CUDA Research Articles

Advances in parallel and distributed computing and its applications

Advances in parallel and distributed computing and its applications

Parallel Dislocation Model Implementation for Earthquake Source Parameter Estimation on Multi-Threaded GPU

Graphics processing units (GPUs) have been in the spotlight in various fields because they can process a massive amount of computation at a relatively low price. This research proposes a performance acceleration framework applied to Monte Carlo method-based earthquake source parameter estimation using multi-threaded compute unified device architecture (CUDA) GPU. The Monte Carlo method takes an exhaustive computational burden because iterative nonlinear optimization is performed more than 1000 times. To alleviate this problem, we parallelize the rectangular dislocation model, i.e., the Okada model, since the model consists of independent point-wise computations and takes up most of the time in the nonlinear optimization. Adjusting the degree of common subexpression elimination, thread block size, and constant caching, we obtained the best CUDA optimization configuration that achieves 134.94×, 14.00×, and 2.99× speedups over sequential CPU, 16-threads CPU, and baseline CUDA GPU implementation from the 1000×1000 mesh size, respectively. Then, we evaluated the performance and correctness of four different line search algorithms for the limited memory Broyden–Fletcher–Goldfarb–Shanno with boundaries (L-BFGS-B) optimization in the real earthquake dataset. The results demonstrated Armijo line search to be the most efficient one among the algorithms. The visualization results with the best-fit parameters finally derived by the proposed framework confirm that our framework also approximates the earthquake source parameters with an excellent agreement with the geodetic data, i.e., at most 0.5 cm root-mean-square-error (RMSE) of residual displacement.

Development of software and algorithms of parallel learning of artificial neural networks using CUDA technologies

The object of research is to parallelize the learning process of artificial neural networks to automate the procedure of medical image analysis using the Python programming language, PyTorch framework and Compute Unified Device Architecture (CUDA) technology. The operation of this framework is based on the Define-by-Run model. The analysis of the available cloud technologies for realization of the task and the analysis of algorithms of learning of artificial neural networks is carried out. A modified U-Net architecture from the MedicalTorch library was used. The purpose of its application was the need for a network that can effectively learn with small data sets, as in the field of medicine one of the most problematic places is the availability of large datasets, due to the requirements for data confidentiality of this nature. The resulting information system is able to implement the tasks set before it, contains the most user-friendly interface and all the necessary tools to simplify and automate the process of visualization and analysis of data. The efficiency of neural network learning with the help of the central processor (CPU) and with the help of the graphic processor (GPU) with the use of CUDA technologies is compared. Cloud technology was used in the study. Google Colab and Microsoft Azure were considered among cloud services. Colab was first used to build a prototype. Therefore, the Azure service was used to effectively teach the finished architecture of the artificial neural network. Measurements were performed using cloud technologies in both services. The Adam optimizer was used to learn the model. CPU duration measurements were also measured to assess the acceleration of CUDA technology. An estimate of the acceleration obtained through the use of GPU computing and cloud technologies was implemented. CPU duration measurements were also measured to assess the acceleration of CUDA technology. The model developed during the research showed satisfactory results according to the metrics of Jaccard and Dyce in solving the problem. A key factor in the success of this study was cloud computing services.

Parallel Makespan Calculation for Flow Shop Scheduling Problem with Minimal and Maximal Idle Time

In this paper, a flow shop scheduling problem with minimal and maximal machine idle time with the goal of minimizing makespan is considered. The mathematical model of the problem is presented. A generalization of the prefix sum, called the job shift scan, for computing required shifts for overlapping jobs is proposed. A work-efficient algorithm for computing the job shift scan in parallel for the PRAM model with n processors is proposed and its time complexity of O(logn) is proven. Then, an algorithm for computing the makespan in time O(mlogn) in parallel using the prefix sum and job shift scan is proposed. Computer experiments on GPU were conducted using the CUDA platform. The results indicate multi-thread GPU vs. single-thread GPU speedups of up to 350 and 1000 for job shift scan and makespan calculation algorithms, respectively. Multi-thread GPU vs. single-thread CPU speedups up to 4.5 and 14.7, respectively, were observed as well. The experiments on the Taillard-based problem instances using a simulated annealing solving method and employing the parallel makespan calculation show that the method is able to perform many more iterations in the given time limit and obtain better results than the non-parallel version.

