Convert packed signed 32-bit integers in a to packed single-precision (32-bit) floating-point elements, store the results in the lower 2 elements, then covert the packed signed 32-bit integers in b to single-precision (32-bit) floating-point element, and store the results in the upper 2 elements.