Search NASA⌕ Search

SEARCH · Search NASA

Results for “Parallel optimization”

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 1,045 records · Page 58

Kokkos GPU Compiler

The Kokkos Clang compiler is a version of the Clang C++ compiler that has been modified to perform targeted code generation for Kokkos constructs in the goal of generating highly optimized code and to provide semantic (domain) awareness throughout the compilation toolchain of these constructs such as parallel for and parallel reduce. This approach is taken to explore the possibilities of exposing the developer’s intentions to the underlying compiler infrastructure (e.g. optimization and analysis passes within the middle stages of the compiler) instead of relying solely on the restricted capabilities of C++ template metaprogramming. To date our current activities have focused on correct GPU code generation and thus we have not yet focused on improving overall performance. The compiler is implemented by recognizing specific (syntactic) Kokkos constructs in order to bypass normal template expansion mechanisms and instead use the semantic knowledge of Kokkos to directly generate code in the compiler’s intermediate representation (IR); which is then translated into an NVIDIA-centric GPU program and supporting runtime calls. In addition, by capturing and maintaining the higher-level semantics of Kokkos directly within the lower levels of the compiler has the potential for significantly improving the ability of the compiler to communicate with the developer in the terms of their original programming model/semantics.

Moss, Nicholas↗

A Large Deviations Analysis of Certain Qualitative Properties of Parallel Tempering and Infinite Swapping Algorithms

Parallel tempering, or replica exchange, is a popular method for simulating complex systems. The idea is to run parallel simulations at different temperatures, and at a given swap rate exchange configurations between the parallel simulations. From the perspective of large deviations it is optimal to let the swap rate tend to infinity and it is possible to construct a corresponding simulation scheme, known as infinite swapping. In this paper we propose a novel use of large deviations for empirical measures for a more detailed analysis of the infinite swapping limit in the setting of continuous time jump Markov processes. Using the large deviations rate function and associated stochastic control problems we consider a diagnostic based on temperature assignments, which can be easily computed during a simulation. Here we show that the convergence of this diagnostic to its a priori known limit is a necessary condition for the convergence of infinite swapping. Finally, the rate function is also used to investigate the impact of asymmetries in the underlying potential landscape, and where in the state space poor sampling is most likely to occur.

97 MATHEMATICS AND COMPUTING↗

Stochastic ordering properties and optimal routing control for a class of finite capacity queueing systems

The problem of routing jobs to parallel queues with identical exponential servers and unequal finite buffer capacities is considered. Stochastic ordering and weak majorization properties on critical performance measures are established by means of event-driven inductions. In particular, it is shown that the intuitive 'join the shortest non-full queue' (SNQ) policy is optimal with respect to an overall function that accounts for holding and blocking costs. Moreover, the buffer allocation problem is solved by proving the intuitive result that, for a fixed total buffer capacity, the optimal allocation scheme is the one in which the difference between the maximum and minimum queue capacities is minimized, i.e., becomes either 0 or 1.

Towsley, Don↗

Optimization of Sparse Matrix-Vector Multiplication on Emerging Multicore Platforms

We are witnessing a dramatic change in computer architecture due to the multicore paradigm shift, as every electronic device from cell phones to supercomputers confronts parallelism of unprecedented scale. To fully unleash the potential of these systems, the HPC community must develop multicore specific-optimization methodologies for important scientific computations. In this work, we examine sparse matrix-vector multiply (SpMV) - one of the most heavily used kernels in scientific computing - across a broad spectrum of multicore designs. Our experimental platform includes the homogeneous AMD quad-core, AMD dual-core, and Intel quad-core designs, the heterogeneous STI Cell, as well as one of the first scientific studies of the highly multithreaded Sun Victoria Falls (a Niagara2 SMP). We present several optimization strategies especially effective for the multicore environment, and demonstrate significant performance improvements compared to existing state-of-the-art serial and parallel SpMV implementations. Additionally, we present key insights into the architectural trade-offs of leading multicore design strategies, in the context of demanding memory-bound numerical algorithms.

