glibc/stdlib/strtod.c
Ulrich Drepper 377a515b4c update from main archive 970304
1997-03-04 04:31  Ulrich Drepper  <drepper@cygnus.com>

	* Makerules: Add rules to handle versioning.
	* config.h.in (DO_VERSIONING): New macro.
	* config.make.in (versioning): New variable.
	* configure.in: Add checks for .symver directive in gas and
	--version-script option to ld.  Define DO_VERSIONING and
	versioning if appropriate.

	* math/Makefile (routines): Add s_signbit, s_fpclassify, s_fmax,
	s_fmin, and s_fdim.
	* math/math.h: Define ISO C 9X constants, macros and functions.
	* math/mathcalls.h: Likewise.
	* sysdeps/libm-ieee754/s_fdim.c: New file.
	* sysdeps/libm-ieee754/s_fdimf.c: New file.
	* sysdeps/libm-ieee754/s_fdiml.c: New file.
	* sysdeps/libm-ieee754/s_fmax.c: New file.
	* sysdeps/libm-ieee754/s_fmaxf.c: New file.
	* sysdeps/libm-ieee754/s_fmaxl.c: New file.
	* sysdeps/libm-ieee754/s_fmin.c: New file.
	* sysdeps/libm-ieee754/s_fminf.c: New file.
	* sysdeps/libm-ieee754/s_fminl.c: New file.
	* sysdeps/libm-ieee754/s_fpclassify.c: New file.
	* sysdeps/libm-ieee754/s_fpclassifyf.c: New file.
	* sysdeps/libm-ieee754/s_fpclassifyl.c: New file.
	* sysdeps/libm-ieee754/s_signbit.c: New file.
	* sysdeps/libm-ieee754/s_signbitf.c: New file.
	* sysdeps/libm-ieee754/s_signbitl.c: New file.

	* stdio-common/printf_fphex.c: Correct printing of decimal point
	character.
	Simplify conversion of mantissa to string.
	* stdio-common/vfscanf.c: Handle %A format.
	Optimize termination of floating-point scanning.
	* stdio-common/tstscanf.c (main): Add new test to scanf to test
	scanning float values with given width.
	* stdlib/strtod.c: Add handling of floating-point numbers in
	hexadecimal notation.

	* stdlib/stdlib.h: Use __USE_ISOC9X feature macro for new long long
	functions.
	Pretty print #if directives.
	* string/string.h: Pretty print #if directives.

	* sysdeps/ieee754/dbl2mpn.c: Update copyright.
	* sysdeps/ieee754/ldbl2mpn.c: Likewise.
	* sysdeps/ieee754/mpn2dbl.c: Likewise.
	* sysdeps/ieee754/mpn2flt.c: Likewise.
	* sysdeps/ieee754/mpn2ldbl.c: Likewise.

	* sysdeps/unix/sysv/linux/poll.c: Implement poll function by
	falling back to select-based implementation if syscall isn't
	available.
	* sysdeps/unix/sysv/linux/syscalls.list: Add s_poll.

	* time/leapseconds: Update from tzdata1997b.
	* time/zic.c: Update from tzcode1997b.

1997-03-01 15:08  Andreas Schwab  <schwab@issan.informatik.uni-dortmund.de>

	* time/Makefile $(tzfiles:%=$(objpfx)z.%): Remove unneeded
	depedencies between installed $(tzlinks) and $(tzbases) files.

1997-03-01 14:27  Andreas Schwab  <schwab@issan.informatik.uni-dortmund.de>

	* math/math.h: Make compatible with traditional preprocessor;
	requires carefull placement of whitespace in macro arguments.
	Use __CONCAT instead of ##.
	Declare long double functions only if __STDC__ or __GNUC__.
	* math/mathcall.h: Avoid whitespace before argument of macro call
	that is used as function name.

	* sysdeps/m68k/fpu/__math.h: Use __CONCAT instead of ##.
	(__m81_u, __m81_inline): Depend on __LIBC_M81_MATH_INLINES instead
	of __NO_M81_MATH_INLINES.
	[!__LIBC_M81_MATH_INLINES]: Don't define internal functions
	starting with __ieee754.
	[!__NO_MATH_INLINES && __OPTIMIZE__]: Define user visible
	functions as inlines.
	(__m81_defun): Put __attribute__ between return type and function
	name.
	* math/math.h: Include <__math.h> also if __LIBC_M81_MATH_INLINES
	is defined.
	* sysdeps/m68k/fpu/e_acos.c: Define __LIBC_M81_MATH_INLINES
	instead of __NO_M81_MATH_INLINES.
	* sysdeps/m68k/fpu/e_fmod.c: Likewise.
	* sysdeps/m68k/fpu/k_cos.c: Likewise.
	* sysdeps/m68k/fpu/k_sin.c: Likewise.
	* sysdeps/m68k/fpu/k_tan.c: Likewise.
	* sysdeps/m68k/fpu/s_atan.c: Likewise. De-ANSI-declify.
	* sysdeps/m68k/fpu/s_frexp.c: Likewise.
	* sysdeps/m68k/fpu/s_ilogb.c: Likewise.
	* sysdeps/m68k/fpu/s_isinf.c: Likewise.
	* sysdeps/m68k/fpu/s_modf.c: Likewise.
	* sysdeps/m68k/fpu/s_scalbn.c: Likewise.

1997-02-27 21:51  Andreas Schwab  <schwab@issan.informatik.uni-dortmund.de>

	* Makefile (tests): Cope with $PATH not including the current
	directory.

1997-02-27 18:04  Andreas Schwab  <schwab@issan.informatik.uni-dortmund.de>

	* sysdeps/unix/Makefile ($(common-objpfx)mk-local_lim): Use
	$(common-objdir-compile).
	($(common-objpfx)make-ioctls): Likewise.
	(mk-local_lim-CFLAGS): Set this instead of local_lim-CFLAGS.
	($(common-objpfx)sys/param.h): Use $(make-target-directory).
	($(addprefix $(common-objpfx),$(sys/param.h-includes))):
	Likewise.
	($(common-objpfx)sys/syscall.h): Likewise.
	($(common-objpfx)local_lim.h): Let make deal with command
	failure.
	($(common-objpfx)param.h.dep): Use temporary file and update
	target atomically.
	($(common-objpfx)errnos): Avoid the Useless Use of cat Award.
	(include $(common-objpfx)param.h.dep): Ignore error.

