[PATCH 2/2] gdb/doublest.c: Use frexpl rather than ldfrexp.

[PATCH 2/2] gdb/doublest.c: Use frexpl rather than ldfrexp.

[Will Newton]

Most modern systems have frexpl and gnulib provides an implementation
for those that don't, so use it instead of the generic but inaccurate
ldfrexp.

gdb/ChangeLog:

2013-06-14  Will Newton  <will.newton@linaro.org>

	* doublest.c (ldfrexp): Remove function.
	(convert_doublest_to_floatformat): Call frexpl instead of
	ldfrexp.
---
 gdb/doublest.c | 49 +------------------------------------------------
 1 file changed, 1 insertion(+), 48 deletions(-)

diff --git a/gdb/doublest.c b/gdb/doublest.c
index 9ddc7a6..2e4c87e 100644
--- a/gdb/doublest.c
+++ b/gdb/doublest.c
@@ -336,53 +336,6 @@ put_field (unsigned char *data, enum floatformat_byteorders order,
     }
 }

-#ifdef HAVE_LONG_DOUBLE
-/* Return the fractional part of VALUE, and put the exponent of VALUE in *EPTR.
-   The range of the returned value is >= 0.5 and < 1.0.  This is equivalent to
-   frexp, but operates on the long double data type.  */
-
-static long double ldfrexp (long double value, int *eptr);
-
-static long double
-ldfrexp (long double value, int *eptr)
-{
-  long double tmp;
-  int exp;
-
-  /* Unfortunately, there are no portable functions for extracting the
-     exponent of a long double, so we have to do it iteratively by
-     multiplying or dividing by two until the fraction is between 0.5
-     and 1.0.  */
-
-  if (value < 0.0l)
-    value = -value;
-
-  tmp = 1.0l;
-  exp = 0;
-
-  if (value >= tmp)		/* Value >= 1.0 */
-    while (value >= tmp)
-      {
-	tmp *= 2.0l;
-	exp++;
-      }
-  else if (value != 0.0l)	/* Value < 1.0  and > 0.0 */
-    {
-      while (value < tmp)
-	{
-	  tmp /= 2.0l;
-	  exp--;
-	}
-      tmp *= 2.0l;
-      exp++;
-    }
-
-  *eptr = exp;
-  return value / tmp;
-}
-#endif /* HAVE_LONG_DOUBLE */
-
-
 /* The converse: convert the DOUBLEST *FROM to an extended float and
    store where TO points.  Neither FROM nor TO have any alignment
    restrictions.  */
@@ -466,7 +419,7 @@ convert_doublest_to_floatformat (CONST struct floatformat *fmt,
     }

 #ifdef HAVE_LONG_DOUBLE
-  mant = ldfrexp (dfrom, &exponent);
+  mant = frexpl (dfrom, &exponent);
 #else
   mant = frexp (dfrom, &exponent);
 #endif
-- 
1.8.1.4

Re: [PATCH 2/2] gdb/doublest.c: Use frexpl rather than ldfrexp.

[Mark Kettenis]

> Date: Fri, 14 Jun 2013 13:33:50 +0100
> From: Will Newton <will.newton@linaro.org>
> 
> Most modern systems have frexpl and gnulib provides an implementation
> for those that don't, so use it instead of the generic but inaccurate
> ldfrexp.

Great!  Hopefully somebody with more gnulib experience can approve the
other diff.

