A Case Study of LLVM-based Analysis for Optimizing SIMD Code Generation
Explore the source record for details and available documents.
SEARCH · Search NASA
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.
Explore the source record for details and available documents.
High-energy physics (HEP) experiments have developed millions of lines of code over decades that are optimized to run on traditional x86 CPU systems. However, we are seeing a rapidly increasing fraction of floating point computing power in leadership-class computing facilities and traditional data centers coming from new accelerator architectures, such as GPUs. HEP experiments are now faced with the untenable prospect of rewriting millions of lines of x86 CPU code, for the increasingly dominant architectures found in these computational accelerators. This task is made more challenging by the architecture-specific languages and APIs promoted by manufacturers such as NVIDIA, Intel and AMD. Producing multiple, architecture-specific implementations is not a viable scenario, given the available person power and code maintenance issues. The Portable Parallelization Strategies team of the HEP Center for Computational Excellence is investigating the use of Kokkos, SYCL, OpenMP, std::execution::parallel and alpaka as potential portability solutions that promise to execute on multiple architectures from the same source code, using representative use cases from major HEP experiments, including the DUNE experiment of the Long Baseline Neutrino Facility, and the ATLAS and CMS experiments of the Large Hadron Collider. This cross-cutting evaluation of portability solutions using real applications will help inform and guide the HEP community when choosing their software and hardware suites for the next generation of experimental frameworks. We present the outcomes of our studies, including performance metrics, porting challenges, API evaluations, and build system integration.
An essential part of burnup analysis is to solve the burnup equations. The burnup equations can be regarded as a first-order linear system and solved by means of matrix exponential methods. Because of its large spectrum, it is difficult to compute the exponential of the burnup matrix. Conventional methods of computing the matrix exponential, such as the truncated Taylor expansion and the Pade approximation, are not applicable to burnup calculations. Recently the Chebyshev Rational Approximation Method (CRAM) has been applied to solve burnup matrix exponential and shown to be robust and accurate. However, the main defect of CRAM is that its coefficients are not easy to obtain. In this paper, an orthogonal polynomial expansion method, called Laguerre Polynomial Approximation Method (LPAM), is proposed to compute the matrix exponential in burnup calculations. The polynomial sequence of LPAM can be easily computed in any order and thus LPAM is quite convenient to be utilized into burnup codes. Two typical test cases with the decay and cross-section data taken from the standard ORIGEN 2.1 libraries are calculated for validation, against the reference results provided by CRAM of 14 order. Numerical results show that, LPAM is sufficiently accurate for burnup calculations. The influences of the parameters on the convergence of LPAM are also discussed. (authors)
The International Energy Agency Technology Collaboration Programme for Ocean Energy Systems (OES) initiated the OES Wave Energy Conversion Modelling Task, which focused on the verification and validation of numerical models for simulating wave energy converters (WECs). The long-term goal is to assess the accuracy of and establish confidence in the use of numerical models used in design as well as power performance assessment of WECs. To establish this confidence, the authors used different existing computational modelling tools to simulate given tasks to identify uncertainties related to simulation methodologies: (i) linear potential flow methods; (ii) weakly nonlinear Froude–Krylov methods; and (iii) fully nonlinear methods (fully nonlinear potential flow and Navier–Stokes models). This article summarizes the code-to-code task and code-to-experiment task that have been performed so far in this project, with a focus on investigating the impact of different levels of nonlinearities in the numerical models. Two different WECs were studied and simulated. The first was a heaving semi-submerged sphere, where free-decay tests and both regular and irregular wave cases were investigated in a code-to-code comparison. The second case was a heaving float corresponding to a physical model tested in a wave tank. We considered radiation, diffraction, and regular wave cases and compared quantities, such as the WEC motion, power output and hydrodynamic loading.
