Tuning and optimization for a variety of many-core architectures without changing a single line of implementation code using the Alpaka library

摘要

We present an analysis on optimizing performance of a single C++11 sourcecode using the Alpaka hardware abstraction library. For this we use the generalmatrix multiplication (GEMM) algorithm in order to show that compilers canoptimize Alpaka code effectively when tuning key parameters of the algorithm.We do not intend to rival existing, highly optimized DGEMM versions, but merelychoose this example to prove that Alpaka allows for platform-specific tuningwith a single source code. In addition we analyze the optimization potentialavailable with vendor-specific compilers when confronted with the heavilytemplated abstractions of Alpaka. We specifically test the code for bleedingedge architectures such as Nvidia's Tesla P100, Intel's Knights Landing (KNL)and Haswell architecture as well as IBM's Power8 system. On some of these weare able to reach almost 50\% of the peak floating point operation performanceusing the aforementioned means. When adding compiler-specific #pragmas we areable to reach 5 TFLOPS/s on a P100 and over 1 TFLOPS/s on a KNL system.