97 MATHEMATICS AND COMPUTING↗

A Combination of Structural and Empirical Analyses Delineates the Key Contacts Mediating Stability and Affinity Increases in an Optimized Biotherapeutic Single-chain Fv (scFv)

Fully-human single-chain Fv (scFv) proteins are key potential building blocks of bispecific therapeutic antibodies, but they often suffer from manufacturability and clinical development limitations such as instability and aggregation. Here, the causes of these scFv instability problems, in proteins that should be theoretically stable, remains poorly understood. To inform the future development of such molecules, we carried out a comprehensive structural analysis of the highly stabilized anti-CXCL13 scFv E10. E10 was derived from the parental 3B4 using complementarity-determining region (CDR)-restricted mutagenesis and tailored selection and screening strategies, and carries four mutations in V L -CDR3. High-resolution crystal structures of parental 3B4 and optimized E10 scFvs were solved in the presence and absence of human CXCL13. In parallel, a series of scFv mutants was generated to interrogate the individual contribution of each of the four mutations to stability and affinity improvements. In combination, these analyses demonstrated that the optimization of E10 was primarily mediated by removing clashes between both the V L and the V H , and between the V L and CXCL13. Importantly, a single, germline-encoded V L -CDR3 residue mediated the key difference between the stable and unstable forms of the scFv. This work demonstrates that, aside from being the critical mediators of specificity and affinity, CDRs may also be the primary drivers of biotherapeutic developability.

59 BASIC BIOLOGICAL SCIENCES↗

A connectionist model for diagnostic problem solving

A competition-based connectionist model for solving diagnostic problems is described. The problems considered are computationally difficult in that (1) multiple disorders may occur simultaneously and (2) a global optimum in the space exponential to the total number of possible disorders is sought as a solution. The diagnostic problem is treated as a nonlinear optimization problem, and global optimization criteria are decomposed into local criteria governing node activation updating in the connectionist model. Nodes representing disorders compete with each other to account for each individual manifestation, yet complement each other to account for all manifestations through parallel node interactions. When equilibrium is reached, the network settles into a locally optimal state. Three randomly generated examples of diagnostic problems, each of which has 1024 cases, were tested, and the decomposition plus competition plus resettling approach yielded very high accuracy.

Peng, Yun↗

A Generalized Stress Inversion Approach With Application to Residual Stress Estimation

In this work develop a generalized stress inversion technique (or the generalized inversion method) capable of recovering stresses in linear elastic bodies subjected to arbitrary cuts. Specifically, given a set of displacement measurements found experimentally from digital image correlation (DIC), we formulate a stress estimation inverse problem as a partial differential equation-constrained optimization problem. We use gradient-based optimization methods, and we accordingly derive the necessary gradient and Hessian information in a matrix-free form to allow for parallel, large-scale operations. By using a combination of finite elements, DIC, and a matrix-free optimization framework, the generalized inversion method can be used on any arbitrary geometry, provided that the DIC camera can view a sufficient part of the surface. We present numerical simulations and experiments, and we demonstrate that the generalized inversion method can be applied to estimate residual stress.

