Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “memory efficiency”

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 811 records · Page 45
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Singleton Sieving: Overcoming the Memory/Speed Trade-Off in Exascale k-mer Analysis

Traditional filter data structures, such as Bloom filters, do not offer necessary features that modern high-performance data analytics applications need in order to efficiently perform complex data analysis tasks. For example, MetaHipMer, a de novo metagenome assembler, can use filters to weed out singleton k-mers and reduce memory usage by 30%-70%. However, the filter needs the ability to associate values with k-mers in order to perform the analysis in a single communication pass. Bloom filters do not support value associations and cause the application to perform an extra communication pass, thereby increasing the run time. Therefore, MetaHipMer faces a trade off between memory and speed due to the limited capabilities of traditional filters. In this paper, we overcome the memory and speed trade off in MetaHipMer by integrating a GPU-based feature-rich filter, the Two-Choice filter (TCF), in the MetaHipMer pipeline. The TCF uses key-value association to approximately store k-mers with extensions. This allows MetaHipMer to perform k-mer analysis on the GPUs in a single communication pass. Our empirical analysis shows a 50% reduction in memory usage in k-mer analysis on each node in MetaHipMer without any effect on the overall run time or assembly quality. The memory reduction in turn results in a 43% reduction in the number of nodes required to assemble datasets and enables MetaHipMer to scale to much larger datasets.

McCoy, Hunter↗
NASA NTRS1982

On nonlinear finite element analysis in single-, multi- and parallel-processors

Numerical solution of nonlinear equilibrium problems of structures by means of Newton-Raphson type iterations is reviewed. Each step of the iteration is shown to correspond to the solution of a linear problem, therefore the feasibility of the finite element method for nonlinear analysis is established. Organization and flow of data for various types of digital computers, such as single-processor/single-level memory, single-processor/two-level-memory, vector-processor/two-level-memory, and parallel-processors, with and without sub-structuring (i.e. partitioning) are given. The effect of the relative costs of computation, memory and data transfer on substructuring is shown. The idea of assigning comparable size substructures to parallel processors is exploited. Under Cholesky type factorization schemes, the efficiency of parallel processing is shown to decrease due to the occasional shared data, just as that due to the shared facilities.

Utku, S.↗
NASA NTRS1992

Content addressable memory project

A parameterized version of the tree processor was designed and tested (by simulation). The leaf processor design is 90 percent complete. We expect to complete and test a combination of tree and leaf cell designs in the next period. Work is proceeding on algorithms for the computer aided manufacturing (CAM), and once the design is complete we will begin simulating algorithms for large problems. The following topics are covered: (1) the practical implementation of content addressable memory; (2) design of a LEAF cell for the Rutgers CAM architecture; (3) a circuit design tool user's manual; and (4) design and analysis of efficient hierarchical interconnection networks.

Hall, J. Storrs↗
NASA NTRS1994

Master-slave system with force feedback based on dynamics of virtual model

A master-slave system can extend manipulating and sensing capabilities of a human operator to a remote environment. But the master-slave system has two serious problems: one is the mechanically large impedance of the system; the other is the mechanical complexity of the slave for complex remote tasks. These two problems reduce the efficiency of the system. If the slave has local intelligence, it can help the human operator by using its good points like fast calculation and large memory. The authors suggest that the slave is a dextrous hand with many degrees of freedom able to manipulate an object of known shape. It is further suggested that the dimensions of the remote work space be shared by the human operator and the slave. The effect of the large impedance of the system can be reduced in a virtual model, a physical model constructed in a computer with physical parameters as if it were in the real world. A method to determine the damping parameter dynamically for the virtual model is proposed. Experimental results show that this virtual model is better than the virtual model with fixed damping.

