Avoid precision errors in round(), lround() and llround() functions.

This commit is contained in:
Keith Marshall 2008-06-03 18:42:21 +00:00
parent 585cd03044
commit 47140f3cdb
14 changed files with 325 additions and 150 deletions

View File

@ -1,3 +1,34 @@
2008-06-02 Keith Marshall <keithmarshall@users.sourceforge.net>
Avoid precision errors in round(), lround() and llround() functions.
* mingwex/math/round_generic.c: New file; it replaces...
* mingwex/math/round.c: ...this; delete it.
* mingwex/math/roundf.c: ...and this; ditto.
* mingwex/math/roundl.c: ...and this; ditto.
* mingwex/math/lround_generic.c: New file; it replaces...
* mingwex/math/lround.c: ...this; delete it.
* mingwex/math/lroundf.c: ...and this; ditto.
* mingwex/math/lroundl.c: ...and this; ditto.
* mingwex/math/llround.c: ...and this; ditto.
* mingwex/math/llroundf.c: ...and this; ditto.
* mingwex/math/llroundl.c: ...and this; ditto.
* mingwex/math/round_internal.h: New file; it provides generic common
code, which is shared by `round_generic.c' and `lround_generic.c'; the
implementation is based on techniques suggested by Danny Smith and
Greg Chicares.
* mingwex/Makefile.in (MATH_DISTFILES): Remove `round.c', `roundf.c',
`roundl.c', `lround.c', `lroundf.c', `lroundl.c', `llround.c',
`llroundf.c' and `llroundl.c'; replace by `round_internal.h',
`round_generic.c' and `lround_generic.c'.
(MATH_OBJS): Factor out files listed in...
(MATH_ROUND_OBJS, MATH_LROUND_OBJS, MATH_LLROUND_OBJS): ...these new
macros; define them; specify dependencies and build rules; add to...
(LIB_OBJS): ...this list.
2008-05-22 Danny Smith <dannysmith@users.sourceforge.net>
[ mingw-Bugs-1961893 ]

