Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Graphics processing units”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 487 records · Page 27
All sourcesNASA NTRSDOE OSTI
DOE OSTI2015

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.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2015

Fully iterative scatter corrected digital breast tomosynthesis using GPU-based fast Monte Carlo simulation and composition ratio update

Purpose: In digital breast tomosynthesis (DBT), scatter correction is highly desirable, as it improves image quality at low doses. Because the DBT detector panel is typically stationary during the source rotation, antiscatter grids are not generally compatible with DBT; thus, a software-based scatter correction is required. This work proposes a fully iterative scatter correction method that uses a novel fast Monte Carlo simulation (MCS) with a tissue-composition ratio estimation technique for DBT imaging. Methods: To apply MCS to scatter estimation, the material composition in each voxel should be known. To overcome the lack of prior accurate knowledge of tissue composition for DBT, a tissue-composition ratio is estimated based on the observation that the breast tissues are principally composed of adipose and glandular tissues. Using this approximation, the composition ratio can be estimated from the reconstructed attenuation coefficients, and the scatter distribution can then be estimated by MCS using the composition ratio. The scatter estimation and image reconstruction procedures can be performed iteratively until an acceptable accuracy is achieved. For practical use, (i) the authors have implemented a fast MCS using a graphics processing unit (GPU), (ii) the MCS is simplified to transport only x-rays in the energy range of 10–50 keV, modeling Rayleigh and Compton scattering and the photoelectric effect using the tissue-composition ratio of adipose and glandular tissues, and (iii) downsampling is used because the scatter distribution varies rather smoothly. Results: The authors have demonstrated that the proposed method can accurately estimate the scatter distribution, and that the contrast-to-noise ratio of the final reconstructed image is significantly improved. The authors validated the performance of the MCS by changing the tissue thickness, composition ratio, and x-ray energy. The authors confirmed that the tissue-composition ratio estimation was quite accurate under a variety of conditions. Our GPU-based fast MCS implementation took approximately 3 s to generate each angular projection for a 6 cm thick breast, which is believed to make this process acceptable for clinical applications. In addition, the clinical preferences of three radiologists were evaluated; the preference for the proposed method compared to the preference for the convolution-based method was statistically meaningful (p < 0.05, McNemar test). Conclusions: The proposed fully iterative scatter correction method and the GPU-based fast MCS using tissue-composition ratio estimation successfully improved the image quality within a reasonable computational time, which may potentially increase the clinical utility of DBT.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2015

Analytic reconstruction algorithms for triple-source CT with horizontal data truncation

Purpose: This paper explores a triple-source imaging method with horizontal data truncation to enlarge the field of view (FOV) for big objects. Methods: The study is conducted by using theoretical analysis, mathematical deduction, and numerical simulations. The proposed algorithms are implemented in c + + and MATLAB. While the basic platform is constructed in MATLAB, the computationally intensive segments are coded in c + +, which are linked via a MEX interface. Results: A triple-source circular scanning configuration with horizontal data truncation is developed, where three pairs of x-ray sources and detectors are unevenly distributed on the same circle to cover the whole imaging object. For this triple-source configuration, a fan-beam filtered backprojection-type algorithm is derived for truncated full-scan projections without data rebinning. The algorithm is also extended for horizontally truncated half-scan projections and cone-beam projections in a Feldkamp-type framework. Using their method, the FOV is enlarged twofold to threefold to scan bigger objects with high speed and quality. The numerical simulation results confirm the correctness and effectiveness of the developed algorithms. Conclusions: The triple-source scanning configuration with horizontal data truncation cannot only keep most of the advantages of a traditional multisource system but also cover a larger FOV for big imaging objects. In addition, because the filtering is shift-invariant, the proposed algorithms are very fast and easily parallelized on graphic processing units.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

Accelerated fast iterative shrinkage thresholding algorithms for sparsity-regularized cone-beam CT image reconstruction

