Sparsity: Optimization Framework for Sparse Matrix Kernels

By Eun-Jin Im, K. Yelick, R. Vuduc. The international journal of high performance computing applications, 2004

378 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Networks and Communications, Hardware and Architecture

View Paper: https://www.semanticscholar.org/paper/66ba6dd1b746aa0e10c3a4f62a5c4dd6b955b503

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. J. Demmel
2. R. Vuduc
3. Aydin Buluç
4. K. Yelick
5. Nathan Bell
6. Leonid Oliker
7. Shoaib Kamil
8. Samuel Williams

In Computer Networks and Communications

1. J. Demmel
2. Aydin Buluç
3. R. Vuduc
4. K. Yelick
5. Leonid Oliker
6. Michael Garland
7. Nathan Bell
8. Shoaib Kamil

In Hardware and Architecture

1. J. Demmel
2. Aydin Buluç
3. R. Vuduc
4. K. Yelick
5. Leonid Oliker
6. Michael Garland
7. Shoaib Kamil
8. Nathan Bell

Citations in the corpus

• Optimization of sparse matrix-vector multiplication on emerging multicore platforms (2007, methodology): “These patterns include blocks [10], variable or mixtures of differently-sized blocks [6] diagonals, which may be especially well-suited to machines with SIMD and vector units [19], dense subtriangles arising in sparse triangular solve [22], and symmetry [11], and combinations.”
• SPIRAL: Code Generation for DSP Transforms (2005, methodology): “The specific problem of sparse matrix vector multiplications is addressed in SPARSITY [12], [15], again by applying empirical search to determine the best blocking strategy for a given sparse matrix.”
• OSKI: A library of automatically tuned sparse matrix kernels (2005, methodology): “OSKI reflects the need for and cost of run-time tuning, as extensively documented in our recent work on automatic tuning of sparse kernels using the Sparsity framework [4, 5, 1, 6, 7, 8, 9, 10, 11].”
• Parallel sparse matrix-vector and matrix-transpose-vector multiplication using compressed sparse blocks (2009, methodology): “A blocked variant of CSR, calledBCSR, has been used for improving register reuse [24].”
• Implementing sparse matrix-vector multiplication on throughput-oriented processors (2009, methodology): “Block formats such as Block CSR or Variable-Block CSR [12] can deliver higher performance on both CPU [23] and GPU [8] architectures, particularly for matrices arising in vector-valued problems.”
• The potential of the cell processor for scientific computing (2005, methodology): “2 Evaluation Matrices In order to effectively evaluate SpMV performance, we examine six synthetic matrices, as well as ten real matrices used in numerical calculations from the Bebop SPARSITY suite [3,5] (four unsymmetric and six symmetric).”
• Automatic performance tuning of sparse matrix kernels (2003, methodology): “We build on the Sparsity system for generating highly-tuned implementations of the SpMV kernel y ← y+Ax, where A is a sparse matrix and x, y are dense vectors.”
• Sparse Convolutional Neural Networks (2015, background): “Blocking is adopted in register [11] and cache [18] level to improve the spatial locality of sparse matrix.”
• Model-driven autotuning of sparse matrix-vector multiply on GPUs (2010, methodology): “Like related work, we use offline benchmarking [11, 19], but in contrast, we use this data to determine various parameters in the model (e.”