Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “shared memory”

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 19 records

AnalyzeThat: A Programmable Shared-Memory System for an Array of Processing-In-Memory Devices

Processing In Memory (PIM), the concept of integrating processing directly with memory, has been attracting a lot of attention since PIM can assist in overcoming the throughput limitation caused by data movement between CPU and memory. The challenge, however, is that it requires the programmers to have a deep understanding of the PIM architecture to maximize the benefits such as data locality and parallel thread execution on multiple PIM devices. In this study, we present AnalyzeThat, a programmable shared-memory system for parallel data processing with PIM devices. Thematic to AnalyzeThat is a rich PIM-Aware Data Structure (PADS), which is an encapsulation that integrally ties together the data, the analysis tasks and the runtime needed to interface with the PIM device array. The PADS abstraction provides (i) a key-value data container that allows programmers to easily store data on multiple PIMs, (ii) a suite of parallel operations with which users can easily implement data analysis applications, and (iii) a runtime, hidden to programmers, which provides the mechanisms needed to overlay both the data and the tasks on the PIM device array in an intelligent fashion, based on PIM-specific information collected from the hardware. We have developed a PIM emulation framework called AnalyzeThat. Our experimental evaluation with representative data analytics applications suggests that the proposed system can significantly reduce the PIM programming effort without losing its technology benefits.

97 MATHEMATICS AND COMPUTING↗

Experimental Characterization of OpenMP Offloading Memory Operations and Unified Shared Memory Support

The OpenMP specification recently introduced support for unified shared memory, allowing implementation to leverage underlying system software to provide a simpler GPU offloading model where explicit mapping of variables is optional. Support for this feature is becoming more available in different OpenMP implementations on several hardware platforms. A deeper understanding of the different implementation’s execution profile and performance is crucial for applications as they consider the performance portability implications of adopting a unified memory offloading programming style. This work introduces a benchmark tool to characterize unified memory support in several OepnMP compilers and runtimes, with emphasis on identifying discrepancies between different OpenMP implementations as to how they various memory allocation strategies interact with unified shared memory. The benchmark tool is used to characterize OpenMP compilers on three leading High Performance Computing platforms supporting different CPU and device architectures. The benchmark tool is used to assess the impact of enabling unified shared memory on the performance of memory-bound code, highlighting implementation differences that should be accounted for when applications consider performance portability across platforms and compilers.

Elwasif, Wael↗

JACC.shared: Leveraging HPC Metaprogramming and Performance Portability for Computations That Use Shared Memory GPUs

In this work, we present JACC.shared, a new feature of Julia for ACCelerators (JACC), which is the performanceportable and metaprogramming model of the just-in-time and LLVM-based Julia language. This new feature allows JACC applications to leverage the high-performance computing (HPC) capabilities of high-bandwidth, on-chip GPU memory. Historically, exploiting high-bandwidth, shared-memory GPUs has not been a priority for high-level programming solutions. JACC.shared covers that gap for the first time, thereby providing a highlevel, portable, and easy-to-use solution for programmers to exploit this memory and supporting all current major accelerator architectures. Well-known HPC and AI workloads, such as multi/hyperspectral imaging and AI convolutions, have been used to evaluate JACC.shared on two exascale GPU architectures hosted by some of the most powerful US Department of Energy supercomputers: Perlmutter (NVIDIA A100) and Frontier (AMD MI250X). The performance evaluation reports speedup of up to 3.5× by adding only one line of code to the base codes, thus providing important accelerators in a simple, portable, and transparent way and elevating the programming productivity and performance-portability capabilities for Julia/JACC HPC, AI, and scientific applications.

Valero Lara, Pedro [ORNL] (ORCID:0000000214794310)↗

Reducing Memory Consumption in Calico with Shared Memory

This document details the work to reduce memory consumption in Calico. Calico is SimTools’ Constructive Solid Geometry (CSG) and geometry painting library. It is primarily used to paint material volume fractions in the Eulerian meshes of the physics codes. Calico provides point-in-body checks for the geometry supplied by an Oso model, which are then aggregated by the host codes. In addition, Calico can be used to build Oso models and is used by Ingen for that purpose. Oso models, and thus Calico, provide support for various CSG primitives such as spheres, cylinders, surfaces generated by rotating tabular curve data, and STL files as well as binary combinations of those primitives. Prior to refactoring Calico will run out of memory on CTS-1 machines when 36 MPI ranks are used per node when reading STL models on the order of 1.5 GB. This limitation is a bottleneck in designer workflow. This problem has been alleviated through the use of data structures to both reduce memory consumption and to leverage MPI-3 shared memory. This report details the data structures targeted for refactoring in Calico, the methods and implementation details for reducing memory consumption and leveraging shared memory, and results for one test problem. Results show a memory reduction when loading a 1 GB STL file by a factor of 27.5, from 93.4 to 3.4 GB.

