1
0
mirror of https://github.com/f4exb/sdrangel.git synced 2024-11-25 09:18:54 -05:00

Floating point to floating point decimator optimization using the even/odd algorithm

This commit is contained in:
f4exb 2018-04-30 11:08:08 +02:00
parent 48cc6df8a7
commit efa168ec77
4 changed files with 42 additions and 38 deletions

View File

@ -32,7 +32,7 @@ void DecimatorsFF::decimate1(FSampleVector::iterator* it, const float* buf, qint
void DecimatorsFF::decimate2_cen(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double intbuf[2];
float intbuf[2];
for (int pos = 0; pos < nbIAndQ - 3; pos += 4)
{
@ -54,7 +54,7 @@ void DecimatorsFF::decimate2_cen(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate2_inf(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal, yimag;
float xreal, yimag;
for (int pos = 0; pos < nbIAndQ - 7; pos += 8)
{
@ -74,7 +74,7 @@ void DecimatorsFF::decimate2_inf(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate2_sup(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal, yimag;
float xreal, yimag;
for (int pos = 0; pos < nbIAndQ - 7; pos += 8)
{
@ -94,7 +94,7 @@ void DecimatorsFF::decimate2_sup(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate4_inf(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal, yimag;
float xreal, yimag;
for (int pos = 0; pos < nbIAndQ - 7; pos += 8)
{
@ -116,7 +116,7 @@ void DecimatorsFF::decimate4_sup(FSampleVector::iterator* it, const float* buf,
// Inf (LSB):
// x y x y x y x y / x -> 0,-3,-4,7 / y -> 1,2,-5,-6
// [ rotate: 0, 1, -3, 2, -4, -5, 7, -6]
double xreal, yimag;
float xreal, yimag;
for (int pos = 0; pos < nbIAndQ - 7; pos += 8)
{
@ -132,7 +132,7 @@ void DecimatorsFF::decimate4_sup(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate8_inf(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal[2], yimag[2];
float xreal[2], yimag[2];
for (int pos = 0; pos < nbIAndQ - 15; pos += 8)
{
@ -154,7 +154,7 @@ void DecimatorsFF::decimate8_inf(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate8_sup(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal[2], yimag[2];
float xreal[2], yimag[2];
for (int pos = 0; pos < nbIAndQ - 15; pos += 8)
{
@ -178,7 +178,7 @@ void DecimatorsFF::decimate16_inf(FSampleVector::iterator* it, const float* buf,
{
// Offset tuning: 4x downsample and rotate, then
// downsample 4x more. [ rotate: 0, 1, -3, 2, -4, -5, 7, -6]
double xreal[4], yimag[4];
float xreal[4], yimag[4];
for (int pos = 0; pos < nbIAndQ - 31; )
{
@ -205,7 +205,7 @@ void DecimatorsFF::decimate16_sup(FSampleVector::iterator* it, const float* buf,
{
// Offset tuning: 4x downsample and rotate, then
// downsample 4x more. [ rotate: 1, 0, -2, 3, -5, -4, 6, -7]
double xreal[4], yimag[4];
float xreal[4], yimag[4];
for (int pos = 0; pos < nbIAndQ - 31; )
{
@ -230,7 +230,7 @@ void DecimatorsFF::decimate16_sup(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate32_inf(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal[8], yimag[8];
float xreal[8], yimag[8];
for (int pos = 0; pos < nbIAndQ - 63; )
{
@ -260,7 +260,7 @@ void DecimatorsFF::decimate32_inf(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate32_sup(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal[8], yimag[8];
float xreal[8], yimag[8];
for (int pos = 0; pos < nbIAndQ - 63; )
{
@ -290,7 +290,7 @@ void DecimatorsFF::decimate32_sup(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate64_inf(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal[16], yimag[16];
float xreal[16], yimag[16];
for (int pos = 0; pos < nbIAndQ - 127; )
{
@ -329,7 +329,7 @@ void DecimatorsFF::decimate64_inf(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate64_sup(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double xreal[16], yimag[16];
float xreal[16], yimag[16];
for (int pos = 0; pos < nbIAndQ - 127; )
{
@ -368,7 +368,7 @@ void DecimatorsFF::decimate64_sup(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate4_cen(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double intbuf[4];
float intbuf[4];
for (int pos = 0; pos < nbIAndQ - 7; pos += 8)
{
@ -402,7 +402,7 @@ void DecimatorsFF::decimate4_cen(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate8_cen(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double intbuf[8];
float intbuf[8];
for (int pos = 0; pos < nbIAndQ - 15; pos += 16)
{
@ -461,7 +461,7 @@ void DecimatorsFF::decimate8_cen(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate16_cen(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double intbuf[16];
float intbuf[16];
for (int pos = 0; pos < nbIAndQ - 31; pos += 32)
{
@ -569,7 +569,7 @@ void DecimatorsFF::decimate16_cen(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate32_cen(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double intbuf[32];
float intbuf[32];
for (int pos = 0; pos < nbIAndQ - 63; pos += 64)
{
@ -774,7 +774,7 @@ void DecimatorsFF::decimate32_cen(FSampleVector::iterator* it, const float* buf,
void DecimatorsFF::decimate64_cen(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ)
{
double intbuf[64];
float intbuf[64];
for (int pos = 0; pos < nbIAndQ - 127; pos += 128)
{

