Name
Title | ||
|---|---|---|
Polyhedral fragments: an efficient representation for symbolically generating code for processor arrays |
Abstract | ||
|---|---|---|
To leverage the vast parallelism of loops, embedded loop accelerators often take the form of processor arrays with many, but simple processing elements. Each processing element executes a subset of a loop's iterations in parallel using instruction- and datalevel parallelism by tightly scheduling iterations using software pipelining and packing instructions into compact, individual programs. However, loop bounds are often unknown until runtime, which complicates the static generation of programs because they influence each program's control flow.
Existing solutions, like generating and storing all possible programs or full just-in-time compilation, are prohibitively expensive, especially in embedded systems. As a remedy, we propose a hybrid approach introducing a tree-like program representation, whose generation front-loads all intractable sub-problems to compile time, and from which all concrete program variants can efficiently be stitched together at runtime. The tree consists of so-called polyhedral fragments that represent concrete program parts and are annotated with iteration-dependent conditions.
We show that both this representation is both space- and time-efficient: it requires polynomial space to store---whereas storing all possibly generated programs is non-polynomial---and polynomial time to evaluate---whereas just-in-time compilation requires solving NP-hard problems. In a case study, we show for a representative loop program that using a tree of polyhedral fragments saves 98.88 % of space compared to storing all program variants.
|
Year | DOI | Venue |
|---|---|---|
2019 | 10.1145/3359986.3361205 | Proceedings of the 17th ACM-IEEE International Conference on Formal Methods and Models for System Design |
Field | DocType | ISBN |
Computer science,Theoretical computer science,Computational science | Conference | 978-1-4503-6997-8 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
3 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Michael Witterauf | 1 | 15 | 6.36 |
| Frank Hannig | 2 | 595 | 75.66 |
| Juergen Teich | 3 | 90 | 18.01 |