This paper introduces Pandora, a parallel algorithm for computing dendrograms, the hierarchical cluster trees for single linkage clustering (SLC). Current parallel approaches construct dendrograms by partitioning a minimum spanning tree and removing edges. However, they struggle with skewed, hard-to-parallelize real-world dendrograms. Consequently, computing dendrograms is the sequential bottleneck in HDBSCAN*[21], a popular SLC variant.

This study presents the first constrained sparse tensor factorization (cSTF) framework that optimizes and fully offloads computation to massively parallel GPU architectures, and the first performance characterization of cSTF on GPU architectures. In contrast to prior work on tensor factorization, where the matricized tensor times Khatri-Rao product (MTTKRP) is the primary performance bottleneck, our systematic analysis of the cSTF algorithm on GPUs reveals that adding constraints creates an additional bottleneck in the update operation for many real-world sparse tensors. While executing the update operation on the GPU brings significant speedup over its CPU counterpart, it remains a significant bottleneck. To further accelerate the update operation, we propose cuADMM, a new update algorithm that leverages algorithmic and code optimization strategies to minimize both computation and data movement on GPUs. As a result, our framework delivers significantly improved performance compared to prior state-of-the-art. On 10 real-world sparse tensors, our framework achieves geometric mean speedup of 5.1 × (max 41.59 ×) and 7.01 × (max 58.05 ×) on the NIVIDA A100 and H100 GPUs, respectively, over the state-of-the-art SPLATT library running on a 26-core Intel Ice Lake Xeon CPU.

The new reality of exascale computing faces many challenges in achieving optimal performance on large numbers of nodes. A key challenge is the efficient utilization of the message-passing interface (MPI), a critical component for process communication. This paper explores communication optimization strategies to harness the GPU-accelerated architectures of these supercomputers. We focus on MPI applications where processors form a two-dimensional process grid, a common arrangement in applications involving dense matrix operations. This configuration offers a unique opportunity to implement innovative strategies to improve performance and maintain effective load distribution. We study two applications— Dist-FW (Apsp:all-pair-shortest-path) and HPL-MxP (LU factorization with Mixed precision)—on two accelerated systems: Summit (IBM Power and NVIDIA V100) and Frontier (AMD EPYC and MI250X). These supercomputers are operated by the Oak Ridge Leadership Computing Facility (OLCF) and are currently ranked #1 and #5 on the Top500 list. We show how to scale up both applications to exascale levels and tackle the MPI challenges related to implementation, synchronization, and performance. We also compare the performance of several communication strategies at an unprecedented scale. Accurately predicting application performance becomes crucial for cost reduction as the computational scale grows. To address this, we suggest a hyperbolic model as a better alternative to the traditional one-sided asymptotic model for predicting future application performance at such large scales.

This paper presents a unified communication optimization framework for sparse triangular solve (SpTRSV) algorithms on CPU and GPU clusters. The framework builds upon a 3D communication-avoiding (CA) layout of Px × Py × Pz processes that divides a sparse matrix into Pz submatrices, each handled by a Px × Py 2D grid with block-cyclic distribution. We propose three communication optimization strategies: First, a new 3D SpTRSV algorithm is developed, which trades the inter-grid communication and synchronization with replicated computation. This design requires only one inter-grid synchronization, and the inter-grid communication is efficiently implemented with sparse allreduce operations. Second, broadcast and reduction communication trees are used to reduce message latency of the intra-grid 2D communication on CPU clusters. Finally, we leverage GPU-initiated one-sided communication to implement the communication trees on GPU clusters. With these nested inter- and intra-grid communication optimization strategies, the proposed 3D SpTRSV algorithm can attain up to 3.45x speedups compared to the baseline 3D SpTRSV algorithm using up to 2048 Cori Haswell CPU cores. In addition, the proposed GPU 3D SpTRSV algorithm can achieve up to 6.5x speedups compared to the proposed CPU 3D SpTRSV algorithm with Pz up to 64. Finally it is remarkable that the proposed GPU 3D SpTRSV can scale to 256 GPUs using the Perlmutter system while the existing 2D SpTRSV algorithm can only scale up to 4 GPUs.

We introduce a new parallel algorithm for solving sparse LU factorization of planar matrices, which commonly arise in the finite element method for 2D PDEs. Existing scalable methods, such as the multifrontal approach with subtree-to-subcube mapping by Gupta et al. [1] and right-looking with 3D mapping by Sao et al. [2] fail to achieve optimal communication costs for these matrices. Our new algorithm combines 3D mapping and subtree-to-subcube mapping to minimize communication costs while allowing trade-offs between extra memory and reduced communication. We demonstrate that our proposed algorithm attains the communication lower bound up to a factor of O(log log n) in the memory-optimal case and up to a factor of O(log P) in the memory-independent case for an n-dimensional planar sparse matrix on P processors.

