SEARCH · Engineering Papers
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.
R-stream release 3.22.3 including the Glow and TVM front-ends, and the features developed in the Portacy 2B, X-MAT and STAPL projects. R-Stream is an automatic parallelization, source-to-source compiler. The TVM and Glow front-ends to R-Stream were developed to automatically map neural net models to parallel computers using openMP and CUDA (for Glow).
This paper presents a methodology for using LLVM-based tools to tune the DCA++ (dynamical cluster approximation) application that targets the new ARM A64FX processor. The goal is to describe the changes required for the new architecture and generate efficient single instruction/multiple data (SIMD) instructions that target the new Scalable Vector Extension instruction set. During manual tuning, the authors used the LLVM tools to improve code parallelization by using OpenMP SIMD, refactored the code and applied transformation that enabled SIMD optimizations, and ensured that the correct libraries were used to achieve optimal performance. By applying these code changes, code speed was increased by 1.98× and 78 GFlops were achieved on the A64FX processor. The authors aim to automatize parts of the efforts in the OpenMP Advisor tool, which is built on top of existing and newly introduced LLVM tooling.
The design of distributed shared memory (DSM) computers liberates users from the duty to distribute data across processors and allows for the incremental development of parallel programs using, for example, OpenMP or Java threads. DSM architecture greatly simplifies the development of parallel programs having good performance on a few processors. However, to achieve a good program scalability on DSM computers requires that the user understand data flow in the application and use various techniques to avoid data traffic congestions. In this paper we discuss a number of such techniques, including data blocking, data placement, data transposition and page size control and evaluate their efficiency on the NAS (NASA Advanced Supercomputing) Parallel Benchmarks. We also present a tool which automates the detection of constructs causing data congestions in Fortran array oriented codes and advises the user on code transformations for improving data traffic in the application.
With the appearance of multi-/many core machines, applications and runtime systems have evolved in order to exploit the new on-node concurrency brought by new software paradigms. POSIX threads (Pthreads) was widely-adopted for that purpose and it remains as the most used threading solution in current hardware. Lightweight thread (LWT) libraries emerged as an alternative offering lighter mechanisms to tackle the massive concurrency of current hardware. In this article, we analyze in detail the most representative threading libraries including Pthread- and LWT-based solutions. In addition, to examine the suitability of LWTs for different use cases, we develop a set of microbenchmarks consisting of OpenMP patterns commonly found in current parallel codes, and we compare the results using threading libraries and OpenMP implementations. Moreover, we study the semantics offered by threading libraries in order to expose the similarities among different LWT application programming interfaces and their advantages over Pthreads. This article exposes that LWT libraries outperform solutions based on operating system threads when tasks and nested parallelism are required.
The development of a parallel blade-element rotor model and its implementation into an adaptive Cartesian method is described. The unsteady version of the rotor model applies a body force to all cells contained in the swept space-time volume at each timestep and special care is taken to maintain axisymmetry on the Cartesian grid. Mesh convergence of rotor thrust and torque is obtained with around 10000 cells in the disk for the steady model. Parallelization is accomplished using OpenMP and the rotor force computation is distributed across all available nodes. Simulations of an isolated XV-15 rotor in hover show good correlation with experimental data and predictions of multi-rotor thrust variation closely match previous high fidelity simulations. The final paper will also include results from the unsteady rotor model and parallel scaling tests.
