From Serial Loops to Parallel Execution on Distributed Systems

Abstract

Abstract. Programmability and performance portability are two major challenges in today's dynamic environment. Algorithm designers targeting efficient algorithms should focus on designing high-level algorithms exhibiting maximum parallelism, while relying on compilers and runtime systems to discover and exploit this parallelism, delivering sustainable performance on a variety of hardware. The compiler tool presented in this paper can analyze the data flow of serial codes with imperfectly nested, affine loop-nests and if statements, commonly found in scientific applications. This tool operates as the front-end compiler for the DAGuE run-time system by automatically converting serial codes into the symbolic representation of their data flow. We show how the compiler analyzes the data flow, and demonstrate that scientifically important, dense linear algebra operations can benefit from this analysis, and deliver high performance on large scale platforms.

Evaluation of dataflow programming models for electronic structure theory

“…While the metadata discovery phase is akin to the task submission process that takes place in implicit dataflow programming models, the development of the PTG representation of the dataflow increases the complexity of code development significantly and diverges from the original program. In some cases, even the PTG creation can be automated (by a polyhedral analysis compiler tool), but these cases are limited to highly regular affine codes, and NWChem includes functions with side effects and statically undecidable branches, and so it is far from being affine and out of reach for current compiler technologies.…”

Polyhedral Dataflow Programming: A Case Study

“…This approach is more a runtime approach than a language approach, and is specific to GPUs. In [37], the authors propose a polyhedral-based precompilation phase for their runtime system DAGuE, in order to expose data exchange information that is further used by the runtime. Our approach is more "friendly" to the compiler since the SigmaC language makes some of these communications explicit.…”

Automatic Task-Based Code Generation for High Performance Domain Specific Embedded Language

“…• PaRSEC [7] (or DAGuE) is a generic framework developed by the ICL (University of Tennessee) that can be used to extract data-flow patterns from a sequential C Code, generating a DAG representation at compiletime. PaRSEC uses a proper runtime to instantiate this DAG in the form of computation tasks.…”