Search NASA⌕ Search

SEARCH · Search NASA

Results for “Kernel methods”

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 919 records · Page 51

Modelling the Lyman-α forest with Eulerian and SPH hydrodynamical methods

ABSTRACT We compare two state-of-the-art numerical codes to study the overall accuracy in modelling the intergalactic medium and reproducing Lyman-α forest observables for DESI and high-resolution data sets. The codes employ different approaches to solving both gravity and modelling the gas hydrodynamics. The first code, Nyx, solves the Poisson equation using the Particle-Mesh (PM) method and the Euler equations using a finite-volume method. The second code, CRK-HACC , uses a Tree-PM method to solve for gravity, and an improved Lagrangian smoothed particle hydrodynamics (SPH) technique, where fluid elements are modelled with particles, to treat the intergalactic gas. We compare the convergence behaviour of the codes in flux statistics as well as the degree to which the codes agree in the converged limit. We find good agreement overall with differences being less than observational uncertainties, and a particularly notable ≲1 per cent agreement in the 1D flux power spectrum. This agreement was achieved by applying a tessellation methodology for reconstructing the density in CRK-HACC instead of using an SPH kernel as is standard practice. We show that use of the SPH kernel can lead to significant and unnecessary biases in flux statistics; this is especially prominent at high redshifts, z ∼ 5, as the Lyman-α forest mostly comes from lower-density regions that are intrinsically poorly sampled by SPH particles.

79 ASTRONOMY AND ASTROPHYSICS↗

New method for analysis of X-ray computed tomography scans of TRISO fuel forms

X-ray computed tomography (XCT) is a powerful characterization tool that has been used for examination of both tristructural-isotropic (TRISO) coated particles and fuel forms. Rapid, nondestructive imaging of complete specimens by XCT provides significant characterization advantages in both speed and comprehensiveness over traditional materialographic methods. This paper discusses a method that has been developed to analyze XCT images of pebble fuel forms containing ~18,000 TRISO particles each. The XCT images are three-dimensional with contrast based on relative x-ray transparency. In the case of TRISO fuel forms, XCT image intensity distinguishes between the high-density kernel, the medium-density SiC, and the low-density matrix and carbon layers. Image analysis software has been developed to process XCT data sets to extract spatial distribution information, including particle nearest-neighbor distances, local packing fraction, and fuel-free zone thickness in TRISO fuel forms. Finally, although this type of high-resolution XCT analysis is likely not feasible as a quality control measurement for commercial TRISO fuel, it is highly useful in generating feedback for the development of fuel form fabrication processes.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Machine learning approaches for structural and thermodynamic properties of a Lennard-Jones fluid

Predicting the functional properties of many molecular systems relies on understanding how atomistic interactions give rise to macroscale observables. However, current attempts to develop predictive models for the structural and thermodynamic properties of condensed-phase systems often rely on extensive parameter fitting to empirically selected functional forms whose effectiveness is limited to a narrow range of physical conditions. Here, we illustrate how these traditional fitting paradigms can be superseded using machine learning. Specifically, we use the results of molecular dynamics simulations to train machine learning protocols that are able to produce the radial distribution function, pressure, and internal energy of a Lennard-Jones fluid with increased accuracy in comparison to previous theoretical methods. The radial distribution function is determined using a variant of the segmented linear regression with the multivariate function decomposition approach developed by Craven et al. [J. Phys. Chem. Lett. 11, 4372 (2020)]. The pressure and internal energy are determined using expressions containing the learned radial distribution function and also a kernel ridge regression process that is trained directly on thermodynamic properties measured in simulation. The presented results suggest that the structural and thermodynamic properties of fluids may be determined more accurately through machine learning than through human-guided functional forms.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Extracting the spectrum of a flow by spatial filtering

We show that the spectrum of a flow field can be extracted within a local region by straightforward filtering in physical space. We find that for a flow with a certain level regularity, the filtering kernel must have a sufficient number of vanishing moments in order for the "filtering spectrum" to be meaningful. Our derivation follows a similar analysis by [1] for the wavelet spectrum, where we show that the filtering kernel has to have at least p vanishing moments in order to correctly extract a spectrum k (-alpha) with (alpha) less/equal to p + 2. For example, any ow with a spectrum shallower than k (-3) can be extracted by a straightforward average on grid-cells of a stencil. We construct two new "simple stencil" kernels, MI and MII, with only two and three fixed stencil weight coefficients, respectively, and that have sufficient vanishing moments to allow for extracting spectra steeper than k(-3). We demonstrate our results using synthetic fields, 2D turbulence from a Direct Numerical Simulation, and 3D turbulence from the JHU Database. Our method guarantees energy conservation and can be useful in both simulations and experiments when a straightforward Fourier analysis is not justified, such as within coherent flow structures covering non-rectangular regions or in geophysical flows on Earth's curved surface.

