Compute Unified Device Architecture Language Research Articles

DEM–SPH Coupling Method for Landslide Surge Based on a GPU Parallel Acceleration Technique

A high-performance DEM–SPH coupling method for landslide surge is proposed based on a graphics processing unit (GPU) parallel acceleration technique. A GPU-based simulation program, namely, G-DeSp, is developed using C++ and Compute Unified Device Architecture language, which can simulate 3D landslide surge with millions of particles precisely and efficiently. A two-phase flow dam-break test is adopted to validate the accuracy and efficiency of G-DeSp. Three-dimensional landslide surge simulations are conducted with the different diameter of solid particles and the maximum climb height of the surge by granular slide and rigid slide is analyzed. Results show that G-DeSp accurately simulates landslide surge, such as the movement and deformation of the landslide, the generation and propagation of surge waves, and the violent interaction of the fluid–solid. The velocity and displacement of the granular slides and surge decrease with the increase in the diameter of solid particles. The maximum climbing height of surge by the rigid slide is 6.5 times that of surge by the granular slide with the same volume, which indicates that the surge hazard induced by the rigid slide is much more serious than that induced by the granular slide.

An augmented HLLEM ADER numerical model parallel on GPU for the porous Shallow Water Equations

This paper presents a novel explicit finite volume scheme for the solution of the two-dimensional Shallow Water Equations (SWEs) with porosity based on the use of an augmented Riemann problem. Bottom and porous source terms are treated as non-conservative fluxes by including two additional equations to the porous SWEs. The HLLEM approximate Riemann solver is chosen to compute the fluxes of the non-conservative system of equations in order to ensure robustness, computational efficiency and entropy enforcement also in the presence of significant discontinuities in the porosity field. Moreover, the proposed solver is extended to second order in time and space adopting the ADER time-integration scheme, together with a Total Variation Diminishing (TVD) spatial reconstruction. The model is implemented in CUDA (Compute Unified Device Architecture) language to exploit the computational power of modern Graphic Processing Units. The theoretical and numerical verification of the C-property is achieved also when discontinuities characterize the porosity and bottom fields. The scheme accurately reproduces Riemann problems generated by the presence of porous discontinuities. Numerical tests also confirm that the proposed algorithm retains roughly the same numerical efficiency of an alternative formulation based on a non-augmented Riemann solver, while providing more accurate results.

비압축성 입자법 내 압력해법에 관한 GPU 가속화에 대한 연구

In the Moving Particle Simulation (MPS) method, which is a gridless particle method with simple algorithm for solving Navier-Stokes equation, insufficient number of particles in the vicinity of boundary has a considerable influence on the numerical calculation, which causes directly the flow field to become unstable and inaccurate. In particular, the Poisson Equation for pressure is solved iteratively at every time step to obtain an approximate solution. This gratly incereases the amount of computation when performing large-scale particle simulations, which requires high-performance computing. The main objective of this study is to develop a GPU(Graphics Processing Unit)-accelerated MPS code using CUDA(Compute Unified Device Architecture) language to solve the Poisson equation. Furthermore in order to compare the performance of the computational speed, a 3-dimensional calculation of broken-dam break problem has been carried out using the GPU-accelerated MPS method. It can be seen that the performance of the speed ratio in the GPU-based code is improved by about 10 times compared to that of the CPU-based code.

Optimisation of HEVC motion estimation exploiting SAD and SSD GPU‐based implementation

The new High-Efficiency Video Coding (HEVC) standard doubles the video compression ratio compared to the previous H.264/AVC at the same video quality and without any degradation. However, this important performance is achieved by increasing the encoder computational complexity. That's why HEVC complexity is a crucial subject. The most time consuming and the most intensive computing part of HEVC is the motion estimation based principally on the sum of absolute differences (SAD) or the sum of square differences (SSD) algorithms. For these reasons, the authors proposed an implementation of these algorithms on a low cost NVIDIA GPU (graphics processing unit) using the Fermi architecture developed with Compute Unified Device Architecture language. The proposed algorithm is based on the parallel-difference and the parallel-reduction process. The investigational results show a significant speed-up in terms of execution time for most 64 × 64 pixel blocks. In fact, the proposed parallel algorithm permits a significant reduction in the execution time that reaches up to 56.17 and 30.4%, compared to the CPU, for SAD and SSD algorithms, respectively. This improvement proves that parallelising the algorithm with the new proposed reduction process for the Fermi-GPU generation leads to better results. These findings are based on a static study that determines the PU percentage utilisation for each dimension in the HEVC. This study shows that the larger PUs are the most utilised in temporal levels 3 and 4, which attain 84.56% for class E. This improvement is accompanied by an average peak signal-to-noise ratio loss of 0.095 dB and a decrease of 0.64% in terms of BitRate.

