The paper treats a method of code generations which provides precise solution of code selection and scheduling task under restrictions on the number of available registers. The method allows for maximal use of instruction-level parallelism. In case of register pressure, the method provides optimal spill code generation or recalculation of register values.