Program Semantics Visualizations (PSVs) are dynamic visualizations of program execution that help programmers understand language semantics. The Sidewinder research program develops design criteria for PSV tools, provides a formal theory for their construction, and reifies that theory in a PSV prototyping framework.
Theia is a generic framework for program state visualization that uses abstract machine definitions to generate complete visualizations.
The paper discusses the relationship between abstract machines and notional machines and identifies completeness, continuity, and consistency as crucial properties for dynamic program visualizations.
Relay is an intermediate representation (IR) for the machine learning compiler stack TVM. TVM powers machine learning inference tasks at Amazon and underpins OctoML Inc.
This paper discusses early work on Relay and the potential benefits of a functional IR over graph IRs underlying PyTorch and TensorFlow.
Machine learning framework IRs must be expressive, composable, and portable. Relay’s design addresses these design criteria without compromising performance.