High-performance implementations of graph algorithms are challenging to implement on new parallel hardware such as GPUs because of three challenges: (1) the difficulty of coming up with graph building blocks, (2) load imbalance on parallel hardware, and (3) graph problems having low arithmetic intensity. To address some of these challenges, GraphBLAS is an innovative, on-going effort by the graph analytics community to propose building blocks based on sparse linear algebra, which will allow graph algorithms to be expressed in a performant, succinct, composable and portable manner. In this paper, we examine the performance challenges of a linear-algebra-based approach to building graph frameworks and describe new design principles for overcoming these bottlenecks. Among the new design principles is exploiting input sparsity, which allows users to write graph algorithms without specifying push and pull direction. Exploiting output sparsity allows users to tell the backend which values of the output in a single vectorized computation they do not want computed. Load-balancing is an important feature for balancing work amongst parallel workers. We describe the important load-balancing features for handling graphs with different characteristics. The design principles described in this paper have been implemented in "GraphBLAST", the first high-performance linear algebra-based graph framework on NVIDIA GPUs that is open-source. The results show that on a single GPU, GraphBLAST has on average at least an order of magnitude speedup over previous GraphBLAS implementations SuiteSparse and GBTL, comparable performance to the fastest GPU hardwired primitives and shared-memory graph frameworks Ligra and Gunrock, and better performance than any other GPU graph framework, while offering a simpler and more concise programming model.
Graph Neural Networks (GNNs) have achieved significant improvements in various domains. Sparse Matrix-Matrix multiplication (SpMM) is a fundamental operator in GNNs, which performs a multiplication between a sparse matrix and a dense matrix. Accelerating SpMM on parallel hardware like GPUs can face the following challenges: From the GNN application perspective, the compatibility needs to be considered. General GNN algorithms require SpMM-like operations (e.g., pooling) between matrices, which are not supported in current high-performance GPU libraries (e.g., Nvidia cuSPARSE). Moreover, the sop…
With the announcement that the Aurora Supercomputer will be composed of general purpose Intel CPUs complemented by discrete high performance Intel GPUs, and the deployment of the oneAPI ecosystem, Intel has committed to enter the arena of discrete high performance GPUs. A central requirement for the scientific computing community is the availability of production-ready software stacks and a glimpse of the performance they can expect to see on Intel high performance GPUs. In this paper, we present the first platform-portable open source math library supporting Intel GPUs via the DPC++ programmi…
In this paper, we investigate GPU based parallel triangular solvers systematically. The parallel triangular solvers are fundamental to incomplete LU factorization family preconditioners and algebraic multigrid solvers. We develop a new matrix format suitable for GPU devices. Parallel lower triangular solvers and upper triangular solvers are developed for this new data structure. With these solvers, ILU preconditioners and domain decomposition preconditioners are developed. Numerical results show that we can speed triangular solvers around seven times faster.
Sparse linear algebra kernels play a critical role in numerous applications, covering from exascale scientific simulation to large-scale data analytics. Offloading linear algebra kernels on one GPU will no longer be viable in these applications, simply because the rapidly growing data volume may exceed the memory capacity and computing power of a single GPU. Multi-GPU systems nowadays being ubiquitous in supercomputers and data-centers present great potentials in scaling up large sparse linear algebra kernels. In this work, we design a novel sparse matrix representation framework for multi-GPU…
A new scalable parallel math library, dMath, is presented in this paper that demonstrates leading scaling when using intranode, or internode, hybrid-parallelism for deep-learning. dMath provides easy-to-use distributed base primitives and a variety of domain-specific algorithms. These include matrix multiplication, convolutions, and others allowing for rapid development of highly scalable applications, including Deep Neural Networks (DNN), whereas previously one was restricted to libraries that provided effective primitives for only a single GPU, like Nvidia cublas and cudnn or DNN primitives …
GPUs are dedicated processors used for complex calculations and simulations and they can be effectively used for tropical algebra computations. Tropical algebra is based on max-plus algebra and min-plus algebra. In this paper we proposed and designed a library based on Tropical Algebra which is used to provide standard vector and matrix operations namely Basic Tropical Algebra Subroutines (BTAS). The testing of BTAS library is conducted by implementing the sequential version of Floyd Warshall Algorithm on CPU and furthermore parallel version on GPU. The developed library for tropical algebra d…
As part of the Exascale Computing Project (ECP), a recent focus of development efforts for the SUite of Nonlinear and DIfferential/ALgebraic equation Solvers (SUNDIALS) has been to enable GPU-accelerated time integration in scientific applications at extreme scales. This effort has resulted in several new GPU-enabled implementations of core SUNDIALS data structures, support for programming paradigms which are aware of the heterogeneous architectures, and the introduction of utilities to provide new points of flexibility. In this paper, we discuss our considerations, both internal and external,…
This paper details an extensible OpenCL framework that allows Stan to utilize heterogeneous compute devices. It includes GPU-optimized routines for the Cholesky decomposition, its derivative, other matrix algebra primitives and some commonly used likelihoods, with more additions planned for the near future. Stan users can now benefit from large speedups offered by GPUs with little effort and without changes to their existing Stan code. We demonstrate the practical utility of our work with two examples - logistic regression and Gaussian Process regression.
