Hello,

This patch introduces two utility functions:

	align_up (v, n);
	align_down (v, n);

for [re]aligning addresses vis:

+   addr = align_up (addr, 8); -- VALUE needs 8 byte alignment
+   write_memory (addr, value, len);
+   addr += len;

+   sp = align_down (sp - len, 16); -- Keep SP 16 byte aligned
+   write_memory (sp, value, len);

It then goes through and replaces all occurances of round_up / 
round_down and align_up / align_down with these globals.

You'll notice that I chose align_XXX rather than round_XXX.  I think 
this better reflects the intended usage.  I've noticed a lot of code doing:

	write_memory (addr, data, len);
	addr += round_up (len, 16);

instead of:

	addr = align_up (addr, 16);
	write_memory (addr, data, len);
	addr += len;

as the former may not result in ADDR having the required alignment.  The 
PPC SVr4 Altivec ABI, for instance, switches between 8 and 16 byte 
alignment making the former code very wrong.

anyway, thoughts?

I'll pick it up in a week,
Andrew