos;i++) fft->smp[i+2*bufsize-start_pos]=new_smps[i];
*/
//compute the output spectrum
fft->applywindow(window_type);
fft->smp2freq();
//for (int i=0;i<bufsize;i++) outfft->freq[i]=fft->freq[i];
FloatVectorOperations::copy(outfft->freq.data(), fft->freq.data(), bufsize);
//for (int i=0;i<bufsize;i++) outfft->freq[i]=infft->freq[i]*remained_samples+old_freq[i]*(1.0-remained_samples);
process_spectrum(outfft->freq.data());
outfft->freq2smp();
//make the output buffer
REALTYPE tmp=(float)(1.0/(float) bufsize*c_PI);
REALTYPE hinv_sqrt2=0.853553390593f;//(1.0+1.0/sqrt(2))*0.5;
REALTYPE ampfactor=2.0f;
//remove the resulted unwanted amplitude modulation (caused by the interference of N and N+1 windowed buffer and compute the output buffer
for (int i=0;i<bufsize;i++) {
REALTYPE a=(float)((0.5+0.5*cos(i*tmp)));
REALTYPE out=(float)(outfft->smp[i+bufsize]*(1.0-a)+old_out_smps[i]*a);
out_buf[i]=(float)(out*(hinv_sqrt2-(1.0-hinv_sqrt2)*cos(i*2.0*tmp))*ampfactor);
};
//copy the current output buffer to old buffer
//for (int i=0;i<bufsize*2;i++) old_out_smps[i]=outfft->smp[i];
FloatVectorOperations::copy(old_out_smps.data(), outfft->smp.data(), 2*bufsize);
};
if (!freezing){
long double used_rap=rap*get_stretch_multiplier(c_pos_percents);
long double r=1.0/used_rap;
if (extra_onset_time_credit>0){
REALTYPE credit_get=(float)(0.5*r);//must be smaller than r
extra_onset_time_credit-=credit_get;
if (extra_onset_time_credit<0.0) extra_onset_time_credit=0.0;
r-=credit_get;
};
//long double old_remained_samples_test=remained_samples;
remained_samples+=r;
//int result=0;
if (remained_samples>=1.0){
skip_samples=(int)(floor(remained_samples-1.0)*bufsize);
remained_samples=remained_samples-floor(remained_samples);
require_new_buffer=true;
}else{
require_new_buffer=false;
};
};
// long double rf_test=remained_samples-old_remained_samples_test;//this value should be almost like "rf" (for most of the time with the exception of changing the "ri" value) for extremely long stretches (otherwise the shown stretch value is not accurate)
//for stretch up to 10^18x "long double" must have at least 64 bits in the fraction part (true for gcc compiler on x86 and macosx)
return onset;
};
void Stretch::set_onset_detection_sensitivity(REALTYPE detection_sensitivity)
{
onset_detection_sensitivity = detection_sensitivity;
if (detection_sensitivity<1e-3) extra_onset_time_credit = 0.0;
}
void Stretch::here_is_onset(REALTYPE onset){
if (freezing) return;
if (onset>0.5){
require_new_buffer=true;
extra_onset_time_credit+=1.0-remained_samples;
remained_samples=0.0;
skip_samples=0;
};
};
int Stretch::get_nsamples(REALTYPE current_pos_percents){
if (bypass) return bufsize;
if (freezing) return 0;
c_pos_percents=current_pos_percents;
return require_new_buffer?bufsize:0;
};
int Stretch::get_nsamples_for_fill(){
return get_max_bufsize();
};
int Stretch::get_skip_nsamples(){
if (freezing||bypass) return 0;
return skip_samples;
};
REALTYPE Stretch::get_stretch_multiplier(REALTYPE /*pos_percents*/){
return 1.0;
};
Reference in New Issue View Git Blame Copy Permalink