Re: attribute((mode)) pointers broken by STRIP_USELESS_TYPE_CONVERSION

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

On Oct 20, 2004, at 10:47 AM, Ulrich Weigand wrote:

> Hello,
>
> using the __attribute__ ((mode)) feature to declare pointer variables
> of non-ptr_mode mode appears to have been broken by tree-ssa.
>
> The following simple test case (on s390x):
>
> struct test;
> typedef struct test * __attribute__ ((mode (SI))) test_ptr32;
> typedef struct test * test_ptr64;
>
> int
> test (test_ptr32 a, test_ptr64 b)
> {
>   return a == b;
> }
>
> aborts in copy_to_mode_reg (explow.c:635) because the tree optimizers
> have optimized away the type conversion inserted by the front end,
> and pass a direct comparison between different modes to the expander.

Try removing the following condition:
  /* Pointers and references are equivalent once we get to GENERIC,
     so strip conversions that just switch between them.  */
  else if (POINTER_TYPE_P (inner_type)
           && POINTER_TYPE_P (outer_type)
           && lang_hooks.types_compatible_p (TREE_TYPE (inner_type),
                                             TREE_TYPE (outer_type)))
    return true;

Note you might also have to fix the C++ front-end's types_compatible_p
also, not to ignore modes.

-- Pinski