In this work, a parallel algorithm is described for the coupled-cluster singles and doubles method augmented with a perturbative correction for triple excitations [CCSD(T)] using the resolution-of-the-identity (RI) approximation for two-electron repulsion integrals (ERIs). The algorithm bypasses the storage of four-center ERIs by adopting an integral-direct strategy. The CCSD amplitude equations are given in a compact quasi-linear form by factorizing them in terms of amplitude-dressed three-center intermediates. A hybrid MPI/OpenMP parallelization scheme is employed, which uses the OpenMP-based shared memory model for intranode parallelization and the MPI-based distributed memory model for internode parallelization. Parallel efficiency has been optimized for all terms in the CCSD amplitude equations. Two different algorithms have been implemented for the rate-limiting terms in the CCSD amplitude equations that entail and -scaling computational costs, where N O and N V denote the number of correlated occupied and virtual orbitals, respectively. One of the algorithms assembles the four-center ERIs requiring N V 4 and N O 2 N V 2 -scaling memory costs in a distributed manner on a number of MPI ranks, while the other algorithm completely bypasses the assembling of quartic memory-scaling ERIs and thus largely reduces the memory demand. It is demonstrated that the former memory-expensive algorithm is faster on a few hundred cores, while the latter memory-economic algorithm shows a better strong scaling in the limit of a few thousand cores. The program is shown to exhibit a near-linear scaling, in particular for the compute-intensive triples correction step, on up to 8000 cores. The performance of the program is demonstrated via calculations involving molecules with 24–51 atoms and up to 1624 atomic basis functions. As the first application, the complete basis set (CBS) limit for the interaction energy of the π-stacked uracil dimer from the S66 data set has been investigated. This work reports the first calculation of the interaction energy at the CCSD(T)/aug-cc-pVQZ level without local orbital approximation. The CBS limit for the CCSD correlation contribution to the interaction energy was found to be -8.01 kcal/mol, which agrees very well with the value -7.99 kcal/mol reported by Schmitz, Hättig, and Tew [ Phys. Chem. Chem. Phys. 2014 , 16 , 22167-22178]. The CBS limit for the total interaction energy was estimated to be -9.64 kcal/mol.
The shared-memory programming model is a very effective way to achieve parallelism on shared memory parallel computers. Historically, the lack of a programming standard for using directives and the rather limited performance due to scalability have affected the take-up of this programming model approach. Significant progress has been made in hardware and software technologies, as a result the performance of parallel programs with compiler directives has also made improvements. The introduction of an industrial standard for shared-memory programming with directives, OpenMP, has also addressed the issue of portability. In this study, we have extended the computer aided parallelization toolkit (developed at the University of Greenwich), to automatically generate OpenMP based parallel programs with nominal user assistance. We outline the way in which loop types are categorized and how efficient OpenMP directives can be defined and placed using the in-depth interprocedural analysis that is carried out by the toolkit. We also discuss the application of the toolkit on the NAS Parallel Benchmarks and a number of real-world application codes. This work not only demonstrates the great potential of using the toolkit to quickly parallelize serial programs but also the good performance achievable on up to 300 processors for hybrid message passing and directive-based parallelizations.
Software developers must adapt to keep up with the changing capabilities of platforms so that they can utilize the power of High-Performance Computers (HPC), including exascale systems. OpenMP, a directive-based parallel programming model, allows developers to include directives to existing C, C++, or Fortran code to allow node level parallelism without compromising performance. This paper describes our CI/CD efforts to provide easy evaluation of the support of OpenMP across different compilers using existing testsuites and benchmark suites on HPC platforms. Our main contributions include (1) the set of a Continuous Integration (CI) and Continuous Development (CD) workflow that captures bugs and provides faster feedback to compiler developers, (2) an evaluation of OpenMP (offloading) implementations supported by AMD, HPE, GNU, LLVM, and Intel, and (3) evaluation of the quality of compilers across different heterogeneous HPC platforms. With the comprehensive testing through the CI/CD workflow, we aim to provide a comprehensive understanding of the current state of OpenMP (offloading) support in different compilers and heterogeneous platforms consisting of CPUs and GPUs from NVIDIA, AMD, and Intel.
The HPC industry is inexorably moving towards an era of extremely heterogeneous architectures, with more devices configured on any given HPC platform and potentially more kinds of devices, some of them highly specialized. Writing a separate code suitable for each target system for a given HPC application is not practical. The better solution is to use directive-based parallel programming models such as OpenMP. OpenMP provides a number of options for offloading a piece of code to devices like GPUs. To select the best option from such options during compilation, most modern compilers use analytical models to estimate the cost of executing the original code and the different offloading code variants. Building such an analytical model for compilers is a difficult task that necessitates a lot of effort on the part of a compiler engineer. Recently, machine learning techniques have been successfully applied to build cost models for a variety of compiler optimization problems. In this paper, we present COMPOFF, a cost model which uses the multi-layer perceptrons to statically estimates the Cost of OpenMP OFFloading. We used six different transformations on a parallel code of Wilson Dslash Operator to support GPU offloading, and we predicted their cost of execution on different GPUs using COMPOFF during compile time. Our results show that this model can predict offloading costs with a root mean squared error in prediction of less than 0.5 seconds. Our preliminary findings indicate that this work will make it much easier and faster for scientists and compiler developers to port legacy HPC applications that use OpenMP to new heterogeneous computing environment.