97 MATHEMATICS AND COMPUTING↗

Performing an allreduce operation using shared memory

Methods, apparatus, and products are disclosed for performing an allreduce operation using shared memory that include: receiving, by at least one of a plurality of processing cores on a compute node, an instruction to perform an allreduce operation; establishing, by the core that received the instruction, a job status object for specifying a plurality of shared memory allreduce work units, the plurality of shared memory allreduce work units together performing the allreduce operation on the compute node; determining, by an available core on the compute node, a next shared memory allreduce work unit in the job status object; and performing, by that available core on the compute node, that next shared memory allreduce work unit.

97 MATHEMATICS AND COMPUTING↗

Performing an allreduce operation using shared memory

Methods, apparatus, and products are disclosed for performing an allreduce operation using shared memory that include: receiving, by at least one of a plurality of processing cores on a compute node, an instruction to perform an allreduce operation; establishing, by the core that received the instruction, a job status object for specifying a plurality of shared memory allreduce work units, the plurality of shared memory allreduce work units together performing the allreduce operation on the compute node; determining, by an available core on the compute node, a next shared memory allreduce work unit in the job status object; and performing, by that available core on the compute node, that next shared memory allreduce work unit.

97 MATHEMATICS AND COMPUTING↗

IMPACC: A Tightly Integrated MPI+OpenACC Framework Exploiting Shared Memory Parallelism

We propose IMPACC, an MPI+OpenACC framework for heterogeneous accelerator clusters. IMPACC tightly integrates MPI and OpenACC, while exploiting the shared memory parallelism in the target system. IMPACC dynamically adapts the input MPI+OpenACC applications on the target heterogeneous accelerator clusters to fully exploit target system-specific features. IMPACC provides the programmers with the unified virtual address space, automatic NUMA-friendly task-device mapping, efficient integrated communication routines, seamless streamlining of asynchronous executions, and transparent memory sharing. We have implemented IMPACC and evaluated its performance using three heterogeneous accelerator systems, including Titan supercomputer. Results show that IMPACC can achieve easier programming, higher performance, and better scalability than the current MPI+OpenACC model.

Kim, Jungwon↗

A parallel WLS state estimator on shared memory computers

This paper describes our experience with developing a parallel weighted- least-square (WLS) state estimation (SE) program for shared-memory parallel computers. Since the key computational kernel of the WLS algorithm based on the Newton-Raphson approach is a solver of sparse linear equations, a significant part of our effort was focused on selecting, implementing and evaluating this algorithm. An optimized shared memory version of the conjugate gradient (CG) algorithm was found to be competitive to state-of-the-art implementation of LU solvers for the SE problem on the SGI Altix, a shared memory architecture. We also ported the full SE algorithm including CG to the Cray MTA-2 shared memory multithreaded architecture and investigated its performance.

97 MATHEMATICS AND COMPUTING↗

Contention Modeling for Multithreaded Distributed Shared Memory Machines: The Cray XMT

Distributed Shared Memory (DSM) machines are a wide class of multi-processor computing systems where a large virtually-shared address space is mapped on a network of physically distributed memories. High memory latency and network contention are two of the main factors that limit performance scaling of such architectures. Modern high-performance computing DSM systems have evolved toward exploitation of massive hardware multi-threading and fine-grained memory hashing to tolerate irregular latencies, avoid network hot-spots and enable high scaling. In order to model the performance of such large-scale machines, parallel simulation has been proved to be a promising approach to achieve good accuracy in reasonable times. One of the most critical factors in solving the simulation speed-accuracy trade-off is network modeling. The Cray XMT is a massively multi-threaded supercomputing architecture that belongs to the DSM class, since it implements a globally-shared address space abstraction on top of a physically distributed memory substrate. In this paper, we discuss the development of a contention-aware network model intended to be integrated in a full-system XMT simulator. We start by measuring the effects of network contention in a 128-processor XMT machine and then investigate the trade-off that exists between simulation accuracy and speed, by comparing three network models which operate at different levels of accuracy. The comparison and model validation is performed by executing a string-matching algorithm on the full-system simulator and on the XMT, using three datasets that generate noticeably different contention patterns.

