bug-gnulib
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[PATCH] ftoastr: new module, for lossless conversion of floats to short


From: Paul Eggert
Subject: [PATCH] ftoastr: new module, for lossless conversion of floats to short strings
Date: Wed, 17 Nov 2010 13:00:56 -0800
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.1.15) Gecko/20101027 Thunderbird/3.0.10

* lib/ftoastr.h, lib/ftoastr.c, lib/dtoastr.c, lib/ldtoastr.c:
* modules/ftoastr: New files.
---
 ChangeLog       |    6 ++
 lib/dtoastr.c   |    2 +
 lib/ftoastr.c   |   89 ++++++++++++++++++++++++++++++++++
 lib/ftoastr.h   |  144 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/ldtoastr.c  |    2 +
 modules/ftoastr |   25 ++++++++++
 6 files changed, 268 insertions(+), 0 deletions(-)
 create mode 100644 lib/dtoastr.c
 create mode 100644 lib/ftoastr.c
 create mode 100644 lib/ftoastr.h
 create mode 100644 lib/ldtoastr.c
 create mode 100644 modules/ftoastr

diff --git a/ChangeLog b/ChangeLog
index 0e6240c..ea0610a 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,9 @@
+2010-11-17  Paul Eggert  <address@hidden>
+
+       ftoastr: new module, for lossless conversion of floats to short strings
+       * lib/ftoastr.h, lib/ftoastr.c, lib/dtoastr.c, lib/ldtoastr.c:
+       * modules/ftoastr: New files.
+
 2010-11-15  Paul Eggert  <address@hidden>
 
