Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “OpenMP”

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 199 records · Page 11

Optimizing the hypre solver for manycore and GPU architectures

The solution of large-scale combustion problems with codes such as Uintah on modern computer architectures requires the use of multithreading and GPUs to achieve performance. Uintah uses a low-Mach number approximation that requires iteratively solving a large system of linear equations. The Hypre iterative solver has solved such systems in a scalable way for Uintah, but the use of OpenMP with Hypre leads to at least slowdown due to OpenMP overheads. The proposed solution uses the MPI Endpoints within Hypre, where each team of threads acts as a different MPI rank. This approach minimizes OpenMP synchronization overhead and performs as fast or (up to 1.44) faster than Hypre's MPI-only version, and allows the rest of Uintah to be optimized using OpenMP. The profiling of the GPU version of Hypre shows the bottleneck to be the launch overhead of thousands of micro-kernels. The GPU performance was improved by fusing these micro-kernels and was further optimized by using Cuda-aware MPI, resulting in an overall speedup of 1.16—1.44 compared to the baseline GPU implementation. The above optimization strategies were published in the International Conference on Computational Science 2020 [1]. This work extends the previously published research by carrying out the second phase of communication-centered optimizations in Hypre to improve its scalability on large-scale supercomputers. Additionally, this includes an efficient non-blocking inter-thread communication scheme, communication-reducing patch assignment, and expression of logical communication parallelism to a new version of the MPICH library that utilizes the underlying network parallelism [2]. The above optimizations avoid communication bottlenecks previously observed during strong scaling and improve performance by up to 2 on 256 nodes of Intel Knight's Landing processor.

97 MATHEMATICS AND COMPUTING↗

Distributed Order Recording Techniques for Efficient Record-and-Replay of Multi-threaded Programs

After all these years and all these other shared memory programming frameworks, OpenMP is still the most popular one. However, its greater levels of non-deterministic execution makes debugging and testing more challenging. The ability to record and deterministically replay the program execution is key to address this challenge. However, scalably replaying OpenMP programs is still an unresolved problem. In this paper, we propose two novel techniques that use Distributed Clock (DC) and Distributed Epoch (DE) recording schemes to eliminate excessive thread synchronization for OpenMP record and replay. Our evaluation on representative HPC applications with ReOMP, which we used to realize DC and DE recording, shows that our approach is 2-5x more efficient than traditional approaches that synchronize on every shared-memory access. Furthermore, we demonstrate that our approach can be easily combined with MPI-level replay tools to replay non-trivial MPI+OpenMP applications. We achieve this by integrating ReOMP into ReMPI, an existing scalable MPI record-and-replay tool, with only a small MPI-scale-independent runtime overhead.

Fu, Xiang↗

Reaching bandwidth saturation using transparent injection parallelization

Although logically available, applications may not exploit enough instantaneous communication concurrency to maximize network utilization on HPC systems. This is exacerbated in hybrid programming models that combine single program multiple data with OpenMP or CUDA. We present the design of a “multi-threaded” runtime able to transparently increase the instantaneous network concurrency and to provide near saturation bandwidth, independent of the application configuration and dynamic behavior. The runtime offloads communication requests from application level tasks to multiple communication servers. The servers use system specific performance models to attain network saturation. Our techniques alleviate the need for spatial and temporal application level message concurrency optimizations. Experimental results show improved message throughput and bandwidth by as much as 150% for 4 KB messages on InfiniBand and by as much as 120% for 4 KB messages on Cray Aries. For more complex operations such as all-to-all collectives, we observe as much as 30% speedup. This translates into 23% speedup on 12,288 cores for a NAS FT implemented using FFTW. We observe as much as 76% speedup on 1500 cores for an already optimized UPC+OpenMP geometric multigrid application using hybrid parallelism. For the geometric multigrid GPU implementation, we observe as much as 44% speedup on 512 GPUs.

