FT8 support: optimize FFT buffers and plan allocation

This commit is contained in:
f4exb 2023-01-25 16:30:40 +01:00
parent c1c64b23e4
commit c5a9f5fe12
7 changed files with 165 additions and 49 deletions

View File

@ -2,6 +2,7 @@ project(ft8)
set(ft8_SOURCES
fft.cpp
fftbuffers.cpp
ft8.cpp
ft8plan.cpp
ft8plans.cpp
@ -14,6 +15,7 @@ set(ft8_SOURCES
set(ft8_HEADERS
fft.h
fftbuffers.h
ft8.h
ft8plan.h
ft8plans.h

View File

@ -26,14 +26,19 @@
#include "util.h"
#include "ft8plan.h"
#include "ft8plans.h"
#include "fftbuffers.h"
namespace FT8 {
FFTEngine::FFTEngine()
{}
{
m_fftBuffers = new FFTBuffers();
}
FFTEngine::~FFTEngine()
{}
{
delete m_fftBuffers;
}
//
// do just one FFT on samples[i0..i0+block]
@ -57,15 +62,13 @@ std::vector<std::complex<float>> FFTEngine::one_fft(
// assert((int)samples.size() - i0 >= block);
int m_in_allocated = 0;
float *m_in = (float *)samples.data() + i0;
if ((((unsigned long long)m_in) % 16) != 0)
{
// m_in must be on a 16-byte boundary for FFTW.
m_in = (float *)fftwf_malloc(sizeof(float) * p->n_);
m_in = m_fftBuffers->getR(p->n_);
// assert(m_in);
m_in_allocated = 1;
for (int i = 0; i < block; i++)
{
if (i0 + i < nsamples)
@ -79,7 +82,7 @@ std::vector<std::complex<float>> FFTEngine::one_fft(
}
}
fftwf_complex *m_out = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * ((p->n_ / 2) + 1));
fftwf_complex *m_out = m_fftBuffers->getC(p->n_);
// assert(m_out);
fftwf_execute_dft_r2c(plan, m_in, m_out);
@ -93,12 +96,6 @@ std::vector<std::complex<float>> FFTEngine::one_fft(
out[bi] = std::complex<float>(re, im);
}
if (m_in_allocated) {
fftwf_free(m_in);
}
fftwf_free(m_out);
return out;
}
@ -124,8 +121,8 @@ FFTEngine::ffts_t FFTEngine::ffts(const std::vector<float> &samples, int i0, int
fftwf_plan plan = p->fwd_;
// allocate our own b/c using p->m_in and p->m_out isn't thread-safe.
float *m_in = (float *)fftwf_malloc(sizeof(float) * p->n_);
fftwf_complex *m_out = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * ((p->n_ / 2) + 1));
float *m_in = m_fftBuffers->getR(p->n_);
fftwf_complex *m_out = m_fftBuffers->getC(p->n_);
// assert(m_in && m_out);
// float *m_in = p->r_;
@ -158,9 +155,6 @@ FFTEngine::ffts_t FFTEngine::ffts(const std::vector<float> &samples, int i0, int
}
}
fftwf_free(m_in);
fftwf_free(m_out);
return bins;
}
@ -183,8 +177,8 @@ std::vector<std::complex<float>> FFTEngine::one_fft_c(
Plan *p = FT8Plans::GetInstance()->getPlan(block);