42 ENGINEERING↗

3-D Semantic Information Inference from Airborne Video. Final Report

We developed a full 3D modeling pipeline focusing on an important application scenario: city-scale 3D reconstruction from aerial imagery. We propose an approach to solve camera pose estimation and dense reconstruction from Wide Area Aerial Surveillance (WAAS) videos captured by an airborne platform. Our approach solves them in an online fashion: it incrementally updates a sparse 3D map and estimates the camera pose as each new frame arrives; depth maps of selected key frames are computed using a variational method and integrated to produce a full 3D model via volumetric reconstruction. In practice, aerial imagery is usually captured using a multicamera system. We propose an approach for camera pose estimation of multi-camera aerial imagery which is parallelized on multiple GPUs for efficiency. The approach is also extended for progressive 3D model acquisition with a hand-held camera. We also developed an offline approach since online approach is not a necessity and accuracy has higher priority over efficiency in many scenarios. We presented MeshRecon, a mesh-based offline system composed of three modules: a dense point cloud is generated using multi-resolution plane sweep method; an initial mesh model is extracted from the point cloud via global optimization considering visibility information of all images; the mesh model is then iteratively refined to capture structural details by optimizing the photometric consistency and spatial regularization. The major processes are parallelized on GPU for efficiency. For the aerial imagery case, we evaluate our system on several real-world multicamera aerial imagery datasets, each covering an urban scenario of several square kilometers. Quantitative result shows that the reconstructed geometric 3D model is highly accurate with error smaller than 1 meter over the entire city. Besides aerial imagery, we also evaluate its performance on general geometric 3D model acquisition of real-world objects. Result shows that the system is robust and flexible for various types of objects at different scales in both indoor and outdoor environments. Based on city 3D models reconstructed at different times, we present a system for city-scale geometry change detection by performing comparisons at the 3D geometry level. Our system is able to detect geometry changes at different scales, ranging from a building cluster to small-scale vegetation changes, with high accuracy.

97 MATHEMATICS AND COMPUTING↗

Computational studies and optimization of wakefield accelerators

Laser- and particle beam-driven plasma wakefield accelerators produce accelerating fields thousands of times higher than radio-frequency accelerators, offering compactness and ultrafast bunches to extend the frontiers of high energy physics and to enable laboratory-scale radiation sources. Large-scale kinetic simulations provide essential understanding of accelerator physics to advance beam performance and stability and show and predict the physics behind recent demonstration of narrow energy spread bunches. Benchmarking between codes is establishing validity of the models used and, by testing new reduced models, is extending the reach of simulations to cover upcoming meter-scale multi-GeV experiments. This includes new models that exploit Lorentz boosted simulation frames to speed calculations. Simulations of experiments showed that recently demonstrated plasma gradient injection of electrons can be used as an injector to increase beam quality by orders of magnitude. Simulations are now also modeling accelerator stages of tens of GeV, staging of modules, and new positron sources to design next-generation experiments and to use in applications in high energy physics and light sources.

43 PARTICLE ACCELERATORS↗

Fast and Accurate Support Vector Machines on Large Scale Systems

Support Vector Machines (SVM) is a supervised Machine Learning and Data Mining (MLDM) algorithm, which has become ubiquitous largely due to its high accuracy and obliviousness to dimensionality. The objective of SVM is to find an optimal boundary --- also known as hyperplane --- which separates the samples (examples in a dataset) of different classes by a maximum margin. Usually, very few samples contribute to the definition of the boundary. However, existing parallel algorithms use the entire dataset for finding the boundary, which is sub-optimal for performance reasons. In this paper, we propose a novel distributed memory algorithm to eliminate the samples which do not contribute to the boundary definition in SVM. We propose several heuristics, which range from early (aggressive) to late (conservative) elimination of the samples, such that the overall time for generating the boundary is reduced considerably. In a few cases, a sample may be eliminated (shrunk) pre-emptively --- potentially resulting in an incorrect boundary. We propose a scalable approach to synchronize the necessary data structures such that the proposed algorithm maintains its accuracy. We consider the necessary trade-offs of single/multiple synchronization using in-depth time-space complexity analysis. We implement the proposed algorithm using MPI and compare it with libsvm--- de facto sequential SVM software --- which we enhance with OpenMP for multi-core/many-core parallelism. Our proposed approach shows excellent efficiency using up to 4096 processes on several large datasets such as UCI HIGGS Boson dataset and Offending URL dataset.

