HybridTuner: Tuning with Hybrid Derivative-Free Optimization Initialization Strategies

To utilize the full potential of advanced computer architectures, algorithms often need to be tuned to the architecture being used. We propose two hybrid derivative-free optimization (DFO) methods to maximize the performance of an algorithm after evaluating a small number of possible algorithmic configurations. Our autotuner (a) reduces the execution time of dense matrix multiplication by a factor of 1.4x compared to state-of-the-art autotuners, (b) identifies high-quality tuning parameters within only 5% of the computational effort required by other autotuners and (c) can be applied to any computer architecture.