View File

@ -54,17 +54,16 @@ MATH_DISTFILES = \
fpclassify.c fpclassifyf.c fpclassifyl.c \
frexpf.c frexpl.S fucom.c hypotf.c hypotl.c ilogb.S ilogbf.S \
ilogbl.S isnan.c isnanf.c isnanl.c ldexpf.c ldexpl.c \
lgamma.c lgammaf.c lgammal.c llrint.c \
llrintf.c llrintl.c llround.c llroundf.c llroundl.c \
lgamma.c lgammaf.c lgammal.c llrint.c llrintf.c llrintl.c \
log10f.S log10l.S log1p.S log1pf.S log1pl.S log2.S log2f.S \
log2l.S logb.c logbf.c logbl.c logf.S logl.S lrint.c lrintf.c \
lrintl.c lround.c lroundf.c lroundl.c modff.c modfl.c \
lrintl.c lround_generic.c modff.c modfl.c \
nearbyint.S nearbyintf.S nearbyintl.S \
nextafterf.c nextafterl.c nexttowardf.c nexttoward.c \
powf.c powi.c powif.c powil.c powl.c \
remainder.S remainderf.S remainderl.S remquo.S \
remquof.S remquol.S rint.c rintf.c rintl.c round.c roundf.c \
roundl.c scalbn.S scalbnf.S scalbnl.S s_erf.c sf_erf.c \
remquof.S remquol.S rint.c rintf.c rintl.c round_internal.h \
round_generic.c scalbn.S scalbnf.S scalbnl.S s_erf.c sf_erf.c \
signbit.c signbitf.c signbitl.c sinf.S sinhf.c sinhl.c sinl.S \
sqrtf.c sqrtl.c tanf.S tanhf.c tanhl.c tanl.S tgamma.c \
tgammaf.c tgammal.c trunc.c truncf.c truncl.c \
@ -153,22 +152,24 @@ MATH_OBJS = \
fpclassify.o fpclassifyf.o fpclassifyl.o \
frexpf.o frexpl.o fucom.o hypotf.o hypotl.o ilogb.o ilogbf.o \
ilogbl.o isnan.o isnanf.o isnanl.o ldexpf.o ldexpl.o \
lgamma.o lgammaf.o lgammal.o llrint.o \
llrintf.o llrintl.o llround.o llroundf.o llroundl.o \
lgamma.o lgammaf.o lgammal.o llrint.o llrintf.o llrintl.o \
log10f.o log10l.o log1p.o log1pf.o log1pl.o log2.o log2f.o \
log2l.o logb.o logbf.o logbl.o logf.o logl.o lrint.o lrintf.o \
lrintl.o lround.o lroundf.o lroundl.o modff.o modfl.o \
log2l.o logb.o logbf.o logbl.o logf.o logl.o \
lrint.o lrintf.o lrintl.o modff.o modfl.o \
nearbyint.o nearbyintf.o nearbyintl.o \
nextafterf.o nextafterl.o nexttowardf.o nexttoward.o \
powf.o powi.o powif.o powil.o powl.o \
remainder.o remainderf.o remainderl.o remquo.o \
remquof.o remquol.o rint.o rintf.o rintl.o round.o roundf.o \
roundl.o scalbn.o scalbnf.o scalbnl.o s_erf.o sf_erf.o \
remquof.o remquol.o rint.o rintf.o rintl.o \
scalbn.o scalbnf.o scalbnl.o s_erf.o sf_erf.o \
signbit.o signbitf.o signbitl.o sinf.o sinhf.o sinhl.o sinl.o \
sqrtf.o sqrtl.o tanf.o tanhf.o tanhl.o tanl.o tgamma.o \
tgammaf.o tgammal.o trunc.o truncf.o truncl.o \
acosh.o acoshf.o acoshl.o asinh.o asinhf.o asinhl.o \
atanh.o atanhf.o atanhl.o
MATH_ROUND_OBJS = round.o roundf.o roundl.o
MATH_LROUND_OBJS = lround.o lroundf.o lroundl.o
MATH_LLROUND_OBJS = llround.o llroundf.o llroundl.o
FENV_OBJS = fesetround.o fegetround.o \
fegetenv.o fesetenv.o feupdateenv.o \
feclearexcept.o feholdexcept.o fegetexceptflag.o \
@ -196,7 +197,8 @@ GDTOA_OBJS = \
mingw_snprintf.o
LIB_OBJS = $(Q8_OBJS) $(CTYPE_OBJS) $(STDLIB_STUB_OBJS) \
$(STDIO_OBJS) $(MATH_OBJS) $(FENV_OBJS) \
$(STDIO_OBJS) $(MATH_OBJS) $(MATH_ROUND_OBJS) \
$(MATH_LROUND_OBJS) $(MATH_LLROUND_OBJS) $(FENV_OBJS) \
$(POSIX_OBJS) $(REPLACE_OBJS) $(COMPLEX_OBJS) \
$(GDTOA_OBJS)
@ -210,6 +212,14 @@ $(LIBMINGWEX_A): $(LIB_OBJS)
$(AR) $(ARFLAGS) $@ $(LIB_OBJS)
$(RANLIB) $@
$(MATH_ROUND_OBJS): round_generic.c
$(CC) $(ALL_CFLAGS) -I$(srcdir)/math -c -o $@ \
-D FUNCTION=$* $(srcdir)/math/round_generic.c
$(MATH_LROUND_OBJS) $(MATH_LLROUND_OBJS): lround_generic.c
$(CC) $(ALL_CFLAGS) -I$(srcdir)/math -c -o $@ \
-D FUNCTION=$* $(srcdir)/math/lround_generic.c
Makefile: Makefile.in config.status configure
$(SHELL) config.status
@ -255,6 +265,8 @@ mbrtowc.o wcrtomb.o wcstof.o wcstold.o: mb_wc_common.h
$(GDTOA_OBJS): gd_arith.h gdtoa.h gdtoaimp.h gd_qnan.h
$(MATH_ROUND_OBJS) $(MATH_LROUND_OBJS) $(MATH_LLROUND_OBJS): round_internal.h
dist:
mkdir $(distdir)/mingwex
chmod 755 $(distdir)/mingwex

View File

@ -1,19 +0,0 @@
#include <math.h>
#include <limits.h>
#include <errno.h>
long long
llround (double x)
{
/* Add +/- 0.5, then round towards zero. */
double tmp = trunc (x + (x >= 0.0 ? 0.5 : -0.5));
if (!isfinite (tmp)
|| tmp > (double)LONG_LONG_MAX
|| tmp < (double)LONG_LONG_MIN)
{
errno = ERANGE;
/* Undefined behaviour, so we could return anything. */
/* return tmp > 0.0 ? LONG_LONG_MAX : LONG_LONG_MIN; */
}
return (long long)tmp;
}

View File

@ -1,19 +0,0 @@
#include <math.h>
#include <limits.h>
#include <errno.h>
long long
llroundf (float x)
{
/* Add +/- 0.5, then round towards zero. */
float tmp = truncf (x + (x >= 0.0F ? 0.5F : -0.5F));
if (!isfinite (tmp)
|| tmp > (float)LONG_LONG_MAX
|| tmp < (float)LONG_LONG_MIN)
{
errno = ERANGE;
/* Undefined behaviour, so we could return anything. */
/* return tmp > 0.0F ? LONG_LONG_MAX : LONG_LONG_MIN; */
}
return (long long)tmp;
}

