SEARCH · Search NASA
Results for “kernel Fusion”
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.
Kernel fusion in atomistic spin dynamics simulations on Nvidia GPUs using tensor core
In atomistic spin dynamics simulations, the time cost of constructing the space- and time-displaced pair correlation function in real space increases quadratically as the number of spins N, leading to significant computational effort. The GEMM subroutine can be adopted to accelerate the calculation of the dynamical spin-spin correlation function, but the computational cost of simulating large spin systems (>40000 spins) on CPUs remains expensive. In this work, we perform the simulation on the graphics processing unit (GPU), a hardware solution widely used as an accelerator for scientific computing and deep learning. Here we show that GPUs can accelerate the simulation up to 25-fold compared to multi-core CPUs when using the GEMM subroutine on both. To hide memory latency, we fuse the element-wise operation into the GEMM kernel using CUTLASS that can improve the performance by 26% ~ 33% compared to implementation based on cuBLAS. Furthermore, we perform the on-the-fly calculation in the epilogue of the GEMM subroutine to avoid saving intermediate results on global memory, which makes the large-scale atomistic spin dynamics simulation feasible and affordable.
RAP: Resource-aware Automated GPU Sharing for Multi-GPU Recommendation Model Training and Input Preprocessing
Ensuring high-quality recommendations for newly onboarded users requires the continuous retraining of Deep Learning Recommendation Models (DLRMs) with freshly generated data. To serve the online DLRM retraining, existing solutions use hundreds of CPU computing nodes designated for input preprocessing, causing significant power consumption that surpasses even the power usage of GPU trainers. To this end, we propose RAP, an end-to-end DLRM training framework that supports Resource-aware Automated GPU sharing for DLRM input Preprocessing and Training. The core idea of RAP is to accurately capture the remaining GPU computing resources during DLRM training for input preprocessing, achieving superior training efficiency without requiring additional resources. Specifically, RAP utilizes a co-running cost model to efficiently assess the costs of various input preprocessing operations, and it implements a resource-aware horizontal fusion technique that adaptively merges smaller kernels according to GPU availability, circumventing any interference with DLRM training. In addition, RAP leverages a heuristic searching algorithm that jointly optimizes both the input preprocessing graph mapping and the co-running schedule to maximize the end-to-end DLRM training throughput. The comprehensive evaluation shows that RAP achieves 78.3× speedup on average over CPU-based DLRM input preprocessing frameworks. In addition, the end-to-end training throughput of RAP is only 2.04% lower than the ideal case, which has no input preprocessing overhead.
Efficient screening of rare large pit anomalies on polished surfaces using a minimalist sampling scheme
Lawrence Livermore National Laboratory (LLNL) has made significant strides in generating clean energy through its inertial confinement fusion (ICF) experiments. These experiments rely on high-density carbon (HDC) coated shells to encapsulate the fusion fuel. The success of these experiments is heavily dependent on the surface quality of these shells, as even minor imperfections, such as deep pits, can negatively impact fusion yield. Ensuring the required smoothness involves an extensive surface-finishing process that spans approximately 20 stages, making it both time-intensive and resource-demanding. A critical challenge in this process is the need for high-resolution scans to detect rare deep pits, which can be costly and impractical if performed on every shell. This highlights the necessity of developing more efficient scanning methods to optimize time and cost without compromising accuracy. To address these challenges, we introduce a novel approach that employs the multivariate Dvoretzky–Kiefer–Wolfowitz (DKW) inequality to provide a probabilistic upper bound on the error in estimating pit distribution characteristics via a Kernel Density Estimator (KDE). This error bound enables efficient and reliable estimation of pit distribution characteristics at a specified statistical confidence level using a minimal number of surface scans. The integrated DKW-KDE approach was validated through surface-finishing experiments across two batches of HDC-coated shells, demonstrating consistent and robust performance across multiple stages of the surface-finishing experiments. The validation studies suggest that the integrated DKW-KDE approach achieves comparable accuracy in estimating the risk of deleterious large pits with six scans, thus conserving time and resources. Further evaluations show that performance remains consistent across batches and over multiple polishing stages. In conclusion, based on these findings, one can leverage the minimal-scan insights to strategically improve the bottleneck inspection process, thus enhancing the productivity and quality of shell polishing and similar challenging manufacturing processes.
