Toward a Better Performance Portability Metric

Marowka, Ami

doi:10.1109/pdp52278.2021.00036

Cited by 8 publications

(4 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…This metric provides an overall picture of an application's performance portability, and has proved to be a useful way to summarize performance portability data. Aspects of the metric, such as normalizing performance across platforms and accounting for unsupported platforms are widely accepted, but discussion continues about ways to further improve and extend the metric [7], [8], [9].…”

Section: Measuring Performancementioning

confidence: 99%

Navigating Performance, Portability, and Productivity

et al. 2021

View full text Add to dashboard Cite

The phrase "performance portability" is commonly used, but may mean different things to different people. Developing a better appreciation of the needs of different software developers and a framework for talking about these needs improves our ability to define goals, design experiments and make forward progress. This article discusses a methodology for quantifying, summarizing, visualizing, and understanding application performance portability and programmer productivity.

show abstract

Section: Measuring Performancementioning

confidence: 99%

Navigating Performance, Portability, and Productivity

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Since the Ⴔ metric was proposed, many in the HPC community have studied it and used it in their research. As often happens with innovative ideas, over time, the limitations and shortcomings of the new Ⴔ metric were discovered 4‐10 . Some critics argue that there are significant flaws in its definition, that is difficult to understand the theory behind it, and that it is inconvenient to use.…”

Section: Introductionmentioning

confidence: 99%

“…This article compares and analyzes in detail the differences between the Ⴔ metric and the proposed

\overline{Ⴔ}

metric 4,5 . The

\overline{Ⴔ}

metric was designed to solve the problems and shortcomings that have arisen with regard to the Ⴔ metric in many studies that have used it during the 5 years since it was first published.…”

Section: Introductionmentioning

confidence: 99%

A comparison of two performance portability metrics

Marowka¹

2023

Concurrency and Computation

Self Cite

View full text Add to dashboard Cite

SummaryThe rise in the demand for new performance portability frameworks for heterogeneous computing systems has brought with it a number of proposals of workable metrics for evaluating the performance portability of applications. The aim of this article is twofold. First, we analyze the underlying principles of the criteria and definition of the revised Ⴔ metric and show that these principles are partially correct and suffer from a lack of a solid and clear performance portability model. We prove mathematically and demonstrate it practically that the principles are only correct for the architectural efficiency approach based on throughputs but are incorrect for the popular application efficiency approach based on run‐times. Second, we are examining whether the Ⴔ and metrics meet the requirements of consistency, proportionality, and lossless information. We use examples from the scientific literature to show the reader that the Ⴔ metric loses information while the metric and other metrics do not lose information.

show abstract

“…La evaluación de la portabilidad se puede dividir en dos aspectos principales: la portabilidad funcional y la portabilidad de rendimiento [101,102].…”

Section: Portabilidadunclassified

Estudio de viabilidad de SYCL como modelo de programación unificado para sistemas heterogéneos basados en GPUs en bioinformática

Costanzo

View full text Add to dashboard Cite

Por un lado, la computación de alto rendimiento (HPC) requiere de modelos de programación que aprovechen el paralelismo masivo de arquitecturas heterogéneas modernas como las plataformas CPU-GPU. Sin embargo, modelos de programación como CUDA y OpenCL presentan limitaciones en portabilidad y productividad. SYCL surge como alternativa prometedora al unificar la programación en C++ y abstraer las particularidades del hardware. Por otro lado, la bioinformática y la biología computacional representan dos campos que han estado explotando las GPUs durante más de dos décadas, y muchas de sus implementaciones se basan en CUDA, lo que impone limitaciones significativas en cuanto a la portabilidad en una amplia gama de arquitecturas heterogéneas. Es por lo que esta tesis doctoral propone evaluar la viabilidad de SYCL como modelo de programación paralelo unificado, portable y eficiente para sistemas heterogéneos con GPUs, específicamente en bioinformática. Considerando que el alineamiento de secuencias biológicas representa una operación fundamental con amplias aplicaciones en diversas áreas de la biología y la medicina, se seleccionó la suite SW\# como caso de estudio por su relevancia y por estar desarrollado en CUDA. Mediante la herramienta SYCLomatic se migró completamente el código de SW\# de CUDA a SYCL. Este proceso involucró la ejecución de la herramienta, modificación del código generado, corrección de errores, verificación funcional, optimizaciones y estandarización SYCL. Posteriormente, se realizaron múltiples experimentos en un amplio conjunto de GPUs y CPUs de diferentes tipos y fabricantes para evaluar la portabilidad de rendimiento del código migrado en contextos individuales, tanto de GPU como de CPU, multi-GPU y CPU+GPU, en un entorno híbrido. Los resultados de esta tesis muestran, en primer lugar, que la herramienta SYCLomatic resulta efectiva y útil para la migración automática, aunque no pueda considerarse una solución final. En segundo lugar, las pruebas realizadas revelan que SYCL presenta una eficiencia comparable a la de CUDA en las GPUs NVIDIA y ofrece una amplia portabilidad funcional con una aceptable portabilidad de rendimiento a GPUs y CPUs de otros fabricantes. En conclusión, SYCL se posiciona como una alternativa viable como modelo unificado de programación heterogénea, portable y eficiente, estableciendo un precedente importante para futuros desarrollos en el área.

show abstract

Toward a Better Performance Portability Metric

Abstract: The adoption of heterogeneous computing systems based on diverse architectures to achieve exascale computing power has worsened the performance portability problem of scientific applications that were designed to run on these platforms.

Cited by 8 publications

References 29 publications

Navigating Performance, Portability, and Productivity

Navigating Performance, Portability, and Productivity

A comparison of two performance portability metrics

Estudio de viabilidad de SYCL como modelo de programación unificado para sistemas heterogéneos basados en GPUs en bioinformática

Contact Info

Product

Resources

About