Cooperative and out-of-core execution of the irregular wavefront propagation pattern on hybrid machines with IntelⓇ Xeon Phi™

摘要

The Irregular Wavefront Propagation Pattern (IWPP) is a core computing structure in severalrnimage analysis operations. Efficient implementation of IWPP on the Intel Xeon Phi is difficultrnbecause of the irregular data access and computation characteristics. The traditional IWPPrnalgorithm relies on atomic instructions, which are not available in the SIMD set of the Intel Phi.rnTo overcome this limitation, we have proposed a new IWPP algorithm that can take advantagernof non-atomic SIMD instructions supported on the Intel Xeon Phi. We have also developed andrnevaluated methods to useCPUand IntelPhi cooperatively for parallel execution of theIWPPalgorithms.rnOur new cooperative IWPP version is also able to handle large out-of-core images thatrnwould not fit into the memory of the accelerator. The new IWPP algorithm is used to implementrnthe Morphological Reconstruction and Fill Holes operations, which are operations commonlyrnfound in image analysis applications. The vectorization implemented with the new IWPP hasrnattained improvements of up to about 5×on top of the original IWPPand significant gains as comparedrnto state-of-the-art the CPU and GPU versions. The new version running on an Intel Phi isrn6.21× and 3.14× faster than running on a 16-core CPU and on a GPU, respectively. Finally, therncooperative execution using two Intel Phi devices and a multi-coreCPUhas reached performancerngains of 2.14× as compared to the execution using a single Intel Xeon Phi.