[patch, libgfortran] Some more transformational intrinsic speedups

[Dominique Dhumieres]

> Here's the revised patch, which should fix the issues that Dominique
> discovered (also the warnings, by casting the pointer to unsigned long
> int before masking it off).  Regression-tested on i686-pc-linux-gnu.
...
> +#define GFC_UNALIGNED_2(x) (((unsigned long int)(x)) & \
> +			    (__alignof__(GFC_INTEGER_2) - 1))
...

Although I did not applied the new patch, I wonder if the use of "unsigned 
long" for pointers is really platform independent (32 vs. 64 bit). Is not 
there a better choice among the zillion variants of parameters/macros 
defined at configure time (I tried to find the type of __alignof__(*), but 
did not succeed)?

Dominique