mirror of
https://github.com/f4exb/sdrangel.git
synced 2024-11-08 17:46:03 -05:00
2769 lines
107 KiB
C++
2769 lines
107 KiB
C++
///////////////////////////////////////////////////////////////////////////////////
|
|
// Copyright (C) 2015 Edouard Griffiths, F4EXB //
|
|
// //
|
|
// 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 //
|
|
// //
|
|
// 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 INCLUDE_GPL_DSP_DECIMATORS_H_
|
|
#define INCLUDE_GPL_DSP_DECIMATORS_H_
|
|
|
|
#include "dsp/dsptypes.h"
|
|
#ifdef USE_SSE4_1
|
|
#include "dsp/inthalfbandfiltereo1.h"
|
|
#else
|
|
#include "dsp/inthalfbandfilterdb.h"
|
|
#endif
|
|
|
|
#define DECIMATORS_HB_FILTER_ORDER 64
|
|
|
|
template<uint SdrBits, uint InputBits>
|
|
struct decimation_shifts
|
|
{
|
|
static const uint pre1 = 0;
|
|
static const uint pre2 = 0;
|
|
static const uint post2 = 0;
|
|
static const uint pre4 = 0;
|
|
static const uint post4 = 0;
|
|
static const uint pre8 = 0;
|
|
static const uint post8 = 0;
|
|
static const uint pre16 = 0;
|
|
static const uint post16 = 0;
|
|
static const uint pre32 = 0;
|
|
static const uint post32 = 0;
|
|
static const uint pre64 = 0;
|
|
static const uint post64 = 0;
|
|
};
|
|
|
|
template<>
|
|
struct decimation_shifts<16, 16>
|
|
{
|
|
static const uint pre1 = 0;
|
|
static const uint pre2 = 0;
|
|
static const uint post2 = 1;
|
|
static const uint pre4 = 0;
|
|
static const uint post4 = 2;
|
|
static const uint pre8 = 0;
|
|
static const uint post8 = 3;
|
|
static const uint pre16 = 0;
|
|
static const uint post16 = 4;
|
|
static const uint pre32 = 0;
|
|
static const uint post32 = 5;
|
|
static const uint pre64 = 0;
|
|
static const uint post64 = 6;
|
|
};
|
|
|
|
template<>
|
|
struct decimation_shifts<16, 12>
|
|
{
|
|
static const uint pre1 = 4;
|
|
static const uint pre2 = 3;
|
|
static const uint post2 = 0;
|
|
static const uint pre4 = 2;
|
|
static const uint post4 = 0;
|
|
static const uint pre8 = 1;
|
|
static const uint post8 = 0;
|
|
static const uint pre16 = 0;
|
|
static const uint post16 = 0;
|
|
static const uint pre32 = 0;
|
|
static const uint post32 = 1;
|
|
static const uint pre64 = 0;
|
|
static const uint post64 = 2;
|
|
};
|
|
|
|
template<>
|
|
struct decimation_shifts<16, 8>
|
|
{
|
|
static const uint pre1 = 6;
|
|
static const uint pre2 = 5;
|
|
static const uint post2 = 0;
|
|
static const uint pre4 = 4;
|
|
static const uint post4 = 0;
|
|
static const uint pre8 = 3;
|
|
static const uint post8 = 0;
|
|
static const uint pre16 = 2;
|
|
static const uint post16 = 0;
|
|
static const uint pre32 = 1;
|
|
static const uint post32 = 0;
|
|
static const uint pre64 = 0;
|
|
static const uint post64 = 0;
|
|
};
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
class Decimators
|
|
{
|
|
public:
|
|
// interleaved I/Q input buffer
|
|
void decimate1(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate2_u(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate2_inf(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate2_sup(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate2_cen(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate4_inf(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate4_sup(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate4_cen(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate8_inf(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate8_sup(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate8_cen(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate16_inf(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate16_sup(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate16_cen(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate32_inf(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate32_sup(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate32_cen(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate64_inf(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate64_sup(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
void decimate64_cen(SampleVector::iterator* it, const T* buf, qint32 len);
|
|
// separate I and Q input buffers
|
|
void decimate1(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate2_u(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate2_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate2_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate2_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate4_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate4_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate4_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate8_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate8_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate8_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate16_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate16_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate16_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate32_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate32_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate32_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate64_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate64_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
void decimate64_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len);
|
|
|
|
private:
|
|
#ifdef USE_SSE4_1
|
|
IntHalfbandFilterEO1<DECIMATORS_HB_FILTER_ORDER> m_decimator2; // 1st stages
|
|
IntHalfbandFilterEO1<DECIMATORS_HB_FILTER_ORDER> m_decimator4; // 2nd stages
|
|
IntHalfbandFilterEO1<DECIMATORS_HB_FILTER_ORDER> m_decimator8; // 3rd stages
|
|
IntHalfbandFilterEO1<DECIMATORS_HB_FILTER_ORDER> m_decimator16; // 4th stages
|
|
IntHalfbandFilterEO1<DECIMATORS_HB_FILTER_ORDER> m_decimator32; // 5th stages
|
|
IntHalfbandFilterEO1<DECIMATORS_HB_FILTER_ORDER> m_decimator64; // 6th stages
|
|
#else
|
|
IntHalfbandFilterDB<DECIMATORS_HB_FILTER_ORDER> m_decimator2; // 1st stages
|
|
IntHalfbandFilterDB<DECIMATORS_HB_FILTER_ORDER> m_decimator4; // 2nd stages
|
|
IntHalfbandFilterDB<DECIMATORS_HB_FILTER_ORDER> m_decimator8; // 3rd stages
|
|
IntHalfbandFilterDB<DECIMATORS_HB_FILTER_ORDER> m_decimator16; // 4th stages
|
|
IntHalfbandFilterDB<DECIMATORS_HB_FILTER_ORDER> m_decimator32; // 5th stages
|
|
IntHalfbandFilterDB<DECIMATORS_HB_FILTER_ORDER> m_decimator64; // 6th stages
|
|
#endif
|
|
};
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate1(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 1; pos += 2)
|
|
{
|
|
xreal = buf[pos+0];
|
|
yimag = buf[pos+1];
|
|
(**it).setReal(xreal << decimation_shifts<SdrBits, InputBits>::pre1); // Valgrind optim (2 - comment not repeated)
|
|
(**it).setImag(yimag << decimation_shifts<SdrBits, InputBits>::pre1);
|
|
++(*it); // Valgrind optim (comment not repeated)
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate1(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len; pos += 1)
|
|
{
|
|
xreal = bufI[pos];
|
|
yimag = bufQ[pos];
|
|
(**it).setReal(xreal << decimation_shifts<SdrBits, InputBits>::pre1); // Valgrind optim (2 - comment not repeated)
|
|
(**it).setImag(yimag << decimation_shifts<SdrBits, InputBits>::pre1);
|
|
++(*it); // Valgrind optim (comment not repeated)
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_u(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 8)
|
|
{
|
|
xreal = (buf[pos+0] - buf[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (buf[pos+1] + buf[pos+2] - 255) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
|
|
xreal = (buf[pos+7] - buf[pos+4]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (255 - buf[pos+5] - buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_u(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 3; pos += 4)
|
|
{
|
|
// 0: I[0] 1: Q[0] 2: I[1] 3: Q[1]
|
|
xreal = (bufI[pos] - bufQ[pos+1]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (bufQ[pos] + bufI[pos+1] - 255) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
|
|
// 4: I[2] 5: Q[2] 6: I[3] 7: Q[3]
|
|
xreal = (bufQ[pos+3] - bufI[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (255 - bufQ[pos+2] - bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_inf(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 8)
|
|
{
|
|
xreal = (buf[pos+0] - buf[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (buf[pos+1] + buf[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
|
|
xreal = (buf[pos+7] - buf[pos+4]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (- buf[pos+5] - buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 3; pos += 4)
|
|
{
|
|
// 0: I[0] 1: Q[0] 2: I[1] 3: Q[1]
|
|
xreal = (bufI[pos] - bufQ[pos+1]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (bufQ[pos] + bufI[pos+1]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
|
|
// 4: I[2] 5: Q[2] 6: I[3] 7: Q[3]
|
|
xreal = (bufQ[pos+3] - bufI[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (- bufQ[pos+2] - bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_sup(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 8)
|
|
{
|
|
xreal = (buf[pos+1] - buf[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (- buf[pos+0] - buf[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
|
|
xreal = (buf[pos+6] - buf[pos+5]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (buf[pos+4] + buf[pos+7]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 3; pos += 4)
|
|
{
|
|
// 0: I[0] 1: Q[0] 2: I[1] 3: Q[1]
|
|
xreal = (bufQ[pos] - bufI[pos+1]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (- bufI[pos] - bufQ[pos+1]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
|
|
// 4: I[2] 5: Q[2] 6: I[3] 7: Q[3]
|
|
xreal = (bufI[pos+3] - bufQ[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
yimag = (bufI[pos+2] + bufQ[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate4_inf(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 8)
|
|
{
|
|
xreal = (buf[pos+0] - buf[pos+3] + buf[pos+7] - buf[pos+4]) << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
yimag = (buf[pos+1] - buf[pos+5] + buf[pos+2] - buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate4_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 3; pos += 4)
|
|
{
|
|
xreal = (bufI[pos] - bufQ[pos+1] + bufQ[pos+3] - bufI[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
yimag = (bufQ[pos] - bufQ[pos+2] + bufI[pos+1] - bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate4_sup(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
// Sup (USB):
|
|
// x y x y x y x y / x -> 1,-2,-5,6 / y -> -0,-3,4,7
|
|
// [ rotate: 1, 0, -2, 3, -5, -4, 6, -7]