Chaimov, Nicholas↗

Deploy threading in Nalu solver stack

The goal of the ExaWind project is to enable predictive simulations of wind farms composed of many MW-scale turbines situated in complex terrain. Predictive simulations will require computational fluid dynamics (CFD) simulations for which the mesh resolves the geometry of the turbines, and captures the rotation and large deflections of blades. Whereas such simulations for a single turbine are arguably petascale class, multi-turbine wind farm simulations will require exascale-class resources. The primary code in the ExaWind project is Nalu, which is an unstructured-grid solver for the acousticallyincompressible Navier-Stokes equations, and mass continuity is maintained through pressure projection. The model consists of the mass-continuity Poisson-type equation for pressure and a momentum equation for the velocity. For such modeling approaches, simulation times are dominated by linear-system setup and solution for the continuity and momentum systems. For the ExaWind challenge problem, the moving meshes greatly affect overall solver costs as re-initialization of matrices and re-computation of preconditioners is required at every time step In this Milestone, we examine the effect of threading on the solver stack performance against flat-MPI results obtained from previous milestones using Haswell performance data full-turbine simulations. Whereas the momentum equations are solved only with the Trilinos solvers, we investigate two algebraic-multigrid preconditioners for the continuity equations: Trilinos/Muelu and HYPRE/BoomerAMG. These two packages embody smoothed-aggregation and classical Ruge-Stiiben AMG methods, respectively. In our FY18 Q2 report, we described our efforts to improve setup and solve of the continuity equations under flat-MPI parallelism. While significant improvement was demonstrated in the solve phase, setup times remained larger than expected. Starting with the optimized settings described in the Q2 report, we explore here simulation performance where OpenMP threading is employed in the solver stack. For Trilinos, threading is acheived through the Kokkos abstraction where, whereas HYPRE/BoomerAMG employs straight OpenMP. We examined results for our mid-resolution baseline turbine simulation configuration (229M DOF). Simulations on 2048 Haswell cores explored the effect of decreasing the number of MPI ranks while increasing the number of threads. Both HYPRE and Trilinos exhibited similar overal solution times, and both showed dramatic increases in simulation time in the shift from MPI ranks to OpenMP threads. This increase is attributed to the large amount of work per MPI rank starting at the single-thread configuration. Decreasing MPI ranks, while increasing threads, may be increasing simulation time due to thread synchronization and start-up overhead contributing to the latency and serial time in the model. These result showed that an MPI+OpenMP parallel decomposition will be more effective as the amount per MPI rank computation per MPI rank decreases and the communication latency increases. This idea was demonstrated in a strong scaling study of our low-resolution baseline model (29M DOF) with the Trilinos-HYPRE configuration. While MPI-only results showed scaling improvement out to about 1536 cores, engaging threading carried scaling improvements out to 4128 cores — roughly 7000 DOF per core. This is an important result as improved strong scaling is needed for simulations to be executed over sufficiently long simulated durations (i.e., for many timesteps). In addition to threading work described above, the team examined solver-performance improvements by exploring communication-overhead in the HYPRE-GMRES implementation through a communicationoptimal- GMRE algorithm (CO-GMRES), and offloading compute-intensive solver actions to GPUs. To those ends, a HYPRE mini-app was allow us to easily test different solver approaches and HYPRE parameter settings without running the entire Nalu code. With GPU acceleration on the Summitdev supercomputer, a 20x speedup was achieved for the overall preconditioner and solver execution time for the mini-app. A study on Haswell processors showed that CO-GMRES provides benefits as one increases MPI ranks.

17 WIND ENERGY↗

Parallel Programming in MCNP6

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.