	* sysdeps/posix/Makefile ($(common-objpfx)mk-stdiolim): Use
	$(common-objdir-compile).
	(mk-stdiolim-CFLAGS): Renamed from cded-objdir-includes, use
	$(shell pwd) instead of $$cwd.

	* sysdeps/generic/Makefile ($(common-objpfx)det_endian): Use
	$(common-objdir-compile).
	($(objpfx)make_siglist): Use $(native-compile).
	(make_siglist-CFLAGS): New variable.

	* Makerules (ALL_BUILD_CFLAGS): Renamed from BUILD_CFLAGS, leaving
	the old name for the user to pass additional flags to the host
	compiler.  Fix reference to config header.
	(native-compile, common-objdir-compile): Rewritten to make more
	generally usable.
	* sysdeps/unix/sysv/sysv4/solaris2/Makefile: Set ALL_BUILD_CFLAGS
	instead of BUILD_CFLAGS.

	* sysvips/sys/ipc.h: Warn if needed feature select macro are not
	defined.

1997-02-27 17:11  Andreas Schwab  <schwab@issan.informatik.uni-dortmund.de>

	* sunrpc/Makefile ($(objpfx)rpc-proto.d, $(objpfx)rpc-proto.c):
	New rules to generate dependencies for the RPC service objects.

1997-02-27 16:26  Andreas Schwab  <schwab@issan.informatik.uni-dortmund.de>

	* argp/argp-parse.c (parser_finalize): Always set *END_INDEX if
	supplied.

1997-02-28 03:27  Ulrich Drepper  <drepper@cygnus.com>

	* stdlib/strtod.c (STRTOF): Make sure return value is large enough
	so that clearing second word is necessary.

	* sysdeps/unix/sysv/linux/netinet/in_systm.h: Don't use kernel
	header since it is wrong for 64 bit systems.
	Patch by a sun <asun@zoology.washington.edu>.

1997-02-27 10:34:11  Richard Henderson  <rth@tamu.edu>

	* sysdeps/unix/sysv/linux/alpha/brk.S: Support both the Linux/i386
	and OSF/1 style brk syscalls.  We may want to change Linux/Alpha
	for the benefit of running foreign binaries.

1997-03-01 20:21  Miles Bader  <miles@gnu.ai.mit.edu>
1997-02-25 19:42  Miles Bader  <miles@gnu.ai.mit.edu>
1997-03-04 05:53:28 +00:00

