Select precise floating-point model, otherwise doubles are less than 64-bit wide

icc test adjustment:

  Select c89 dialect