99 GENERAL AND MISCELLANEOUS↗

Parallel-vector algorithms for particle simulations on shared-memory multiprocessors

Over the last few decades, the computational demands of massive particle-based simulations for both scientific and industrial purposes have been continuously increasing. Hence, considerable efforts are being made to develop parallel computing techniques on various platforms. In such simulations, particles freely move within a given space, and so on a distributed-memory system, load balancing, i.e., assigning an equal number of particles to each processor, is not guaranteed. However, shared-memory systems achieve better load balancing for particle models, but suffer from the intrinsic drawback of memory access competition, particularly during (1) paring of contact candidates from among neighboring particles and (2) force summation for each particle. Here, novel algorithms are proposed to overcome these two problems. For the first problem, the key is a pre-conditioning process during which particle labels are sorted by a cell label in the domain to which the particles belong. Then, a list of contact candidates is constructed by pairing the sorted particle labels. For the latter problem, a table comprising the list indexes of the contact candidate pairs is created and used to sum the contact forces acting on each particle for all contacts according to Newton's third law. With just these methods, memory access competition is avoided without additional redundant procedures. The parallel efficiency and compatibility of these two algorithms were evaluated in discrete element method (DEM) simulations on four types of shared-memory parallel computers: a multicore multiprocessor computer, scalar supercomputer, vector supercomputer, and graphics processing unit. The computational efficiency of a DEM code was found to be drastically improved with our algorithms on all but the scalar supercomputer. Thus, the developed parallel algorithms are useful on shared-memory parallel computers with sufficient memory bandwidth.

97 MATHEMATICS AND COMPUTING↗

Direct access inter-process shared memory

A technique for directly sharing physical memory between processes executing on processor cores is described. The technique includes loading a plurality of processes into the physical memory for execution on a corresponding plurality of processor cores sharing the physical memory. An address space is mapped to each of the processes by populating a first entry in a top level virtual address table for each of the processes. The address space of each of the processes is cross-mapped into each of the processes by populating one or more subsequent entries of the top level virtual address table with the first entry in the top level virtual address table from other processes.

97 MATHEMATICS AND COMPUTING↗

Scalable Triadic Analysis of Large-Scale Graphs: Multi-Core vs. Multi-Processor vs. Multi-Threaded Shared Memory Architectures

Triadic analysis encompasses a useful set of graph mining methods that is centered on the concept of a triad, which is a subgraph of three nodes and the configuration of directed edges across the nodes. Such methods are often applied in the social sciences as well as many other diverse fields. Triadic methods commonly operate on a triad census that counts the number of triads of every possible edge configuration in a graph. Like other graph algorithms, triadic census algorithms do not scale well when graphs reach tens of millions to billions of nodes. To enable the triadic analysis of large-scale graphs, we developed and optimized a triad census algorithm to efficiently execute on shared memory architectures. We will retrace the development and evolution of a parallel triad census algorithm. Over the course of several versions, we continually adapted the code’s data structures and program logic to expose more opportunities to exploit parallelism on shared memory that would translate into improved computational performance. We will recall the critical steps and modifications that occurred during code development and optimization. Furthermore, we will compare the performances of triad census algorithm versions on three specific systems: Cray XMT, HP Superdome, and AMD multi-core NUMA machine. These three systems have shared memory architectures but with markedly different hardware capabilities to manage parallelism.

Triadic analysis, triads, triad census, large-scal↗

A Comparison of Shared Memory Parallel Programming Models

The dominant parallel programming models for shared memory computers, Pthreads and OpenMP, are both thread-centric in that they are based on explicit management of tasks and enforce data dependencies and output ordering through task management. By comparison, the Cray XMT programming model is data-centric where the primary concern of the programmer is managing data dependencies, allowing threads to progress in a data flow fashion. The XMT implements this programming model by associating tag bits with each word of memory, affording efficient fine grained data synchronization independent of the number of processors or how tasks are scheduled. When task management is implicit and synchronization is abundant, efficient, and easy to use, programmers have viable alternatives to traditional thread-centric algorithms. In this paper we compare the amount of available parallelism relative to the amount of work in a variety of different algorithms and data structures when synchronization does not need to be rationed, as well as identify opportunities for platform and performance portability of the data-centric programming model on multi-core processors.