1338 lines
37 KiB
C
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/* Read decimal floating point numbers.
This file is part of the GNU C Library.
Copyright (C) 1995, 1996, 1997 Free Software Foundation, Inc.
Contributed by Ulrich Drepper <drepper@gnu.ai.mit.edu>, 1995.
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. */
/* Configuration part. These macros are defined by `strtold.c',
`strtof.c', `wcstod.c', `wcstold.c', and `wcstof.c' to produce the
`long double' and `float' versions of the reader. */
#ifndef FLOAT
# define FLOAT double
# define FLT DBL
# ifdef USE_WIDE_CHAR
# define STRTOF wcstod
# else
# define STRTOF strtod
# endif
# define MPN2FLOAT __mpn_construct_double
# define FLOAT_HUGE_VAL HUGE_VAL
#endif
#ifdef USE_WIDE_CHAR
# include <wctype.h>
# include <wchar.h>
# define STRING_TYPE wchar_t
# define CHAR_TYPE wint_t
# define L_(Ch) L##Ch
# define ISSPACE(Ch) iswspace (Ch)
# define ISDIGIT(Ch) iswdigit (Ch)
# define ISXDIGIT(Ch) iswxdigit (Ch)
# define TOLOWER(Ch) towlower (Ch)
# define STRNCASECMP(S1, S2, N) __wcsncasecmp ((S1), (S2), (N))
#else
# define STRING_TYPE char
# define CHAR_TYPE char
# define L_(Ch) Ch
# define ISSPACE(Ch) isspace (Ch)
# define ISDIGIT(Ch) isdigit (Ch)
# define ISXDIGIT(Ch) isxdigit (Ch)
# define TOLOWER(Ch) tolower (Ch)
# define STRNCASECMP(S1, S2, N) __strncasecmp ((S1), (S2), (N))
#endif
/* End of configuration part. */
#include <ctype.h>
#include <errno.h>
#include <float.h>
#include <ieee754.h>
#include "../locale/localeinfo.h"
#include <math.h>
#include <stdlib.h>
#include <string.h>
/* The gmp headers need some configuration frobs. */
#define HAVE_ALLOCA 1
#include "gmp.h"
#include "gmp-impl.h"
#include <gmp-mparam.h>
#include "longlong.h"
#include "fpioconst.h"
#define NDEBUG 1
#include <assert.h>
/* Constants we need from float.h; select the set for the FLOAT precision. */
#define MANT_DIG PASTE(FLT,_MANT_DIG)
#define DIG PASTE(FLT,_DIG)
#define MAX_EXP PASTE(FLT,_MAX_EXP)
#define MIN_EXP PASTE(FLT,_MIN_EXP)
#define MAX_10_EXP PASTE(FLT,_MAX_10_EXP)
#define MIN_10_EXP PASTE(FLT,_MIN_10_EXP)
/* Extra macros required to get FLT expanded before the pasting. */
#define PASTE(a,b) PASTE1(a,b)
#define PASTE1(a,b) a##b
/* Function to construct a floating point number from an MP integer
containing the fraction bits, a base 2 exponent, and a sign flag. */
extern FLOAT MPN2FLOAT (mp_srcptr mpn, int exponent, int negative);
/* Definitions according to limb size used. */
#if BITS_PER_MP_LIMB == 32
# define MAX_DIG_PER_LIMB 9
# define MAX_FAC_PER_LIMB 1000000000UL
#elif BITS_PER_MP_LIMB == 64
# define MAX_DIG_PER_LIMB 19
# define MAX_FAC_PER_LIMB 10000000000000000000UL
#else
# error "mp_limb_t size " BITS_PER_MP_LIMB "not accounted for"
#endif
/* Local data structure. */
static const mp_limb_t _tens_in_limb[MAX_DIG_PER_LIMB + 1] =
{ 0, 10, 100,
1000, 10000, 100000,
1000000, 10000000, 100000000,
1000000000
#if BITS_PER_MP_LIMB > 32
, 10000000000, 100000000000,
1000000000000, 10000000000000, 100000000000000,
1000000000000000, 10000000000000000, 100000000000000000,
1000000000000000000, 10000000000000000000U
#endif
#if BITS_PER_MP_LIMB > 64
#error "Need to expand tens_in_limb table to" MAX_DIG_PER_LIMB
#endif
};
#ifndef howmany
#define howmany(x,y) (((x)+((y)-1))/(y))
#endif
#define SWAP(x, y) ({ typeof(x) _tmp = x; x = y; y = _tmp; })
#define NDIG (MAX_10_EXP - MIN_10_EXP + 2 * MANT_DIG)
#define HEXNDIG ((MAX_EXP - MIN_EXP + 7) / 8 + 2 * MANT_DIG)
#define RETURN_LIMB_SIZE howmany (MANT_DIG, BITS_PER_MP_LIMB)
#define RETURN(val,end) \
do { if (endptr != NULL) *endptr = (STRING_TYPE *) (end); \
return val; } while (0)
/* Maximum size necessary for mpn integers to hold floating point numbers. */
#define MPNSIZE (howmany (MAX_EXP + 2 * MANT_DIG, BITS_PER_MP_LIMB) \
+ 2)
/* Declare an mpn integer variable that big. */
#define MPN_VAR(name) mp_limb_t name[MPNSIZE]; mp_size_t name##size
/* Copy an mpn integer value. */
#define MPN_ASSIGN(dst, src) \
memcpy (dst, src, (dst##size = src##size) * sizeof (mp_limb_t))
/* Return a floating point number of the needed type according to the given
multi-precision number after possible rounding. */
static inline FLOAT
round_and_return (mp_limb_t *retval, int exponent, int negative,
mp_limb_t round_limb, mp_size_t round_bit, int more_bits)
{
if (exponent < MIN_EXP - 1)
{
mp_size_t shift = MIN_EXP - 1 - exponent;
if (shift > MANT_DIG)
{
__set_errno (EDOM);
return 0.0;
}
more_bits |= (round_limb & ((((mp_limb_t) 1) << round_bit) - 1)) != 0;
if (shift == MANT_DIG)
/* This is a special case to handle the very seldom case where
the mantissa will be empty after the shift. */
{
int i;
round_limb = retval[RETURN_LIMB_SIZE - 1];
round_bit = BITS_PER_MP_LIMB - 1;
for (i = 0; i < RETURN_LIMB_SIZE; ++i)
more_bits |= retval[i] != 0;
MPN_ZERO (retval, RETURN_LIMB_SIZE);
}
else if (shift >= BITS_PER_MP_LIMB)
{
int i;
round_limb = retval[(shift - 1) / BITS_PER_MP_LIMB];
round_bit = (shift - 1) % BITS_PER_MP_LIMB;
for (i = 0; i < (shift - 1) / BITS_PER_MP_LIMB; ++i)
more_bits |= retval[i] != 0;
more_bits |= ((round_limb & ((((mp_limb_t) 1) << round_bit) - 1))
!= 0);
(void) __mpn_rshift (retval, &retval[shift / BITS_PER_MP_LIMB],
RETURN_LIMB_SIZE - (shift / BITS_PER_MP_LIMB),
shift % BITS_PER_MP_LIMB);
