RFC: PR37472 bad output on default-format write of double in common block with -m64

[Steve Kargl]

On Sun, Dec 21, 2008 at 11:09:34PM -0800, Jerry DeLisle wrote:
> Steve Kargl wrote:
> 
> >I suspect the problem lies in 0.1.  It isn't exactly representable
> >and you're subject to rounding errors.  Can you use 
> >
> >temp = (calculate_exp_ ## x (mid) - 5 * calculate_exp_ ## x (mid - d - 1));
> >temp /= 10;
> >
> 
> Steve, you hit the mark.
> 
> This patch works:
> 
> Index: write_float.def
> ===================================================================
> --- write_float.def	(revision 142883)
> +++ write_float.def	(working copy)
> @@ -640,7 +640,7 @@ output_float_FMT_G_ ## x (st_parameter_d
>        GFC_REAL_ ## x temp;\
>        mid = (low + high) / 2;\
>  \
> -      temp = 0.1 * calculate_exp_ ## x (mid) - 0.5\
> +      temp = calculate_exp_ ## x (mid)/10 - 0.5\
>  	     * calculate_exp_ ## x (mid - d - 1);\
>  \
>        if (m < temp)\

Looks good to me.  For some unknown reason, I tend to think about
floating point and rounding errors more than I should.

-- 
Steve