OpenMP Directives Research Articles

Performance of a Code Migration for the Simulation of Supersonic Ejector Flow to SMP, MIC, and GPU Using OpenMP, OpenMP+LEO, and OpenACC Directives

A serial source code for simulating a supersonic ejector flow is accelerated using parallelization based on OpenMP and OpenACC directives. The purpose is to reduce the development costs and to simplify the maintenance of the application due to the complexity of the FORTRAN source code. This research follows well-proven strategies in order to obtain the best performance in both OpenMP and OpenACC. OpenMP has become the programming standard for scientific multicore software and OpenACC is one true alternative for graphics accelerators without the need of programming low level kernels. The strategies using OpenMP are oriented towards reducing the creation of parallel regions, tasks creation to handle boundary conditions, and a nested control of the loop time for the programming in offload mode specifically for the Xeon Phi. In OpenACC, the strategy focuses on maintaining the data regions among the executions of the kernels. Experiments for performance and validation are conducted here on a 12-core Xeon CPU, Xeon Phi 5110p, and Tesla C2070, obtaining the best performance from the latter. The Tesla C2070 presented an acceleration factor of 9.86X, 1.6X, and 4.5X compared against the serial version on CPU, 12-core Xeon CPU, and Xeon Phi, respectively.

Overhead Analysis of Loop Parallelization with OpenMP Directives

OpenMP is one of the most popular parallel programming technique nowadays. Using openMP, programmers parallelized their serial programs by inserting directives. However, in addition to parallel algorithm related overhead which depends on how to design the parallel program, performing openMP also leads to overhead. In this paper, we give a model to evaluate the overhead, and design 5 groups of experiment to evaluate the overhead caused by starting openMP, scheduling and synchronizations. We also analyze the changes of the overhead with different scale of the problems and the number of parallel threads.

NDL-v2.0: A new version of the numerical differentiation library for parallel architectures

NDL-v2.0: A new version of the numerical differentiation library for parallel architectures

Simulating the filtration combustion of gases on multi-core computers

Under study is the performance of some computational models of filtration combustion of gases on multi-core computers. The analysis is restricted to the models based on explicit difference schemes. In particular, an explicit two-level parallel algorithm with an adaptive mesh is constructed. The two shared memory parallelization methods are applied: the straightforward application of OpenMP directives and special distribution of data among the threads. The simulation shows that the last method has a substantial performance advantage.

Hybrid parallelism in MFIX CFD-DEM using OpenMP

This paper supplements previous work done by Gopalakrishnan and Tafti [21] for the parallel implementation of the Discrete Element Method (DEM) in the open source code Multiphase Flow with Interphase eXchange (MFIX) for distributed memory architectures using MPI. The current study extends the parallel implementation to shared memory architectures using OpenMP making the code amenable to hybrid MPI+OpenMP computations on multicore multiprocessor architectures. The contact force computation, the drag force computation, and locating the particle in fluid cell for neighbor searches are identified to be some of the major time consuming routines and instrumented with OpenMP directives. The OpenMP instrumented code is shown to give a speedup of 7 on 8 threads in a 3D bubbling fluidized bed with 80,000 particles and 18,000 fluid cells. Furthermore, the hybrid performance of MPI+OpenMP is evaluated for a large scale CFD-DEM system of 5.12million particles. The strong scaling analysis shows good scalability with a speed up of 185 on 256 processors (20,000particles/processor) for hybrid versus a speed up of 138 for a standalone MPI.

Exploration on a Fast EHL Computing Technology for Analyzing Journal Bearings with Engineered Surface Textures

Solving elastohydrodynamic lubrication (EHL) problems is a complex and time-consuming process due to the interactive solutions of the Reynolds equation and contact elasticity. Analyzing journal bearing EHL problems is even more difficult due to the scale difference in the structural and surface features, which may span four orders of magnitude. This article presents a fast EHL computing technology utilizing a parallel numerical iterative method (the red–black successive overrelaxation method) and multithreaded computing scheme conducted by OpenMP directives. The fast computational approaches allow the construction of high-density EHL meshes for effective descriptions of important texture features of journal bearing surfaces.