|
|
// 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]
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 8)
|
|
{
|
|
xreal = (buf[pos+1] - buf[pos+2] - buf[pos+5] + buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
yimag = (- buf[pos+0] - buf[pos+3] + buf[pos+4] + buf[pos+7]) << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate4_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal, yimag;
|
|
|
|
for (int pos = 0; pos < len - 3; pos += 4)
|
|
{
|
|
xreal = (bufQ[pos] - bufI[pos+1] - bufQ[pos+2] + bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
yimag = (- bufI[pos] - bufQ[pos+1] + bufI[pos+2] + bufQ[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
|
|
(**it).setReal(xreal >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
(**it).setImag(yimag >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate8_inf(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal[2], yimag[2];
|
|
|
|
for (int pos = 0; pos < len - 15; pos += 8)
|
|
{
|
|
xreal[0] = (buf[pos+0] - buf[pos+3] + buf[pos+7] - buf[pos+4]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
yimag[0] = (buf[pos+1] - buf[pos+5] + buf[pos+2] - buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
pos += 8;
|
|
|
|
xreal[1] = (buf[pos+0] - buf[pos+3] + buf[pos+7] - buf[pos+4]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
yimag[1] = (buf[pos+1] - buf[pos+5] + buf[pos+2] - buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
|
|
(**it).setReal(xreal[1] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
(**it).setImag(yimag[1] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate8_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal[2], yimag[2];
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 4)
|
|
{
|
|
xreal[0] = (bufI[pos] - bufQ[pos+1] + bufQ[pos+3] - bufI[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
yimag[0] = (bufQ[pos] - bufQ[pos+2] + bufI[pos+1] - bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
pos += 4;
|
|
|
|
xreal[1] = (bufI[pos] - bufQ[pos+1] + bufQ[pos+3] - bufI[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
yimag[1] = (bufQ[pos] - bufQ[pos+2] + bufI[pos+1] - bufQ[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
|
|
(**it).setReal(xreal[1] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
(**it).setImag(yimag[1] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate8_sup(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal[2], yimag[2];
|
|
|
|
for (int pos = 0; pos < len - 15; pos += 8)
|
|
{
|
|
xreal[0] = (buf[pos+1] - buf[pos+2] - buf[pos+5] + buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
yimag[0] = (- buf[pos+0] - buf[pos+3] + buf[pos+4] + buf[pos+7]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
pos += 8;
|
|
|
|
xreal[1] = (buf[pos+1] - buf[pos+2] - buf[pos+5] + buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
yimag[1] = (- buf[pos+0] - buf[pos+3] + buf[pos+4] + buf[pos+7]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
|
|
(**it).setReal(xreal[1] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
(**it).setImag(yimag[1] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate8_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal[2], yimag[2];
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 4)
|
|
{
|
|
xreal[0] = (bufQ[pos] - bufI[pos+1] - bufQ[pos+2] + bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
yimag[0] = (- bufI[pos] - bufQ[pos+1] + bufI[pos+2] + bufQ[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
pos += 4;
|
|
|
|
xreal[1] = (bufQ[pos] - bufI[pos+1] - bufQ[pos+2] + bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
yimag[1] = (- bufI[pos] - bufQ[pos+1] + bufI[pos+2] + bufQ[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
|
|
(**it).setReal(xreal[1] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
(**it).setImag(yimag[1] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate16_inf(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
// Offset tuning: 4x downsample and rotate, then
|
|
// downsample 4x more. [ rotate: 0, 1, -3, 2, -4, -5, 7, -6]
|
|
qint32 xreal[4], yimag[4];
|
|
|
|
for (int pos = 0; pos < len - 31; )
|
|
{
|
|
for (int i = 0; i < 4; i++)
|
|
{
|
|
xreal[i] = (buf[pos+0] - buf[pos+3] + buf[pos+7] - buf[pos+4]) << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
yimag[i] = (buf[pos+1] - buf[pos+5] + buf[pos+2] - buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
pos += 8;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
|
|
(**it).setReal(xreal[3] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
(**it).setImag(yimag[3] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate16_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
// Offset tuning: 4x downsample and rotate, then
|
|
// downsample 4x more. [ rotate: 0, 1, -3, 2, -4, -5, 7, -6]
|
|
qint32 xreal[4], yimag[4];
|
|
|
|
for (int pos = 0; pos < len - 15; )
|
|
{
|
|
for (int i = 0; i < 4; i++)
|
|
{
|
|
xreal[i] = (bufI[pos] - bufI[pos+1] + bufQ[pos+3] - bufI[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
yimag[i] = (bufQ[pos] - bufQ[pos+2] + bufI[pos+1] - bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
pos += 4;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
|
|
(**it).setReal(xreal[3] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
(**it).setImag(yimag[3] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate16_sup(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
// Offset tuning: 4x downsample and rotate, then
|
|
// downsample 4x more. [ rotate: 1, 0, -2, 3, -5, -4, 6, -7]
|
|
qint32 xreal[4], yimag[4];
|
|
|
|
for (int pos = 0; pos < len - 31; )
|
|
{
|
|
for (int i = 0; i < 4; i++)
|
|
{
|
|
xreal[i] = (buf[pos+1] - buf[pos+2] - buf[pos+5] + buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
yimag[i] = (buf[pos+4] + buf[pos+7] - buf[pos+0] - buf[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
pos += 8;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
|
|
(**it).setReal(xreal[3] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
(**it).setImag(yimag[3] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate16_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
// Offset tuning: 4x downsample and rotate, then
|
|
// downsample 4x more. [ rotate: 1, 0, -2, 3, -5, -4, 6, -7]
|
|
qint32 xreal[4], yimag[4];
|
|
|
|
for (int pos = 0; pos < len - 15; )
|
|
{
|
|
for (int i = 0; i < 4; i++)
|
|
{
|
|
xreal[i] = (bufQ[pos+0] - bufI[pos+1] - bufQ[pos+2] + bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
yimag[i] = (bufI[pos+2] + bufQ[pos+3] - bufI[pos+0] - bufQ[pos+1]) << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
pos += 4;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
|
|
(**it).setReal(xreal[3] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
(**it).setImag(yimag[3] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate32_inf(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal[8], yimag[8];
|
|
|
|
for (int pos = 0; pos < len - 63; )
|
|
{
|
|
for (int i = 0; i < 8; i++)
|
|
{
|
|
xreal[i] = (buf[pos+0] - buf[pos+3] + buf[pos+7] - buf[pos+4]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
yimag[i] = (buf[pos+1] - buf[pos+5] + buf[pos+2] - buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
pos += 8;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
m_decimator2.myDecimate(xreal[4], yimag[4], &xreal[5], &yimag[5]);
|
|
m_decimator2.myDecimate(xreal[6], yimag[6], &xreal[7], &yimag[7]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
m_decimator4.myDecimate(xreal[5], yimag[5], &xreal[7], &yimag[7]);
|
|
|
|
m_decimator8.myDecimate(xreal[3], yimag[3], &xreal[7], &yimag[7]);
|
|
|
|
(**it).setReal(xreal[7] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
(**it).setImag(yimag[7] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate32_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal[8], yimag[8];
|
|
|
|
for (int pos = 0; pos < len - 31; )
|
|
{
|
|
for (int i = 0; i < 8; i++)
|
|
{
|
|
xreal[i] = (bufI[pos+0] - bufQ[pos+1] + bufQ[pos+3] - bufI[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
yimag[i] = (bufQ[pos+0] - bufQ[pos+2] + bufI[pos+1] - bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
pos += 4;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
m_decimator2.myDecimate(xreal[4], yimag[4], &xreal[5], &yimag[5]);
|
|
m_decimator2.myDecimate(xreal[6], yimag[6], &xreal[7], &yimag[7]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
m_decimator4.myDecimate(xreal[5], yimag[5], &xreal[7], &yimag[7]);
|
|
|
|
m_decimator8.myDecimate(xreal[3], yimag[3], &xreal[7], &yimag[7]);
|
|
|
|
(**it).setReal(xreal[7] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
(**it).setImag(yimag[7] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate32_sup(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal[8], yimag[8];
|
|
|
|
for (int pos = 0; pos < len - 63; )
|
|
{
|
|
for (int i = 0; i < 8; i++)
|
|
{
|
|
xreal[i] = (buf[pos+1] - buf[pos+2] - buf[pos+5] + buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
yimag[i] = (buf[pos+4] + buf[pos+7] - buf[pos+0] - buf[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
pos += 8;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
m_decimator2.myDecimate(xreal[4], yimag[4], &xreal[5], &yimag[5]);
|
|
m_decimator2.myDecimate(xreal[6], yimag[6], &xreal[7], &yimag[7]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
m_decimator4.myDecimate(xreal[5], yimag[5], &xreal[7], &yimag[7]);
|
|
|
|
m_decimator8.myDecimate(xreal[3], yimag[3], &xreal[7], &yimag[7]);
|
|
|
|
(**it).setReal(xreal[7] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
(**it).setImag(yimag[7] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate32_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal[8], yimag[8];
|
|
|
|
for (int pos = 0; pos < len - 31; )
|
|
{
|
|
for (int i = 0; i < 8; i++)
|
|
{
|
|
xreal[i] = (bufQ[pos+0] - bufI[pos+1] - bufQ[pos+2] + bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
yimag[i] = (bufI[pos+2] + bufQ[pos+3] - bufI[pos+0] - bufQ[pos+1]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
pos += 4;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
m_decimator2.myDecimate(xreal[4], yimag[4], &xreal[5], &yimag[5]);
|
|
m_decimator2.myDecimate(xreal[6], yimag[6], &xreal[7], &yimag[7]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
m_decimator4.myDecimate(xreal[5], yimag[5], &xreal[7], &yimag[7]);
|
|
|
|
m_decimator8.myDecimate(xreal[3], yimag[3], &xreal[7], &yimag[7]);
|
|
|
|
(**it).setReal(xreal[7] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
