high performance computing on graphics processing units: hgpu.org

In this paper we will present a detailed study of implementing double-precision matrix-matrix multiplication (DGEMM) utilizing the Intel Xeon Phi Coprocessor. We discuss a DGEMM algorithm implementation running "natively" on the coprocessor, minimizing communication with the host CPU. We will run DGEMM across a range of matrix sizes natively as well using Intel Math Kernel Library. Our optimizations were designed to support maximal reuse of on-die cache, which significantly reduces transfer from GDDR. Finally we analyze the improvement of a classic matrix multiplication implementation based on Cauchy algorithm compared to the latest results achieved using the Intel Math Kernel Library DGEMM subroutine.