Sparse compiler is a promising solution for sparse tensor algebra optimization. In compiler implementation, reduction in sparse-dense hybrid algebra plays a key role in performance. Though GPU provides various reduction semantics that can better utilize the parallel computing and memory bandwidth capacity, the central question is: how to elevate the flexible reduction semantics to sparse compilation theory that assumes serial execution. Specifically, we have to tackle two main challenges: (1) there are wasted parallelism by adopting static synchronization granularity (2) static reduction strat…
Linear algebra operations have been widely used in big data analytics and scientific computations. Many works have been done on optimizing linear algebra operations on GPUs with regular-shaped input. However, few works focus on fully utilizing GPU resources when the input is not regular-shaped. Current optimizations do not consider fully utilizing the memory bandwidth and computing power; therefore, they can only achieve sub-optimal performance. In this paper, we propose two efficient algorithms -- TSM2R and TSM2L -- for two classes of tall-and-skinny matrix-matrix multiplications on GPUs. Bot…
The paper presents the aspect of use of modern graphics accelerators supporting CUDA technology for high-performance computing in the field of linear algebra. Fully programmable graphic cards have been available for several years for both ordinary users and research units. They provide the capability of performing virtually any computing with high performance, which is often beyond the reach of conventional CPUs. GPU architecture, also in case of classical problems of linear algebra which is the basis for many calculations, can bring many benefits to the developer. Performance increase, observ…
Basic Linear Algebra Subprograms (BLAS) are a set of low level linear algebra kernels widely adopted by applications involved with the deep learning and scientific computing. The massive and economic computing power brought forth by the emerging GPU architectures drives interest in implementation of compute-intensive level 3 BLAS on multi-GPU systems. In this paper, we investigate existing multi-GPU level 3 BLAS and present that 1) issues, such as the improper load balancing, inefficient communication, insufficient GPU stream level concurrency and data caching, impede current implementations f…
Fast computation of singular value decomposition (SVD) is of great interest in various machine learning tasks. Recently, SVD methods based on randomized linear algebra have shown significant speedup in this regime. This paper attempts to further accelerate the computation by harnessing a modern computing architecture, namely graphics processing unit (GPU), with the goal of processing large-scale data that may not fit in the GPU memory. It leads to a new block randomized algorithm that fully utilizes the power of GPUs and efficiently processes large-scale data in an out-of- core fashion. Our ex…
Image Processing is a specialized area of Digital Signal Processing which contains various mathematical and algebraic operations such as matrix inversion, transpose of matrix, derivative, convolution, Fourier Transform etc. Operations like those require higher computational capabilities than daily usage purposes of computers. At that point, with increased image sizes and more complex operations, CPUs may be unsatisfactory since they use Serial Processing by default. GPUs are the solution that come up with greater speed compared to CPUs because of their Parallel Processing/Computation nature. A…
In this work we present a performance exploration on Eager K-truss, a linear-algebraic formulation of the K-truss graph algorithm. We address performance issues related to load imbalance of parallel tasks in symmetric, triangular graphs by presenting a fine-grained parallel approach to executing the support computation. This approach also increases available parallelism, making it amenable to GPU execution. We demonstrate our fine-grained parallel approach using implementations in Kokkos and evaluate them on an Intel Skylake CPU and an Nvidia Tesla V100 GPU. Overall, we observe between a 1.261…
Reduction operations are extensively employed in many computational problems. A reduction consists of, given a finite set of numeric elements, combining into a single value all elements in that set, using for this a combiner function. A parallel reduction, in turn, is the reduction operation concurrently performed when multiple execution units are available. The current work reports an investigation on this subject and depicts a GPU-based parallel approach for it. Employing techniques like Loop Unrolling, Persistent Threads and Algebraic Expressions to avoid thread divergence, the presented ap…
While many of the architectural details of future exascale-class high performance computer systems are still a matter of intense research, there appears to be a general consensus that they will be strongly heterogeneous, featuring "standard" as well as "accelerated" resources. Today, such resources are available as multicore processors, graphics processing units (GPUs), and other accelerators such as the Intel Xeon Phi. Any software infrastructure that claims usefulness for such environments must be able to meet their inherent challenges: massive multi-level parallelism, topology, asynchronici…
LISN, CNRS, Université Paris-Saclay（巴黎萨克雷大学国家科学研究中心LISN实验室）；Brno University of Technology（布尔诺理工大学）