(**it).setImag(yimag[7] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate64_inf(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal[16], yimag[16];
|
|
|
|
for (int pos = 0; pos < len - 127; )
|
|
{
|
|
for (int i = 0; i < 16; i++)
|
|
{
|
|
xreal[i] = (buf[pos+0] - buf[pos+3] + buf[pos+7] - buf[pos+4]) << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
yimag[i] = (buf[pos+1] - buf[pos+5] + buf[pos+2] - buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
pos += 8;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
m_decimator2.myDecimate(xreal[4], yimag[4], &xreal[5], &yimag[5]);
|
|
m_decimator2.myDecimate(xreal[6], yimag[6], &xreal[7], &yimag[7]);
|
|
m_decimator2.myDecimate(xreal[8], yimag[8], &xreal[9], &yimag[9]);
|
|
m_decimator2.myDecimate(xreal[10], yimag[10], &xreal[11], &yimag[11]);
|
|
m_decimator2.myDecimate(xreal[12], yimag[12], &xreal[13], &yimag[13]);
|
|
m_decimator2.myDecimate(xreal[14], yimag[14], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
m_decimator4.myDecimate(xreal[5], yimag[5], &xreal[7], &yimag[7]);
|
|
m_decimator4.myDecimate(xreal[9], yimag[9], &xreal[11], &yimag[11]);
|
|
m_decimator4.myDecimate(xreal[13], yimag[13], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator8.myDecimate(xreal[3], yimag[3], &xreal[7], &yimag[7]);
|
|
m_decimator8.myDecimate(xreal[11], yimag[11], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator16.myDecimate(xreal[7], yimag[7], &xreal[15], &yimag[15]);
|
|
|
|
(**it).setReal(xreal[15] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
(**it).setImag(yimag[15] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate64_inf(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal[16], yimag[16];
|
|
|
|
for (int pos = 0; pos < len - 63; )
|
|
{
|
|
for (int i = 0; i < 16; i++)
|
|
{
|
|
xreal[i] = (bufI[pos+0] - bufQ[pos+1] + bufQ[pos+3] - bufI[pos+2]) << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
yimag[i] = (bufQ[pos+0] - bufQ[pos+2] + bufI[pos+1] - bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
pos += 4;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
m_decimator2.myDecimate(xreal[4], yimag[4], &xreal[5], &yimag[5]);
|
|
m_decimator2.myDecimate(xreal[6], yimag[6], &xreal[7], &yimag[7]);
|
|
m_decimator2.myDecimate(xreal[8], yimag[8], &xreal[9], &yimag[9]);
|
|
m_decimator2.myDecimate(xreal[10], yimag[10], &xreal[11], &yimag[11]);
|
|
m_decimator2.myDecimate(xreal[12], yimag[12], &xreal[13], &yimag[13]);
|
|
m_decimator2.myDecimate(xreal[14], yimag[14], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
m_decimator4.myDecimate(xreal[5], yimag[5], &xreal[7], &yimag[7]);
|
|
m_decimator4.myDecimate(xreal[9], yimag[9], &xreal[11], &yimag[11]);
|
|
m_decimator4.myDecimate(xreal[13], yimag[13], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator8.myDecimate(xreal[3], yimag[3], &xreal[7], &yimag[7]);
|
|
m_decimator8.myDecimate(xreal[11], yimag[11], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator16.myDecimate(xreal[7], yimag[7], &xreal[15], &yimag[15]);
|
|
|
|
(**it).setReal(xreal[15] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
(**it).setImag(yimag[15] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate64_sup(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 xreal[16], yimag[16];
|
|
|
|
for (int pos = 0; pos < len - 127; )
|
|
{
|
|
for (int i = 0; i < 16; i++)
|
|
{
|
|
xreal[i] = (buf[pos+1] - buf[pos+2] - buf[pos+5] + buf[pos+6]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
yimag[i] = (buf[pos+4] + buf[pos+7] - buf[pos+0] - buf[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
pos += 8;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
m_decimator2.myDecimate(xreal[4], yimag[4], &xreal[5], &yimag[5]);
|
|
m_decimator2.myDecimate(xreal[6], yimag[6], &xreal[7], &yimag[7]);
|
|
m_decimator2.myDecimate(xreal[8], yimag[8], &xreal[9], &yimag[9]);
|
|
m_decimator2.myDecimate(xreal[10], yimag[10], &xreal[11], &yimag[11]);
|
|
m_decimator2.myDecimate(xreal[12], yimag[12], &xreal[13], &yimag[13]);
|
|
m_decimator2.myDecimate(xreal[14], yimag[14], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
m_decimator4.myDecimate(xreal[5], yimag[5], &xreal[7], &yimag[7]);
|
|
m_decimator4.myDecimate(xreal[9], yimag[9], &xreal[11], &yimag[11]);
|
|
m_decimator4.myDecimate(xreal[13], yimag[13], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator8.myDecimate(xreal[3], yimag[3], &xreal[7], &yimag[7]);
|
|
m_decimator8.myDecimate(xreal[11], yimag[11], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator16.myDecimate(xreal[7], yimag[7], &xreal[15], &yimag[15]);
|
|
|
|
(**it).setReal(xreal[15] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
(**it).setImag(yimag[15] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate64_sup(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 xreal[16], yimag[16];
|
|
|
|
for (int pos = 0; pos < len - 63; )
|
|
{
|
|
for (int i = 0; i < 16; i++)
|
|
{
|
|
xreal[i] = (bufQ[pos+0] - bufI[pos+1] - bufQ[pos+2] + bufI[pos+3]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
yimag[i] = (bufI[pos+2] + bufQ[pos+3] - bufI[pos+0] - bufQ[pos+1]) << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
pos += 4;
|
|
}
|
|
|
|
m_decimator2.myDecimate(xreal[0], yimag[0], &xreal[1], &yimag[1]);
|
|
m_decimator2.myDecimate(xreal[2], yimag[2], &xreal[3], &yimag[3]);
|
|
m_decimator2.myDecimate(xreal[4], yimag[4], &xreal[5], &yimag[5]);
|
|
m_decimator2.myDecimate(xreal[6], yimag[6], &xreal[7], &yimag[7]);
|
|
m_decimator2.myDecimate(xreal[8], yimag[8], &xreal[9], &yimag[9]);
|
|
m_decimator2.myDecimate(xreal[10], yimag[10], &xreal[11], &yimag[11]);
|
|
m_decimator2.myDecimate(xreal[12], yimag[12], &xreal[13], &yimag[13]);
|
|
m_decimator2.myDecimate(xreal[14], yimag[14], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator4.myDecimate(xreal[1], yimag[1], &xreal[3], &yimag[3]);
|
|
m_decimator4.myDecimate(xreal[5], yimag[5], &xreal[7], &yimag[7]);
|
|
m_decimator4.myDecimate(xreal[9], yimag[9], &xreal[11], &yimag[11]);
|
|
m_decimator4.myDecimate(xreal[13], yimag[13], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator8.myDecimate(xreal[3], yimag[3], &xreal[7], &yimag[7]);
|
|
m_decimator8.myDecimate(xreal[11], yimag[11], &xreal[15], &yimag[15]);
|
|
|
|
m_decimator16.myDecimate(xreal[7], yimag[7], &xreal[15], &yimag[15]);
|
|
|
|
(**it).setReal(xreal[15] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
(**it).setImag(yimag[15] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 intbuf[2];
|
|
|
|
for (int pos = 0; pos < len - 3; pos += 4)
|
|
{
|
|
intbuf[0] = buf[pos+2] << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
intbuf[1] = buf[pos+3] << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
|
|
m_decimator2.myDecimate(
|
|
buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre2,
|
|
buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre2,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
|
|
(**it).setReal(intbuf[0] >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(intbuf[1] >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 intbuf[2];
|
|
|
|
for (int pos = 0; pos < len - 1; pos += 2)
|
|
{
|
|
intbuf[0] = bufI[pos+1] << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
intbuf[1] = bufQ[pos+1] << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+0] << decimation_shifts<SdrBits, InputBits>::pre2,
|
|
bufQ[pos+0] << decimation_shifts<SdrBits, InputBits>::pre2,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
|
|
(**it).setReal(intbuf[0] >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(intbuf[1] >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate4_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 intbuf[4];
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 8)
|
|
{
|
|
intbuf[0] = buf[pos+2] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
intbuf[1] = buf[pos+3] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
intbuf[2] = buf[pos+6] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
intbuf[3] = buf[pos+7] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
|
|
m_decimator2.myDecimate(
|
|
buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre4,
|
|
buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre4,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+4] << decimation_shifts<SdrBits, InputBits>::pre4,
|
|
buf[pos+5] << decimation_shifts<SdrBits, InputBits>::pre4,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
|
|
(**it).setReal(intbuf[2] >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
(**it).setImag(intbuf[3] >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate4_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 intbuf[4];
|
|
|
|
for (int pos = 0; pos < len - 3; pos += 4)
|
|
{
|
|
intbuf[0] = bufI[pos+1] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
intbuf[1] = bufQ[pos+1] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
intbuf[2] = bufI[pos+3] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
intbuf[3] = bufQ[pos+3] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+0] << decimation_shifts<SdrBits, InputBits>::pre4,
|
|
bufQ[pos+0] << decimation_shifts<SdrBits, InputBits>::pre4,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+2] << decimation_shifts<SdrBits, InputBits>::pre4,
|
|
bufQ[pos+2] << decimation_shifts<SdrBits, InputBits>::pre4,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
|
|
(**it).setReal(intbuf[2] >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
(**it).setImag(intbuf[3] >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate8_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 intbuf[8];
|
|
|
|
for (int pos = 0; pos < len - 15; pos += 16)
|
|
{
|
|
intbuf[0] = buf[pos+2] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[1] = buf[pos+3] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[2] = buf[pos+6] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[3] = buf[pos+7] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[4] = buf[pos+10] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[5] = buf[pos+11] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[6] = buf[pos+14] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[7] = buf[pos+15] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
|
|
m_decimator2.myDecimate(
|
|
buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+4] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
buf[pos+5] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+8] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
buf[pos+9] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
