Evaluation of Automatic Power Reduction with OSCAR Compiler on Intel Haswell and ARM Cortex-A9 Multicores

Reducing power dissipation is one of the most important issues that need to be addressed to improve the performance of all computing systems, such as supercomputers, cloud servers, desktop PCs, medical systems, and wearable devices. Exploiting parallelism and decreasing redundant power dissipation by fine grain power control for multicore/manycore systems are promising approaches, which can ensure continuous performance improvements and reduce power dissipation. However, the manual development of parallelized applications and the embedding of power control code are both time-consuming and errorprone. The OSCAR automatic parallelization compiler has been developed to overcome these problems, which facilitates automatic low-power optimization in addition to parallelization. Though the OSCAR compiler allows these optimization, the suitability of the power optimization method for various platforms is unclear because each architecture has its own power control functionality interface. Therefore, we investigated low-power optimization with the OSCAR compiler on Intel Haswell and ARM multicore platforms to determine the efficiency of the compiler in exploiting the power control functionality of these platforms. The evaluations showed that the power consumption was reduced by 44.2% on the Intel Haswell platform having three-cores with the H.264 decoder and by 68.4% with Optical Flow on three-cores with power control compared with three-cores without power control. On the ARM cortex-A9, having three-cores with power control obtained a power reduction of 57.9% with the H.264 decoder and 67.2% with Optical Flow. These results show that the OSCAR multi-platform API resolves differences between architectures and reduces the power consumption on multiple platforms.