Re: Some VRP improvements

[Diego Novillo <dnovillo at redhat dot com>]

On Thu, Jun 02, 2005 at 10:51:15PM +0200, Andreas Jaeger wrote:

> #1  0x0000000000ee5bac in compare_values (val1=0x2aaaab1d7b60, 
> val2=0x2aaaaae00270) at /cvs/gcc/gcc/tree-vrp.c:484
> 484       gcc_assert (POINTER_TYPE_P (TREE_TYPE (val1))
> (gdb) p val1
> $1 = 0x2aaaab1d7b60
> (gdb) pt
>  <ssa_name 0x2aaaab1d7b60
>     type <pointer_type 0x2aaaaadfe5b0
>         type <void_type 0x2aaaaadfe4e0 void sizes-gimplified visited VOID
>             align 8 symtab 0 alias set -1
>             pointer_to_this <pointer_type 0x2aaaaadfe5b0>>
>         public unsigned DI
>         size <integer_cst 0x2aaaaadeac00 constant invariant 64>
>         unit size <integer_cst 0x2aaaaadeac30 constant invariant 8>
>         align 64 symtab 0 alias set -1>
>     visited var <var_decl 0x2aaaaaeb2680 D.869> def_stmt <modify_expr 
> 0x2aaaaaeb3280>
>     version 70
>     ptr-info 0x2aaaaaed0b60>
> (gdb) p val2
> $2 = 0x2aaaaae00270
> (gdb) pt
>  <integer_cst 0x2aaaaae00270 type <integer_type 0x2aaaaadf0a90 integer> 
> constant invariant 0>
> 
Yeah, exactly the same stunt that the Fortran FE was pulling on
me.

That NULL value in VAL2 should be of the same type as the SSA
name.  See http://gcc.gnu.org/ml/fortran/2005-05/msg00306.html
and the resulting patch that fixed the FE.

The C FE is doing something similar in PR 21858.


Diego.