------- Comment #4 from rguenth at gcc dot gnu dot org 2006-03-10 14:31 -------
with 4.1 we start with
((intD.0) (short intD.7) ((short unsigned intD.8) a1D.1280 - 10000) + 10000) *
2
which convert builds from converting a - 10000 to short:
(short intD.7) ((short unsigned intD.8) a1D.1280 - 10000)
this prevents the (now latent) problem in extract_muldiv to show up.
I believe
case CONVERT_EXPR: case NON_LVALUE_EXPR: case NOP_EXPR:
/* If op0 is an expression ... */
if ((COMPARISON_CLASS_P (op0)
|| UNARY_CLASS_P (op0)
|| BINARY_CLASS_P (op0)
|| EXPRESSION_CLASS_P (op0))
/* ... and is unsigned, and its type is smaller than ctype,
then we cannot pass through as widening. */
&& ((TYPE_UNSIGNED (TREE_TYPE (op0))
&& ! (TREE_CODE (TREE_TYPE (op0)) == INTEGER_TYPE
&& TYPE_IS_SIZETYPE (TREE_TYPE (op0)))
&& (GET_MODE_SIZE (TYPE_MODE (ctype))
> GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (op0)))))
/* ... or this is a truncation (t is narrower than op0),
then we cannot pass through this narrowing. */
|| (GET_MODE_SIZE (TYPE_MODE (type))
< GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (op0))))
the last check needs to read !=, but this extract_muldiv looks like a mess...
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26630
