Computing On Graphics Processing Units Research Articles

Near Video-Rate Optical Coherence Elastography by Acceleration With a Graphics Processing Unit

We present a graphics processing unit (GPU)-accelerated optical coherence elastography (OCE) system capable of generating strain images (elastograms) of soft tissue at near video-rates. The system implements phase-sensitive compression OCE using a pipeline of GPU kernel functions to enable a highly parallel implementation of OCE processing using the OpenCL framework. Developed on a commercial-grade GPU and desktop computer, the system achieves a processing rate of 21 elastograms per second at an image size of 960 × 400 pixels, enabling high-rate visualization during acquisition. The system is demonstrated on both tissue-simulating phantoms and fresh ex vivo mouse muscle. To the best of our knowledge, this is the first implementation of near video-rate OCE and the fastest reported OCE processing rate, enabling, for the first time, a system capable of computing and displaying OCE elastograms interactively during acquisition. This advance provides new opportunities for medical imaging of soft tissue stiffness using optical methods.

Mixed Linear Model Approaches of Association Mapping for Complex Traits Based on Omics Variants.

Precise prediction for genetic architecture of complex traits is impeded by the limited understanding on genetic effects of complex traits, especially on gene-by-gene (GxG) and gene-by-environment (GxE) interaction. In the past decades, an explosion of high throughput technologies enables omics studies at multiple levels (such as genomics, transcriptomics, proteomics, and metabolomics). The analyses of large omics data, especially two-loci interaction analysis, are very time intensive. Integrating the diverse omics data and environmental effects in the analyses also remain challenges. We proposed mixed linear model approaches using GPU (Graphic Processing Unit) computation to simultaneously dissect various genetic effects. Analyses can be performed for estimating genetic main effects, GxG epistasis effects, and GxE environment interaction effects on large-scale omics data for complex traits, and for estimating heritability of specific genetic effects. Both mouse data analyses and Monte Carlo simulations demonstrated that genetic effects and environment interaction effects could be unbiasedly estimated with high statistical power by using the proposed approaches.

Efficient matrix-free GPU implementation of Fixed Grid Finite Element Analysis

This paper proposes a strategy for the efficient implementation of Fixed Grid Finite Element Analysis (FGFEA) method on Graphics Processing Units (GPUs). Such a strategy makes use of grid regularity of FGFEA to reduce drastically both the memory required by the implementation and the memory transactions to perform the operations with the common elemental stiffness matrix. The matrix-free method is adopted (i) to reduce the memory requirements obviating the assembly process of FEA and (ii) to exploit the parallelization potential of GPU architectures performing matrix–vector products at the Degree of Freedom (DoF) level. The underlying idea is to exploit data locality and maximize the use of on-chip memory, which increase notably the performance of GPU computing. The numerical experiments show that the proposed matrix-free GPU instance of FGFEA can achieve significant speedup over classical sparse-matrix CPU implementation using similar iterative solver.

A GPU-accelerated adaptive mesh refinement for immersed boundary methods

A new patch-based adaptive mesh refinement (AMR) code is developed. The fully threaded tree (FTT) data structure (Khokhlov, 1998) originally developed for cell-based AMR is extended to organize the patch-based adaptive meshes. We demonstrate the accuracy of the AMR code by performing the simulations of the two and three-dimensional Sedov blast wave problems. We measure the performance of the code by solving the same problems in different grid resolutions on a single CPU, where detailed timing analyses provide comparison between the current AMR code and a conventional oct-tree-based AMR code (FLASH 4.0.1) (Fryxell et al., 2000). For the two-dimensional cases, maximum speed-up ratio of 58.23 is obtained versus FLASH 4.0.1 with 5122 effective grid resolution. For the three-dimensional cases, maximum speed-up ratios of 32.16 is achieved over FLASH 4.0.1 with 1283 effective grid resolution. We present the implementation and performance of a patch-based AMR algorithm with the immersed boundary method (IBM) on Graphic Processing Units (GPU). We also applied several optimizations in GPU computation, including the asynchronous memory copy, the concurrent execution between CPU and GPU and hybrid MPI/OpenMP/GPU parallelization. Performance benchmarks are conducted on the GPU cluster on SHARCNET (https://www.sharcnet.ca/) using 1–8 Tesla M2070 GPUs. Maximum speed-up factors of 22.3 and 20.5 are demonstrated using one GPU and 8 GPUs with 20483 effective resolution, respectively.

