查看文章

Recode Software is developed in C++ and Python to evaluate Recoding engine (aka Unstructured Data Processor or UDP)[1] on a wide range of applications. Recoding engine is an efficient, software programmable, data recoding accelerator that is orders of magnitude more energyefficient (1,900 x) and area-efficient than conventional CPU cores on recoding tasks [1-4]. With Recoding engine, systems/applications can select the right encoding of data for each stage of the computation, and in each part of the systems (eg, core, LLC, main memory and storage). Please check [1] to learn more about the Recoding Engine (UDP) architectures. With Recode Software, users can simply call library functions that have been built for the Recoding engine and perform an efficient data transformation and encodings on diverse workloads. Currently, the Recode Software supports the data transformation workloads including CSV parsing, Huffman Encoding/Decoding, Dictionary, Dictionary-RLE (Run-length Encoding) and Snappy Compression and Decompression [1].

The target users for Recode Software are application software developers and studiers who are familiar with C-level source interface. These users can call the standard Recode Software library for C++ program, and use the interfaces to accelerate data transformations workloads and generate performance results. Moreover, Recode Software has been also provided with a basis programming environment for software tool chain developers to be able to develop a compiler which will recognize patterns, eg, sparse matrix operations, and automatically target the Recode engine.