42 ENGINEERING↗

SU-F-J-133: Adaptive Radiation Therapy with a Four-Dimensional Dose Calculation Algorithm That Optimizes Dose Distribution Considering Breathing Motion

Purpose: To model patient motion and produce four-dimensional (4D) optimized dose distributions that consider motion-artifacts in the dose calculation during the treatment planning process. Methods: An algorithm for dose calculation is developed where patient motion is considered in dose calculation at the stage of the treatment planning. First, optimal dose distributions are calculated for the stationary target volume where the dose distributions are optimized considering intensity-modulated radiation therapy (IMRT). Second, a convolution-kernel is produced from the best-fitting curve which matches the motion trajectory of the patient. Third, the motion kernel is deconvolved with the initial dose distribution optimized for the stationary target to produce a dose distribution that is optimized in four-dimensions. This algorithm is tested with measured doses using a mobile phantom that moves with controlled motion patterns. Results: A motion-optimized dose distribution is obtained from the initial dose distribution of the stationary target by deconvolution with the motion-kernel of the mobile target. This motion-optimized dose distribution is equivalent to that optimized for the stationary target using IMRT. The motion-optimized and measured dose distributions are tested with the gamma index with a passing rate of >95% considering 3% dose-difference and 3mm distance-to-agreement. If the dose delivery per beam takes place over several respiratory cycles, then the spread-out of the dose distributions is only dependent on the motion amplitude and not affected by motion frequency and phase. This algorithm is limited to motion amplitudes that are smaller than the length of the target along the direction of motion. Conclusion: An algorithm is developed to optimize dose in 4D. Besides IMRT that provides optimal dose coverage for a stationary target, it extends dose optimization to 4D considering target motion. This algorithm provides alternative to motion management techniques such as beam-gating or breath-holding and has potential applications in adaptive radiation therapy.

60 APPLIED LIFE SCIENCES↗

Data-driven parameterization of the generalized Langevin equation

Significance The generalized Langevin equation (GLE) provides a precise description of coarse-grained variable dynamics in reduced dimension models. However, computation of the memory kernel poses a major challenge to the practical use of the GLE. This paper presents a data-driven approach to compute the memory kernel, relying on a hierarchy of parameterized rational approximations in terms of the Laplace transform, which can be expanded to arbitrarily high order as necessary. This parameterization makes it convenient to represent the GLE via an extended stochastic model where the memory term is eliminated by properly introducing auxiliary variables. The present method is well-suited for constructing reduced models for nonequilibrium properties of complex systems such as biomolecules, chemical reaction networks, and climate simulations.

97 MATHEMATICS AND COMPUTING↗

Effect of time-of-flight and point spread function modeling on detectability of myocardial defects in PET

Purpose: A study was designed to investigate the impact of time-of-flight (TOF) and point spread function (PSF) modeling on the detectability of myocardial defects. Methods: Clinical FDG-PET data were used to generate populations of defect-present and defect-absent images. Defects were incorporated at three contrast levels, and images were reconstructed by ordered subset expectation maximization (OSEM) iterative methods including ordinary Poisson, alone and with PSF, TOF, and PSF+TOF. Channelized Hotelling observer signal-to-noise ratio (SNR) was the surrogate for human observer performance. Results: For three iterations, 12 subsets, and no postreconstruction smoothing, TOF improved overall defect detection SNR by 8.6% as compared to its non-TOF counterpart for all the defect contrasts. Due to the slow convergence of PSF reconstruction, PSF yielded 4.4% less SNR than non-PSF. For reconstruction parameters (iteration number and postreconstruction smoothing kernel size) optimizing observer SNR, PSF showed larger improvement for faint defects. The combination of TOF and PSF improved mean detection SNR as compared to non-TOF and non-PSF counterparts by 3.0% and 3.2%, respectively. Conclusions: For typical reconstruction protocol used in clinical practice, i.e., less than five iterations, TOF improved defect detectability. In contrast, PSF generally yielded less detectability. For large number of iterations, TOF+PSF yields the best observer performance.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Grassmannian Diffusion Maps--Based Dimension Reduction and Classification for High-Dimensional Data

