[Fwd: -malign-double oddities]

[Andy Polyakov]

>   > > (alignment of second, third, etc. arguments is concern of application
>   > > programmer, not compiler);
>   >
>   > Idea is that HPC programmer should inject dummy arguments to or
>   > rearrange the argument list to assure double-vise alignment. E.g.: 'func
>   > (int i, double f)' should be reimplemented as 'func (int i, int notused,
>   > double f)' or 'func(double f, int i)'.
>   >
> We don't do argument list rearrangement yet.  It's not even clear that it can
> be profitably and safely done.
To me it's *crystal* clear that it can *not* be done. Well, the only
case I can think about is statically declared functions, but does it
worth the effort? In general case the very best one can do is to assure
alignment of the first argument (by *padding* the list) and *document*
the rest in kind of "HPC notes for EGCS programmers" or something. My
whole point is that it *should* be concern of *application programmer*,
*not* compiler.

Andy.