Nojima, Shuji↗
NASA NTRS2011

NASA Tech Briefs, July 2011

Topics covered include: 1) Collaborative Clustering for Sensor Networks; 2) Teleoperated Marsupial Mobile Sensor Platform Pair for Telepresence Insertion Into Challenging Structures; 3) Automated Verification of Spatial Resolution in Remotely Sensed Imagery; 4) Electrical Connector Mechanical Seating Sensor; 5) In Situ Aerosol Detector; 6) Multi-Parameter Aerosol Scattering Sensor; 7) MOSFET Switching Circuit Protects Shape Memory Alloy Actuators; 8) Optimized FPGA Implementation of Multi-Rate FIR Filters Through Thread Decomposition; 9) Circuit for Communication Over Power Lines; 10) High-Efficiency Ka-Band Waveguide Two-Way Asymmetric Power Combiner; 11) 10-100 Gbps Offload NIC for WAN, NLR, and Grid Computing; 12) Pulsed Laser System to Simulate Effects of Cosmic Rays in Semiconductor Devices; 13) Flight Planning in the Cloud; 14) MPS Editor; 15) Object-Oriented Multi Disciplinary Design, Analysis, and Optimization Tool; 16) Cryogenic-Compatible Winchester Connector Mount and Retaining System for Composite Tubes; 17) Development of Position-Sensitive Magnetic Calorimeters for X-Ray Astronomy; 18) Planar Rotary Piezoelectric Motor Using Ultrasonic Horns; 19) Self-Rupturing Hermetic Valve; 20) Explosive Bolt Dual-Initiated from One Side; 21) Dampers for Stationary Labyrinth Seals; 22) Two-Arm Flexible Thermal Strap; 23) Carbon Dioxide Removal via Passive Thermal Approaches; 24) Polymer Electrolyte-Based Ambient Temperature Oxygen Microsensors for Environmental Monitoring; 25) Pressure Shell Approach to Integrated Environmental Protection; 26) Image Quality Indicator for Infrared Inspections; 27) Micro-Slit Collimators for X-Ray/Gamma-Ray Imaging; 28) Scatterometer-Calibrated Stability Verification Method; 29) Test Port for Fiber-Optic-Coupled Laser Altimeter; 30) Phase Retrieval System for Assessing Diamond Turning and Optical Surface Defects; 31) Laser Oscillator Incorporating a Wedged Polarization Rotator and a Porro Prism as Cavity Mirror; 32) Generic, Extensible, Configurable Push-Pull Framework for Large-Scale Science Missions; 33) Dynamic Loads Generation for Multi-Point Vibration Excitation Problems; 34) Optimal Control via Self-Generated Stochasticity; 35) Space-Time Localization of Plasma Turbulence Using Multiple Spacecraft Radio Links; 36) Surface Contact Model for Comets and Asteroids; 37) Dust Mitigation Vehicle; 38) Optical Coating Performance for Heat Reflectors of the JWST-ISIM Electronic Component; 39) SpaceCube Demonstration Platform; 40) Aperture Mask for Unambiguous Parity Determination in Long Wavelength Imagers; 41) Spaceflight Ka-Band High-Rate Radiation-Hard Modulator; 42) Enabling Disabled Persons to Gain Access to Digital Media; 43) Cytometer on a Chip; 44) Principles, Techniques, and Applications of Tissue Microfluidics; and 45) Two-Stage Winch for Kites and Tethered Balloons or Blimps.

Source record↗
NASA NTRS2014

Statistical Symbolic Execution with Informed Sampling

Symbolic execution techniques have been proposed recently for the probabilistic analysis of programs. These techniques seek to quantify the likelihood of reaching program events of interest, e.g., assert violations. They have many promising applications but have scalability issues due to high computational demand. To address this challenge, we propose a statistical symbolic execution technique that performs Monte Carlo sampling of the symbolic program paths and uses the obtained information for Bayesian estimation and hypothesis testing with respect to the probability of reaching the target events. To speed up the convergence of the statistical analysis, we propose Informed Sampling, an iterative symbolic execution that first explores the paths that have high statistical significance, prunes them from the state space and guides the execution towards less likely paths. The technique combines Bayesian estimation with a partial exact analysis for the pruned paths leading to provably improved convergence of the statistical analysis. We have implemented statistical symbolic execution with in- formed sampling in the Symbolic PathFinder tool. We show experimentally that the informed sampling obtains more precise results and converges faster than a purely statistical analysis and may also be more efficient than an exact symbolic analysis. When the latter does not terminate symbolic execution with informed sampling can give meaningful results under the same time and memory limits.

