High-level mapping framework for heterogeneous architectures with FPGAs and GPUs
| Grant number: | 16/18937-7 |
| Support type: | Scholarships in Brazil - Doctorate (Direct) |
| Effective date (Start): | June 01, 2017 |
| Effective date (End): | February 28, 2021 |
| Field of knowledge: | Physical Sciences and Mathematics - Computer Science |
| Principal Investigator: | Vanderlei Bonato |
| Grantee: | Andre Bannwart Perina |
| Home Institution: | Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil |
| Associated scholarship(s): | 18/22289-6 - High-level mapping framework for heterogeneous architectures with FPGAs and GPUs, BE.EP.DD |
Abstract Over many decades, a broad spectrum of digital architectures were developed for solving different types of problems in terms of their characteristics. Some years ago, trends have emerged towards Heterogeneous Computing, where different architectures (e.g. CPUs, GPUs, FPGAs) are used for accelerating large-scale problems. In this method of computation, parts of a source code are separated (usually by wrapping them to functions, known as kernels) to be executed on the heterogeneous accelerators. The task of making such separation efficiently, however, is not trivial. There are many investigations in this area, such as kernel mapping (assigning defined kernels to the cores based on parameters such as data dependency, control flow, etc.), kernel transformation (transforming the code between kernels in order to increase performance, e.g. kernel fusion or fission) among others. Reconfigurable architectures are suitable for heterogeneous computing due to its acceptable performance associated with low energy consumption. In this project, we propose the creation of an integrated design space exploration framework involving mainly GPUs and FPGAs, where kernels mapping and transformations will be applied not only to increase performance but also reducing energy consumption, moving towards the era of green computing. (AU) | |