A Discrete Hankel Transform Approach to Nuclear Data Processing for Fusion Applications
This study introduces advancements to the numerical solutions employed in the processing of nuclear data for fusion applications. It leverages the convolution theorem and Fourier transform techniques to enhance computational efficiency and broaden applicability. Building upon a previously reported discrete Hankel transform approach for Doppler broadening, this work refines the solution of convolution integrals central to these applications. The methodology provides a general and unified framework for evaluating any convolution operation, regardless of whether the underlying problem involves temperature effects in nuclear reactions. The applicability to the nuclear data processing for fusion is demonstrated by deriving the convolution integrals for some of the fusion-related quantities. As before, the convolution operation utilizes a Gaussian-based kernel; however, the discrete Hankel transform of order $𝛼$ = $\frac{1}{2}$ is now applied to the forward Fourier transform of the nonkernel argument, rather than the inverse Fourier transform. This modification eliminates the need for the integration of the nonkernel, cross section–based function, which is a step that posed challenges for certain pointwise cross-section representations. It also removes the requirement for cross-section linearization. Optimized for graphics processing unit architectures, the approach significantly improves computational performance. These advancements are currently under evaluation as the foundation for the next-generation thermonuclear data file processing codes being developed at Lawrence Livermore National Laboratory.
Full wave modeling of radio-frequency beams in tokamaks in the electron cyclotron frequency range
Simulation of full wave, without paraxial approximation, high-resolution solution of wave equations in frequency domain in the electron cyclotron resonance (ECR) frequency range for realistic Tokamak plasma parameters became possible by using recently formulated hybrid iterative algorithm [Svidzinski et al., Phys. Plasmas 25, 082509 (2018)] for numerically solving discretized wave equations. This approach combines time evolution and iterative relaxation techniques into iteration cycles. This algorithm is implemented in 2D code FullWave, solving wave equations in Tokamaks in cold and hot plasma models, and it has been tested in 3D full wave iterative RF beams simulation tool, which is presently being developed to model 3D ECRH RF beams in fusion devices using dynamic grid adaptation. The results of 2D full wave modeling, assuming specified toroidal mode number, of ECRH RF beams in DIII-D plasma, performed in the cold and hot plasma models for outboard and top launch scenarios using FullWave are presented. Nonlocal hot plasma response model, based on accurate numerical solution of linearized Vlasov equation, is used to model beam propagation and absorption in the 2nd electron cyclotron harmonic region. Demonstration of capability of the hybrid iterative algorithm to model ECRH RF beams in 3D is made by simulating a substantial part of realistic beam in DIII-D, launched from outboard side of the machine. All relevant physics of RF beam propagation, most of which is not captured in paraxial approximation, such as beam's divergence, interference between the X and O modes in the beam, X-O mode conversion, beam splitting into the X and O mode beams, transformation of beam's cross section, and absorption at the 2nd electron cyclotron harmonic, is captured in the simulations. A numerical technique to find an optimal beam polarization at the launcher to launch a nearly pure X or O mode beam in plasma is developed and tested.
A Performance-Portable MultiGPU Implementation of 3D Euler Equations using ProtoX and IRIS
Computational scientists often face challenges when developing and optimizing code for high-performance computing (HPC), especially when trying to leverage GPUs. Given the heterogeneity of the nodes that comprise many modern HPC facilities, considerable demand exists for performance portable solutions for the core computational kernels used in many scientific computing libraries. In this work, we demonstrate a fourth-order finite volume method–based implementation of the Euler equations, which are an integral part of computational fluid dynamics. Our performance-portable multiGPU implementation for Euler equations uses ProtoX to generate kernels and IRIS for portability. ProtoX is a domain-specific language that uses a structured-grid partial differential equation library called Proto as its front end and the SPIRAL code generation system as its back end to generate optimized kernels for different architectures. Optimized kernels generated by ProtoX are orchestrated through the IRIS intelligent runtime system to provide portability. Two levels of optimizations within the IRIS runtime— directed acyclic graph fusion and task fusion—are explored to efficiently utilize computing resources in a multiGPU environment. Performance improvement through these optimizations is showcased by comparing the base ProtoX-IRIS implementation on AMD GPUs (Frontier node) and on NVIDIA GPUs (NVIDIA DGX-1).
cuAlign: Scalable Network Alignment on GPU Accelerators
Given two graphs, the objective of network alignment is to find the best one-to-one mapping of vertices in one graph (??) to vertices in the other (??), such that the number of overlaps is maximized. We say that edges(??, ??) ???and(??', ??') ??? are overlapped if ?? is mapped to ??' and ?? is mapped to??'. Network alignment is an important optimization problem with several applications in bioinformatics, computer vision and ontology matching. Since it is an NP-hard problem, efficient heuristics and scalable implementations are necessary. In this work, we introduce a new framework that combines the concepts of intra-network proximity using vertex embedding,Belief Propagation (BP) and approximate weighted matching, and provides qualitative improvements up to22%over state-of-the-art approaches. We also provide scalable implementations on GPU accelerators, demonstrating up to19×speedup for Belief Propagation and 3× speedup for approximate weighted matching relative to previous multithreaded implementation. A combination of combinatorial and algebraic kernels within the network alignment algorithm poses significant hurdles for parallelization. Load imbalance and irregular DRAM traffic limit achievable performance on GPUs. Our novel approach identifies and exploits unique structural proper-ties of the BP-based algorithm and employs code fusion to reduce data movement between different steps of the algorithm. Using a diverse set of inputs, we demonstrate qualitative improvements of our algorithms, and performance gains of our GPU-accelerated implementation. We believe that our work will enable algorithmic improvements and practical applications of network alignment.
