mirror of
https://github.com/f4exb/sdrangel.git
synced 2024-11-10 10:33:29 -05:00
408 lines
10 KiB
C++
408 lines
10 KiB
C++
#ifndef LEANSDR_GENERIC_H
|
|
#define LEANSDR_GENERIC_H
|
|
|
|
#include <sys/types.h>
|
|
#include <unistd.h>
|
|
|
|
#include "leansdr/math.h"
|
|
|
|
namespace leansdr
|
|
{
|
|
|
|
//////////////////////////////////////////////////////////////////////
|
|
// Simple blocks
|
|
//////////////////////////////////////////////////////////////////////
|
|
|
|
// [file_reader] reads raw data from a file descriptor into a [pipebuf].
|
|
// If the file descriptor is seekable, data can be looped.
|
|
|
|
template<typename T>
|
|
struct file_reader: runnable
|
|
{
|
|
file_reader(scheduler *sch, int _fdin, pipebuf<T> &_out) :
|
|
runnable(sch, _out.name), loop(false), fdin(_fdin), out(_out)
|
|
{
|
|
}
|
|
void run()
|
|
{
|
|
size_t size = out.writable() * sizeof(T);
|
|
if (!size)
|
|
return;
|
|
|
|
again: ssize_t nr = read(fdin, out.wr(), size);
|
|
if (nr < 0)
|
|
{
|
|
fatal("leansdr::file_reader::run: read");
|
|
return;
|
|
}
|
|
if (!nr)
|
|
{
|
|
if (!loop)
|
|
return;
|
|
if (sch->debug)
|
|
fprintf(stderr, "leansdr::file_reader::run: %s looping\n", name);
|
|
off_t res = lseek(fdin, 0, SEEK_SET);
|
|
if (res == (off_t) -1)
|
|
{
|
|
fatal("leansdr::file_reader::run: lseek");
|
|
return;
|
|
}
|
|
goto again;
|
|
}
|
|
|
|
// Always stop at element boundary (may block)
|
|
size_t partial = nr % sizeof(T);
|
|
size_t remain = partial ? sizeof(T) - partial : 0;
|
|
while (remain)
|
|
{
|
|
if (sch->debug)
|
|
fprintf(stderr, "+");
|
|
ssize_t nr2 = read(fdin, (char*) out.wr() + nr, remain);
|
|
if (nr2 <= 0)
|
|
{
|
|
fatal("leansdr::file_reader::run: partial read");
|
|
return;
|
|
}
|
|
nr += nr2;
|
|
remain -= nr2;
|
|
}
|
|
|
|
out.written(nr / sizeof(T));
|
|
}
|
|
bool loop;
|
|
private:
|
|
int fdin;
|
|
pipewriter<T> out;
|
|
};
|
|
|
|
// [file_writer] writes raw data from a [pipebuf] to a file descriptor.
|
|
|
|
template<typename T>
|
|
struct file_writer: runnable
|
|
{
|
|
file_writer(scheduler *sch, pipebuf<T> &_in, int _fdout) :
|
|
runnable(sch, _in.name), in(_in), fdout(_fdout)
|
|
{
|
|
}
|
|
void run()
|
|
{
|
|
int size = in.readable() * sizeof(T);
|
|
if (!size)
|
|
return;
|
|
int nw = write(fdout, in.rd(), size);
|
|
if (!nw)
|
|
{
|
|
fatal("leansdr::file_writer::run: pipe");
|
|
return;
|
|
}
|
|
if (nw < 0)
|
|
{
|
|
fatal("leansdr::file_writer::run: write");
|
|
return;
|
|
}
|
|
if (nw % sizeof(T))
|
|
{
|
|
fatal("leansdr::file_writer::run:partial write");
|
|
return;
|
|
}
|
|
in.read(nw / sizeof(T));
|
|
}
|
|
private:
|
|
pipereader<T> in;
|
|
int fdout;
|
|
};
|
|
|
|
// [file_printer] writes data from a [pipebuf] to a file descriptor,
|
|
// with printf-style formatting and optional scaling.
|
|
|
|
template<typename T>
|
|
struct file_printer: runnable
|
|
{
|
|
file_printer(scheduler *sch, const char *_format, pipebuf<T> &_in, int _fdout, int _decimation = 1) :
|
|
runnable(sch, _in.name), scale(1), decimation(_decimation), in(_in), format(_format), fdout(_fdout), phase(0)
|
|
{
|
|
}
|
|
void run()
|
|
{
|
|
int n = in.readable();
|
|
T *pin = in.rd(), *pend = pin + n;
|
|
for (; pin < pend; ++pin)
|
|
{
|
|
if (++phase >= decimation)
|
|
{
|
|
phase -= decimation;
|
|
char buf[256];
|
|
int len = snprintf(buf, sizeof(buf), format, (*pin) * scale);
|
|
if (len < 0)
|
|
{
|
|
fatal("leansdr::file_printer::run: obsolete glibc");
|
|
return;
|
|
}
|
|
int nw = write(fdout, buf, len);
|
|
if (nw != len)
|
|
{
|
|
fatal("leansdr::file_printer::run: partial write");
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
in.read(n);
|
|
}
|
|
T scale;
|
|
int decimation;
|
|
private:
|
|
pipereader<T> in;
|
|
const char *format;
|
|
int fdout;
|
|
int phase;
|
|
};
|
|
|
|
// [file_carrayprinter] writes all data available from a [pipebuf]
|
|
// to a file descriptor on a single line.
