Technology trends prompting architects to consider greater heterogeneity and hardware specialization have exposed an increasing need for vertically integrated research methodologies that can effectively assess perform...
详细信息
ISBN:
(纸本)9781479969982
Technology trends prompting architects to consider greater heterogeneity and hardware specialization have exposed an increasing need for vertically integrated research methodologies that can effectively assess performance, area, and energy metrics of future architectures. However, constructing such a methodology with existing tools is a significant challenge due to the unique languages, design patterns, and tools used in functional-level (FL), cycle-level (CL), and register-transfer-level (RTL) modeling. We introduce a new framework called PyMTL that aims to close this computer architecture research methodology gap by providing a unified design environment for FL, CL, and RTL modeling. PyMTL leverages the Python programming language to create a highly productive domain-specific embedded language for concurrent-structural modeling and hardware design. While the use of Python as a modeling and framework implementation language provides considerable benefits in terms of productivity, it comes at the cost of significantly longer simulation times. We address this performance-productivity gap with a hybrid JIT compilation and JIT specialization approach. We introduce SimJIT, a custom JIT specialization engine that automatically generates optimized C++ for CL and RTL models. To reduce the performance impact of the remaining unspecialized code, we combine SimJIT with an off-the-shelf Python interpreter with a meta-tracing JIT compiler (PyPy). SimJIT+PyPy provides speedups of up to 72x for CL models and 200x for RTL models, bringing us within 4-6x of optimized C++ code while providing significant benefits in terms of productivity and usability.
As graph models are applied to more widely varying fields, researchers struggle with tools for exploring and analyzing these structures. We describe GUESS, a novel system for graph exploration that combines an interpr...
详细信息
ISBN:
(纸本)9781595933720
As graph models are applied to more widely varying fields, researchers struggle with tools for exploring and analyzing these structures. We describe GUESS, a novel system for graph exploration that combines an interpreted language with a graphical front end that allows researchers to rapidly prototype and deploy new visualizations. GUESS also contains a novel, interactive interpreter that connects the language and interface in a way that facilities exploratory visualization tasks. Our language, Gython, is a domain-specific embedded language which provides all the advantages of Python with new, graph specific operators, primitives, and shortcuts. We highlight key aspects of the system in the context of a large user survey and specific, real-world, case studies ranging from social and knowledge networks to distributed computer network analysis.
暂无评论