99 GENERAL AND MISCELLANEOUS↗

LU Factorization with Partial Pivoting for a Multi-CPU, Multi-GPU Shared Memory System

LU factorization with partial pivoting is a canonical numerical procedure and the main component of the High Performance LINPACK benchmark. This article presents an implementation of the algorithm for a hybrid, shared memory, system with standard CPU cores and GPU accelerators. Performance in excess of one TeraFLOPS is achieved using four AMD Magny Cours CPUs and four NVIDIA Fermi GPUs.

97 MATHEMATICS AND COMPUTING↗

Conditional load and store in a shared memory

A method, system and computer program product for implementing load-reserve and store-conditional instructions in a multi-processor computing system. The computing system includes a multitude of processor units and a shared memory cache, and each of the processor units has access to the memory cache. In one embodiment, the method comprises providing the memory cache with a series of reservation registers, and storing in these registers addresses reserved in the memory cache for the processor units as a result of issuing load-reserve requests. In this embodiment, when one of the processor units makes a request to store data in the memory cache using a store-conditional request, the reservation registers are checked to determine if an address in the memory cache is reserved for that processor unit. If an address in the memory cache is reserved for that processor, the data are stored at this address.

97 MATHEMATICS AND COMPUTING↗

Multi-core and many-core shared-memory parallel raycasting volume rendering optimization and tuning

Given the computing industry trend of increasing processing capacity by adding more cores to a chip, the focus of this work is tuning the performance of a staple visualization algorithm, raycasting volume rendering, for shared-memory parallelism on multi-core CPUs and many-core GPUs. Our approach is to vary tunable algorithmic settings, along with known algorithmic optimizations and two different memory layouts, and measure performance in terms of absolute runtime and L2 memory cache misses. Our results indicate there is a wide variation in runtime performance on all platforms, as much as 254% for the tunable parameters we test on multi-core CPUs and 265% on many-core GPUs, and the optimal configurations vary across platforms, often in a non-obvious way. For example, our results indicate the optimal configurations on the GPU occur at a crossover point between those that maintain good cache utilization and those that saturate computational throughput. This result is likely to be extremely difficult to predict with an empirical performance model for this particular algorithm because it has an unstructured memory access pattern that varies locally for individual rays and globally for the selected viewpoint. Our results also show that optimal parameters on modern architectures are markedly different from those in previous studies run on older architectures. In addition, given the dramatic performance variation across platforms for both optimal algorithm settings and performance results, there is a clear benefit for production visualization and analysis codes to adopt a strategy for performance optimization through auto-tuning. These benefits will likely become more pronounced in the future as the number of cores per chip and the cost of moving data through the memory hierarchy both increase.

97 MATHEMATICS AND COMPUTING↗

Multi-core and Many-core Shared-memory Parallel Raycasting Volume Rendering Optimization and Tuning

Given the computing industry trend of increasing processing capacity by adding more cores to a chip, the focus of this work is tuning the performance of a staple visualization algorithm, raycasting volume rendering, for shared-memory parallelism on multi-core CPUs and many-core GPUs. Our approach is to vary tunable algorithmic settings, along with known algorithmic optimizations and two different memory layouts, and measure performance in terms of absolute runtime and L2 memory cache misses. Our results indicate there is a wide variation in runtime performance on all platforms, as much as 254% for the tunable parameters we test on multi-core CPUs and 265% on many-core GPUs, and the optimal configurations vary across platforms, often in a non-obvious way. For example, our results indicate the optimal configurations on the GPU occur at a crossover point between those that maintain good cache utilization and those that saturate computational throughput. This result is likely to be extremely difficult to predict with an empirical performance model for this particular algorithm because it has an unstructured memory access pattern that varies locally for individual rays and globally for the selected viewpoint. Our results also show that optimal parameters on modern architectures are markedly different from those in previous studies run on older architectures. And, given the dramatic performance variation across platforms for both optimal algorithm settings and performance results, there is a clear benefit for production visualization and analysis codes to adopt a strategy for performance optimization through auto-tuning. These benefits will likely become more pronounced in the future as the number of cores per chip and the cost of moving data through the memory hierarchy both increase.

97 MATHEMATICS AND COMPUTING↗