Purpose: The development of iterative image reconstruction algorithms for cone-beam computed tomography (CBCT) remains an active and important research area. Even with hardware acceleration, the overwhelming majority of the available 3D iterative algorithms that implement nonsmooth regularizers remain computationally burdensome and have not been translated for routine use in time-sensitive applications such as image-guided radiation therapy (IGRT). In this work, two variants of the fast iterative shrinkage thresholding algorithm (FISTA) are proposed and investigated for accelerated iterative image reconstruction in CBCT. Methods: Algorithm acceleration was achieved by replacing the original gradient-descent step in the FISTAs by a subproblem that is solved by use of the ordered subset simultaneous algebraic reconstruction technique (OS-SART). Due to the preconditioning matrix adopted in the OS-SART method, two new weighted proximal problems were introduced and corresponding fast gradient projection-type algorithms were developed for solving them. We also provided efficient numerical implementations of the proposed algorithms that exploit the massive data parallelism of multiple graphics processing units. Results: The improved rates of convergence of the proposed algorithms were quantified in computer-simulation studies and by use of clinical projection data corresponding to an IGRT study. The accelerated FISTAs were shown to possess dramatically improved convergence properties as compared to the standard FISTAs. For example, the number of iterations to achieve a specified reconstruction error could be reduced by an order of magnitude. Volumetric images reconstructed from clinical data were produced in under 4 min. Conclusions: The FISTA achieves a quadratic convergence rate and can therefore potentially reduce the number of iterations required to produce an image of a specified image quality as compared to first-order methods. We have proposed and investigated accelerated FISTAs for use with two nonsmooth penalty functions that will lead to further reductions in image reconstruction times while preserving image quality. Moreover, with the help of a mixed sparsity-regularization, better preservation of soft-tissue structures can be potentially obtained. The algorithms were systematically evaluated by use of computer-simulated and clinical data sets.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

SU-D-206-02: Evaluation of Partial Storage of the System Matrix for Cone Beam Computed Tomography Using a GPU Platform

Purpose: Iterative reconstruction algorithms in computed tomography (CT) require a fast method for computing the intersections between the photons’ trajectories and the object, also called ray-tracing or system matrix computation. This work evaluates different ways to store the system matrix, aiming to reconstruct dense image grids in reasonable time. Methods: We propose an optimized implementation of the Siddon’s algorithm using graphics processing units (GPUs) with a novel data storage scheme. The algorithm computes a part of the system matrix on demand, typically, for one projection angle. The proposed method was enhanced with accelerating options: storage of larger subsets of the system matrix, systematic reuse of data via geometric symmetries, an arithmetic-rich parallel code and code configuration via machine learning. It was tested on geometries mimicking a cone beam CT acquisition of a human head. To realistically assess the execution time, the ray-tracing routines were integrated into a regularized Poisson-based reconstruction algorithm. The proposed scheme was also compared to a different approach, where the system matrix is fully pre-computed and loaded at reconstruction time. Results: Fast ray-tracing of realistic acquisition geometries, which often lack spatial symmetry properties, was enabled via the proposed method. Ray-tracing interleaved with projection and backprojection operations required significant additional time. In most cases, ray-tracing was shown to use about 66 % of the total reconstruction time. In absolute terms, tracing times varied from 3.6 s to 7.5 min, depending on the problem size. The presence of geometrical symmetries allowed for non-negligible ray-tracing and reconstruction time reduction. Arithmetic-rich parallel code and machine learning permitted a modest reconstruction time reduction, in the order of 1 %. Conclusion: Partial system matrix storage permitted the reconstruction of higher 3D image grid sizes and larger projection datasets at the cost of additional time, when compared to the fully pre-computed approach. This work was supported in part by the Fonds de recherche du Quebec - Nature et technologies (FRQ-NT). The authors acknowledge partial support by the CREATE Medical Physics Research Training Network grant of the Natural Sciences and Engineering Research Council of Canada (Grant No. 432290).

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

SU-F-T-209: Multicriteria Optimization Algorithm for Intensity Modulated Radiation Therapy Using Pencil Proton Beam Scanning

Purpose: To provide a multicriteria optimization algorithm for intensity modulated radiation therapy using pencil proton beam scanning. Methods: Intensity modulated radiation therapy using pencil proton beam scanning requires efficient optimization algorithms to overcome the uncertainties in the Bragg peaks locations. This work is focused on optimization algorithms that are based on Monte Carlo simulation of the treatment planning and use the weights and the dose volume histogram (DVH) control points to steer toward desired plans. The proton beam treatment planning process based on single objective optimization (representing a weighted sum of multiple objectives) usually leads to time-consuming iterations involving treatment planning team members. We proved a time efficient multicriteria optimization algorithm that is developed to run on NVIDIA GPU (Graphical Processing Units) cluster. The multicriteria optimization algorithm running time benefits from up-sampling of the CT voxel size of the calculations without loss of fidelity. Results: We will present preliminary results of Multicriteria optimization for intensity modulated proton therapy based on DVH control points. The results will show optimization results of a phantom case and a brain tumor case. Conclusion: The multicriteria optimization of the intensity modulated radiation therapy using pencil proton beam scanning provides a novel tool for treatment planning. Work support by a grant from Varian Inc.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

