OpenCL Parallel Research Articles

OclCUB: an OpenCL parallel computing library for deep learning operators

oclCUB: an OpenCL parallel computing library for deep learning operators

GPU parallel computing based on PF-LBM method for simulating dendrites growth under natural convection conditions

This study introduces a GPU-based parallel computing approach that combines the phase-field model (PF) and the lattice Boltzmann model (LBM). By establishing a coupled multiphase field model incorporating physical external fields such as flow field, temperature field, and solute field, the research simulates the growth of single grains and multiple grains under the influence of natural convection. The variations in dendritic morphology, flow field, and solute field during dendritic solidification processes are observed. Initially, the study analyzes the morphology of equiaxed dendrites and the growth patterns of primary dendrites arms under natural convection conditions. The evolution of equiaxed dendrites in single grains and multiple grains under various conditions is investigated. Furthermore, the study explores the impact of different anisotropy strengths on the growth of single grains and multiple grains under natural convection. Notably, a distinct “necking” phenomenon is observed when the anisotropy strength of a single grain is 0.05. In the case of multiple grains, where competition between dendrites is present in addition to the influence of natural convection, a pronounced “necking” phenomenon is evident at an anisotropy strength of 0.03. Moreover, OpenCL parallel technology is designed on the GPU platform to accelerate the solution of the model. The parallelization of the phase-field model coupled with the LBM model on the GPU demonstrates a clear advantage. The parallel computation based on GPU not only exhibits absolute superiority but also shows more significant acceleration effects as the computational domain increases.

Low-Power FPGA Architecture Based Monitoring Applications in Precision Agriculture

Today’s on-chip systems technology has grounded impressive advances in computing power and energy consumption. The choice of the right architecture depends on the application. In our case, we were studying vegetation monitoring algorithms in precision agriculture. This study presents a system based on a monitoring algorithm for agricultural fields, an electronic architecture based on a CPU-FPGA SoC system and the OpenCL parallel programming paradigm. We focused our study on our own dataset of agricultural fields to validate the results. The fields studied in our case are in the Guelmin-Oued noun region in the south of Morocco. These fields are divided into two areas, with a total surface of 3.44 Ha2 for the first field and 3.73 Ha2 for the second. The images were collected using a DJI-type unmanned aerial vehicle and an RGB camera. Performance evaluation showed that the system could process up to 86 fps versus 12 fps or 20 fps in C/C++ and OpenMP implementations, respectively. Software optimizations have increased the performance to 107 fps, which meets real-time constraints.

Three-dimensional multi-phase-field simulation of eutectoid alloy based on OpenCL parallel

Seeking high-performance computing methods to solve the problem of a large amount of calculation, low calculation efficiency, and small simulation scale on the traditional single central processing unit (CPU) platform is of great value to the simulation study of micro-structure. In this study, based on the three-dimensional multi-phase-field model of KKSO coupling phase-field and solute field, the open computing language (OpenCL) + graphics processing unit (GPU) heterogeneous parallel computing technology is used to simulate the eutectoid growth of Fe-C alloy and the end growth process of pearlite under pure diffusion. The effects of initial supercooling and different diffusion coefficients on the growth morphology of lamellar pearlite were investigated. The results show that ferrite and cementite are perpendicular to the front of the solid-solid interface and are coupled and coordinated to grow, and there is no leading phase under the initial supercooling degree of 20 K. With the continuous increase of the initial supercooling degree (19 K–22 K), the morphology changes of the eutectoid layer are as follows: cementite stops growing → slice amplitude increases → regular symmetric growth → oblique growth → layer merge. With the increase of the diffusion coefficient from 3×10−13 m2·s−1 to 15×10−13 m2·s−1, the growth rate of the microstructure of the lamellar pearlite increases linearly, and there is no obvious change in the frontal appearance of the pearlite.

Simulation of three-dimensional eutectic growth multi-phase field based on OpenCL parallel

