GPGPU Research Articles

OpenCL and OpenGL Implementation of Simultaneous Localization and Mapping Algorithm using High-End GPU

Simultaneous Localization And Mapping (SLAM) algorithms are being used in many robotic applications and autonomous navigation systems. The FastSLAM2.0 addresses an issue of the SLAM problem and allows a robot to navigate in an unknown environment. Several works have presented many algorithmic optimizations to reduce the computational complexity of such algorithm. In this paper, a GPGPU (general-purpose computing on graphics processing units) is exploited to achieve a parallel implementation of the FastSLAM2.0. The GPGPU acceleration is done using two different implementations for parallel programming. The first implementation used OpenGL shading language which is based on the characteristics of graphics hardwares. The second implementation used OpenCL which allows hardware acceleration across heterogeneous architectures. We also explored the impact of the two approaches on the the resulting GPGPU implementation. A comparison related to processing-time and localization accuracy is made using a real indoor dataset. Our results show a significant speedup of the GPGPU implementation over a Quad-Core CPU. We show also that, by adopting the same optimization methodology using the two approachs, the OpenCL implementation is faster and suitable for GPGPU accelerated SLAM algorithms.

Research on Programming Model and Compilation Optimization Technology of Multi-Core GPU

GPGPU (General Purpose Computing on Graphics Processing Units) has been widely applied to high performance computing. However, GPU architecture and programming model are different from that of traditional CPU. Accordingly, it is rather challenging to develop efficient GPU applications. This paper focuses on the key techniques of programming model and compiler optimization for many-core GPU, and addresses a number of key theoretical and technical issues. This paper proposes a many-threaded programming model ab-Stream, which would transparentize architecture differences and provide an easy to parallel, easy to program, easy to extend and easy to tune programming model. In addition, this paper proposes memory optimization and data transfer transformation according to data classification. Firstly, this paper proposes data layout pruning based on classification memory, and then proposes Ta T (Transfer after Transformed) for transferring Strided data between CPU and GPU. Experimental results demonstrate that proposed techniques would significantly improve performance for GPGPU applications.

Efficient Implementation of Lightweight Hash Functions on GPU and Quantum Computers for IoT Applications

Secure communication is important for Internet of Things (IoT) applications, to avoid cybersecurity attacks. One of the key security aspects is data integrity, which can be protected by employing cryptographic hash functions. Recently, US National Institute of Standards and Technology (NIST) announced a competition to standardize lightweight hash functions, which can be used in IoT applications. IoT communication involves various hardware platforms, from low-end microcontrollers to high-end cloud servers with GPU accelerators. Since many sensor nodes are connected to the gateway devices and cloud servers, performing high throughput integrity check is important to secure IoT applications. However, this is a time consuming task even for high-end servers, which may affect the response time in IoT systems. Moreover, no prior work had evaluated the performance of NIST candidates on contemporary processors like GPU and quantum computers. In this study, we showed that with carefully crafted implementation techniques, all the finalist hash function candidates in the NIST standardization competition can achieve high throughput (up-to 1,000 Gbps) on a RTX 3080 GPU. This research output can be used by IoT gateway devices and cloud servers to perform data integrity checks at high speed, thus ensuring a timely response. In addition, this is also the first study that showcase the implementation of NIST lightweight hash functions on a quantum computer (ProjectQ). Besides securing the communication in IoT, these efficient implementations on a GPU and quantum computer can be used to evaluate the strength of respective hash functions against brute-force attack.

Testing an Abnormal Grain Growth Theory via Large-scale Phase-field simulations

The mean-field theory proposed by Humphreys is frequently used to predict abnormal grain growth caused by nonuniform grain boundary properties. This theory enables the abnormal growth conditions of a particular grain to be described using only three parameters: the size ratio, grain boundary energy ratio, and mobility ratio between the particular grain and its surrounding matrix grains. However, detailed validation of the theory has yet to be reported. In this study, we elucidate the validity of the mean-field theory by performing very large-scale phase-field simulations of 2D and 3D abnormal grain growth based on parallel GPU computing. Systematic simulations are carried out while varying the initial size ratio, boundary energy ratio, and mobility ratio of a particular grain and matrix grains. Comparisons of the simulated theoretical results demonstrate that the abnormal growth behaviors (i.e., whether the abnormal growth occurs and the limiting size that can be reached by an abnormally growing grain) is well described by the mean-field theory.

