PR tree-optimization/56918
	PR tree-optimization/56920
	* fold-const.c (int_const_binop_1): Use op1.mul_with_sign (op2, ...)
	instead of op1 - op2.  Pass 2 * TYPE_PRECISION (type) as second
	argument to rshift method.  For 2 * HOST_BITS_PER_WIDE_INT precision
	use wide_mul_with_sign method.

	* gcc.dg/vect/pr56918.c: New test.
	* gcc.dg/vect/pr56920.c: New test.

From-SVN: r197846