Graphics Processing Unit Platform Research Articles

Directive-based GPU programming for computational fluid dynamics

Directive-based programming of graphics processing units (GPUs) has recently appeared as a viable alternative to using specialized low-level languages such as CUDA C and OpenCL for general-purpose GPU programming. This technique, which uses “directive” or “pragma” statements to annotate source codes written in traditional high-level languages, is designed to permit a unified code base to serve multiple computational platforms. In this work we analyze the popular OpenACC programming standard, as implemented by the PGI compiler suite, in order to evaluate its utility and performance potential in computational fluid dynamics (CFD) applications. We examine the process of applying the OpenACC Fortran API to a test CFD code that serves as a proxy for a full-scale research code developed at Virginia Tech; this test code is used to asses the performance improvements attainable for our CFD algorithm on common GPU platforms, as well as to determine the modifications that must be made to the original source code in order to run efficiently on the GPU. Performance is measured on several recent GPU architectures from NVIDIA and AMD (using both double and single precision arithmetic) and the accelerator code is benchmarked against a multithreaded CPU version constructed from the same Fortran source code using OpenMP directives. A single NVIDIA Kepler GPU card is found to perform approximately 20× faster than a single CPU core and more than 2× faster than a 16-core Xeon server. An analysis of optimization techniques for OpenACC reveals cases in which manual intervention by the programmer can improve accelerator performance by up to 30% over the default compiler heuristics, although these optimizations are relevant only for specific platforms. Additionally, the use of multiple accelerators with OpenACC is investigated, including an experimental high-level interface for multi-GPU programming that automates scheduling tasks across multiple devices. While the overall performance of the OpenACC code is found to be satisfactory, we also observe some significant limitations and restrictions imposed by the OpenACC API regarding certain useful features of modern Fortran (2003/8); these are sufficient for us to conclude that it would not be practical to apply OpenACC to our full research code at this time due to the amount of refactoring required.

Fast parallel algorithm for three-dimensional distance-driven model in iterative computed tomography reconstruction**Projected supported by the National High Technology Research and Development Program of China (Grant No. 2012AA011603) and the National Natural Science Foundation of China (Grant No. 61372172).

The projection matrix model is used to describe the physical relationship between reconstructed object and projection. Such a model has a strong influence on projection and backprojection, two vital operations in iterative computed tomographic reconstruction. The distance-driven model (DDM) is a state-of-the-art technology that simulates forward and back projections. This model has a low computational complexity and a relatively high spatial resolution; however, it includes only a few methods in a parallel operation with a matched model scheme. This study introduces a fast and parallelizable algorithm to improve the traditional DDM for computing the parallel projection and backprojection operations. Our proposed model has been implemented on a GPU (graphic processing unit) platform and has achieved satisfactory computational efficiency with no approximation. The runtime for the projection and backprojection operations with our model is approximately 4.5 s and 10.5 s per loop, respectively, with an image size of 256×256×256 and 360 projections with a size of 512×512. We compare several general algorithms that have been proposed for maximizing GPU efficiency by using the unmatched projection/backprojection models in a parallel computation. The imaging resolution is not sacrificed and remains accurate during computed tomographic reconstruction.

Spatial resolution recovery utilizing multi-ray tracing and graphic processing unit in PET image reconstruction

Depth-of-interaction (DOI) poses a major challenge for a PET system to achieve uniform spatial resolution across the field-of-view, particularly for small animal and organ-dedicated PET systems. In this work, we implemented an analytical method to model system matrix for resolution recovery, which was then incorporated in PET image reconstruction on a graphical processing unit platform, due to its parallel processing capacity. The method utilizes the concepts of virtual DOI layers and multi-ray tracing to calculate the coincidence detection response function for a given line-of-response. The accuracy of the proposed method was validated for a small-bore PET insert to be used for simultaneous PET/MR breast imaging. In addition, the performance comparisons were studied among the following three cases: 1) no physical DOI and no resolution modeling; 2) two physical DOI layers and no resolution modeling; and 3) no physical DOI design but with a different number of virtual DOI layers. The image quality was quantitatively evaluated in terms of spatial resolution (full-width-half-maximum and position offset), contrast recovery coefficient and noise. The results indicate that the proposed method has the potential to be used as an alternative to other physical DOI designs and achieve comparable imaging performances, while reducing detector/system design cost and complexity.