Innovative meshless approach for shaped charges applications

We focus here on modelling shaped charges. Combining large deformations, numerous interfaceproductions, and strong damage mechanisms, those events are particularly challenging from a numerical point of view. Eulerian finite element methods are classically used for such modeling.However, they induce very long computation times, accuracy losses (projection algorithms), anddifficulties with opening criteria related to jet fragmentation. Among the Lagrangian approaches, the meshless method called Smoothed Particle Hydrodynamics (SPH) appears as a relevant alternative to prevent such shortcomings. Based on a set of moving interpolation points, it disregards any connectivity between its elements which makes it naturally well suited to handle material failure. Nevertheless, SPH schemes suffer from well-known instabilities questioning their accuracy and activating nonphysical processes,such as numerical fragmentation. Many stabilizing tools are available in the literature however, they either raise conservation and consistency issues or drastically increase the computation times. We propose then to use an alternative scheme called γ-SPH-ALE. Based on the ALE framework, it achieves robust and consistent stabilization through an arbitrary description of motion. Thanks to CFL-like conditions obtained through a nonlinear stability analysis, the scheme stability is ensured. By preventing spurious oscillations in elastic waves and correcting the so-called tensile instability, both stability and accuracy are increased regarding classical approaches. Also, taking advantage of GPU computing, such results are achieved in reduced computation times contrary to classical CPUimplementations. Its implementation on a “Viper” shaped charge shows that the scheme handles the jet generation process as well as its resulting interaction with a target.

Reducing GPU Energy Consumption by Packing Narrow-Width Operands

In this paper, we study the use of an Operand-Width-Aware Register (OWAR) packing mechanism for graphics processing unit (GPU) energy saving. In order to efficiently use the GPU register file (RF), OWAR employs a power gating method to shut down unused register sub-arrays in order to reduce dynamic and leakage energy consumption of RF. As the number of register accesses was reduced due to the packing of the narrow width operands, the dynamic energy dissipation was further decreased. Finally, with the help of RF usage optimized by register packing, OWAR allowed GPUs to support more thread-level parallelism (TLP) through assigning additional thread blocks on streaming multiprocessors (SMs) for general-purpose GPU (GPGPU) applications that suffered from the deficiency of register resources. The extra TLP opens opportunities for hiding more memory latencies and thus reducing the overall execution time, which can lower the overall energy consumption. We evaluated OWAR using a set of representative GPU benchmarks. The experimental results showed that compared to the baseline without optimization, OWAR can reduce the GPGPU’s total energy up to 29.6% and 9.5% on average. In addition, OWAR achieved performance improvement up to 1.97X and 1.18X on average.

Integrating tsunami simulations in web applications using BROWNI, an open source client-side GPU-powered tsunami simulation library

Tsunami simulation software is a key component of state-of-the-art early warning systems but the inherent complexities in phases of installation, execution, pre and post-processing prevent their use in other areas of risk management such as communication and education. Recent advances in software and computational capacities such as the efficiency of GPU computing and the ubiquity of web browsers bring new opportunities to bridge the gap between expert and non-expert users. Here we present a Javascript library to enable a web browser to facilitate gathering and analyzing data from tsunami simulations, by means of interactive and efficient visualizations. At its core, the library uses WebGL, the browser’s standard 3D graphics API, to run GPU accelerated computations of a tsunami model. A far-field tsunami model is implemented (linear shallow water equations discretized on spherical coordinates), and its implementation is validated against real tsunami observations, and benchmarked with two other tsunami software-packages. Two software platforms that use this library are presented to illustrate the powerful applications that can be developed for risk communication and education. These applications are characterized by their interactivity and fast computations, which enable users to focus on the understanding of the phenomena of tsunami propagation and iterate quickly to assess different scenarios and potential implications to tsunami risk management. Some limitations on this approach are discussed, in aspects such as scalability, performance, multi-threading and batch-processing, that can be relevant for other users. In our experience, the before mentioned benefits very well compensate the discussed limitations for this kind of applications. The library has an open source license, and is meant to be imported without modifying its source code to facilitate the creation of new applications as the ones herein presented.

