Efficient vectorization of SIMD programs with non-aligned and irregular data access hardware