This work introduces the Grassmannian diffusion maps (GDMaps), a novel nonlinear dimensionality reduction technique that defines the affinity between points through their representation as low-dimensional subspaces corresponding to points on the Grassmann manifold. Here, the method is designed for applications, such as image recognition and data-based classification of constrained high-dimensional data where each data point itself is a high-dimensional object (i.e., a large matrix) that can be compactly represented in a lower-dimensional subspace. The GDMaps is composed of two stages. The first is a pointwise linear dimensionality reduction wherein each high-dimensional object is mapped onto the Grassmann manifold representing the low-dimensional subspace on which it resides. The second stage is a multipoint nonlinear kernel-based dimension reduction using diffusion maps to identify the subspace structure of the points on the Grassmann manifold. To this end, an appropriate Grassmannian kernel is used to construct the transition matrix of a random walk on a graph connecting points on the Grassmann manifold. Spectral analysis of the transition matrix yields low-dimensional Grassmannian diffusion coordinates embedding the data into a low-dimensional reproducing kernel Hilbert space. Further, a novel data classification/recognition technique is developed based on the construction of an overcomplete dictionary of reduced dimension whose atoms are given by the Grassmannian diffusion coordinates. Three examples are considered. First, a "toy" example shows that the GDMaps can identify an appropriate parametrization of structured points on the unit sphere. The second example demonstrates the ability of the GDMaps to revealing the intrinsic subspace structure of high-dimensional random field data. In the last ex- ample, a face recognition problem is solved considering face images subject to varying illumination conditions, changes in face expressions, and occurrence of occlusions. The technique presented high recognition rates (i.e., 95% in the best case) using a fraction of the data required by conventional methods.

42 ENGINEERING↗

Simulation of macromolecule self-assembly in solution: A multiscale approach

One of the most common processes to produce polymer nanoparticles is to induce self-assembly by using the solvent-displacement method, in which the polymer is dissolved in a “good” solvent and the solution is then mixed with an “anti-solvent”. The polymer ability to self-assemble in solution is therefore determined by its structural and transport properties in solutions of the pure solvents and at the intermediate compositions. In this work, we focus on poly-ε-caprolactone (PCL) which is a biocompatible polymer that finds widespread application in the pharmaceutical and biomedical fields, performing simulation at three different scales using three different computational tools: full atomistic molecular dynamics (MD), population balance modeling (PBM) and computational fluid dynamics (CFD). Simulations consider PCL chains of different molecular weight in solution of pure acetone (good solvent), of pure water (anti-solvent) and their mixtures, and mixing at different rates and initial concentrations in a confined impinging jets mixer (CIJM). Our MD simulations reveal that the nano-structuring of one of the solvents in the mixture leads to an unexpected identical polymer structure irrespectively of the concentration of the two solvents. In particular, although in pure solvents the behavior of the polymer is, as expected, very different, at intermediate compositions, the PCL chain shows properties very similar to those found in pure acetone as a result of the clustering of the acetone molecules in the vicinity of the polymer chain. We derive an analytical expression to predict the polymer structural properties in solution at different solvent compositions and use it to formulate an aggregation kernel to describe the self-assembly in the CIJM via PBM and CFD. Simulations are eventually validated against experiments.