Performance analysis of SSE and AVX instructions in multi-core CPUs and GPU computing on FDTD scheme for solid and fluid vibration problems

In this work a unified treatment of solid and fluid vibration problems is developed by means of the Finite-Difference Time-Domain (FDTD). The scheme here proposed takes advantage from a scaling factor in the velocity fields that improves the performance of the method and the vibration analysis in heterogenous media. Moreover, the scheme has been extended in order to simulate both the propagation in porous media and the lossy solid materials. In order to accurately reproduce the interaction of fluids and solids in FDTD both time and spatial resolutions must be reduced compared with the set up used in acoustic FDTD problems. This aspect implies the use of bigger grids and hence more time and memory resources. For reducing the time simulation costs, FDTD code has been adapted in order to exploit the resources available in modern parallel architectures. For CPUs the implicit usage of the advanced vectorial extensions (AVX) in multi-core CPUs has been considered. In addition, the computation has been distributed along the different cores available by means of OpenMP directives. Graphic Processing Units have been also considered and the degree of improvement achieved by means of this parallel architecture has been compared with the highly-tuned CPU scheme by means of the relative speed up. The speed up obtained by the parallel versions implemented were up to 3 (AVX and OpenMP) and 40 (CUDA) times faster than the best sequential version for CPU that also uses OpenMP with auto-vectorization techniques, but non includes implicitely vectorial instructions. Results obtained with both parallel approaches demonstrate that massive parallel programming techniques are mandatory in solid-vibration problems with FDTD.

Modeling of the turbulent mixing on basis of the large eddy simulation by using parallel computing

In the research the numerical simulation of the room’s ventilation was performed on the basis of the large eddy simulation. The numerical algorithm was developed by using the physical parameters splitting scheme. Poisson equation for pressure field is solved by Fourier method in combination with tridiagonal matrix method for determination of Fourier coefficients. The parallel programming technique MPI in combination with directives of OpenMP was applied to solve the assign task, as the most relevant for today and the most efficient a view of the productivity improvement of complex calculations. Simulation data are represented as three-dimensional graphs.

Comparing high performance techniques for the automatic generation of efficient solvers of cardiac cell models

In silico experiments have been used for a better understanding of the electrical activity of cardiac myocytes, usually via models based on nonlinear systems of ordinary differential equations. Many different models for cardiac myocytes are available that vary on the level of complexity, depending on how detailed the phenomena is described. Long simulations of realistic and complex models are computationally expensive. To cope with this problem, this work compares different techniques to automatically speed up the numerical solution of cardiac models: (a) adaptive time step method, (b) Partial Evaluation (PE) and Lookup Tables (LUTs), and (c) an automatic way to find and exploit code concurrency via OpenMP directives. All the techniques were implemented as part of an automatic code generator for the numerical solution of models that are described in the CellML markup language. Experimental results demonstrated that the adaptive time step simulations were up to 32 times faster than the traditional Euler that use fixed time step. Combined with parallel computing on a multicore processor the execution time was further decreased and simulations were 41 times faster. Finally, the LUTs and PE techniques resulted in a 117-fold improvement in computation time over the Euler method and 72-fold improvement when compared to the traditional Rush–Larsen method.

Asynchronous Approach to Memory Management in Sparse Multifrontal Methods on Multiprocessors

This research covers the Intel? Direct Sparse Solver for Clusters, the software that implements a direct method for solving the Ax = b equation with sparse symmetric matrix A on a cluster. This method, researched by Intel, is based on Cholesky decomposition and could be considered as extension of functionality PARDISO from Intel? MKL. To achieve an efficient work balance on a large number of processes, the so-called “multifrontal” approach to Cholesky decomposition is implemented. This software implements parallelization that is based on nodes of the dependency tree and uses MPI, as well as parallelization inside a node of the tree that uses OpenMP directives. The article provides a high-level description of the algorithm to distribute the work between both computational nodes and cores within a single node, and between different computational nodes. A series of experiments shows that this implementation causes no growth of the computational time and decreases the amount of memory needed for the computations.