MPN_ZERO (&retval[RETURN_LIMB_SIZE - (shift / BITS_PER_MP_LIMB)],
shift / BITS_PER_MP_LIMB);
}
else if (shift > 0)
{
round_limb = retval[0];
round_bit = shift - 1;
(void) __mpn_rshift (retval, retval, RETURN_LIMB_SIZE, shift);
}
exponent = MIN_EXP - 2;
}
if ((round_limb & (((mp_limb_t) 1) << round_bit)) != 0
&& (more_bits || (retval[0] & 1) != 0
|| (round_limb & ((((mp_limb_t) 1) << round_bit) - 1)) != 0))
{
mp_limb_t cy = __mpn_add_1 (retval, retval, RETURN_LIMB_SIZE, 1);
if (((MANT_DIG % BITS_PER_MP_LIMB) == 0 && cy) ||
((MANT_DIG % BITS_PER_MP_LIMB) != 0 &&
(retval[RETURN_LIMB_SIZE - 1]
& (((mp_limb_t) 1) << (MANT_DIG % BITS_PER_MP_LIMB))) != 0))
{
++exponent;
(void) __mpn_rshift (retval, retval, RETURN_LIMB_SIZE, 1);
retval[RETURN_LIMB_SIZE - 1]
|= ((mp_limb_t) 1) << ((MANT_DIG - 1) % BITS_PER_MP_LIMB);
}
else if (exponent == MIN_EXP - 2
&& (retval[RETURN_LIMB_SIZE - 1]
& (((mp_limb_t) 1) << ((MANT_DIG - 1) % BITS_PER_MP_LIMB)))
!= 0)
/* The number was denormalized but now normalized. */
exponent = MIN_EXP - 1;
}
if (exponent > MAX_EXP)
return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
return MPN2FLOAT (retval, exponent, negative);
}
/* Read a multi-precision integer starting at STR with exactly DIGCNT digits
into N. Return the size of the number limbs in NSIZE at the first
character od the string that is not part of the integer as the function
value. If the EXPONENT is small enough to be taken as an additional
factor for the resulting number (see code) multiply by it. */
static inline const STRING_TYPE *
str_to_mpn (const STRING_TYPE *str, int digcnt, mp_limb_t *n, mp_size_t *nsize,
int *exponent)
{
/* Number of digits for actual limb. */
int cnt = 0;
mp_limb_t low = 0;
mp_limb_t start;
*nsize = 0;
assert (digcnt > 0);
do
{
if (cnt == MAX_DIG_PER_LIMB)
{
if (*nsize == 0)
n[0] = low;
else
{
mp_limb_t cy;
cy = __mpn_mul_1 (n, n, *nsize, MAX_FAC_PER_LIMB);
cy += __mpn_add_1 (n, n, *nsize, low);
if (cy != 0)
n[*nsize] = cy;
}
++(*nsize);
cnt = 0;
low = 0;
}
/* There might be thousands separators or radix characters in
the string. But these all can be ignored because we know the
format of the number is correct and we have an exact number
of characters to read. */
while (*str < L_('0') || *str > L_('9'))
++str;
low = low * 10 + *str++ - L_('0');
++cnt;
}
while (--digcnt > 0);
if (*exponent > 0 && cnt + *exponent <= MAX_DIG_PER_LIMB)
{
low *= _tens_in_limb[*exponent];
start = _tens_in_limb[cnt + *exponent];
*exponent = 0;
}
else
start = _tens_in_limb[cnt];
if (*nsize == 0)
{
n[0] = low;
*nsize = 1;
}
else
{
mp_limb_t cy;
cy = __mpn_mul_1 (n, n, *nsize, start);
cy += __mpn_add_1 (n, n, *nsize, low);
if (cy != 0)
n[(*nsize)++] = cy;
}
return str;
}
/* Shift {PTR, SIZE} COUNT bits to the left, and fill the vacated bits
with the COUNT most significant bits of LIMB.
Tege doesn't like this function so I have to write it here myself. :)
--drepper */
static inline void
__mpn_lshift_1 (mp_limb_t *ptr, mp_size_t size, unsigned int count,
mp_limb_t limb)
{
if (count == BITS_PER_MP_LIMB)
{
/* Optimize the case of shifting by exactly a word:
just copy words, with no actual bit-shifting. */
mp_size_t i;
for (i = size - 1; i > 0; --i)
ptr[i] = ptr[i - 1];
ptr[0] = limb;
}
else
{
(void) __mpn_lshift (ptr, ptr, size, count);
ptr[0] |= limb >> (BITS_PER_MP_LIMB - count);
}
}
#define INTERNAL(x) INTERNAL1(x)
#define INTERNAL1(x) __##x##_internal
/* This file defines a function to check for correct grouping. */
#include "grouping.h"
/* Return a floating point number with the value of the given string NPTR.
Set *ENDPTR to the character after the last used one. If the number is
smaller than the smallest representable number, set `errno' to ERANGE and
return 0.0. If the number is too big to be represented, set `errno' to
ERANGE and return HUGE_VAL with the appropriate sign. */
FLOAT
INTERNAL (STRTOF) (nptr, endptr, group)
const STRING_TYPE *nptr;
STRING_TYPE **endptr;
int group;
{
int negative; /* The sign of the number. */
MPN_VAR (num); /* MP representation of the number. */
int exponent; /* Exponent of the number. */
/* Numbers starting `0X' or `0x' have to be processed with base 16. */
int base = 10;
/* When we have to compute fractional digits we form a fraction with a
second multi-precision number (and we sometimes need a second for
temporary results). */
MPN_VAR (den);
/* Representation for the return value. */
mp_limb_t retval[RETURN_LIMB_SIZE];
/* Number of bits currently in result value. */
int bits;
/* Running pointer after the last character processed in the string. */
const STRING_TYPE *cp, *tp;
/* Start of significant part of the number. */
const STRING_TYPE *startp, *start_of_digits;
/* Points at the character following the integer and fractional digits. */
const STRING_TYPE *expp;
/* Total number of digit and number of digits in integer part. */
int dig_no, int_no, lead_zero;
/* Contains the last character read. */
CHAR_TYPE c;
/* We should get wint_t from <stddef.h>, but not all GCC versions define it
there. So define it ourselves if it remains undefined. */
#ifndef _WINT_T
typedef unsigned int wint_t;
#endif
/* The radix character of the current locale. */
wint_t decimal;
/* The thousands character of the current locale. */
wint_t thousands;