SU‐E‐T‐423: Fast Photon Convolution Calculation with a 3D‐Ideal Kernel On the GPU

Purpose:The calculation time is a trade‐off for improving the accuracy of convolution dose calculation with fine calculation spacing of the KERMA kernel. We investigated to accelerate the convolution calculation using an ideal kernel on the Graphic Processing Units (GPU).Methods:The calculation was performed on the AMD graphics hardware of Dual FirePro D700 and our algorithm was implemented using the Aparapi that convert Java bytecode to OpenCL. The process of dose calculation was separated with the TERMA and KERMA steps. The dose deposited at the coordinate (x, y, z) was determined in the process. In the dose calculation running on the central processing unit (CPU) of Intel Xeon E5, the calculation loops were performed for all calculation points. On the GPU computation, all of the calculation processes for the points were sent to the GPU and the multi‐thread computation was done. In this study, the dose calculation was performed in a water equivalent homogeneous phantom with 1503 voxels (2 mm calculation grid) and the calculation speed on the GPU to that on the CPU and the accuracy of PDD were compared.Results:The calculation time for the GPU and the CPU were 3.3 sec and 4.4 hour, respectively. The calculation speed for the GPU was 4800 times faster than that for the CPU. The PDD curve for the GPU was perfectly matched to that for the CPU.Conclusion:The convolution calculation with the ideal kernel on the GPU was clinically acceptable for time and may be more accurate in an inhomogeneous region. Intensity modulated arc therapy needs dose calculations for different gantry angles at many control points. Thus, it would be more practical that the kernel uses a coarse spacing technique if the calculation is faster while keeping the similar accuracy to a current treatment planning system.

TU‐G‐BRA‐05: Predicting Volume Change of the Tumor and Critical Structures Throughout Radiation Therapy by CT‐CBCT Registration with Local Intensity Correction

Purpose:The purpose of this study is to develop an accurate and effective technique to predict and monitor volume changes of the tumor and organs at risk (OARs) from daily cone‐beam CTs (CBCTs).Methods:While CBCT is typically used to minimize the patient setup error, its poor image quality impedes accurate monitoring of daily anatomical changes in radiotherapy. Reconstruction artifacts in CBCT often cause undesirable errors in registration‐based contour propagation from the planning CT, a conventional way to estimate anatomical changes. To improve the registration and segmentation accuracy, we developed a new deformable image registration (DIR) that iteratively corrects CBCT intensities using slice‐based histogram matching during the registration process. Three popular DIR algorithms (hierarchical B‐spline, demons, optical flow) augmented by the intensity correction were implemented on a graphics processing unit for efficient computation, and their performances were evaluated on six head and neck (HN) cancer cases. Four trained scientists manually contoured nodal gross tumor volume (GTV) on the planning CT and every other fraction CBCTs for each case, to which the propagated GTV contours by DIR were compared. The performance was also compared with commercial software, VelocityAI (Varian Medical Systems Inc.).Results:Manual contouring showed significant variations, [‐76, +141]% from the mean of all four sets of contours. The volume differences (mean±std in cc) between the average manual segmentation and four automatic segmentations are 3.70±2.30(B‐spline), 1.25±1.78(demons), 0.93±1.14(optical flow), and 4.39±3.86 (VelocityAI). In comparison to the average volume of the manual segmentations, the proposed approach significantly reduced the estimation error by 9%(B‐spline), 38%(demons), and 51%(optical flow) over the conventional mutual information based method (VelocityAI).Conclusion:The proposed CT‐CBCT registration with local CBCT intensity correction can accurately predict the tumor volume change with reduced errors. Although demonstrated only on HN nodal GTVs, the results imply improved accuracy for other critical structures.This work was supported by NIH/NCI under grant R42CA137886.