The unsteady cross flow analogy reduces the steady three dimensional separation flow problem into an unsteady two dimensional flow problem in which the section shape changes with time. The two dimensional VORSEP code is extended to the case of arbitrary body growth rates in order to generate the initial vortex structures for the three dimensional free vortex sheet (FVS) code. Automatic procedures to reduce the wing geometry definition to a set of cross flow plane sections corresponding to the locations of the time step solutions and to generate the effective source distribution on each cross flow section to represent the section normal growth across the following steps are incorporated in the VORSEP code. Also, the wake shedding model is improved by adopting a redistribution scheme which improves the stability of the free sheet development with time. The improved wake shedding model combined with the redistribution scheme alleviated the numerical instabilities associated with the vortex rollup.
Combustors in modern aerospace propulsion systems are highly optimized devices and further enhancements to their performance will require advanced predictive modeling techniques. The LES approach has potential to be such a technique, but demonstration of its true predictive capability remains a challenge. In this paper, a methodical study is performed to assess the impact of numerical error on the predictive capabilities of LES and to distinguish it from error arising from physical modeling. Four codes are employed to simulate the bluff body stabilized flame experiment of Volvo Flygmotor AB. The physical models (governing equations and sub-grid models) are kept identical between the codes so that the physical modeling error is the same. For the same reason, computational grids are also kept identical since the sub-grid models are grid dependent. Results indicate that all codes produce a very similar solution for the non-reacting flow in terms of large scale unsteady vortex shedding behavior, mean profiles and second order turbulence statistics, on a suitably refined grid. This solution is also very close to the experimentally observed behavior of the flow. In contrast, for the reacting case, solutions from the various codes exhibit significant differences when analyzed under the same metrics and with similar or even finer grid resolution than that used for the non-reacting case. These differences are large enough to change the macroscopic behavior of the flow, e.g. some codes predict symmetric vortex shedding while others predict asymmetric shedding and some codes show strong recirculation zones while others show stagnant regions behind the bluff body. It is also shown that none of the solutions truly predicts the experimentally observed flow. Given that the physical sub-grid models and computational grids are the same in all the simulations, the inconsistency between results points to a significant impact of numerical methods and, hence, numerical error. The need for mitigation of this impact is highlighted, along with the need for metrics detailing how LES should be applied to reacting flows. It is suggested that both of these issues should be the focus of future work.
The results are presented for the stagnation-point heat-transfer rates used in the design process of the Aeroassist Flight Experiment (AFE) vehicle over its entire aeropass trajectory. The prediction methods used in this investigation demonstrate the application of computational fluid dynamics (CFD) techniques to a wide range of flight conditions and their usefulness in a design process. The heating rates were computed by a viscous-shock-layer (VSL) code at the lower altitudes and by a Navier-Stokes (N-S) code for the higher altitude cases. For both methods, finite-rate chemically reacting gas was considered, and a temperature-dependent wall-catalysis model was used. The wall temperature for each case was assumed to be radiative equilibrium temperature, based on total heating. The radiative heating was estimated by using a correlation equation. Wall slip was included in the N-S calculation method, and this method implicitly accounts for shock slip. The N-S/VSL combination of projection methods was established by comparison with the published benchmark flow-field code LAURA results at lower altitudes, and the direct simulation Monte Carlo results at higher altitude cases. To obtain the design heating rate over the entire forward face of the vehicle, a boundary-layer method (BLIMP code) that employs reacting chemistry and surface catalysis was used. The ratio of the VSL or N-S method prediction to that obtained from the boundary-layer method code at the stagnation point is used to define an adjustment factor, which accounts for the errors involved in using the boundary-layer method.
This document is a user’s guide for the jet noise prediction code GAA_JET_FV, which can be used to make predictions of turbulent mixing noise in high-speed free jets (ie. in the absence of any solid surfaces) of arbitrary cross section. The code requires a Reynolds-averaged Navier-Stokes (RANS) solution for the mean flow and turbulence as input. A script is provided in the code package which can be used to interpolate structured or unstructured RANS solutions onto a structured grid suitable for the noise calculations. Output file formats for two commonly used RANS solvers are currently supported by this script. The document describes how the code can be obtained and installed on a user’s system. A simple test case is provided that can be run with minimal user knowledge of the code details. General instructions for running the interpolation script and the main code are given along with descriptions of the input and output data files and individual code modules. Several additional test cases are provided which allow the user to exercise additional features of the code. This document is Version 1, Revision 0 of the User’s Guide, which contains examples of round and non-axisymmetric unheated jet test cases. Future versions are planned which will include additional functionality for the code and more complex test cases.