Real-Time 10,000 km Straight-Line Transmission Using a Software-Defined GPU-Based Receiver

Real-time 10,000 km transmission over a straight-line link is achieved using a software-defined multi-modulation format receiver implemented on a commercial off-the-shelf general-purpose graphics processing unit (GPU). Minimum phase 1 GBaud (QAM) signals are transmitted over 10,000 km and successfully received after detection with a Kramers-Kronig (KK) coherent receiver. 8-, 16-, 32-, and 64-QAM are successfully transmitted over 7600, 5600, 3600, and 1600 km, respectively.

Parallel SVD Algorithm for a Three-Diagonal Matrix on a Video Card Using the Nvidia CUDA Architecture

SVD (Singular Value Decomposition) algorithm is used in recommendation systems, machine learning, image processing, and in various algorithms for working with matrices which can be very large and Big Data, so, given the peculiarities of this algorithm, it can be performed on a large number of computing threads that have only video cards.CUDA is a parallel computing platform and application programming interface model created by Nvidia. It allows software developers and software engineers to use a CUDA-enabled graphics processing unit for general purpose processing – an approach termed GPGPU (general-purpose computing on graphics processing units). The GPU provides much higher instruction throughput and memory bandwidth than the CPU within a similar price and power envelope. Many applications leverage these higher capabilities to run faster on the GPU than on the CPU. Other computing devices, like FPGAs, are also very energy efficient, but they offer much less programming flexibility than GPUs.The developed modification uses the CUDA architecture, which is intended for a large number of simultaneous calculations, which allows to quickly process matrices of very large sizes. The algorithm of parallel SVD for a three-diagonal matrix based on the Givents rotation provides a high accuracy of calculations. Also the algorithm has a number of optimizations to work with memory and multiplication algorithms that can significantly reduce the computation time discarding empty iterations.This article proposes an approach that will reduce the computation time and, consequently, resources and costs. The developed algorithm can be used with the help of a simple and convenient API in C ++ and Java, as well as will be improved by using dynamic parallelism or parallelization of multiplication operations. Also the obtained results can be used by other developers for comparison, as all conditions of the research are described in detail, and the code is in free access.

A comparative study on SoC embedded low power GPUs for real‐time edge‐based automated traffic surveillance

SummaryAchieving real‐time processing for automated traffic surveillance is a major challenge due to the huge amount of data generated by a large number of surveillance cameras. For this, centralized computing has been a standard architecture for many years. But due to the increasing need for real‐time processing and limitations of centralized computing architecture (such as network congestion due to limited bandwidth and the need for costly high‐end servers), the paradigm is shifting toward edge processing. This article compares the suitability of two popular but different types of SoCs (System on Chip) (from NVIDIA and Qualcomm) embedded with low‐power GPUs as processing units for edge devices. These GPUs can be programmed as general‐purpose GPUs (GPGPU) to speed up compute‐intensive tasks, so as to achieve real‐time processing at the edge of the network. The article also discusses the architectural features and differences between these GPUs and recommends optimization techniques to leverage them. For quantitative comparison, we implement a wrong‐way vehicle detection algorithm (using background‐subtraction‐based moving object detection and Kalman‐filter‐based trajectory tracking), and optimized it for these SoCs. The experimental results show that real‐time processing can be achieved for HD videos with implementations optimized for these SoCs.

The steady-state diffusional flux to isolated square cuboids in solution and supported on an inert substrate

• 3D finite difference simulation of the diffusional flux to a square cuboid. • Supporting a cube on a surface decreases the diffusional flux by ca. 35%. • Equivalent sphere size is not simply a function of the particle’s inherent size and shape. • Numerical computation undertaken using a general purpose graphic processor unit. Numerical simulation is used to characterise the steady-state diffusional flux to square cuboid particles both isolated in the solution phase and supported on an inert substrate. The diffusional flux is expressed as that to a sphere with an ‘effective’ or ‘equivalent’ radius allowing an experimentalist to determine the expected total flux to the square cuboid. For any given square cuboid particle, the effective radius depends on the particle shape, size and whether the square cuboid is supported or not.