View File

@ -1,19 +0,0 @@
#include <math.h>
#include <limits.h>
#include <errno.h>
long long
llroundl (long double x)
{
/* Add +/- 0.5, then round towards zero. */
long double tmp = truncl (x + (x >= 0.0L ? 0.5L : -0.5L));
if (!isfinite (tmp)
|| tmp > (long double)LONG_LONG_MAX
|| tmp < (long double)LONG_LONG_MIN)
{
errno = ERANGE;
/* Undefined behaviour, so we could return anything. */
/* return tmp > 0.0L ? LONG_LONG_MAX : LONG_LONG_MIN; */
}
return (long long)tmp;
}

View File

@ -1,19 +0,0 @@
#include <math.h>
#include <limits.h>
#include <errno.h>
long
lround (double x)
{
/* Add +/- 0.5 then then round towards zero. */
double tmp = trunc (x + (x >= 0.0 ? 0.5 : -0.5));
if (!isfinite (tmp)
|| tmp > (double)LONG_MAX
|| tmp < (double)LONG_MIN)
{
errno = ERANGE;
/* Undefined behaviour, so we could return anything. */
/* return tmp > 0.0 ? LONG_MAX : LONG_MIN; */
}
return (long)tmp;
}

View File

@ -0,0 +1,64 @@
/*
* lround_generic.c
*
* $Id$
*
* Provides a generic implementation for the `lround()', `lroundf()',
* `lroundl()', `llround()', `llroundf()' and `llroundl()' functions;
* compile with `-D FUNCTION=name', with `name' set to each of these
* six in turn, to create separate object files for each of the six
* functions.
*
* Written by Keith Marshall <keithmarshall@users.sourceforge.net>
*
* This is free software. You may redistribute and/or modify it as you
* see fit, without restriction of copyright.
*
* This software is provided "as is", in the hope that it may be useful,
* but WITHOUT WARRANTY OF ANY KIND, not even any implied warranty of
* MERCHANTABILITY, nor of FITNESS FOR ANY PARTICULAR PURPOSE. At no
* time will the author accept any form of liability for any damages,
* however caused, resulting from the use of this software.
*
*/
#ifndef FUNCTION
/*
* Normally specified with `-D FUNCTION=name', on the command line.
* Valid FUNCTION names are `lround', `lroundf', `lroundl', `llround'
* `llroundf' and `llroundl'; specifying anything else will most likely
* cause a compilation error. If user did not specify an appropriate
* FUNCTION name, default to `lround'.
*/
#define FUNCTION lround
#endif
#include "round_internal.h"
#include <limits.h>
#include <errno.h>
/* Generic implementation.
* The user is required to specify the FUNCTION name;
* the RETURN_TYPE and INPUT_TYPE macros resolve to appropriate
* type declarations, to match the selected FUNCTION prototype,
* while RETURN_MAX and RETURN_MIN map to the correspondingly
* appropriate limits.h manifest values, to establish the
* valid range for the RETURN_TYPE.
*/
RETURN_TYPE FUNCTION( INPUT_TYPE x )
{
if( !isfinite( x ) || !isfinite( x = round_internal( x ) )
|| (x > MAX_RETURN_VALUE) || (x < MIN_RETURN_VALUE) )
/*
* Undefined behaviour...
* POSIX requires us to report a domain error; ANSI C99 says we
* _may_ report a range error, and previous MinGW implementation
* set `errno = ERANGE' here; we change that, conforming to the
* stricter requiremment of the POSIX standard.
*/
errno = EDOM;
return (RETURN_TYPE)(x);
}
/* $RCSfile$$Revision$: end of file */

View File

@ -1,19 +0,0 @@
#include <math.h>
#include <limits.h>
#include <errno.h>
long
lroundf (float x)
{
/* Add +/- 0.5, then round towards zero. */
float tmp = truncf (x + (x >= 0.0F ? 0.5F : -0.5F));
if (!isfinite (tmp)
|| tmp > (float)LONG_MAX
|| tmp < (float)LONG_MIN)
{
errno = ERANGE;
/* Undefined behaviour, so we could return anything. */
/* return tmp > 0.0F ? LONG_MAX : LONG_MIN; */
}
return (long)tmp;
}