Identification of material properties - efficient modelling approach based on guided wave propagation and spatial multiple signal classification

Modern structures are often designed using new types of lightweight materials of interesting properties. Accurate information on physical properties of these materials is a key element of every stage of lifecycle from design through maintenance to retirement. Although there are numerous experimental methods that can be used for material testing, only a small handful of these methods provide required information on material parameters in a nondestructive and in-operational manner, assuring high level of accuracy. The paper demonstrates application of a method that can be used to estimate material properties of engineering structures. The method is based on guided wave propagation and dispersion characteristics. The proposed approach combines three recently developed elements, that is, efficient numerical, experimental and image processing analyses: (i) wave propagation modelling is based on two finite difference approaches, that is, the semianalytical finite difference method and the 3-D local interaction simulation approach implemented with a multigeneral-purpose computing on graphics processing units platform to avoid numerical discrepancies and to reduce the computational effort; (ii) experimental testing utilises noncontact, scanning laser vibrometry; and (iii) image processing involves spatial multiple signal classification to improve dispersion curve estimation. This unique combination offers a reliable approach for material parameter estimation. The proposed method is fully nondestructive and can be performed online under varying operational conditions. The method is demonstrated using Young's modulus estimation of an aluminium plate. The results are compared using the traditional destructive approach based on a three-point bending test. Copyright © 2015 John Wiley & Sons, Ltd.

Performance evaluation of a 3D multi-view-based particle filter for visual object tracking using GPUs and multicore CPUs

This paper presents a deep and extensive performance analysis of the particle filter (PF) algorithm for a very compute intensive 3D multi-view visual tracking problem. We compare different implementations and parameter settings of the PF algorithm in a CPU platform taking advantage of the multithreading capabilities of the modern processors and a graphics processing unit (GPU) platform using NVIDIA CUDA computing environment as developing framework. We extend our experimental study to each individual stage of the PF algorithm, and evaluate the quality versus performance trade-off among different ways to design these stages. We have observed that the GPU platform performs better than the multithreaded CPU platform when handling a large number of particles, but we also demonstrate that hybrid CPU/GPU implementations can run almost as fast as only GPU solutions.

Adaptation of fluid model EULAG to graphics processing unit architecture

SummaryThe goal of this study is to adapt the multiscale fluid solver EULerian or LAGrangian framewrok (EULAG) to future graphics processing units (GPU) platforms. The EULAG model has the proven record of successful applications, and excellent efficiency and scalability on conventional supercomputer architectures. Currently, the model is being implemented as the new dynamical core of the COSMO weather prediction framework. Within this study, two main modules of EULAG, namely the multidimensional positive definite advection transport algorithm (MPDATA) and the variational generalized conjugate residual, elliptic pressure solver Generalized Conjugate Residual (GCR) are analyzed and optimized. In this paper, a method is proposed, which ensures a comprehensive analysis of the resource consumption including registers, shared, and global memories. This method allows us to identify bottlenecks of the algorithm, including data transfers between host and global memory, global and shared memories, as well as GPU occupancy. We put the emphasis on providing a fixed memory access pattern, padding as well as organizing computation in the MPDATA algorithm. The testing and validation of the new GPU implementation have been carried out based on modeling decaying turbulence of a homogeneous incompressible fluid in a triply‐periodic cube. Simulations performed using the standard version of EULAG and its new GPU implementation give similar solutions. Preliminary results show a promising increase in terms of computational efficiency. Copyright © 2014 John Wiley & Sons, Ltd.