상용 3 차원 고속연산 엔진과 지도정보 연동 도심 환경의 GNSS 다중경로신호 제거 및 실시간 항법해 개선 연구

This paper presents a technique to improve the accuracy of navigation of a satellite navigation system in the urban environment by effectively detecting and removing multipath signals in real time using 3D map information. To improve the satellite navigation accuracy, we designed a system that can implement an effective real-time ray-tracing calculation with 3D high-speed computation engine. In this process, objective multipath satellites can be identified and separated by collecting the actual satellite data around a complex urban environment and around the partial shadow areas during the satellite navigation. The Unity3D Engine ensures a real-time and efficient computing performance in large sized complex map environments and multiple access situations. By adapting the functionality of Unity3D, the GPU computation for satellite visibility is efficiently implemented in a complex map environment, resulting into the implementation of a real-time ray tracing function, even in a mobile platform. Consequently, the computation time for detecting a multipath signal with one set of satellite data is shortened. In addition, the root mean square error in estimating the horizontal position using the GPS L1 pseudorange least square navigation solution, which eliminates multipath signals, is reduced.

SAR Image Simulation of Complex Target including Multiple Scattering

We present a GPU-based computation for simulating the synthetic aperture radar (SAR) image of the complex target. To be more realistic, we included the multiple scattering field and antenna pattern tracking in producing the SAR echo signal for both Stripmap and Spotlight modes. Of the signal chains, the computation of the backscattering field is the most computationally intensive. To resolve the issue, we implement a computation parallelization for SAR echo signal generation. By profiling, the overall processing was identified to find which is the heavy loading stage. To further accommodate the hardware structure, we made extensive modifications in the CUDA kernel function. As a result, the computation efficiency is much improved, with over 224 times the speed up. The computation complexity by comparing the CPU and GPU computations was provided. We validated the proposed simulation algorithm using canonical targets, including a perfectly electric conductor (PEC), dielectric spheres, and rotated/unrotated dihedral corner reflectors. Additionally, the targets can be a multi-layered dielectric coating or a layered medium. The latter case aimed to evaluate the polarimetric response quantitively. Then, we simulated a complex target with various poses relative to the SAR imaging geometry. We show that the simulated images have high fidelity in geometric and radiometric specifications. The decomposition of images from individual scattering bounce offers valuable exploitation of the scattering mechanisms responsible for imaging certain target features.

Sparse attention block: Aggregating contextual information for object detection

It is well recognized that the contextual information of surrounding objects is beneficial for object detection. Such contextual information can often be obtained from long-range dependencies. This paper proposes a sparse attention block to capture long-range dependencies in an efficient way. Unlike the conventional non-local block, which generates a dense attention map to characterize the dependency between any two positions of the input feature map, our sparse attention block samples the most representative positions for contextual information aggregation. After searching for local peaks in a heat map of the given input feature map, it adaptively selects a sparse set of positions to represent the relationship between query and key elements. With the obtained sparse positions, our sparse attention block can well model long-range dependencies, and greatly improve the object detection performance at the additional cost of <2% GPU memory and computation of the conventional non-local block. This sparse attention block can be easily plugged into various object detection frameworks, such as Faster R-CNN, RetinaNet and Mask R-CNN. Experiments on COCO benchmark confirm that our sparse attention block can boost the detection accuracy with significant gains ranging from 1.4% to 1.9% and negligible overhead of computation and memory usage.

