Re: SSE ABI change proof for Jan's old patch

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Dec 29, 2003, at 20:29, Richard Henderson wrote:

> On Mon, Dec 29, 2003 at 06:58:32PM -0500, Daniel Berlin wrote:
> > This is SSE. arch=i686 doesn't support SSE, so you'd never be able to
> > compile the same code anyway.
>
> You forget about our generalized vector support.
>
> 	typedef __attribute__((vector_size(16))) int v4si;
>
> 	v4si foo(v4si x, v4si y)
> 	{
> 	  return x + y;
> 	}
>
> 	void bar (v4si x, v4si y, v4si *p)
> 	{
> 	  *p = x + y;
> 	}
>
> I do note that we already warn:
>
> z.c: In function `foo':
> z.c:4: warning: SSE vector return without SSE enabled changes the ABI
>
> so I suppose a similar warning for parameter passing would be
> acceptable.

But GCC does not warn for use of generalized vector when using on PPC 
which does
change ABI when you use altivec, either with return values or parameter 
passing.
Why do we want one for x86 or should we add one for PPC also?

Thanks,
Andrew Pinski