The high cost of software production is driving development organizations to adopt more automated design and analysis methods such as rapid prototyping, computer-aided software engineering (CASE) tools, and high-level code generators. Even developers of safety-critical software system have adopted many of these new methods while striving to achieve high levels Of quality and reliability. While these new methods may enhance productivity and quality in many cases, we examine some of the risks involved in the use of new methods in safety-critical contexts. We examine a case study involving the use of a CASE tool that automatically generates code from high-level system designs. We show that while high-level testing on the system structure is highly desirable, significant risks exist in the automatically generated code and in re-validating releases of the generated code after subsequent design changes. We identify these risks and suggest process improvements that retain the advantages of rapid, automated development methods within the quality and reliability contexts of safety-critical projects.
The NCC code was validated for a case involving staged transverse injection into Mach 2 flow behind a rearward facing step. Comparisons with experimental data and with solutions from the FPVortex code was then used to perform computations to study fuel-air mixing for the combustor of a candidate rocket based combined cycle engine geometry. Comparisons with a one-dimensional analysis and a three-dimensional code (VULCAN) were performed to assess the qualitative and quantitative performance of the NCC solver.
Purpose: To validate the in-house developed CT Monte Carlo calculation tool GMctdospp against reference simulation sets provided by the AAPM in the new report 195. Methods: Deposited energy was calculated in four segments (test 1) and two 10 cm long cylinders (test 2) inside a CTDI phantom (following case #4 of the AAPM report 195). The x-ray point source of a given 120 kVp spectrum was collimated to a fan beam with two thicknesses (10 mm, 80 mm) for a static and a rotational setup. In addition, a given chest geometry was used to calculate deposited energy in several organs for a 0° static and a rotational beam (following case #5 of the AAPM report 195). The results of GMctdospp were compared against the particular mean value of the four quoted Monte Carlo codes (EGSnrc, Geant 4, MCNP and Penelope). Results: Calculated values showed no outliers in any of the cases. Differences between GMctdospp and the particular mean Results: Calculated values showed no outliers in any of the cases. Differences between GMctdospp and the particular mean value were always at similar magnitude compared to the quoted codes. For case #4 (CTDI phantom) the relative differences were within 1.5 %, on average 0.4 % and for case #5 (chest phantom) within 2.5 % and on average 0.85 %. Conclusion: The results confirmed an overall uncertainty of the Monte-Carlo calculation chain in GMctdospp being <2.5 %, for most cases even better. This can be considered small compared to other sources of uncertainties, e.g. virtual source and patient models. The photon transport implemented in GMctdospp inside a voxel-based patient geometry was successfully verified.
Aeroelastic tests require extensive cost and risk. An aeroelastic wind-tunnel experiment is an order of magnitude more expensive than a parallel experiment involving only aerodynamics. By complementing the wind-tunnel experiments with numerical simulations, the overall cost of the development of aircraft can be considerably reduced. In order to accurately compute aeroelastic phenomenon it is necessary to solve the unsteady Euler/Navier-Stokes equations simultaneously with the structural equations of motion. These equations accurately describe the flow phenomena for aeroelastic applications. At ARC a code, ENSAERO, is being developed for computing the unsteady aerodynamics and aeroelasticity of aircraft, and it solves the Euler/Navier-Stokes equations. The purpose of this cooperative agreement was to enhance ENSAERO in both algorithm and geometric capabilities. During the last five years, the algorithms of the code have been enhanced extensively by using high-resolution upwind algorithms and efficient implicit solvers. The zonal capability of the code has been extended from a one-to-one grid interface to a mismatching unsteady zonal interface. The geometric capability of the code has been extended from a single oscillating wing case to a full-span wing-body configuration with oscillating control surfaces. Each time a new capability was added, a proper validation case was simulated, and the capability of the code was demonstrated.