Multi-GPU implementation of a cellular automaton model for dendritic growth of binary alloy

A parallel three-dimensional (3D) cellular automaton (CA) model is developed for dendritic growth simulation during solidification of binary alloy. The CA model is implemented using the graphic processing unit (GPU). Compute Unified Device Architecture (CUDA) is combined with Message Passing Interface (MPI) to perform large-scale simulation on the multi-GPU architecture. A CUDA-Aware OpenMPI is used to reduce the communication cost between different GPU cards. The accuracy and the speedup ratio of the CA model are investigated. First, the simulated steady-state tip velocity under different undercooling is compared with the LGK analytical solution. Then, dendritic growth during directional solidification and equiaxed multi-dendrite growth are simulated. Dendritic growth simulation with a domain size up to 453 million grids is performed. Finally, the speedup ratio of the CA model using four GPUs is compared with that using four CPU processes via MPI, which leads to a maximal speedup ratio of 152.19. The developed CA model achieves high parallel performance compared with the MPI-based parallel model. The developed CA model is capable of predicting the columnar to equiaxed transition (CET) event during directional solidification. The results indicate that increasing cooling rate promotes the occurrence of the equiaxed dendrites in the melt ahead of the solidification front, and equiaxed dendrites are prone to nucleate in the columnar dendrite groove zone when a middle cooling rate is applied.

Investigation of MHD free convection of power‐law fluids in a sinusoidally heated enclosure using the MRT‐LBM

AbstractMagnetohydrodynamical free convective heat transfer flow of non‐Newtonian power‐law fluids has been analyzed numerically. For numerical simulation, the multiple‐relaxation‐time (MRT) lattice Boltzmann method (LBM) has been employed based on graphics process unit (GPU) computing. The computational domain is a two‐dimensional square cavity with thermally adiabatic upper and base walls, uniformly heated left wall, and sinusoidal heat distribution on the right wall. There is no research on non‐Newtonian fluid using MRT‐LBM using this configuration. Hence it is a new study in this field and novel results have been achieved with the magnetic field effect and sinusoidal wall temperature. The numerical simulations are carried out for different vital parameters such as thermal Rayleigh number , Hartmann number , and power‐law index () to study the characteristics of heat transport along with the flow physics inside the square enclosure. The streamlines and isotherms elucidate the distribution of fluid flow and energy within the cavity. The energy transfer phenomena are interpreted as the local Nusselt number distribution and the average Nusselt number along the heated wall. Computational results show that the rate of energy transfer is attenuated due to the augmentation of the magnetic field. However, exhibits direct correspondence with the Rayleigh number and an inverse relation with the power‐law index . The elevation of Hartmann number results in the diminution of and for pseudoplastic fluid () by 82.65% and 86.04%, respectively. Moreover, the temperature distribution () of amplifies by 95.83% compared to as the fluid transitions from pseudoplastic to dilation behavior. For GPU computing, the NVIDIA CUDA C programming platform is used in the present study.

Fast fringe enhancement by improved bidimensional sinusoids-assisted empirical mode decomposition

Fringe enhancement is necessary for optical measurement, which includes the elimination of noise and background from a fringe pattern. Bidimensional empirical mode decomposition (BEMD) algorithms have been used for fringe enhancement, but they usually decompose a fringe pattern into many bidimensional intrinsic mode functions (BIMFs), which is not only inefficient but also leads to difficulties for classifying the BIMFs to distinguish noise, the useful signal and background. In this paper, the bidimensional sinusoids-assisted empirical mode decomposition (BSEMD) algorithm is firstly studied in depth, and a key factor is found in deciding the results. Then an improved BSEMD (iBSEMD) algorithm is proposed, where the key factor is estimated easily and automatically. The proposed method largely reduces the number of BIMFs resulted, making the fringe enhancement become faster and have a more robust effect. Some optimized strategies are given additionally to implement the iBSEMD algorithm by compute unified device architecture (CUDA) for practical applications. Comparison experiments are conducted with state-of-the-art methods to prove the efficiency and effectiveness of our method.