36 MATERIALS SCIENCE↗

HELIOSEISMOLOGY OF A REALISTIC MAGNETOCONVECTIVE SUNSPOT SIMULATION

We compare helioseismic travel-time shifts measured from a realistic magnetoconvective sunspot simulation using both helioseismic holography and time-distance helioseismology, and measured from real sunspots observed with the Helioseismic and Magnetic Imager instrument on board the Solar Dynamics Observatory and the Michelson Doppler Imager instrument on board the Solar and Heliospheric Observatory. We find remarkable similarities in the travel-time shifts measured between the methodologies applied and between the simulated and real sunspots. Forward modeling of the travel-time shifts using either Born or ray approximation kernels and the sound-speed perturbations present in the simulation indicates major disagreements with the measured travel-time shifts. These findings do not substantially change with the application of a correction for the reduction of wave amplitudes in the simulated and real sunspots. Overall, our findings demonstrate the need for new methods for inferring the subsurface structure of sunspots through helioseismic inversions.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Temporal ensemble learning of univariate methods for short term load forecasting

Short term load forecasting (STLF) is a fundamental component of demand response programs in smart grids. Recent literature has focused on complex neural networks and deep learning models for solving STLF. While these models work well for load forecasting with complex non-linear relationships, they have been shown to be less effective than simpler univariate models for STLF problems with under a 6 hours horizon. Given the lack of multivariate data (such as temperature) in many practical datasets, we need better univariate prediction models for STLF. By partitioning the dataset by temporal features, we develop a novel ensemble learning method that exploits daily seasonality in electricity consumption data to improve accuracy of popular univariate models. We train an ensemble of models from the dataset partitions. We develop a variety of methods, including Ridge Regression, to increase the robustness of the ensemble prediction. To show the effectiveness of our approach, we perform detailed evaluation using an aggregated user electricity consumption dataset collected by the Los Angeles Department of Water and Power (LADWP). We select four popular prediction algorithms in literature for our experiments, including Kernel Regression (KR), Support Vector Regression (SVR) and neural network approaches. We compare the performance of these algorithms applying our ensemble approach to training only one single model. Our approach leads to an 11.2% decrease in mean absolute percentage error (MAPE) and 21.3% decrease in root mean squared error (RMSE) over the single model approach for KR, and a 30% and 32.4% decrease in MAPE and RMSE respectively for SVR. These ensemble models also outperform the neural network approaches.

14 SOLAR ENERGY↗

Prediction of Adsorption Energies for Chemical Species on Metal Catalyst Surfaces Using Machine Learning

Computational catalyst screening has the potential to significantly accelerate heterogeneous catalyst discovery. Typically, this involves developing microkinetic reactor models that are based on parameters obtained from density functional theory and transition-state theory. To reduce the large computational cost involved in computing various adsorption and transition-state energies of all possible surface states on a large number of catalyst models, linear scaling relations for surface intermediates and transition states have been developed that only depend on a few, typically one or two descriptors, such as the carbon atom adsorption energy. As a result, only the descriptor values have to be computed for various active site models to generate volcano curves in activity or selectivity. Unfortunately, for more complex chemistries the predictability of linear scaling relations is unknown. Also, the selection of descriptors is essentially a trial and error process. Here, using a database of adsorption energies of the surface species involved in the decarboxylation and decarbonylation of propionic acid over eight monometalic transition-metal catalyst surfaces (Ni, Pt, Pd, Ru, Rh, Re, Cu, Ag), we tested if nonlinear machine learning (ML) models can outperform the linear scaling relations in prediction accuracy when predicting the adsorption energy for various species on a metal surface based on data from the rest of the metal surfaces. We found linear scaling relations to hold well for predictions across metals with a mean-absolute error of 0.12 eV, and ML methods being unable to outperform linear scaling relations when the training dataset contains a complete set of energies for all of the species on various metal surfaces. Only when the training dataset is incomplete, namely, contains a random subset of species’ energies for each metal, a currently unlikely scenario for catalyst screening, do kernel-based ML models significantly outperform linear scaling relations. We also found that simple coordinate-free species descriptors, such as bond counts, achieve as good results as sophisticated coordinate-based descriptors. Finally, we propose an approach for automatic discovery of appropriate metal descriptors using principal component analysis.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Adiabatic-connection fluctuation-dissipation DFT for the structural properties of solids—The renormalized ALDA and electron gas kernels