Automated Derivation of the Adjoint of High-Level Transient Finite Element Programs

In this paper we demonstrate a new technique for deriving discrete adjoint and tangent linear models of finite element models. The technique is significantly more efficient and automatic than standard algorithmic differentiation techniques. The approach relies on a high-level symbolic representation of the forward problem. In contrast to developing a model directly in Fortran or C++, high-level systems allow the developer to express the variational problems to be solved in near-mathematical notation. As such, these systems have a key advantage: since the mathematical structure of the problem is preserved, they are more amenable to automated analysis and manipulation. The framework introduced here is implemented in a freely available software package named dolfin-adjoint, based on the FEniCS Project. Our approach to automated adjoint derivation relies on run-time annotation of the temporal structure of the model, and employs the FEniCS finite element form compiler to automatically generate the low-level code for the derived models. The approach requires only trivial changes to a large class of forward models, including complicated time-dependent nonlinear models. The adjoint model automatically employs optimal checkpointing schemes to mitigate storage requirements for nonlinear models, without any user management or intervention. Furthermore, both the tangent linear and adjoint models naturally work in parallel, without any need to differentiate through calls to MPI or to parse OpenMP directives. The generality, applicability and efficiency of the approach are demonstrated with examples from a wide range of scientific applications.

Towards petascale spectral simulations for transition analysis in wall bounded flow

SUMMARYAn efficient parallel spectral method for direct numerical simulations of transitional and turbulent flows is described in this paper. The parallelization is classically based on a bidimensional domain decomposition, but has been specifically developed for a solenoidal Fourier–Chebyshev spectral approximation where in one Fourier direction, the number of modes is very large compared with the two other directions. The approach therefore differs from classical libraries developed for cubic Fourier boxes. The strategy uses message‐passing interface (MPI) for message‐passing among nodes and is fairly portable. One of the originalities of this paper is the use of an efficient hybrid programming with MPI for internodes communications and a coarse grain parallelism using OpenMP for core shared‐memory computation, instead of the classical hybrid programming with MPI and a fine granularity parallelism at the loop level with OpenMP directives. This hybrid parallelism has been tested on the recent generation of high‐performance parallel supercomputers involving a few tens of cores per node. Performances are evaluated on different low‐frequency and high‐frequency processors massively parallel platforms. We demonstrate that spectral methods, which are known to be inherently ill‐fitted for the new generation of high‐performance distributed‐memory computers, can be implemented efficiently using this hybrid programming with good scalability and a very fast wall‐clock time per iteration. New numerical experiments are therefore now accessible on petascale computers, while keeping the attractive features of spectral methods such as accuracy, exponential convergence, computational efficiency and conservative properties. This is illustrated by a direct numerical simulation of the transition of the boundary layers developing from the entrance section of a plane channel and interacting to merge into a fully turbulent flow. Copyright © 2012 John Wiley & Sons, Ltd.

Parallel simulation of Brownian dynamics on shared memory systems with OpenMP and Unified Parallel C

The simulation of particle dynamics is an essential method to analyze and predict the behavior of molecules in a given medium. This work presents the design and implementation of a parallel simulation of Brownian dynamics with hydrodynamic interactions for shared memory systems using two approaches: (1) OpenMP directives and (2) the Partitioned Global Address Space (PGAS) paradigm with the Unified Parallel C (UPC) language. The structure of the code is described, and different techniques for work distribution are analyzed in terms of efficiency, in order to select the most suitable strategy for each part of the simulation. Additionally, performance results have been collected from two representative NUMA systems, and they are studied and compared against the original sequential code.

CorrelaGenes: a new tool for the interpretation of the human transcriptome

CorrelaGenes: a new tool for the interpretation of the human transcriptome

Performance analysis of the FDTD method applied to holographic volume gratings: Multi-core CPU versus GPU computing

The finite-difference time-domain method (FDTD) allows electromagnetic field distribution analysis as a function of time and space. The method is applied to analyze holographic volume gratings (HVGs) for the near-field distribution at optical wavelengths. Usually, this application requires the simulation of wide areas, which implies more memory and time processing. In this work, we propose a specific implementation of the FDTD method including several add-ons for a precise simulation of optical diffractive elements. Values in the near-field region are computed considering the illumination of the grating by means of a plane wave for different angles of incidence and including absorbing boundaries as well. We compare the results obtained by FDTD with those obtained using a matrix method (MM) applied to diffraction gratings. In addition, we have developed two optimized versions of the algorithm, for both CPU and GPU, in order to analyze the improvement of using the new NVIDIA Fermi GPU architecture versus highly tuned multi-core CPU as a function of the size simulation. In particular, the optimized CPU implementation takes advantage of the arithmetic and data transfer streaming SIMD (single instruction multiple data) extensions (SSE) included explicitly in the code and also of multi-threading by means of OpenMP directives. A good agreement between the results obtained using both FDTD and MM methods is obtained, thus validating our methodology. Moreover, the performance of the GPU is compared to the SSE+OpenMP CPU implementation, and it is quantitatively determined that a highly optimized CPU program can be competitive for a wider range of simulation sizes, whereas GPU computing becomes more powerful for large-scale simulations.