We present the new features available in the recent release of SuperLU_DIST , Version 8.1.1. SuperLU_DIST is a distributed-memory parallel sparse direct solver. The new features include (1) a 3D communication-avoiding algorithm framework that trades off inter-process communication for selective memory duplication, (2) multi-GPU support for both NVIDIA GPUs and AMD GPUs, and (3) mixed-precision routines that perform single-precision LU factorization and double-precision iterative refinement. Apart from the algorithm improvements, we also modernized the software build system to use CMake and Spack package installation tools to simplify the installation procedure. Throughout the article, we describe in detail the pertinent performance-sensitive parameters associated with each new algorithmic feature, show how they are exposed to the users, and give general guidance of how to set these parameters. We illustrate that the solver’s performance both in time and memory can be greatly improved after systematic tuning of the parameters, depending on the input sparse matrix and underlying hardware.

Computing the Euclidean minimum spanning tree (Emst) is a computationally demanding step of many algorithms. While work-efficient serial and multithreaded algorithms for computing Emst are known, designing an efficient GPU algorithm is challenging due to a complex branching structure, data dependencies, and load imbalances. In this paper, we propose a single-tree Borůvka-based algorithm for computing Emst on GPUs. We use an efficient nearest neighbor algorithm and reduce the number of the required distance calculations by avoiding traversing subtrees with leaf nodes in the same component. The developed algorithms are implemented in a performance portable way using ArborX, an open-source geometric search library based on the Kokkos framework. We evaluate the proposed algorithm on various 2D and 3D datasets, show and compare it with the current state-of-the-art open-source CPU implementations. We demonstrate 4-24 × speedup over the fastest multi-threaded implementation. We prove the portability of our implementation by providing results on a variety of hardware: AMD EPYC 7763, Nvidia A100 and AMD MI250X. We show scalability of the implementation, computing Emst for 37 million 3D cosmological dataset in under a 0.5 second on a single A100 Nvidia GPU.

We are motivated by newly proposed methods for mining large-scale corpora of scholarly publications (e.g., full biomedical literature), which consists of tens of millions of papers spanning decades of research. In this setting, analysts seek to discover relationships among concepts. They construct graph representations from annotated text databases and then formulate the relationship-mining problem as an all-pairs shortest paths (APSP) and validate connective paths against curated biomedical knowledge graphs (e.g., Spoke). In this context, we present Coast (Exascale Communication-Optimized All-Pairs Shortest Path) and demonstrate 1.004 EF/s on 9,200 Frontier nodes (73,600 GCDs). We develop hyperbolic performance models (HYPERMOD), which guide optimizations and parametric tuning. The proposed Coast algorithm achieved the memory constant parallel efficiency of 99% in the single-precision tropical semiring. Looking forward, Coast will enable the integration of scholarly corpora like PubMed into the Spoke biomedical knowledge graph.

We present an optimized Floyd-Warshall (Floyd-Warshall) algorithm that computes the All-pairs shortest path (APSP) for GPU accelerated clusters. The Floyd-Warshall algorithm due to its structural similarities to matrix-multiplication is well suited for highly parallel GPU architectures. To achieve high parallel efficiency, we address two key algorithmic challenges: reducing high communication overhead and addressing limited GPU memory. To reduce high communication costs, we redesign the parallel (a) to expose more parallelism, (b) aggressively overlap communication and computation with pipelined and asynchronous scheduling of operations, and (c) tailored MPI-collective. To cope with limited GPU memory, we employ an offload model, where the data resides on the host and is transferred to GPU on-demand. The proposed optimizations are supported with detailed performance models for tuning. Our optimized parallel Floyd-Warshall implementation is up to 5x faster than a strong baseline and achieves 8.1 PetaFLOPS/sec on 256~nodes of the Summit supercomputer at Oak Ridge National Laboratory. This performance represents 70% of the theoretical peak and 80% parallel efficiency. The offload algorithm can handle 2.5x larger graphs with a 20% increase in overall running time.

We show how to exploit graph sparsity in the Floyd-Warshall algorithm for the all-pairs shortest path (Apsp) problem. Floyd-Warshall is an attractive choice for Apsp on high-performing systems due to its structural similarity to solving dense linear systems and matrix multiplication. However, if sparsity of the input graph is not properly exploited, Floyd-Warshall will perform unnecessary asymptotic work and thus may not be a suitable choice for many input graphs. To overcome this limitation, the key idea in our approach is to use the known algebraic relationship between Floyd-Warshall and Gaussian elimination, and import several algorithmic techniques from sparse Cholesky factorization, namely, fill-in reducing ordering, symbolic analysis, supernodal traversal, and elimination tree parallelism. When combined, these techniques reduce computation, improve locality and enhance parallelism. We implement these ideas in an efficient shared memory parallel prototype that is orders of magnitude faster than an efficient multi-threaded baseline Floyd-Warshall that does not exploit sparsity. Our experiments suggest that the Floyd-Warshall algorithm can compete with Dijkstra's algorithm (the algorithmic core of Johnson's algorithm) for several classes sparse graphs.

