glibc/sysdeps/powerpc/w_sqrtf.c
Ulrich Drepper 118bad87bd Update.
1998-12-31  Ulrich Drepper  <drepper@cygnus.com>

	* iconvdata/Makefile: Use rpath $ORIGIN for modules which use any of
	the conversion libraries.

	* include/features.h: Mention latest extensions in _POSIX_C_SOURCE
	description.

	* libio/genops.c (list_all_lock): New variable.
	(_IO_un_link, _IO_link_in): Acquire lock before modifying _IO_list_all.
	[PR libc/911].

	* sysdeps/i386/i686/strtok.s: Add missing cld.

	* sysdeps/i386/i686/mempcpy.S: New file.

1998-12-30  Geoff Keating  <geoffk@ozemail.com.au>

	Suppress parentheses warnings:
	* sysdeps/unix/sysv/linux/powerpc/dl-sysdep.c
	(DL_FIND_ARG_COMPONENTS): here,
	* sysdeps/powerpc/dl-machine.c: here,
	* sysdeps/powerpc/fclrexcpt.c (feclearexcept): here,
	* sysdeps/powerpc/fesetround.c (fesetround): here,
	* sysdeps/powerpc/feupdateenv.c (feupdateenv): here,
	* sysdeps/powerpc/fraiseexcpt.c (feraiseexcept): here,
	* sysdeps/powerpc/fsetexcptflg.c (fesetexceptflag): here,
	* sysdeps/powerpc/s_rint.c (__rint): here,
	* sysdeps/powerpc/s_rintf.c (__rintf): here,
	* sysdeps/powerpc/w_sqrt.c (__sqrt): here,
	* sysdeps/powerpc/w_sqrtf.c (__sqrtf): and here.

1998-12-30  Geoff Keating  <geoffk@ozemail.com.au>

	* wcsmbs/wcstold.c [__NO_LONG_DOUBLE_MATH]: Include wchar.h.
	* wcsmbs/wcstold_l.c [__NO_LONG_DOUBLE_MATH]: Likewise.  Also define
	appropropriate prototypes, correct procedure names.

1998-12-30  Geoff Keating  <geoffk@ozemail.com.au>

	* sysdeps/generic/crypt-entry.c: Prototype __crypt_r, suppress
	warning.
	(__crypt_r): Add __restrict.
	* sysdeps/generic/crypt.h: Remove internal routine.  Add __restrict.

1998-12-31  Ulrich Drepper  <drepper@cygnus.com>

	* stdlib/longlong.h: Add missing #else in SPARC definitions.

1998-12-30  Andreas Jaeger  <aj@arthur.rhein-neckar.de>

	Patches by Ralf Baechle <ralf@gnu.org>:
	* sysdeps/unix/sysv/linux/mips/sgidefs.h: Removed.
	* sysdeps/unix/sysv/linux/mips/sys/asm.h:  Removed.
	* sysdeps/unix/sysv/linux/mips/sys/regdef.h:  Removed.
	* sysdeps/unix/sysv/linux/mips/sys/fpregdef.h:  Removed.
	* sysdeps/unix/sysv/linux/mips/regdef.h:  Removed.
	* sysdeps/unix/sysv/linux/mips/fpregdef.h: Removed.

	* sysdeps/unix/sysv/linux/mips/Makefile: Remove delete files.
	* sysdeps/unix/sysv/linux/mips/Dist: Remove deleted files.

	* sysdeps/mips/sys/fpregdef.h: New files, enhanced versions of deleted
	linux specific files.
	* sysdeps/mips/sys/asm.h: Likewise.
	* sysdeps/mips/sgidefs.h: Likewise.
	* sysdeps/mips/fpregdef.h: Likewise.
	* sysdeps/mips/regdef.h: Likewise.

	* sysdeps/mips/Makefile: Add new headers.
	* sysdeps/mips/Dist: Add new files.

