⟦167af0be3⟧

TextFile

// This may look like C code, but it is really -*- C++ -*-
/* 
Copyright (C) 1988 Free Software Foundation
    written by Kurt Baudendistel (gt-eedsp!baud@gatech.edu)
    adapted for libg++ by Doug Lea (dl@rocky.oswego.edu)

This file is part of GNU CC.

GNU CC is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY.  No author or distributor
accepts responsibility to anyone for the consequences of using it
or for whether it serves any particular purpose or works at all,
unless he says so in writing.  Refer to the GNU CC General Public
License for full details.

Everyone is granted permission to copy, modify and redistribute
GNU CC, but only under the conditions described in the
GNU CC General Public License.   A copy of this license is
supposed to have been given to you along with GNU CC so you
can know your rights and responsibilities.  It should be in a
file named COPYING.  Among other things, the copyright notice
and this notice must be preserved on all copies.  
*/

//
// Fix24.cc : fixed precision class support functions
//

#include <Fix24.h>

const double
  Fix24_fs = 2147483648.,
  Fix24_mult = Fix24_fs,
  Fix24_div = 1./Fix24_fs,
  Fix24_max = 1. - .5/Fix24_fs,
  Fix24_min = -1.;
	  
const unsigned long
  Fix24_msb = 0x80000000L,
  Fix24_lsb = 0x00000100L,
  Fix24_m_max = 0x7fffff00L,
  Fix24_m_min = 0x80000000L;
		  
const double
  Fix48_fs = 4611686018427387904.,
  Fix48_max = 1. - .5/Fix48_fs,
  Fix48_min = -1.,
  Fix48_div_u = 1./Fix24_fs,
  Fix48_div_l = 1./Fix48_fs;
   
const twolongs
  Fix48_msb = { 0x80000000L, 0L },
  Fix48_lsb = { 0L, 0x00000100L },
  Fix48_m_max = { 0x7fffff00L, 0xffffff00L },
  Fix48_m_min = { 0x80000000L, 0L };
		  
// basic operators too large to be inline

long Fix24::assign(double d) 
{ 
  if (d == 1.0)
    return Fix24_m_max;
  else if (d > Fix24_max)
  {
    long i = Fix24_m_max;
    range_error(i);
    return i;
  }
  else if (d < Fix24_min)
  {
    long i = Fix24_m_min;
    range_error(i);
    return i;
  }
  else {
    d *= Fix24_mult;
    return (long)(d + ((d >= 0)? 0.5 : -0.5));
  }
}

twolongs Fix48::assign(double d) 
{ 
  if (d == 1.0)
    return Fix48_m_max;
  else if (d > Fix48_max)
  {
    twolongs i = Fix48_m_max;
    range_error(i);
    return i;
  }
  else if (d < Fix48_min)
  {
    twolongs i = Fix48_m_min;
    range_error(i);
    return i;
  }
  else {
    twolongs i;
    i.u = (long)(d *= Fix24_mult);
    i.l = (long)((d - (double)i.u) * Fix24_mult);
    return i;
  }
}


Fix48 operator * (Fix24& a, Fix24& b)
{
// break a and b into lo and hi parts, and do a multiple-precision
// multiply, with rounding

  int apos = (a.m >= 0);
  unsigned long ua = (apos)? a.m : - a.m;
  ua <<= 1; // ua is biased so result will be 47 bit mantissa, not 46:
  unsigned long hi_a = (ua >> 16) & ((1 << 16) - 1);
  unsigned long lo_a = ua & ((1 << 16) - 1);

  int bpos = (b.m >= 0);
  unsigned long ub = (bpos)? b.m : -b.m;
  unsigned long hi_b = (ub >> 16) & ((1 << 16) - 1);
  unsigned long lo_b = ub & ((1 << 16) - 1);

  unsigned long 
    hi_r = hi_a * hi_b,
    mi_r = hi_a * lo_b + lo_a * hi_b,
    lo_r = lo_a * lo_b,
    rl = ((hi_r << 16) & 0x00ffffffL) + (mi_r & 0x00ffffffL) + (lo_r >> 16);
  twolongs r = {
    (hi_r & 0xffffff00L) + ((mi_r >> 16) & 0x0000ff00L)
      + ((rl >> 16) & 0x0000ff00L),
    rl << 8
  };

  if ( apos != bpos ) {
    unsigned long l = r.l;
    r.l = -r.l;
    r.u = ~r.u + ((l ^ r.l) & Fix24_msb ? 0 : Fix24_lsb);
  }
  return r;
}

Fix24 operator / (Fix24& a, Fix24& b)
{
  long q;
  int apos = (a.m >= 0);
  unsigned long la = (apos)? a.m : -a.m;
  int bpos = (b.m >= 0);
  unsigned long lb = (bpos)? b.m: -b.m;
  if (la >= lb)
  {
    q = (apos == bpos)? Fix24_m_max: Fix24_m_min;
    a.range_error(q);
  }
  else                        // standard shift-based division alg
  {
    q = 0;
    long r = la;

    for (int i = 32; i > 0; i--)
    {
	if (r > lb) {
	    q = (q << 1) | 1;
	    r -= lb;
	}
	else
	    q = (q << 1);
	r <<= 1;
    }

    q += 0x80;			// Round result to 24 bits
    if (apos != bpos) q = -q;	// Fix sign
  }
  return (q & 0xffffff00);
}


Fix48 operator + (Fix48&  f, Fix48&  g)
{
  long lo_r = (f.m.l >> 8) + (g.m.l >> 8);
  twolongs r = {
    f.m.u + g.m.u + (lo_r & 0x01000000L ? 0x00000100L : 0),
    lo_r << 8
  };
  if ( (f.m.u ^ r.u) & (g.m.u ^ r.u) & Fix24_msb )
    f.overflow(r);
  return r;
}

