bug-gnulib
[Top][All Lists]
Advanced

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

roundf on mingw


From: Bruno Haible
Subject: roundf on mingw
Date: Sun, 20 Apr 2008 16:15:39 +0200
User-agent: KMail/1.5.4

The roundf test fails on mingw:

bad round implementation: roundf(0.5(0x1.fffffep-2)) = 0(0x0p+0) or 1(0x1p+0)?
bad round implementation: roundf(8.38861e+006(0x1.000002p+23)) = 
8.38861e+006(0x1.000002p+23) or 8.38861e+006(0x1.000004p+23)?
...

This should fix it (similar to the test done in m4/round.m4).


2008-04-20  Bruno Haible  <address@hidden>

        * m4/roundf.m4 (gl_FUNC_ROUNDF): Add test whether roundf actually
        works.
        * doc/posix-functions/roundf.texi: Mention roundf bug on mingw.

*** doc/posix-functions/roundf.texi.orig        2008-04-20 16:14:11.000000000 
+0200
--- doc/posix-functions/roundf.texi     2008-04-20 16:05:35.000000000 +0200
***************
*** 11,16 ****
--- 11,19 ----
  @item
  This function is missing on some platforms:
  FreeBSD 5.2.1, OpenBSD 3.8, AIX 5.1, HP-UX 11, IRIX 6.5, OSF/1 4.0, Solaris 
9, Interix 3.5.
+ @item
+ This functions returns a wrong result for x = 1/2 - 2^-25 on some platforms:
+ mingw.
  @end itemize
  
  Portability problems not fixed by Gnulib:
*** m4/roundf.m4.orig   2008-04-20 16:14:11.000000000 +0200
--- m4/roundf.m4        2008-04-20 16:11:52.000000000 +0200
***************
*** 1,5 ****
! # roundf.m4 serial 4
! dnl Copyright (C) 2007 Free Software Foundation, Inc.
  dnl This file is free software; the Free Software Foundation
  dnl gives unlimited permission to copy and/or distribute it,
  dnl with or without modifications, as long as this notice is preserved.
--- 1,5 ----
! # roundf.m4 serial 5
! dnl Copyright (C) 2007-2008 Free Software Foundation, Inc.
  dnl This file is free software; the Free Software Foundation
  dnl gives unlimited permission to copy and/or distribute it,
  dnl with or without modifications, as long as this notice is preserved.
***************
*** 13,18 ****
--- 13,53 ----
    if test "$ac_cv_have_decl_roundf" = yes; then
      gl_CHECK_MATH_LIB([ROUNDF_LIBM], [x = roundf (x);])
    fi
+   if test "$ac_cv_have_decl_roundf" = yes && test "$ROUNDF_LIBM" != missing; 
then
+     dnl Test whether roundf() produces correct results. On mingw, for
+     dnl x = 1/2 - 2^-25, the system's roundf() returns a wrong result.
+     AC_REQUIRE([AC_PROG_CC])
+     AC_REQUIRE([AC_CANONICAL_HOST]) dnl for cross-compiles
+     AC_CACHE_CHECK([whether roundf works], [gl_cv_func_roundf_works],
+       [
+         save_LIBS="$LIBS"
+         LIBS="$LIBS $ROUNDF_LIBM"
+         AC_TRY_RUN([
+ #include <float.h>
+ #include <math.h>
+ int main()
+ {
+   /* 2^FLT_MANT_DIG.  */
+   static const float TWO_MANT_DIG =
+     /* Assume FLT_MANT_DIG <= 3 * 31.
+        Use the identity  n = floor(n/3) + floor((n+1)/3) + floor((n+2)/3).  */
+     (float) (1U << (FLT_MANT_DIG / 3))
+     * (float) (1U << ((FLT_MANT_DIG + 1) / 3))
+     * (float) (1U << ((FLT_MANT_DIG + 2) / 3));
+   volatile float x = 0.5f - 0.5f / TWO_MANT_DIG;
+   exit (x < 0.5f && roundf (x) != 0.0f);
+ }], [gl_cv_func_roundf_works=yes], [gl_cv_func_roundf_works=no],
+         [case "$host_os" in
+            mingw*) gl_cv_func_roundf_works="guessing no";;
+            *)      gl_cv_func_roundf_works="guessing yes";;
+          esac
+         ])
+         LIBS="$save_LIBS"
+       ])
+     case "$gl_cv_func_roundf_works" in
+       *no) ROUNDF_LIBM=missing ;;
+     esac
+   fi
    if test "$ac_cv_have_decl_roundf" != yes || test "$ROUNDF_LIBM" = missing; 
then
      REPLACE_ROUNDF=1
      AC_LIBOBJ([roundf])





reply via email to

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