P.: Automatic transformations for effective parallel execution on intel many integrated core. In: TACC-Intel Highly Parallel Computing Symp

We demonstrate in this work the potential effectiveness of a source-to-source framework for automatically optimizing a sub-class of affine programs on the Intel Many Integrated Core Architecture. Data locality is achieved through complex and automated loop trans-formations within the polyhedral framework to enable parallel tiling, and the resulting tiles are processed by an aggressive automatic SIMD vector code generator. We evaluate the effectiveness of this approach on tensor contraction kernels. We show a mean improvement of 1.86 × over existing compiler techniques for single core performance, and combined with automatic parallelization we achieve 14.56 × the performance of Intel’s ICC Compiler on MIC.