Based on the muti-component eutectic multi-phase field model of S. G. Kim, W. T. Kim, T. Suzuki et al. [J. Cryst. Growth 261, 135–158 (2004)], the high-performance computing method of hardware and software architecture of OpenCL + graphics processing unit (GPU) was studied. Taking CBr4–C2Cl6 as an example, the evolution process of large-scale three-dimensional eutectic structure growth is realized by concurrent execution of multiple processes and multiple threads on two heterogeneous platforms of AMD and NVIDIA, respectively. The effects of different initial lamellar spacing and flow on eutectic lamellar morphology were also studied. The results show that, with the increasing of eutectic lamellar spacing, the morphology of eutectic lamellar changes are as follows: the symmetrically steady-state perpendicular to the interface is based on growth, the slightly oscillating state is unstable, and the large oscillating state is unstable; Under the condition of forced convection, the symmetrically alternating growth pattern of the original eutectic lamellar was broken, the melt flow led to the change of eutectic growth morphology, and the eutectic lamellar grew in the opposite direction of the flow. At the same computing scale, compared with the serial algorithm on the central processing unit platform, the acceleration ratio on the single GPU on the heterogeneous platform reaches 24.3 times and 21.6 times respectively, which improves the computing efficiency. At the same time, with its strong floating-point computing power to obtain more accurate simulation results and achieve the dual needs of computational efficiency and portability, it has also proven to solve the problems of a large amount of calculation, low efficiency and limited to qualitative research existing in the traditional phase-field models.

Design space exploration of multi-core RTL via high level synthesis from OpenCL models

As more and more powerful integrated circuits are appearing on the market, more and more applications, with very different requirements and workloads, are making use of available computing power. Designing optimized accelerators that can meet particular requirements has always presented a tremendous challenge to hardware engineers. To do so, designers have to trade off performance for power consumption in a manner such that the final RTL consumes minimum energy to meet the required performance (e.g. FLOPS) target. Moreover, the growing trend towards heterogeneous platforms is crucial to meet time and power consumption constraints of high-performance computing (HPC) applications. The OpenCL parallel programming language and framework enables programming CPU, GPU and recently FPGAs using the high-level synthesis (HLS) methodology. This work presents a design space exploration flow based on execution time, resource utilization and power consumption of OpenCL kernels mapped on FPGAs using the Xilinx high-level synthesis tool chain. Our experiments suggest that the quality of generated solutions, in terms of performance-per-watt, can be determined using analytical formulas prior to implementation, thus enabling fast and accurate DSE by considering on-chip and off-chip sources of parallelism. Moreover, the automated flow suggests design hints to meet a given time constraint within available resources. The proposed technique is demonstrated by optimizing the well known bitonic sorting network from NVIDIA’s OpenCL benchmark. Our results report that FPGAs have at least 20% higher performance-per-watt with respect to two high-end GPUs manufactured in the same technology (28 nm). Additionally, FPGAs with more available resources and using a more modern process (20 nm) can outperform the tested GPUs while consuming at least 55% less power at the cost of more expensive devices.

GPUBlocks: GUI Programming Tool for CUDA and OpenCL

Recent advances in general-purpose graphics processing units (GPGPUs) have resulted in massively parallel hardware that is widely available to achieve high performance in desktop, notebook, and even mobile computer systems. While multicore technology has become the norm of modern computers, programming such systems requires the understanding of underlying hardware architecture and hence posts a great challenge for average programmers, who might be professionals in specific domains, but not experts in parallel programming. This paper presents a GUI tool called GPUBlocks that can facilitate parallel programming on multicore computer systems. GPUBlocks is developed based on the OpenBlocks framework, an extendable tool for graphical programming, to construct the GUI-based programming environment for CUDA and OpenCL parallel computing platforms. Programmers simply need to drag-n-drop blocks, fill the fields of the blocks, and connect them according to array or matrix computations that are specified by algorithms. GPUBlocks can then translate block-based code to CUDA or OpenCL programs. Furthermore, a couple of optimization constructs have also been offered for rapid program optimization. Experimental results have shown that the generated CUDA and OpenCL programs can achieve reasonable speedups on GPUs. Consequently, GPUBlocks can be used as a tool for fast prototyping of GPU applications or a platform for educational parallel programming.

OpenCL-Accelerated Probabilistic Power Flow for Active Distribution Networks

Open computing language (OpenCL) is an open standard developed for general purpose parallel programming of heterogeneous computing devices. OpenCL can be used to exploit the massively parallel architecture of graphics processing units (GPU) in performing non graphic tasks. This paper presents an OpenCL parallel implementation of Monte-Carlo simulation based probabilistic power flow analysis for large active distribution systems. The proposed implementation adopts single precision arithmetic and is independent of third-party linear algebra libraries, which makes it suitable for standalone implementation and use on any mainstream GPU. Moreover, given that it adopts the OpenCL programing model, the proposed approach may also be used on multicore CPUs. The accuracy and speedup of the proposed implementation is validated using several test systems, of sizes ranging from 32 to 10476 buses. The results show that the proposed parallelized implementation can achieve substantial speedups without requiring high performance computing purposed hardware.

