Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userEfficient Sparse Matrix Kernels based on Adaptive Workload-Balancing and Parallel-Reduction
82
0
0.0
(
0
)
Added by
Guyue Huang
Publication date
2021
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
Sparse matrix-vector and matrix-matrix multiplication (SpMV and SpMM) are fundamental in both conventional (graph analytics, scientific computing) and emerging (sparse DNN, GNN) domains. Workload-balancing and parallel-reduction are widely-used design principles for efficient SpMV. However, prior work fails to resolve how to implement and adaptively use the two principles for SpMV/MM. To overcome this obstacle, we first complete the implementation space with optimizations by filling three missing pieces in prior work, including: (1) We show that workload-balancing and parallel-reduction can be combined through a segment-reduction algorithm implemented with SIMD-shuffle primitives. (2) We show that parallel-reduction can be implemented in SpMM through loading the dense-matrix rows with vector memory operations. (3) We show that vectorized loading of sparse rows, being a part of the benefit of parallel-reduction, can co-exist with sequential-reduction in SpMM through temporally caching sparse-matrix elements in the shared memory. In terms of adaptive use, we analyze how the benefit of two principles change with two characteristics from the input data space: the diverse sparsity pattern and dense-matrix width. We find the benefit of the two principles fades along with the increased total workload, i.e. the increased dense-matrix width. We also identify, for SpMV and SpMM, different sparse-matrix features that impact workload-balancing effectiveness. Our design consistently exceeds cuSPARSE by 1.07-1.57x on different GPUs and dense matrix width, and the kernel selection rules involve 5-12% performance loss compared with optimal choices. Our kernel is being integrated into popular graph learning frameworks to accelerate GNN training.
rate research
Read More
Matrix factorizations are among the most important building blocks of scientific computing. State-of-the-art libraries, however, are not communication-optimal, underutilizing current parallel architectures. We present novel algorithms for Cholesky and LU factorizations that utilize an asymptotically communication-optimal 2.5D decomposition. We first establish a theoretical framework for deriving parallel I/O lower bounds for linear algebra kernels, and then utilize its insights to derive Cholesky and LU schedules, both communicating N^3/(P*sqrt(M)) elements per processor, where M is the local memory size. The empirical results match our theoretical analysis: our implementations communicate significantly less than Intel MKL, SLATE, and the asymptotically communication-optimal CANDMC and CAPITAL libraries. Our code outperforms these state-of-the-art libraries in almost all tested scenarios, with matrix sizes ranging from 2,048 to 262,144 on up to 512 CPU nodes of the Piz Daint supercomputer, decreasing the time-to-solution by up to three times. Our code is ScaLAPACK-compatible and available as an open-source library.
Fast domain propagation of linear constraints has become a crucial component of todays best algorithms and solvers for mixed integer programming and pseudo-boolean optimization to achieve peak solving performance. Irregularities in the form of dynamic algorithmic behaviour, dependency structures, and sparsity patterns in the input data make efficient implementations of domain propagation on GPUs and, more generally, on parallel architectures challenging. This is one of the main reasons why domain propagation in state-of-the-art solvers is single thread only. In this paper, we present a new algorithm for domain propagation which (a) avoids these problems and allows for an efficient implementation on GPUs, and is (b) capable of running propagation rounds entirely on the GPU, without any need for synchronization or communication with the CPU. We present extensive computational results which demonstrate the effectiveness of our approach and show that ample speedups are possible on practically relevant problems: on state-of-the-art GPUs, our geometric mean speed-up for reasonably-large instances is around 10x to 20x and can be as high as 180x on favorably-large instances.
Performance and energy are the two most important objectives for optimisation on modern parallel platforms. Latest research demonstrated the importance of workload distribution as a decision variable in the bi-objective optimisation for performance and energy on homogeneous multicore clusters. We show in this work that bi-objective optimisation for performance and energy on heterogeneous processors results in a large number of Pareto-optimal optimal solutions (workload distributions) even in the simple case of linear performance and energy profiles. We then study performance and energy profiles of real-life data-parallel applications and find that their shapes are non-linear, complex and non-smooth. We, therefore, propose an efficient and exact global optimisation algorithm, which takes as an input most general discrete performance and dynamic energy profiles of the heterogeneous processors and solves the bi-objective optimisation problem. The algorithm is also used as a building block to solve the bi-objective optimisation problem for performance and total energy. We also propose a novel methodology to build discrete dynamic energy profiles of individual computing devices, which are input to the algorithm. The methodology is based purely on system-level measurements and addresses the fundamental challenge of accurate component-level energy modelling of a hybrid data-parallel application running on a heterogeneous platform integrating CPUs and accelerators. We experimentally validate the proposed method using two data-parallel applications, matrix multiplication and 2D fast Fourier transform (2D-FFT).
Important workloads, such as machine learning and graph analytics applications, heavily involve sparse linear algebra operations. These operations use sparse matrix compression as an effective means to avoid storing zeros and performing unnecessary computation on zero elements. However, compression techniques like Compressed Sparse Row (CSR) that are widely used today introduce significant instruction overhead and expensive pointer-chasing operations to discover the positions of the non-zero elements. In this paper, we identify the discovery of the positions (i.e., indexing) of non-zero elements as a key bottleneck in sparse matrix-based workloads, which greatly reduces the benefits of compression. We propose SMASH, a hardware-software cooperative mechanism that enables highly-efficient indexing and storage of sparse matrices. The key idea of SMASH is to explicitly enable the hardware to recognize and exploit sparsity in data. To this end, we devise a novel software encoding based on a hierarchy of bitmaps. This encoding can be used to efficiently compress any sparse matrix, regardless of the extent and structure of sparsity. At the same time, the bitmap encoding can be directly interpreted by the hardware. We design a lightweight hardware unit, the Bitmap Management Unit (BMU), that buffers and scans the bitmap hierarchy to perform highly-efficient indexing of sparse matrices. SMASH exposes an expressive and rich ISA to communicate with the BMU, which enables its use in accelerating any sparse matrix computation. We demonstrate the benefits of SMASH on four use cases that include sparse matrix kernels and graph analytics applications.
We introduce a new graph problem, the token dropping game, and we show how to solve it efficiently in a distributed setting. We use the token dropping game as a tool to design an efficient distributed algorithm for stable orientations and more generally for locally optimal semi-matchings. The prior work by Czygrinow et al. (DISC 2012) finds a stable orientation in $O(Delta^5)$ rounds in graphs of maximum degree $Delta$, while we improve it to $O(Delta^4)$ and also prove a lower bound of $Omega(Delta)$.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