Scattering theory in noncanonical phase space: A Drift-Kinetic collision operator for weakly collisional plasmas
After developing a scattering theory for grazing collisions in general noncanonical phase spaces, we introduce a guiding center collision operator in five-dimensional phase space designed for plasma regimes characterized by long wavelengths (relative to the Larmor radius), low frequencies (relative to the cyclotron frequency), and weak collisionality (where repeated Coulomb collisions induce cumulatively small changes in particle magnetic moment). The collision operator is fully determined by the noncanonical Hamiltonian structure of guiding center dynamics and exhibits a metriplectic structure, ensuring the conservation of particle number, momentum, energy, and interior Casimir invariants. It also satisfies an H-theorem, allowing for deviations from an equilibrium Maxwellian distribution due to the nontrivial kernel of the noncanonical guiding center Poisson tensor, spanned by the magnetic moment. We propose that this collision operator and its underlying mathematical structure may offer valuable insight into the study of turbulence, transport, and self-organizing phenomena in both laboratory and astrophysical plasmas.
Building a new multiphysics workflow in MOOSE: application to tritium migration, trapping and advection in TMAP8
Fusion devices are anticipated to produce and consume several kilograms of tritium per year. This rare fuel resource is both highly mobile and radioactive, making tracking inventories a priority for operation and safety. The fusion safety program at the Idaho National Laboratory has been developing the Tritium Migration and Analysis Program (TMAP), of which the latest version is a MOOSE-based application. TMAP8 is verified against its predecessors and possesses additional multi-dimensional tritium migration modeling capabilities. As we extend its capabilities towards both whole device (in multiple dimensions) and whole plant (with multiple components) simulations, the syntax of inputs must become compact, descriptive, compatible with quality assurance processes, and as error-proof as achievable. The new Physics system developed MOOSE can set up equations and instantiating them on plant components. The system permits the automatic definition of complex discretization with a consistency between object parameters achieved programmatically. The Physics system can currently instantiate the equations for heat conduction and Navier Stokes weakly compressible flow. In MOOSE-terms, it automates the definition of kernels, boundary conditions, and several core and helper materials and fields. As part of this effort, Physics classes were developed for tritium migration, trapping and advection within either a multi-dimensional Navier Stokes fluid dynamics simulation, or a 1D thermal hydraulics piping system. In this presentation, we will showcase the new syntax, its application to several verification and validation cases which were already studied using the classical TMAP8 syntax, and a demonstration of the new coupling capabilities for the migration of tritium into blanket coolant channels and the subsequent advection into the coolant loop.
Imposing correct jellium response is key to predict the density response by orbital-free DFT
Orbital-free density functional theory constitutes a computationally highly effective tool for modeling electronic structures of systems ranging from room-temperature materials to warm dense matter. Its accuracy critically depends on the employed kinetic energy (KE) density functional, which has to be supplied as an external input. In this work we consider several nonlocal and Laplacian-level KE functionals and use an external harmonic perturbation to compute the static density response at T = 0 K in the linear and beyond-linear response regimes. We test for the satisfaction of exact conditions in the limit of uniform densities and for how approximate KE functionals reproduce the density response of realistic materials (e.g., Al and Si) against the Kohn-Sham DFT reference, which employs the exact KE. The results illustrate that several functionals violate exact conditions in the uniform electron gas (UEG) limit. We find a strong correlation between the accuracy of the KE functionals in the UEG limit and in the strongly inhomogeneous case. This empirically demonstrates the importance of imposing the limit of UEG response for uniform densities and validates the use of the Lindhard function in the formulation of kernels for nonlocal functionals. This conclusion is substantiated by additional calculations for bulk aluminum (Al) with a face-centered cubic (fcc) lattice and silicon (Si) with an fcc lattice, body-centered cubic (bcc) lattice, and semiconducting crystal diamond state. The analysis of fcc Al, and fcc as well as bcc Si data follows closely the conclusions drawn for the UEG, allowing us to extend our conclusions to realistic systems that are subject to density inhomogeneities induced by ions.
