Three-Dimensional Fluid Code with XcalableMP

In order to adapt parallel computers to general convenient tools for computational scientists, a high-level and easy-to-use portable parallel programming paradigm is mandatory. XcalableMP, which is proposed by the XcalableMP Specification Working Group, is a directive-based language extension for Fortran and C to easily describe parallelization in programs for distributed memory parallel computers. The Omni XcalableMP compiler, which is provided as a reference XcalableMP compiler, is currently implemented as a source-to-source translator. It converts XcalableMP programs to standard MPI programs, which can be easily compiled by the native Fortran compiler and executed on most of parallel computers. A three-dimensional Eulerian fluid code written in Fortran is parallelized by XcalableMP using two different programming models with the ordinary domain decomposition method, and its performances are measured on the K computer. Programs converted by the Omni XcalableMP compiler prevent native Fortran compiler optimizations and show lower performance than that of hand-coded MPI programs. Finally almost the same performances are obtained by using specific compiler options of the native Fortran compiler in the case of a global-view programming model, but performance degradation is not improved by specifying any native compiler options when the code is parallelized by a local-view programming model.

Performance portable parallel programming of heterogeneous stencils across shared-memory platforms with modern Intel processors

In this work, we take up the challenge of performance portable programming of heterogeneous stencil computations across a wide range of modern shared-memory systems. An important example of such computations is the Multidimensional Positive Definite Advection Transport Algorithm (MPDATA), the second major part of the dynamic core of the EULAG geophysical model. For this aim, we develop a set of parametric optimization techniques and four-step procedure for customization of the MPDATA code. Among these techniques are: islands-of-cores strategy, (3+1)D decomposition, exploiting data parallelism and simultaneous multithreading, data flow synchronization, and vectorization. The proposed adaptation methodology helps us to develop the automatic transformation of the MPDATA code to achieve high sustained scalable performance for all tested ccNUMA platforms with Intel processors of last generations. This means that for a given platform, the sustained performance of the new code is kept at a similar level, independently of the problem size. The highest performance utilization rate of about 41–46% of the theoretical peak, measured for all benchmarks, is provided for any of the two-socket servers based on Skylake-SP (SKL-SP), Broadwell, and Haswell CPU architectures. At the same time, the four-socket server with SKL-SP processors achieves the highest sustained performance of around 1.0–1.1 Tflop/s that corresponds to about 33% of the peak.