TU-AB-BRC-09: Fast Dose-Averaged LET and Biological Dose Calculations for Proton Therapy Using Graphics Cards

Purpose: To demonstrate fast and accurate Monte Carlo (MC) calculations of proton dose-averaged linear energy transfer (LETd) and biological dose (BD) on a Graphics Processing Unit (GPU) card. Methods: A previously validated GPU-based MC simulation of proton transport was used to rapidly generate LETd distributions for proton treatment plans. Since this MC handles proton-nuclei interactions on an event-by-event using a Bertini intranuclear cascade-evaporation model, secondary protons were taken into account. The smaller contributions of secondary neutrons and recoil nuclei were ignored. Recent work has shown that LETd values are sensitive to the scoring method. The GPU-based LETd calculations were verified by comparing with a TOPAS custom scorer that uses tabulated stopping powers, following recommendations by other authors. Comparisons were made for prostate and head-and-neck patients. A python script is used to convert the MC-generated LETd distributions to BD using a variety of published linear quadratic models, and to export the BD in DICOM format for subsequent evaluation. Results: Very good agreement is obtained between TOPAS and our GPU MC. Given a complex head-and-neck plan with 1 mm voxel spacing, the physical dose, LETd and BD calculations for 10{sup 8} proton histories can be completed in ∼5 minutes using a NVIDIA Titan X card. The rapid turnover means that MC feedback can be obtained on dosimetric plan accuracy as well as BD hotspot locations, particularly in regards to their proximity to critical structures. In our institution the GPU MC-generated dose, LETd and BD maps are used to assess plan quality for all patients undergoing treatment. Conclusion: Fast and accurate MC-based LETd calculations can be performed on the GPU. The resulting BD maps provide valuable feedback during treatment plan review. Partially funded by Varian Medical Systems.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

TU-FG-BRB-07: GPU-Based Prompt Gamma Ray Imaging From Boron Neutron Capture Therapy

Purpose: 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. Methods: 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. Results: 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). Conclusion: 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 reconstruction using the GPU computation for BNCT simulations.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

Technical Note: SCUDA: A software platform for cumulative dose assessment

Purpose: Accurate tracking of anatomical changes and computation of actually delivered dose to the patient are critical for successful adaptive radiation therapy (ART). Additionally, efficient data management and fast processing are practically important for the adoption in clinic as ART involves a large amount of image and treatment data. The purpose of this study was to develop an accurate and efficient Software platform for CUmulative Dose Assessment (SCUDA) that can be seamlessly integrated into the clinical workflow. Methods: SCUDA consists of deformable image registration (DIR), segmentation, dose computation modules, and a graphical user interface. It is connected to our image PACS and radiotherapy informatics databases from which it automatically queries/retrieves patient images, radiotherapy plan, beam data, and daily treatment information, thus providing an efficient and unified workflow. For accurate registration of the planning CT and daily CBCTs, the authors iteratively correct CBCT intensities by matching local intensity histograms during the DIR process. Contours of the target tumor and critical structures are then propagated from the planning CT to daily CBCTs using the computed deformations. The actual delivered daily dose is computed using the registered CT and patient setup information by a superposition/convolution algorithm, and accumulated using the computed deformation fields. Both DIR and dose computation modules are accelerated by a graphics processing unit. Results: The cumulative dose computation process has been validated on 30 head and neck (HN) cancer cases, showing 3.5 ± 5.0 Gy (mean±STD) absolute mean dose differences between the planned and the actually delivered doses in the parotid glands. On average, DIR, dose computation, and segmentation take 20 s/fraction and 17 min for a 35-fraction treatment including additional computation for dose accumulation. Conclusions: The authors developed a unified software platform that provides accurate and efficient monitoring of anatomical changes and computation of actually delivered dose to the patient, thus realizing an efficient cumulative dose computation workflow. Evaluation on HN cases demonstrated the utility of our platform for monitoring the treatment quality and detecting significant dosimetric variations that are keys to successful ART.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2019

Simulation of Magnetic Systems in the Domain with a Corner

Numerical simulation of magnetic systems having corner points is considered. The literature on this topic is reviewed. A method of encapsulating the function behavior near corner points in the difference scheme is proposed for numerical solution of the 2D magnetostatic problem in the iron domain. For the 3D magnetostatic problem in the iron/vacuum domain, the increase in the magnetic field near the corner point is estimated, and a method for construction of an adaptive mesh is proposed. A parallel algorithm is implemented on the graphics-processing unit (GPU) architecture for faster search for the numerical solution of the magnetostatic problem. Numerical simulation is performed for the magnetic system of the SPD detector to be used at the NICA complex (JINR, Dubna).

