115
reads

Takahiro Katagiri

2013-03-27
10:45:00 - 11:10:00

101 , Mathematics Research Center Building (ori. New Math. Bldg.)

It is said that code optimization by compilers will be too difficult to optimize codes in computer systems in Exa-flops era. One of reasons is due to various architectures in hardware and software on the expected systems in Exa-flops era. This damages execution performance of legacy codes. In particular, the performance drawback of reliable software stacks for numerical computation is serious problem. In this talk, we present a new code optimization framework to solve the performance problem in Exa-flops era. We are now developing ppOpen-HPC, which is numerical middleware for numerical simulation in post peta-scale era. By using real codes from ppOpen- HPC, we are developing new functions. Software that includes this optimization functions is called ppOpen-AT. In this presentation, we firstly show a code optimization example of automatic loops transformation by ppOpen-AT that seems to be frequently used in explicit method. Secondly, we show a research direction for code optimization that towards to Exa-flops era in ppOpen-AT project. The direction focuses on technologies with OpenACC, and related technologies for the Intel Many Integrated Core (MIC) architectures.