/* The numeric grouping specification of the current locale,
in the format described in <locale.h>. */
const char *grouping;
assert (sizeof (wchar_t) == sizeof (wint_t));
if (group)
{
grouping = _NL_CURRENT (LC_NUMERIC, GROUPING);
if (*grouping <= 0 || *grouping == CHAR_MAX)
grouping = NULL;
else
{
/* Figure out the thousands separator character. */
if (mbtowc ((wchar_t *) &thousands,
_NL_CURRENT (LC_NUMERIC, THOUSANDS_SEP),
strlen (_NL_CURRENT (LC_NUMERIC, THOUSANDS_SEP))) <= 0)
thousands = (wint_t) *_NL_CURRENT (LC_NUMERIC, THOUSANDS_SEP);
if (thousands == L'\0')
grouping = NULL;
}
}
else
{
grouping = NULL;
thousands = L'\0';
}
/* Find the locale's decimal point character. */
if (mbtowc ((wchar_t *) &decimal, _NL_CURRENT (LC_NUMERIC, DECIMAL_POINT),
strlen (_NL_CURRENT (LC_NUMERIC, DECIMAL_POINT))) <= 0)
decimal = (wint_t) *_NL_CURRENT (LC_NUMERIC, DECIMAL_POINT);
assert (decimal != L'\0');
/* Prepare number representation. */
exponent = 0;
negative = 0;
bits = 0;
/* Parse string to get maximal legal prefix. We need the number of
characters of the integer part, the fractional part and the exponent. */
cp = nptr - 1;
/* Ignore leading white space. */
do
c = *++cp;
while (ISSPACE (c));
/* Get sign of the result. */
if (c == L_('-'))
{
negative = 1;
c = *++cp;
}
else if (c == L_('+'))
c = *++cp;
/* Return 0.0 if no legal string is found.
No character is used even if a sign was found. */
if ((c < L_('0') || c > L_('9'))
&& ((wint_t) c != decimal || cp[1] < L_('0') || cp[1] > L_('9')))
{
int matched = 0;
/* Check for `INF' or `INFINITY'. */
if (TOLOWER (c) == L_('i') && ((STRNCASECMP (cp, L_("nf"), 2) == 0
&& (matched = 2))
|| (STRNCASECMP (cp, L_("nfinity"), 7)
== 0
&& (matched = 7))))
{
/* Return +/- inifity. */
if (endptr != NULL)
*endptr = (STRING_TYPE *) (cp + matched);
return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
}
if (TOLOWER (c) == L_('n') && STRNCASECMP (cp, L_("an"), 2) == 0)
{
/* Return NaN. */
if (endptr != NULL)
{
cp += 2;
/* Match `(n-char-sequence-digit)'. */
if (*cp == L_('('))
{
const STRING_TYPE *startp = cp;
do
++cp;
while ((*cp >= '0' && *cp <= '9')
|| (TOLOWER (*cp) >= 'a' && TOLOWER (*cp) <= 'z')
|| *cp == L_('_'));
if (*cp != L_(')'))
/* The closing brace is missing. Only match the NAN
part. */
cp = startp;
}
*endptr = (STRING_TYPE *) cp;
}
return NAN;
}
/* It is really a text we do not recognize. */
RETURN (0.0, nptr);
}
/* First look whether we are faced with a hexadecimal number. */
if (c == L_('0') && TOLOWER (cp[1]) == L_('x'))
{
/* Okay, it is a hexa-decimal number. Remember this and skip
the characters. BTW: hexadecimal numbers must not be
grouped. */
base = 16;
cp += 2;
c = *cp;
grouping = NULL;
}
/* Record the start of the digits, in case we will check their grouping. */
start_of_digits = startp = cp;
/* Ignore leading zeroes. This helps us to avoid useless computations. */
while (c == L_('0') || (thousands != L'\0' && (wint_t) c == thousands))
c = *++cp;
/* If no other digit but a '0' is found the result is 0.0.
Return current read pointer. */
if ((c < L_('0') || c > L_('9')) &&
(base == 16 && (c < TOLOWER (L_('a')) || c > TOLOWER (L_('f')))) &&
(wint_t) c != decimal &&
(base == 16 && (cp == start_of_digits || TOLOWER (c) != L_('p'))) &&
(base != 16 && TOLOWER (c) != L_('e')))
{
tp = correctly_grouped_prefix (start_of_digits, cp, thousands, grouping);
/* If TP is at the start of the digits, there was no correctly
grouped prefix of the string; so no number found. */
RETURN (0.0, tp == start_of_digits ? (base == 16 ? cp - 1 : nptr) : tp);
}
/* Remember first significant digit and read following characters until the
decimal point, exponent character or any non-FP number character. */
startp = cp;
dig_no = 0;
while (dig_no < (base == 16 ? HEXNDIG : NDIG) ||
/* If parsing grouping info, keep going past useful digits
so we can check all the grouping separators. */
grouping)
{
if ((c >= L_('0') && c <= L_('9'))
|| (base == 16 && TOLOWER (c) >= L_('a') && TOLOWER (c) <= L_('f')))
++dig_no;
else if (thousands == L'\0' || (wint_t) c != thousands)
/* Not a digit or separator: end of the integer part. */
break;
c = *++cp;
}
if (grouping && dig_no > 0)
{
/* Check the grouping of the digits. */
tp = correctly_grouped_prefix (start_of_digits, cp, thousands, grouping);
if (cp != tp)
{
/* Less than the entire string was correctly grouped. */
if (tp == start_of_digits)
/* No valid group of numbers at all: no valid number. */
RETURN (0.0, nptr);
if (tp < startp)
/* The number is validly grouped, but consists
only of zeroes. The whole value is zero. */
RETURN (0.0, tp);
/* Recompute DIG_NO so we won't read more digits than
are properly grouped. */
cp = tp;
dig_no = 0;
for (tp = startp; tp < cp; ++tp)
if (*tp >= L_('0') && *tp <= L_('9'))
++dig_no;
int_no = dig_no;
lead_zero = 0;
goto number_parsed;
}
}
if (dig_no >= (base == 16 ? HEXNDIG : NDIG))
/* Too many digits to be representable. Assigning this to EXPONENT
allows us to read the full number but return HUGE_VAL after parsing. */
exponent = MAX_10_EXP;
/* We have the number digits in the integer part. Whether these are all or
any is really a fractional digit will be decided later. */
int_no = dig_no;
lead_zero = int_no == 0 ? -1 : 0;
/* Read the fractional digits. A special case are the 'american style'