G-SHOT: GPU accelerated 3D local descriptor for surface matching

Signature of histogram of orientations (SHOT) as a novel 3D object local descriptor can achieves a good balance between descriptiveness and robustness in surface matching. However, its computation workload is much higher than the other 3D local descriptors. This paper investigates the development of suitable massively parallel algorithms on the graphics processing unit (GPU) for computation of high density and large scale 3D object local descriptors through two alternative parallel algorithms; one exact, and one approximate. Both algorithms exhibit outstanding speedup performance. The exact parallel descriptor comes at no cost to the descriptiveness, with a speedup factor of up to 40.70, with respect to the serial SHOT on the central processing unit (CPU). The approximate version achieves a corresponding speedup factor of up to 54 with minor degradation in descriptiveness. The proposed algorithms are integrated into point cloud library (PCL), a open source project for image and point cloud.

Fine-grained GPU implementation of assembly-free iterative solver for finite element problems

This paper proposes a fine-grained implementation of matrix-free Conjugate Gradient (CG) solver for Finite Element Analysis (FEA) using Graphics Processing Unit (GPU) architectures. The use of GPU computing in FEA is today an active research field. This is primary due to current GPU sparse solvers are partially parallelizable and can hardly make use of Data-Level Parallelism (DLP) for which GPU architectures are designed. The proposed GPU instance takes advantage of Massively Parallel Processing (MPP) architectures performing well-balanced parallel calculations at the Degree-of-Freedom (DoF) level of finite elements. The numerical experiments evaluate and analyze the performance of diverse GPU instances of the matrix-free CG solver.

High-performance financial simulation using randomized quasi-Monte Carlo methods

Graphics Processing Unit (GPU) computing has become popular in computational finance, and many financial institutions are moving their CPU-based applications to the GPU platform. Since most Monte Carlo algorithms are embarrassingly parallel, they benefit greatly from parallel implementations, and consequently Monte Carlo has become a focal point in GPU computing. GPU speed-up examples reported in the literature often involve Monte Carlo algorithms, and there are software tools commercially available that help migrate Monte Carlo financial pricing models to GPU. We present a survey of Monte Carlo and randomized quasi-Monte Carlo methods, and discuss existing (quasi) Monte Carlo sequences in GPU libraries. We discuss specific features of GPU architecture relevant for developing efficient (quasi) Monte Carlo methods. We introduce a recent randomized quasi-Monte Carlo method, and compare it with some of the existing implementations on GPU, when they are used in pricing caplets in the LIBOR market model and mortgage-backed securities.

A graphics processing unit implementation of time-domain full-waveform inversion

The graphics processing unit (GPU) has become a popular device for seismic imaging and inversion due to its superior speed-up performance. We implemented GPU-based full-waveform inversion using the wavefield reconstruction strategy. Because computation on the GPU was much faster than CPU-GPU data communication, in our implementation, the boundaries of the forward modeling were saved on the device to avert the issue of data transfer between the host and device. We adopted the Clayton-Enquist absorbing boundary to maintain the efficiency of the GPU computation. A hybrid nonlinear conjugate gradient algorithm combined with the parallel reduction scheme was used to do computation in GPU blocks. The numerical results confirmed the validity of our implementation.