&intbuf[4],
|
|
&intbuf[5]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+12] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
buf[pos+13] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[4],
|
|
intbuf[5],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
|
|
m_decimator8.myDecimate(
|
|
intbuf[2],
|
|
intbuf[3],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
|
|
(**it).setReal(intbuf[6] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
(**it).setImag(intbuf[7] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate8_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 intbuf[8];
|
|
|
|
for (int pos = 0; pos < len - 7; pos += 8)
|
|
{
|
|
intbuf[0] = bufI[pos+1] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[1] = bufQ[pos+1] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[2] = bufI[pos+3] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[3] = bufQ[pos+3] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[4] = bufI[pos+5] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[5] = bufQ[pos+5] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[6] = bufI[pos+7] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
intbuf[7] = bufQ[pos+7] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+0] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
bufQ[pos+0] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+2] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
bufQ[pos+2] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+4] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
bufQ[pos+4] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
&intbuf[4],
|
|
&intbuf[5]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+6] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
bufQ[pos+6] << decimation_shifts<SdrBits, InputBits>::pre8,
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[4],
|
|
intbuf[5],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
|
|
m_decimator8.myDecimate(
|
|
intbuf[2],
|
|
intbuf[3],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
|
|
(**it).setReal(intbuf[6] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
(**it).setImag(intbuf[7] >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate16_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 intbuf[16];
|
|
|
|
for (int pos = 0; pos < len - 31; pos += 32)
|
|
{
|
|
intbuf[0] = buf[pos+2] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[1] = buf[pos+3] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[2] = buf[pos+6] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[3] = buf[pos+7] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[4] = buf[pos+10] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[5] = buf[pos+11] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[6] = buf[pos+14] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[7] = buf[pos+15] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[8] = buf[pos+18] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[9] = buf[pos+19] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[10] = buf[pos+22] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[11] = buf[pos+23] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[12] = buf[pos+26] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[13] = buf[pos+27] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[14] = buf[pos+30] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[15] = buf[pos+31] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
|
|
m_decimator2.myDecimate(
|
|
buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+4] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
buf[pos+5] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+8] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
buf[pos+9] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[4],
|
|
&intbuf[5]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+12] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
buf[pos+13] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+16] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
buf[pos+17] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[8],
|
|
&intbuf[9]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+20] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
buf[pos+21] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+24] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
buf[pos+25] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[12],
|
|
&intbuf[13]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+28] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
buf[pos+29] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[4],
|
|
intbuf[5],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[8],
|
|
intbuf[9],
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[12],
|
|
intbuf[13],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
|
|
m_decimator8.myDecimate(
|
|
intbuf[2],
|
|
intbuf[3],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[10],
|
|
intbuf[11],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
|
|
m_decimator16.myDecimate(
|
|
intbuf[6],
|
|
intbuf[7],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
|
|
(**it).setReal(intbuf[14] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
(**it).setImag(intbuf[15] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate16_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 intbuf[16];
|
|
|
|
for (int pos = 0; pos < len - 15; pos += 16)
|
|
{
|
|
intbuf[0] = bufI[pos+1] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[1] = bufQ[pos+1] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[2] = bufI[pos+3] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[3] = bufQ[pos+3] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[4] = bufI[pos+5] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[5] = bufQ[pos+5] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[6] = bufI[pos+7] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[7] = bufQ[pos+7] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[8] = bufI[pos+9] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[9] = bufQ[pos+9] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[10] = bufI[pos+11] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[11] = bufQ[pos+11] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[12] = bufI[pos+13] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[13] = bufQ[pos+13] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[14] = bufI[pos+15] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
intbuf[15] = bufQ[pos+15] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+0] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
bufQ[pos+0] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+2] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
bufQ[pos+2] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+4] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
bufQ[pos+4] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[4],
|
|
&intbuf[5]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+6] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
bufQ[pos+6] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+8] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
bufQ[pos+8] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[8],
|
|
&intbuf[9]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+10] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
bufQ[pos+10] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+12] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
bufQ[pos+12] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[12],
|
|
&intbuf[13]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+14] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
bufQ[pos+14] << decimation_shifts<SdrBits, InputBits>::pre16,
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[4],
|
|
intbuf[5],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[8],
|
|
intbuf[9],
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[12],
|
|
intbuf[13],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
|
|
m_decimator8.myDecimate(
|
|
intbuf[2],
|
|
intbuf[3],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[10],
|
|
intbuf[11],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
|
|
m_decimator16.myDecimate(
|
|
intbuf[6],
|
|
intbuf[7],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
|
|
(**it).setReal(intbuf[14] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
(**it).setImag(intbuf[15] >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate32_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 intbuf[32];
|
|
|
|
for (int pos = 0; pos < len - 63; pos += 64)
|
|
{
|
|
intbuf[0] = buf[pos+2] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[1] = buf[pos+3] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[2] = buf[pos+6] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[3] = buf[pos+7] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[4] = buf[pos+10] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[5] = buf[pos+11] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[6] = buf[pos+14] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[7] = buf[pos+15] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[8] = buf[pos+18] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[9] = buf[pos+19] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[10] = buf[pos+22] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[11] = buf[pos+23] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[12] = buf[pos+26] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[13] = buf[pos+27] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[14] = buf[pos+30] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[15] = buf[pos+31] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[16] = buf[pos+34] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[17] = buf[pos+35] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[18] = buf[pos+38] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[19] = buf[pos+39] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[20] = buf[pos+42] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[21] = buf[pos+43] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[22] = buf[pos+46] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[23] = buf[pos+47] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[24] = buf[pos+50] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[25] = buf[pos+51] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[26] = buf[pos+54] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[27] = buf[pos+55] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[28] = buf[pos+58] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[29] = buf[pos+59] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[30] = buf[pos+62] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[31] = buf[pos+63] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
