Egeria: A Framework for Auto-Construction of HPC Advising
Tools through Multi-Layered Natural Language Processing
Authors
Event Type
Paper
Natural Language Processing
Performance
Programming Systems
TimeTuesday, November 14th10:30am -
11am
Location301-302-303
DescriptionVendors often provide some detailed programming guides
to assist programmers in developing high performance
programs. However, these guides are frequently hundreds
of pages long, making it difficult for general
programmers to master and memorize all the rules and
guidelines and properly apply them to a specific problem
instance.
In this work, we develop a framework named Egeria to alleviate the difficulty. Through Egeria, one can easily construct an advising tool for a certain high performance computing (HPC) domain (e.g., GPU programming). Egeria is made possible through a distinctive multi-layered design that leverages the properties of HPC domains and overcomes the weaknesses of existing Natural Language Processing (NLP) techniques. Experiments on CUDA and OpenCL programming demonstrate the usefulness of Egeria for HPC both qualitatively and quantitatively.
In this work, we develop a framework named Egeria to alleviate the difficulty. Through Egeria, one can easily construct an advising tool for a certain high performance computing (HPC) domain (e.g., GPU programming). Egeria is made possible through a distinctive multi-layered design that leverages the properties of HPC domains and overcomes the weaknesses of existing Natural Language Processing (NLP) techniques. Experiments on CUDA and OpenCL programming demonstrate the usefulness of Egeria for HPC both qualitatively and quantitatively.
Download PDF:
here
