PING^3 for PR target/17836, PR c/10735, PR c++/16882, PR rtl-optimization/17860[3.4]

[Janis Johnson]

On Tue, Nov 23, 2004 at 03:37:16PM -0800, Janis Johnson wrote:
> On Tue, Nov 23, 2004 at 05:49:01PM -0500, David Edelsohn wrote:
> > >>>>> Janis Johnson writes:
> > 
> > Janis> With this patch, the following testcase gets a warning for "synthetic
> > Janis> vector returned by reference" for -maltivec -mabi=no-altivec.  It's not
> > Janis> a real call, just a call to an AltiVec builtin, so the warning isn't
> > Janis> needed.
> > 
> > Janis> #include <altivec.h>
> > Janis> extern vector int a1, a2, a3;
> > Janis> void
> > Janis> foo ()
> > Janis> {
> > Janis> a3 = vec_add (a1, a2);
> > Janis> }
> > 
> > 	The only reason that it does not cause pass by reference warnings
> > is that we specifically extended the 32-bit non-Altivec ABI when not
> > ALTIVEC_ABI to pass by reference.
> > 
> > 	The two options seem to be:
> > 
> > 1) Implement similar exception for 32-bit non-Altivec ABI and return
> > vectors by reference.
> > 
> > 2) Change test for large vectors to
> > 
> > 	(TARGET_ALTIVEC ||TARGET_ALTIVEC_ABI).
> > 
> > Any opinions?
> 
> My point for this particular example is that the call is to a builtin 
> which is always inlined, so the ABI ought to be irrelevant.  The warning
> is for both C and C++.

I missed the comment about why the test does NOT get warnings, when
in fact my point is that it DOES get warnings and should not, being a
builtin.

Does the 32-bit non-AltiVec ABI cover vectors that are not 16 bytes
(either smaller or larger) or whose component type is not supported
by AltiVec vectors?

Janis