NetBSD/sys/arch/m68k/fpe/fpu_subr.c
briggs 59422960ea PR 7220 from Ken Nakata:
I've fixed most (not all) m68k FPE bugs that give bogus
calculation results, esp. fsqrt instruction.  Also, the internal FP
representation has been reduced from 115-bit mantissa to 67-bit
mantissa which reduced the required mantissa operation roughly by one
fourth.  I've done an extensive (though not exhaustive - it's
impossible!) test on the internal routines by feeding them randomly
generated FP numbers, and found that the new code is more precise than
MC68040 FPU (it seems to have a rounding bug).

[ Only change was to keep fpu_calcea.c's name instead of renaming to
  fpu_ea.c in Ken's patch.  --akb ]
1999-05-30 20:17:48 +00:00

213 lines
5.8 KiB
C

/* $NetBSD: fpu_subr.c,v 1.3 1999/05/30 20:17:49 briggs Exp $ */
/*
* Copyright (c) 1992, 1993
* The Regents of the University of California. All rights reserved.
*
* This software was developed by the Computer Systems Engineering group
* at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
* contributed to Berkeley.
*
* All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the University of
* California, Lawrence Berkeley Laboratory.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)fpu_subr.c 8.1 (Berkeley) 6/11/93
*/
/*
* FPU subroutines.
*/
#include <sys/types.h>
#include <sys/systm.h>
#include <machine/reg.h>
#include "fpu_emulate.h"
#include "fpu_arith.h"
/*
* Shift the given number right rsh bits. Any bits that `fall off' will get
* shoved into the sticky field; we return the resulting sticky. Note that
* shifting NaNs is legal (this will never shift all bits out); a NaN's
* sticky field is ignored anyway.
*/
int
fpu_shr(register struct fpn *fp, register int rsh)
{
register u_int m0, m1, m2, s;
register int lsh;
#ifdef DIAGNOSTIC
if (rsh < 0 || (fp->fp_class != FPC_NUM && !ISNAN(fp)))
panic("fpu_rightshift 1");
#endif
m0 = fp->fp_mant[0];
m1 = fp->fp_mant[1];
m2 = fp->fp_mant[2];
/* If shifting all the bits out, take a shortcut. */
if (rsh >= FP_NMANT) {
#ifdef DIAGNOSTIC
if ((m0 | m1 | m2) == 0)
panic("fpu_rightshift 2");
#endif
fp->fp_mant[0] = 0;
fp->fp_mant[1] = 0;
fp->fp_mant[2] = 0;
#ifdef notdef
if ((m0 | m1 | m2) == 0)
fp->fp_class = FPC_ZERO;
else
#endif
fp->fp_sticky = 1;
return (1);
}
/* Squish out full words. */
s = fp->fp_sticky;
if (rsh >= 32 * 2) {
s |= m2 | m1;
m2 = m0, m1 = 0, m0 = 0;
} else if (rsh >= 32) {
s |= m2;
m2 = m1, m1 = m0, m0 = 0;
}
/* Handle any remaining partial word. */
if ((rsh &= 31) != 0) {
lsh = 32 - rsh;
s |= m2 << lsh;
m2 = (m2 >> rsh) | (m1 << lsh);
m1 = (m1 >> rsh) | (m0 << lsh);
m0 >>= rsh;
}
fp->fp_mant[0] = m0;
fp->fp_mant[1] = m1;
fp->fp_mant[2] = m2;
fp->fp_sticky = s;
return (s);
}
/*
* Force a number to be normal, i.e., make its fraction have all zero
* bits before FP_1, then FP_1, then all 1 bits. This is used for denorms
* and (sometimes) for intermediate results.
*
* Internally, this may use a `supernormal' -- a number whose fp_mant
* is greater than or equal to 2.0 -- so as a side effect you can hand it
* a supernormal and it will fix it (provided fp->fp_mant[2] == 0).
*/
void
fpu_norm(register struct fpn *fp)
{
register u_int m0, m1, m2, sup, nrm;
register int lsh, rsh, exp;
exp = fp->fp_exp;
m0 = fp->fp_mant[0];
m1 = fp->fp_mant[1];
m2 = fp->fp_mant[2];
/* Handle severe subnormals with 32-bit moves. */
if (m0 == 0) {
if (m1) {
m0 = m1;
m1 = m2;
m2 = 0;
exp -= 32;
} else if (m2) {
m0 = m2;
m1 = 0;
m2 = 0;
exp -= 2 * 32;
} else {
fp->fp_class = FPC_ZERO;
return;
}
}
/* Now fix any supernormal or remaining subnormal. */
nrm = FP_1;
sup = nrm << 1;
if (m0 >= sup) {
/*
* We have a supernormal number. We need to shift it right.
* We may assume m2==0.
*/
asm volatile("bfffo %1{#0:#32},%0" : "=d"(rsh) : "g"(m0));
rsh = 31 - rsh - FP_LG;
exp += rsh;
lsh = 32 - rsh;
m2 = m1 << lsh;
m1 = (m1 >> rsh) | (m0 << lsh);
m0 = (m0 >> rsh);
} else if (m0 < nrm) {
/*
* We have a regular denorm (a subnormal number), and need
* to shift it left.
*/
asm volatile("bfffo %1{#0:#32},%0" : "=d"(lsh) : "g"(m0));
lsh = FP_LG - 31 + lsh;
exp -= lsh;
rsh = 32 - lsh;
m0 = (m0 << lsh) | (m1 >> rsh);
m1 = (m1 << lsh) | (m2 >> rsh);
m2 <<= lsh;
}
fp->fp_exp = exp;
fp->fp_mant[0] = m0;
fp->fp_mant[1] = m1;
fp->fp_mant[2] = m2;
}
/*
* Concoct a `fresh' Quiet NaN per Appendix N.
* As a side effect, we set OPERR for the current exceptions.
*/
struct fpn *
fpu_newnan(register struct fpemu *fe)
{
register struct fpn *fp;
fe->fe_fpsr |= FPSR_OPERR;
fp = &fe->fe_f3;
fp->fp_class = FPC_QNAN;
fp->fp_sign = 0;
fp->fp_mant[0] = FP_1 - 1;
fp->fp_mant[1] = fp->fp_mant[2] = ~0;
return (fp);
}