We propose to express the forward-backward algorithm in terms of operations between sparse matrices in a specific semiring. This new perspective naturally leads to a GPU-friendly algorithm which is easy to implement in Julia or any programming languages with native support of semiring algebra. We use this new implementation to train a TDNN with the LF-MMI objective function and we compare the training time of our system with PyChain - a recently introduced C++/CUDA implementation of the LF-MMI loss. Our implementation is about two times faster while not having to use any approximation such as …
The paper presents a parallel math library, dMath, that demonstrates leading scaling when using intranode, internode, and hybrid-parallelism for deep learning (DL). dMath provides easy-to-use distributed primitives and a variety of domain-specific algorithms including matrix multiplication, convolutions, and others allowing for rapid development of scalable applications like deep neural networks (DNNs). Persistent data stored in GPU memory and advanced memory management techniques avoid costly transfers between host and device. dMath delivers performance, portability, and productivity to its s…
We equip dynamic geometry software (DGS) with a user-friendly method that enables massively parallel calculations on the graphics processing unit (GPU). This interplay of DGS and GPU opens up various applications in education and mathematical research. The GPU-aided discovery of mathematical properties, interactive visualizations of algebraic surfaces (raycasting), the mathematical deformation of images and footage in real-time, and computationally demanding numerical simulations of PDEs are examples from the long and versatile list of new domains that our approach makes accessible within a DG…
The widely-adopted practice is to train deep learning models with specialized hardware accelerators, e.g., GPUs or TPUs, due to their superior performance on linear algebra operations. However, this strategy does not employ effectively the extensive CPU and memory resources -- which are used only for preprocessing, data transfer, and scheduling -- available by default on the accelerated servers. In this paper, we study training algorithms for deep learning on heterogeneous CPU+GPU architectures. Our two-fold objective -- maximize convergence rate and resource utilization simultaneously -- make…
We introduce Stream-K, a work-centric parallelization of matrix multiplication (GEMM) and related computations in dense linear algebra. Whereas contemporary decompositions are primarily tile-based, our method operates by partitioning an even share of the aggregate inner loop iterations among physical processing elements. This provides a near-perfect utilization of computing resources, regardless of how efficiently the output tiling for any given problem quantizes across the underlying processing elements. On GPU processors, our Stream-K parallelization of GEMM produces a peak speedup of up to …
Fine-grained workload and resource balancing is the key to high performance for regular and irregular computations on the GPUs. In this dissertation, we conduct an extensive survey of existing load-balancing techniques to build an abstraction that addresses the difficulty of scheduling computations on the GPU. We propose a GPU fine-grained load-balancing abstraction that decouples load balancing from work processing and aims to support both static and dynamic schedules with a programmable interface to implement new load-balancing schedules. Prior to our work, the only way to unleash the GPU's …
Computation in several real-world applications like probabilistic machine learning, sparse linear algebra, and robotic navigation, can be modeled as irregular directed acyclic graphs (DAGs). The irregular data dependencies in DAGs pose challenges to parallel execution on general-purpose CPUs and GPUs, resulting in severe under-utilization of the hardware. This paper proposes DPU, a specialized processor designed for the efficient execution of irregular DAGs. The DPU is equipped with parallel compute units that execute different subgraphs of a DAG independently. The compute units can synchroniz…
We introduce DISTAL, a compiler for dense tensor algebra that targets modern distributed and heterogeneous systems. DISTAL lets users independently describe how tensors and computation map onto target machines through separate format and scheduling languages. The combination of choices for data and computation distribution creates a large design space that includes many algorithms from both the past (e.g., Cannon's algorithm) and the present (e.g., COSMA). DISTAL compiles a tensor algebra domain specific language to a distributed task-based runtime system and supports nodes with multi-core CPU…
Efficient execution of SPARQL queries over large RDF datasets is a topic of considerable interest due to increased use of RDF to encode data. Most of this work has followed either relational or graph-based approaches. In this paper, we propose an alternative query engine, called gSmart, based on matrix algebra. This approach can potentially better exploit the computing power of high-performance heterogeneous architectures that we target. gSmart incorporates: (1) grouped incident edge-based SPARQL query evaluation, in which all unevaluated edges of a vertex are evaluated together using a series…
There is an increased interest in building data analytics frameworks with advanced algebraic capabilities both in industry and academia. Many of these frameworks, e.g., TensorFlow and BIDMach, implement their compute-intensive primitives in two flavors---as multi-thread routines for multi-core CPUs and as highly-parallel kernels executed on GPU. Stochastic gradient descent (SGD) is the most popular optimization method for model training implemented extensively on modern data analytics platforms. While the data-intensive properties of SGD are well-known, there is an intense debate on which of t…