Vishnu, Abhinav↗

High-Level Synthesis of Parallel Specifications Coupling Static and Dynamic Controllers

The increased need for efficient ways to implement domain-specific accelerators is driving design methodologies towards the use of abstractions higher than the Register Transfer Level (RTL). In this scenario, High Level Synthesis (HLS) plays a significant role by enabling the automatic generation of custom hardware accelerators starting from high level descriptions (e.g., C code). Conventional HLS tools exploit parallelism mostly at the Instruction Level (ILP). They statically schedule the input specifications, and build centralized Finite State Machine (FSM) controllers. However, aggressive exploitation of ILP in many applications has diminishing returns and, usually, centralized approaches do not efficiently exploit coarser parallelism because FSMs are inherently serial. In this paper we present a HLS framework able to synthesize applications that, beside ILP, also expose Task Level Parallelism (TLP). An application can expose TLP through annotations that identify the parallel functions (i.e., tasks). To generate accelerators that efficiently execute concur- rent tasks, we need to solve several issues: devise a mechanism to support concurrent execution flows, exploit memory parallelism, and manage synchronization. To support concurrent execution flows, we introduce a novel adaptive controller. The adaptive controller is composed of a set of interacting control elements that independently manage the execution of a single operation or function call. These control elements check dependencies and resource constraints at runtime, enabling as soon as possible execution. To support parallel access to shared memories and synchronization, we introduce a novel Hierarchical Memory Interface (HMI). With respect to previous solutions, the proposed interface supports multi-ported memories and atomic memory operations, which commonly occur in parallel programming. Our framework can generate the hardware implementation of C functions by employing two different approaches, depending on its characteristics. If a function exposes TLP, then the framework generates hardware implementations based on the adaptive controller. Otherwise, the framework implements the function by exploiting a more conventional FSM approach, which is optimized for ILP exploitation. We evaluate our framework on a set of parallel applications, and show substantial performance improvements (average speedup of 4.7) with limited area over- heads (average area increase of 5.48 times).

Castellana, Vito G.↗

High-Level Synthesis of Parallel Specifications Coupling Static and Dynamic Controllers

The increased need for efficient ways to implement domain-specific accelerators is driving design methodologies towards the use of abstractions higher than the Register Transfer Level (RTL). In this scenario, High Level Synthesis (HLS) plays a significant role by enabling the automatic generation of custom hardware accelerators starting from high level descriptions (e.g., C code). Conventional HLS tools exploit parallelism mostly at the Instruction Level (ILP). They statically schedule the input specifications, and build centralized Finite State Machine (FSM) controllers. However, aggressive exploitation of ILP in many applications has diminishing returns and, usually, centralized approaches do not efficiently exploit coarser parallelism because FSMs are inherently serial. In this paper we present a HLS framework able to synthesize applications that, beside ILP, also expose Task Level Parallelism (TLP). An application can expose TLP through annotations that identify the parallel functions (i.e., tasks). To generate accelerators that efficiently execute concur- rent tasks, we need to solve several issues: devise a mechanism to support concurrent execution flows, exploit memory parallelism, and manage synchronization. To support concurrent execution flows, we introduce a novel adaptive controller. The adaptive controller is composed of a set of interacting control elements that independently manage the execution of a single operation or function call. These control elements check dependencies and resource constraints at runtime, enabling as soon as possible execution. To support parallel access to shared memories and synchronization, we introduce a novel Hierarchical Memory Interface (HMI). With respect to previous solutions, the proposed interface supports multi-ported memories and atomic memory operations, which commonly occur in parallel programming. Our framework can generate the hardware implementation of C functions by employing two different approaches, depending on its characteristics. If a function exposes TLP, then the framework generates hardware implementations based on the adaptive controller. Otherwise, the framework implements the function by exploiting a more conventional FSM approach, which is optimized for ILP exploitation. We evaluate our framework on a set of parallel applications, and show substantial performance improvements (average speedup of 4.7) with limited area over- heads (average area increase of 5.48 times).

