Performance analysis of the Kahan-enhanced scalar product oncurrent multi-core and many-core processors

摘要

We investigate the performance characteristics of a numerically enhanced scalar product (dot) kernel loopthat uses the Kahan algorithm to compensate for numerical errors, and describe efficient single instructionmultiple data-vectorized implementations on recent multi-core and many-core processors. Using low-levelinstruction analysis and the execution-cache-memory performance model, we pinpoint the relevant performancebottlenecks for single-core and thread-parallel execution and predict performance and saturationbehavior. We show that the Kahan-enhanced scalar product comes at almost no additional cost comparedwith the naive (non-Kahan) scalar product if appropriate low-level optimizations, notably single instructionmultiple data vectorization and unrolling, are applied. The execution-cache-memory model is extendedappropriately to accommodate not only modern Intel multicore chips but also the Intel Xeon Phi ‘KnightsCorner’ coprocessor and an IBM POWER8 CPU. This allows us to discuss the impact of processor featureson the performance across four modern architectures that are relevant for high performance computing.