The PRIMA-X (Performance Retargeting of Instrumentation, Measurement, and Analysis Technologies for Exascale Computing) project is the successor of the DOE PRIMA (Performance Refactoring of Instrumentation, Measurement, and Analysis Technologies for Petascale Computing) project, which addressed the challenge of creating a core measurement infrastructure that would serve as a common platform for both integrating leading parallel performance systems (notably TAU and Scalasca) and developing next-generation scalable performance tools. The PRIMA-X project shifts the focus away from refactorization of robust performance tools towards a re-targeting of the parallel performance measurement and analysis architecture for extreme scales. The massive concurrency, asynchronous execution dynamics, hardware heterogeneity, and multi-objective prerequisites (performance, power, resilience) that identify exascale systems introduce fundamental constraints on the ability to carry forward existing performance methodologies. In particular, there must be a deemphasis of per-thread observation techniques to significantly reduce the otherwise unsustainable flood of redundant performance data. Instead, it will be necessary to assimilate multi-level resource observations into macroscopic performance views, from which resilient performance metrics can be attributed to the computational features of the application. This requires a scalable framework for node-level and system-wide monitoring and runtime analyses of dynamic performance information. Also, the interest in optimizing parallelism parameters with respect to performance and energy drives the integration of tool capabilities in the exascale environment further. Initially, PRIMA-X was a collaborative project between the University of Oregon (lead institution) and the German Research School for Simulation Sciences (GRS). Because Prof. Wolf, the PI at GRS, accepted a position as full professor at Technische Universitaet Darmstadt (TU Darmstadt) starting February 1st, 2015, the work of GRS was continued at TU Darmstadt. This report covers the work at TU Darmstadt after the transition. The first main accomplishment of TU Darmstadt is the development of different techniques to aggregate performance data on the level of threads. TU Darmstadt evaluated different schemes designed at GRS during the first phase of this project and integrated them into Score-P, a widely used performance measurement framework for HPC applications. The second main accomplishment is a substantial increase of Score-P’s scalability, achieved by improving its internal representation of the underlying system. Third, we extended our performance-modeling tool Extra-P to automatically create models with more than one input parameter (e.g., process count and input size), which was previously impossible. These models allow the performance behavior of an application to be extrapolated to larger machines. Specific applications of the new multi-parameter capability realized in this project include modeling the isoefficiency function of task-based applications and a novel co-design methodology. Further extensions of Extra-P comprise online modeling functionality, automatic refinement of the model search space, and segmented performance modeling. Further contributions include new locality metrics for both shared- and distributed-memory applications, a new way of measuring the impact of inter-application inference on the execution time of an application, and finally an automatic method that selects appropriate OpenMP constructs for the parallelization of sequential code and classifies the affected variables based on their sharing semantics.
The shared-memory programming model is a very effective way to achieve parallelism on shared memory parallel computers. As great progress was made in hardware and software technologies, performance of parallel programs with compiler directives has demonstrated large improvement. The introduction of OpenMP directives, the industrial standard for shared-memory programming, has minimized the issue of portability. In this study, we have extended CAPTools, a computer-aided parallelization toolkit, to automatically generate OpenMP-based parallel programs with nominal user assistance. We outline techniques used in the implementation of the tool and discuss the application of this tool on the NAS Parallel Benchmarks and several computational fluid dynamics codes. This work demonstrates the great potential of using the tool to quickly port parallel programs and also achieve good performance that exceeds some of the commercial tools.
