ReFlO: an interactive tool for pipe-and-filter domain specification and program generation

Abstract

ReFlO is a framework and interactive tool to record and systematize domain knowledge used by experts to derive complex pipe-and-filter (PnF) applications. Domain knowledge is encoded as transformations that alter PnF graphs by refinement (adding more details), flattening (removing modular boundaries), and optimization (substituting inefficient PnF graphs with more efficient ones). All three kinds of transformations arise in reverse-engineering legacy PnF applications. We present the conceptual foundation and tool capabilities of ReFlO, illustrate how parallel PnF applications are designed and generated, and how domain-specific libraries of transformations are developed.

Understanding performance stairs

“…Operations can be either an interface or a primitive. An interface has no implementation details, other than a name, inputs, outputs, preconditions and postconditions [11]. A primitive additionally has a given code implementation (eg an API function call).…”

“…As interfaces can be in loops, this analysis is done for each time the code is run across all iterations of the loop. All of the costs are summed 11. Setting n = 1, n-best reduces to a Greedy heuristic, discussed later in Section 7.3.…”

From software extensions to product lines of dataflow programs

“…In prior work [24] we explored an MDE approach to encode Df P design knowledge as graph transformations so that a simple easy-to-understand Df P could be transformed into a complex and optimized implementation, also expressed as a Df P. That is, we transformed an initial Df P (graph), always preserving its behavior, until we reached another Df P that had the desired implementation properties regarding, for example, efficiency or availability. This Df P could then be mapped to code using a domain-specific code generator.…”

“…To summarize, our initial work [24] focused on multi-step derivations of implementations from specifications. This paper generalizes [24] and [47] to examine and reveal the rich relationships among extensions, refinements, optimizations, derivations, dataflow graphs, and SPLs. The contributions of this paper are to:…”

“…illustrate a pragmatic way to extend dataflow graphs, graph transformations, and PIM-to-PSM derivations; distinguish transformations (refinements and optimizations) from higher-order transformations (extensions); explain how extensions can be encoded to support incremental development (or reverse engineering) of Df Ps and the specification of dataflow SPLs; and describe how extensions were implemented in the ReFlO framework [24].…”