NMF-mGPU: non-negative matrix factorization on multi-GPU systems

BackgroundIn the last few years, the Non-negative Matrix Factorization ( NMF ) technique has gained a great interest among the Bioinformatics community, since it is able to extract interpretable parts from high-dimensional datasets. However, the computing time required to process large data matrices may become impractical, even for a parallel application running on a multiprocessors cluster.In this paper, we present NMF-mGPU, an efficient and easy-to-use implementation of the NMF algorithm that takes advantage of the high computing performance delivered by Graphics-Processing Units ( GPUs ). Driven by the ever-growing demands from the video-games industry, graphics cards usually provided in PCs and laptops have evolved from simple graphics-drawing platforms into high-performance programmable systems that can be used as coprocessors for linear-algebra operations. However, these devices may have a limited amount of on-board memory, which is not considered by other NMF implementations on GPU.Results NMF-mGPU is based on CUDA ( Compute Unified Device Architecture ), the NVIDIA’s framework for GPU computing. On devices with low memory available, large input matrices are blockwise transferred from the system’s main memory to the GPU’s memory, and processed accordingly. In addition, NMF-mGPU has been explicitly optimized for the different CUDA architectures. Finally, platforms with multiple GPUs can be synchronized through MPI ( Message Passing Interface ). In a four-GPU system, this implementation is about 120 times faster than a single conventional processor, and more than four times faster than a single GPU device (i.e., a super-linear speedup).ConclusionsApplications of GPUs in Bioinformatics are getting more and more attention due to their outstanding performance when compared to traditional processors. In addition, their relatively low price represents a highly cost-effective alternative to conventional clusters. In life sciences, this results in an excellent opportunity to facilitate the daily work of bioinformaticians that are trying to extract biological meaning out of hundreds of gigabytes of experimental information. NMF-mGPU can be used “out of the box” by researchers with little or no expertise in GPU programming in a variety of platforms, such as PCs, laptops, or high-end GPU clusters. NMF-mGPU is freely available at https://github.com/bioinfo-cnb/bionmf-gpu.

OpenACC acceleration of an unstructured CFD solver based on a reconstructed discontinuous Galerkin method for compressible flows

SummaryAn OpenACC directive‐based graphics processing unit (GPU) parallel scheme is presented for solving the compressible Navier–Stokes equations on 3D hybrid unstructured grids with a third‐order reconstructed discontinuous Galerkin method. The developed scheme requires the minimum code intrusion and algorithm alteration for upgrading a legacy solver with the GPU computing capability at very little extra effort in programming, which leads to a unified and portable code development strategy. A face coloring algorithm is adopted to eliminate the memory contention because of the threading of internal and boundary face integrals. A number of flow problems are presented to verify the implementation of the developed scheme. Timing measurements were obtained by running the resulting GPU code on one Nvidia Tesla K20c GPU card (Nvidia Corporation, Santa Clara, CA, USA) and compared with those obtained by running the equivalent Message Passing Interface (MPI) parallel CPU code on a compute node (consisting of two AMD Opteron 6128 eight‐core CPUs (Advanced Micro Devices, Inc., Sunnyvale, CA, USA)). Speedup factors of up to 24× and 1.6× for the GPU code were achieved with respect to one and 16 CPU cores, respectively. The numerical results indicate that this OpenACC‐based parallel scheme is an effective and extensible approach to port unstructured high‐order CFD solvers to GPU computing. Copyright © 2015 John Wiley & Sons, Ltd.

Faster, more accurate, parallelized inversion for shape optimization in electroheat problems on a graphics processing unit (GPU) with the real-coded genetic algorithm