> gdb/ChangeLog:
> 
> 2013-06-14  Will Newton  <will.newton@linaro.org>
> 
> 	* doublest.c (ldfrexp): Remove function.
> 	(convert_doublest_to_floatformat): Call frexpl instead of
> 	ldfrexp.
> ---
>  gdb/doublest.c | 49 +------------------------------------------------
>  1 file changed, 1 insertion(+), 48 deletions(-)
> 
> diff --git a/gdb/doublest.c b/gdb/doublest.c
> index 9ddc7a6..2e4c87e 100644
> --- a/gdb/doublest.c
> +++ b/gdb/doublest.c
> @@ -336,53 +336,6 @@ put_field (unsigned char *data, enum floatformat_byteorders order,
>      }
>  }
> 
> -#ifdef HAVE_LONG_DOUBLE
> -/* Return the fractional part of VALUE, and put the exponent of VALUE in *EPTR.
> -   The range of the returned value is >= 0.5 and < 1.0.  This is equivalent to
> -   frexp, but operates on the long double data type.  */
> -
> -static long double ldfrexp (long double value, int *eptr);
> -
> -static long double
> -ldfrexp (long double value, int *eptr)
> -{
> -  long double tmp;
> -  int exp;
> -
> -  /* Unfortunately, there are no portable functions for extracting the
> -     exponent of a long double, so we have to do it iteratively by
> -     multiplying or dividing by two until the fraction is between 0.5
> -     and 1.0.  */
> -
> -  if (value < 0.0l)
> -    value = -value;
> -
> -  tmp = 1.0l;
> -  exp = 0;
> -
> -  if (value >= tmp)		/* Value >= 1.0 */
> -    while (value >= tmp)
> -      {
> -	tmp *= 2.0l;
> -	exp++;
> -      }
> -  else if (value != 0.0l)	/* Value < 1.0  and > 0.0 */
> -    {
> -      while (value < tmp)
> -	{
> -	  tmp /= 2.0l;
> -	  exp--;
> -	}
> -      tmp *= 2.0l;
> -      exp++;
> -    }
> -
> -  *eptr = exp;
> -  return value / tmp;
> -}
> -#endif /* HAVE_LONG_DOUBLE */
> -
> -
>  /* The converse: convert the DOUBLEST *FROM to an extended float and
>     store where TO points.  Neither FROM nor TO have any alignment
>     restrictions.  */
> @@ -466,7 +419,7 @@ convert_doublest_to_floatformat (CONST struct floatformat *fmt,
>      }
> 
>  #ifdef HAVE_LONG_DOUBLE
> -  mant = ldfrexp (dfrom, &exponent);
> +  mant = frexpl (dfrom, &exponent);
>  #else
>    mant = frexp (dfrom, &exponent);
>  #endif
> -- 
> 1.8.1.4
> 
> 

Re: [PATCH 2/2] gdb/doublest.c: Use frexpl rather than ldfrexp.

[Will Newton]

On 14 June 2013 13:33, Will Newton <will.newton@linaro.org> wrote:
>
> Most modern systems have frexpl and gnulib provides an implementation
> for those that don't, so use it instead of the generic but inaccurate
> ldfrexp.
>
> gdb/ChangeLog:
>
> 2013-06-14  Will Newton  <will.newton@linaro.org>
>
>         * doublest.c (ldfrexp): Remove function.
>         (convert_doublest_to_floatformat): Call frexpl instead of
>         ldfrexp.
> ---
>  gdb/doublest.c | 49 +------------------------------------------------
>  1 file changed, 1 insertion(+), 48 deletions(-)

Ping?

--
Will Newton
Toolchain Working Group, Linaro

Re: [PATCH 2/2] gdb/doublest.c: Use frexpl rather than ldfrexp.

[Joel Brobecker]

> 2013-06-14  Will Newton  <will.newton@linaro.org>
> 
> 	* doublest.c (ldfrexp): Remove function.
> 	(convert_doublest_to_floatformat): Call frexpl instead of
> 	ldfrexp.

FWIW, I read Mark's message as approval of this patch, but this looks
like an improvement for me as well. Goodbye ldfrexp :).

Thank you,
-- 
Joel

Re: [PATCH 2/2] gdb/doublest.c: Use frexpl rather than ldfrexp.

[Will Newton]

On 21 June 2013 16:56, Joel Brobecker <brobecker@adacore.com> wrote:
>> 2013-06-14  Will Newton  <will.newton@linaro.org>
>>
>>       * doublest.c (ldfrexp): Remove function.
>>       (convert_doublest_to_floatformat): Call frexpl instead of
>>       ldfrexp.
>
> FWIW, I read Mark's message as approval of this patch, but this looks
> like an improvement for me as well. Goodbye ldfrexp :).

Thanks, both patches applied.

--
Will Newton
Toolchain Working Group, Linaro