high performance computing on graphics processing units: hgpu.org

An auto-tuning framework for GPU devices is presented for tuning application kernels of OpenCL. The GPU tuner employs multi-objective optimization methodology to improve the performance and power consumption of applications. It efficiently explores a user defined solution space comprising of possible tunable algorithmic and hardware counter variations through code transformations. The methodology targets GPU code tuning situations where performance and energy consumption are critical. The proposed framework is evaluated for 2D convolution kernels. It utilizes a non-dominated sorting Genetic Algorithm with hardware power sensor data for application code transformation through code rewrite and validation. Various algorithmic variations such as loop unrolling, caching, workgroup size and memory utilization are applied. The final pareto optimal configurations code utilized around 30% less power and 4% faster execution time. The analysis shows the convergence of optimization, and 45% improvement in standard deviation.