Design space exploration of SW beamformer on GPU

SummaryUltrasound imaging has become one of the most widely used modalities in medical diagnosis today. However, real‐time ultrasound imaging requires large amount of data transfer and massive computation and therefore mainly relies on a complex dedicated hardware system. A recent trend of a graphics processing unit (GPU) based software‐based approach offers the advantages of flexibility and quick implementation. The GPUs have been reported as excellent accelerators across a wide range of applications. For best exploiting outstanding computational power and high memory bandwidth of a GPU, the paper explores the design space of implementing an ultrasound beamformer on a GPU platform. The design spaces are expanded by applying different optimization strategies to the beamformer on a GPU platform, and we also discuss the performance evaluation results on the various GPUs whose architectural characteristics are different to each others. The performance analysis shows that by optimizing CUDA code, our real‐time‐GPU‐based beamformer can be successfully implemented with 181 frames per second (fps) and speedup of 230.6X compared with the single‐threaded implementation on a high‐performance CPU platform. Copyright © 2014 John Wiley & Sons, Ltd.

Accelerated Harmonic-Balance Analysis Using a Graphical Processing Unit Platform

This paper describes a new approach to accelerate the simulation of the steady-state response of nonlinear circuits using the harmonic-balance (HB) technique. The approach presented in this paper focuses on the direct factorization of the Jacobian matrix, of the HB nonlinear equations, using a graphical processing unit (GPU) platform. The computational core of the proposed approach is based on developing a block-wise version of the KLU factorization algorithm, where scalar arithmetic operations are replaced by block-aware matrix operations. For a large number of harmonics, or excitation tones, or both, the Block-KLU (BKLU) approach effectively raises the ratio of floating-point operations to other operations and, therefore, becomes an ideal vehicle for implementation on a GPU-based platform. Motivated by this fact, we develop a GPU-based framework to implement the BKLU. The proposed approach yields speedup by up to 89 times over conventional direct factorization on CPU.

SU-D-BRD-03: A Gateway for GPU Computing in Cancer Radiotherapy Research

Purpose: Graphics Processing Unit (GPU) has become increasingly important in radiotherapy. However, it is still difficult for general clinical researchers to access GPU codes developed by other researchers, and for developers to objectively benchmark their codes. Moreover, it is quite often to see repeated efforts spent on developing low-quality GPU codes. The goal of this project is to establish an infrastructure for testing GPU codes, cross comparing them, and facilitating code distributions in radiotherapy community. Methods: We developed a system called Gateway for GPU Computing in Cancer Radiotherapy Research (GCR2). A number of GPU codes developed by our group and other developers can be accessed via a web interface. To use the services, researchers first upload their test data or use the standard data provided by our system. Then they can select the GPU device on which the code will be executed. Our system offers all mainstream GPU hardware for code benchmarking purpose. After the code running is complete, the system automatically summarizes and displays the computing results. We also released a SDK to allow the developers to build their own algorithm implementation and submit their binary codes to the system. The submitted code is then systematically benchmarked using a variety of GPU hardware and representative data provided by our system. The developers can also compare their codes with others and generate benchmarking reports. Results: It is found that the developed system is fully functioning. Through a user-friendly web interface, researchers are able to test various GPU codes. Developers also benefit from this platform by comprehensively benchmarking their codes on various GPU platforms and representative clinical data sets. Conclusion: We have developed an open platform allowing the clinical researchers and developers to access the GPUs and GPU codes. This development will facilitate the utilization of GPU in radiation therapy field.

Optimizing Hopfield Neural Network for Spectral Mixture Unmixing on GPU Platform

The Hopfield neural network (HNN) has been demonstrated to be an effective tool for the spectral mixture unmixing of hyperspectral images. However, it is extremely time consuming for such per-pixel algorithm to be utilized in real-world applications. In this letter, the implementation of a multichannel structure of HNN (named as MHNN) on a graphics processing unit (GPU) platform is proposed. According to the unmixing procedure of MHNN, three levels of parallelism, including thread, block, and stream, are designed to explore the peak computing capacity of a GPU device. In addition, constant and texture memories are utilized to further improve its computational performance. Experiments on both synthetic and real hyperspectral images demonstrated that the proposed GPU-based implementation works on the peak computing ability of a GPU device and obtains several hundred times of acceleration versus the CPU-based implementation while its unmixing performance remains unchanged.

