GPGPU Research Articles

Block red–black MILU(0) preconditioner with relaxation on GPU

To accelerate the Krylov subspace-based linear equation solvers on Graphics Processing Units (GPUs), a stable, efficient and highly parallel preconditioner is essential. One of the strong candidates for such a preconditioner is the combination of the block red–black ordering and the relaxed modified incomplete LU factorization without fill-ins (MILU(0)). In this paper, we present techniques for implementing this type of preconditioner on General-purpose computing on GPU (GPGPU) using OpenACC. Our implementation is designed for 3-dimensional finite-difference computations with 7-point stencil, and the matrix storage format is optimized to realize coalesced memory access. Also, mixed-precision computation is employed to exploit the high single-precision performance of GPUs without sacrificing the accuracy of the computed solution. Extensive numerical tests were performed and the optimal values of various tunable parameters such as the number of blocks in each direction and the number of workers specified in OpenACC clauses are discussed. Performance comparison on NVIDIA Quadro GP100 and Tesla K40t GPUs shows that our solver is much faster than existing libraries like cuSPARSE, MAGMA, ViennaCL, and Ginkgo, especially when multiple linear equations with coefficient matrices sharing the same nonzero pattern are solved.

Computational Performance of MATLAB and Python for Electromagnetic Applications

MATLAB and Python are two commonly used scripting languages for prototyping electromagnetic problems today. Each of these languages provides access to computationally efficient functions allowing a user to easily run many math heavy problems with minimal programming. In this paper we will discuss the usage of MATLAB and a variety of libraries in Python capable of running these efficient computations. Tests will be run in both languages to compare both CPU and GPU computations. The runtimes of a variety of problems using each of these platforms will also be compared for a variety of mathematical operations typically used in electromagnetic problems. Finally, a simple angle of arrival calculation using conventional beamforming will be performed to show these speeds on a realistic problem.

WITHDRAWN: GPGPU-Based Fast Calculation Of Large Matrices And Dynamic Prediction Of Sports Injury Images

Abstract Sports injuries are usually direct impact and application of force than the body is structurally acceptable, is caused by excessive use. Common injuries, bruises, sprains, strains will include joint injury and nose bleeding. General-Purpose Computation on Graphics Processing Units (GPGPU) is Central Processing Units (CPU) usually using a graphics processing a processing unit a computer graphics calculations Graphics Processing Units (GPU), traditionally the central processing unit run the calculation of processing applications. Dynamic prediction, the patient is assumed to have been exposed to danger at a predictable point in time, has an intuitive representation associated with the estimated probability of the occurrence. The Drawback of the existing method does not give proper sports injury images and calculation. The proposed method of the Binary Support Vector System (BSVS) algorithm is to give the proper classification, result, and prediction.

How Effective are Retro-Inverso Peptides? Insights from MD Supported by Paramagnetic NMR Data

Early MD simulations were largely limited to observation of fast structural fluctuations in protein molecules. With recent advent of GPU computing, it became possible to model far more relevant dynamic processes: protein folding, ligand binding, allostery, etc. In our study, we use long unbiased MD simulations to model binding of the so-called retro-inverso (ri) peptide to its protein target. Retro-inverso peptides are designed from native peptides by replacing L-amino acids with D-amino acids and inverting the amino-acid sequence. They are believed to retain the binding affinity of their parent peptides, while being resistant to proteolytic degradation. We investigated the binding of retro-inverso peptide ri-Sos1 (rrrppvpppv) to N-terminal SH3 domain from adaptor protein Grb2. The peptide was initially placed in a random position inside a large simulation cell. During the course of the simulation, we observed the binding of ri-Sos1 at the canonical SH3 binding site. The resulting complex features a number of key hydrogen bonds that are similar to the ones found in the native Sos1/Grb2 N-SH3 structure. However, the pattern of these interactions has undergone some noticeable rearrangements. We argue that the concept of retro-inversion lacks any solid theoretical justification and, therefore, ri-peptides are only poor imitators of their parent peptides. To validate our findings, we conducted a number of NMR experiments. In particular, we used the peptides crrrppvpppv and rrrppvpppvc tagged with paramagnetic label MTSL to measure paramagnetic relaxation enhancements (PRE) in Grb2 N-SH3. The results were compared with the PRE rates rigorously calculated from the MD trajectories involving the two respective MTSL-tagged peptides. The 1HN,15N-HSQC-based titration data showed that ri-Sos1 is a much weaker binder than the original Sos1, thus confirming the tenuous nature of ri-peptide ligands. We acknowledge grant support from SPbU 51142660.