|
|
m_decimator2.myDecimate(
|
|
buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+4] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+5] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+8] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+9] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[4],
|
|
&intbuf[5]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+12] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+13] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+16] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+17] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[8],
|
|
&intbuf[9]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+20] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+21] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+24] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+25] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[12],
|
|
&intbuf[13]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+28] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+29] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+32] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+33] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[16],
|
|
&intbuf[17]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+36] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+37] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[18],
|
|
&intbuf[19]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+40] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+41] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[20],
|
|
&intbuf[21]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+44] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+45] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+48] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+49] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[24],
|
|
&intbuf[25]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+52] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+53] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[26],
|
|
&intbuf[27]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+56] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+57] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[28],
|
|
&intbuf[29]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+60] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
buf[pos+61] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[4],
|
|
intbuf[5],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[8],
|
|
intbuf[9],
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[12],
|
|
intbuf[13],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[16],
|
|
intbuf[17],
|
|
&intbuf[18],
|
|
&intbuf[19]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[20],
|
|
intbuf[21],
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[24],
|
|
intbuf[25],
|
|
&intbuf[26],
|
|
&intbuf[27]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[28],
|
|
intbuf[29],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
m_decimator8.myDecimate(
|
|
intbuf[2],
|
|
intbuf[3],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[10],
|
|
intbuf[11],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[18],
|
|
intbuf[19],
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[26],
|
|
intbuf[27],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
m_decimator16.myDecimate(
|
|
intbuf[6],
|
|
intbuf[7],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator16.myDecimate(
|
|
intbuf[22],
|
|
intbuf[23],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
m_decimator32.myDecimate(
|
|
intbuf[14],
|
|
intbuf[15],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
(**it).setReal(intbuf[30] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
(**it).setImag(intbuf[31] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate32_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 intbuf[32];
|
|
|
|
for (int pos = 0; pos < len - 31; pos += 32)
|
|
{
|
|
intbuf[0] = bufI[pos+1] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[1] = bufQ[pos+1] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[2] = bufI[pos+3] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[3] = bufQ[pos+3] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[4] = bufI[pos+5] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[5] = bufQ[pos+5] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[6] = bufI[pos+7] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[7] = bufQ[pos+7] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[8] = bufI[pos+9] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[9] = bufQ[pos+9] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[10] = bufI[pos+11] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[11] = bufQ[pos+11] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[12] = bufI[pos+13] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[13] = bufQ[pos+13] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[14] = bufI[pos+15] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[15] = bufQ[pos+15] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[16] = bufI[pos+17] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[17] = bufQ[pos+17] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[18] = bufI[pos+19] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[19] = bufQ[pos+19] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[20] = bufI[pos+21] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[21] = bufQ[pos+21] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[22] = bufI[pos+23] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[23] = bufQ[pos+23] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[24] = bufI[pos+25] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[25] = bufQ[pos+25] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[26] = bufI[pos+27] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[27] = bufQ[pos+27] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[28] = bufI[pos+29] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[29] = bufQ[pos+29] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[30] = bufI[pos+31] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
intbuf[31] = bufQ[pos+31] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+0] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+0] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+2] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+2] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+4] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+4] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[4],
|
|
&intbuf[5]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+6] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+6] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+8] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+8] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[8],
|
|
&intbuf[9]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+10] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+10] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+12] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+12] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[12],
|
|
&intbuf[13]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+14] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+14] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+16] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+16] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[16],
|
|
&intbuf[17]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+18] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+18] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[18],
|
|
&intbuf[19]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+20] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+20] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[20],
|
|
&intbuf[21]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+22] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+22] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+24] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+24] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[24],
|
|
&intbuf[25]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+26] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+26] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[26],
|
|
&intbuf[27]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+28] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+28] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[28],
|
|
&intbuf[29]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+30] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
bufQ[pos+30] << decimation_shifts<SdrBits, InputBits>::pre32,
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[4],
|
|
intbuf[5],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[8],
|
|
intbuf[9],
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[12],
|
|
intbuf[13],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[16],
|
|
intbuf[17],
|
|
&intbuf[18],
|
|
&intbuf[19]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[20],
|
|
intbuf[21],
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[24],
|
|
intbuf[25],
|
|
&intbuf[26],
|
|
&intbuf[27]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[28],
|
|
intbuf[29],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
m_decimator8.myDecimate(
|
|
intbuf[2],
|
|
intbuf[3],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[10],
|
|
intbuf[11],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[18],
|
|
intbuf[19],
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[26],
|
|
intbuf[27],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
m_decimator16.myDecimate(
|
|
intbuf[6],
|
|
intbuf[7],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator16.myDecimate(
|
|
intbuf[22],
|
|
intbuf[23],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
m_decimator32.myDecimate(
|
|
intbuf[14],
|
|
intbuf[15],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
|
|
(**it).setReal(intbuf[30] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
