|PARRAVICINI ALBERTO||Cycle: XXXIV |
Section: Computer Science and Engineering
Tutor: MIRANDOLA RAFFAELA
Advisor: SANTAMBROGIO MARCO DOMENICO Major Research topic
:Gospel - Leveraging runtime environments for heterogeneous hardware accelerationAbstract:
The end goal of the research project is to leverage managed runtime environments (such as GraalVM) to provide heterogeneous hardware acceleration (especially GPU-based acceleration) to high-level languages, improving code performance and lowering the barrier of access to hardware acceleration.
It is studied how to exploit Just-In-Time compilation of GPU code to provide the GPU with information known only at run time, such as array sizes: through this information, it is possible to modify the GPU code Intermediate Representation to introduce additional checks against out-of-bounds array accesses, improving the robustness of GPU code without any detriment to performance.
In this research, it is also addressed the problem of GPU multi-task computations, through the development of an optimized GPU scheduler that is capable of computing dependencies between tasks using information available at run time to transparently enable parallel execution of computations on GPU and asynchronous CPU-GPU computations.