Fast Automatic Differentiation for Large Scale Bundle Adjustment

A parallelized implementation of automatic differentiation that derives from the problem of bundle adjustment is proposed in this paper. Reverse mode of automatic differentiation is more efficient to compute the derivatives of functions with $n$ real-value parameters, which is the case of computing the Jacobian matrix in bundle adjustment problem. By reason of a large number of small derivative computing tasks being needed in bundle adjustment problem, we implement an automatic differentiation library based on operator overloading and OpenCL parallel framework. In order to parallelize the computation in the framework of OpenCL, we generate forward and reverse computational sequences from computational graph by topological sorting. This library enables us to write down the function formula elegantly and then evaluate the derivatives rapidly and automatically. Finally, large scale bundle adjustment data sets is used to evaluate our proposed implementation. The result shows that our implementation runs about 3.6 times faster than Ceres Solver, which utilizes OpenMP parallel programming model.

Abstract ID: 184 OpenDNA: An OpenCL-based GPU Monte Carlo simulation code for microdosimetry

Introduction Geant4-DNA [1] , [2] is a Geant4 module that simulates the track structure of charged particles (electron/positron, proton and ions) in water as well as the radiolysis process that follows. Unlike the condensed-history technique in normal Geant4 simulations, Geant4-DNA implements an analogue Monte Carlo (AMC) technique that gives the event-by-event history of primary particle and all its descendants. Due to this, the simulation is extremely slow. To provide a faster alternative, this work implements a GPU-based AMC using the OpenCL parallel programming framework, OpenDNA. Methods For electrons and positron, ionization, excitation and elastic scattering were implemented. For protons, ionization and excitation were implemented. The low energy cut is 10 eV (10 keV) for electron (proton). The physical models and cross section data were extracted from Geant4-DNA. OpenDNA was preliminarily benchmarked against Geant4-DNA by radial dose distribution (RDD) simulations. For electron, a spherical geometry was used for a isotropic point source of E = 30 keV for both codes. For proton, a cylindrical geometry was used and a E = 800 keV proton was initiated along its axis. The ratio of OpenDNA RDD over Geant4-DNA RDD was computed to evaluate the consistency. 1,000,000 (and 100,000) primary histories were simulated for the electron (proton) case. The codes were run in a Mac Pro with two AMD FirePro D700 GPUs (for OpenDNA, using 100 threads) and a 2.7 GHz 12-Core Intel Xeon E5 CPU (Geant-DNA). Results The RDDs of OpenDNA and Geant4-DNA are consistent with each other within 1% for both particles. The simulation time of the electron (proton) case was reduced to 1/12 (1/10). Conclusion An analogue MC code was implemented using OpenCL and therefore is supposed to run in all OpenCL-conformed GPU. Future work include the simulation of physicochemical and chemical processes.

Simulating electron wave dynamics in graphene superlattices exploiting parallel processing advantages

This work introduces a parallel computing framework to characterize the propagation of electron waves in graphene-based nanostructures. The electron wave dynamics is modeled using both “microscopic” and effective medium formalisms and the numerical solution of the two-dimensional massless Dirac equation is determined using a Finite-Difference Time-Domain scheme. The propagation of electron waves in graphene superlattices with localized scattering centers is studied, and the role of the symmetry of the microscopic potential in the electron velocity is discussed. The computational methodologies target the parallel capabilities of heterogeneous multi-core CPU and multi-GPU environments and are built with the OpenCL parallel programming framework which provides a portable, vendor agnostic and high throughput-performance solution. The proposed heterogeneous multi-GPU implementation achieves speedup ratios up to 75x when compared to multi-thread and multi-core CPU execution, reducing simulation times from several hours to a couple of minutes. Program summaryProgram title: GslSim.Program Files doi:http://dx.doi.org/10.17632/prmfv63nj6.1Licensing provisions: GPLv3.Programming language: C, OpenCL and Matlab for results analysis.Nature of problem: Computing the time evolution of electron waves in graphene superlattices is a time consuming process due to the high number of necessary nodes to discretize the spatial and time domains.Solution method: We develop a simulator based on the C/OpenCL standards to study the time evolution of electron waves in graphene superlattices by exploiting hardware architectures such as graphics processing units (GPUs) to speedup the computation of the pseudospinor.