View File

@ -1,19 +0,0 @@
#include <math.h>
#include <limits.h>
#include <errno.h>
long
lroundl (long double x)
{
/* Add +/- 0.5, then round towards zero. */
long double tmp = truncl (x + (x >= 0.0L ? 0.5L : -0.5L));
if (!isfinite (tmp)
|| tmp > (long double)LONG_MAX
|| tmp < (long double)LONG_MIN)
{
errno = ERANGE;
/* Undefined behaviour, so we could return anything. */
/* return tmp > 0.0L ? LONG_MAX : LONG_MIN; */
}
return (long)tmp;
}

View File

@ -1,8 +0,0 @@
#include <math.h>
double
round (double x)
{
/* Add +/- 0.5 then then round towards zero. */
return trunc ( x + (x >= 0.0 ? 0.5 : -0.5));
}

View File

@ -0,0 +1,51 @@
/*
* round_generic.c
*
* $Id$
*
* Provides a generic implementation for the `round()', `roundf()'
* and `roundl()' functions; compile with `-D FUNCTION=name', with
* `name' set to each of these three in turn, to create separate
* object files for each of the three functions.
*
* Written by Keith Marshall <keithmarshall@users.sourceforge.net>
*
* This is free software. You may redistribute and/or modify it as you
* see fit, without restriction of copyright.
*
* This software is provided "as is", in the hope that it may be useful,
* but WITHOUT WARRANTY OF ANY KIND, not even any implied warranty of
* MERCHANTABILITY, nor of FITNESS FOR ANY PARTICULAR PURPOSE. At no
* time will the author accept any form of liability for any damages,
* however caused, resulting from the use of this software.
*
*/
#ifndef FUNCTION
/*
* Normally specified with `-D FUNCTION=name', on the command line.
* Valid FUNCTION names are `round', `roundf' and `roundl'; specifying
* anything else will most likely cause a compilation error. If user
* did not specify any FUNCTION name, default to `round'.
*/
#define FUNCTION round
#endif
#include "round_internal.h"
/* Generic implementation.
* The user is required to specify the FUNCTION name;
* the RETURN_TYPE and INPUT_TYPE macros resolve to appropriate
* type declarations, to match the selected FUNCTION prototype.
*/
RETURN_TYPE FUNCTION( INPUT_TYPE x )
{
/* Round to nearest integer, away from zero for half-way.
*
* We split it with the `round_internal()' function in
* a private header file, so that it may be shared by this,
* `lround()' and `llround()' implementations.
*/
return isfinite( x ) ? round_internal( x ) : x;
}
/* $RCSfile$$Revision$: end of file */