|
|
// Special case for complex.
|
|
|
|
template<typename T>
|
|
struct file_carrayprinter: runnable
|
|
{
|
|
file_carrayprinter(scheduler *sch, const char *_head, const char *_format, const char *_sep, const char *_tail, pipebuf<complex<T> > &_in, int _fdout) :
|
|
runnable(sch, _in.name), scale(1), fixed_size(0), in(_in), head(_head), format(_format), sep(_sep), tail(_tail), fout(fdopen(_fdout, "w"))
|
|
{
|
|
}
|
|
void run()
|
|
{
|
|
int n, nmin = fixed_size ? fixed_size : 1;
|
|
while ((n = in.readable()) >= nmin)
|
|
{
|
|
if (fixed_size)
|
|
n = fixed_size;
|
|
if (fout)
|
|
{
|
|
fprintf(fout, head, n);
|
|
complex<T> *pin = in.rd();
|
|
for (int i = 0; i < n; ++i)
|
|
{
|
|
if (i)
|
|
fprintf(fout, "%s", sep);
|
|
fprintf(fout, format, pin[i].re * scale, pin[i].im * scale);
|
|
}
|
|
fprintf(fout, "%s", tail);
|
|
}
|
|
fflush(fout);
|
|
in.read(n);
|
|
}
|
|
}
|
|
T scale;
|
|
int fixed_size; // Number of elements per batch, or 0.
|
|
private:
|
|
pipereader<complex<T> > in;
|
|
const char *head, *format, *sep, *tail;
|
|
FILE *fout;
|
|
};
|
|
|
|
template<typename T, int N>
|
|
struct file_vectorprinter: runnable
|
|
{
|
|
file_vectorprinter(scheduler *sch, const char *_head, const char *_format, const char *_sep, const char *_tail, pipebuf<T[N]> &_in, int _fdout) :
|
|
runnable(sch, _in.name), scale(1), in(_in), head(_head), format(_format), sep(_sep), tail(_tail)
|
|
{
|
|
fout = fdopen(_fdout, "w");
|
|
if (!fout)
|
|
{
|
|
fatal("leansdr::file_vectorprinter::file_vectorprinter: fdopen");
|
|
}
|
|
}
|
|
void run()
|
|
{
|
|
while (in.readable() >= 1)
|
|
{
|
|
fprintf(fout, head, N);
|
|
T (*pin)[N] = in.rd();
|
|
for (int i = 0; i < N; ++i)
|
|
{
|
|
if (i)
|
|
fprintf(fout, "%s", sep);
|
|
fprintf(fout, format, (*pin)[i] * scale);
|
|
}
|
|
fprintf(fout, "%s", tail);
|
|
in.read(1);
|
|
}
|
|
fflush(fout);
|
|
}
|
|
T scale;
|
|
private:
|
|
pipereader<T[N]> in;
|
|
const char *head, *format, *sep, *tail;
|
|
FILE *fout;
|
|
};
|
|
|
|
// [itemcounter] writes the number of input items to the output [pipebuf].
|
|
// [Tout] must be a numeric type.
|
|
|
|
template<typename Tin, typename Tout>
|
|
struct itemcounter: runnable
|
|
{
|
|
itemcounter(scheduler *sch, pipebuf<Tin> &_in, pipebuf<Tout> &_out) :
|
|
runnable(sch, "itemcounter"), in(_in), out(_out)
|
|
{
|
|
}
|
|
void run()
|
|
{
|
|
if (out.writable() < 1)
|
|
return;
|
|
unsigned long count = in.readable();
|
|
if (!count)
|
|
return;
|
|
out.write(count);
|
|
in.read(count);
|
|
}
|
|
private:
|
|
pipereader<Tin> in;
|
|
pipewriter<Tout> out;
|
|
};
|
|
|
|
// [decimator] forwards 1 in N sample.