We present calculations of the correlation energies of crystalline solids and isolated systems within the adiabatic-connection fluctuation-dissipation formulation of density-functional theory. We perform a quantitative comparison of a set of model exchange-correlation kernels originally derived for the homogeneous electron gas (HEG), including the recently introduced renormalized adiabatic local-density approximation (rALDA) and also kernels which (a) satisfy known exact limits of the HEG, (b) carry a frequency dependence, or (c) display a 1/k{sup 2} divergence for small wavevectors. After generalizing the kernels to inhomogeneous systems through a reciprocal-space averaging procedure, we calculate the lattice constants and bulk moduli of a test set of 10 solids consisting of tetrahedrally bonded semiconductors (C, Si, SiC), ionic compounds (MgO, LiCl, LiF), and metals (Al, Na, Cu, Pd). We also consider the atomization energy of the H{sub 2} molecule. We compare the results calculated with different kernels to those obtained from the random-phase approximation (RPA) and to experimental measurements. We demonstrate that the model kernels correct the RPA’s tendency to overestimate the magnitude of the correlation energy whilst maintaining a high-accuracy description of structural properties.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Charge transfer excitations from particle-particle random phase approximation—Opportunities and challenges arising from two-electron deficient systems

The particle-particle random phase approximation (pp-RPA) is a promising method for studying charge transfer (CT) excitations. Through a detailed analysis on two-electron deficient systems, we show that the pp-RPA is always able to recover the long-distance asymptotic –1/R trend for CT excitations as a result of the concerted effect between orbital energies and the pp-RPA kernel. We also provide quantitative results for systems with relatively short donor-acceptor distances. With conventional hybrid or range-separated functionals, the pp-RPA performs much better than time-dependent density functional theory (TDDFT), although it still gives underestimated results which are not as good as TDDFT with system-dependent tuned functionals. For pp-RPA, there remain three great challenges in dealing with CT excitations. First, the delocalized frontier orbitals in strongly correlated systems often lead to difficulty with self-consistent field convergence as well as an incorrect picture with about half an electron transferred. Second, the commonly used density functionals often underestimate the energy gap between the highest occupied molecular orbital and the lowest unoccupied molecular orbital (LUMO) for the two-electron deficient species, resulting in systems with delocalized orbitals. Third, the performance of pp-RPA greatly depends on the energy difference between the LUMO and a higher virtual orbital. However, the meaning of the orbital energies for higher virtual orbitals is still not clear. We also discuss the performance of an approximate pp-RPA scheme that uses density functional tight binding (pp-DFTB) as reference and demonstrate that the aforementioned challenges can be overcome by adopting suitable range-separated hybrid functionals. Here, the pp-RPA and pp-DFTB are thus promising general approaches for describing charge transfer excitations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Nek5000 developments in support of industry and the NRC

This year, the Nuclear Energy Advanced Modeling Simulation program (NEAMS) thermal-hydraulics verification and validation (V&V) work has focused in three areas of Nek5000 V&V-driven development. First, in a close collaborative effort with the U. S. Nuclear Regulatory Commission (NRC) staff, we have continued V&V efforts for the HYMERES-2 project using the OECD/NEA sponsored testing in the PSI PANDA facility. This year’s focus of ANL-NRC collaboration involves Nek5000 setups and validation for a range of problems relevant to and including the HYMERES-2 benchmark from PSI. The primary outcome of this year efforts is a more efficient geometry and inlet modeling simplification after a careful sensitivity study of the inlet profiles and pipe geometries. The resulting modeling choice of a short recycling/fully-developed turbulent inlet is within the experimental uncertainty estimate. This finding simplifies the next step of the cross-V&V HYMERES-2 project. In addition, the ANL team continue to provide assistance to the NRC staff in the form of Nek5000 application support in general and on the use of the HPC platforms of ALCF and INL in particular. This supports the NRC’s assessment of Nek5000 for use with the NRC Blue CRAB code suite. Second, we have implemented and tested more robust model of URANS, namely the k – τ model, a variant of the k-ω model, along with other improvements to RANS Nek5000 modeling in general. Because of its demonstrated robustness and stability, the k – τ model is the only RANS model that has been implemented in the new GPU version of the Nek5000 code, nekRS. Lastly, we report the initial implementation of Jacobian-free Newton Krylov approach to the direct Newton method for steady fluid solvers aimed at acceleration of RANS modeling and at IC improvement for LES campaigns. Also leveraging the Exascale Computing Project (ECP) ANL/CEED & SMR team’s software development effort to support NEAMS problems at large scale of the advanced computing architectures, NekRS, a GPU variant of Nek5000, built on top of kernels from libParanumal using OCCA for portability, has been successfully run on the full system of Summit (4608 nodes, 27648 GPUs).