Accelerating turbo similarity searching in chemoinformatics on multicore and GPU platforms

The increase in the database size of chemical compound requires a longer time in processing for any searching algorithms [1]. With the focus on an algorithm called Turbo Similarity Searching which have been proven to have good accuracy in retrieving actives [2], we propose that this algorithm be implemented on the widely-used many-cores and multi-cores processors which are easily available at lower cost. This would help medicinal chemist runs virtual screening faster while maintaining the accuracy. The many-cores processors are on-chip processors that could run simultaneously at a clock-cycle. Whilst the multi-cores processors are originally being developed to support graphics processing hence being called Graphics Processing Unit (GPU). However, the usage of GPU is now being generalized to include other general purpose operation [3]. Many works on the parallel field have tried implementing computational algorithms on this unit [4]. Taken into consideration on the compute intensive characteristics of TSS, we investigate the best method to parallelize it for better execution time. TSS is a two-phase algorithm in which its second phase is a compute intensive portion of the algorithm. Hence, if this phase could be accelerated, TSS would run faster and searching through a larger database would not be less a problem. This poster describes our experimental details and results on the matter.

Real-time motion tracking using optical flow on multiple GPUs

Abstract Motion tracking algorithms are widely used in computer vision related research. However, the new video standards, especially those in high resolutions, cause that current implementations, even running on modern hardware, no longer meet the needs of real-time processing. To overcome this challenge several GPU (Graphics Processing Unit) computing approaches have recently been proposed. Although they present a great potential of a GPU platform, hardly any is able to process high definition video sequences efficiently. Thus, a need arose to develop a tool being able to address the outlined problem. In this paper we present software that implements optical flow motion tracking using the Lucas-Kanade algorithm. It is also integrated with the Harris corner detector and therefore the algorithm may perform sparse tracking, i.e. tracking of the meaningful pixels only. This allows to substantially lower the computational burden of the method. Moreover, both parts of the algorithm, i.e. corner selection and tracking, are implemented on GPU and, as a result, the software is immensely fast, allowing for real-time motion tracking on videos in Full HD or even 4K format. In order to deliver the highest performance, it also supports multiple GPU systems, where it scales up very well

GPU accelerated lattice Boltzmann simulation for rotational turbulence

In this work, we numerically study decaying isotropic turbulence in periodic cubes with frame rotation using the lattice Boltzmann method (LBM) and present the results of rotation effects on turbulence. The implementation of LBM is on a GPU (Graphic Processing Unit) platform using CUDA (Compute Unified Device Architecture). Through the accelerated GPU-LBM simulation, we look into various effects of frame rotation on turbulence. It has been observed that rotation slows down the decay of kinetic energy and enstrophy. Rotation also breaks isotropy and induces vortex tubes in the direction of frame rotation. Characteristics related to velocity and its derivatives have been studied with and without rotation. Without rotation, the kinetic energy and enstrophy decay follow −10/7 and −17/7 scaling respectively whereas in the presence of rotation with the relatively small Rossby number (large rotation intensity), the energy decay slows down to −5/21 scaling when the initial isotropic turbulence energy spectrum is scaled to k4. These scalings with and without rotation are in quantitative agreements with the predictions from Kolmogorov hypotheses respectively. The skewness and kurtosis are seen more fluctuating in rotational turbulence, which agrees with the results from NS-based computation. Using this accelerated and validated GPU-LBM computation tool, we are further studying the inverse energy transfer behavior with and without rotation aiming to quantify the effects of rotation on the inverse energy transfer to reveal underlying physics of a particular stage of the turbulence development. The results will be presented in near future.

A parallel lattice Boltzmann method for large eddy simulation on multiple GPUs

To improve the simulation efficiency of turbulent fluid flows at high Reynolds numbers with large eddy dynamics, a CUDA-based simulation solution of lattice Boltzmann method for large eddy simulation (LES) using multiple graphics processing units (GPUs) is proposed. Our solution adopts the "collision after propagation" lattice evolution way and puts the misaligned propagation phase at global memory read process. The latest GPU platform allows a single CPU thread to control up to four GPUs that run in parallel. In order to make use of multiple GPUs, the whole working set is evenly partitioned into sub-domains. We implement Smagorinsky model and Vreman model respectively to verify our multi-GPU solution. These two LES models have different relaxation time calculation behavior and lead to different CUDA implementation characteristics. The implementation based on Smagorinsky model achieves 190 times speedup over the sequential implementation on CPU, while the implementation based on Vreman model archives more than 90 times speedup. The experimental results show that the parallel performance of our multi-GPU solution scales very well on multiple GPUs. Therefore large-scale (up to 10,240 $$\times $$ × 10,240 lattices) LES---LBM simulation becomes possible at a low cost, even using double-precision floating point calculation.