|
|
|
|
template<typename T>
|
|
struct decimator: runnable
|
|
{
|
|
unsigned int d;
|
|
|
|
decimator(scheduler *sch, int _d, pipebuf<T> &_in, pipebuf<T> &_out) :
|
|
runnable(sch, "decimator"), d(_d), in(_in), out(_out)
|
|
{
|
|
}
|
|
void run()
|
|
{
|
|
unsigned long count = min(in.readable() / d, out.writable());
|
|
T *pin = in.rd(), *pend = pin + count * d, *pout = out.wr();
|
|
for (; pin < pend; pin += d, ++pout)
|
|
*pout = *pin;
|
|
in.read(count * d);
|
|
out.written(count);
|
|
}
|
|
private:
|
|
pipereader<T> in;
|
|
pipewriter<T> out;
|
|
};
|
|
|
|
// [rate_estimator] accumulates counts of two quantities
|
|
// and periodically outputs their ratio.
|
|
|
|
template<typename T>
|
|
struct rate_estimator: runnable
|
|
{
|
|
int sample_size;
|
|
|
|
rate_estimator(scheduler *sch, pipebuf<int> &_num, pipebuf<int> &_den, pipebuf<float> &_rate) :
|
|
runnable(sch, "rate_estimator"), sample_size(10000), num(_num), den(_den), rate(_rate), acc_num(0), acc_den(0)
|
|
{
|
|
}
|
|
|
|
void run()
|
|
{
|
|
if (rate.writable() < 1)
|
|
return;
|
|
int count = min(num.readable(), den.readable());
|
|
int *pnum = num.rd(), *pden = den.rd();
|
|
for (int n = count; n--; ++pnum, ++pden)
|
|
{
|
|
acc_num += *pnum;
|
|
acc_den += *pden;
|
|
}
|
|
num.read(count);
|
|
den.read(count);
|
|
if (acc_den >= sample_size)
|
|
{
|
|
rate.write((float) acc_num / acc_den);
|
|
acc_num = acc_den = 0;
|
|
}
|
|
}
|
|
|
|
private:
|
|
pipereader<int> num, den;
|
|
pipewriter<float> rate;
|
|
T acc_num, acc_den;
|
|
};
|
|
|
|
// SERIALIZER
|
|
|
|
template<typename Tin, typename Tout>
|
|
struct serializer: runnable
|
|
{
|
|
serializer(scheduler *sch __attribute__((unused)), pipebuf<Tin> &_in, pipebuf<Tout> &_out) :
|
|
nin(max((size_t) 1, sizeof(Tin) / sizeof(Tout))), nout(max((size_t) 1, sizeof(Tout) / sizeof(Tin))), in(_in), out(_out, nout)
|
|
{
|
|
if (nin * sizeof(Tin) != nout * sizeof(Tout))
|
|
{
|
|
fail("serializer::serializer", "incompatible sizes");
|
|
return;
|
|
}
|
|
}
|
|
void run()
|
|
{
|
|
while (in.readable() >= nin && out.writable() >= nout)
|
|
{
|
|
memcpy(out.wr(), in.rd(), nout * sizeof(Tout));
|
|
in.read(nin);
|
|
out.written(nout);
|
|
}
|
|
}
|
|
private:
|
|
int nin, nout;
|
|
pipereader<Tin> in;
|
|
pipewriter<Tout> out;
|
|
};
|
|
// serializer
|
|
|
|
// [buffer_reader] reads from a user-supplied buffer.
|
|
|
|
template<typename T>
|
|
struct buffer_reader: runnable
|
|
{
|
|
buffer_reader(scheduler *sch, T *_data, int _count, pipebuf<T> &_out) :
|
|
runnable(sch, "buffer_reader"), data(_data), count(_count), out(_out), pos(0)
|
|
{
|
|
}
|
|
void run()
|
|
{
|
|
int n = min(out.writable(), (unsigned long) (count - pos));
|
|
memcpy(out.wr(), &data[pos], n * sizeof(T));
|
|
pos += n;
|
|
out.written(n);
|
|
}
|
|
private:
|
|
T *data;
|
|
int count;
|
|
pipewriter<T> out;
|
|
int pos;
|
|
};
|
|
// buffer_reader
|
|
|
|
// [buffer_writer] writes to a user-supplied buffer.
|
|
|
|
template<typename T>
|
|
struct buffer_writer: runnable
|
|
{
|
|
buffer_writer(scheduler *sch, pipebuf<T> &_in, T *_data, int _count) :
|
|
runnable(sch, "buffer_reader"), in(_in), data(_data), count(_count), pos(0)
|
|
{
|
|
}
|
|
void run()
|
|
{
|
|
int n = min(in.readable(), (unsigned long) (count - pos));
|
|
memcpy(&data[pos], in.rd(), n * sizeof(T));
|
|
in.read(n);
|
|
pos += n;
|
|
}
|
|
private:
|
|
pipereader<T> in;
|
|
T *data;
|
|
int count;
|
|
int pos;
|
|
};
|
|
// buffer_writer
|
|
|
|
}// namespace
|
|
|
|
#endif // LEANSDR_GENERIC_H
|