There are command line options within the <offload> compiler which allow for the output of performance advice. This advice can help developers avoid expensive reads through the software cache on SPU, or rearrange data for better alignment. What is important is that the compiler conveys this information so the programmers can take advantage of this knowledge through as series of iterative steps from initial offload block implementation all the way through to shipped <offload> powered product.

See for example see article 80.