View File

@ -17,7 +17,7 @@
#ifndef SDRBASE_DSP_DECIMATORSFF_H_
#define SDRBASE_DSP_DECIMATORSFF_H_
#include "dsp/inthalfbandfilterdbf.h"
#include "dsp/inthalfbandfiltereof.h"
#include "export.h"
#define DECIMATORSFF_HB_FILTER_ORDER 64
@ -46,12 +46,12 @@ public:
void decimate64_sup(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ);
void decimate64_cen(FSampleVector::iterator* it, const float* buf, qint32 nbIAndQ);
IntHalfbandFilterDBF<double, float, DECIMATORSFF_HB_FILTER_ORDER> m_decimator2; // 1st stages
IntHalfbandFilterDBF<double, float, DECIMATORSFF_HB_FILTER_ORDER> m_decimator4; // 2nd stages
IntHalfbandFilterDBF<double, float, DECIMATORSFF_HB_FILTER_ORDER> m_decimator8; // 3rd stages
IntHalfbandFilterDBF<double, float, DECIMATORSFF_HB_FILTER_ORDER> m_decimator16; // 4th stages
IntHalfbandFilterDBF<double, float, DECIMATORSFF_HB_FILTER_ORDER> m_decimator32; // 5th stages
IntHalfbandFilterDBF<double, float, DECIMATORSFF_HB_FILTER_ORDER> m_decimator64; // 6th stages
IntHalfbandFilterEOF<DECIMATORSFF_HB_FILTER_ORDER> m_decimator2; // 1st stages
IntHalfbandFilterEOF<DECIMATORSFF_HB_FILTER_ORDER> m_decimator4; // 2nd stages
IntHalfbandFilterEOF<DECIMATORSFF_HB_FILTER_ORDER> m_decimator8; // 3rd stages
IntHalfbandFilterEOF<DECIMATORSFF_HB_FILTER_ORDER> m_decimator16; // 4th stages
IntHalfbandFilterEOF<DECIMATORSFF_HB_FILTER_ORDER> m_decimator32; // 5th stages
IntHalfbandFilterEOF<DECIMATORSFF_HB_FILTER_ORDER> m_decimator64; // 6th stages
};