Blockwise Euclidean likelihood for spatio-temporal covariance models

A spatio-temporal blockwise Euclidean likelihood method for the estimation of covariance models when dealing with large spatio-temporal Gaussian data is proposed. The method uses moment conditions coming from the score of the pairwise composite likelihood. The blockwise approach guarantees considerable computational improvements over the standard pairwise composite likelihood method. In order to further speed up computation, a general purpose graphics processing unit implementation using OpenCL is implemented. The asymptotic properties of the proposed estimator are derived and the finite sample properties of this methodology by means of a simulation study highlighting the computational gains of the OpenCL graphics processing unit implementation. Finally, there is an application of the estimation method to a wind component data set.

Modelling of dynamic rock fracture process using the finite-discrete element method with a novel and efficient contact activation scheme

Within the framework of combined finite-discrete element method (FDEM), we propose a novel and efficient semi-adaptive contact activation approach (semi-ACAA). The semi-ACAA adaptively activates contact calculations for continuum solid elements around the cohesive element which has just been subjected to shear softening while its softening function has just satisfied a prescribed threshold. The semi-ACAA is implemented in a three-dimensional (3D) FDEM code parallelized based on general-purpose graphic processing units (GPGPU) to model dynamic fracture processes of marbles in dynamic Brazilian tensile strength (BTS) and uniaxial compressive strength (UCS) tests. The semi-ACAA not only overcomes spurious fracturing mode associated with FDEM simulations using ACAA but also is more physically sound and computationally efficient compared with brute-force contact activation approach (BCAA), which has been prevalent in FDEMs with intrinsic cohesive zone model (ICZM)-based cohesive elements. Furthermore, it is proven that the spurious fracturing mode is caused by unphysical movements of mesh due to missing contacts during the shear softening of cohesive elements instead of the sudden activation of contact force calculations as indicated in some literature. Finally, a series of simulations of the dynamic BTS and UCS tests conducted by split Hopkinson pressure bar (SHPB) facility reveals that, if the threshold is set to around unity, not only the spurious fracturing mode can be overcome but also the modelling precision of stress waves in intact rocks can be improved. The speed-up times of semi-ACAA compared with BCAA are between 1.23 and 13.5 depending on the intensity of shear cracking in the simulations including mixed-mode cracking. Since the proposed semi-ACAA is simple and can be easily implemented into any FDEM codes with ICZM-based cohesive elements, it is expected to accelerate future developments and applications of FDEM with ICZM-based cohesive elements in rock dynamics such as drilling and blasting.

Enabling Large-Scale Simulations of Quantum Transport with Manycore Computing

The non-equilibrium Green’s function (NEGF) is being utilized in the field of nanoscience to predict transport behaviors of electronic devices. This work explores how much performance improvement can be driven for quantum transport simulations with the aid of manycore computing, where the core numerical operation involves a recursive process of matrix multiplication. Major techniques adopted for performance enhancement are data restructuring, matrix tiling, thread scheduling, and offload computing, and we present technical details on how they are applied to optimize the performance of simulations in computing hardware, including Intel Xeon Phi Knights Landing (KNL) systems and NVIDIA general purpose graphic processing unit (GPU) devices. With a target structure of a silicon nanowire that consists of 100,000 atoms and is described with an atomistic tight-binding model, the effects of optimization techniques on the performance of simulations are rigorously tested in a KNL node equipped with two Quadro GV100 GPU devices, and we observe that computation is accelerated by a factor of up to ∼20 against the unoptimized case. The feasibility of handling large-scale workloads in a huge computing environment is also examined with nanowire simulations in a wide energy range, where good scalability is procured up to 2048 KNL nodes.

