Stalled Patches

[Paul Brook]

On Friday 02 April 2004 12:51, Joost VandeVondele wrote:
> It has nothing to do with the accuracy of the math libraries, the question
> is if the constant needs to be converted to real*4 before
> assignment in
>
> real*8 x=1.7
>
> as done in most compilers, or if it is OK to assign the real*8
> approximation.

Ah, I see. I think the GMP->tree conversion is the wrong place to be doing 
this, so the patch is still ok.

The frontend doesn't accurately model the target floating point arithmetic, so 
there are probably other places where we may have excess precision or 
incorrect rounding.

It has previously been mentioned that the GMP MPFR routines may provide the 
functionality to do this with relatively few changes. I don't think anyone 
has actually attempted to implement this though.

Paul