	* sysdeps/mips/fpu_control.h (_FPU_DEFAULT): Change value.

1998-12-28  Andreas Schwab  <schwab@issan.cs.uni-dortmund.de>

	* iconvdata/Makefile: Remove variable assignments and rules that
	are now auto-generated.  Include $(objpfx)iconv-rules instead.
	Btw, that removes a few typos.
	(charmaps): New variable.
	($(objpfx)iconv-rules): New target.
	(gen-8bit-modules): Renamed from sed-generated-headers, remove .h
	suffixes.
	(gen-8bit-gap-modules): Renamed from awk-generated-headers, remove
	.h suffixes.
	(gen-special-modules): New variable.
	(generated-modules): New variable.
	(headers): Use it.
	(before-compile): Likewise.
	(generated): Likewise, and add iconv-rules.
	* iconvdata/euc-cn.c: Renamed from euccn.c.
	* iconvdata/euc-jp.c: Renamed from eucjp.c.
	* iconvdata/euc-kr.c: Renamed from euckr.c.
	* iconvdata/euc-tw.c: Renamed from euctw.c.
	* iconvdata/iso_6937.c: Renamed from iso6937.c.
	* iconvdata/iso_6937-2.c: Renamed from iso6937-2.c.
	* iconvdata/t.61.c: Renamed from t61.c.
	* iconvdata/Makefile (distribute): Adjusted for those renames.
	Remove $(objpfx) from names.

1998-12-28  Andreas Schwab  <schwab@issan.cs.uni-dortmund.de>

	* sysdeps/generic/hp-timing.h: Fix comment.
	* sysdeps/generic/dl-hash.h: Likewise.
	* sysdeps/i386/i686/hp-timing.h: Likewise.

1998-12-28  Andreas Schwab  <schwab@issan.cs.uni-dortmund.de>

	* posix/fnmatch.c (fnmatch): Always pass unsigned char values to
	FOLD macro.

1998-12-28  Andreas Schwab  <schwab@issan.cs.uni-dortmund.de>

	* nss/db-Makefile ($(VAR_DB)/shadow.db): Fix last change.

1998-12-29  Andreas Jaeger  <aj@arthur.rhein-neckar.de>

	* sysdeps/unix/sysv/linux/mips/bits/sigaction.h: Sync with
	current Linux 2.1.132 kernel sources.

1998-12-28  Andreas Jaeger  <aj@arthur.rhein-neckar.de>

	Patches by Ralf Baechle <ralf@gnu.org> for Linux/MIPS:
	* sysdeps/unix/sysv/linux/mips/syscalls.list: Add ipc, change
	pread/pwrite and llseek calls.

	* sysdeps/unix/sysv/linux/mips/sysdep.S: New file.

1998-12-29  Wolfram Gloger  <wmglo@dent.med.uni-muenchen.de>

	* malloc/thread-m.h [_LIBC] : Fix for NO_THREADS case.
1998-12-31 18:19:19 +00:00