Parallel computation of satellite orbit acceleration

We present a formulation of parallel computation of the satellite orbit acceleration due to non-spherical gravitational field. The key techniques are the vector computation of associated Legendre functions and the do-loop folding. These are easily implemented by a slight modification of existing programs and by embedding a few OpenMP directives. The formulation significantly speeds up the orbit integrations of low-altitude satellites. For example, that of GRACE under the influence of EGM2008, a 2190×2159 geopotential model runs around 4 times faster than the serial computation if conducted at a PC with a quad-core 8-thread processor.

Automatic Parallelization of Array-oriented Programs for a Multi-core Machine

We present the work on automatic parallelization of array-oriented programs for multi-core machines. Source programs written in standard APL are translated by a parallelizing APL-to-C compiler into parallelized C code, i.e. C mixed with OpenMP directives. We describe techniques such as virtual operations and data-partitioning used to effectively exploit parallelism structured around array-primitives. We present runtime performance data, showing the speedup of the resulting parallelized code, using different numbers of threads and different problem sizes, on a 4-core machine, for several examples.

Workstation Computing of Discretized Reynolds Equations

In this study, the solution methods suitable for use in a workstation to solve the discretized incompressible and compressible fluid Reynolds equations are examined. The workstation used for computing consists of dual six-core central processing units (CPUs) and a 256-core graphics processing unit (GPU) dedicated for computing. To compare the computational performance, the Reynolds equations are solved by a parallel iterative method using multithreaded and GPU computing. Multithreaded computing is conducted by OpenMP directives and GPU computing is executed using either NVIDIA's compute unified device architecture (CUDA) programming or the accelerator programming model (APM). The lubrication models used are an inclined surface slider with or without a central recess and an air journal bearing. In GPU computing, both the CUDA and APM are subjected to communication latency to achieve efficient fine-grained computations. The results show that the performance of the less expensive GPU computing is close to that of multithreaded computing in which the grid sizes are large in the analyses. The performance of the OpenMP-like APM programming is compatible with CUDA computing in the slider cases but is significantly worse for air bearings. This study illustrates the workstation computing for lubrication analyses in which the computing devices can be multi-core CPUs or many-core GPUs.

Low-energy electron collisions with glycine

We report cross sections for elastic electron scattering by gas phase glycine (neutral form), obtained with the Schwinger multichannel method. The present results are the first obtained with a new implementation that combines parallelization with OpenMP directives and pseudopotentials. The position of the well known π* shape resonance ranged from 2.3 eV to 2.8 eV depending on the polarization model and conformer. For the most stable isomer, the present result (2.4 eV) is in fair agreement with electron transmission spectroscopy assignments (1.93 ± 0.05 eV) and available calculations. Our results also point out a shape resonance around 9.5 eV in the A' symmetry that would be weakly coupled to vibrations of the hydroxyl group. Since electron attachment to a broad and lower lying σ* orbital located on the OH bond has been suggested the underlying mechanism leading to dissociative electron attachment at low energies, we sought for a shape resonance around ~4 eV. Though we obtained cross sections with the target molecule at the equilibrium geometry and with stretched OH bond lengths, least-squares fits to the calculated eigenphase sums did not point out signatures of this anion state (though, in principle, it could be hidden in the large background). The low energy (~1 eV) integral cross section strongly scales as the bond length is stretched, and this could indicate a virtual state pole, since dipole supported bound states are not expected at the geometries addressed here.

광선추적법 기반의 적분구 분석 시뮬레이터에서 OpenMP 지시어를 이용한 속도 향상 및 몬테카를로 방법의 무작위성 보장

In order to improve the calculation speed of an integrating-sphere simulator based on a ray-tracing method, parallel processing with OpenMP directive was implemented into the simulator and the randomness of Monte Carlo method was guaranteed by utilizing a parallel random number generator. It was confirmed that simulation results obtained with more than rays showed good agreement with theoretical results within the error range of 0.5%, and that the calculation speed improved as the number of threads increased. Finally, the spatial response distribution functions of a real integrating sphere were simulated and compared with previous results.