97 MATHEMATICS AND COMPUTING↗

Thread-level parallelization and optimization of NWChem for the Intel MIC architecture

In the multicore era it was possible to exploit the increase in on-chip parallelism by simply running multiple MPI processes per chip. Unfortunately, manycore processors' greatly increased thread- and data-level parallelism coupled with a reduced memory capacity demand an altogether different approach. In this paper we explore augmenting two NWChem modules, triples correction of the CCSD(T) and Fock matrix construction, with OpenMP in order that they might run efficiently on future manycore architectures. As the next NERSC machine will be a self-hosted Intel MIC (Xeon Phi) based supercomputer, we leverage an existing MIC testbed at NERSC to evaluate our experiments. In order to proxy the fact that future MIC machines will not have a host processor, we run all of our experiments in native mode. We found that while straightforward application of OpenMP to the deep loop nests associated with the tensor contractions of CCSD(T) was sufficient in attaining high performance, significant e ort was required to safely and efeciently thread the TEXAS integral package when constructing the Fock matrix. Ultimately, our new MPI+OpenMP hybrid implementations attain up to 65× better performance for the triples part of the CCSD(T) due in large part to the fact that the limited on-card memory limits the existing MPI implementation to a single process per card. Additionally, we obtain up to 1.6× better performance on Fock matrix constructions when compared with the best MPI implementations running multiple processes per card.

CCSD(T)↗

Spray: Sparse Reductions of Arrays in OPEN MP

We present SPRAY, an open-source header-only C++ library for sparse reductions of arrays. SPRAY is meant for applications in which a large array is collaboratively updated by multiple threads using an associative and commutative operation such as +=. Especially when each thread accesses only parts of the array, SPRAY can perform significantly better than OPENMP's built-in reduction clause or atomic updates, while also using less memory than the former. SPRAY provides both an easy-to-use interface that can serve as a drop-in replacement for OPENMP reductions and a selection of reducer objects that accumulate the final result in different thread-safe ways. We demonstrate SPRAY through multiple test cases including the LULESH shock hydrodynamics code and a transpose-matrix-vector multiplication for sparse matrices stored in CSR format. SPRAY reductions outperform built-in OPENMP reductions consistently, in some cases improving run time and memory overhead by 20X, and even beating domain-specific approaches such as Intel MKI, by over 2X in some cases. Furthermore, SPRAY reductions have a minimal impact on the code base, requiring only a few lines of source code changes. Once in place, SPRAY reduction schemes can be switched easily, allowing performance portability and tuning opportunities by separating performance-critical implementation details from application code.

OpenMP↗

Thread-Level Parallelization and Optimization of NWChem for the Intel MIC Architecture

In the multicore era it was possible to exploit the increase in on-chip parallelism by simply running multiple MPI processes per chip. Unfortunately, manycore processors' greatly increased thread- and data-level parallelism coupled with a reduced memory capacity demand an altogether different approach. In this paper we explore augmenting two NWChem modules, triples correction of the CCSD(T) and Fock matrix construction, with OpenMP in order that they might run efficiently on future manycore architectures. As the next NERSC machine will be a self-hosted Intel MIC (Xeon Phi) based supercomputer, we leverage an existing MIC testbed at NERSC to evaluate our experiments. In order to proxy the fact that future MIC machines will not have a host processor, we run all of our experiments in tt native mode. We found that while straightforward application of OpenMP to the deep loop nests associated with the tensor contractions of CCSD(T) was sufficient in attaining high performance, significant effort was required to safely and efficiently thread the TEXAS integral package when constructing the Fock matrix. Ultimately, our new MPI OpenMP hybrid implementations attain up to 65x better performance for the triples part of the CCSD(T) due in large part to the fact that the limited on-card memory limits the existing MPI implementation to a single process per card. Additionally, we obtain up to 1.6x better performance on Fock matrix constructions when compared with the best MPI implementations running multiple processes per card.