(**it).setImag(intbuf[31] >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate64_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
qint32 intbuf[64];
|
|
|
|
for (int pos = 0; pos < len - 127; pos += 128)
|
|
{
|
|
intbuf[0] = buf[pos+2] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[1] = buf[pos+3] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[2] = buf[pos+6] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[3] = buf[pos+7] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[4] = buf[pos+10] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[5] = buf[pos+11] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[6] = buf[pos+14] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[7] = buf[pos+15] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[8] = buf[pos+18] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[9] = buf[pos+19] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[10] = buf[pos+22] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[11] = buf[pos+23] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[12] = buf[pos+26] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[13] = buf[pos+27] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[14] = buf[pos+30] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[15] = buf[pos+31] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[16] = buf[pos+34] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[17] = buf[pos+35] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[18] = buf[pos+38] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[19] = buf[pos+39] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[20] = buf[pos+42] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[21] = buf[pos+43] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[22] = buf[pos+46] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[23] = buf[pos+47] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[24] = buf[pos+50] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[25] = buf[pos+51] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[26] = buf[pos+54] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[27] = buf[pos+55] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[28] = buf[pos+58] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[29] = buf[pos+59] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[30] = buf[pos+62] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[31] = buf[pos+63] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
|
|
intbuf[32] = buf[pos+66] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[33] = buf[pos+67] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[34] = buf[pos+70] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[35] = buf[pos+71] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[36] = buf[pos+74] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[37] = buf[pos+75] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[38] = buf[pos+78] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[39] = buf[pos+79] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[40] = buf[pos+82] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[41] = buf[pos+83] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[42] = buf[pos+86] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[43] = buf[pos+87] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[44] = buf[pos+90] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[45] = buf[pos+91] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[46] = buf[pos+94] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[47] = buf[pos+95] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[48] = buf[pos+98] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[49] = buf[pos+99] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[50] = buf[pos+102] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[51] = buf[pos+103] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[52] = buf[pos+106] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[53] = buf[pos+107] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[54] = buf[pos+110] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[55] = buf[pos+111] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[56] = buf[pos+114] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[57] = buf[pos+115] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[58] = buf[pos+118] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[59] = buf[pos+119] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[60] = buf[pos+122] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[61] = buf[pos+123] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[62] = buf[pos+126] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[63] = buf[pos+127] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
|
|
m_decimator2.myDecimate(
|
|
buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+4] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+5] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+8] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+9] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[4],
|
|
&intbuf[5]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+12] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+13] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+16] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+17] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[8],
|
|
&intbuf[9]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+20] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+21] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+24] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+25] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[12],
|
|
&intbuf[13]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+28] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+29] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+32] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+33] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[16],
|
|
&intbuf[17]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+36] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+37] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[18],
|
|
&intbuf[19]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+40] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+41] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[20],
|
|
&intbuf[21]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+44] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+45] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+48] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+49] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[24],
|
|
&intbuf[25]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+52] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+53] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[26],
|
|
&intbuf[27]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+56] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+57] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[28],
|
|
&intbuf[29]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+60] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+61] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+64] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+65] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[32],
|
|
&intbuf[33]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+68] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+69] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[34],
|
|
&intbuf[35]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+72] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+73] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[36],
|
|
&intbuf[37]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+76] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+77] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[38],
|
|
&intbuf[39]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+80] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+81] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[40],
|
|
&intbuf[41]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+84] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+85] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[42],
|
|
&intbuf[43]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+88] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+89] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[44],
|
|
&intbuf[45]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+92] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+93] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[46],
|
|
&intbuf[47]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+96] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+97] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[48],
|
|
&intbuf[49]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+100] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+101] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[50],
|
|
&intbuf[51]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+104] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+105] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[52],
|
|
&intbuf[53]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+108] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+109] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[54],
|
|
&intbuf[55]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+112] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+113] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[56],
|
|
&intbuf[57]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+116] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+117] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[58],
|
|
&intbuf[59]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+120] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+121] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[60],
|
|
&intbuf[61]);
|
|
m_decimator2.myDecimate(
|
|
buf[pos+124] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
buf[pos+125] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[4],
|
|
intbuf[5],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[8],
|
|
intbuf[9],
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[12],
|
|
intbuf[13],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[16],
|
|
intbuf[17],
|
|
&intbuf[18],
|
|
&intbuf[19]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[20],
|
|
intbuf[21],
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[24],
