This article presents an automatic approach to quickly derive a good solution for hardware resource partition and task granularity for task-based parallel applications on heterogeneous many-core architectures. Our approach employs a performance model to estimate the resulting performance of the target application under a given resource partition and task granularity configuration. The model is used as a utility to quickly search for a good configuration at runtime. Instead of hand-crafting an analytical model that requires expert insights into low-level hardware details, we employ machine learning techniques to automatically learn it. We achieve this by first learning a predictive model offline using training programs. The learnt model can then be used to predict the performance of any unseen program at runtime. We apply our approach to 39 representative parallel applications and evaluate it on two representative heterogeneous many-core platforms: a CPU-XeonPhi platform and a CPU-GPU platform. Compared to the single-stream version, our approach achieves, on average, a 1.6x and 1.1x speedup on the XeonPhi and the GPU platform, respectively. These results translate to over 93% of the performance delivered by a theoretically perfect predictor.
Task parallelism as employed by the OpenMP task construct, although ideal for tackling irregular problems or typical producer/consumer schemes, bears some potential for performance bottlenecks if locality of data access is important, which is typically the case for memory-bound code on ccNUMA systems. We present a programming technique which ameliorates adverse effects of dynamic task distribution by sorting tasks into locality queues, each of which is preferably processed by threads that belong to the same locality domain. Dynamic scheduling is fully preserved inside each domain, and is prefe…
Dynamic parallelism on GPUs allows GPU threads to dynamically launch other GPU threads. It is useful in applications with nested parallelism, particularly where the amount of nested parallelism is irregular and cannot be predicted beforehand. However, prior works have shown that dynamic parallelism may impose a high performance penalty when a large number of small grids are launched. The large number of launches results in high launch latency due to congestion, and the small grid sizes result in hardware underutilization. To address this issue, we propose a compiler framework for optimizing th…
Streaming computations on massive data sets are an attractive candidate for parallelization, particularly when they exhibit independence (and hence data parallelism) between items in the stream. However, some streaming computations are stateful, which disrupts independence and can limit parallelism. In this work, we consider how to extract data parallelism from streaming computations with a common, limited form of statefulness. The stream is assumed to be divided into variably-sized regions, and items in the same region are processed in a common context of state. In general, the computation to…
We present a model of multithreaded computation, combining fork-join and single-instruction-multiple-data parallelisms, with an emphasis on estimating parallelism overheads of programs written for modern many-core architectures. We establish a Graham-Brent theorem for this model so as to estimate execution time of programs running on a given number of streaming multiprocessors. We evaluate the benefits of our model with four fundamental algorithms from scientific computing. In each case, our model is used to minimize parallelism overheads by determining an appropriate value range for a given p…
Reactive molecular dynamics simulations are computationally demanding. Reaching spatial and temporal scales where interesting scientific phenomena can be observed requires efficient and scalable implementations on modern hardware. In this paper, we focus on optimizing the performance of the widely used LAMMPS/ReaxC package for multi-core architectures. As hybrid parallelism allows better leverage of the increasing on-node parallelism, we adopt thread parallelism in the construction of bonded and nonbonded lists, and in the computation of complex ReaxFF interactions. To mitigate the I/O overhea…
An important objective for analyzing real-world graphs is to achieve scalable performance on large, streaming graphs. A challenging and relevant example is the graph partition problem. As a combinatorial problem, graph partition is NP-hard, but existing relaxation methods provide reasonable approximate solutions that can be scaled for large graphs. Competitive benchmarks and challenges have proven to be an effective means to advance state-of-the-art performance and foster community collaboration. This paper describes a graph partition challenge with a baseline partition algorithm of sub-quadra…
Using parallel embedded systems these days is increasing. They are getting more complex due to integrating multiple functionalities in one application or running numerous ones concurrently. This concerns a wide range of applications, including streaming applications, commonly used in embedded systems. These applications must implement adaptable and reliable algorithms to deliver the required performance under varying circumstances (e.g., running applications on the platform, input data, platform variety, etc.). Given the complexity of streaming applications, target systems, and adaptivity requ…
University of Nevada, Las Vegas（内华达大学拉斯维加斯分校）；Southern China Institute of Technology（华南理工学院）
The increasing popularity of deep neural network (DNN) applications demands high computing power and efficient hardware accelerator architecture. DNN accelerators use a large number of processing elements (PEs) and on-chip memory for storing weights and other parameters. As the communication backbone of a DNN accelerator, networks-on-chip (NoC) play an important role in supporting various dataflow patterns and enabling processing with communication parallelism in a DNN accelerator. However, the widely used mesh-based NoC architectures inherently cannot support the efficient one-to-many and man…
