Type-Directed Program Synthesis and Constraint Generation for Library Portability

Abstract

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

Automatically harnessing sparse acceleration

“…The paper focuses on code detection, with manual data marshaling. Recent work [13] uses type-guided program synthesis to model library routines, which are then detected by a solver. Again, data marshaling is not taken into account.…”

M 3

“…No further information about the target function is required. For example, base cases or semantic annotations (such as those used by λ 2 [19] or in the type-directed synthesis procedure demonstrated by Collie et al [11]) are not required by our implementation, and we do not require manually created inputs to test candidate programs as other synthesizers such as Simpl [47] or SketchAdapt [33] do.…”

“…We Figure 4: Proportion of each library's API that we were able to successfully synthesize, across all functions in the library as well as those with compatible type signatures. Results are also shown for TypeDirect [11].…”

“…We then selected five mathematical libraries with slightly different areas of application and platform support in order to evaluate the ability of M 3 to discover cross-vendor or cross-platform migrations. Other work [11] identifies the usefulness of this type of migration. Full details of the selected libraries are given in Table 1b.…”

“…We were only able to identify one other program synthesizer with library functions and migrations as an explicitly stated goal. In that work, partial semantic knowledge and type information is used to guide a synthesizer (TypeDirect) towards synthesized implementations of performance bottleneck functions [11] Evaluation of TypeDirect is limited to 12 such functions, with synthesis guided by annotations that specify semantic properties of the target functions [11]. We restated our set of synthesis problems for TypeDirect and recorded how many it could synthesize.…”