Fix48 operator - (Fix48&  f, Fix48&  g)
{
  long lo_r = (f.m.l >> 8) - (g.m.l >> 8);
  twolongs r = {
    f.m.u - g.m.u - (lo_r & 0x01000000L ? 0x00000100L: 0),
    lo_r << 8
  };
  if ( (f.m.u ^ r.u) & (-g.m.u ^ r.u) & Fix24_msb )
    f.overflow(r);
  return r;
}

Fix48 operator * (Fix48& a, int b)
{
  twolongs r;
  int bpos = (b >= 0);
  unsigned ub = (bpos)? b : -b;
  if ( ub >= 65536L ) {
    r = (bpos)? Fix48_m_max : Fix48_m_min;
    a.range_error(r);
  }
  else {
    unsigned long 
      lo_r = (a.m.l & 0xffff) * ub,
      mi_r = ((a.m.l >> 16) & 0xffff) * ub,
      hi_r = a.m.u * ub;
    r.l = lo_r + (mi_r << 16);
    r.u = hi_r + ((mi_r >> 8) & 0x00ffff00L);
    if ( !bpos ) {
      unsigned long l = r.l;
      r.l = -r.l;
      r.u = ~r.u + ((l ^ r.l) & Fix24_msb ? 0 : Fix24_lsb);
    }
  }
  return r;
}

Fix48 operator << (Fix48& a, int b)
{
  twolongs r = { 0L, 0L };
  if ( b >= 0 )
    if ( b < 24 ) {
      r.u = (a.m.u << b) + ((a.m.l >> (24 - b)) & 0xffffff00L);
      r.l = a.m.l << b;
    }
    else if ( b < 48 ) {
      r.u = a.m.l << (b - 24);
    }
  return r;
}

Fix48 operator >> (Fix48& a, int b)
{
  twolongs r = { 0L, 0L };
  if ( b >= 0 )
    if ( b < 24 ) {
      r.l = (a.m.u << (24 - b)) + ((a.m.l >> b) & 0xffffff00L);
      r.u = (a.m.u >> b) & 0xffffff00L;
    }
    else if ( b < 48 ) {
      r.l = (a.m.u >> (b - 24)) & 0xffffff00L;
      r.u = (a.m.u >> 24) & 0xffffff00L;
    }
    else {
      r.l = (a.m.u >> 24) & 0xffffff00L;
      r.u = r.l;
    }
  return r;
}

// error handling

void Fix24::overflow(long& i)
{
  (*Fix24_overflow_handler)(i);
}

void Fix48::overflow(twolongs& i)
{
  (*Fix48_overflow_handler)(i);
}

void Fix24::range_error(long& i)
{
  (*Fix24_range_error_handler)(i);
}

void Fix48::range_error(twolongs& i)
{
  (*Fix48_range_error_handler)(i);
}

// data definitions

Fix24_peh Fix24_overflow_handler = Fix24_overflow_saturate;
Fix48_peh Fix48_overflow_handler = Fix48_overflow_saturate;

Fix24_peh Fix24_range_error_handler = Fix24_warning;
Fix48_peh Fix48_range_error_handler = Fix48_warning;

//function definitions

Fix24_peh set_Fix24_overflow_handler(Fix24_peh new_handler) {
  Fix24_peh old_handler = Fix24_overflow_handler;
  Fix24_overflow_handler = new_handler;
  return old_handler;
}

Fix48_peh set_Fix48_overflow_handler(Fix48_peh new_handler) {
  Fix48_peh old_handler = Fix48_overflow_handler;
  Fix48_overflow_handler = new_handler;
  return old_handler;
}

void set_overflow_handler(Fix24_peh handler24, Fix48_peh handler48) {
  set_Fix24_overflow_handler(handler24);
  set_Fix48_overflow_handler(handler48);
}

Fix24_peh set_Fix24_range_error_handler(Fix24_peh new_handler) {
  Fix24_peh old_handler = Fix24_range_error_handler;
  Fix24_range_error_handler = new_handler;
  return old_handler;
}

Fix48_peh set_Fix48_range_error_handler(Fix48_peh new_handler) {
  Fix48_peh old_handler = Fix48_range_error_handler;
  Fix48_range_error_handler = new_handler;
  return old_handler;
}

void set_range_error_handler(Fix24_peh handler24, Fix48_peh handler48) {
  set_Fix24_range_error_handler(handler24);
  set_Fix48_range_error_handler(handler48);
}

void Fix24_overflow_saturate(long& i)
  { i = (i > 0 ? Fix24_m_min : Fix24_m_max); }
void Fix24_ignore(long&) {}
void Fix24_warning(long&)
  { cerr << "warning: Fix24 result out of range\n"; }
void Fix24_overflow_warning_saturate(long& i)
  { cerr << "warning: Fix24 result out of range\n"; 
   Fix24_overflow_saturate(i); }
void Fix24_abort(long&)
  { cerr << "error: Fix24 result out of range\n"; abort(); }

void Fix48_ignore(twolongs&) {}
void Fix48_overflow_saturate(twolongs& i)
  { i = (i.u > 0 ? Fix48_m_min : Fix48_m_max); }
void Fix48_warning(twolongs&)
  { cerr << "warning: Fix48 result out of range\n"; }
void Fix48_overflow_warning_saturate(twolongs& i)
  { cerr << "warning: Fix48 result out of range\n"; 
   Fix48_overflow_saturate(i); }
void Fix48_abort(twolongs&)
  { cerr << "error: Fix48 result out of range\n"; abort(); }
DataMuseum.dk

DKUUG/EUUG Conference tapes

⟦167af0be3⟧ TextFile

Derivation

TextFile