numbers like `16.' i.e. with decimal but without trailing digits. */
if ((wint_t) c == decimal)
{
c = *++cp;
while (c >= L_('0') && c <= L_('9') ||
(base == 16 && TOLOWER (c) >= L_('a') && TOLOWER (c) <= L_('f')))
{
if (c != L_('0') && lead_zero == -1)
lead_zero = dig_no - int_no;
++dig_no;
c = *++cp;
}
}
/* Remember start of exponent (if any). */
expp = cp;
/* Read exponent. */
if (TOLOWER (c) == (base == 16 ? L_('p') : L_('e')))
{
int exp_negative = 0;
c = *++cp;
if (c == L_('-'))
{
exp_negative = 1;
c = *++cp;
}
else if (c == L_('+'))
c = *++cp;
if (c >= L_('0') && c <= L_('9'))
{
int exp_limit;
/* Get the exponent limit. */
if (base == 16)
exp_limit = (exp_negative ?
-MIN_EXP + MANT_DIG - 4 * int_no :
MAX_EXP - 4 * int_no + lead_zero);
else
exp_limit = (exp_negative ?
-MIN_10_EXP + MANT_DIG - int_no :
MAX_10_EXP - int_no + lead_zero);
do
{
exponent *= 10;
if (exponent > exp_limit)
/* The exponent is too large/small to represent a valid
number. */
{
FLOAT result;
/* Overflow or underflow. */
__set_errno (ERANGE);
result = (exp_negative ? 0.0 :
negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL);
/* Accept all following digits as part of the exponent. */
do
++cp;
while (*cp >= L_('0') && *cp <= L_('9'));
RETURN (result, cp);
/* NOTREACHED */
}
exponent += c - L_('0');
c = *++cp;
}
while (c >= L_('0') && c <= L_('9'));
if (exp_negative)
exponent = -exponent;
}
else
cp = expp;
}
/* We don't want to have to work with trailing zeroes after the radix. */
if (dig_no > int_no)
{
while (expp[-1] == L_('0'))
{
--expp;
--dig_no;
}
assert (dig_no >= int_no);
}
number_parsed:
/* The whole string is parsed. Store the address of the next character. */
if (endptr)
*endptr = (STRING_TYPE *) cp;
if (dig_no == 0)
return 0.0;
if (lead_zero)
{
/* Find the decimal point */
while ((wint_t) *startp != decimal)
++startp;
startp += lead_zero + 1;
exponent -= base == 16 ? 4 * lead_zero : lead_zero;
dig_no -= lead_zero;
}
/* If the BASE is 16 we can use a simpler algorithm. */
if (base == 16)
{
static const int nbits[16] = { 0, 1, 2, 2, 3, 3, 3, 3,
4, 4, 4, 4, 4, 4, 4, 4 };
int idx = (MANT_DIG - 1) / BITS_PER_MP_LIMB;
int pos = (MANT_DIG - 1) % BITS_PER_MP_LIMB;
mp_limb_t val;
while (!ISXDIGIT (*startp))
++startp;
if (ISDIGIT (*startp))
val = *startp++ - L_('0');
else
val = 10 + TOLOWER (*startp++) - L_('a');
bits = nbits[val];
if (pos + 1 >= 4)
{
/* We don't have to care for wrapping. This is the normal
case so we add this optimization. */
retval[idx] = val << (pos - bits + 1);
pos -= bits;
}
else
{
if (pos + 1 >= bits)
{
retval[idx] = val << (pos - bits + 1);
pos -= bits;
}
else
{
retval[idx--] = val >> (bits - pos - 1);
retval[idx] = val << (BITS_PER_MP_LIMB - (bits - pos - 1));
pos = BITS_PER_MP_LIMB - 1 - (bits - pos - 1);
}
}
while (--dig_no > 0 && idx >= 0)
{
while (!ISXDIGIT (*startp))
++startp;
if (ISDIGIT (*startp))
val = *startp++ - L_('0');
else
val = 10 + TOLOWER (*startp++) - L_('a');
if (pos + 1 >= 4)
{
retval[idx] |= val << (pos - 4 + 1);
pos -= 4;
}
else
{
retval[idx--] |= val >> (4 - pos - 1);
val <<= BITS_PER_MP_LIMB - (4 - pos - 1);
if (idx < 0)
return round_and_return (retval, exponent, negative, val,
BITS_PER_MP_LIMB - 1, dig_no > 0);
retval[idx] = val;
pos = BITS_PER_MP_LIMB - 1 - (4 - pos - 1);
}
}
/* We ran out of digits. */
MPN_ZERO (retval, idx);
return round_and_return (retval, exponent, negative, 0, 0, 0);
}
/* Now we have the number of digits in total and the integer digits as well
as the exponent and its sign. We can decide whether the read digits are
really integer digits or belong to the fractional part; i.e. we normalize
123e-2 to 1.23. */
{
register int incr = (exponent < 0 ? MAX (-int_no, exponent)
: MIN (dig_no - int_no, exponent));
int_no += incr;
exponent -= incr;
}
if (int_no + exponent > MAX_10_EXP + 1)
{
__set_errno (ERANGE);
return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
}
if (exponent < MIN_10_EXP - (DIG + 1))
{
__set_errno (ERANGE);
return 0.0;
}
if (int_no > 0)
{
/* Read the integer part as a multi-precision number to NUM. */
startp = str_to_mpn (startp, int_no, num, &numsize, &exponent);
if (exponent > 0)
{
/* We now multiply the gained number by the given power of ten. */
mp_limb_t *psrc = num;
mp_limb_t *pdest = den;
int expbit = 1;
const struct mp_power *ttab = &_fpioconst_pow10[0];
do
{
if ((exponent & expbit) != 0)
{
mp_limb_t cy;
exponent ^= expbit;
/* FIXME: not the whole multiplication has to be
done. If we have the needed number of bits we
only need the information whether more non-zero
bits follow. */
if (numsize >= ttab->arraysize - _FPIO_CONST_OFFSET)
cy = __mpn_mul (pdest, psrc, numsize,
&ttab->array[_FPIO_CONST_OFFSET],
ttab->arraysize - _FPIO_CONST_OFFSET);
else
cy = __mpn_mul (pdest, &ttab->array[_FPIO_CONST_OFFSET],
ttab->arraysize - _FPIO_CONST_OFFSET,
psrc, numsize);
numsize += ttab->arraysize - _FPIO_CONST_OFFSET;
if (cy == 0)
--numsize;
SWAP (psrc, pdest);
}
expbit <<= 1;
++ttab;
}
while (exponent != 0);
if (psrc == den)
memcpy (num, den, numsize * sizeof (mp_limb_t));
}
/* Determine how many bits of the result we already have. */
count_leading_zeros (bits, num[numsize - 1]);
bits = numsize * BITS_PER_MP_LIMB - bits;
/* Now we know the exponent of the number in base two.
Check it against the maximum possible exponent. */