Reliability↗
NASA NTRS1988

Data traffic reduction schemes for sparse Cholesky factorizations

Load distribution schemes are presented which minimize the total data traffic in the Cholesky factorization of dense and sparse, symmetric, positive definite matrices on multiprocessor systems with local and shared memory. The total data traffic in factoring an n x n sparse, symmetric, positive definite matrix representing an n-vertex regular 2-D grid graph using n (sup alpha), alpha is equal to or less than 1, processors are shown to be O(n(sup 1 + alpha/2)). It is O(n(sup 3/2)), when n (sup alpha), alpha is equal to or greater than 1, processors are used. Under the conditions of uniform load distribution, these results are shown to be asymptotically optimal. The schemes allow efficient use of up to O(n) processors before the total data traffic reaches the maximum value of O(n(sup 3/2)). The partitioning employed within the scheme, allows a better utilization of the data accessed from shared memory than those of previously published methods.

Naik, Vijay K.↗
DOE OSTI2025

Teacher-student training improves the accuracy and efficiency of machine learning interatomic potentials

Machine learning interatomic potentials (MLIPs) are revolutionizing the field of molecular dynamics (MD) simulations. Recent MLIPs have tended towards more complex architectures trained on larger datasets. The resulting increase in computational and memory costs may prohibit the application of these MLIPs to perform large-scale MD simulations. Herein, we present a teacher-student training framework in which the latent knowledge from the teacher (atomic energies) is used to augment the students' training. We show that the light-weight student MLIPs have faster MD speeds at a fraction of the memory footprint compared to the teacher models. Remarkably, the student models can even surpass the accuracy of the teachers, even though both are trained on the same quantum chemistry dataset. Our work highlights a practical method for MLIPs to reduce the resources required for large-scale MD simulations.

36 MATERIALS SCIENCE↗
NASA NTRS1988

Efficient use of bit planes in the generation of motion stimuli

The production of animated motion sequences on computer-controlled display systems presents a technical problem because large images cannot be transferred from disk storage to image memory at conventional frame rates. A technique is described in which a single base image can be used to generate a broad class of motion stimuli without the need for such memory transfers. This technique was applied to the generation of drifting sine-wave gratings (and by extension, sine wave plaids). For each drifting grating, sine and cosine spatial phase components are first reduced to 1 bit/pixel using a digital halftoning technique. The resulting pairs of 1-bit images are then loaded into pairs of bit planes of the display memory. To animate the patterns, the display hardware's color lookup table is modified on a frame-by-frame basis; for each frame the lookup table is set to display a weighted sum of the spatial sine and cosine phase components. Because the contrasts and temporal frequencies of the various components are mutually independent in each frame, the sine and cosine components can be counterphase modulated in temporal quadrature, yielding a single drifting grating. Using additional bit planes, multiple drifting gratings can be combined to form sine-wave plaid patterns. A large number of resultant plaid motions can be produced from a single image file because the temporal frequencies of all the components can be varied independently. For a graphics device having 8 bits/pixel, up to four drifting gratings may be combined, each having independently variable contrast and speed.

Mulligan, Jeffrey B.↗
DOE OSTI2022

ASAP: Automatic Synthesis of Area-Efficient and Precision-Aware CGRAs

Coarse-grained reconfigurable accelerators (CGRAs) are a promising accelerator design choice that strikes a balance between performance and adaptability to different computing patterns across various applications domains. Designing a CGRA for a specific application domain involves enormous software/hardware engineering effort. Recent research works explore loop transformations, functional unit types, network topology, and memory size to identify optimal CGRA designs given a set of kernels from a specific application do- main. Unfortunately, the impact of functional units with different precision support has rarely been investigated. To address this gap, we propose ASAP – a hardware/software co-design framework that automatically identifies and synthesizes optimal precision-aware CGRA for a set of applications of interest. Our evaluation shows that ASAP generates specialized designs 3.2×, 4.21×, and 5.8× more efficient (in terms of performance per unit of energy or area) than non-specialized homogeneous CGRAs, for the scientific computing, embedded, and edge machine learning domains, respectively, with limited accuracy loss. Moreover, ASAP provides more efficient designs than other state-of-the-art synthesis frameworks for specialized CGRAs.

artificial intelligence↗
NASA NTRS2002

Advanced Engine Cycles Analyzed for Turbofans With Variable-Area Fan Nozzles Actuated by a Shape Memory Alloy