Optimizing Neural Networks for Efficient FPGA Implementation: A Survey

The deep learning has become the key for artificial intelligence applications development. It was successfully used to solve computer vision tasks. But the deep learning algorithms are based on Deep Neural Networks (DNN) with many hidden layers which need a huge computation effort and a big storage space. Thus, the general-purpose graphical processing units (GPGPU) are the best candidate for DNN development and inference because of their high number of processing core and the big integrated memory. In the other side, the disadvantage of the GPGPU is high-power consumption. In a real-world application, the processing unit is an embedded system based on limited power and computation resources. In recent years, Field Programmable Gate Array (FPGA) becomes a serious solution that can outperform GPGPU because of their flexible architecture and low power consumption. The FPGA is equipped with a very small integrated memory and a low bandwidth. To make DNNs fit into FPGA we need a lot of optimization techniques at different levels such as the network level, the hardware level, and the implementation tools level. In this paper, we will cite the existing optimization techniques and evaluate them to provide a complete overview of FPGA based DNN accelerators.

Field Trial of a Flexible Real-Time Software-Defined GPU-Based Optical Receiver

We introduce a flexible, software-defined real-time multi-modulation format receiver implemented on an off-the-shelf general-purpose graphics processing unit (GPU). The flexible receiver is able to process 2 GBaud 2-, 4-, 8-, and 16-ary pulse-amplitude modulation (PAM) signals as well as 1 GBaud 4-, 16- and 64-ary quadrature amplitude modulation (QAM) signals, with the latter detected using a Kramers-Kronig (KK) coherent receiver. Experimental performance evaluation is shown for back-to-back. In addition, by using the JGN high speed R&D network testbed, performance is evaluated after transmission over 91 km field-deployed optical fiber and reconfigurable optical add-drop multiplexers (ROADMs).

Parallelization of Frequent Itemset Mining Methods with FP-tree: An Experiment with PrePost+ Algorithm

Parallel processing has turn to be a common programming practice because of its efficiency and thus becomes an interesting field for researchers. With the introduction of multi- core processors as well as general purpose graphics processing units, parallel programming has become affordable. This leads to the parallelization of many of the complex data processing algorithms including algorithms in data mining. In this paper, a study on parallel PrePost+ is presented. PrePost+ is an efficient frequent itemset mining algorithm. The algorithm has been modified as a parallel algorithm and the obtained result is compared with the result of sequential PrePost+ algorithm

A General Technique for Real-Time Robotic Simulation in Manufacturing System

This paper describes a real-time simulator that allows the user in the factories to simulate arbitrary interaction between machinery and equipment. We discussed in details not only the general technique for developing such a real-time simulator but also the implementation of the simulator in its actual use. As such, people on the production line could benefit from observing and controlling robots in factories for preventing or reducing the severity of a collision, using the proposed simulator and its related technique. For that purpose, we divided the simulator into two main models: the real-time communication model and the simulation model. For the communication model, we present the distributed messaging server structure and the database semaphore technique for handling the real-time peer-to-peer (P2P) connections and control. For the simulation model, we used parallel programming and the general-purpose GPU (GPGPU) technique to speed up the processing time of the simulation. In the paper, we exhaustively explain how the simulator’s techniques are designed and implemented and verify two real world models with robots and robot-operating systems (ROS). Moreover, we compare the results between our technique and the traditional method.

RNA Modeling with the Computational Energy Landscape Framework.

