This is the mail archive of the mailing list for the glibc project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] powerpc: Fix lgammal_r overflow warnings

On 10-12-2014 19:17, Joseph Myers wrote:
> On Wed, 10 Dec 2014, Adhemerval Zanella wrote:
>> +static const long double huge = HUGE_VALL;
> No, HUGE_VALL is not correct; that's infinity, which would lose the 
> overflow exceptions and rounding-mode-dependent results.  (And the 
> ChangeLog entry should name the variable whose initializer is being 
> changed.)

Indeed I overlooked it, below corrected patch:

	* sysdeps/ieee754/ldbl-128/e_lgammal_r.c: Remove overflow warnings for
	constants definition.
	(huge): Change value to LDBL_MAX.


diff --git a/sysdeps/ieee754/ldbl-128/e_lgammal_r.c b/sysdeps/ieee754/ldbl-128/e_lgammal_r.c
index 1961355..2124d62 100644
--- a/sysdeps/ieee754/ldbl-128/e_lgammal_r.c
+++ b/sysdeps/ieee754/ldbl-128/e_lgammal_r.c
@@ -70,12 +70,20 @@

 #include <math.h>
 #include <math_private.h>
+#include <libc-internal.h>
+#include <float.h>

+/* BZ#16347: ldbl-128ibm uses this file as is, however the MAXLGM
+   definition overflows for IBM long double.  This directive prevents the
+   overflow warnings until IBM long double version is not fixed.  */
 static const long double PIL = 3.1415926535897932384626433832795028841972E0L;
+DIAG_IGNORE_NEEDS_COMMENT (4.6, "-Woverflow");
 static const long double MAXLGM = 1.0485738685148938358098967157129705071571E4928L;
 static const long double one = 1.0L;
 static const long double zero = 0.0L;
-static const long double huge = 1.0e4000L;
+static const long double huge = LDBL_MAX;

 /* log gamma(x) = ( x - 0.5 ) * log(x) - x + LS2PI + 1/x P(1/x^2)
    1/x <= 0.0741 (x >= 13.495...)

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]