Numerical Library with High-Performance/Adaptive-Precision/High-Reliability: Extension of ppOpen-HPC towards the Post Moore Era
Math Library Design
Post Moore’s Law Computing
TimeMonday, June 25th4:40pm - 5pm
DescriptionPower consumption is one of the most critical issues in supercomputing towards the Post Moore Era. Although computations with lower precision, such as approximate computing, is one of the reasonable choices for reducing power consumption, such computations may provide results with less accuracy. Therefore, accuracy verification is very important for computations with lower precision. In the present work, we are focusing on preconditioned iterative solvers with sparse coefficient matrices derived from real applications, and are developing a practical method for accuracy verification in adaptive/trans-precision computations based on iterative refinement. Method for automatic selection of the optimum precision, which minimizes computation time and power consumption under certain target accuracy, will be developed, and it is implemented to “ppOpen-HPC”.
“ppOpen-HPC" is an open source infrastructure for development and execution of optimized and reliable simulation code on post-peta-scale (pp) parallel computers based on many-core architectures, and it consists of various types of libraries, which cover general procedures for scientific computation. “ppOpen-HPC" has been supported by Japanese Government (FY.2011-2018).
In this talk, we will overview the strategy for the new numerical library with High-Performance/Adaptive-Precision/High-Reliability, and show preliminary results.