97 MATHEMATICS AND COMPUTING↗

Large language model evaluation for high–performance computing software development

We apply AI-assisted large language model (LLM) capabilities of GPT-3 targeting high-performance computing (HPC) kernels for (i) code generation, and (ii) auto-parallelization of serial code in C ++, Fortran, Python and Julia. Our scope includes the following fundamental numerical kernels: AXPY, GEMV, GEMM, SpMV, Jacobi Stencil, and CG, and language/programming models: (1) C++ (e.g., OpenMP [including offload], OpenACC, Kokkos, SyCL, CUDA, and HIP), (2) Fortran (e.g., OpenMP [including offload] and OpenACC), (3) Python (e.g., numpy, Numba, cuPy, and pyCUDA), and (4) Julia (e.g., Threads, CUDA.jl, AMDGPU.jl, and KernelAbstractions.jl). Kernel implementations are generated using GitHub Copilot capabilities powered by the GPT-based OpenAI Codex available in Visual Studio Code given simple + + prompt variants. To quantify and compare the generated results, we propose a proficiency metric around the initial 10 suggestions given for each prompt. For auto-parallelization, we use ChatGPT interactively giving simple prompts as in a dialogue with another human including simple “prompt engineering” follow ups. Results suggest that correct outputs for C++ correlate with the adoption and maturity of programming models. For example, OpenMP and CUDA score really high, whereas HIP is still lacking. We found that prompts from either a targeted language such as Fortran or the more general-purpose Python can benefit from adding language keywords, while Julia prompts perform acceptably well for its Threads and CUDA.jl programming models. Finally, we expect to provide an initial quantifiable point of reference for code generation in each programming model using a state-of-the-art LLM. Overall, understanding the convergence of LLMs, AI, and HPC is crucial due to its rapidly evolving nature and how it is redefining human-computer interactions.

97 MATHEMATICS AND COMPUTING↗

ChatPORT: Fine-Tuned LLM for Easy Code {PORT}ing

Fine-tuning existing LLMs for specialized tasks has become a very attractive alternative due to its low cost and quick development cycle. With many pre-trained LLMs available, it is an increasingly complex task to choose the correct model as the starting point or base model. In this work we discuss ChatPORT - a specialized fine-tuned LLM geared towards providing correctly translated codes from one programming model to another. We evaluate a number of base models and compare and contrast their features and characteristics that make them a viable starting point. In this paper, we focus on the OpenMP offload porting capabilities of ChatPORT. We build our training data using kernels from the Heterogeneous Computing Benchmarks (HeCBench) [12] and the OpenMP Validation and Verification suite [5] to fine-tune the base models. We then test the model using unseen kernels extracted from the HeCBench benchmark suite. Our results show that: (1) not all open LLMs geared towards HPC are aware of programming models like OpenMP, (2) although all base models benefit from fine-tuning they learn differently and produce different correctness rates, (3) depending on the memory size and compute resource available, different base models can be used for fine-tuning without significantly affecting the quality of transpiled code they generate, (4) fine-tuning improved the correctness rate of the LLM by an average of 43.2%, and (5) feedback-based training data further increased the correctness rate by an average of 6% over the LLMs tested.

Pophale, Swaroop [ORNL] (ORCID:0000000185446367)↗

PyOMP: Multithreaded Parallel Programming in Python

We know that Python is a widely used language in scientific computing. When the goal is high performance, however, Python lags far behind low-level languages such as C and Fortran. To support applications that stress performance, Python needs to access the full capabilities of modern CPUs. That means support for parallel multithreading. In this paper, we describe PyOMP, a system that enables OpenMP in Python. Programmers write code in Python with OpenMP, Numba generates code that compiles to LLVM, and the resulting programs run with performance that approaches that from code written with C and OpenMP. In this paper we provide an update on the PyOMP project and explain how to install it and use it to write parallel multithreaded code in Python.