Parallel implementation and optimization of high definition video real-time dehazing

In some warning applications, such as aircraft taking-off and landing, ship sailing, and traffic guidance in foggy weather, the high definition (HD) and rapid dehazing of images and videos is increasingly necessary. Existing technologies for the dehazing of videos or images have not completely exploited the parallel computing capacity of modern multi-core CPU and GPU, and leads to the long dehazing time or the low frame rate of video dehazing which cannot meet the real-time requirement. In this paper, we propose a parallel implementation and optimization method for the real-time dehazing of the high definition videos based on a single image haze removal algorithm. Our optimization takes full advantage of the modern CPU+GPU architecture, which increases the parallelism of the algorithm, and greatly reduces the computational complexity and the execution time. The optimized OpenCL parallel implementation is integrate into FFmpeg as an independent module. The experimental results show that for a single image, the performance of the optimized OpenCL algorithm is improved approximately 500% compared with the existing algorithm, and approximately 153% over the basic OpenCL algorithm. The 1080p (1920 × 1080) high definition hazy video can also processed at a real-time rate (more than 41 frames per second).

Numerical simulation of polariton Bose gas thermalization

In this work, we present the numerical simulation of the process a Bose gas thermalization and the formation of the condensate. Our approach is based on kinetic equations and “Fermi's golden rule” in the incoherent approximation. Direct summation of terms is performed using GPGPU OpenCL parallel code using AMD Radeon HD 7970.

OpenCL parallel integration of ordinary differential equations: Applications in computational dynamics

In many physical problems the use of numerical simulations presents the only path to obtain insight into the behavior and evolution of the system of interest. GPU, CPU and MIC technologies are frequently employed for simulations on computational dynamics and we present results comparing different schemes for the numerical integration of ordinary differential systems (ODEs) in these architectures. The use of adapted methods with low memory storage (Low storage Runge–Kutta methods) gives good results for low precision studies, whereas the Taylor series method provides a powerful technique for high precision. We show how the computation of several dynamics indicators, such as a fast chaos indicator (FLI) or a phase shift indicator in small neuron networks (Central Pattern Generators), can be efficiently computed on these architectures by means of the numerical ODE methods executed through OpenCL. This high computational time reduction allows real-time simulations or generating video media. Program summaryProgram title: OCL-TIDES, OCL-RKCatalogue identifier: AEVW_v1_0Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEVW_v1_0.htmlProgram obtainable from: CPC Program Library, Queen’s University, Belfast, N. IrelandLicensing provisions: Standard CPC licence, http://cpc.cs.qub.ac.uk/licence/licence.htmlNo. of lines in distributed program, including test data, etc.: 1836No. of bytes in distributed program, including test data, etc.: 10,617Distribution format: tar.gzProgramming language: C and OpenCL.Computer: Any computer with a CPU or a GPU or a Xeon Phi.Operating system: Linux, MacOS, Windows.Has the code been vectorized or parallelized?: Yes, using the parallelization methods from OpenCL.RAM: Problem dependentSupplementary material: A video with a Fast Lyapunov indicator simulation is available (see Appendix A).Classification: 4.3, 6.5.External routines: OpenCL version 1.2Nature of problem: Solution of ODE problems in generic OpenCL environment oriented to large scale independent sets of initial conditions.Solution method: OpenCL parallel integrator based in Runge–Kutta or Taylor Series Method suitable for large sets of independent sets of initial conditions. Both are able to run either on CPU or in a GPU, using the parallelization methods from OpenCL.Running time: Problem dependent, sample tests take less than a minute to run.

OpenCL Parallel Blocked Approach for Solving All Pairs Shortest Path Problem on GPU

Pairs Shortest Path Problem (APSP) finds a large number of practical applications in real world. This paper presents a blocked parallel approach for APSP using an open standard framework OpenCL, which provides development environment for utilizing heterogeneous computing elements of computer system and to take advantage of massive parallel capabilities of multi-core processors such as graphics processing unit (GPU) and CPU. This blocked parallel approach exploits the local shared memory of GPU, thereby enhancing the overall performance. The proposed solution is for directed and dense graphs with no negative cycles and is based on blocked Floyd Warshall (FW) and Kleenes algorithm. Like Floyd Warshall this approach is also in-place and therefore requires no extra memory.

