Energy-efficient execution of data-parallel applications on heterogeneous mobile platforms

Abstract

Abstract-State-of-the-art mobile system-on-chips (SoC) include heterogeneity in various forms for accelerated and energyefficient execution of diverse range of applications. The modern SoCs now include programmable cores such as CPU and GPU with very different functionality. The SoCs also integrate performance heterogeneous cores with different power-performance characteristics but the same instruction-set architecture such as ARM big.LITTLE. In this paper, we first explore and establish the combined benefits of functional heterogeneity and performance heterogeneity in improving power-performance behavior of data parallel applications. Next, given an application specified in OpenCL, we present a static partitioning strategy to execute the application kernel across CPU and GPU cores along with voltage-frequency setting for individual cores so as to obtain the best power-performance tradeoff. We achieve over 19% runtime improvement by exploiting the functional and performance heterogeneities concurrently. In addition, energy saving of 36% is achieved by using appropriate voltage-frequency setting without significantly degrading the runtime improvement from concurrent execution.

Reliable mapping and partitioning of performance-constrained openCL applications on CPU-GPU MPSoCs

“…In [20], a similar open source framework, FreeOCL [3] is used for the ARM CPU that acts as both the host processor and an OpenCL device. This enables concurrent use of CPU and GPU to execute an application threads, but in [20], a static partitioning is performed by using all the CPU and GPU cores.…”

“…OpenCL [2] provides an opportunity to write programs that can execute across heterogeneous cores including CPUs and GPUs [13,15,17,20]. However, depending upon the kind of parallelism dominant in the application, the performance, energy consumption and temperature will vary when it is allocated onto only CPU, only GPU, or both CPU and GPU cores.…”

“…The application codes are slightly modified to launch them only on CPU cores, only on GPU cores, or on both CPU and GPU cores. Additionally, an appropriate work-group size for each application is selected as in [20].…”

Energy-Efficient Run-Time Mapping and Thread Partitioning of Concurrent OpenCL Applications on CPU-GPU MPSoCs

“…They usually consider Pthreads programming model and cannot be used to simultaneously exploit cores of di erent ISAs such as CPU and GPU. For a single application, simultaneous exploitation of CPU and GPU cores has been performed by employing OpenCL programming model [30], but it leads to poor results when applied to concurrent applications (shown in the next section). Additionally, some works exploit either CPU or GPU for an application [37], which is not e cient in terms of execution time and energy consumption as shown in Figure 1.…”

“…Additionally, CPU cores typically handle task and thread level parallelisms, whereas GPU cores handle data level parallelism. OpenCL [7] provides an opportunity to write programs that can execute across heterogeneous cores including CPUs and GPUs [18,19,24,30]. However, depending upon the kind of parallelism dominant in the application, the performance and energy consumption will vary when it is allocated onto only CPU, only GPU, or both CPU and GPU cores.…”