Implementation of a parallel high-order WENO-type Euler equation solver using a CUDA PTX paradigm

ABSTRACT This study proposes the optimization of a low-level assembly code to reconstruct the flux for a splitting flux Harten–Lax–van Leer (SHLL) scheme on high-end graphic processing units. The proposed solver is implemented using the weighted essentially non-oscillatory reconstruction method to simulate compressible gas flows that are derived using an unsteady Euler equation. Instructions in the low-level assembly code, i.e. parallel thread execution and instruction set architecture in compute unified device architecture (CUDA), are used to optimize the CUDA kernel for the flux reconstruction method. The flux reconstruction method is a fifth-order one that is used to process the high-resolution intercell flux for achieving a highly localized scheme, such as the high-order implementation of SHLL scheme. Many benchmarking test cases including shock-tube and four-shock problems are demonstrated and compared. The results show that the reconstruction method is computationally very intensive and can achieve excellent performance up to 5183 GFLOP/s, ∼66% of peak performance of NVIDIA V100, using the low-level CUDA assembly code. The computational efficiency is twice the value as compared with the previous studies. The CUDA assembly code reduces 26.7% calculation and increases 37.5% bandwidth. The results show that the optimal kernel reaches up to 990 GB/s for the bandwidth. The overall efficiency of bandwidth and computation performance achieves 127% of the predicted performance based on the HBM2-memory roofline model estimated by Empirical Roofline Tool.

Performance evaluation of GPU- and cluster-computing for parallelization of compute-intensive tasks

PurposeThis paper aims to evaluate different approaches for the parallelization of compute-intensive tasks. The study compares a Java multi-threaded algorithm, distributed computing solutions with MapReduce (Apache Hadoop) and resilient distributed data set (RDD) (Apache Spark) paradigms and a graphics processing unit (GPU) approach with Numba for compute unified device architecture (CUDA).Design/methodology/approachThe paper uses a simple but computationally intensive puzzle as a case study for experiments. To find all solutions using brute force search, 15! permutations had to be computed and tested against the solution rules. The experimental application comprises a Java multi-threaded algorithm, distributed computing solutions with MapReduce (Apache Hadoop) and RDD (Apache Spark) paradigms and a GPU approach with Numba for CUDA. The implementations were benchmarked on Amazon-EC2 instances for performance and scalability measurements.FindingsThe comparison of the solutions with Apache Hadoop and Apache Spark under Amazon EMR showed that the processing time measured in CPU minutes with Spark was up to 30% lower, while the performance of Spark especially benefits from an increasing number of tasks. With the CUDA implementation, more than 16 times faster execution is achievable for the same price compared to the Spark solution. Apart from the multi-threaded implementation, the processing times of all solutions scale approximately linearly. Finally, several application suggestions for the different parallelization approaches are derived from the insights of this study.Originality/valueThere are numerous studies that have examined the performance of parallelization approaches. Most of these studies deal with processing large amounts of data or mathematical problems. This work, in contrast, compares these technologies on their ability to implement computationally intensive distributed algorithms.

A novel GPGPU-parallelized contact detection algorithm for combined finite-discrete element method

This paper presents a novel algorithm for contact detection for the three-dimensional combined finite-discrete element method (3D FDEM). The contact detection consists of two phases, neighbor search and fine search. Both phases are fully parallelized with Compute Unified Device Architecture (CUDA). In contrast to the non-binary search (NBS) algorithm, the proposed neighbor search algorithm is not sensitive to the element size distribution. To avoid the algorithm being performed at every step, an efficient GPGPU (general purpose graphic processing unit) -parallelized velocity reduction algorithm is employed. For the fine search, the concepts of “positive faces” and “positive edges” are proposed, and the process is considerably simplified and runs faster than the 3D separation axis algorithm. Numerical tests are performed to validate the efficiency and effectiveness of the proposed algorithms. The result shows that for both phases, the computation time is linearly proportional to the number of potential contact pairs regardless of the element size distribution. Furthermore, by using NVIDIA Telsa V100, the resultant overall speed-up ratios of the proposed contact detection algorithms relative to the original Y3D version for uniform and non-uniform element size distributions can reach up to 1982.6 and 13,894.7, respectively. For both quasi-static and dynamic problems, the simulated fracture patterns are in good agreement with the results generated by the NBS algorithm. The proposed methodology can be also employed for 2D FDEM and discrete element method (DEM).