Adaptive discrete cosine transform-based image compression method on a heterogeneous system platform using Open Computing Language

Discrete cosine transform (DCT) is one of the major operations in image compression standards and it requires intensive and complex computations. Recent computer systems and handheld devices are equipped with high computing capability devices such as a general-purpose graphics processing unit (GPGPU) in addition to the traditional multicores CPU. We develop an optimized parallel implementation of the forward DCT algorithm for the JPEG image compression using the recently proposed Open Computing Language (OpenCL). This OpenCL parallel implementation combines a multicore CPU and a GPGPU in a single solution to perform DCT computations in an efficient manner by applying certain optimization techniques to enhance the kernel execution time and data movements. A separate optimal OpenCL kernel code was developed (CPU-based and GPU-based kernels) based on certain appropriate device-based optimization factors, such as thread-mapping, thread granularity, vector-based memory access, and the given workload. The performance of DCT is evaluated on a heterogeneous environment and our OpenCL parallel implementation results in speeding up the execution of the DCT by the factors of 3.68 and 5.58 for different image sizes and formats in terms of workload allocations and data transfer mechanisms. The obtained speedup indicates the scalability of the DCT performance.

Accelerating numerical modeling of wave propagation through 2-D anisotropic materials using OpenCL

We present an implementation of the numerical modeling of elastic waves propagation, in 2D anisotropic materials, using the new parallel computing devices (PCDs). Our study is aimed both to model laboratory experiments and explore the capabilities of the emerging PCDs by discussing performance issues. In the experiments a sample plate of an anisotropic material placed inside a water tank is rotated and, for every angle of rotation it is subjected to an ultrasonic wave (produced by a large source transducer) that propagates in the water and through the material producing some reflection and transmission signals that are recording by a “point-like” receiver. This experiment is numerically modeled by running a finite difference code covering a set of angles θ∈[−50°, 50°], and recorded the signals for the transmission and reflection results. Transversely anisotropic and weakly orthorhombic materials are considered. We accelerated the computation using an open-source toolkit called PyOpenCL, which lets one to easily access the OpenCL parallel computation API’s from the high-level programming environment of Python. A speedup factor over 19 using the GPU is obtained when compared with the execution of the same program in parallel using a CPU multi-core (in this case we use the 4-cores that has the CPU). The performance for different graphic cards and operating systems is included together with the full 2-D finite difference code with PyOpenCL.

OpenCL‐based acceleration of the FDTD method in computational electromagnetics

SUMMARYThe evolution of processors into multi‐core architectures has led to the acceleration of scientific codes using numerous highly specialized processors, that is, multi‐core central processing units (CPUs), graphics processing units (GPUs) and also devices that merge both technologies in a single‐die chip. Development of parallel codes that are both scalable and portable between the processor architectures is challenging. To overcome this limitation, we investigated the acceleration of the finite‐difference time‐domain (FDTD) method in computational electromagnetics on modern computing architectures, that is, multi‐core CPUs and GPUs, through the use of Open Computing Language (OpenCL). Further extension of the OpenCL parallel programing model with the Message Passing Interface allows for the targeting of standard distributed memory computer clusters as well as clusters accelerated by GPUs. Portability between hardware manufactured by different vendors and highly specialized and parallel computing architectures is the main advantage of the developed FDTD solvers. The codes were coupled with a commercial simulation platform to evaluate the performance of the solvers in real‐world industrial scenarios. Although the portability resulted in a slightly reduced performance (10–35%) of the OpenCL‐accelerated FDTD simulations compared with the native Compute Unified Device Architecture or Open Multiprocessing implementations, the obtained benchmarking results of the OpenCL FDTD solvers on distributed memory systems show that the communication overhead can be hidden by computations for sufficiently large simulation domains with a scaling efficiency higher than 90%. Copyright © 2012 John Wiley & Sons, Ltd.

Research and Application of Parallel Computing Technologies based on CUDA and OpenCL

The increased computational performance in science and engineering has led to the strong need for arithmetic intensive parallel computing, CUDA (Compute Unified Device Architecture) and OpenCL (Open Computing Language) are parallel computing technologies proposed in recent years, which both have a great expanse of application prospect in the field of high performance computing. In this paper, we discuss the parallel computing technologies based on CUDA and OpenCL by applying them on the smoking simulation based on real-time physics. Real-time physics requires high real-time quality and authenticity. We build the 3D smoking simulation model using the Computational Fluid Dynamics method and render the smoke using 3D texture volume rendering method to represent the 3D smoking simulation veridical and accurately. In order to ensure the real-time quality of 3D smoking simulation, we respectively adopt CUDA and OpenCL parallel computing technologies applied on different hardware platforms, achieving the solving of 3D smoking simulation in parallel. Experimental results show that compared with CPU serial programming, CUDA and OpenCL parallel computing technologies have predominant operability and inevitable efficiency advantages on arithmetic intensive parallel computing.