|
|
intbuf[25],
|
|
&intbuf[26],
|
|
&intbuf[27]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[28],
|
|
intbuf[29],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[32],
|
|
intbuf[33],
|
|
&intbuf[34],
|
|
&intbuf[35]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[36],
|
|
intbuf[37],
|
|
&intbuf[38],
|
|
&intbuf[39]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[40],
|
|
intbuf[41],
|
|
&intbuf[42],
|
|
&intbuf[43]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[44],
|
|
intbuf[45],
|
|
&intbuf[46],
|
|
&intbuf[47]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[48],
|
|
intbuf[49],
|
|
&intbuf[50],
|
|
&intbuf[51]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[52],
|
|
intbuf[53],
|
|
&intbuf[54],
|
|
&intbuf[55]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[56],
|
|
intbuf[57],
|
|
&intbuf[58],
|
|
&intbuf[59]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[60],
|
|
intbuf[61],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator8.myDecimate(
|
|
intbuf[2],
|
|
intbuf[3],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[10],
|
|
intbuf[11],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[18],
|
|
intbuf[19],
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[26],
|
|
intbuf[27],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[34],
|
|
intbuf[35],
|
|
&intbuf[38],
|
|
&intbuf[39]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[42],
|
|
intbuf[43],
|
|
&intbuf[46],
|
|
&intbuf[47]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[50],
|
|
intbuf[51],
|
|
&intbuf[54],
|
|
&intbuf[55]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[58],
|
|
intbuf[59],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator16.myDecimate(
|
|
intbuf[6],
|
|
intbuf[7],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator16.myDecimate(
|
|
intbuf[22],
|
|
intbuf[23],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator16.myDecimate(
|
|
intbuf[38],
|
|
intbuf[39],
|
|
&intbuf[46],
|
|
&intbuf[47]);
|
|
m_decimator16.myDecimate(
|
|
intbuf[54],
|
|
intbuf[55],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator32.myDecimate(
|
|
intbuf[14],
|
|
intbuf[15],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator32.myDecimate(
|
|
intbuf[46],
|
|
intbuf[47],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator64.myDecimate(
|
|
intbuf[30],
|
|
intbuf[31],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
(**it).setReal(intbuf[62] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
(**it).setImag(intbuf[63] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate64_cen(SampleVector::iterator* it, const T* bufI, const T* bufQ, qint32 len)
|
|
{
|
|
qint32 intbuf[64];
|
|
|
|
for (int pos = 0; pos < len - 63; pos += 64)
|
|
{
|
|
intbuf[0] = bufI[pos+1] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[1] = bufQ[pos+1] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[2] = bufI[pos+3] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[3] = bufQ[pos+3] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[4] = bufI[pos+5] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[5] = bufQ[pos+5] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[6] = bufI[pos+7] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[7] = bufQ[pos+7] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[8] = bufI[pos+9] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[9] = bufQ[pos+9] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[10] = bufI[pos+11] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[11] = bufQ[pos+11] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[12] = bufI[pos+13] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[13] = bufQ[pos+13] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[14] = bufI[pos+15] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[15] = bufQ[pos+15] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[16] = bufI[pos+17] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[17] = bufQ[pos+17] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[18] = bufI[pos+19] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[19] = bufQ[pos+19] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[20] = bufI[pos+21] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[21] = bufQ[pos+21] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[22] = bufI[pos+23] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[23] = bufQ[pos+23] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[24] = bufI[pos+25] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[25] = bufQ[pos+25] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[26] = bufI[pos+27] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[27] = bufQ[pos+27] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[28] = bufI[pos+29] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[29] = bufQ[pos+29] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[30] = bufI[pos+31] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[31] = bufQ[pos+31] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
|
|
intbuf[32] = bufI[pos+33] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[33] = bufQ[pos+33] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[34] = bufI[pos+35] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[35] = bufQ[pos+35] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[36] = bufI[pos+37] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[37] = bufQ[pos+37] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[38] = bufI[pos+39] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[39] = bufQ[pos+39] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[40] = bufI[pos+41] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[41] = bufQ[pos+41] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[42] = bufI[pos+43] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[43] = bufQ[pos+43] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[44] = bufI[pos+45] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[45] = bufQ[pos+45] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[46] = bufI[pos+47] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[47] = bufQ[pos+47] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[48] = bufI[pos+49] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[49] = bufQ[pos+49] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[50] = bufI[pos+51] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[51] = bufQ[pos+51] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[52] = bufI[pos+53] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[53] = bufQ[pos+53] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[54] = bufI[pos+55] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[55] = bufQ[pos+55] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[56] = bufI[pos+57] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[57] = bufQ[pos+57] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[58] = bufI[pos+59] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[59] = bufQ[pos+59] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[60] = bufI[pos+61] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[61] = bufQ[pos+61] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[62] = bufI[pos+63] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
intbuf[63] = bufQ[pos+63] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+0] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+0] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[0],
|
|
&intbuf[1]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+2] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+2] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+4] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+4] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[4],
|
|
&intbuf[5]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+6] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+6] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+8] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+8] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[8],
|
|
&intbuf[9]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+10] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+10] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+12] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+12] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[12],
|
|
&intbuf[13]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+14] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+14] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+16] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+16] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[16],
|
|
&intbuf[17]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+18] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+18] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[18],
|
|
&intbuf[19]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+20] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+20] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[20],
|
|
&intbuf[21]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+22] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+22] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+24] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+24] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[24],
|
|
&intbuf[25]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+26] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+26] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[26],
|
|
&intbuf[27]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+28] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+28] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[28],
|
|
&intbuf[29]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+30] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+30] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+32] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+32] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[32],
|
|
&intbuf[33]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+34] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+34] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[34],
|
|
&intbuf[35]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+36] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+36] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[36],
|
|
&intbuf[37]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+38] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+38] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[38],
