Workshop on Numerical Computing and Software Productivity in Nagoya (WNCSP2023)

Date:
August 30th (Wed) 2023

Venue:
2F Lecture Room, Information Technology Center, Nagoya University
MAP:https://icts.nagoya-u.ac.jp/en/center/about/map.html

Support:
Joint Usage/Research Center for Interdisciplinary Large-scale Information Infrastructures (JHPCN)" and "High Performance Computing Infrastructure (HPCI)" in Japan
(Project ID: jh230005) “New Horizon Technology of Auto-tuning by Software Engineering”

Program:

13:00 - 13:10 Opening

Session 1: Software Productivity
13:10 - 13:40 Elaine Raybourn (Sandia National Labs)
Title: Sociotechnical aspects of code collaborations

In this talk the speaker will describe the Exascale Computing Project
(ECP) Team of Teams research on code collaboration and how the ECP organizational model for a multi-institute vision ritual organization can be applied in other contexts such as university code teams, perhaps with some equal success due to the sociotechical emphasis on teaming.

This talk would be a high level treatment of the organizational structure and collaboration. Time permitting, the speaker will also review the productivity & sustainability improvement process (PSIP) and how it is used for developer productivity. The speaker will also tie this the HPC Best Practices webinar and the Strategies for Working Remotely panel series. Key take always will be thought provoking ideas for collaboration, teaming, and productivity.

13:40 - 14:10 Shuji Morisaki (Nagoya University)
Title: Test case prioritization studies in software engineering

The scale of software testing is increasing due to the growing scale of software and the complexity of implementation. Performing large-scale testing requires more time and effort. Test case prioritization enables to decrease the time and effort to detect and fix defects because executing all the test cases is not always necessary to detect and fix defects unless the test cases can detect no defect. This talk will introduce test case prioritization studies that can lead to time and effort reduction for test execution. The speaker will also show a usage example of the test case prioritization studies.

14:10 - 14:20 Break

Session 2: ECP Products
14:20 - 14:50 Osni Marques (Laurence Berkeley National Laboratory)
Title: ECP Overview

The goal of the US Department of Energy’s Exascale Computing Project (ECP) is to ensure that all the necessary pieces are in place for the nation’s first exascale systems, to enable the solution of problems in scientific discovery, manufacturing R&D, and national security at levels of complexity and performance that previously were not possible. This presentation will give an overview of the activities carried out by ECP, including applications, software tools and libraries, and training efforts.

14:50 - 15:20 Makoto Morishita (Nagoya University), Osni Marques (Laurence Berkeley National Laboratory), Takahiro Katagiri (Nagoya University)
Title: Auto-tuning of Numerical Library by GPTune

To maximize performance of numerical libraries, there are several performance parameters to be tuned. For example, the block size for cache optimization, and the process grid for parallel distributed computing. In this presentation, we talk about GPTune, which is an auto-tuning software developed by the US Department of Energy's Exascale Computing Project (ECP) to tune performance parameters on numerical libraries. We evaluated the effect of applying GPTune to two ScaLAPACK routines, which are QR decomposition and LU decomposition.

15:20 - 15:30 Break

Session 3: Example of Numerical Software
15:30 - 15:50 Takahiro Katagiri (Nagoya University)
Title: Autotuning for Sparse Iterative Solvers and Quantum Computing Related Technology

To establish high performance computing, performance tuning is still high cost in viewpoint of software productivity. Autotuning (AT) technology is one of promising way to solve the productivity problem. In this presentation, a state-in-the-art AT technology is presented in several cases for adaptation of AT. The adaptation includes AT for performance parameters for typical sparse iterative solver (i.e. ICCG method), and related process for quantum computing on GPU supercomputer environments, such as inspired quantum annealer and simulation of quantum circuits.

15:50 - 16:10 Masatoshi Kawai (Nagoya University)
Title: Provides dynamic load balancing with dynamic core binding

We are conducting research on dynamic load balancing for MPI/OpenMP parallelized applications with load imbalance. To support dynamic load balancing, we propose a dynamic core binding (DCB) environment, which is changed the number of cores bound to each process, and equalize the load at the core level. To provide the DCB environment, we are implementing the library because of the requirement of special system calls. In this presentation, we discuss the interface of the library being implemented.

16:10 - 16:30 Tetsuya Hoshino (Nagoya University)
Title: Acceleration of Hierarchical Matrix Library HACApK

Hierarchical matrices (H-matrices) are methods for approximating dense matrices such as those appearing as coefficients in boundary element methods. HACApK is a Fortran-based MPI+OpenMP parallelized H-matrix library. In this presentation, we will introduce the speed-up method by SIMD considering the user interface of HACApK and the speed-up method by GPU.

16:30 - 16:40 Closing

17:30- Reception at near Nagoya University

---------------------------
[Latest updates] 4th August, 2023