Advanced, large commercial turbofan engines using low-fan-pressure-ratio, very high bypass ratio thermodynamic cycles can offer significant fuel savings over engines currently in operation. Several technological challenges must be addressed, however, before these engines can be designed. To name a few, the high-diameter fans associated with these engines pose a significant packaging and aircraft installation challenge, and a large, heavy gearbox is often necessary to address the differences in ideal operating speeds between the fan and the low-pressure turbine. Also, the large nacelles contribute aerodynamic drag penalties and require long, heavy landing gear when mounted on conventional, low wing aircraft. Nevertheless, the reduced fuel consumption rates of these engines are a compelling economic incentive, and fans designed with low pressure ratios and low tip speeds offer attractive noise-reduction benefits. Another complication associated with low-pressure-ratio fans is their need for variable flow-path geometry. As the design fan pressure ratio is reduced below about 1.4, an operational disparity is set up in the fan between high and low flight speeds. In other words, between takeoff and cruise there is too large a swing in several key fan parameters-- such as speed, flow, and pressure--for a fan to accommodate. One solution to this problem is to make use of a variable-area fan nozzle (VAFN). However, conventional, hydraulically actuated variable nozzles have weight, cost, maintenance, and reliability issues that discourage their use with low-fan-pressure-ratio engine cycles. United Technologies Research, in cooperation with NASA, is developing a revolutionary, lightweight, and reliable shape memory alloy actuator system that can change the on-demand nozzle exit area by up to 20 percent. This "smart material" actuation technology, being studied under NASA's Ultra-Efficient Engine Technology (UEET) Program and Revolutionary Concepts in Aeronautics (RevCon) Program, has the potential to enable the next generation of efficient, quiet, very high bypass ratio turbofans. NASA Glenn Research Center's Propulsion Systems Analysis Office, along with NASA Langley Research Center's Systems Analysis Branch, conducted an independent analytical assessment of this new technology to provide strategic guidance to UEET and RevCon. A 2010-technology-level high-spool engine core was designed for this evaluation. Two families of low-spool components, one with and one without VAFN's, were designed to operate with the core. This "constant core" approach was used to hold most design parameters constant so that any performance differences between the VAFN and fixed nozzle cycles could be attributed to the VAFN technology alone. In this manner, the cycle design regimes that offer a performance payoff when VAFN's are used could be identified. The NASA analytical model of a performance-optimized VAFN turbofan with a fan pressure ratio of 1.28 is shown. Mission analyses of the engines were conducted using the notional, long-haul, advanced commercial twinjet shown. A high wing design was used to accommodate the large high-bypassratio engines. The mission fuel reduction benefit of very high bypass shape-memory-alloy VAFN aircraft was calculated to be 8.3 percent lower than a moderate bypass cycle using a conventional fixed nozzle. Shape-memory-alloy VAFN technology is currently under development in NASA's UEET and RevCon Programs.

Berton, Jeffrey J.↗
NASA NTRS1999

Multiplexed Holographic Data Storage in Bacteriorhodopsin

Biochrome photosensitive films in particular Bacteriorhodopsin exhibit features which make these materials an attractive recording medium for optical data storage and processing. Bacteriorhodopsin films find numerous applications in a wide range of optical data processing applications; however the short-term memory characteristics of BR limits their applications for holographic data storage. The life-time of the BR can be extended using cryogenic temperatures [1], although this method makes the system overly complicated and unstable. Longer life-times can be provided in one modification of BR - the "blue" membrane BR [2], however currently available films are characterized by both low diffraction efficiency and difficulties in providing photoreversible recording. In addition, as a dynamic recording material, the BR requires different wavelengths for recording and reconstructing of optical data in order to prevent the information erasure during its readout. This fact also put constraints on a BR-based Optical Memory, due to information loss in holographic memory systems employing the two-lambda technique for reading-writing thick multiplexed holograms.

Mehrl, David J.↗
DOE OSTI2021

Effective field theory of stochastic diffusion from gravity

Planar black holes in AdS have long-lived quasinormal modes which capture the physics of charge and momentum diffusion in the dual field theory. How should we characterize the effective dynamics of a probe system coupled to the conserved currents of the dual field theory? Specifically, how would such a probe record the long-lived memory of the black hole and its Hawking fluctuations? We address this question by exhibiting a universal gauge invariant framework which captures the physics of stochastic diffusion in holography: a designer scalar with a gravitational coupling governed by a single parameter, the Markovianity index. We argue that the physics of gauge and gravitational perturbations of a planar Schwarzschild-AdS black hole can be efficiently captured by such designer scalars. We demonstrate that this framework allows one to decouple, at the quadratic order, the long-lived quasinormal and Hawking modes from the short-lived ones. It furthermore provides a template for analyzing fluctuating open quantum field theories with memory. In particular, we use this set-up to analyze the diffusive Hawking photons and gravitons about a planar Schwarzschild-AdS black hole and derive the quadratic effective action that governs fluctuating hydrodynamics of the dual CFT. Along the way we also derive results relevant for probes of hyperscaling violating backgrounds at finite temperature.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Performance Improvements of the Griffin Solvers in FY24