97 MATHEMATICS AND COMPUTING↗

Towards Achieving Performance Portability Using Directives for Accelerators

In this paper we explore the performance portability of directives provided by OpenMP 4 and OpenACC to program various types of node architectures with attached accelerators, both self-hosted multicore and offload multicore/GPU. Our goal is to examine how successful OpenACC and the newer offload features of OpenMP 4.5 are for moving codes between architectures, how much tuning might be required and what lessons we can learn from this experience. To do this, we use examples of algorithms with varying computational intensities for our evaluation, as both compute and data access efficiency are important considerations for overall application performance. We implement these kernels using various methods provided by newer OpenACC and OpenMP implementations, and we evaluate their performance on various platforms including both X86_64 with attached NVIDIA GPUs, self-hosted Intel Xeon Phi KNL, as well as an X86_64 host system with Intel Xeon Phi coprocessors. In this paper, we explain what factors affected the performance portability such as how to pick the right programming model, its programming style, its availability on different platforms, and how well compilers can optimize and target to multiple platforms.

97 MATHEMATICS AND COMPUTING↗

A multi-threaded version of MCFM

We report on our findings modifying MCFM using OpenMP to implement multi-threading. By using OpenMP, the modified MCFM will execute on any processor, automatically adjusting to the number of available threads. We then modified the integration routine VEGAS to distribute the event evaluation over the threads, while combining all events at the end of every iteration to optimize the numerical integration. Furthermore, we took special care so that the results of the Monte Carlo integration were independent of the number of threads used, to facilitate the validation of the OpenMP version of MCFM.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Critical Path-Based Thread Placement for NUMA Systems

Multicore multiprocessors use a Non Uniform Memory Architecture (NUMA) to improve their scalability. However, NUMA introduces performance penalties due to remote memory accesses. Without efficiently managing data layout and thread mapping to cores, scientific applications, even if they are optimized for NUMA, may suffer performance loss. In this paper, we present algorithms and a runtime system that optimize the execution of OpenMP applications on NUMA architectures. By collecting information from hardware counters, the runtime system directs thread placement and reduces performance penalties by minimizing the critical path of OpenMP parallel regions. The runtime system uses a scalable algorithm that derives placement decisions with negligible overhead. We evaluate our algorithms and runtime system with four NPB applications implemented in OpenMP. On average the algorithms achieve between 8.13% and 25.68% performance improvement compared to the default Linux thread placement scheme. The algorithms miss the optimal thread placement in only 8.9% of the cases.

97 MATHEMATICS AND COMPUTING↗

Evaluation of OpenAI Codex for HPC Parallel Programming Models Kernel Generation

We evaluate AI-assisted generative capabilities on fundamental numerical kernels in high-performance computing (HPC), including AXPY, GEMV, GEMM, SpMV, Jacobi Stencil, and CG. We test the generated kernel codes for a variety of language-supported programming models, including (1) C++ (e.g., OpenMP [including offload], OpenACC, Kokkos, SyCL, CUDA, and HIP), (2) Fortran (e.g., OpenMP [including offload] and OpenACC), (3) Python (e.g., numpy, Numba, cuPy, and pyCUDA), and (4) Julia (e.g., Threads, CUDA.jl, AMDGPU.jl, and KernelAbstractions.jl). We use the GitHub Copilot capabilities powered by the GPT-based OpenAI Codex available in Visual Studio Code as of April 2023 to generate a vast amount of implementations given simple + + prompt variants. To quantify and compare the results, we propose a proficiency metric around the initial 10 suggestions given for each prompt. Results suggest that the OpenAI Codex outputs for C++ correlate with the adoption and maturity of programming models. For example, OpenMP and CUDA score really high, whereas HIP is still lacking. We found that prompts from either a targeted language such as Fortran or the more general purpose Python can benefit from adding code keywords, while Julia prompts perform acceptably well for its mature programming models (e.g., Threads and CUDA.jl). We expect for these benchmarks to provide a point of reference for each programming model's community. Overall, understanding the convergence of large language models, AI, and HPC is crucial due to its rapidly evolving nature and how it is redefining human-computer interactions.

