https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71149
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Status|UNCONFIRMED |NEW
Last reconfirmed| |2016-05-16
Ever confirmed|0 |1
Severity|normal |enhancement
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #1)
> Most likely the opt needs to be moved from fold-const to match.pd . Should
> be a simple patch.
Yes:
/* If this is an NE or EQ comparison of zero against the result of a
signed MOD operation whose second operand is a power of 2, make
the MOD operation unsigned since it is simpler and equivalent. */
if (integer_zerop (arg1)
&& !TYPE_UNSIGNED (TREE_TYPE (arg0))
&& (TREE_CODE (arg0) == TRUNC_MOD_EXPR
|| TREE_CODE (arg0) == CEIL_MOD_EXPR
|| TREE_CODE (arg0) == FLOOR_MOD_EXPR
|| TREE_CODE (arg0) == ROUND_MOD_EXPR)
&& integer_pow2p (TREE_OPERAND (arg0, 1)))
{
tree newtype = unsigned_type_for (TREE_TYPE (arg0));
tree newmod = fold_build2_loc (loc, TREE_CODE (arg0), newtype,
fold_convert_loc (loc, newtype,
TREE_OPERAND (arg0, 0)),
fold_convert_loc (loc, newtype,
TREE_OPERAND (arg0,
1)));
return fold_build2_loc (loc, code, type, newmod,
fold_convert_loc (loc, newtype, arg1));
}
Something like:
(for mod (ceil_mod floor_mod round_mod trunc_mod)
(for eqne (ne eq)
(simplify (eqne (mod @1 integer_pow2p@2) integer_zerop)
(if (!TYPE_UNSIGNED (TREE_TYPE (@1)))
( .... ))))
I am lazy to fill in the ... part right now but you get the idea.