The Griffin code is a MOOSE-based reactor physics application jointly developed by Idaho National Laboratory and Argonne National Laboratory under the Department of Energy Office of Nuclear Energy Nuclear Energy Advanced Modeling and Simulation Program. This fiscal year, we have made significant efforts to improve the performance of transport solver options and cross-section generation for the efficient use of Griffin in advanced reactor applications. For the HFEM-PN solver, the residual evaluations of HFEM kernels were optimized by utilizing the pre- computed averaged cross sections for individual elements. Numerical integration involving the evaluation of basis functions at quadrature points was bypassed by facilitating precomputed element mass matrices for response matrices. Red-black iterations were improved by introducing a new generalized minimum residual based solver. The memory usage of response matrix storage was significantly reduced by applying basis function rotations on interfaces and calculating volumetric odd-parity moments on the fly. Additionally, the adjoint flux and transient calculation capabilities of the HFEM-PN solver were successfully implemented and verified using the TWIGL benchmark problem. For the DFEM-SN solver, memory footprint and computation time were significantly reduced by not treating angular flux vectors as the MOOSE nonlinear system vectors. Specifically for IQS, scalar adjoint weighting was introduced to further eliminate angular adjoint flux storage in the MOOSE auxiliary system. It was demonstrated through the three-dimensional Advanced Burner Test Reactor core problem that the memory usage for transient calculations with the IQS method was reduced by over 7.5× compared to before the optimizations. For the self-shielding application programming interface, a new double-heterogeneity treatment method, named the Bell Function-Based Analytic Two-Region Slowing Down Method, was developed to efficiently flux-volume homogenize TRISO particles with the matrix. Additionally, optimizations were made to hyper- fine group (HFG) slowing down calculations by pretabulating collision probability coefficients and grouping isotopes, significantly reducing the computational time for calculating scattering sources per HFG. Lastly, the pin power reconstruction module was extended to account for temporal behavior in a microreactor analysis problem, specifically for a control drum transient. Verification tests for each of these improvements demonstrated significant performance enhancements and memory reduction.

22 - GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2022

SMALE: Enhancing Scalability of Machine Learning Algorithms on Extreme-Scale Computing Platforms

Deployment and execution of machine learning tasks on extreme-scale computing platforms face several significant technical challenges: 1) High computing cost incurred by dense networks – The computing workload of deep networks with densely-connected topology increases rapidly with the network size, imposing a non-scalable computing model of extreme-scale computing platforms; 2) Non-optimized workload distribution – Many advanced deep learning algorithms, e.g., sparsification and irregular net-work topology, produce very unbalanced workload distribution on extreme-scale computing platforms. The computation efficiency is greatly hindered by the incurred data and computation redundancies as well as long tails of the node with extensive workload; 3) Constraints in data movement and I/O bottle-neck – Inter-node data movement in extreme-scale computing platforms are associated with high energy and latency costs, and subject to the constraints of I/O bandwidth; and 4) Generalization of algorithm realization and acceleration on computing platforms – The large varieties of machine learning algorithms and structures of extreme-scale computing platforms make the derivation of a generalized algorithm realization and acceleration method very challenging, which, however, is the requirement by domain scientists and interested users. We call the above challenges Smale’s Problems in Machine Learning and Understanding for High-Performance Computing Scientific Discovery. The objective of our three-year research project is to develop a holistic innovation set at structure, assembly, and acceleration layers of machine learning algorithms to address the above challenges in algorithm deployment and execution. Three tasks are particularly performed, including: At the algorithm structure level, we investigate the techniques that can structurally sparsify on the topology of deep networks for computing workload reduction. We also study clustering and pruning techniques that can optimize the workload distributions over the extreme-scale computing platforms; At the algorithm assembly level, we derive a unified learning framework for unsupervised transfer learning and dynamic growing capabilities. Novel training methods are also exploited to enhance the training efficiency of the proposed framework; At the algorithm acceleration level, we will develop a series of techniques that can accelerate the computation of sparse matrix operations, which are one of the core executions in deep learning and optimize memory access of the concerned platforms. Our proposed techniques attack the fundamental problems in machine learning algorithms running on extreme-scale computing platforms by vertically integrating the solutions at three closely entangled layers, paving the long-term scaling path of machine learning applications under DOE context. Three tasks corresponding to the above respective research orientations are performed during the three-year project period with our collaborators at ORNL. The outcome of the proposed project is anticipated to form a holistic solution set of novel algorithms and network topologies, efficient training techniques, and fast acceleration methods to promote the computing scalability of the machine learning applications of particular interest to DOE.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2017