A Probabilistic Scheme for Semilinear Nonlocal Diffusion Equations with Volume Constraints
This work presents a probabilistic scheme for solving semilinear nonlocal diffusion equations with volume constraints and integrable kernels. The nonlocal model of interest is defined by a time-dependent semilinear partial integro-differential equation (PIDE), in which the integro-differential operator consists of both local convection-diffusion and nonlocal diffusion operators. Here, our numerical scheme is based on the direct approximation of the nonlinear Feynman–Kac formula that establishes a link between nonlinear PIDEs and stochastic differential equations. The exploitation of the Feynman–Kac representation avoids solving dense linear systems arising from nonlocal operators. Compared with existing stochastic approaches, our method can achieve first-order convergence after balancing the temporal and spatial discretization errors, which is a significant improvement of existing probabilistic/stochastic methods for nonlocal diffusion problems. Error analysis of our numerical scheme is established. The effectiveness of our approach is shown in two numerical examples. The first example considers a three-dimensional nonlocal diffusion equation to numerically verify the error analysis results. The second example presents a physics problem motivated by the study of heat transport in magnetically confined fusion plasmas.
Plasma assisted NH 3 /H 2 /air ignition in nanosecond discharges with non-equilibrium energy transfer
Ammonia (NH 3 ), with its high energy density and easiness to store and transport as a hydrogen carrier, has become a promising alternative green fuel. However, its adoption in power generation is hindered by challenges such as low burning velocity, slow low-temperature oxidation, high NO x emissions, and ignition difficulty. Here, this work computationally investigates the effects of non-equilibrium energy transfer by nanosecond discharges on NH 3 ignition and flame propagation in an NH 3 /H 2 /air flow at 700 K and 1 atm. The simulation results demonstrate that NH 3 /air mixtures require a large ignition energy due to their large critical ignition radius. It is shown that adding 30 % hydrogen significantly reduces the critical ignition radius and minimum ignition energy. Two-dimensional modeling further shows a non-monotonic dependence of ignition kernel volume on the applied voltage and reduced electric field. The optimum ignition enhancement occurs at 200 Td where the generation of electronically excited species and radicals including N 2 (B), O( 1 D) and OH becomes most efficient. Higher voltages divert electron energy toward ionization, which makes it less effective for NH 3 ignition. The study also identifies an optimal electrode gap size for a given pulse energy. Smaller gap sizes increase deposited energy density, raising temperature and radical concentrations. However, excessive reduction of the gap distance reduces flame propagation speed due to the flame stretch effect in rich mixtures with the effective Lewis number greater than unity. A nonlinear relationship between pulse repetition frequency and ignition kernel volume is observed in a nanosecond pulsed high frequency discharge (NPHFD). An optimal frequency range of 200 kHz to 2 MHz is found when two pulses are used. In addition, an optimal number of pulses exists for each pulse repetition frequency, with higher frequencies requiring more pulses to maximize the overlap region. These findings provide critical insights on developing controlled plasma discharge techniques for efficient NH 3 ignition in reactive flows within internal combustion engines and gas turbines.
Simulation of electron Bernstein waves using FullWave with a 2D non-local hot plasma model
Hot plasma wave simulation capability is expanded in the FullWave code by updating the hybrid iterative solver in the code with a semi-implicit time stepping method. The new approach is used to simulate Electron Bernstein Wave (EBW) heating in over-dense spherical tokamak plasmas. The code’s hybrid iterative solver circumvents the prohibitive memory cost of direct methods by combining a time evolution of Maxwell’s equations with frequency-domain relaxation, while the conductivity kernel, calculated via 3D particle tracking, captures the essential non-local wave–particle interactions. One-dimensional EBW simulations verify the algorithm’s accuracy by demonstrating mode conversion from X-mode wave to EBW at the upper hybrid resonance and a strong cyclotron damping near the plasma core. Two-dimensional simulation reproduces the predicted short EBW wavelength and quantitatively matches the hot-plasma dispersion relation. This study demonstrates the fidelity of the hybrid solver for the electron cyclotron frequency range.
Shock Hugoniot calculations using on-the-fly machine learned force fields with ab initio accuracy
We present a framework for computing the shock Hugoniot using on-the-fly machine learned force field (MLFF) molecular dynamics simulations. In particular, we employ an MLFF model based on the kernel method and Bayesian linear regression to compute the free energy, atomic forces, and pressure, in conjunction with a linear regression model between the internal and free energies to compute the internal energy, with all training data generated from Kohn–Sham density functional theory (DFT). We verify the accuracy of the formalism by comparing the Hugoniot for carbon with recent Kohn–Sham DFT results in the literature. In so doing, we demonstrate that Kohn–Sham calculations for the Hugoniot can be accelerated by up to two orders of magnitude, while retaining ab initio accuracy. We apply this framework to calculate the Hugoniots of 14 materials in the FPEOS database, comprising 9 single elements and 5 compounds, between temperatures of 10 kK and 2 MK. We find good agreement with first principles results in the literature while providing tighter error bars. In addition, we confirm that the inter-element interaction in compounds decreases with temperature.