This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Prelude Patch to My Small AltiVec Patch...
On Tue, Feb 10, 2004 at 05:25:09PM -0800, Ziemowit Laski wrote:
> Here is my understanding of it:
>
> when you have code like
>
> int __attribute__((mode(v4si))) foo(void);
>
> the compiler first constructs
>
> int foo(void);
>
> and only _then_ calls the attribute handler. So, you have to have
> a way of "disassembling" the foo decl and then "reassembling" it with
> the corrected return type. Ditto for arrays & pointers.
>
> Anyhoo, the bootstrap is happily finishing, so I'll commit this
> shortly.
Ah, this makes sense if you take a look at the C (or C++) grammars and
the way GCC butchers them. This is a declaration, so the attribute is
a declspec, so it applies to the function type. Obviously for
__attribute__((mode)) the useful interpretation is to apply it to the
return type, instead.
At least, that's my understanding from the last time I had my head in
the grammar.
--
Daniel Jacobowitz
MontaVista Software Debian GNU/Linux Developer