fftwf_plan plan = p->cfwd_;
fftwf_complex *m_in = (fftwf_complex *)fftwf_malloc(block * sizeof(fftwf_complex));
fftwf_complex *m_out = (fftwf_complex *)fftwf_malloc(block * sizeof(fftwf_complex));
fftwf_complex *m_in = m_fftBuffers->getCCI(block);
fftwf_complex *m_out = m_fftBuffers->getCCO(block);
// assert(m_in && m_out);
for (int i = 0; i < block; i++)
@ -214,9 +208,6 @@ std::vector<std::complex<float>> FFTEngine::one_fft_c(
out[bi] = c;
}
fftwf_free(m_in);
fftwf_free(m_out);
return out;
}
@ -234,8 +225,8 @@ std::vector<std::complex<float>> FFTEngine::one_fft_cc(
Plan *p = FT8Plans::GetInstance()->getPlan(block);
fftwf_plan plan = p->cfwd_;
fftwf_complex *m_in = (fftwf_complex *)fftwf_malloc(block * sizeof(fftwf_complex));
fftwf_complex *m_out = (fftwf_complex *)fftwf_malloc(block * sizeof(fftwf_complex));
fftwf_complex *m_in = m_fftBuffers->getCCI(block);
fftwf_complex *m_out = m_fftBuffers->getCCO(block);
// assert(m_in && m_out);
for (int i = 0; i < block; i++)
@ -266,9 +257,6 @@ std::vector<std::complex<float>> FFTEngine::one_fft_cc(
out[bi] = c;
}
fftwf_free(m_in);
fftwf_free(m_out);
return out;
}
@ -281,8 +269,8 @@ std::vector<std::complex<float>> FFTEngine::one_ifft_cc(
Plan *p = FT8Plans::GetInstance()->getPlan(block);
fftwf_plan plan = p->crev_;
fftwf_complex *m_in = (fftwf_complex *)fftwf_malloc(block * sizeof(fftwf_complex));
fftwf_complex *m_out = (fftwf_complex *)fftwf_malloc(block * sizeof(fftwf_complex));
fftwf_complex *m_in = m_fftBuffers->getCCI(block);
fftwf_complex *m_out = m_fftBuffers->getCCO(block);
// assert(m_in && m_out);
for (int bi = 0; bi < block; bi++)
@ -306,9 +294,6 @@ std::vector<std::complex<float>> FFTEngine::one_ifft_cc(
out[i] = c;
}
fftwf_free(m_in);
fftwf_free(m_out);
return out;
}
@ -320,8 +305,8 @@ std::vector<float> FFTEngine::one_ifft(const std::vector<std::complex<float>> &b
Plan *p = FT8Plans::GetInstance()->getPlan(block);
fftwf_plan plan = p->rev_;
fftwf_complex *m_in = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * ((p->n_ / 2) + 1));
float *m_out = (float *)fftwf_malloc(sizeof(float) * p->n_);
fftwf_complex *m_in = m_fftBuffers->getC(p->n_);
float *m_out = m_fftBuffers->getR(p->n_);
for (int bi = 0; bi < nbins; bi++)
{
@ -339,9 +324,6 @@ std::vector<float> FFTEngine::one_ifft(const std::vector<std::complex<float>> &b
out[i] = m_out[i];
}
fftwf_free(m_in);
fftwf_free(m_out);
return out;
}