Heterogeneous CPU+GPU parallelization for high-accuracy scale-resolving simulations of compressible turbulent flows on hybrid supercomputers

A heterogeneous parallel algorithm for simulation of compressible turbulent flows and its portable software implementation are presented. The underlying numerical method is based on a family of higher accuracy edge-based reconstruction schemes on unstructured mixed-element meshes. The proposed parallel solution can engage a large number of computing devices of most of the existing computing architectures used in modern supercomputers, including manycore CPUs and GPUs. It is capable of co-execution on both CPUs and accelerators simultaneously. The multilevel parallel algorithm combines: MPI for distributing workload among hybrid cluster nodes and between devices inside nodes; OpenMP for manycore CPUs and other supporting devices, such as Intel Xeon Phi; OpenCL for massively-parallel accelerators, such as GPUs of various vendors, including NVIDIA, AMD, Intel. The main focus is on the adaptation of the numerical method and its computational algorithm to the stream processing parallel paradigm. The very limited device memory inherent in GPU computing is also taken into account. A detailed description of the parallel algorithm is presented, as well as the techniques used for its efficient parallel implementation. Special attention is paid to implicit time integration with its linear solver and calculation of convective fluxes and viscous terms. The use of mixed floating-point precision and overlapping communications and computations is also discussed. Parallel performance is demonstrated in practical applications on different kinds of supercomputers using up to 10 thousand cores and multiple GPUs of comparable overall performance.

Neural network fast-classifies biological images through features selecting to power automated microscopy.

Artificial intelligence is nowadays used for cell detection and classification in optical microscopy during post-acquisition analysis. The microscopes are now fully automated and next expected to be smart by making acquisition decisions based on the images. It calls for analysing them on the fly. Biology further imposes training on a reduced data set due to cost and time to prepare the samples and have the data sets annotated by experts. We propose a real-time image processing compliant with these specifications by balancing accurate detection and execution performance. We characterised the images using a generic, high-dimensional feature extractor. We then classified the images using machine learning to understand the contribution of each feature in decision and execution time. We found that the non-linear-classifier random forests outperformed Fisher's linear discriminant. More importantly, the most discriminant and time-consuming features could be excluded without significant accuracy loss, offering a substantial gain in execution time. It suggests a feature-group redundancy likely related to the biology of the observed cells. We offer a method to select fast and discriminant features. In our assay, a 79.6 2.4% accurate classification of a cell took 68.7 3.5 ms (mean SD, 5-fold cross-validation nested in 10 bootstrap repeats), corresponding to 14 cells per second, dispatched into eight phases of the cell cycle, using 12 feature groups and operating a consumer market ARM-based embedded system. A simple neural network offered similar performances paving the way to faster training and classification, using parallel execution on a general-purpose graphic processing unit. Finally, this strategy is also usable for deep neural networks paving the way to optimizing these algorithms for smartmicroscopy.

Combined DNA Damage Repair Inhibition and Ion Beam Therapy: Development, Benchmark and Clinical Implications of a Mechanistic Biological Model

Cellular radiosensitivity is tightly linked to DNA damage repair (DDR) capability. In a clinical context, inhibition of key DDR enzymes within tumor results in improved tumor control after radiotherapy and many preclinical and clinical trials of radiosensitizing drugs targeting DDR enzymes have been recently conducted or launched. Concurrently, interest in ion beam therapy continues to grow globally and its high precision may provide ideal treatment attributes in combination with tumor radiosensitizers. However, most studies of tumor radiosensitization based on DDR inhibition have been or are conducted using prevalent photon beams. Here, a mechanistic model of biological radiation response is presented to accurately translate findings in DDR inhibition studies to the case of charged particle radiation.The mechanistic "UNIfied and VErsatile bio-response Engine" (UNIVERSE) and its model of radiosensitization by DDR inhibition is extended to charged particles by implementing the heterogeneous dose distributions of protons and helium ions utilizing GPU computation. Predictions for monoenergetic and mixed fields over clinically relevant dose and LET range are compared to cell survival data from the literature and own dedicated experiments, including different cell lines and DDR inhibited variants. Our own experimental data represents the first comprehensive measurement of cell survival of repair competent and deficient cell lines in a helium spread-out bragg peak. Ultimately, the predictions of UNIVERSE are considered in a concrete clinical situation based on patient plan recalculations.UNIVERSE is able to precisely predict cell survival of DDR competent and deficient cell lines in clinically relevant settings of ion beam therapy based only on up to 3 parameters derived solely from conventional photon data. With increasing LET the effect of DDR deficiency reduces. This results in a diminished RBE of ion beam radiation for DDR deficient cells in comparison to DDR proficient cells.The capabilities of UNIVERSE could lead towards more personalized and "biological" planning in charged particle therapy, that could be guided by bio-marker based imaging in the future. By increasing the coverage of a tumor with high-LET using modifications of the treatment planning optimizer or advanced irradiation techniques, one could exploit the reduced impact of DDR inhibition at high LET to reduce the adverse effects of heterogeneous tumor radiosensitivity. UNIVERSE could also aid the appraisal of clinical viability of combined administration of radiosensitizing drugs and charged particle therapy and the identification of patients with known DDR inhibition which might not benefit from ion beam therapy, facilitating the management of limited capacities at therapy centers.

