Type-Directed Program Synthesis and Constraint Generation for Library Portability.

摘要: Fast numerical libraries have been a cornerstone of scientific computing for decades, but this comes at price. Programs may be tied to vendor specific software ecosystems resulting in polluted, non-portable code. As we enter an era heterogeneous computing, there is explosion the number accelerator required harness specialized hardware. We need system that allows developers exploit ever-changing libraries, without over-specializing their As cannot know behavior future ahead time, paper develops scheme assists matching code new requiring source these libraries. Furthermore, it can recover equivalent from programs use existing and automatically port them interfaces. It first uses program synthesis determine meaning library, then maps synthesized description into generalized constraints which are used search replacement opportunities present developer. We applied approach large applications deep learning domains. Using our approach, show speedups ranging 1.1$\times$ over 10$\times$ on end performance when using