        bootstrap: port to Solaris sed
diff --git a/lib/dtoastr.c b/lib/dtoastr.c
new file mode 100644
index 0000000..aed181d
--- /dev/null
+++ b/lib/dtoastr.c
@@ -0,0 +1,2 @@
+#define LENGTH 2
+#include "ftoastr.c"
diff --git a/lib/ftoastr.c b/lib/ftoastr.c
new file mode 100644
index 0000000..25e0705
--- /dev/null
+++ b/lib/ftoastr.c
@@ -0,0 +1,89 @@
+/* floating point to accurate string
+
+   Copyright (C) 2010 Free Software Foundation, Inc.
+
+   This program 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 3 of the License, or
+   (at your option) any later version.
+
+   This program 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 program.  If not, see <http://www.gnu.org/licenses/>.  */
+
+/* Written by Paul Eggert.  */
+
+#include "ftoastr.h"
+
+#include "intprops.h"
+#include <float.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#if LENGTH == 3
+# define FLOAT long double
+# define FLOAT_DIG LDBL_DIG
+# define FLOAT_MIN LDBL_MIN
+# define FLOAT_PREC_BOUND _GL_LDBL_PREC_BOUND
+# define FTOASTR ldtoastr
+# define STRTOF strtold
+#elif LENGTH == 2
+# define FLOAT double
+# define FLOAT_DIG DBL_DIG
+# define FLOAT_MIN DBL_MIN
+# define FLOAT_PREC_BOUND _GL_DBL_PREC_BOUND
+# define FTOASTR dtoastr
+# define STRTOF strtod
+#else
+# define LENGTH 1
+# define FLOAT float
+# define FLOAT_DIG FLT_DIG
+# define FLOAT_MIN FLT_MIN
+# define FLOAT_PREC_BOUND _GL_FLT_PREC_BOUND
+# define FTOASTR ftoastr
+# define STRTOF strtof
+#endif
+
+int
+FTOASTR (char *buf, size_t bufsize, int flags, int width, FLOAT x)
+{
+  /* The following method is simple but slow.
+     For ideas about speeding things up, please see:
+
+     Florian Loitsch, Printing floating-point numbers quickly and accurately
+     with integers.  ACM SIGPLAN notices 46, 6 (June 2010), 233-243
+     <http://dx.doi.org/10.1145/1809028.1806623>.  */
+
+  char format[sizeof "%-+ 0*.*Lg"];
+  FLOAT abs_x = x < 0 ? -x : x;
+  int prec;
+
+  char *p = format;
+  *p++ = '%';
+
+  /* Support flags that generate output parsable by strtof.  */
+  *p = '-'; p += (flags & FTOASTR_LEFT_JUSTIFY  ) != 0;
+  *p = '+'; p += (flags & FTOASTR_ALWAYS_SIGNED ) != 0;
+  *p = ' '; p += (flags & FTOASTR_SPACE_POSITIVE) != 0;
+  *p = '0'; p += (flags & FTOASTR_ZERO_PAD      ) != 0;
+
+  *p++ = '*';
+  *p++ = '.';
+  *p++ = '*';
+  *p = 'L'; p += 2 < LENGTH;
+  *p++ = flags & FTOASTR_UPPER_E ? 'G' : 'g';
+  *p = '\0';
+
+  for (prec = abs_x < FLOAT_MIN ? 1 : FLOAT_DIG; ; prec++)
+    {
+      int n = snprintf (buf, bufsize, format, width, prec, x);
+      if (n < 0
+          || FLOAT_PREC_BOUND <= prec
+          || (n < bufsize && STRTOF (buf, NULL) == x))
+        return n;
+    }
+}
diff --git a/lib/ftoastr.h b/lib/ftoastr.h
new file mode 100644
index 0000000..4c4f890
--- /dev/null
+++ b/lib/ftoastr.h
@@ -0,0 +1,144 @@
+/* floating point to accurate string
+
+   Copyright (C) 2010 Free Software Foundation, Inc.
+
+   This program 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 3 of the License, or
+   (at your option) any later version.
+
+   This program 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 program.  If not, see <http://www.gnu.org/licenses/>.  */
+
+/* Written by Paul Eggert.  */
+
+#ifndef _GL_FTOASTR_H
+
+#include "intprops.h"
+#include <float.h>
+#include <stddef.h>
+
+/* Store into BUF (of size BUFSIZE) an accurate minimal-precision
+   string representation of a floating point number.  FLAGS affect the
+   formatting of the number.  Pad the output string with spaces as
+   necessary to width WIDTH bytes, in the style of printf.  X is the
+   floating-point number to be converted.
+
+   Return the number of bytes stored into BUF, not counting the
+   terminating null.  However, do not overrun BUF: if BUF is too
+   small, return a fairly tight (but not necessarily exact) upper
+   bound on the value that would have been returned if BUF had been
+   big enough.  If SIZE is zero, BUF may be a null pointer.  On error
+   (e.g., returned value would exceed INT_MAX), return -1 and set
+   errno.
+
+   Example:
+
+     char buf[DBL_ABUFSIZE_BOUND];
+     int r = dtoastr (buf, sizeof buf, "%*.*g", 0, 0.1);
+
+   In the C locale, this sets R to 3 and stores "0.1" into BUF.  */
+
+int  ftoastr (char *buf, size_t bufsize, int flags, int width,       float x);
+int  dtoastr (char *buf, size_t bufsize, int flags, int width,      double x);
+int ldtoastr (char *buf, size_t bufsize, int flags, int width, long double x);
+
+/* Flag values for ftoastr etc.  These can be ORed together.  */
+enum
+  {
+    /* Left justify within the width; the default is right justification.  */
+    FTOASTR_LEFT_JUSTIFY = 1,
+
+    /* Output "+" before positive numbers; the default outputs nothing.  */