if (bits > MAX_EXP)
{
__set_errno (ERANGE);
return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
}
/* We have already the first BITS bits of the result. Together with
the information whether more non-zero bits follow this is enough
to determine the result. */
if (bits > MANT_DIG)
{
int i;
const mp_size_t least_idx = (bits - MANT_DIG) / BITS_PER_MP_LIMB;
const mp_size_t least_bit = (bits - MANT_DIG) % BITS_PER_MP_LIMB;
const mp_size_t round_idx = least_bit == 0 ? least_idx - 1
: least_idx;
const mp_size_t round_bit = least_bit == 0 ? BITS_PER_MP_LIMB - 1
: least_bit - 1;
if (least_bit == 0)
memcpy (retval, &num[least_idx],
RETURN_LIMB_SIZE * sizeof (mp_limb_t));
else
{
for (i = least_idx; i < numsize - 1; ++i)
retval[i - least_idx] = (num[i] >> least_bit)
| (num[i + 1]
<< (BITS_PER_MP_LIMB - least_bit));
if (i - least_idx < RETURN_LIMB_SIZE)
retval[RETURN_LIMB_SIZE - 1] = num[i] >> least_bit;
}
/* Check whether any limb beside the ones in RETVAL are non-zero. */
for (i = 0; num[i] == 0; ++i)
;
return round_and_return (retval, bits - 1, negative,
num[round_idx], round_bit,
int_no < dig_no || i < round_idx);
/* NOTREACHED */
}
else if (dig_no == int_no)
{
const mp_size_t target_bit = (MANT_DIG - 1) % BITS_PER_MP_LIMB;
const mp_size_t is_bit = (bits - 1) % BITS_PER_MP_LIMB;
if (target_bit == is_bit)
{
memcpy (&retval[RETURN_LIMB_SIZE - numsize], num,
numsize * sizeof (mp_limb_t));
/* FIXME: the following loop can be avoided if we assume a
maximal MANT_DIG value. */
MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize);
}
else if (target_bit > is_bit)
{
(void) __mpn_lshift (&retval[RETURN_LIMB_SIZE - numsize],
num, numsize, target_bit - is_bit);
/* FIXME: the following loop can be avoided if we assume a
maximal MANT_DIG value. */
MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize);
}
else
{
mp_limb_t cy;
assert (numsize < RETURN_LIMB_SIZE);
cy = __mpn_rshift (&retval[RETURN_LIMB_SIZE - numsize],
num, numsize, is_bit - target_bit);
retval[RETURN_LIMB_SIZE - numsize - 1] = cy;
/* FIXME: the following loop can be avoided if we assume a
maximal MANT_DIG value. */
MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize - 1);
}
return round_and_return (retval, bits - 1, negative, 0, 0, 0);
/* NOTREACHED */
}
/* Store the bits we already have. */
memcpy (retval, num, numsize * sizeof (mp_limb_t));
#if RETURN_LIMB_SIZE > 1
if (numsize < RETURN_LIMB_SIZE)
retval[numsize] = 0;
#endif
}
/* We have to compute at least some of the fractional digits. */
{
/* We construct a fraction and the result of the division gives us
the needed digits. The denominator is 1.0 multiplied by the
exponent of the lowest digit; i.e. 0.123 gives 123 / 1000 and
123e-6 gives 123 / 1000000. */
int expbit;
int cnt;
int neg_exp;
int more_bits;
mp_limb_t cy;
mp_limb_t *psrc = den;
mp_limb_t *pdest = num;
const struct mp_power *ttab = &_fpioconst_pow10[0];
assert (dig_no > int_no && exponent <= 0);
/* For the fractional part we need not process too many digits. One
decimal digits gives us log_2(10) ~ 3.32 bits. If we now compute
ceil(BITS / 3) =: N
digits we should have enough bits for the result. The remaining
decimal digits give us the information that more bits are following.
This can be used while rounding. (One added as a safety margin.) */
if (dig_no - int_no > (MANT_DIG - bits + 2) / 3 + 1)
{
dig_no = int_no + (MANT_DIG - bits + 2) / 3 + 1;
more_bits = 1;
}
else
more_bits = 0;
neg_exp = dig_no - int_no - exponent;
/* Construct the denominator. */
densize = 0;
expbit = 1;
do
{
if ((neg_exp & expbit) != 0)
{
mp_limb_t cy;
neg_exp ^= expbit;
if (densize == 0)
{
densize = ttab->arraysize - _FPIO_CONST_OFFSET;
memcpy (psrc, &ttab->array[_FPIO_CONST_OFFSET],
densize * sizeof (mp_limb_t));
}
else
{
cy = __mpn_mul (pdest, &ttab->array[_FPIO_CONST_OFFSET],
ttab->arraysize - _FPIO_CONST_OFFSET,
psrc, densize);
densize += ttab->arraysize - _FPIO_CONST_OFFSET;
if (cy == 0)
--densize;
SWAP (psrc, pdest);
}
}
expbit <<= 1;
++ttab;
}
while (neg_exp != 0);
if (psrc == num)
memcpy (den, num, densize * sizeof (mp_limb_t));
/* Read the fractional digits from the string. */
(void) str_to_mpn (startp, dig_no - int_no, num, &numsize, &exponent);
/* We now have to shift both numbers so that the highest bit in the
denominator is set. In the same process we copy the numerator to
a high place in the array so that the division constructs the wanted
digits. This is done by a "quasi fix point" number representation.
num: ddddddddddd . 0000000000000000000000
|--- m ---|
den: ddddddddddd n >= m
|--- n ---|
*/
count_leading_zeros (cnt, den[densize - 1]);
(void) __mpn_lshift (den, den, densize, cnt);
cy = __mpn_lshift (num, num, numsize, cnt);
if (cy != 0)
num[numsize++] = cy;
/* Now we are ready for the division. But it is not necessary to
do a full multi-precision division because we only need a small
number of bits for the result. So we do not use __mpn_divmod
here but instead do the division here by hand and stop whenever
the needed number of bits is reached. The code itself comes
from the GNU MP Library by Torbj\"orn Granlund. */
exponent = bits;
switch (densize)
{
case 1:
{
mp_limb_t d, n, quot;
int used = 0;
n = num[0];
d = den[0];
assert (numsize == 1 && n < d);
do
{
udiv_qrnnd (quot, n, n, 0, d);