Simulation on system configuration for stationary head CT using linear carbon nanotube x-ray source arrays.

Purpose: The invention of carbon nanotube (CNT) x-ray source arrays has enabled the development of novel imaging systems, including stationary tomosynthesis and stationary computed tomography (CT) with fast data acquisition, mechanically robust structures, and reduced image blur from source-detector motion. In this work, we report the results of simulation studies of potential system configurations for a stationary head CT (s-HCT) using linear CNT x-ray sources and detector arrays. Approach: We explored s-HCT configurations that utilize one, two, and three linear CNT source arrays. Simulations were implemented using three digital phantoms with both CPU and GPU computing. Sinogram coverage was used for qualitative evaluation of the CT projection collection efficiency for each configuration. A modified low-contrast Shepp-Logan (SL) phantom was implemented for image quality assessment using quantitative metrics. Different iterative reconstruction (IR) methods were compared with both qualitative and quantitative assessments. Results: Sinogram coverage of s-HCT configurations was sensitive to the number of CNT source arrays and geometry. The simulations suggest that a s-HCT configuration with three planes gives near complete sinogram coverage. Such a configuration enables accurate reconstruction of the low-contrast SL phantom and considerably diminished artifacts caused by the system geometry. Conclusions: An optimized s-HCT system configuration with three linear CNT x-ray source arrays is feasible. IR algorithms can diminish artifacts caused by sparse and asymmetrical scans. The proposed s-HCT system configuration is currently under construction.

Predictive modeling and analysis of runout distance of physical mudflows based on a discrete element method

ABSTRACT Both visualization simulation and numerical simulation play important roles in mudflow simulation. However, the computational bottlenecks and the correctness of mudflow simulation supporting a large-scale particle system are still key problems to be solved in modeling of mudflow dynamics. In this article, we propose a discrete element method that uses many-core graphics processing units to accelerate the simulation of such a depositional process and to predict runout distances for physical mudflows. A model constructed using this method was verified for accuracy and applicability by a comparative analysis of physical mudflow flume tests and corresponding simulation tests. Our results showed that the runout distance prediction model, constructed using the discrete element method, can simulate the depositional processes of physical mudflows (supporting the simulation of tens of millions of particles), and can simulate their runout distance and deposit morphologies. Because the heterogeneous CPU–GPU computing approach can effectively improve the simulation accuracy by using more mudflow particles, the error in the simulation of the runout distance of the physical mudflows was within a small and acceptable range. Thus, the model can be used for further studies on the prediction of risks associated with mudflows.

A User-Centric Data Protection Method for Cloud Storage Based on Invertible DWT

Protection on end users’ data stored in Cloud servers becomes an important issue in today’s Cloud environments. In this paper, we present a novel data protection method combining Selective Encryption (SE) concept with fragmentation and dispersion on storage. Our method is based on the invertible Discrete Wavelet Transform (DWT) to divide agnostic data into three fragments with three different levels of protection. Then, these three fragments can be dispersed over different storage areas with different levels of trustworthiness to protect end users’ data by resisting possible leaks in Clouds. Thus, our method optimizes the storage cost by saving expensive, private, and secure storage spaces and utilizing cheap but low trustworthy storage space. We have intensive security analysis performed to verify the high protection level of our method. Additionally, the efficiency is proved by implementation of deploying tasks between CPU and General Purpose Graphic Processing Unit (GPGPU) in an optimized manner.