Login

Subversion Repositories NedoOS

Rev

Blame | Last modification | View Log | Download | RSS feed

/*-------------------------------------------------------------------------
   _fsdiv.c - Floating point library in optimized assembly for 8051

   Copyright (c) 2004, Paul Stoffregen, paul@pjrc.com

   This library 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; either version 2, or (at your option) any
   later version.

   This library 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 for more details.

   You should have received a copy of the GNU General Public License
   along with this library; see the file COPYING. If not, write to the
   Free Software Foundation, 51 Franklin Street, Fifth Floor, Boston,
   MA 02110-1301, USA.

   As a special exception, if you link this library with other files,
   some of which are compiled with SDCC, to produce an executable,
   this library does not by itself cause the resulting executable to
   be covered by the GNU General Public License. This exception does
   not however invalidate any other reasons why the executable file
   might be covered by the GNU General Public License.
-------------------------------------------------------------------------*/



#define __SDCC_FLOAT_LIB
#include <float.h>


#ifdef FLOAT_ASM_MCS51

// float __fsdiv (float a, float b) __reentrant
static void dummy(void) __naked
{
        __asm
        .globl  ___fsdiv
___fsdiv:
        // extract the two inputs, placing them into:
        //      sign     exponent   mantiassa
        //      ----     --------   ---------
        //  a:  sign_a   exp_a      r4/r3/r2
        //  b:  sign_b   exp_b      r7/r6/r5

        lcall   fsgetargs

        // compute final sign bit
        jnb     sign_b, 00001$
        cpl     sign_a
00001$:

        // if divisor is zero, ...
        cjne    r7, #0, 00003$
        // if dividend is also zero, return NaN
        cjne    r4, #0, 00002$
        ljmp    fs_return_nan
00002$:
        // but dividend is non-zero, return infinity
        ljmp    fs_return_inf
00003$:
        // if dividend is zero, return zero
        cjne    r4, #0, 00004$
        ljmp    fs_return_zero
00004$:
        // if divisor is infinity, ...
        mov     a, exp_b
        cjne    a, #0xFF, 00006$
        // and dividend is also infinity, return NaN
        mov     a, exp_a
        cjne    a, #0xFF, 00005$
        ljmp    fs_return_nan
00005$:
        // but dividend is not infinity, return zero
        ljmp    fs_return_zero
00006$:
        // if dividend is infinity, return infinity
        mov     a, exp_a
        cjne    a, #0xFF, 00007$
        ljmp    fs_return_inf
00007$:

        // subtract exponents
        clr     c
        subb    a, exp_b
        // if no carry then no underflow
        jnc     00008$
        add     a, #127
        jc      00009$
        ljmp    fs_return_zero

00008$:
        add     a, #128
        dec     a
        jnc     00009$
        ljmp    fs_return_inf

00009$:
        mov     exp_a, a

        // need extra bits on a's mantissa
#ifdef FLOAT_FULL_ACCURACY
        clr     c
        mov     a, r5
        subb    a, r2
        mov     a, r6
        subb    a, r3
        mov     a, r7
        subb    a, r4
        jc      00010$
        dec     exp_a
        clr     c
        mov     a, r2
        rlc     a
        mov     r1, a
        mov     a, r3
        rlc     a
        mov     r2, a
        mov     a, r4
        rlc     a
        mov     r3, a
        clr     a
        rlc     a
        mov     r4, a
        sjmp    00011$
00010$:
#endif
        clr     a
        xch     a, r4
        xch     a, r3
        xch     a, r2
        mov     r1, a
00011$:

        // begin long division
        push    exp_a
#ifdef FLOAT_FULL_ACCURACY
        mov     b, #25
#else
        mov     b, #24
#endif
00012$:
        // compare
        clr     c
        mov     a, r1
        subb    a, r5
        mov     a, r2
        subb    a, r6
        mov     a, r3
        subb    a, r7
        mov     a, r4
        subb    a, #0           // carry==0 if mant1 >= mant2

#ifdef FLOAT_FULL_ACCURACY
        djnz    b, 00013$
        sjmp    00015$
00013$:
#endif
        jc      00014$
        // subtract
        mov     a, r1
        subb    a, r5
        mov     r1, a
        mov     a, r2
        subb    a, r6
        mov     r2, a
        mov     a, r3
        subb    a, r7
        mov     r3, a
        mov     a, r4
        subb    a, #0
        mov     r4, a
        clr     c