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2025

The critical importance of software for HEP

Particle physics has an ambitious and broad global experimental programme for the coming decades. Large investments in building new facilities are already underway or under consideration. Scaling the present processing power and data storage needs by the foreseen increase in data rates in the next decade for HL-LHC is not sustainable within the current budgets. As a result, a more efficient usage of computing resources is required in order to realise the physics potential of future experiments. Software and computing are an integral part of experimental design, trigger and data acquisition, simulation, reconstruction, and analysis, as well as related theoretical predictions. A significant investment in computing and software is therefore critical. Advances in software and computing, including artificial intelligence (AI) and machine learning (ML), will be key for solving these challenges. Making better use of new processing hardware such as graphical processing units (GPUs) or ARM chips is a growing trend. This forms part of a computing solution that makes efficient use of facilities and contributes to the reduction of the environmental footprint of HEP computing. The HEP community already provided a roadmap for software and computing for the last EPPSU, and this paper updates that, with a focus on the most resource critical parts of our data processing chain.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Full event interpretation with machine-learning-based particle-flow reconstruction in the CMS detector

The particle-flow (PF) algorithm constructs a global description of each particle collision by producing a comprehensive list of final-state particles, and is central to event reconstruction in the CMS experiment at the CERN LHC. The existing PF implementation relies on physics-motivated heuristics and assumptions that can be replaced by machine-learning (ML) models trained directly on simulated data and naturally suited to modern graphics processing units (GPUs). A state-of-the-art ML-based PF (MLPF) reconstruction algorithm, implemented within the CMS software framework, is presented. The MLPF algorithm performs a learnable full-event reconstruction on GPUs, generalizes across detector conditions and collision energies, and replaces multiple modular reconstruction steps with a single unified model. Physics performance comparable to standard PF reconstruction is achieved in both simulation and data, with improved jet energy resolution and inference time. In simulated top quark-antiquark events under LHC Run-3 (2023-2024) conditions, the jet energy resolution improves by 10-20% for jets with transverse momentum between 30-100 GeV. Inference time is evaluated using simulated multijet events, with a median of $20\,\hbox {ms}$ per event on an Nvidia L4 GPU, compared to approximately $110\,\hbox {ms}$ for the standard CMS PF reconstruction.

Hayrapetyan, Aram [Yerevan Phys. Inst.]↗
DOE OSTI2013

Efficient Scheduling of Recursive Control Flow on GPUs

Graphics processing units (GPUs) have rapidly emerged as a very significant player in high performance computing. Single instruction multiple thread (SIMT) pipelines are typically used in GPUs to exploit parallelism and maximize performance. Although support for unstructured control flow has been included in GPUs, efficiently managing thread divergence for arbitrary parallel programs remains a critical challenge. In this paper, we focus on the problem of supporting recursion in modern GPUs. We design and comparatively evaluate various algorithms to manage thread divergence encountered in recursive programs. The results improve upon traditional post-dominator based reconvergence mechanisms designed to handle thread divergence due to control flow within a procedure.

GPUs↗
DOE OSTI2013

Radiative signatures of the relativistic Kelvin-Helmholtz instability

We present a particle-in-cell simulation of the relativistic Kelvin-Helmholtz Instability (KHI) that for the first time delivers angularly resolved radiation spectra of the particle dynamics during the formation of the KHI. This enables studying the formation of the KHI with unprecedented spatial, angular and spectral resolution. Our results are of great importance for understanding astrophysical jet formation and comparable plasma phenomena by relating the particle motion observed in the KHI to its radiation signature. The innovative methods presented here on the implementation of the particle-in-cell algorithm on graphic processing units can be directly adapted to any many-core parallelization of the particle-mesh method. With these methods we see a peak performance of 7.176 PFLOP/s (double-precision) plus 1.449 PFLOP/s (single-precision), an efficiency of 96% when weakly scaling from 1 to 18432 nodes, an efficiency of 68.92% and a speed up of 794 (ideal: 1152) when strongly scaling from 16 to 18432 nodes.