+    FTOASTR_ALWAYS_SIGNED = 2,
+
+    /* Output " " before positive numbers; ignored if
+       FTOASTER_ALWAYS_SIGNED is also given.  */
+    FTOASTR_SPACE_POSITIVE = 4,
+
+    /* Pad with zeros instead of spaces; ignored if FTOASTR_LEFT_JUSTIFY
+       is also given.  */
+    FTOASTR_ZERO_PAD = 8,
+
+    /* Use 'E' instead of 'e' before the exponent.  */
+    FTOASTR_UPPER_E = 16
+  };
+
+
+/* _GL_FLT_PREC_BOUND is an upper bound on the precision needed to
+   represent a float value without losing information.  Likewise for
+   _GL_DBL_PREC_BOUND and double, and _GL_LDBL_PREC_BOUND and long double.  */
+
+#if FLT_RADIX == 10 /* decimal floating point */
+ enum {  _GL_FLT_PREC_BOUND =  FLT_MANT_DIG };
+ enum {  _GL_DBL_PREC_BOUND =  DBL_MANT_DIG };
+ enum { _GL_LDBL_PREC_BOUND = LDBL_MANT_DIG };
+#else
+
+/* An upper bound on the number of bits needed to represent a single
+   digit in a floating-point fraction.  */
+# if FLT_RADIX == 2 /* IEEE 754 floating point, VAX floating point, etc. */
+#  define _GL_FLOAT_DIG_BITS_BOUND 1
+# elif FLT_RADIX <= 16 /* IBM hex floating point has FLT_RADIX == 16.  */
+#  define _GL_FLOAT_DIG_BITS_BOUND 4
+# else /* no machine is this bad, but let's be complete */
+#  define _GL_FLOAT_DIG_BITS_BOUND (CHAR_BIT * (int) sizeof (int) - 1)
+# endif
+
+/* An upper bound on the number of decimal digits needed to represent
+   a floating point number accurately, assuming a fraction contains
+   DIG digits.  For why the "+ 1" is needed, see "Binary to Decimal
+   Conversion" in David Goldberg's paper "What Every Computer
+   Scientist Should Know About Floating-Point Arithmetic"
+   <http://docs.sun.com/source/806-3568/ncg_goldberg.html>.  */
+# define _GL_FLOAT_PREC_BOUND(dig) \
+   (INT_BITS_STRLEN_BOUND ((dig) * _GL_FLOAT_DIG_BITS_BOUND) + 1)
+
+ enum {  _GL_FLT_PREC_BOUND = _GL_FLOAT_PREC_BOUND ( FLT_MANT_DIG) };
+ enum {  _GL_DBL_PREC_BOUND = _GL_FLOAT_PREC_BOUND ( DBL_MANT_DIG) };
+ enum { _GL_LDBL_PREC_BOUND = _GL_FLOAT_PREC_BOUND (LDBL_MANT_DIG) };
+#endif
+
+
+/* Bound on the number of bytes printed for an exponent in the range
+   MIN..MAX, where MIN < 0 < MAX; printf always prints a sign and at
+   least 2 digits.  Although the maximum known exponent is 4932 for
+   IEEE 754 binary128, support tight bounds for exponents up to a
+   million, just in case.  */
+#define _GL_FLOAT_EXPONENT_STRLEN_BOUND(min, max)  \
+  (      -100 < (min) && (max) <     100 ? 3       \
+   :    -1000 < (min) && (max) <    1000 ? 4       \
+   :   -10000 < (min) && (max) <   10000 ? 5       \
+   :  -100000 < (min) && (max) <  100000 ? 6       \
+   : -1000000 < (min) && (max) < 1000000 ? 7       \
+   : INT_STRLEN_BOUND (int) /* not a tight bound */)
+
+/* A reasonably tight bound on the length of a type-T floating value
+   formatted with ftoastr etc.  Room is needed for sign, fraction
+   digits, decimal point, "e", and exponent.  POINTLEN should be a
+   reasonably tight bound on the string length of the decimal
+   point.  */
+#define _GL_FLOAT_STRLEN_BOUND_L(t, pointlen)                          \
+  (1 + _GL_##t##_PREC_BOUND + pointlen + 1                             \
+   + _GL_FLOAT_EXPONENT_STRLEN_BOUND (t##_MIN_10_EXP, t##_MAX_10_EXP))
+#define  FLT_STRLEN_BOUND_L(pointlen) _GL_FLOAT_STRLEN_BOUND_L ( FLT, pointlen)
+#define  DBL_STRLEN_BOUND_L(pointlen) _GL_FLOAT_STRLEN_BOUND_L ( DBL, pointlen)
+#define LDBL_STRLEN_BOUND_L(pointlen) _GL_FLOAT_STRLEN_BOUND_L (LDBL, pointlen)
+
+/* Looser bounds that are locale-independent and are integral constant
+   expressions.  */
+#define  FLT_STRLEN_BOUND  FLT_STRLEN_BOUND_L (MB_LEN_MAX)
+#define  DBL_STRLEN_BOUND  DBL_STRLEN_BOUND_L (MB_LEN_MAX)
+#define LDBL_STRLEN_BOUND LDBL_STRLEN_BOUND_L (MB_LEN_MAX)
+
+/* Looser, locale-independent bounds that include the trailing null byte.  */
+#define  FLT_BUFSIZE_BOUND ( FLT_STRLEN_BOUND + 1)
+#define  DBL_BUFSIZE_BOUND ( DBL_STRLEN_BOUND + 1)
+#define LDBL_BUFSIZE_BOUND (LDBL_STRLEN_BOUND + 1)
+
+#endif /* _GL_FTOASTR_H */
diff --git a/lib/ldtoastr.c b/lib/ldtoastr.c
new file mode 100644
index 0000000..bf54a35
--- /dev/null
+++ b/lib/ldtoastr.c
@@ -0,0 +1,2 @@
+#define LENGTH 3
+#include "ftoastr.c"
diff --git a/modules/ftoastr b/modules/ftoastr
new file mode 100644
index 0000000..fbc0cba
--- /dev/null
+++ b/modules/ftoastr
@@ -0,0 +1,25 @@
+Description:
+Convert floating point to accurate string
+
+Files:
+lib/ftoastr.h
+lib/ftoastr.c
+lib/dtoastr.c
+lib/ldtoastr.c
+
+Depends-on:
+intprops
+
+configure.ac:
+
+Makefile.am:
+lib_SOURCES += ftoastr.h ftoastr.c dtoastr.c ldtoastr.c
+
+Include:
+"ftoastr.h"
+
+License:
+GPL
+
+Maintainer:
+Paul Eggert
-- 
1.7.2




reply via email to

[Prev in Thread] Current Thread [Next in Thread]