Monday 3rd

Registration | Reception

At the Nara Hotel (not the conference center) from 4:30pm. Nara Hotel. Travel Info.

Time Activity
16:30 – 17:00


17:00 – 19:00

Welcome Drinks Reception

Tuesday 4th

Two Keynotes | Six Technical Presentations | Panel Discussions | Conference Dinner

At the Nara Kasugano International Forum in Nara Park, Japan. Travel Info.

Time Presentation / Speaker
08:30 – 09:00


09:00 – 09:15

Welcome and Introduction

09:15- 10:15
  • KEYNOTE:  Simon McIntosh-Smith
    Evaluating OpenMP’s Effectiveness in the Many-Core Era
    University of Bristol
  • OpenMP has had a significant impact on high performance computing since its introduction in 1997, but today it is being stretched far beyond its original goals. Modern CPUs are heading towards hundreds of cores, and since OpenMP 4, the standard is even targeting GPUs. In this talk we will review OpenMP’s performance on the latest architectures, investigating how well modern compilers are vectorising and scaling on today’s processors. We will finish with a “score card” summarising OpenMP’s current achievements, and look forward to OpenMP’s future challenges.
10:15 – 10:45

Coffee Break | Posters | Sponsor Table Top demos

10:45 – 11:15
  • Obtaining the Last Values of Conditionally Assigned Privates
    Hideki Saito, Serge Preis, Aleksei Cherkasov and Xinmin Tian
11:15 – 11:45
  • Using Fortran 2003 Features in OpenMP Programs
    Kelvin Li (IBM)
  • OpenMP has been used in Fortran scientific and engineering applications for many years. The current OpenMP specification supports Fortran 95 and partial Fortran 2003 standard. The Fortran subcommittee has been working on rebasing the OpenMP specification to Fortran 2003 standard over several releases. In this talk, we will discuss major features being added in the latest OpenMP specification and also discusses the details about how the features are supported in the OpenMP context. Some hints and tips are also provided in the talks.
11:45 – 12:15
  • OpenMP in Classifier Design
    Bogusław Cyganek (AGH University of Science and Technology) and Michal Wozniak (Wroclaw University of Technology)
  • Classifiers are algorithms that, given features, respond class of unknown objects. These can be supervised, i.e. trained from known examples, or unsupervised which can discover some regularities in data. Their examples are ample, from spam filters, text search engines, up to face recognition, car security systems with road signs recognition, driver sleepiness alert, unmanned vehicle maneuvering, and many more. Main parameters of classifiers are accuracy and operation speed. In this talk we concentrate on the latter and provide some design patterns and recipes on parallel implementations of selected classifiers using OpenMP. Our main application domain are classifiers used in computer vision and streams of big data, though the presented methods can be easily exploited in other branches of classification problems.
12:15 – 13:45

Lunch Break

13:45 – 14:45
  • KEYNOTE: Michael Wong
    Towards Heterogeneous Programming in C++17 and beyond

  • This talk will present the recent rise in demand for heterogeneous programming in C++ and the new challenges that this brings to the C++ community. It will then take a look at some of the different programming languages and models that have arrived in the heterogeneous landscape, the motivations behind them, how they have attempted to solve those challenges and what we can learn from them. Finally it will take a look at the future of heterogeneous programming in C++ and what that might look like.  Read more
14:45 – 15:15

Coffee Break | Posters | Sponsor Table Top demos

15:15 – 15:45
  • User-Guided Speculative Locks for Algebraic Multigrid Smoothers
    Barna Bihari, Ulrike Yang and Bronis de Supinski (Lawrence Livermore National Laboratory), James Cownie and Hansang Bae (Intel Corporation)
  • We extend the implementation of transactional memory for algebraic multigrid (AMG) smoothers to the Intel Xeon processor (former codename “Broadwell”) by employing a user-guided locking API for OpenMP that provides a general and unified user interface. This work also builds upon our earlier investigations, but it is now applied to an industrial-grade linear solver library: hypre. Since the Intel Transactional Synchronization Extensions (TSX) provide two different options for speculation — Hardware Lock Elision (HLE) and Restricted Transactional Memory (RTM) — we compare three different run modes: (i) HLE, (ii) RTM, and (iii) OpenMP critical to some of the established AMG smoothers. By testing our implementation on a 3-D problem we find that HLE offers the best performance of all methods when using the measure introduced in our earlier work.
15:45 – 16:15
  • Improving the Performance of the Global Atmospheric Model Using Novel OpenMP Features
    Kwangjae Sung, Ki-Hwan Kim and Junghan Kim, Korea Institute of Atmospheric Prediction Systems (KIAPS)
  • KIAPS is developing a next generation global atmospheric model (KIAPS Integrated Model, KIM).  KIM simulates the atmospheric flow on earth for a weather forecast. It requires huge computational resources. However, the model currently cannot utilize the modern processors, such as GPU, MIC, and FPGA, which are low-power and high-performance.  Our main concern is to apply OpenMP 4.0 features, such as thread affinity, vectorization, and heterogeneous computing, to KIM while using MPI processes. In this study, we present an enhanced global model based on MPI APIs and novel OpenMP directives for parallelization, SIMD programming, and work offloading called an OpenMP-based KIM (ompKIM). Compared to the existing KIM, while ompKIM can achieve competitive computational performance and energy-efficiency, it is simple and intuitive to be implemented with SIMD and HPC accelerators (e.g., GPUs and Intel MIC Coprocessors) due to the use of OpenMP. Additionally, since ompKIM is surppoted by a variety of compilers, it can provide portability for various architectures and systems. Experimenatal results through HPC and accelerators show that our model can offer the performance of more effective and energy-efficient computation than the atmospheric model based on only the MPI features.
16:15 – 16:45
  • Programming-Model Centric Debugging for OpenMP
    Kevin Pouget (STMicroelectronics / Université de Grenoble), Jean-François Mehaut (Laboratoire LIG) and Miguel Santana (STMicroelectronics)
  • In this presentation, we introduce a new approach for debugging
    parallel applications based on the OpenMP programming standard. This
    approach is centered on the programming model, by opposition with more
    traditionnal debugging methods focused on the programming language and
    machine code. We detail the main axes of programming-model centric debugging: providing a structural representation of the application, following
    the execution dynamic behaviors and allowing interactions with the
    abstract and physical machines; and we explain how they apply to
    OpenMP fork-join and task-based programming.

Panel Discussion

Conference Dinner in Local Restaurant

Wednesday 5th

Tutorials sessions run in conjunction with IWOMP.  Visit the IWOMP Tutorial page for more information.