137 lines
4.9 KiB
C

/* Single-precision floating point square root.
Copyright (C) 1997 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 Library General Public License as
published by the Free Software Foundation; either version 2 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
Library General Public License for more details.
You should have received a copy of the GNU Library General Public
License along with the GNU C Library; see the file COPYING.LIB. If not,
write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA. */
#include <math.h>
#include <math_private.h>
#include <fenv_libc.h>
#include <inttypes.h>
static const float almost_half = 0.50000006; /* 0.5 + 2^-24 */
static const uint32_t a_nan = 0x7fc00000;
static const uint32_t a_inf = 0x7f800000;
static const float two48 = 281474976710656.0;
static const float twom24 = 5.9604644775390625e-8;
extern const float __t_sqrt[1024];
/* The method is based on a description in
Computation of elementary functions on the IBM RISC System/6000 processor,
P. W. Markstein, IBM J. Res. Develop, 34(1) 1990.
Basically, it consists of two interleaved Newton-Rhapson approximations,
one to find the actual square root, and one to find its reciprocal
without the expense of a division operation. The tricky bit here
is the use of the POWER/PowerPC multiply-add operation to get the
required accuracy with high speed.
The argument reduction works by a combination of table lookup to
obtain the initial guesses, and some careful modification of the
generated guesses (which mostly runs on the integer unit, while the
Newton-Rhapson is running on the FPU). */
float
__sqrtf(float x)
{
const float inf = *(const float *)&a_inf;
/* x = f_washf(x); *//* This ensures only one exception for SNaN. */
if (x > 0)
{
if (x != inf)
{
/* Variables named starting with 's' exist in the
argument-reduced space, so that 2 > sx >= 0.5,
1.41... > sg >= 0.70.., 0.70.. >= sy > 0.35... .
Variables named ending with 'i' are integer versions of
floating-point values. */
float sx; /* The value of which we're trying to find the square
root. */
float sg,g; /* Guess of the square root of x. */
float sd,d; /* Difference between the square of the guess and x. */
float sy; /* Estimate of 1/2g (overestimated by 1ulp). */
float sy2; /* 2*sy */
float e; /* Difference between y*g and 1/2 (note that e==se). */
float shx; /* == sx * fsg */
float fsg; /* sg*fsg == g. */
fenv_t fe; /* Saved floating-point environment (stores rounding
mode and whether the inexact exception is
enabled). */
uint32_t xi, sxi, fsgi;
const float *t_sqrt;
GET_FLOAT_WORD (xi, x);
fe = fegetenv_register ();
relax_fenv_state ();
sxi = (xi & 0x3fffffff) | 0x3f000000;
SET_FLOAT_WORD (sx, sxi);
t_sqrt = __t_sqrt + (xi >> (23-8-1) & 0x3fe);
sg = t_sqrt[0];
sy = t_sqrt[1];
/* Here we have three Newton-Rhapson iterations each of a
division and a square root and the remainder of the
argument reduction, all interleaved. */
sd = -(sg*sg - sx);
fsgi = (xi + 0x40000000) >> 1 & 0x7f800000;
sy2 = sy + sy;
sg = sy*sd + sg; /* 16-bit approximation to sqrt(sx). */
e = -(sy*sg - almost_half);
SET_FLOAT_WORD (fsg, fsgi);
sd = -(sg*sg - sx);
sy = sy + e*sy2;
if ((xi & 0x7f800000) == 0)
goto denorm;
shx = sx * fsg;
sg = sg + sy*sd; /* 32-bit approximation to sqrt(sx),
but perhaps rounded incorrectly. */
sy2 = sy + sy;
g = sg * fsg;
e = -(sy*sg - almost_half);
d = -(g*sg - shx);
sy = sy + e*sy2;
fesetenv_register (fe);
return g + sy*d;
denorm:
/* For denormalised numbers, we normalise, calculate the
square root, and return an adjusted result. */
fesetenv_register (fe);
return __sqrtf(x * two48) * twom24;
}
}
else if (x < 0)
{
#ifdef FE_INVALID_SQRT
feraiseexcept (FE_INVALID_SQRT);
/* For some reason, some PowerPC processors don't implement
FE_INVALID_SQRT. I guess no-one ever thought they'd be
used for square roots... :-) */
if (!fetestexcept (FE_INVALID))
#endif
feraiseexcept (FE_INVALID);
#ifndef _IEEE_LIBM
if (_LIB_VERSION != _IEEE_)
x = __kernel_standard(x,x,126);
else
#endif
x = *(const float*)&a_nan;
}
return f_washf(x);
}
weak_alias (__sqrtf, sqrtf)
/* Strictly, this is wrong, but the only places where _ieee754_sqrt is
used will not pass in a negative result. */
strong_alias(__sqrtf,__ieee754_sqrtf)