View File

@ -27,7 +27,8 @@ MainBench::MainBench(qtwebapp::LoggerWithFile *logger, const ParserBench& parser
QObject(parent),
m_logger(logger),
m_parser(parser),
m_uniform_distribution(-1.0, 1.0)
m_uniform_distribution_f(-1.0, 1.0),
m_uniform_distribution_s16(-32768,32767)
{
qDebug() << "MainBench::MainBench: start";
m_instance = this;
@ -62,22 +63,24 @@ void MainBench::run()
void MainBench::testDecimateII()
{
QElapsedTimer timer;
qint64 nsecs;
qint64 nsecs = 0;
qDebug() << "MainBench::testDecimateII: create test data";
qint16 *buf = new qint16[m_parser.getNbSamples()*2];
m_convertBuffer.resize(m_parser.getNbSamples()/(1<<m_parser.getLog2Factor()));
auto my_rand = std::bind(m_uniform_distribution_s16, m_generator);
std::generate(buf, buf + m_parser.getNbSamples()*2 - 1, my_rand);
qDebug() << "MainBench::testDecimateII: run test";
timer.start();
for (uint32_t i = 0; i < m_parser.getRepetition(); i++)
{
timer.start();
decimateII(buf, m_parser.getNbSamples()*2);
nsecs += timer.nsecsElapsed();
}
nsecs = timer.nsecsElapsed();
printResults("MainBench::testDecimateII", nsecs);
qDebug() << "MainBench::testDecimateII: cleanup test data";
@ -87,24 +90,24 @@ void MainBench::testDecimateII()
void MainBench::testDecimateFI()
{
QElapsedTimer timer;
qint64 nsecs;
qint64 nsecs = 0;
qDebug() << "MainBench::testDecimateFI: create test data";
float *buf = new float[m_parser.getNbSamples()*2];
m_convertBuffer.resize(m_parser.getNbSamples()/(1<<m_parser.getLog2Factor()));
auto my_rand = std::bind(m_uniform_distribution, m_generator);
auto my_rand = std::bind(m_uniform_distribution_f, m_generator);
std::generate(buf, buf + m_parser.getNbSamples()*2 - 1, my_rand); // make sure data is in [-1.0..1.0] range
qDebug() << "MainBench::testDecimateFI: run test";
timer.start();
for (uint32_t i = 0; i < m_parser.getRepetition(); i++)
{
timer.start();
decimateFI(buf, m_parser.getNbSamples()*2);
nsecs += timer.nsecsElapsed();
}
nsecs = timer.nsecsElapsed();
printResults("MainBench::testDecimateFI", nsecs);
qDebug() << "MainBench::testDecimateFI: cleanup test data";
@ -114,24 +117,24 @@ void MainBench::testDecimateFI()
void MainBench::testDecimateFF()
{
QElapsedTimer timer;
qint64 nsecs;
qint64 nsecs = 0;
qDebug() << "MainBench::testDecimateFF: create test data";
float *buf = new float[m_parser.getNbSamples()*2];
m_convertBufferF.resize(m_parser.getNbSamples()/(1<<m_parser.getLog2Factor()));
auto my_rand = std::bind(m_uniform_distribution, m_generator);
auto my_rand = std::bind(m_uniform_distribution_f, m_generator);
std::generate(buf, buf + m_parser.getNbSamples()*2 - 1, my_rand); // make sure data is in [-1.0..1.0] range
qDebug() << "MainBench::testDecimateFF: run test";
timer.start();
for (uint32_t i = 0; i < m_parser.getRepetition(); i++)
{
timer.start();
decimateFF(buf, m_parser.getNbSamples()*2);
nsecs += timer.nsecsElapsed();
}
nsecs = timer.nsecsElapsed();
printResults("MainBench::testDecimateFF", nsecs);
qDebug() << "MainBench::testDecimateFF: cleanup test data";

View File

@ -58,7 +58,8 @@ private:
qtwebapp::LoggerWithFile *m_logger;
const ParserBench& m_parser;
std::mt19937 m_generator;
std::uniform_real_distribution<float> m_uniform_distribution;
std::uniform_real_distribution<float> m_uniform_distribution_f;
std::uniform_int_distribution<qint16> m_uniform_distribution_s16;
#ifdef SDR_RX_SAMPLE_24BIT
Decimators<qint32, qint16, SDR_RX_SAMP_SZ, 12> m_decimatorsII;