GPU-accelerated multitiered iterative phasing algorithm for fluctuation X-ray scattering.

The multitiered iterative phasing (MTIP) algorithm is used to determine the biological structures of macromolecules from fluctuation scattering data. It is an iterative algorithm that reconstructs the electron density of the sample by matching the computed fluctuation X-ray scattering data to the external observations, and by simultaneously enforcing constraints in real and Fourier space. This paper presents the first ever MTIP algorithm acceleration efforts on contemporary graphics processing units (GPUs). The Compute Unified Device Architecture (CUDA) programming model is used to accelerate the MTIP algorithm on NVIDIA GPUs. The computational performance of the CUDA-based MTIP algorithm implementation outperforms the CPU-based version by an order of magnitude. Furthermore, the Heterogeneous-Compute Interface for Portability (HIP) runtime APIs are used to demonstrate portability by accelerating the MTIP algorithm across NVIDIA and AMD GPUs.

Numerical modelling of interaction between aluminium structure and explosion in soil

In this paper, a graphics processing unit-accelerated smoothed particle hydrodynamics solver is presented to simulate the three-dimensional explosions in soils and their damage to aluminium structures. To achieve this objective, a number of equations of state and constitutive models required to close the governing equations are incorporated into the proposed smoothed particle hydrodynamics framework, including the Jones-Wilkins-Lee equation of state for explosive materials, the Grüneisen equation of state for metals, the elastic-perfectly plastic constitutive model for metals, and the elastoplastic and elasto-viscoplastic constitutive models for soils. The proposed smoothed particle hydrodynamics methodology was implemented using the Compute Unified Device Architecture programming interface on an NVIDIA graphics processing unit in order to improve the computational efficiency. The various components of the proposed methodology were validated using four test cases, namely, a C4 detonation and an aluminium bar expanded by denotation to validate the modelling of explosion, a cylindrical Taylor bar impact test case to validate the modelling of large deformation in metals, a sand collapse test for the modelling of soils. Following the validation, the proposed method was used to simulate the detonation of an explosive material (C4) in soil and the concomitant deformation of an aluminium plate resulting from this explosion. The predicted results of this simulation are shown to be in good conformance with available experimental data. Finally, it is shown that the proposed graphics processing unit-accelerated SPH solver is able to model interaction problems involving millions of particles in a reasonable time.

Efficient parallelization of SPH algorithm on modern multi-core CPUs and massively parallel GPUs

Smoothed Particle Hydrodynamics (SPH) is fast emerging as a practically useful computational simulation tool for a wide variety of engineering problems. SPH is also gaining popularity as the back bone for fast and realistic animations in graphics and video games. The Lagrangian and mesh-free nature of the method facilitates fast and accurate simulation of material deformation, interface capture, etc. Typically, particle-based methods would necessitate particle search and locate algorithms to be implemented efficiently, as continuous creation of neighbor particle lists is a computationally expensive step. Hence, it is advantageous to implement SPH, on modern multi-core platforms with the help of High-Performance Computing (HPC) tools. In this work, the computational performance of an SPH algorithm is assessed on multi-core Central Processing Unit (CPU) as well as massively parallel General Purpose Graphical Processing Units (GP-GPU). Parallelizing SPH faces several challenges such as, scalability of the neighbor search process, force calculations, minimizing thread divergence, achieving coalesced memory access patterns, balancing workload, ensuring optimum use of computational resources, etc. While addressing some of these challenges, detailed analysis of performance metrics such as speedup, global load efficiency, global store efficiency, warp execution efficiency, occupancy, etc. is evaluated. The OpenMP and Compute Unified Device Architecture[Formula: see text] parallel programming models have been used for parallel computing on Intel Xeon[Formula: see text] E5-[Formula: see text] multi-core CPU and NVIDIA Quadro M[Formula: see text] and NVIDIA Tesla p[Formula: see text] massively parallel GPU architectures. Standard benchmark problems from the Computational Fluid Dynamics (CFD) literature are chosen for the validation. The key concern of how to identify a suitable architecture for mesh-less methods which essentially require heavy workload of neighbor search and evaluation of local force fields from neighbor interactions is addressed.