Shape Memory Alloys (SMAs) for Non-Pneumatic Tires

Conventional mobility elements, such as pneumatic tires, suffer from a number of issues related to reliability. Two of the more prevalent problems are the high likelihood of single point failure owing to puncture (i.e. flat tire), and loss of efficiency due to reduction in tire pressure over time. In order to overcome these limitations, alternative compliant tire designs not requiring pneumatics have been developed. However, although current designs have significantly reduced the aforementioned issues, they tend to have their own set of limitations. First, non-pneumatic tires designed for high load applications often have restricted envelopment capability, making their performance less than optimal, especially on uneven terrain. Second, tires designed with larger envelopment capability tend to suffer from large amounts of plasticity (permanent deformation) or failure (rupture). Both of these limitations are the direct result of the choice of material being used for the design; conventional metals undergo plastic deformation at low strain while elastomer based designs are often too rigid for the localized deformations needed for high envelopment. Recent advancements at the NASA Glenn research center in a unique class of metals know as shape memory alloys (SMAs) has opened the design space for non-pneumatic compliant tire technologies allowing designs to incorporate orders of magnitude more deformation without damage. The work presented herein highlights the advantages of using SMAs as compared to conventional metals. Additionally, the development of a unique SMA compliant tire design capable of carrying up to 8.9 kN (2000 lbf) with reversible, local deformations on the order of the side wall height will be presented.

Padula, Santo↗
DOE OSTI2020

Loss circulation prevention in geothermal drilling by shape memory polymer

Geothermal formations are naturally fractured with large fracture openings and networks. This can often lead to frequent drilling fluid loss events which is a major contributor to the cost and non-productive time in geothermal drilling. Development of smarter technologies and methods to tackle problem lost circulation is vital for geothermal drilling cost reduction required for geothermal energy to be recognized as a competitive alternative energy source. Mitigation of this problem and other drilling problems such as stuck pipe can minimize overall project cost. In this report, a thermoset shape memory polymer that can be activated by formation natural heat was assessed to seal near wellbore fractures in geothermal wells. The performance of the shape memory polymer (SMP) was evaluated using artificial fractures created in aluminum discs and cylindrical granite cores. Rheology and particle size distribution were considered. A novel testing setup was built for this work, which allows testing of sealing efficiency under dynamic conditions at high temperature. Analysis showed that the SMP has efficiently succeeded in forming a strong plug inside the fractures and stopped fluid loss at high sealing pressure. This smart loss circulation material can expand within the fractures to reduces non-drilling time and strengthen the wellbore in high-temperature drilling operations.

58 GEOSCIENCES↗
DOE OSTI2020

Population Count on Intel® CPU, GPU, and FPGA

Population count is a primitive used in many applications. Commodity processors have dedicated instructions for achieving high-performance population count. Motivated by the productivity of high-level synthesis and the importance of population count, in this paper we investigated the OpenCL implementations of population count algorithms, and evaluated their performance and resource utilizations on an FPGA. Based on the results, we select the most efficient implementation. Then we derived a reduction pattern from a representative application of population count. We parallelized the reduction with atomic functions, and optimized it with vectorized memory accesses, tree reduction, and compute-unit duplication. We evaluated the performance of the reduction kernel on an Intel (R) Xeon (R) CPU and an Intel (R) IrisT Pro integrated GPU, and an FPGA card that features an Intel (R) Arria (R) 10 FPGA. When DRAM memory bandwidth is comparable on the three computing platforms, the FPGA can achieve the highest kernel performance for large workload. On the other hand, we described performance bottlenecks on the FPGA. To make FPGAs more competitive in raw performance compared to high-performant CPU and GPU platforms, it is important to increase external memory bandwidth, minimize data movement between a host and a device, and reduce OpenCL runtime overhead on an FPGA.

heterogeneous computing↗