00014$:
        // shift result
        cpl     c
        mov     a, r0
        rlc     a
        mov     r0, a
        mov     a, dpl
        rlc     a
        mov     dpl, a
        mov     a, dph
        rlc     a
        mov     dph, a

        // shift partial remainder
        clr     c
        mov     a, r1
        rlc     a
        mov     r1, a
        mov     a, r2
        rlc     a
        mov     r2, a
        mov     a, r3
        rlc     a
        mov     r3, a
        mov     a, r4
        rlc     a
        mov     r4, a

#ifdef FLOAT_FULL_ACCURACY
        sjmp    00012$
00015$:
#else
        djnz    b, 00012$
#endif

        // now we've got a division result, so all we need to do
        // is round off properly, normalize and output a float

#ifdef FLOAT_FULL_ACCURACY
        cpl     c
        clr     a
        mov     r1, a
        addc    a, r0
        mov     r2, a
        clr     a
        addc    a, dpl
        mov     r3, a
        clr     a
        addc    a, dph
        mov     r4, a
        pop     exp_a
        jnc     00016$
        inc     exp_a
        // incrementing exp_a without checking carry is dangerous
        mov     r4, #0x80
00016$:
#else
        mov     r1, #0
        mov     a, r0
        mov     r2, a
        mov     r3, dpl
        mov     r4, dph
        pop     exp_a
#endif

        lcall   fs_normalize_a
        ljmp    fs_zerocheck_return
        __endasm;
}

#else

/*
** libgcc support for software floating point.
** Copyright (C) 1991 by Pipeline Associates, Inc.  All rights reserved.
** Permission is granted to do *anything* you want with this file,
** commercial or otherwise, provided this message remains intact.  So there!
** I would appreciate receiving any updates/patches/changes that anyone
** makes, and am willing to be the repository for said changes (am I
** making a big mistake?).
**
** Pat Wood
** Pipeline Associates, Inc.
** pipeline!phw@motown.com or
** sun!pipeline!phw or
** uunet!motown!pipeline!phw
*/


/* (c)2000/2001: hacked a little by johan.knol@iduna.nl for sdcc */

union float_long
  {
    float f;
    long l;
  };

/* divide two floats */
static float __fsdiv_org (float a1, float a2)
{
  volatile union float_long fl1, fl2;
  volatile long result;
  volatile unsigned long mask;
  volatile long mant1, mant2;
  volatile int exp;
  char sign;
 
  fl1.f = a1;
  fl2.f = a2;

  /* subtract exponents */
  exp = EXP (fl1.l) ;
  exp -= EXP (fl2.l);
  exp += EXCESS;

  /* compute sign */
  sign = SIGN (fl1.l) ^ SIGN (fl2.l);

  /* divide by zero??? */
  if (!fl2.l)
    {/* return NaN or -NaN */
      fl2.l = 0x7FC00000;
      return (fl2.f);
    }

  /* numerator zero??? */
  if (!fl1.l)
    return (0);

  /* now get mantissas */
  mant1 = MANT (fl1.l);
  mant2 = MANT (fl2.l);

  /* this assures we have 25 bits of precision in the end */
  if (mant1 < mant2)
    {
      mant1 <<= 1;
      exp--;
    }

  /* now we perform repeated subtraction of fl2.l from fl1.l */
  mask = 0x1000000;
  result = 0;
  while (mask)
    {
      if (mant1 >= mant2)
        {
          result |= mask;
          mant1 -= mant2;
        }
      mant1 <<= 1;
      mask >>= 1;
    }

  /* round */
  result += 1;

  /* normalize down */
  exp++;
  result >>= 1;

  result &= ~HIDDEN;

  /* pack up and go home */
  if (exp >= 0x100)
    fl1.l = (sign ? SIGNBIT : 0) | __INFINITY;
  else if (exp < 0)
    fl1.l = 0;
  else
    fl1.l = PACK (sign ? SIGNBIT : 0 , exp, result);
  return (fl1.f);
}

float __fsdiv (float a1, float a2)
{
  float f;
  unsigned long *p = (unsigned long *) &f;

  if (a2 == 0.0f && a1 > 0.0f)
    *p = 0x7f800000; // inf
  else if (a2 == 0.0f && a1 < 0.0f)
    *p = 0xff800000; // -inf
  else if (a2 == 0.0f && a1 == 0.0f)
    *p = 0xffc00000; // nan
  else
    f = __fsdiv_org (a1, a2);

  return f;
}

#endif