Speakers
Description
This lecture presents notable Fortran features such as array slices and array expressions (incl. caveats with temporaries), elemental functions, and extends the principles to explicit SIMD vectorization based on the OpenMP standard. This part concludes on how to target modern vector extension using compiler flags beyond an umbrella optimization setting.
OpenMP constructs targeting thread-level parallelism may be covered by an own lecture; however a collection of common code patterns is presented that can help to improve multicore scalability (synchronization, thread-local storage). The lecture concludes with an Application Performance Snapshot using Intel VTune Amplifier, a tool that helps to not only speculate about performance, but rather analyze a given code at an algorithmic as well as target-specific level.
