2006-01-28 00:15:15 +00:00
|
|
|
/* nearbyint long double.
|
|
|
|
IBM extended format long double version.
|
|
|
|
Copyright (C) 2004, 2006 Free Software Foundation, Inc.
|
|
|
|
This file is part of the GNU C Library.
|
|
|
|
|
|
|
|
The GNU C Library is free software; you can redistribute it and/or
|
|
|
|
modify it under the terms of the GNU Lesser General Public
|
|
|
|
License as published by the Free Software Foundation; either
|
|
|
|
version 2.1 of the License, or (at your option) any later version.
|
|
|
|
|
|
|
|
The GNU C 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
|
|
|
|
Lesser General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU Lesser General Public
|
2012-02-09 23:18:22 +00:00
|
|
|
License along with the GNU C Library; if not, see
|
|
|
|
<http://www.gnu.org/licenses/>. */
|
2006-01-28 00:15:15 +00:00
|
|
|
|
|
|
|
#include <sysdep.h>
|
|
|
|
#include <math_ldbl_opt.h>
|
|
|
|
|
|
|
|
.section ".toc","aw"
|
|
|
|
.LC0: /* 2**52 */
|
|
|
|
.tc FD_43300000_0[TC],0x4330000000000000
|
|
|
|
.section ".text"
|
|
|
|
|
|
|
|
/* long double [fp1,fp2] nearbyintl (long double x [fp1,fp2])
|
|
|
|
IEEE 1003.1 nearbyintl function. nearbyintl is simular to the rintl
|
|
|
|
but does raise the "inexact" exception. This implementation is
|
|
|
|
based on rintl but explicitly maskes the inexact exception on entry
|
|
|
|
and clears any pending inexact before restoring the exception mask
|
|
|
|
on exit.
|
|
|
|
|
|
|
|
PowerPC64 long double uses the IBM extended format which is
|
|
|
|
represented two 64-floating point double values. The values are
|
|
|
|
non-overlapping giving an effective precision of 106 bits. The first
|
|
|
|
double contains the high order bits of mantisa and is always rounded
|
|
|
|
to represent a normal rounding of long double to double. Since the
|
|
|
|
long double value is sum of the high and low values, the low double
|
|
|
|
normally has the opposite sign to compensate for the this rounding.
|
|
|
|
|
|
|
|
For long double there are two cases:
|
|
|
|
1) |x| < 2**52, all the integer bits are in the high double.
|
|
|
|
floor the high double and set the low double to -0.0.
|
|
|
|
2) |x| >= 2**52, Rounding involves both doubles.
|
|
|
|
See the comment before lable .L2 for details.
|
|
|
|
*/
|
|
|
|
ENTRY (__nearbyintl)
|
|
|
|
mffs fp11 /* Save current FPSCR. */
|
|
|
|
lfd fp13,.LC0@toc(2)
|
|
|
|
fabs fp0,fp1
|
|
|
|
mtfsb0 28 /* Disable "inexact" exceptions. */
|
|
|
|
fsub fp12,fp13,fp13 /* generate 0.0 */
|
|
|
|
fabs fp9,fp2
|
|
|
|
fcmpu cr7,fp0,fp13 /* if (fabs(x) > TWO52) */
|
|
|
|
fcmpu cr6,fp1,fp12 /* if (x > 0.0) */
|
|
|
|
bnl- cr7,.L2
|
|
|
|
fmr fp2,fp12
|
|
|
|
bng- cr6,.L4
|
|
|
|
fadd fp1,fp1,fp13 /* x+= TWO52; */
|
|
|
|
fsub fp1,fp1,fp13 /* x-= TWO52; */
|
|
|
|
b .L9
|
|
|
|
.L4:
|
|
|
|
bnl- cr6,.L9 /* if (x < 0.0) */
|
|
|
|
fsub fp1,fp13,fp1 /* x = TWO52 - x; */
|
|
|
|
fsub fp0,fp1,fp13 /* x = - (x - TWO52); */
|
|
|
|
fneg fp1,fp0
|
|
|
|
.L9:
|
|
|
|
mtfsb0 6 /* Clear any pending "inexact" exceptions. */
|
|
|
|
mtfsf 0x01,fp11 /* restore exception mask. */
|
|
|
|
blr
|
|
|
|
|
|
|
|
/* The high double is > TWO52 so we need to round the low double and
|
|
|
|
perhaps the high double. This gets a bit tricky so we use the
|
|
|
|
following algorithm:
|
|
|
|
|
|
|
|
tau = floor(x_high/TWO52);
|
|
|
|
x0 = x_high - tau;
|
|
|
|
x1 = x_low + tau;
|
|
|
|
r1 = nearbyint(x1);
|
|
|
|
y_high = x0 + r1;
|
|
|
|
y_low = r1 - tau;
|
|
|
|
return y; */
|
|
|
|
.L2:
|
|
|
|
fcmpu cr7,fp9,fp13 /* if (|x_low| > TWO52) */
|
|
|
|
fcmpu cr0,fp9,fp12 /* || (|x_low| == 0.0) */
|
|
|
|
bge- cr7,.L9 /* return x; */
|
|
|
|
beq- cr0,.L9
|
|
|
|
fdiv fp8,fp1,fp13 /* x_high/TWO52 */
|
|
|
|
fctidz fp0,fp8
|
|
|
|
fcfid fp8,fp0 /* tau = floor(x_high/TWO52); */
|
|
|
|
fsub fp3,fp1,fp8 /* x0 = x_high - tau; */
|
|
|
|
fadd fp4,fp2,fp8 /* x1 = x_low + tau; */
|
|
|
|
|
|
|
|
fcmpu cr6,fp4,fp12 /* if (x1 > 0.0) */
|
|
|
|
bng- cr6,.L8
|
|
|
|
fadd fp5,fp4,fp13 /* r1 = x1 + TWO52; */
|
|
|
|
fsub fp5,fp5,fp13 /* r1 = r1 - TWO52; */
|
|
|
|
b .L6
|
|
|
|
.L8:
|
|
|
|
fmr fp5,fp4
|
|
|
|
bge- cr6,.L6 /* if (x1 < 0.0) */
|
|
|
|
fsub fp5,fp13,fp4 /* r1 = TWO52 - x1; */
|
|
|
|
fsub fp0,fp5,fp13 /* r1 = - (r1 - TWO52); */
|
|
|
|
fneg fp5,fp0
|
|
|
|
.L6:
|
|
|
|
fadd fp1,fp3,fp5 /* y_high = x0 + r1; */
|
|
|
|
fsub fp2,fp5,fp8 /* y_low = r1 - tau; */
|
|
|
|
b .L9
|
|
|
|
END (__nearbyintl)
|
|
|
|
|
|
|
|
long_double_symbol (libm, __nearbyintl, nearbyintl)
|