42 ENGINEERING↗
DOE OSTI2016

Understanding GPU Power. A Survey of Profiling, Modeling, and Simulation Methods

Modern graphics processing units (GPUs) have complex architectures that admit exceptional performance and energy efficiency for high throughput applications.Though GPUs consume large amounts of power, their use for high throughput applications facilitate state-of-the-art energy efficiency and performance. Consequently, continued development relies on understanding their power consumption. Our work is a survey of GPU power modeling and profiling methods with increased detail on noteworthy efforts. Moreover, as direct measurement of GPU power is necessary for model evaluation and parameter initiation, internal and external power sensors are discussed. Hardware counters, which are low-level tallies of hardware events, share strong correlation to power use and performance. Statistical correlation between power and performance counters has yielded worthwhile GPU power models, yet the complexity inherent to GPU architectures presents new hurdles for power modeling. Developments and challenges of counter-based GPU power modeling is discussed. Often building on the counter-based models, research efforts for GPU power simulation, which make power predictions from input code and hardware knowledge, provide opportunities for optimization in programming or architectural design. Noteworthy strides in power simulations for GPUs are included along with their performance or functional simulator counterparts when appropriate. Lastly, possible directions for future research are discussed.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2017

Efficient Simulation of Nested Hollow Sphere Intersections: for Dynamically Nested Compartmental Models in Cell Biology

In the particle-based simulation of cell-biological systems in continuous space, a key performance bottleneck is the computation of all possible intersections between particles. These typically rely for collision detection on solid sphere approaches. The behavior of cell biological systems is influenced by dynamic hierarchical nesting, such as the forming of, the transport within, and the merging of vesicles. Existing collision detection algorithms are found not to be designed for these types of spatial cell-biological models, because nearly all existing high performance parallel algorithms are focusing on solid sphere interactions. The known algorithms for solid sphere intersections return more intersections than actually occur with nested hollow spheres. Here we define a new problem of computing the intersections among arbitrarily nested hollow spheres of possibly different sizes, thicknesses, positions, and nesting levels. We describe a new algorithm designed to solve this nested hollow sphere intersection problem and implement it for parallel execution on graphical processing units (GPUs). We present first results about the runtime performance and scaling to hundreds of thousands of spheres, and compare the performance with that from a leading solid object intersection package also running on GPUs.

Koester, Till↗
DOE OSTI2018

A Study of Complex Deep Learning Networks on High-Performance, Neuromorphic, and Quantum Computers

Current deep learning approaches have been very successful using convolutional neural networks trained on large graphical-processing-unit-based computers. Three limitations of this approach are that (1) they are based on a simple layered network topology, i.e., highly connected layers, without intra-layer connections; (2) the networks are manually configured to achieve optimal results, and (3) the implementation of the network model is expensive in both cost and power. In this paper, we evaluate deep learning models using three different computing architectures to address these problems: quantum computing to train complex topologies, high performance computing to automatically determine network topology, and neuromorphic computing for a low-power hardware implementation. We use the MNIST dataset for our experiment, due to input size limitations of current quantum computers. Our results show the feasibility of using the three architectures in tandem to address the above deep learning limitations. Finally, we show that a quantum computer can find high quality values of intra-layer connection weights in a tractable time as the complexity of the network increases, a high performance computer can find optimal layer-based topologies, and a neuromorphic computer can represent the complex topology and weights derived from the other architectures in low power memristive hardware.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2018

Juggler: a dependence-aware task-based execution framework for GPUs

Scientific applications with single instruction, multiple data (SIMD) computations show considerable performance improvements when run on today's graphics processing units (GPUs). However, the existence of data dependences across thread blocks may significantly impact the speedup by requiring global synchronization across multiprocessors (SMs) inside the GPU. To efficiently run applications with interblock data dependences, we need fine-granular task-based execution models that will treat SMs inside a GPU as stand-alone parallel processing units. Such a scheme will enable faster execution by utilizing all internal computation elements inside the GPU and eliminating unnecessary waits during device-wide global barriers.In this paper, we propose Juggler, a task-based execution scheme for GPU workloads with data dependences. The Juggler framework takes applications embedding OpenMP 4.5 tasks as input and executes them on the GPU via an efficient in-device runtime, hence eliminating the need for kernel-wide global synchronization. Juggler requires no or little modification to the source code, and once launched, the runtime entirely runs on the GPU without relying on the host through the entire execution. We have evaluated Juggler on an NVIDIA Tesla P100 GPU and obtained up to 31% performance improvement against global barrier based implementation, with minimal runtime overhead.

Belviranli, Mehmet E.↗