Re: PATCH: Relax AltiVec integral parameter/literal type checking

[Ziemowit Laski <zlaski at apple dot com>]

On 12 May, 2004, at 12.53, Fariborz Jahanian wrote:

> On May 12, 2004, at 12:32 PM, Ziemowit Laski wrote:
>
> > On 12 May, 2004, at 8.48, Aldy Hernandez wrote:
> >
> > > > > > > > "Ziemowit" == Ziemowit Laski <zlaski@apple.com> writes:
> > >
> > > > I haven't modified the casting logic in the compiler, so if the
> > > > compiler allows
> > > > a cast to int or const int, then it's probably OK. :-)
> > >
> > > Please test that this is actually the case.
> >
> > IIUC, you want me to test that type casting/conversion in C and C++ 
> > work correctly, which
> > is completely orthogonal to the AltiVec issue, and also gives off a 
> > slight
> > "halting problem" vibe.  :-)  Can you be more specific?
>
> What I would like to see is that incorrect AltiVec calls are caught; 
> for example, passing
> a non-intergal argument when an integral type is expected.

That will require either a __builtin_integral_type_p, or an argument 
test right inside the
expand logic in rs6000.c.  Which one do you prefer, and do you _really_ 
think it is worth
the candle?  I don't.

>  P.S. What is IIUC (
> I have been thinking for a minute and cannot come up with the sentence 
> :).

If I Understand Correctly. :-)

--------------------------------------------------------------
Ziemowit Laski                 1 Infinite Loop, MS 301-2K
Mac OS X Compiler Group        Cupertino, CA USA  95014-2083
Apple Computer, Inc.           +1.408.974.6229  Fax .5477