Transparent Application Acceleration by Intelligent Scheduling of Shared Library Calls on Heterogeneous Systems

Transparent application acceleration in heterogeneous systems can be performed by automatically intercepting shared libraries calls and by efficiently orchestrating the execution across all processing devices. To fully exploit the available computing power, the intercepted calls must be replaced with faster accelerator-based implementations and intelligent scheduling algorithms must be incorporated. When compared with previous approaches, the framework herein proposed does not only transparently intercepts and redirects the library calls, but it also incorporates state-of-art scheduling algorithms, for both divisible and indivisible applications. When compared with highly optimized implementations for multi-core CPUs (e.g., MKL and FFTW), the obtained experimental results demonstrate that, by applying appropriate light-weight scheduling and load-balancing mechanisms, performance speedups as high as 7.86 (matrix multiplication) and 4.6 (FFT) can be achieved.