Fast GPU-based Monte Carlo code for SPECT/CT reconstructions generates improved 177Lu images

BackgroundFull Monte Carlo (MC)-based SPECT reconstructions have a strong potential for correcting for image degrading factors, but the reconstruction times are long. The objective of this study was to develop a highly parallel Monte Carlo code for fast, ordered subset expectation maximum (OSEM) reconstructions of SPECT/CT images. The MC code was written in the Compute Unified Device Architecture language for a computer with four graphics processing units (GPUs) (GeForce GTX Titan X, Nvidia, USA). This enabled simulations of parallel photon emissions from the voxels matrix (1283 or 2563). Each computed tomography (CT) number was converted to attenuation coefficients for photo absorption, coherent scattering, and incoherent scattering. For photon scattering, the deflection angle was determined by the differential scattering cross sections. An angular response function was developed and used to model the accepted angles for photon interaction with the crystal, and a detector scattering kernel was used for modeling the photon scattering in the detector. Predefined energy and spatial resolution kernels for the crystal were used. The MC code was implemented in the OSEM reconstruction of clinical and phantom 177Lu SPECT/CT images. The Jaszczak image quality phantom was used to evaluate the performance of the MC reconstruction in comparison with attenuated corrected (AC) OSEM reconstructions and attenuated corrected OSEM reconstructions with resolution recovery corrections (RRC).ResultThe performance of the MC code was 3200 million photons/s. The required number of photons emitted per voxel to obtain a sufficiently low noise level in the simulated image was 200 for a 1283 voxel matrix. With this number of emitted photons/voxel, the MC-based OSEM reconstruction with ten subsets was performed within 20 s/iteration. The images converged after around six iterations. Therefore, the reconstruction time was around 3 min. The activity recovery for the spheres in the Jaszczak phantom was clearly improved with MC-based OSEM reconstruction, e.g., the activity recovery was 88% for the largest sphere, while it was 66% for AC-OSEM and 79% for RRC-OSEM.ConclusionThe GPU-based MC code generated an MC-based SPECT/CT reconstruction within a few minutes, and reconstructed patient images of 177Lu-DOTATATE treatments revealed clearly improved resolution and contrast.

Gnss Signal Processing in Gpu

ABSTRACT Signal processing of the global navigation satellite systems (GNSS) is a computationally demanding task due to the wide bandwidth of the signals and their complicated modulation schemes. The classical GNSS receivers therefore utilize tailored digital signal processors (DSP) not being flexible in nature. Fortunately, the up-to-date parallel processors or graphical processing units (GPUs) dispose sufficient computational power for processing of not only relatively narrow band GPS L1 C/A signal but also the modernized GPS, GLONASS, Galileo and COMPASS signals. The performance improvement of the modern processors is based on the constantly increasing number of cores. This trend is evident not only from the development of the central processing units (CPUs), but also from the development of GPUs that are nowadays equipped with up to several hundreds of cores optimized for video signals. GPUs include special vector instructions that support implementation of massive parallelism. The new GPUs, named as general-purpose computation on graphics processing units (GPGPU), are able to process both graphic and general data, thus making the GNSS signal processing possible. Application programming interfaces (APIs) supporting GPU parallel processing have been developed and standardized. The most general one, Open Computing Language (Open CL), is now supported by most of the GPU vendors. Next, Compute Unified Device Architecture (CUDA) language was developed for NVidia graphic cards. The CUDA language features optimized signal processing libraries including efficient implementation of the fast Fourier transform (FFT). In this paper, we study the applicability of the GPU approach in GNSS signal acquisition. Two common parallel DSP methods, parallel code space search (PCSS) and double-block zero padding (DBZP), have been investigated. Implementations in the C language for CPU and the CUDA language for GPU are discussed and compared with respect to the acquisition time. It is shown that for signals with long ranging codes (with 10230 number of chips - Galileo E5, GPS L5 etc.). Paper presented at the "European Navigation Conference 2012", held in Gdansk, Poland

GPU-acceleration for Moving Particle Semi-Implicit method

The MPS (Moving Particle Semi-implicit) method has been proven useful in computation free-surface hydrodynamic flows. Despite its applicability, one of its drawbacks in practical application is the high computational load. On the other hand, Graphics Processing Unit (GPU), which was originally developed for acceleration of computer graphics, now provides unprecedented capability for scientific computations. The main objective of this study is to develop a GPU-accelerated MPS code using CUDA (Compute Unified Device Architecture) language. Several techniques have been shown to optimize calculations in CUDA. In order to promote the acceleration by GPU, particular attentions are given to both the search of neighboring particles and the iterative solution of simultaneous linear equations in the Poisson Pressure Equation. In this paper, 2-dimensional calculations of elliptical drop evolution and dam break flow have been carried out by the GPU-accelerated MPS method, and the accuracy and performance of GPU-based code are investigated by comparing the results with those by CPU. It is shown that results of GPU-based calculations can be obtained much faster with the same reliability as the CPU-based ones.