Purpose – Inverting electroheat problems involves synthesizing the electromagnetic arrangement of coils and geometries to realize a desired heat distribution. To this end two finite element problems need to be solved, first for the magnetic fields and the joule heat that the associated eddy currents generate and then, based on these heat sources, the second problem for heat distribution. This two-part problem needs to be iterated on to obtain the desired thermal distribution by optimization. Being a time consuming process, the purpose of this paper is to parallelize the process using the graphics processing unit (GPU) and the real-coded genetic algorithm, each for both speed and accuracy. Design/methodology/approach – This coupled problem represents a heavy computational load with long wait-times for results. The GPU has recently been demonstrated to enhance the efficiency and accuracy of the finite element computations and cut down solution times. It has also been used to speedup the naturally parallel genetic algorithm. The authors use the GPU to perform coupled electroheat finite element optimization by the genetic algorithm to achieve computational efficiencies far better than those reported for a single finite element problem. In the genetic algorithm, coding objective functions in real numbers rather than binary arithmetic gives added speed and accuracy. Findings – The feasibility of the method proposed to reduce computational time and increase accuracy is established through the simple problem of shaping a current carrying conductor so as to yield a constant temperature along a line. The authors obtained a speedup (CPU time to GPU time ratio) saturating to about 28 at a population size of 500 because of increasing communications between threads. But this far better than what is possible on a workstation. Research limitations/implications – By using the intrinsically parallel genetic algorithm on a GPU, large complex coupled problems may be solved very quickly. The method demonstrated here without accounting for radiation and convection, may be trivially extended to more completely modeled electroheat systems. Since the primary purpose here is to establish methodology and feasibility, the thermal problem is simplified by neglecting convection and radiation. While that introduces some error, the computational procedure is still validated. Practical implications – The methodology established has direct applications in electrical machine design, metallurgical mixing processes, and hyperthermia treatment in oncology. In these three practical application areas, the authors need to compute the exciting coil (or antenna) arrangement (current magnitude and phase) and device geometry that would accomplish a desired heat distribution to achieve mixing, reduce machine heat or burn cancerous tissue. This process presented does it more accurately and speedily. Social implications – Particularly the above-mentioned application in oncology will alleviate human suffering through use in hyperthermia treatment planning in cancer treatment. The method presented provides scope for new commercial software development and employment. Originality/value – Previous finite element shape optimization of coupled electroheat problems by this group used gradient methods whose difficulties are explained. Others have used analytical and circuit models in place of finite elements. This paper applies the massive parallelization possible with GPUs to the inherently parallel genetic algorithm, and extends it from single field system problems to coupled problems, and thereby realizes practicable solution times for such a computationally complex problem. Further, by using GPU computations rather than CPU, accuracy is enhanced. And then by using real number rather than binary coding for object functions, further accuracy and speed gains are realized.

GPU-based prompt gamma ray imaging from boron neutron capture therapy.

The purpose of this research is to perform the fast reconstruction of a prompt gamma ray image using a graphics processing unit (GPU) computation from boron neutron capture therapy (BNCT) simulations. To evaluate the accuracy of the reconstructed image, a phantom including four boron uptake regions (BURs) was used in the simulation. After the Monte Carlo simulation of the BNCT, the modified ordered subset expectation maximization reconstruction algorithm using the GPU computation was used to reconstruct the images with fewer projections. The computation times for image reconstruction were compared between the GPU and the central processing unit (CPU). Also, the accuracy of the reconstructed image was evaluated by a receiver operating characteristic (ROC) curve analysis. The image reconstruction time using the GPU was 196 times faster than the conventional reconstruction time using the CPU. For the four BURs, the area under curve values from the ROC curve were 0.6726 (A-region), 0.6890 (B-region), 0.7384 (C-region), and 0.8009 (D-region). The tomographic image using the prompt gamma ray event from the BNCT simulation was acquired using the GPU computation in order to perform a fast reconstruction during treatment. The authors verified the feasibility of the prompt gamma ray image reconstruction using the GPU computation for BNCT simulations.

A Fading Channel Simulator Implementation Based on GPU Computing Techniques

Channel simulators are powerful tools that permit performance tests of the individual parts of a wireless communication system. This is relevant when new communication algorithms are tested, because it allows us to determine if they fulfill the communications standard requirements. One of these tests consists of evaluating the system performance when a communication channel is considered. In this sense, it is possible to model the channel as an FIR filter with time-varying random coefficients. If the number of coefficients is increased, then a better approach to real scenarios can be achieved; however, in that case, the computational complexity is increased. In order to address this issue, a design methodology for computing the time-varying coefficients of the fading channel simulators using consumer-designed graphic processing units (GPUs) is proposed. With the use of GPUs and the proposed methodology, it is possible for nonspecialized users in parallel computing to accelerate their simulation developments when compared to conventional software. Implementation results show that the proposed approach allows the easy generation of communication channels while reducing the processing time. Finally, GPU-based implementation takes precedence when compared with the CPU-based implementation, due to the scattered nature of the channel.

