A Built-In Adaptive NDA for High-Level Language Acceleration
TimeWednesday, July 13th6pm - 7pm PDT
LocationLevel 2 Lobby
DescriptionNDAs have proved to tackle the memory wall effects of data-intensive applications.
Nevertheless, covering the plethora of processing demands, including algorithm, dataset, and especially the overhead of VMs that the usage of interpreted languages entails, prohibitively increases the accelerator design time.
We present an adaptive NDA for the efficient processing of data-intensive applications.
This work innovates in accelerating high-level interpreted languages, relying on built-in HW resources to identify acceleration opportunities concerning the execution demands at CPU and cache levels.
Our experiments achieve EDP improvements of up to 3.71x over a static NDA, without modifications in the software source code.