Enabling performance portability of data-parallel OpenMP applications on asymmetric multicore processors

Abstract

Asymmetric multicore processors (AMPs) couple high-performance big cores and low-power small cores with the same instruction-set architecture but different features, such as clock frequency or microarchitecture. Previous work has shown that asymmetric designs may deliver higher energy efficiency than symmetric multicores for diverse workloads. Despite their benefits, AMPs pose significant challenges to runtime systems of parallel programming models. While previous work has mainly explored how to efficiently execute task-based parallel applications on AMPs, via enhancements in the runtime system, improving the performance of unmodified data-parallel applications on these architectures is still a big challenge. In this work we analyze the particular case of loop-based OpenMP applications, which are widely used today in scientific and engineering domains, and constitute the dominant application type in many parallel benchmark suites used for performance evaluation on multicore systems. We observed that conventional loop-scheduling OpenMP approaches are unable to efficiently cope with the load imbalance that naturally stems from the different performance delivered by big and small cores.To address this shortcoming, we propose Asymmetric Iteration Distribution (AID), a set of novel loop-scheduling methods for AMPs that distribute iterations unevenly across worker threads to efficiently deal with performance asymmetry. We implemented AID in libgomp -the GNU OpenMP runtime system-, and evaluated it on two different asymmetric multicore platforms. Our analysis reveals that the AID methods constitute effective replacements of the static and dynamic methods on AMPs, and are capable of improving performance over these conventional strategies by up to 56% and 16.8%, respectively. CCS CONCEPTS• Computer systems organization → Multicore architectures; Heterogeneous (hybrid) systems; • Software and its engineering → Runtime environments; Parallel programming languages.

Flexible system software scheduling for asymmetric multicore systems with PMCSched: A case for Intel Alder Lake

“…For legacy platforms not officially supporting kernel versions newer than v5.8. y (such as the ARM-based Odroid-XU4 board 11,52 ), PMCSched can still be used, but requires a patched Linux kernel that either includes the necessary ftrace (backported) bugfixes, or incorporates the changes of the old PMCTrack patch.…”

“…2,[7][8][9][10] When a single multithreaded application runs alone on an AMP system, smart user-level scheduling within the runtime system is the key to making the most out of its heterogeneous cores. 7,11 However, in multi-application scenarios, and especially under the presence of legacy programs, the OS scheduler plays an essential role in transparently delivering the benefits of AMPs to the end user. 2,9,12,13 In our work, we focus on designing and implementing effective OS and runtime-level schedulers to deal with single-and multi-application scenarios, by leveraging smart interaction between system software layers when possible.…”

“…Many of these applications are broadly used in scientific and engineering domains, and constitute the dominant application type in many parallel benchmark suites extensively used for performance evaluation on symmetric and asymmetric multicore systems. 7,11,[14][15][16][17][18][19] Some of our insights on effective OS-runtime interaction techniques could be leveraged for other types of programs, such as the increasingly popular task-based parallel applications, where runtime scheduling has also drawn special attention. 10,[20][21][22] Previous research has demonstrated that the runtime system and the OS scheduler can perform optimizations on AMPs by leveraging hardware features that are directly controlled by the OS kernel and exposed to user space, such as performance monitoring counters (PMCs) 4,12,23 or dynamic voltage and frequency scaling (DVFS).…”

“…• We propose flexible asymmetric-iteration distribution (AID), a set of enhancements for improved AID 11 of parallel loops in OpenMP programs.…”

Preparing for the Future -- Rethinking Proxy Apps

“…Unfortunately, and despite best efforts by the HPC community, achieving performance portability across slightly or widely varying architectures is still an unobtainable goal. Modern features in OpenMP or alternative programming paradigms, such as OpenCL, demonstrate that applications can be written (often without separate code paths for different architectures) in a way to easily migrate between different CPUs and GPUs and vendors while preserving correctness, however usually such migration results in severe performance drops [39][40][41] . Hence, manual code refactoring is still required, to change algorithms, data layouts, parallelization strategies, etc., to fully utilize any given 3/10 architecture.…”