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Imaging Wavepackets in Real-Space & Disentangling Ultrafast Solvation Dynamics with High-Energy Ultrafast X-ray Scattering

The microscopic information of solute and solute-solvent motions can be measured using ultrafast diffuse x-ray scattering, and compared to molecular dynamics simulations, as demonstrated in studies of solvated molecular systems done in X-ray free-electron lasers. However, for the typical photon energies used in these experiments, the photon momentum transfer range was limited to lower Q values where the signals arising from different parts of the studied system overlap significantly. Recently, this limitation was greatly alleviated by extending the photon energy at the Linac Coherent Light-Source (LCLS), significantly increasing the accessible momentum transfer range. This improvement is transformative for ultrafast diffuse x-ray scattering measurements, enabling for the first time to disentangle the solute scattering difference signal, that persists to higher Q values, from the bulk solvent and solute-solvent cross-terms difference signals. In addition, it enhances the prediction ability of modeling and simulation methods such as the hybrid QM/MM approach. The extended Q range also opens the way to resolve in real-space details regarding coherent wavepackets motions beyond their center-of-mass positions. Here, we present the first results on high-energy (18keV) diffuse x-ray scattering in solution, demonstrating high fidelity time-resolved scattering and analysis of the photoexcited model photocatalysts PtPOP ([Pt2(POP)4]4-), and IrDimen ([Ir2(dimen)4]2+) in several solvents. These complexes provide ideal systems for demonstrating the ability of high-Q ultrafast scattering and QM/MM simulations to decompose ultrafast solvation dynamics into specific changes in the solute-solvent pair distribution function with a particular focus on how electronically excited states change the interaction between the solvent and photo-catalytically active metal sites. We introduce an approach to further utilize the high-energy capability and demonstrate a single-shot ultra-wide-angle X-ray Scattering modality using two perpendicular detectors, spanning a scattering angle range of more than 100 degrees, allowing to extend the accessible momentum transfer range up to Q~14 Å-1. We develop a model-free method to invert the x-ray scattering signals and enable the recovery of multiple pair-density motions that happen simultaneously, allowing us to directly measure in real-space nuclear wavepacket motions. . [1] Natan, Adi. "Real-Space Inversion and Super-Resolution of Ultrafast X-ray Scattering using Natural Scattering Kernels." arXiv preprint arXiv:2107.05576 (2021)

Natan, Adi↗

Towards Enhancing Coding Productivity for GPU Programming Using Static Graphs

The main contribution of this work is to increase the coding productivity of GPU programming by using the concept of Static Graphs. GPU capabilities have been increasing significantly in terms of performance and memory capacity. However, there are still some problems in terms of scalability and limitations to the amount of work that a GPU can perform at a time. To minimize the overhead associated with the launch of GPU kernels, as well as to maximize the use of GPU capacity, we have combined the new CUDA Graph API with the CUDA programming model (including CUDA math libraries) and the OpenACC programming model. We use as test cases two different, well-known and widely used problems in HPC and AI: the Conjugate Gradient method and the Particle Swarm Optimization. In the first test case (Conjugate Gradient) we focus on the integration of Static Graphs with CUDA. In this case, we are able to significantly outperform the NVIDIA reference code, reaching an acceleration of up to 11x thanks to a better implementation, which can benefit from the new CUDA Graph capabilities. In the second test case (Particle Swarm Optimization), we complement the OpenACC functionality with the use of CUDA Graph, achieving again accelerations of up to one order of magnitude, with average speedups ranging from 2x to 4x, and performance very close to a reference and optimized CUDA code. Our main target is to achieve a higher coding productivity model for GPU programming by using Static Graphs, which provides, in a very transparent way, a better exploitation of the GPU capacity. The combination of using Static Graphs with two of the current most important GPU programming models (CUDA and OpenACC) is able to reduce considerably the execution time w.r.t. the use of CUDA and OpenACC only, achieving accelerations of up to more than one order of magnitude. Finally, we propose an interface to incorporate the concept of Static Graphs into the OpenACC Specifications.

58 GEOSCIENCES↗