Castellana, Vito G.↗

Optimizing the Performance of Reactive Molecular Dynamics Simulations for Multi-core Architectures

Hybrid parallelism allows high performance computing applications to better leverage the increasing on-node parallelism of modern supercomputers. In this paper, we present a hybrid parallel implementation of the widely used LAMMPS/ReaxC package, where the construction of bonded and nonbonded lists and evaluation of complex ReaxFF interactions are implemented efficiently using OpenMP parallelism. Additionally, the performance of the QEq charge equilibration scheme is examined and a dual-solver is implemented. We present the performance of the resulting ReaxC-OMP package on a state-of-the-art multi-core architecture Mira, an IBM BlueGene/Q supercomputer. For system sizes ranging from 32 thousand to 16.6 million particles, speedups in the range of 1.5-4.5x are observed using the new ReaxC-OMP software. Sustained performance improvements have been observed for up to 262,144 cores (1,048,576 processes) of Mira with a weak scaling efficiency of 91.5% in larger simulations containing 16.6 million particles.

Aktulga, Hasan Metin↗

Multigrid Reduction in Time for Chaotic and Hyperbolic Problems (Final Report)

The coming massive parallelism of exascale computing presents a pressing challenge for the many DOE simulations of time-dependent partial differential equations (PDEs), which typically use traditional sequential time stepping methods. Since this traditional approach is inherently serial, it presents a sequential bottleneck when moving to exascale computing, because future performance gains will come through greater concurrency, not faster clock speeds. Thus, the goal of this work is to research parallelism in time, i.e., methods that compute multiple time values simultaneously, not sequentially. The focus will be on hyperbolic and chaotic problems of interest to DOE, with the goal of enabling scalable simulations of time-dependent hyperbolic and chaotic problems on future architectures. The chosen methodology for solving these problems parallel-in-time is multigrid, because multigrid (when it works) is a powerful, optimal, and scalable solver for discretized PDEs. Multigrid is already commonly used in many DOE simulations for scalably and optimally solving space-only PDE problems. The areas of hyperbolic and chaotic problems are chosen because of their relevance to problems of programmatic interest to DOE. However, these problems are also well-known to be difficult for parallel-in-time methods, with the most common method, parareal, diverging in many cases. The current state of-the-art for parallel-in-time at LLNL is the multigrid reduction in time (MGRIT) XBraid package, which also struggles for such problems, while still showing some improvement over parareal. In summary, new methods are needed for an efficient parallel-in-time scheme for hyperbolic and chaotic problems, and this work shall research promising new multigrid methods in this area. In particular, this work shall continue researching the directions from the current collaboration with Dr. Falgout, which are laid out in the work Toward Parallel in Time for Chaotic Dynamical Systems and showed the first known results of a parallel-in-time speedup for a chaotic problem. This work outlines two key improvements to XBraid for chaotic problems, the so-called “theta” and “delta-correction” methods. Here, these two improvements will be further researched and improved (including with a new relaxation method inspired by on Least Squares Shadowing (LSS)) and explored for more complicated problems.

97 MATHEMATICS AND COMPUTING↗

Optimizing inference of segmentation on high-resolution images in MLExchange

MLExchange is a machine learning (ML) operations platform providing web user-interfaces (UIs) for data visualization and analysis pipelines at synchrotron facilities. Among these UIs is the segmentation app which helps synchrotron users utilize ML algorithms to automatically segment high-resolution scientific images with minimal manual annotation effort. In this work, we share code optimizations that significantly speed up the segmentation inference workflow of large data in short time. By optimizing the sequence of CPU-GPU data transfers and introducing CPU parallelization to key operations, we improve the per-device, per-image frame computational efficiency and observe close to 3×$$\times$$ speedup over the original segmentation inference workflow run time when utilizing a single GPU. Further adaptations enabling multi-GPU inference yield more than 40×$$\times$$ speedup with 100 GPUs compared to the optimized single GPU inference workflow. This acceleration of the segmentation inference workflow will provide MLExchange users with easy access to segmentation results with little wait time.