Label Propagation is not only a well-known machine learning algorithm for classification but also an effective method for discovering communities and connected components in networks. We propose a new Direction-optimizing Label Propagation Algorithm (DOLPA) framework that enhances the performance of the standard Label Propagation Algorithm (LPA), increases its scalability, and extends its versatility and application scope. As a central feature, the DOLPA framework relies on the use of frontiers and alternates between label push and label pull operations to attain high performance. It is formulated in such a way that the same basic algorithm can be used for finding communities or connected components in graphs by only changing the objective function used. Additionally, DOLPA has parameters for tuning the processing order of vertices in a graph to reduce the number of edges visited and improve the quality of solution obtained. We present the design and implementation of the enhanced algorithm as well as our shared-memory parallelization of it using OpenMP. We also present an extensive experimental evaluation of our implementations using the LFR benchmark and real-world networks drawn from various domains. Compared with an implementation of LPA for community detection available in a widely used network analysis software, we achieve at most five times the F-Score while maintaining similar runtime for graphs with overlapping communities. We also compare DOLPA against an implementation of the Louvain method for community detection using the same LFR-graphs and show that DOLPA achieves about three times the F-Score at just 10% of the runtime. For connected component decomposition, our algorithm achieves orders of magnitude speedups over the basic LP-based algorithm on large-diameter graphs, up to 13.2× speedup over the Shiloach-Vishkin algorithm, and up to 1.6× speedup over Afforest on an Intel Xeon processor using 40 threads.
Originally used Spark and HadoopFS. Collected interesting results, but this method had its issues: limited application, too much overhead to gauge CSDs’ raw performance. Solution? Rewrite our benchmarks without Spark using Serial Python and Serial & Parallel C++ (Combinations of OpenMP & OpenMPI). Compared to Spark and Python, C++ implementation is a lot faster. Caveat: an expert with Spark or Python would likely be able to improve the performance of those implementations. Computational power of our CSDs seem to be much lower than the host machine. Using all 4 cores of a single CSD, the job takes ~6.8x longer than using just one core on the host machine. Host also seems to scale better with increasing file size. Resulting Question: When, if ever, would it make sense to use CSDs for compute rather than a much-faster host?
In this poster, a GPU-accelerated sparse multifrontal solver for structurally symmetric matrices is described. The implementation is tested on the Summit supercomputer against the current version, which is parallelized via MPI and OpenMP on CPUs. The GPU-accelerated implementation achieves significant speedup from the original code.
Monte Carlo N-Particle (MCNP)1 is a general-purpose Monte Carlo particle transport code developed by Los Alamos National Laboratory (LANL). To efficiently handle long simulations, MCNP version 6 (MCNP6) supports parallel execution using two primary programming models: • Shared-memory task-based threading using OpenMP (Open Multi-Processing), and • Distributed-memory calculations using MPI (Message Passing Interface). The OpenMP and MPI programming models enable MCNP6 to scale from desktop systems to high-performance computing (HPC) clusters, allowing users to run MCNP in one of three parallel modes: • OpenMP-only, • MPI-only, and • Hybrid (MPI + OpenMP). The choice of parallelization mode depends on the underlying computer architecture and the characteristics of the simulation problem.
This paper presents software advances to easily exploit computer architectures consisting of a multi-core CPU and CPU+GPU to accelerate diverse types of high-performance computing (HPC) applications using a single code implementation. The paper describes and demonstrates the performance of the open-source C++ matrix and array (MATAR) library that uniquely offers: (1) a straightforward syntax for programming productivity, (2) usable data structures for data-oriented programming (DOP) for performance, and (3) a simple interface to the open-source C++ Kokkos library for portability and memory management across CPUs and GPUs. The portability across architectures with a single code implementation is achieved by automatically switching between diverse fine-grained parallelism backends (e.g., CUDA, HIP, OpenMP, pthreads, etc.) at compile time. The MATAR library solves many longstanding challenges associated with easily writing software that can run in parallel on any computer architecture. This work benefits projects seeking to write new C++ codes while also addressing the challenges of quickly making existing Fortran codes performant and portable over modern computer architectures with minimal syntactical changes from Fortran to C++. We demonstrate the feasibility of readily writing new C++ codes and modernizing existing codes with MATAR to be performant, parallel, and portable across diverse computer architectures.
Parallel adaptive mesh refinement (AMR) is an important numerical technique that leads to the efficient solution of many physical and engineering problems.
Currently, the Weather Research and Forecasting model (WRF) utilizes shared memory (OpenMP) and distributed memory (MPI) parallelisms. To take advantage of GPU resources on the Perlmutter supercomputer at NERSC, we port parts of the computationally expensive routine Fast Spectral Bin Microphysics (FSBM) to NVIDIA GPUs using OpenMP device offloading directives. To facilitate this process, we explore a workflow for optimization which uses both runtime profilers and a static code inspection tool Codee to refactor the subroutine. We observe an 2.24x overall speedup for the CONUS-12km storm test case.