Using the loop chain abstraction to schedule across loops in existing code

Bertolacci, Ian J.; Strout, Michelle Mills; Riley, Jordan; Guzik, Stephen M.; Davis, Eddie C.; Olschanowsky, Catherine

doi:10.1504/ijhpcn.2019.097053

Cited by 3 publications

(3 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, determining when inter-loop scheduling transformations such as loop fusion are legal requires an analysis of the dependences between those loops. Other loop chain approaches require specification of data access patterns by hand, which effectively rewrites the loop [10]. Our symbolic evaluation mechanism eliminates this effort for the developer by leveraging RAJA's array wrapping views abstraction.…”

Section: Gathering Access Informationmentioning

confidence: 99%

“…Figures 2a and 2b show an example of shifting the loops to adjust negative dependences. Prior work showed how to convert the dependence relations into constraints of an ILP optimization problem [10]. We minimize the sum of the non-negative shift amounts S 1,1 ,S 1,2 ...,S 2,1 ,...,S l,n with the following constraints.…”

Section: Loop Shifting For Stencil Computationsmentioning

confidence: 99%

“…Bertolacci et al [9,10] propose extensions to OpenMP pragmas to express and schedule loop chains. The approach requires annotations about data accesses in the pragmas.…”

Section: Staging and Other Loop Chain Approachesmentioning

confidence: 99%

See 2 more Smart Citations

Inter-loop optimization in RAJA using loop chains

Neth

Scogland

Supinski

et al. 2021

Proceedings of the ACM International Conference on Supercomputing

Self Cite

View full text Add to dashboard Cite

Typical parallelization approaches such as OpenMP and CUDA provide constructs for parallelizing and blocking for data locality for individual loops. By focusing on each loop separately, these approaches fail to leverage sources of data locality possible due to inter-loop data reuse. The loop chain abstraction provides a framework for reasoning about and applying inter-loop optimizations. In this work, we incorporate the loop chain abstraction into RAJA, a performance portability layer for high-performance computing applications. Using the loop-chain-extended RAJA, or RAJALC, developers can have the RAJA library apply loop transformations like loop fusion and overlapped tiling while maintaining the original structure of their programs. By introducing targeted symbolic evaluation capabilities, we can collect and cache data access information required to verify loop transformations. We evaluate the performance improvement and refactoring costs of our extension. Overall, our results demonstrate 85-98% of the performance improvements of hand-optimized kernels with dramatically fewer code changes. CCS CONCEPTS• Software and its engineering → Software libraries and repositories.

show abstract

Section: Gathering Access Informationmentioning

confidence: 99%

Section: Loop Shifting For Stencil Computationsmentioning

confidence: 99%