This paper presents simulations of bump tests conducted as part of the P2M (Power to Melt and Maneuverability) project within the OECD/NEA FIDES-II framework (Framework for Irradiation Experiments). To prepare fuel performance codes for analyzing the P2M experiments, an international Simulation Exercise (SE) was initiated under the P2M project. The objective of this exercise was to model the AN3 and AN10 bump tests carried out in the 1980s at the DR3 reactor (3rd RISØ Fission Gas Project), during which instrumented fuel rodlets (equipped with thermocouples and pressure sensors) were subjected to prolonged power transients and power dips. The ongoing simulation exercise involves seven organizations from industry, research, and regulatory bodies, each using different fuel performance codes. This paper provides detailed descriptions of the test cases and simulation results from six fuel performance codes (ALCYONE, BISON, FAST, FEMAXI, FRAPCON, and TRANSURANUS), with a particular focus on fission gas release kinetics.
Direct Numerical Simulations (DNS) are performed using the FUN3D code ( https://fun3d.larc.nasa.gov) for three validation cases: (1) flow through a plane channel, (2) flow through a channel with a constriction, and (3) flow over a flat plate. FUN3D is a node-centered finite-volume code developed at the NASA Langley Research Center that solves the three-dimensional compressible Navier-Stokes equations on unstructured computational grids. The simulations are performed employing the 2nd-order unstructured monotonic upstream scheme for conservation laws (UMUSCL). The results are compared with available experimental and numerical data. The effect of the UMUSCL reconstruction parameter (κ) is assessed, and the results indicate that κ = 0.9 yields satisfactory results in terms of accuracy and robustness compared to available data. Further analyses of the results, along with additional test cases and grids will be presented in the final manuscript.
Monte-Carlo nuclear reaction and transport codes are widely used to devise accelerator-based nuclear physics experiments; at the same time, many experiments are performed to validate the Monte-Carlo codes, which can be used for the design of full-scale nuclear power applications or the design of new benchmark experiments. Dedicated model benchmark studies investigate a broad range of nuclear reactions and quantities. Examples of these include isotope formation or secondary particle fluxes that result from the interactions of GeV-range hadrons with monoisotopic targets, which can be used to assess the respective systematic uncertainty of models. Such benchmark studies, as well as many nuclear application experiments and simulations carried out by various groups over the last few decades, enable us to draw methodological lessons. In this work, model uncertainty determined based on available experimental data allow us to identify the effects of practitioner expertise as well as the design of codes (user access to micro-scale parameters) on the range of uncertainties. We found that in cases when simulations are performed by code developers or users that are very experienced in performing simulations, the model to experiment quantity ratios generally agree with the limits determined by dedicated benchmark studies. In other cases, the ratios generally tend to be either smaller (underestimation of model error) or larger (overestimation of model error). A plausible explanation of the aforementioned effects is suggested.
The frame synchronization performance characteristics are defined that can be measured under actual operating conditions. The frame synchronization process is explained, and test data on the performance of frame synchronizers and frame synchronization codes are presented. The application of frame synchronization methods in specific cases is considered in the light of code selection, frame length, and data recovery.
Quantum communication systems harness modern physics through state-of-the-art optical engineering to provide revolutionary capabilities. An important concern for quantum communication engineering is designing and prototyping these systems to prototype proposed capabilities. We apply the paradigm of software-defined communica- tion for engineering quantum communication systems to facilitate rapid prototyping and prototype comparisons. We detail how to decompose quantum communication terminals into functional layers defining hardware, software, and middleware concerns, and we describe how each layer behaves. Using the super-dense coding protocol as a test case, we describe implementations of both the transmitter and receiver, and we present results from numerical simulations of the behavior. We find that while the theoretical benefits of super dense coding are maintained, there is a classical overhead associated with the full implementation.