Re: broken arithmetic operations on vectors (c and c++)

[Mark Mitchell <mark at codesourcery dot com>]

Aldy Hernandez wrote:
> > Why not just the moral equivalent of:
> >
> >  same_type_p (langhooks.signed_type(t1), langhooks.signed_type(t2))
>
>
> Oooo.  Nice.
>
> Ok, I used the equality operator, since that's how it seems we compare
> types throughout the FEs.
>
> Here is an updated patch which I'm testing.  It's for the C and C++
> front ends.  C was borked as well.
>
> In testing I noticed gcc.dg/simd-1b.c had most of the tests
> XFAILed.  They now work.  I've re-enabled them.
>
> How does this look?

I'm concerned about the pointer-equality test, but, because 
c_common_signed_type grabs the TYPE_MAIN_VARIANT, it should work for 
scalars, as RTH says.  I think that to make this a bit more bulletproof, 
you should (a) add a comment noting that c_common_signed_type does that, 
and (b) assert that the inputs are INTEGER_TYPE or REAL_TYPE.  Perhaps 
even change the name of the function to 
same_scalar_type_ignoring_signedness.

The point of all that is that "signed_type (array_type1) == signed_type 
(array_type2)" might not be true, even if the array types are the same, 
and that the C signed type hooks will accept such input, even though its 
weird.  So, an assertion to make clear that your function won't accept 
these would be nice.

The C++ change is OK.

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304