Applications Acceleration through Adaptive Hardware Components
Vito Giovanni Castellana, Fabrizio Ferrandi · 2013
High Level Synthesis (HLS) provides automatic flows for the generation of hardware accelerators starting from their behavioral description. HLS guarantees results comparable to hand-written design for some applications domains such as Digital Signal Processing. However, it is not yet able to cope with performance requirements when scaling the application complexity. One of the biggest limitation is an execution paradigm still based on the construction of a centralized Finite State Machine (FSM). Parallelism exploitation is thus bound to Instruction Level Parallelism within a single execution flow. This is in contrast to the current trends for hardware architectures and programming languages, which are progressively moving towards execution paradigms dominated other type of parallelisms, such as Task or Thread Level Parallelism. This work proposes a novel adaptive accelerator design, not based on the FSM execution paradigm, which provides support to dynamic parallel execution. Execution is parallel, because no pre-computed scheduled is considered. Operations are directly managed by dedicated lightweight hardware modules, which directly communicate to notify execution completion and to start other dependent operations. Execution is parallel, because several execution flows may run concurrently. The proposed design targets different application domains, from Embedded Systems accelerators to hybrid high-performance architectures.