The recent advances in computational abilities, such as the enormous speed-ups provided by GPU computing, allow for large scale computational studies of RNA molecules at an atomic level of detail. As RNA molecules are known to adopt multiple conformations with comparable energies, but different two-dimensional structures, all-atom models are necessary to better describe the structural ensembles for RNA molecules. This point is important because different conformations can exhibit different functions, and their regulation or mis-regulation is linked to a number of diseases. Problematically, the energy barriers between different conformational ensembles are high, resulting in long time scales for interensemble transitions. The computational potential energy landscape framework was designed to overcome this problem of broken ergodicity by use of geometry optimization. Here, we describe the algorithms used in the energy landscape explorations with the OPTIM and PATHSAMPLE programs, and how they are used in biomolecular simulations. We present a recent case study of the 5'-hairpin of RNA 7SK to illustrate how the method can be applied to interpret experimental results, and to obtain a detailed description of molecular properties.

HPC and Physics―A New World Opened by GPU Computing

HPC and Physics―A New World Opened by GPU Computing

A FAST AND ENERGY EFFICIENT PARALLEL IMAGE FILTERING IMPLEMENTATION ON RASPBERRY PI'S GPU

This paper presents a powerful processing technique for fast and energy-efficient image filtering algorithm focusing energy and time-sensitive embedded and robotic platforms. Digital video processing is getting more and more popular in battery-powered devices like mobile robots and smartphones whereas in most cases, it leads overhead on the main central processing unit (CPU) and it consumes a significant amount of energy from the battery. It is suitable for parallelism since there is no data dependency between the steps of the two-dimensional convolution algorithm. We propose a vector version of the two-dimensional convolution algorithm, which can run parallel on embedded processors that has general purpose graphic processing unit (GPGPU), to reduce computation time and energy consumption. Our in-depth experiments shows that using GPGPU could reduce the execution time while guaranteeing lower power consumption and offloading the system CPU. Experimental results showed that we achieved up to 105 times faster operation and 100 times less energy consumption compared to the CPU implementation. Besides, we reduced the CPU overhead up to 10 times.

Efficient Nearest-Neighbor Data Sharing in GPUs

Stencil codes (a.k.a. nearest-neighbor computations) are widely used in image processing, machine learning, and scientific applications. Stencil codes incur nearest-neighbor data exchange because the value of each point in the structured grid is calculated as a function of its value and the values of a subset of its nearest-neighbor points. When running on Graphics Processing Unit (GPUs), stencil codes exhibit a high degree of data sharing between nearest-neighbor threads. Sharing is typically implemented through shared memories, shuffle instructions, and on-chip caches and often incurs performance overheads due to the redundancy in memory accesses. In this article, we propose Neighbor Data (NeDa), a direct nearest-neighbor data sharing mechanism that uses two registers embedded in each streaming processor (SP) that can be accessed by nearest-neighbor SP cores. The registers are compiler-allocated and serve as a data exchange mechanism to eliminate nearest-neighbor shared accesses. NeDa is embedded carefully with local wires between SP cores so as to minimize the impact on density. We place and route NeDa in an open-source GPU and show a small area overhead of 1.3%. The cycle-accurate simulation indicates an average performance improvement of 21.8% and power reduction of up to 18.3% for stencil codes in General-Purpose Graphics Processing Unit (GPGPU) standard benchmark suites. We show that NeDa’s performance is within 13.2% of an ideal GPU with no overhead for nearest-neighbor data exchange.

Algorithm for replica redistribution in an implementation of the population annealing method on a hybrid supercomputer architecture

The population annealing method is a promising approach for large-scale simulations because it is potentially scalable on any parallel architecture. We present an implementation of the algorithm on a hybrid program architecture combining CUDA and MPI. The problem is to keep all general-purpose graphics processing unit devices as busy as possible by efficiently redistributing replicas. We provide details of testing on hardware based the Intel Skylake/Nvidia V100 running more than two million replicas of the Ising model sample in parallel. The results are quite encouraging because the acceleration grows toward the perfect line as the complexity of the simulated system increases.