Lu, Shizhao↗

The cost of conservative synchronization in parallel discrete event simulations

The performance of a synchronous conservative parallel discrete-event simulation protocol is analyzed. The class of simulation models considered is oriented around a physical domain and possesses a limited ability to predict future behavior. A stochastic model is used to show that as the volume of simulation activity in the model increases relative to a fixed architecture, the complexity of the average per-event overhead due to synchronization, event list manipulation, lookahead calculations, and processor idle time approach the complexity of the average per-event overhead of a serial simulation. The method is therefore within a constant factor of optimal. The analysis demonstrates that on large problems--those for which parallel processing is ideally suited--there is often enough parallel workload so that processors are not usually idle. The viability of the method is also demonstrated empirically, showing how good performance is achieved on large problems using a thirty-two node Intel iPSC/2 distributed memory multiprocessor.

Nicol, David M.↗

Self-optimized construction of transition rate matrices from accelerated atomistic simulations with Bayesian uncertainty quantification

Here, a massively parallel method to build large transition rate matrices from temperature-accelerated molecular dynamics trajectories is presented. Bayesian Markov model analysis is used to estimate the expected residence time in the known state space, providing crucial uncertainty quantification for higher-scale simulation schemes such as kinetic Monte Carlo or cluster dynamics. The estimators are additionally used to optimize where exploration is performed and the degree of temperature acceleration on the fly, giving an autonomous, optimal procedure to explore the state space of complex systems. The method is tested against exactly solvable models and used to explore the dynamics of C15 interstitial defects in iron. Our uncertainty quantification scheme allows for accurate modeling of the evolution of these defects over timescales of several seconds.

36 MATERIALS SCIENCE↗

Adding GPU Support to the Markov Chain Monte Carlo Code Catmip

In geophysics, we are confronted with many under-determined inverse problems. For example, all of our observations of earthquakes are made at the Earth’s surface. So, when we try to infer how slip during an earthquake evolves in space and time, we find that there are many potential slip histories that are consistent with our limited observations and our understanding of earthquake physics. One way to approach these problems is with Bayesian analysis which allows us to infer the ensemble of all potential slip models that satisfy the observations and our prior knowledge of earthquake physics. In Bayesian analysis, our prior knowledge is known as the prior probability density function or prior PDF, the fit to the data is known as the data likelihood, and the target PDF that satisfies both the prior PDF and data likelihood is known as the posterior PDF. However, simulating the posterior PDF typically requires using Markov Chain Monte Carlo (MCMC) to draw tens of billions of random realizations of earthquake slip models, which may not be computationally feasible. To make this and similar geophysical inversions computationally tractable, we developed the Cascading Adaptive Transitional Metropolis In Parallel (CATMIP) algorithm. CATMIP is an efficient parallel Markov Chain Monte Carlo (MCMC) sampler that is used for model fitting and uncertainty quantification in geophysics. Example use cases are earthquake rupture modeling, determining mineral composition on Mars, reconstructing the history of ocean salinity, and historical earthquake relocation. CATMIP employs many parallel instances of the Metropolis algorithm for sampling in a transitioning framework. Transitioning is a process in which a set of random samples at equilibrium with a known probability density function (PDF) are used as seeds for the Markov chains to sample successive target PDFs that incrementally move the distribution from the starting seeds to the final desired PDF that describes the relative plausibility of potential values for the model parameters. The algorithm is implemented as a Master-Worker model employing MPI for communication. The worker processes are loosely coupled with global parameters periodically optimized by the master process. This provides a very high amount of parallelism with little communication between updates. During the presentation we will discuss the history of the algorithm and elaborate the earthquake rupture modeling use case for the CATMIP package. Our first step toward GPU optimization was to optimize the code for the CPU. CPU profiling revealed that most of the compute time is spent in calls to level 2 BLAS routines and calls to GSL random number generators. We revised the algorithm to employ level 3 BLAS routines instead. In our presentation we will describe how this was accomplished. Adding GPU support to CATMIP consisted mostly of replacing the calls to GSL with calls to GPU vendor-provided library routines. A small number of loops were directly implemented in CUDA. In the presentation will provide implementation details. Finally, we will discuss methods for profiling and opportunities for further optimizing GPU execution. By creating a code with the flexibility to run on either a CPU or GPU architecture, CATMIP can be used on systems ranging from large CPU-based HPC environments to single servers with GPU acceleration and everything in between.

HECC↗