Re: PATCH [RFC] for Altivec Motorola breakage

[Fariborz Jahanian <fjahanian at apple dot com>]

On Aug 9, 2004, at 3:17 PM, Janis Johnson wrote:

> On Sat, Aug 07, 2004 at 11:55:28AM -0700, Fariborz Jahanian wrote:
> > Mainline FSF, along with local implementation of Motorola's Altivec
> > syntax for
> > initialization (construction) of vector variables has broken this
> > feature, as in the
> > following test case:
> >
> > void VecMult( vector unsigned int *pResult, int Index)
> > {
> >   pResult[Index] = (vector unsigned int)(0,0,0,0);
> > }
>
> This test case, derived from yours:
>
> void VecMult (__vector unsigned int *pResult, int Index)
> {
>     pResult[Index] = (__vector unsigned int){0,0,0,0};
> }
>
> ICEs for mainline g++ with:
>
> elm3b11% /opt/gcc-nightly/mline-20040809/bin/g++ -c -maltivec fj.c
> fj.c: In function `void VecMult(unsigned int __vector__*, int)':
> fj.c:2: error: Invalid operand to binary operator
> (__vector unsigned int<D1311> *)((unsigned int<D6>)Index<D1781> * 16);
>
> fj.c:2: internal compiler error: verify_stmts failed.
> Please submit a full bug report,
> with preprocessed source if appropriate.
> See <URL:http://gcc.gnu.org/bugs.html> for instructions.
>
> Your proposed patch fixes this.  It could be useful to have a test
> case that's relevant for mainline sources.

Thanks. I tried, but apparently not hard enough :).

- Fariborz

> Janis