[Bug tree-optimization/21173] [4.0/4.1 regression] miscompiled pointer subtraction broke Linux kernel

["stevenb at suse dot de" <gcc-bugzilla at gcc dot gnu dot org>]

------- Additional Comments From stevenb at suse dot de  2005-04-24 09:23 -------
Subject: Re:  [4.0/4.1 regression] miscompiled pointer subtraction broke Linux kernel

On Sunday 24 April 2005 05:36, dberlin at dberlin dot org wrote:
> Uh, because it causes things to become non-invariant when they were
> originally invariant (because it will *always* create a new name for
> them).

Well, the comment before force_gimple_operand says it should not:

/* Expands EXPR to list of gimple statements STMTS.  If SIMPLE is true,
   force the result to be either ssa_name or an invariant, otherwise
   just force it to be a rhs expression.  If VAR is not NULL, make the
   base variable of the final destination be VAR if suitable.  */

tree
force_gimple_operand (tree expr, tree *stmts, bool simple, tree var)
{


tree-ssa-pre uses force_gimple_operand with SIMPLE==false, so if
expr is already a valid rhs, force_gimple_operand should do nothing.
If it does, I consider that to be a bug.

Your patch to use unshare_expr is IMHO unnecessarily expensive.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21173