http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51799
Ira Rosen <irar at il dot ibm.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |ASSIGNED CC| |irar at il dot ibm.com AssignedTo|unassigned at gcc dot |irar at gcc dot gnu.org |gnu.org | --- Comment #4 from Ira Rosen <irar at il dot ibm.com> 2012-01-12 10:48:11 UTC --- This is actually the same problem as in pr 51301, and the fix of 51301 looks incomplete: we want an over-promoted sequence to end with type demotion operation, so we need to check that properly: Index: tree-vect-patterns.c =================================================================== --- tree-vect-patterns.c (revision 182840) +++ tree-vect-patterns.c (working copy) @@ -1186,13 +1186,15 @@ { use_lhs = gimple_assign_lhs (use_stmt); use_type = TREE_TYPE (use_lhs); - /* Support only type promotion or signedess change. Check that USE_TYPE - is not bigger than the original type. */ + /* Support only type demotion or signedess change. */ if (!INTEGRAL_TYPE_P (use_type) - || TYPE_PRECISION (new_type) > TYPE_PRECISION (use_type) - || TYPE_PRECISION (type) < TYPE_PRECISION (use_type)) + || TYPE_PRECISION (type) <= TYPE_PRECISION (use_type)) return NULL; + /* Check that NEW_TYPE is not bigger than the conversion result. */ + if (TYPE_PRECISION (new_type) > TYPE_PRECISION (use_type)) + return NULL; + if (TYPE_UNSIGNED (new_type) != TYPE_UNSIGNED (use_type) || TYPE_PRECISION (new_type) != TYPE_PRECISION (use_type)) { I am going to test this patch.