https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127645
Drea Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|unassigned at gcc dot gnu.org |pinskia at gcc dot
gnu.org
Status|NEW |ASSIGNED
--- Comment #7 from Drea Pinski <pinskia at gcc dot gnu.org> ---
So bitint has the same issue but bitfields are fine.
The check should be very similar to one in gimple_needing_rewrite_undefined:
```
/* Boolean loads need special handling as they are treated as a full MODE
load
and don't mask off the bits for the precision. */
if (gimple_assign_load_p (stmt)
/* Booleans are the integral type which has this non-masking issue. */
&& TREE_CODE (lhs_type) == BOOLEAN_TYPE
/* Only non mode precision booleans are need the masking. */
&& !type_has_mode_precision_p (lhs_type)
/* BFR should be the correct thing and just grab the precision. */
&& TREE_CODE (rhs) != BIT_FIELD_REF
/* Bit-fields loads don't need a rewrite as the masking
happens for them. */
&& (TREE_CODE (rhs) != COMPONENT_REF
|| !DECL_BIT_FIELD (TREE_OPERAND (rhs, 1))))
return true;
```
Except for the removal of BOOLEAN_TYPE, it should just be INTEGRAL_TYPE_P
instead. (note the check in gimple_needing_rewrite_undefined might also need a
similar change too but I have not checked it yet).
