OpenMP, the de-facto standard for intra-node parallelism, is designed for ease of parallel programming on most modern hardware. It provides users with presets of scheduling and configurations so that users can choose one of those settings for their applications with their domain knowledge. However, it doesn’t give users enough freedom to apply knowledge in scheduling of constructs in OpenMP. This functionality is especially useful for HPC users who have domain specific knowledge which can be used for better optimized scheduling. In this project, we propose some APIs where user can apply their own scheduling algorithms with domain knowledge on applications.

Bio: PhD student in Georgia Tech, transferred from UIUC this summer. Working on integrated programming models and task level parallelism. Integrated OpenMP into Charm++ / AMPI programming framework and presented in CCGrid 2018. Currently studying various scheduling issues in programming runtime especially on OpenMP.