Development of a real-time magnetic island reconstruction system based on PCIe platform for HL-2A tokamak

A real-time magnetic island reconstruction (MIR) system based on PCI express platform for HL-2A tokamak is introduced. The front-end analog circuit and high performance analog-to-digital converters complete high-precision synchronous sampling of 18 channel Mirnov signals, and the application of PCIe platform and direct memory access technology enables high speed data transmission between graphics processing unit and field programmable gate array (FPGA). FPGA, as a mainstream high speed parallelizable computing tool, was used to implement the MIR algorithm, while a parameter table is established in an external double data rate SDRAM to improve the computational efficiency. The software of the MIR system is developed with Compute Unified Device Architecture 8.0 in Centos 6 system, which mainly realizes driver development, data transmission, network communication, parameter calculation and system control. This system has been tested in HL-2A plasma discharge experiment, and the reconstructed magnetic island structure can achieve a spatial resolution of 1.02 cm while the time resolution can reach 2 ms.

An Efficient Method to Compute EM Scattering From Target Covered With Honeycomb Composite Material

A new electromagnetic scattering model of target with honeycomb composite material is proposed in this letter by combining equivalent reflection coefficient model with high frequency method. First, the honeycomb composite is regarded as a uniaxial media based on strong interference theory. Then, the reflection coefficient of honeycomb composite with perfect electric conductor is derived with generalized propagation matrix method. Finally, the derived reflection coefficients are introduced in shooting and bouncing rays and physical optics (SBR-PO) method to obtain the scattering characteristic of complex target covered with honeycomb material. Moreover, the SBR-PO method is accelerated by the Compute Unified Device Architecture and OptiX from Nvidia. The accuracy of our method was demonstrated by comparing with full-wave method and measured data, respectively. The scattering of target covered with honeycomb material are simulated and discussed in this letter.

Boundary condition enforcement for renormalised weakly compressible meshless Lagrangian methods

This paper introduces a boundary condition scheme for weakly compressible (WC) renormalised first-order accurate meshless Lagrangian methods (MLM) by considering both solid and free surface conditions.A hybrid meshless Lagrangian method-finite difference (MLM-FD) scheme on prescribed boundary nodes is proposed to enforce Neumann boundary conditions. This is used to enforce symmetry boundary conditions and the implied Neumann pressure boundary conditions on solid boundaries in a manner consistent with the Navier-Stokes equation leading to the accurate recovery of surface pressures. The free surface boundary conditions allow all differential operators to be approximated by the same renormalised scheme while also efficiently determining free surface particles.The boundary conditions schemes are implemented for two renormalised MLMs. A WC smoothed particle hydrodynamics (SPH) solver is compared to a WC generalised finite difference (GFD) solver. Applications in both 2D and 3D are explored. A substantial performance benefit was found when comparing the WCGFD solver to the WCSPH solver with the WCGFD solver realising a maximum speedup in the range of three times over WCSPH in both 2D and 3D configurations. The solvers were implemented in C++ and used the NVIDIA CUDA 10.1 toolkit for the parallelisation of the solvers.