View File

@ -5,6 +5,10 @@
// written by Robert Morris, AB1HL //
// reformatted and adapted to Qt and SDRangel context //
// //
// Caution: this is intentionally not thread safe and one such engine should //
// be allocated by thread. Due to optimization of FFT buffers these buffers are //
// not shared among threads. //
// //
// This program is free software; you can redistribute it and/or modify //
// it under the terms of the GNU General Public License as published by //
// the Free Software Foundation as version 3 of the License, or //
@ -30,6 +34,8 @@
namespace FT8
{
class FFTBuffers;
class FT8_API FFTEngine
{
public:
@ -47,6 +53,7 @@ public:
private:
std::vector<std::complex<float>> analytic(const std::vector<float> &x);
FFTBuffers *m_fftBuffers;
}; // FFTEngine
} // namespace FT8

78
ft8/fftbuffers.cpp Normal file
View File

@ -0,0 +1,78 @@
///////////////////////////////////////////////////////////////////////////////////
// Copyright (C) 2023 Edouard Griffiths, F4EXB. //
// //
// This is the code from ft8mon: https://github.com/rtmrtmrtmrtm/ft8mon //
// written by Robert Morris, AB1HL //
// reformatted and adapted to Qt and SDRangel context //
// //
// This program is free software; you can redistribute it and/or modify //
// it under the terms of the GNU General Public License as published by //
// the Free Software Foundation as version 3 of the License, or //
// (at your option) any later version. //
// //
// 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 V3 for more details. //
// //
// You should have received a copy of the GNU General Public License //
// along with this program. If not, see <http://www.gnu.org/licenses/>. //
///////////////////////////////////////////////////////////////////////////////////
#include "fftbuffers.h"
namespace FT8
{
FFTBuffers::~FFTBuffers()
{
for (auto& mapitem : m_rs) {
fftwf_free(mapitem.second);
}
for (auto& mapitem : m_cs) {
fftwf_free(mapitem.second);
}
for (auto& mapitem : m_ccis) {
fftwf_free(mapitem.second);
}
for (auto& mapitem : m_ccos) {
fftwf_free(mapitem.second);
}
}
float* FFTBuffers::getR(int n)
{
if (m_rs.find(n) == m_rs.end()) {
m_rs[n] = (float *) fftwf_malloc(sizeof(float) * n);
}
return m_rs[n];
}
fftwf_complex *FFTBuffers::getC(int n)
{
if (m_cs.find(n) == m_cs.end()) {
m_cs[n] = (fftwf_complex *) fftwf_malloc(sizeof(fftwf_complex) * ((n / 2) + 1));
}
return m_cs[n];
}
fftwf_complex *FFTBuffers::getCCI(int n)
{
if (m_ccis.find(n) == m_ccis.end()) {
m_ccis[n] = (fftwf_complex *) fftwf_malloc(n * sizeof(fftwf_complex));
}
return m_ccis[n];
}
fftwf_complex *FFTBuffers::getCCO(int n)
{
if (m_ccos.find(n) == m_ccos.end()) {
m_ccos[n] = (fftwf_complex *) fftwf_malloc(n * sizeof(fftwf_complex));
}
return m_ccos[n];
}
} // nemespace FT8

52
ft8/fftbuffers.h Normal file
View File

@ -0,0 +1,52 @@
///////////////////////////////////////////////////////////////////////////////////
// Copyright (C) 2023 Edouard Griffiths, F4EXB. //
// //
// This is the code from ft8mon: https://github.com/rtmrtmrtmrtm/ft8mon //
// written by Robert Morris, AB1HL //
// reformatted and adapted to Qt and SDRangel context //
// //
// This program is free software; you can redistribute it and/or modify //
// it under the terms of the GNU General Public License as published by //
// the Free Software Foundation as version 3 of the License, or //
// (at your option) any later version. //
// //
// 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 V3 for more details. //
// //
// You should have received a copy of the GNU General Public License //
// along with this program. If not, see <http://www.gnu.org/licenses/>. //
///////////////////////////////////////////////////////////////////////////////////
#ifndef FFTBUFFERS_H
#define FFTBUFFERS_H
#include <map>
#include <fftw3.h>
#include "export.h"
namespace FT8
{
class FT8_API FFTBuffers
{
public:
~FFTBuffers();
float* getR(int n);
fftwf_complex *getC(int n);
fftwf_complex *getCCI(int n);
fftwf_complex *getCCO(int n);
private:
std::map<int, float*> m_rs; //!< R2C inputs or C2R inputs by size
std::map<int, fftwf_complex*> m_cs; //!< R2C outputs or C2R inputs by size
std::map<int, fftwf_complex*> m_ccis; //!< C2C inputs/outputs by size
std::map<int, fftwf_complex*> m_ccos; //!< C2C outputs/inputs by size
};
} // FFTBuffers
#endif // FFTBUFFERS_H

View File

@ -38,7 +38,7 @@ FT8Plans::~FT8Plans()
qDebug("FT8::FT8Plans::~FT8Plans: %lu plans to delete", m_plans.size());
for (auto& plan : m_plans) {
delete plan;
delete plan.second;
}
}
@ -55,19 +55,14 @@ Plan *FT8Plans::getPlan(int n)
{
QMutexLocker mlock(&m_globalPlanMutex);
for (auto& plan : m_plans)
{
if ((plan->n_ == n) && (plan->type_ == Plan::M_FFTW_TYPE)) {
return plan;
}
if (m_plans.find(n) != m_plans.end()) {
return m_plans[n];
}
fftwf_set_timelimit(5);
Plan *p = new Plan(n);
m_plans.push_back(p);
return p;
m_plans[n] = new Plan(n);
return m_plans[n];
}
}

View File

@ -21,7 +21,7 @@
#ifndef ft8plans_h
#define ft8plans_h
#include <vector>
#include <map>
#include <QMutex>
#include "export.h"
@ -45,7 +45,7 @@ protected:
static FT8Plans *m_instance;
private:
std::vector<Plan*> m_plans;
std::map<int, Plan*> m_plans;
static QMutex m_globalPlanMutex;
};