A Learning Approach to Introducing GPU Computing in Undergraduate Engineering Program

The graphics processing unit (GPU) learning initiative is developed within a project awarded by the Moroccan Fulbright Alumni Association (MFAA), entitled “GPU Acceleration of Human Genome Sequencing”. This project involves undergraduate students at Abdelmalek Essaadi University, and is conducted in collaboration with the High Performance Computing Lab (HPCL) at the George Washington University in U.S. The study brings together two of the most important topics and challenges for the medical field, Genomics, and information technology, parallel computing specially with Graphical Processing Units. The potential outcomes from the project will make very valuable contributions to medical and information technology research and will enrich the academic experience of the students. General Terms GPU programming, Learning-by-doing.

Neighborhood grid: A novel data structure for fluids animation with GPU computing

This paper introduces a novel and efficient data structure, called neighborhood grid, capable of supporting large number of particle based elements on GPUs (graphics processing units), and is used for optimizing fluid animation with the use of GPU computing. The presented fluid simulation approach is based on SPH (smoothed particle hydrodynamics) and uses a unique algorithm for the neighborhood gathering. The brute force approach to neighborhood gathering of n particles has complexity O(n2), since it involves proximity queries of all pairs of fluid particles in order to compute the relevant mutual interactions. Usually, the algorithm is optimized by using spatial data structures which subdivide the environment in cells and then classify the particles among the cells based on their position, which is not efficient when a large number of particles are grouped in the same cell. Instead of using such approach, this work presents a novel and efficient data structure that maintains the particles into another form of proximity data structure, called neighborhood grid. In this structure, each cell contains only one particle and does not directly represent a discrete spatial subdivision. The neighborhood grid does process an approximate spatial neighborhood of the particles, yielding promising results for real time fluid animation, with results that goes up to 9 times speedup, when compared to traditional GPU approaches, and up to 100 times when compared against CPU implementations.

Parallel computing on graphics processing units and heterogeneous platforms

Parallel computing on graphics processing units and heterogeneous platforms

Large-scale neural circuit mapping data analysis accelerated with the graphical processing unit (GPU).

Modern neuroscience research demands computing power. Neural circuit mapping studies such as those using laser scanning photostimulation (LSPS) produce large amounts of data and require intensive computation for post hoc processing and analysis. Here we report on the design and implementation of a cost-effective desktop computer system for accelerated experimental data processing with recent GPU computing technology. A new version of Matlab software with GPU enabled functions is used to develop programs that run on Nvidia GPUs to harness their parallel computing power. We evaluated both the central processing unit (CPU) and GPU-enabled computational performance of our system in benchmark testing and practical applications. The experimental results show that the GPU-CPU co-processing of simulated data and actual LSPS experimental data clearly outperformed the multi-core CPU with up to a 22× speedup, depending on computational tasks. Further, we present a comparison of numerical accuracy between GPU and CPU computation to verify the precision of GPU computation. In addition, we show how GPUs can be effectively adapted to improve the performance of commercial image processing software such as Adobe Photoshop. To our best knowledge, this is the first demonstration of GPU application in neural circuit mapping and electrophysiology-based data processing. Together, GPU enabled computation enhances our ability to process large-scale data sets derived from neural circuit mapping studies, allowing for increased processing speeds while retaining data precision.

Optimization of compute unified device architecture for real-time ultrahigh-resolution optical coherence tomography

We propose an optimized signal processing scheme that utilizes the compute unified device architecture (CUDA) for real-time spectral domain optical coherence tomography (OCT). Because linear spline interpolation and the direct spectral reshaping method have low data and control dependencies, these algorithms maximally utilize graphic processing unit (GPU) resources for dispersion control. In addition, data transfer between main memory and GPU, regarded as one of the most wasteful and time-consuming processes in GPU computing, is executed in parallel with the signal processing by overlapping kernel execution and data transfers. Experimental results obtained from application of the proposed scheme to a laboratory constructed OCT system comprising five spectrally shifted SLDs indicate that the OCT system has an axial resolution of 4.8μm and transverse resolution of 13μm in air. Further, coherence artifacts are reduced by 3–14dB over the side-lobes in the point spread function. The optimization of CUDA enables OCT imaging rates up to 350kHz (A-lines/sec) with a single GTX680 GPU.