Compile-time Code Parallelization For Self-Adapting Acceleration On Intel Xeon Phi

  • In this thesis we present a prototype compiler capable of automatically detecting paral-lelizable loops and of creating code that leverages a target platform’s parallelism features in a way that improves performance. Our ultimate goal will be to provide a tool that lifts the burden of having to write parallel code for loops manually to profit from today’s parallel hardware from a developer’s shoulders.