#define got_limb \
if (bits == 0) \
{ \
register int cnt; \
if (quot == 0) \
cnt = BITS_PER_MP_LIMB; \
else \
count_leading_zeros (cnt, quot); \
exponent -= cnt; \
if (BITS_PER_MP_LIMB - cnt > MANT_DIG) \
{ \
used = MANT_DIG + cnt; \
retval[0] = quot >> (BITS_PER_MP_LIMB - used); \
bits = MANT_DIG + 1; \
} \
else \
{ \
/* Note that we only clear the second element. */ \
/* The conditional is determined at compile time. */ \
if (RETURN_LIMB_SIZE > 1) \
retval[1] = 0; \
retval[0] = quot; \
bits = -cnt; \
} \
} \
else if (bits + BITS_PER_MP_LIMB <= MANT_DIG) \
__mpn_lshift_1 (retval, RETURN_LIMB_SIZE, BITS_PER_MP_LIMB, \
quot); \
else \
{ \
used = MANT_DIG - bits; \
if (used > 0) \
__mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, quot); \
} \
bits += BITS_PER_MP_LIMB
got_limb;
}
while (bits <= MANT_DIG);
return round_and_return (retval, exponent - 1, negative,
quot, BITS_PER_MP_LIMB - 1 - used,
more_bits || n != 0);
}
case 2:
{
mp_limb_t d0, d1, n0, n1;
mp_limb_t quot = 0;
int used = 0;
d0 = den[0];
d1 = den[1];
if (numsize < densize)
{
if (num[0] >= d1)
{
/* The numerator of the number occupies fewer bits than
the denominator but the one limb is bigger than the
high limb of the numerator. */
n1 = 0;
n0 = num[0];
}
else
{
if (bits <= 0)
exponent -= BITS_PER_MP_LIMB;
else
{
if (bits + BITS_PER_MP_LIMB <= MANT_DIG)
__mpn_lshift_1 (retval, RETURN_LIMB_SIZE,
BITS_PER_MP_LIMB, 0);
else
{
used = MANT_DIG - bits;
if (used > 0)
__mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, 0);
}
bits += BITS_PER_MP_LIMB;
}
n1 = num[0];
n0 = 0;
}
}
else
{
n1 = num[1];
n0 = num[0];
}
while (bits <= MANT_DIG)
{
mp_limb_t r;
if (n1 == d1)
{
/* QUOT should be either 111..111 or 111..110. We need
special treatment of this rare case as normal division
would give overflow. */
quot = ~(mp_limb_t) 0;
r = n0 + d1;
if (r < d1) /* Carry in the addition? */
{
add_ssaaaa (n1, n0, r - d0, 0, 0, d0);
goto have_quot;
}
n1 = d0 - (d0 != 0);
n0 = -d0;
}
else
{
udiv_qrnnd (quot, r, n1, n0, d1);
umul_ppmm (n1, n0, d0, quot);
}
q_test:
if (n1 > r || (n1 == r && n0 > 0))
{
/* The estimated QUOT was too large. */
--quot;
sub_ddmmss (n1, n0, n1, n0, 0, d0);
r += d1;
if (r >= d1) /* If not carry, test QUOT again. */
goto q_test;
}
sub_ddmmss (n1, n0, r, 0, n1, n0);
have_quot:
got_limb;
}
return round_and_return (retval, exponent - 1, negative,
quot, BITS_PER_MP_LIMB - 1 - used,
more_bits || n1 != 0 || n0 != 0);
}
default:
{
int i;
mp_limb_t cy, dX, d1, n0, n1;
mp_limb_t quot = 0;
int used = 0;
dX = den[densize - 1];
d1 = den[densize - 2];
/* The division does not work if the upper limb of the two-limb
numerator is greater than the denominator. */
if (__mpn_cmp (num, &den[densize - numsize], numsize) > 0)
num[numsize++] = 0;
if (numsize < densize)
{
mp_size_t empty = densize - numsize;
if (bits <= 0)
{
register int i;
for (i = numsize; i > 0; --i)
num[i + empty] = num[i - 1];
MPN_ZERO (num, empty + 1);
exponent -= empty * BITS_PER_MP_LIMB;
}
else
{
if (bits + empty * BITS_PER_MP_LIMB <= MANT_DIG)
{
/* We make a difference here because the compiler
cannot optimize the `else' case that good and
this reflects all currently used FLOAT types
and GMP implementations. */
register int i;
#if RETURN_LIMB_SIZE <= 2
assert (empty == 1);
__mpn_lshift_1 (retval, RETURN_LIMB_SIZE,
BITS_PER_MP_LIMB, 0);
#else
for (i = RETURN_LIMB_SIZE; i > empty; --i)
retval[i] = retval[i - empty];
#endif
#if RETURN_LIMB_SIZE > 1
retval[1] = 0;
#endif
for (i = numsize; i > 0; --i)
num[i + empty] = num[i - 1];
MPN_ZERO (num, empty + 1);
}
else
{
used = MANT_DIG - bits;
if (used >= BITS_PER_MP_LIMB)
{
register int i;
(void) __mpn_lshift (&retval[used
/ BITS_PER_MP_LIMB],
retval, RETURN_LIMB_SIZE,
used % BITS_PER_MP_LIMB);
for (i = used / BITS_PER_MP_LIMB; i >= 0; --i)
retval[i] = 0;
}
else if (used > 0)
__mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, 0);
}
bits += empty * BITS_PER_MP_LIMB;
}
}
else
{
int i;
assert (numsize == densize);
for (i = numsize; i > 0; --i)
num[i] = num[i - 1];
}
den[densize] = 0;
n0 = num[densize];
while (bits <= MANT_DIG)
{
if (n0 == dX)
/* This might over-estimate QUOT, but it's probably not
worth the extra code here to find out. */
quot = ~(mp_limb_t) 0;
else
{
mp_limb_t r;
udiv_qrnnd (quot, r, n0, num[densize - 1], dX);
umul_ppmm (n1, n0, d1, quot);
while (n1 > r || (n1 == r && n0 > num[densize - 2]))
{
--quot;
r += dX;
if (r < dX) /* I.e. "carry in previous addition?" */
break;
n1 -= n0 < d1;
n0 -= d1;
}
}
/* Possible optimization: We already have (q * n0) and (1 * n1)
after the calculation of QUOT. Taking advantage of this, we
could make this loop make two iterations less. */
cy = __mpn_submul_1 (num, den, densize + 1, quot);
if (num[densize] != cy)
{
cy = __mpn_add_n (num, num, den, densize);
assert (cy != 0);
--quot;
}
n0 = num[densize] = num[densize - 1];
for (i = densize - 1; i > 0; --i)
num[i] = num[i - 1];
got_limb;
}
for (i = densize; num[i] == 0 && i >= 0; --i)
;
return round_and_return (retval, exponent - 1, negative,
quot, BITS_PER_MP_LIMB - 1 - used,
more_bits || i >= 0);
}
}
}
/* NOTREACHED */
}
/* External user entry point. */
FLOAT
#ifdef weak_function
weak_function
#endif
STRTOF (nptr, endptr)
const STRING_TYPE *nptr;
STRING_TYPE **endptr;
{
return INTERNAL (STRTOF) (nptr, endptr, 0);
}