Performance Analysis of Thread Block Schedulers in GPGPU and Its Implications

GPGPU (General-Purpose Graphics Processing Unit) consists of hardware resources that can execute tens of thousands of threads simultaneously. However, in reality, the parallelism is limited as resource allocation is performed by the base unit called thread block, which is not managed judiciously in the current GPGPU systems. To schedule threads in GPGPU, a specialized hardware scheduler allocates thread blocks to the computing unit called SM (Stream Multiprocessors) in a Round-Robin manner. Although scheduling in hardware is simple and fast, we observe that the Round-Robin scheduling is not efficient in GPGPU, as it does not consider the workload characteristics of threads and the resource balance among SMs. In this article, we present a new thread block scheduling model that has the ability of analyzing and quantifying the performances of thread block scheduling. We implement our model as a GPGPU scheduling simulator and show that the conventional thread block scheduling provided in GPGPU hardware does not perform well as the workload becomes heavy. Specifically, we observe that the performance degradation of Round-Robin can be eliminated by adopting DFA (Depth First Allocation), which is simple but scalable. Moreover, as our simulator consists of modular forms based on the framework and we publicly open it for other researchers to use, various scheduling policies can be incorporated into our simulator for evaluating the performance of GPGPU schedulers.

Learning Guided Convolutional Network for Depth Completion.

Dense depth perception is critical for autonomous driving and other robotics applications. However, modern LiDAR sensors only provide sparse depth measurement. It is thus necessary to complete the sparse LiDAR data, where a synchronized guidance RGB image is often used to facilitate this completion. Many neural networks have been designed for this task. However, they often naïvely fuse the LiDAR data and RGB image information by performing feature concatenation or element-wise addition. Inspired by the guided image filtering, we design a novel guided network to predict kernel weights from the guidance image. These predicted kernels are then applied to extract the depth image features. In this way, our network generates content-dependent and spatially-variant kernels for multi-modal feature fusion. Dynamically generated spatially-variant kernels could lead to prohibitive GPU memory consumption and computation overhead. We further design a convolution factorization to reduce computation and memory consumption. The GPU memory reduction makes it possible for feature fusion to work in multi-stage scheme. We conduct comprehensive experiments to verify our method on real-world outdoor, indoor and synthetic datasets. Our method produces strong results. It outperforms state-of-the-art methods on the NYUv2 dataset and ranks 1st on the KITTI depth completion benchmark at the time of submission. It also presents strong generalization capability under different 3D point densities, various lighting and weather conditions as well as cross-dataset evaluations. The code will be released for reproduction.

Carrier Phase Estimation Softwarized on GPU Using Decision-Aided Phase Unwrapping for Flexible Optical Coherent Access Systems

We present a unique carrier phase recovery algorithm which permits the softwarization of optical coherent receiver digital signal processing (DSP). The proposed algorithm with decision-aided phase unwrapping overcomes the serial dependency issue that has, up to now, prevented the softwarization of optical coherent systems. Its performance is verified through extensive Monte Carlo (MC) simulations and hardware experiments on quadrature phase-shift keying (QPSK) signaling. The proposed algorithm is implemented on a general-purpose graphic processing unit (GPU) and achieves 5-Gbit/s throughput for real-time streaming signals, as well as −44.5-dBm receiver sensitivity. The softwarization of carrier phase recovery function is demonstrated for the first time; our enhancement of disaggregated optical coherent receivers will significantly improve system flexibility and scalability, towards the realization of a new access networking that exploits general purpose hardware and virtualized transport layer functions.

Modeling of stereometric figures in a browser using WebGL

The article is devoted to WebGL technology. Previously, web browsers could only display simple graphics using а processor. Today's adopted hardware acceleration support allows browsers to access the computer's GPU to render increasingly complex 3D graphics. To facilitate working with WebGL, you can use ready-made libraries, for example three.js. Thе article provides an overview of the three.js library tools for creating 3D shapes in the browser. Snippets of JavaScript code that implement basic 3D primitives are provided. Ways to apply textures are shown.The results presented in the article will be useful to teachers of informatics when used in additional classes or when organizing project and research activities of schoolchildren in informatics. Students will be able to familiarize themselves with the promising JavaScript language and gain basic 3D modeling skills.