Efficient parallel implementation of three‐point viterbi decoding algorithm on CPU, GPU, and FPGA

SUMMARYIn wireless communication, Viterbi decoding algorithm (VDA) is the one of most popular channel decoding algorithms, which is widely used in WLAN, WiMAX, or 3G communications. However, the throughput of Viterbi decoder is constrained by the convolutional characteristic. Recently, the three‐point VDA (TVDA) was proposed to solve this problem. In TVDA, the whole procedure can be divided into three phases, the forward, trace‐back, and decoding phases. In this paper, we analyze the parallelism of TVDA and propose parallel TVDA on the multi‐core CPU, graphics processing unit (GPU), and field programmable gate array (FPGA). We demonstrate approaches that fully exploit its performance potential on CPU, GPU, and FPGA computing platforms. For CPU platforms, we perform two optimization methods, single instruction multiple data and multithreading to gain over 145 × speedup over the naive CPU version on a quad‐core CPU platform. For GPU platforms, we propose the combination of cached memory optimization, coalesced global memory accesses, codeword packing scheme, and asynchronous data transition, achieving the throughput of 404.65 Mbps and 12 × speedup over initial GPU versions on an NVIDIA GeForce GTX580 card and 7 × speedup over Intel quad‐core CPU i5‐2300, under the same manufacturing year and both with fully optimized schemes. In addition, for FPGA platforms, we customize a radix‐4 pipelined architecture for the TVDA in a 45‐nm FPGA chip from Xilinx (XC6VLX760). Under 209.15‐MHz clock rate, it achieves a throughput of 418.30 Mbps. Finally, we also discuss the performance evaluation and efficiency comparison of different flexible architectures for real‐time Viterbi decoding in terms of the decoding throughput, power consumption, optimization schemes, programming costs, and price costs.Copyright © 2013 John Wiley & Sons, Ltd.

SU‐E‐T‐242: A Gateway for GPU Computations in Radiotherapy

Purpose: Graphics Processing Unit (GPU) has become increasingly important in radiotherapy. However, it is still difficult for general clinical researchers to access GPU codes developed by other researchers with GPU expertise, and also for developers to benchmark their codes. It is quite often to see repeated efforts spent on developing GPU codes with low quality. The goal of this project is to establish an infrastructure for sharing GPU codes in the community. Methods: We deployed a GPU code sharing infrastructure on a GPU cluster. A number of codes developed in our group can be accessed via a web interface. To use the services, researchers first upload their test data or use the data provided by our system. Then they have to select the GPU device they are going to run the codes. Our system offers all mainstream GPU hardware for code benchmarking purpose. After the code running is complete, the system will automatically summarize and display the computing results. We will also release a SDK to allow the developers to build their own algorithm implementation and submit their binary codes to the system. The submitted code will be systematically benchmarked using a variety of GPU hardware and representative clinical data provided by our system. Results: This project provides a platform to the public to access a variety of GPU codes for radiotherapy research via a web interface. With the help of this platform, researchers are able to focus their efforts on clinical research. Developers will also benefit from this platform by benchmarking their codes on various GPU platforms and clinical data sets and comparing with other people's codes for the same application. Conclusion: The gateway for GPU code developers and clinical research users can greatly facilitate the adoption of GPU codes in radiotherapy.

A Parallel Reanalysis Method Based on Approximate Inverse Matrix for Complex Engineering Problems

The combined approximations (CA) method is an effective reanalysis approach providing high quality results. The CA method is suitable for a wide range of structural optimization problems including linear reanalysis, nonlinear reanalysis and eigenvalue reanalysis. However, with increasing complexity and scale of engineering problems, the efficiency of the CA method might not be guaranteed. A major bottleneck of the CA is how to obtain reduced basis vectors efficiently. Therefore, a modified CA method, based on approximation of the inverse matrix, is suggested. Based on the symmetric successive over-relaxation (SSOR) and compressed sparse row (CSR), the efficiency of CA method is shown to be much improved and corresponding storage space markedly reduced. In order to further improve the efficiency, the suggested strategy is implemented on a graphic processing unit (GPU) platform. To verify the performance of the suggested method, several case studies are undertaken. Compared with the popular serial CA method, the results demonstrate that the suggested GPU-based CA method is an order of magnitude faster for the same level of accuracy.