Godoy, William↗

Characterizing and Mitigating Work Time Inflation in Task Parallel Programs

Task parallelism raises the level of abstraction in shared memory parallel programming to simplify the development of complex applications. However, task parallel applications can exhibit poor performance due to thread idleness, scheduling overheads, and work time inflation – additional time spent by threads in a multithreaded computation beyond the time required to perform the same work in a sequential computation. We identify the contributions of each factor to lost efficiency in various task parallel OpenMP applications and diagnose the causes of work time inflation in those applications. Increased data access latency can cause significant work time inflation in NUMA systems. Our locality framework for task parallel OpenMP programs mitigates this cause of work time inflation. Our extensions to the Qthreads library demonstrate that locality-aware scheduling can improve performance up to 3X compared to the Intel OpenMP task scheduler.

Olivier, Stephen L.↗

RAJA Performance Suite

The RAJA Performance Suite is designed to evaluate performance of the RAJA performance portability library on a wide variety of important high performance computing (HPC) algorithmic lulmels. These kernels assess compiler optimizations and various parallel programming model backends accessible through RAJA, such as OpenMP, CUDA, etc. The Initial version of the suite contains 25 computational kernels, each of which appears in 6 variants: Baseline SequcntiaJ, RAJA SequentiaJ, Baseline OpenMP, RAJA OpenMP, Baseline CUDA, RAJA CUDA. All variants of each kernel perform essentially the same mathematical operations and the loop body code for each kernel is identical across all variants. There are a few kernels, such as those that contain reduction operations, that require CUDA-specific coding for their CUDA variants. ActuaJ computer instructions executed and how they run in parallel differs depending on the parallel programming model backend used and which optimizations are perfonned by the compiler used to build the Perfonnance Suite executable. The Suite will be used primarily by RAJA developers to perform regular assessments of RAJA performance across a range of hardware platforms and compilers as RAJA features are being developed. It will also be used by LLNL hardware and software vendor panners for new defining requirements for future computing platform procurements and acceptance testing. In particular, the RAJA Performance Suite will be used for compiler acceptance testing of the upcoming CORAUSierra machine {initial LLNL delivery expected in late-2017/early 2018) and the CORAL-2 procurement. The Suite will aJso be used to generate concise source code reproducers of compiler and runtime issues we uncover so that we may provide them to relevant vendors to be fixed.

Hornung, RichardD.↗

A feasibility study on porting the community land model onto accelerators using OpenACC

As environmental models (such as Accelerated Climate Model for Energy (ACME), Parallel Reactive Flow and Transport Model (PFLOTRAN), Arctic Terrestrial Simulator (ATS), etc.) became more and more complicated, we are facing enormous challenges regarding to porting those applications onto hybrid computing architecture. OpenACC appears as a very promising technology, therefore, we have conducted a feasibility analysis on porting the Community Land Model (CLM), a terrestrial ecosystem model within the Community Earth System Models (CESM)). Specifically, we used automatic function testing platform to extract a small computing kernel out of CLM, then we apply this kernel into the actually CLM dataflow procedure, and investigate the strategy of data parallelization and the benefit of data movement provided by current implementation of OpenACC. Even it is a non-intensive kernel, on a single 16-core computing node, the performance (based on the actual computation time using one GPU) of OpenACC implementation is 2.3 time faster than that of OpenMP implementation using single OpenMP thread, but it is 2.8 times slower than the performance of OpenMP implementation using 16 threads. On multiple nodes, MPI_OpenACC implementation demonstrated very good scalability on up to 128 GPUs on 128 computing nodes. This study also provides useful information for us to look into the potential benefits of “deep copy” capability and “routine” feature of OpenACC standards. In conclusion, we believe that our experience on the environmental model, CLM, can be beneficial to many other scientific research programs who are interested to porting their large scale scientific code using OpenACC onto high-end computers, empowered by hybrid computing architecture.

54 ENVIRONMENTAL SCIENCES↗