Study and evaluation of improved automatic GPU offloading method

With the slowing down of Moore's law, the use of hardware other than CPUs, such as graphics processing units (GPUs) or field-Programmable gate arrays (FPGAs), is increasing. However, when using heterogeneous hardware other than CPUs, barriers to technical skills, such for compute unified device architecture (CUDA) and open computing language (OpenCL), are high. Therefore, I previously proposed environment adaptive software that enables automatic conversion, configuration, and high-performance operation of once written code according to the hardware to be placed. As part of environment adaptive software, I also proposed a method to offload loop statements of applications to GPUs automatically. In this paper, I improved upon this automatic GPU offloading method to expand its applicability to more applications and enhance offloading performance. I implemented the improved method to evaluate its effectiveness for multiple applications.

Safer and More Efficient Parallel Cryptographic Algorithm and its Implementation in the GPU

In the digital world, the demand for data security during communication has increased. Hash functions are one of the cryptographic algorithms that provide data security in terms of data authenticity and integrity. Nowadays, most online applications require user authentication. These authentications are done on the server-side, which he must manage. As the number of applications increases, building a one-way function will be faster for calculating a hash value for small data such as passwords. In this paper, we will present a sequential cryptographic algorithm and its parallel implementation. We performed security analyses, executed comparisons for different amounts of data, and provided steps for further developing this algorithm. With the construction of this one-way function, we have provided the calculation of hash value in a shorter time for data in small quantities, which speeds up the authentication process on the server and thus speeds up the online services provided by the respective applications. A comparison was made between sequential implementation, parallel implementation on the CPU, and parallel implementation on the GPU using CUDA (Computer Unified Device Architecture) platform.

Implementation of a Framelet-Based Spectral Reconstruction for Multi-Slice Spiral CT

Spectral CT utilizes spectral information of X-ray sources to reconstruct energy-resolved X-ray images and has wide medical applications. Compared with conventional energy-integrated CT scanners, however, spectral CT faces serious technical difficulties in hardware, and hence its clinical use has been expensive and limited. The goal of this paper is to present a software solution and an implementation of a framelet-based spectral reconstruction algorithm for multi-slice spiral scanning based on a conventional energy-integrated CT hardware platform. In the present work, we implement the framelet-based spectral reconstruction algorithm using compute unified device architecture (CUDA) with bowtie filtration. The platform CUDA enables fast execution of the program, while the bowtie filter reduces radiation exposure. We also adopt an order-subset technique to accelerate the convergence. The multi-slice spiral scanning geometry with these additional features will make the framelet-based spectral reconstruction algorithm more powerful for clinical applications. The method provides spectral information from just one scan with a standard energy-integrating detector and produces color CT images, spectral curves of the attenuation coefficient at every point inside the object, and photoelectric images, which are all valuable imaging tools in cancerous diagnosis. The proposed algorithm is tested with a Catphan phantom and real patient data sets for its performance. In experiments with the Catphan 504 phantom, the synthesized color image reveals changes in the level of colors and details and the yellow color in Teflon indicates a special spectral property which is invisible in regular CT reconstruction. In experiments with clinical images, the synthesized color images provide some extra details which are helpful for clinical diagnosis, for example, details about the renal pelvis and lumbar join. The numerical studies indicate that the proposed method provides spectral image information which can reveal fine structures in clinical images and that the algorithm is efficient regarding to the computational time. Thus, the proposed algorithm has a great potential in practical application.

Tabu Genetic Cat Swarm Algorithm Analysis of Optimization Arrangement on Mistuned Blades Based on CUDA

Tabu genetic cat swarm optimization algorithm is proposed for optimization arrangement on mistuned blades. Furthermore, it is improved to be an innovative parallel algorithm based on Compute Unified Device Architecture (CUDA), whose performance is analyzed both in continuous and discrete solution space. The lumped parameter model and finite element model of the bladed-disk system are established for dynamics analysis and optimization verification. The complete mistuned parameter identification method and the improved mixed-interface prestressed substructure modal synthesis method are two other highlights. The results indicate that the algorithm in this paper has the advantages of low cost and high efficiency. The vibration localization and amplitude of the mistuned bladed-disk system are both reduced significantly. The optimization analysis method is proved to be applicable to the mistuned bladed-disk system of the aeroengine in engineering practice.