CPU–GPU hybrid parallel strategy for cosmological simulations

SUMMARYGadget is a simulation application for N‐body and smoothed particle hydrodynamics problems in cosmology, and it is widely applied in solving series of cosmological problems. N‐body focuses on the motion of the interaction of N particles, and smoothed particle hydrodynamics is a fluid simulation algorithm that studies the movement of fluid through particle simulation. Most scholars focus their attention on accelerating Gadget on multi‐core CPU or graphics processing units (GPUs) platforms. However, these research activities failed to achieve CPU–GPU hybrid computing, which resulted in tremendous waste of CPU computing resources.In this paper, we propose a CPU–GPU hybrid parallel strategy to accelerate Gadget‐2, a massively parallel structure formation code for cosmological simulations. This strategy uses CPU and GPU to process the calculation of short‐range force. To ensure CPU and GPU workload balance, a dynamic task allocation scheme is proposed according to the computational performance difference between the CPU and GPU.Experimental results showed that our CPU–GPU hybrid parallel strategy achieved an overall speedup factor of 18.6 and a partial speedup factor for short‐range force calculation of 28.35 compared with a single‐core CPU implementation for particles in million‐size magnitudes. Moreover, compared with a GPU platform that contained 12 CPU cores and one GPU, our hybrid parallel strategy obtained overall speedup and partial speedup factors of 6% and 20%, respectively. Furthermore, the scalability of the hybrid strategy is very fine – its performance will be enhanced when the problem scale is increasing. However, this strategy also has its limitation that the performance enhancement will be decreasing if the ratio(the number of CPU cores divides that of the GPU cards) reduces. Finally, in our hybrid strategy, the CPU coefficient of utilization improved by 17.14% or better. Copyright © 2013 John Wiley & Sons, Ltd.

GPU-Based Real-Time Volumetric Ultrasound Image Reconstruction for a Ring Array

Synthetic phased array (SPA) beamforming with Hadamard coding and aperture weighting is an optimal option for real-time volumetric imaging with a ring array, a particularly attractive geometry in intracardiac and intravascular applications. However, the imaging frame rate of this method is limited by the immense computational load required in synthetic beamforming. For fast imaging with a ring array, we developed graphics processing unit (GPU)-based, real-time image reconstruction software that exploits massive data-level parallelism in beamforming operations. The GPU-based software reconstructs and displays three cross-sectional images at 45 frames per second (fps). This frame rate is 4.5 times higher than that for our previously-developed multi-core CPU-based software. In an alternative imaging mode, it shows one B-mode image rotating about the axis and its maximum intensity projection, processed at a rate of 104 fps . This paper describes the image reconstruction procedure on the GPU platform and presents the experimental images obtained using this software.

Parallel unsupervised Synthetic Aperture Radar image change detection on a graphics processing unit

Change detection is now routinely applied in many application domains, such as damage assessment, environmental monitoring and agricultural surveys. As the number of remote sensing images and the complexity of algorithms rise, the demand for processing power is increasing. In this paper, we propose PLog-FLICM , a parallel algorithm for change detection, which includes two steps: (1) generate the difference image based on the log-ratio operator; (2) detect changes in the difference image by using a modified fuzzy c-means clustering algorithm. PLog-FLICM is implemented on AMD Accelerated Parallel Processing SDK based on Open Computing Language. The parallel characteristics and implementation details of the proposed PLog-FLICM algorithm are presented. Experiments on several Synthetic Aperture Radar images demonstrate that the proposed algorithm outperforms other algorithms, and the designed parallel algorithm can greatly reduce the computational time of the change detection algorithm. Furthermore, we investigate the performance portability of PLog-FLICM in the different central processing unit and graphics processing unit platforms. Experimental results show that they have also achieved good parallel performance.