Re: [PATCH] Fix PR middle-end/28862, losing user alignment for vectors

[Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>]

> I was thinking about a portable way to test this does not happen again
> but all I could think of was looking at the asm but now I just
> remembered about the alignof expressions.  Here is a testcase which
> passes now but failed before the patch:
> float x __attribute__((aligned(128),vector_size(8)));
> int f[__alignof__(x) == 128?1:-1];

GCC may complain about too high an alignment.  Maybe just aligned(16)? 
Or a smaller vector:

char x __attribute__((aligned(4),vector_size(2)));
int f[__alignof__(x) == 4?1:-1];

Paolo