View File

@ -0,0 +1,155 @@
#ifndef _ROUND_INTERNAL_H
/*
* round_internal.h
*
* $Id$
*
* Provides a generic implementation of the numerical rounding
* algorithm, which is shared by all functions in the `round()',
* `lround()' and `llround()' families.
*
* Written by Keith Marshall <keithmarshall@users.sourceforge.net>
*
* This is free software. You may redistribute and/or modify it as you
* see fit, without restriction of copyright.
*
* This software is provided "as is", in the hope that it may be useful,
* but WITHOUT WARRANTY OF ANY KIND, not even any implied warranty of
* MERCHANTABILITY, nor of FITNESS FOR ANY PARTICULAR PURPOSE. At no
* time will the author accept any form of liability for any damages,
* however caused, resulting from the use of this software.
*
*/
#define _ROUND_INTERNAL_H
#include <math.h>
#include <fenv.h>
#define TYPE_PASTE( NAME, TYPE ) NAME##TYPE
#define INPUT_TYPE INPUT_TYPEDEF( FUNCTION )
#define INPUT_TYPEDEF( FUNCTION ) TYPE_PASTE( FUNCTION, _input_type )
/*
* The types for the formal parameter, to each of the derived functions.
*/
#define round_input_type double
#define roundf_input_type float
#define roundl_input_type long double
#define lround_input_type double
#define lroundf_input_type float
#define lroundl_input_type long double
#define llround_input_type double
#define llroundf_input_type float
#define llroundl_input_type long double
#define RETURN_TYPE RETURN_TYPEDEF( FUNCTION )
#define RETURN_TYPEDEF( FUNCTION ) TYPE_PASTE( FUNCTION, _return_type )
/*
* The types for the return value, from each of the derived functions.
*/
#define round_return_type double
#define roundf_return_type float
#define roundl_return_type long double
#define lround_return_type long
#define lroundf_return_type long
#define lroundl_return_type long
#define llround_return_type long long
#define llroundf_return_type long long
#define llroundl_return_type long long
#define MAX_RETURN_VALUE RETURN_MAX( FUNCTION )
#define RETURN_MAX( FUNCTION ) TYPE_PASTE( FUNCTION, _return_max )
/*
* The maximum values which may be returned by each of the derived functions
* in the `lround' or the `llround' families.
*/
#define lround_return_max LONG_MAX
#define lroundf_return_max LONG_MAX
#define lroundl_return_max LONG_MAX
#define llround_return_max LLONG_MAX
#define llroundf_return_max LLONG_MAX
#define llroundl_return_max LLONG_MAX
#define MIN_RETURN_VALUE RETURN_MIN( FUNCTION )
#define RETURN_MIN( FUNCTION ) TYPE_PASTE( FUNCTION, _return_min )
/*
* The minimum values which may be returned by each of the derived functions
* in the `lround' or the `llround' families.
*/
#define lround_return_min LONG_MIN
#define lroundf_return_min LONG_MIN
#define lroundl_return_min LONG_MIN
#define llround_return_min LLONG_MIN
#define llroundf_return_min LLONG_MIN
#define llroundl_return_min LLONG_MIN
#define REF_VALUE( VALUE ) REF_TYPE( FUNCTION, VALUE )
#define REF_TYPE( FUNC, VAL ) TYPE_PASTE( FUNC, _ref )( VAL )
/*
* Macros for expressing constant values of the appropriate data type,
* for use in each of the derived functions.
*/
#define round_ref( VALUE ) VALUE
#define lround_ref( VALUE ) VALUE
#define llround_ref( VALUE ) VALUE
#define roundl_ref( VALUE ) TYPE_PASTE( VALUE, L )
#define lroundl_ref( VALUE ) TYPE_PASTE( VALUE, L )
#define llroundl_ref( VALUE ) TYPE_PASTE( VALUE, L )
#define roundf_ref( VALUE ) TYPE_PASTE( VALUE, F )
#define lroundf_ref( VALUE ) TYPE_PASTE( VALUE, F )
#define llroundf_ref( VALUE ) TYPE_PASTE( VALUE, F )
static __inline__
INPUT_TYPE __attribute__(( always_inline )) round_internal( INPUT_TYPE x )
#define ROUND_MODES ( FE_TONEAREST | FE_UPWARD | FE_DOWNWARD | FE_TOWARDZERO )
{
/* Generic helper function, for rounding of the input parameter value to
* the nearest integer value.
*/
INPUT_TYPE z;
unsigned short saved_CW, tmp_required_CW;
/* Rounding method suggested by Danny Smith <dannysmith@users.sf.net>
*
* Save the FPU control word state, set rounding mode TONEAREST, round the
* input value, then restore the original FPU control word state.
*/
__asm__( "fnstcw %0;" : "=m"( saved_CW ));
tmp_required_CW = ( saved_CW & ~ROUND_MODES ) | FE_TONEAREST;
__asm__( "fldcw %0;" :: "m"( tmp_required_CW ));
__asm__( "frndint;" : "=t"( z ) : "0"( x ));
__asm__( "fldcw %0;" :: "m"( saved_CW ));
/* We now have a possible rounded value; unfortunately the FPU uses the
* `round-to-even' rule for exact mid-way cases, where both C99 and POSIX
* require us to always round away from zero, so we need to adjust those
* mid-way cases which the FPU rounded in the wrong direction.
*
* Correction method suggested by Greg Chicares <gchicares@sbcglobal.net>
*/
return x < REF_VALUE( 0.0 )
? /*
* For negative input values, an incorrectly rounded value will be
* exactly 0.5 greater than the original value; when we find such an
* exact rounding offset, we must subtract an additional 1.0 from the
* rounded result, otherwise we return the rounded result unchanged.
*/
z - x == REF_VALUE( 0.5 ) ? z - REF_VALUE( 1.0 ) : z
: /* For positive input values, an incorrectly rounded value will be
* exactly 0.5 less than the original value; when we find such an exact
* rounding offset, we must add an additional 1.0 to the rounded result,
* otherwise we return the rounded result unchanged.
*/
x - z == REF_VALUE( 0.5 ) ? z + REF_VALUE( 1.0 ) : z;
}
#endif /* !defined _ROUND_INTERNAL_H: $RCSfile$: end of file */

View File

@ -1,8 +0,0 @@
#include <math.h>
float
roundf (float x)
{
/* Add +/- 0.5 then then round towards zero. */
return truncf ( x + (x >= 0.0F ? 0.5F : -0.5F));
}

View File

@ -1,8 +0,0 @@
#include <math.h>
long double
roundl (long double x)
{
/* Add +/- 0.5 then then round towards zero. */
return truncl ( x + (x >= 0.0L ? 0.5L : -0.5L));
}