论文信息 - A Square Block Format for Symmetric Band Matrices

A Square Block Format for Symmetric Band Matrices

This contribution describes a Square Block, SB, format for storing a banded symmetric matrix. This is possible by rearranging “in place” LAPACK Band Layout to become a SB layout: store submatrices as a set of square blocks. The new format reduces storage space, provides higher locality of memory accesses, results in regular access patterns, and exposes parallelism.

[1] Robert A. van de Geijn,et al. An Algorithm-by-Blocks for SuperMatrix Band Cholesky Factorization , 2008, VECPAR.

[2] Jack Dongarra,et al. LAPACK Users' Guide, 3rd ed. , 1999 .

[3] Jack J. Dongarra,et al. Solving Systems of Linear Equations on the CELL Processor Using Cholesky Factorization , 2008, IEEE Transactions on Parallel and Distributed Systems.

[4] José R. Herrero. New Data Structures for Matrices and Specialized Inner Kernels: Low Overhead for High Performance , 2007, PPAM.

[5] Lars Karlsson,et al. Parallel and Cache-Efficient In-Place Matrix Storage Format Conversion , 2012, TOMS.

[6] José R. Herrero,et al. Exposing Inner Kernels and Block Storage for Fast Parallel Dense Linear Algebra Codes , 2008 .

[7] Julien Langou,et al. Level-3 Cholesky Factorization Routines Improve Performance of Many Cholesky Algorithms , 2013, TOMS.

[8] Fred G. Gustavson,et al. Algorithms for In-Place Matrix Transposition , 2013, PPAM.

[9] F. Gustavson,et al. Clearer, Simpler and more Efficient LAPACK Routines for Symmetric Positive Definite Band Factorization , 2008 .

[10] Juan J. Navarro,et al. Compiler-Optimized Kernels: An Efficient Alternative to Hand-Coded Inner Kernels , 2006, ICCSA.