Re: bk commit into 5.0 tree (kaa:1.2597) BUG#15936

[Sergei Golubchik]

Hi!

ok to push

On Mar 14, Alexey Kopytov wrote:
> ChangeSet@stripped, 2008-03-14 13:25:37+03:00, kaa@kaamos.(none) +4 -0
>   Fix for bug #15936: "round" differs on Windows to Unix
>   
>   Both of our own implementations of rint(3) were inconsistent with the
>   most common behavior of rint() on those platforms that have it: round
>   to nearest, break ties by rounding to nearest even.

perhaps we need to call fesetround() in mysqld to set
rounding mode to nearest even explicitly ?
   
>   Fixed by leaving just one implementation of rint() in our source tree,
>   and changing its behavior to match the most common native
>   implementations on other platforms.
> 
> diff -Nrup a/include/my_global.h b/include/my_global.h
> --- a/include/my_global.h	2007-11-30 03:37:04 +03:00
> +++ b/include/my_global.h	2008-03-14 13:25:36 +03:00
> @@ -483,9 +483,42 @@ typedef unsigned short ushort;
>  #define test_all_bits(a,b) (((a) & (b)) == (b))
>  #define set_bits(type, bit_count) (sizeof(type)*8 <= (bit_count) ? ~(type) 0 :
> ((((type) 1) << (bit_count)) - (type) 1))
>  #define array_elements(A) ((uint) (sizeof(A)/sizeof(A[0])))
> +
>  #ifndef HAVE_RINT
> -#define rint(A) floor((A)+(((A) < 0)? -0.5 : 0.5))
> -#endif
> +/**
> +  All integers up to this number can be represented exactly as double precision
> +  values (DBL_MANT_DIG == 53 for IEEE 754 hardware).
> +*/
> +#define MAX_EXACT_INTEGER ((1LL << DBL_MANT_DIG) - 1)
> +
> +/**
> +  rint(3) implementation for platforms that do not have it.
> +  Always rounds to the nearest integer with ties being rounded to the nearest
> +  even integer to mimic glibc's rint() behavior in the "round-to-nearest"
> +  FPU mode. Hardware-specific optimizations are possible (frndint on x86).
> +  Unlike this implementation, hardware will also honor the FPU rounding mode.
> +*/
> +
> +inline double rint(double x)

use 'static inline'

> +{
> +  double f, i;
> +  f = modf(x, &i);

assuming, modf() is more portable than rint()...

> +
> +  /*
> +    All doubles with absolute values > MAX_EXACT_INTEGER are even anyway,
> +    no need to check it.
> +  */
> +  if (x > 0.0)
> +    i += (double) ((f > 0.5) || (f == 0.5 &&
> +                                 i <= (double) MAX_EXACT_INTEGER &&
> +                                 (longlong)i % 2));
> +  else
> +    i -= (double) ((f < -0.5) || (f == -0.5 &&
> +                                  i >= (double) -MAX_EXACT_INTEGER &&
> +                                  (longlong)i % 2));
> +  return i;
> +}
> +#endif /* HAVE_RINT */
>  
Regards / Mit vielen Grüssen,
Sergei

-- 
   __  ___     ___ ____  __
  /  |/  /_ __/ __/ __ \/ /   Sergei Golubchik <serg@stripped>
 / /|_/ / // /\ \/ /_/ / /__  Principal Software Developer/Server Architect
/_/  /_/\_, /___/\___\_\___/  MySQL GmbH, Dachauer Str. 37, D-80335 München
       <___/                  Geschäftsführer: Kaj Arnö - HRB
München 162140