We are motivated by newly proposed methods for data mining large-scale corpora of scholarly publications, such as the full biomedical literature, which may consist of tens of millions of papers spanning decades of research. In this setting, analysts seek to discover how concepts relate to one another. They construct graph representations from annotated text databases and then formulate the relationship-mining problem as one of computing all-pairs shortest paths (APSP), which becomes a significant bottleneck. In this context, we present a new high-performance algorithm and implementation of the Floyd-Warshall algorithm for distributed-memory parallel computers accelerated by GPUs, which we call DSNAPSHOT (Distributed Accelerated Semiring All-Pairs Shortest Path). For our largest experiments, we ran DSNAPSHOT on a connected input graph with millions of vertices using 4, 096nodes (24,576GPUs) of the Oak Ridge National Laboratory's Summit supercomputer system. We find DSNAPSHOT achieves a sustained performance of $136\times 10^{15}$ floating-point operations per second (136petaflop/s) at a parallel efficiency of 90% under weak scaling and, in absolute speed, 70% of the best possible performance given our computation (in the single-precision tropical semiring or “min-plus” algebra). Looking forward, we believe this novel capability will enable the mining of scholarly knowledge corpora when embedded and integrated into artificial intelligence-driven natural language processing workflows at scale.

Due to the limited capacity of GPU memory, the majority of prior work on graph applications on GPUs has been restricted to graphs of modest sizes that fit in memory. Recent hardware and software advances make it possible to address much larger host memory transparently as a part of a feature known as unified virtual memory. While accessing host memory over an interconnect is understandably slower, the problem space has not been sufficiently explored in the context of a challenging workload with low computational intensity and an irregular data access pattern such as graph traversal. We analyse the performance of breadth first search (BFS) for several large graphs in the context of unified memory and identify the key factors that contribute to slowdowns. Next, we propose a lightweight offline graph reordering algorithm, HALO (Harmonic Locality Ordering), that can be used as a pre-processing step for static graphs. HALO yields speedups of 1.5x-1.9x over baseline in subsequent traversals. Our method specifically aims to cover large directed real world graphs in addition to undirected graphs whereas prior methods only account for the latter. Additionally, we demonstrate ties between the locality ordering problem and graph compression and show that prior methods from graph compression such as recursive graph bisection can be suitably adapted to this problem.

We present a novel distributed memory algorithm to improve the strong scalability of the solution of a sparse triangular system. This operation appears in the solve phase of direct methods for solving general sparse linear systems, Ax = b. Our 3D sparse triangular solver employs several techniques, including a 3D MPI process grid, elimination tree parallelism, and data replication, all of which reduce the per-process communication when combined. We present analytical models to understand the communication cost of our algorithm and show that our 3D sparse triangular solver can reduce the per-process communication volume asymptotically by a factor of O(n1/4) and O(n1/6) for problems arising from the finite element discretizations of 2D "planar" and 3D "non-planar" PDEs, respectively. We implement our algorithm for use in SuperLU_DIST3D, using a hybrid MPI+OpenMP programming model. Our 3D triangular solve algorithm, when run on 12k cores of Cray XC30, outperforms the current state-of-the-art 2D algorithm by 7.2x for planar and 2.7x for the non-planar sparse matrices, respectively.

We propose a new algorithm to improve the strong scalability of right-looking sparse LU factorization on distributed memory systems. Our 3D sparse LU algorithm uses a three-dimensional MPI process grid, aggressively exploits elimination tree parallelism and trades off increased memory for reduced per-process communication. We also analyze the asymptotic improvements for planar graphs (e.g., from 2D grid or mesh domains) and certain non-planar graphs (specifically for 3D grids and meshes). For planar graphs with n vertices, our algorithm reduces communication volume asymptotically in n by a factor of O{log n} and latency by a factor of O{log n}. For non-planar cases, our algorithm can reduce the per-process communication volume by 3× and latency by O{n^1/3} times. In all cases, the memory needed to achieve these gains is a constant factor. We implemented our algorithm by extending the 2D data structure used in superLU. Our new 3D code achieves speedups up to 27× for planar graphs and up to 3.3× for non-planar graphs over the baseline 2D superLU when run on 24,000 cores of a Cray XC30.

This paper presents the first sparse direct solver for distributed memory systems comprising hybrid multicourse CPU and Intel Xeon Pico-processors. It builds on the algorithmic approach of SuperLU_DIST, which is right-looking and statically pivoted. Our contribution is a novel algorithm, called the HALO. The name is shorthand for highly asynchronous lazy offload, it refers tithe way the algorithm combines highly aggressive use of asynchrony with accelerated offload, lazy updates, and data shadowing (a la halo or ghost zones), all of which serve to hide and reduce communication, whether to local memory, across the network, or over PCIe. We further augment HALO with a model-driven autotuning heuristicthat chooses the intra-node division of labor among CPU and Xeon Pico-processor components. When integrated into SuperLU_DIST and evaluated on a variety of realistic test problems in both single-node and multi-node configurations, the resulting implementation achieves speedups of unto 2.5× over an already efficient multicourse CPU implementation, and achieves up to 83% of a machine-specific upper-bound that we haveestimated. Our analysis quantifies how well our implementation performs and allows us to speculate on the potential speedups that might come from variety of future improvements to the algorithm and system.