Fault tolerance is a property which needs deeper consideration when dealing with streaming jobs requiring high levels of availability and low-latency processing even in case of failures where Quality-of-Service constraints must be adhered to. Typically, systems achieve fault tolerance and the ability to recover automatically from partial failures by implementing Checkpoint and Rollback Recovery. However, this is an expensive operation which impacts negatively on the overall performance of the system and manually optimizing fault tolerance for specific jobs is a difficult and time consuming tas…
Parallel architectures are essential in order to take advantage of the parallelism inherent in streaming applications. One particular branch of these employ hardware SIMD pipelines. In this paper, we analyse several scheduling techniques, namely ad hoc overlapped execution, modulo scheduling and modulo scheduling with unrolling, all of which aim to efficiently utilize the special architecture design. Our investigation focuses on improving throughput while analysing other metrics that are important for streaming applications, such as register pressure, buffer sizes and code size. Through experi…
C is the lingua franca of programming and almost any device can be programmed using C. However, programming mod-ern heterogeneous architectures such as multi-core CPUs and GPUs requires explicitly expressing parallelism as well as device-specific properties such as memory hierarchies. The resulting code is often hard to understand, debug, and modify for different architectures. We propose to lift C programs to a parametric dataflow representation that lends itself to static data-centric analysis and enables automatic high-performance code generation. We separate writing code from optimizing fo…
Spatial computing architectures pose an attractive alternative to mitigate control and data movement overheads typical of load-store architectures. In practice, these devices are rarely considered in the HPC community due to the steep learning curve, low productivity and lack of available libraries for fundamental operations. High-level synthesis (HLS) tools are facilitating hardware programming, but optimizing for these architectures requires factoring in new transformations and resources/performance trade-offs. We present FBLAS, an open-source HLS implementation of BLAS for FPGAs, that enabl…
Task parallelism as employed by the OpenMP task construct or some Intel Threading Building Blocks (TBB) components, although ideal for tackling irregular problems or typical producer/consumer schemes, bears some potential for performance bottlenecks if locality of data access is important, which is typically the case for memory-bound code on ccNUMA systems. We present a thin software layer ameliorates adverse effects of dynamic task distribution by sorting tasks into locality queues, each of which is preferably processed by threads that belong to the same locality domain. Dynamic scheduling is…
Occult: Optimizing Collaborative Communication across Experts for Accelerated Parallel MoE Training and Inference
The University of North Carolina at Chapel Hill（北卡罗来纳大学教堂山分校）；University of California, Los Angeles（加州大学洛杉矶分校）；University of Minnesota Twin Cities（明尼苏达大学双城分校）；The Chinese University of Hong Kong（香港中文大学）
While parallelism remains the main source of performance, architectural implementations and programming models change with each new hardware generation, often leading to costly application re-engineering. Most tools for performance portability require manual and costly application porting to yet another programming model. We propose an alternative approach that automatically translates programs written in one programming model (CUDA), into another (CPU threads) based on Polygeist/MLIR. Our approach includes a representation of parallel constructs that allows conventional compiler transformatio…
Telefonica Research（西班牙电信研究）；Qatar Computing Research Institute（卡塔尔计算研究所）；Telecom ParisTech（巴黎高等电信学院）；LARC-SMU（新加坡管理大学智能系统实验室）
IoT Big Data requires new machine learning methods able to scale to large size of data arriving at high speed. Decision trees are popular machine learning models since they are very effective, yet easy to interpret and visualize. In the literature, we can find distributed algorithms for learning decision trees, and also streaming algorithms, but not algorithms that combine both features. In this paper we present the Vertical Hoeffding Tree (VHT), the first distributed streaming algorithm for learning decision trees. It features a novel way of distributing decision trees via vertical parallelis…
Graph partitioning plays a vital role in distributedlarge-scale web graph analytics, such as pagerank and labelpropagation. The quality and scalability of partitioning strategyhave a strong impact on such communication- and computation-intensive applications, since it drives the communication costand the workload balance among distributed computing nodes.Recently, the streaming model shows promise in optimizing graphpartitioning. However, existing streaming partitioning strategieseither lack of adequate quality or fall short in scaling with alarge number of partitions.In this work, we explore …
Region proposal is critical for object detection while it usually poses a bottleneck in improving the computation efficiency on traditional control-flow architectures. We have observed region proposal tasks are potentially suitable for performing pipelined parallelism by exploiting dataflow driven acceleration. In this paper, a scalable pipelined dataflow accelerator is proposed for efficient region proposals on FPGA platform. The accelerator processes image data by a streaming manner with three sequential stages: resizing, kernel computing and sorting. First, Ping-Pong cache strategy is adopt…