|
|
&intbuf[39]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+40] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+40] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[40],
|
|
&intbuf[41]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+42] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+42] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[42],
|
|
&intbuf[43]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+44] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+44] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[44],
|
|
&intbuf[45]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+46] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+46] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[46],
|
|
&intbuf[47]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+48] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+48] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[48],
|
|
&intbuf[49]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+50] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+50] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[50],
|
|
&intbuf[51]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+52] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+52] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[52],
|
|
&intbuf[53]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+54] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+54] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[54],
|
|
&intbuf[55]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+56] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+56] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[56],
|
|
&intbuf[57]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+58] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+58] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[58],
|
|
&intbuf[59]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+60] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+60] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[60],
|
|
&intbuf[61]);
|
|
m_decimator2.myDecimate(
|
|
bufI[pos+62] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
bufQ[pos+62] << decimation_shifts<SdrBits, InputBits>::pre64,
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator4.myDecimate(
|
|
intbuf[0],
|
|
intbuf[1],
|
|
&intbuf[2],
|
|
&intbuf[3]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[4],
|
|
intbuf[5],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[8],
|
|
intbuf[9],
|
|
&intbuf[10],
|
|
&intbuf[11]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[12],
|
|
intbuf[13],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[16],
|
|
intbuf[17],
|
|
&intbuf[18],
|
|
&intbuf[19]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[20],
|
|
intbuf[21],
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[24],
|
|
intbuf[25],
|
|
&intbuf[26],
|
|
&intbuf[27]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[28],
|
|
intbuf[29],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[32],
|
|
intbuf[33],
|
|
&intbuf[34],
|
|
&intbuf[35]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[36],
|
|
intbuf[37],
|
|
&intbuf[38],
|
|
&intbuf[39]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[40],
|
|
intbuf[41],
|
|
&intbuf[42],
|
|
&intbuf[43]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[44],
|
|
intbuf[45],
|
|
&intbuf[46],
|
|
&intbuf[47]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[48],
|
|
intbuf[49],
|
|
&intbuf[50],
|
|
&intbuf[51]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[52],
|
|
intbuf[53],
|
|
&intbuf[54],
|
|
&intbuf[55]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[56],
|
|
intbuf[57],
|
|
&intbuf[58],
|
|
&intbuf[59]);
|
|
m_decimator4.myDecimate(
|
|
intbuf[60],
|
|
intbuf[61],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator8.myDecimate(
|
|
intbuf[2],
|
|
intbuf[3],
|
|
&intbuf[6],
|
|
&intbuf[7]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[10],
|
|
intbuf[11],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[18],
|
|
intbuf[19],
|
|
&intbuf[22],
|
|
&intbuf[23]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[26],
|
|
intbuf[27],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[34],
|
|
intbuf[35],
|
|
&intbuf[38],
|
|
&intbuf[39]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[42],
|
|
intbuf[43],
|
|
&intbuf[46],
|
|
&intbuf[47]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[50],
|
|
intbuf[51],
|
|
&intbuf[54],
|
|
&intbuf[55]);
|
|
m_decimator8.myDecimate(
|
|
intbuf[58],
|
|
intbuf[59],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator16.myDecimate(
|
|
intbuf[6],
|
|
intbuf[7],
|
|
&intbuf[14],
|
|
&intbuf[15]);
|
|
m_decimator16.myDecimate(
|
|
intbuf[22],
|
|
intbuf[23],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator16.myDecimate(
|
|
intbuf[38],
|
|
intbuf[39],
|
|
&intbuf[46],
|
|
&intbuf[47]);
|
|
m_decimator16.myDecimate(
|
|
intbuf[54],
|
|
intbuf[55],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator32.myDecimate(
|
|
intbuf[14],
|
|
intbuf[15],
|
|
&intbuf[30],
|
|
&intbuf[31]);
|
|
m_decimator32.myDecimate(
|
|
intbuf[46],
|
|
intbuf[47],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
m_decimator64.myDecimate(
|
|
intbuf[30],
|
|
intbuf[31],
|
|
&intbuf[62],
|
|
&intbuf[63]);
|
|
|
|
(**it).setReal(intbuf[62] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
(**it).setImag(intbuf[63] >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
++(*it);
|
|
}
|
|
}
|
|
|
|
/*
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate2_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
int pos = 0;
|
|
|
|
while (pos < len - 1)
|
|
{
|
|
qint32 x0 = buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
qint32 y0 = buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre2;
|
|
pos += 2;
|
|
|
|
if (m_decimator2.workDecimateCenter(&x0, &y0))
|
|
{
|
|
(**it).setReal(x0 >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
(**it).setImag(y0 >> decimation_shifts<SdrBits, InputBits>::post2);
|
|
++(*it);
|
|
}
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate4_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
int pos = 0;
|
|
|
|
while (pos < len - 1)
|
|
{
|
|
qint32 x0 = buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
qint32 y0 = buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre4;
|
|
pos += 2;
|
|
|
|
if (m_decimator2.workDecimateCenter(&x0, &y0))
|
|
{
|
|
qint32 x1 = x0;
|
|
qint32 y1 = y0;
|
|
|
|
if (m_decimator4.workDecimateCenter(&x1, &y1))
|
|
{
|
|
(**it).setReal(x0 >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
(**it).setImag(y0 >> decimation_shifts<SdrBits, InputBits>::post4);
|
|
++(*it);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate8_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
int pos = 0;
|
|
|
|
while (pos < len - 1)
|
|
{
|
|
qint32 x0 = buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
qint32 y0 = buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre8;
|
|
pos += 2;
|
|
|
|
if (m_decimator2.workDecimateCenter(&x0, &y0))
|
|
{
|
|
qint32 x1 = x0;
|
|
qint32 y1 = y0;
|
|
|
|
if (m_decimator4.workDecimateCenter(&x1, &y1))
|
|
{
|
|
qint32 x2 = x1;
|
|
qint32 y2 = y1;
|
|
|
|
if (m_decimator8.workDecimateCenter(&x2, &y2))
|
|
{
|
|
(**it).setReal(x2 >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
(**it).setImag(y2 >> decimation_shifts<SdrBits, InputBits>::post8);
|
|
++(*it);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate16_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
int pos = 0;
|
|
|
|
while (pos < len - 1)
|
|
{
|
|
qint32 x0 = buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
qint32 y0 = buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre16;
|
|
pos += 2;
|
|
|
|
if (m_decimator2.workDecimateCenter(&x0, &y0))
|
|
{
|
|
qint32 x1 = x0;
|
|
qint32 y1 = y0;
|
|
|
|
if (m_decimator4.workDecimateCenter(&x1, &y1))
|
|
{
|
|
qint32 x2 = x1;
|
|
qint32 y2 = y1;
|
|
|
|
if (m_decimator8.workDecimateCenter(&x2, &y2))
|
|
{
|
|
qint32 x3 = x2;
|
|
qint32 y3 = y2;
|
|
|
|
if (m_decimator16.workDecimateCenter(&x3, &y3))
|
|
{
|
|
(**it).setReal(x3 >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
(**it).setImag(y3 >> decimation_shifts<SdrBits, InputBits>::post16);
|
|
++(*it);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate32_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
int pos = 0;
|
|
|
|
while (pos < len - 1)
|
|
{
|
|
qint32 x0 = buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
qint32 y0 = buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre32;
|
|
pos += 2;
|
|
|
|
if (m_decimator2.workDecimateCenter(&x0, &y0))
|
|
{
|
|
qint32 x1 = x0;
|
|
qint32 y1 = y0;
|
|
|
|
if (m_decimator4.workDecimateCenter(&x1, &y1))
|
|
{
|
|
qint32 x2 = x1;
|
|
qint32 y2 = y1;
|
|
|
|
if (m_decimator8.workDecimateCenter(&x2, &y2))
|
|
{
|
|
qint32 x3 = x2;
|
|
qint32 y3 = y2;
|
|
|
|
if (m_decimator16.workDecimateCenter(&x3, &y3))
|
|
{
|
|
qint32 x4 = x3;
|
|
qint32 y4 = y3;
|
|
|
|
if (m_decimator32.workDecimateCenter(&x4, &y4))
|
|
{
|
|
(**it).setReal(x4 >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
(**it).setImag(y4 >> decimation_shifts<SdrBits, InputBits>::post32);
|
|
++(*it);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
template<typename T, uint SdrBits, uint InputBits>
|
|
void Decimators<T, SdrBits, InputBits>::decimate64_cen(SampleVector::iterator* it, const T* buf, qint32 len)
|
|
{
|
|
int pos = 0;
|
|
|
|
while (pos < len - 1)
|
|
{
|
|
qint32 x0 = buf[pos+0] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
qint32 y0 = buf[pos+1] << decimation_shifts<SdrBits, InputBits>::pre64;
|
|
pos += 2;
|
|
|
|
if (m_decimator2.workDecimateCenter(&x0, &y0))
|
|
{
|
|
qint32 x1 = x0;
|
|
qint32 y1 = y0;
|
|
|
|
if (m_decimator4.workDecimateCenter(&x1, &y1))
|
|
{
|
|
qint32 x2 = x1;
|
|
qint32 y2 = y1;
|
|
|
|
if (m_decimator8.workDecimateCenter(&x2, &y2))
|
|
{
|
|
qint32 x3 = x2;
|
|
qint32 y3 = y2;
|
|
|
|
if (m_decimator16.workDecimateCenter(&x3, &y3))
|
|
{
|
|
qint32 x4 = x3;
|
|
qint32 y4 = y3;
|
|
|
|
if (m_decimator32.workDecimateCenter(&x4, &y4))
|
|
{
|
|
qint32 x5 = x4;
|
|
qint32 y5 = y4;
|
|
|
|
if (m_decimator64.workDecimateCenter(&x5, &y5))
|
|
{
|
|
(**it).setReal(x5 >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
(**it).setImag(y5 >> decimation_shifts<SdrBits, InputBits>::post64);
|
|
++(*it);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
*/
|
|
|
|
#endif /* INCLUDE_GPL_DSP_DECIMATORS_H_ */
|