Software Automatic Tuning : From Concepts to State-of-the-Art Results

Software Automatic Tuning: From Concepts to State-of-the-Art Results Ken Naono Keita Teranishi John Cavazos Reiji Suda It is well known that carefully tuned programs run much faster than ones consisting of simply written code, and sometimes the difference of speed is more 100X. To make things more complex, well-tuned code for some machines performs badly on others. "Automatic Performance Tuning" is a technology paradigm that enables software to tune itself to its environments so that it performs well on any computer, even on computers unknown to the programmer. This book summarizes the research efforts to date and state of the art of automatic performance tuning. Software developers and researchers in the area of scientific and technical computing, optimized compilers, high performance systems software, and low-power computing will find this book to be an invaluable reference to this powerful new paradigm. •Presents the first English collaboration on the powerful, new software paradigm of Automatic Performance Tuning; •Offers a comprehensive survey of fundamental concepts and state-of-the-art results from the field; •Enables programmers to create software that will tune itself to its environments so that it performs well on any computer. Front Matter....Pages i-xiv Front Matter....Pages 1-1 Software Automatic Tuning: Concepts and State-of-the-Art Results....Pages 3-15 Front Matter....Pages 17-17 ATLAS Version 3.9: Overview and Status....Pages 19-32 Autotuning Method for Deciding Block Size Parameters in Dynamically Load-Balanced BLAS....Pages 33-48 Automatic Tuning for Parallel FFTs....Pages 49-67 Dynamic Programming Approaches to Optimizing the Blocking Strategy for Basic Matrix Decompositions....Pages 69-85 Automatic Tuning of the Division Number in the Multiple Division Divide-and-Conquer for Real Symmetric Eigenproblem....Pages 87-101 Automatically Tuned Mixed-Precision Conjugate Gradient Solver....Pages 103-119 Automatically Tuned Sparse Eigensolvers....Pages 121-133 Systematic Performance Evaluation of Linear Solvers Using Quality Control Techniques....Pages 135-152 Application of Alternating Decision Trees in Selecting Sparse Linear Solvers....Pages 153-173 Toward Automatic Performance Tuning for Numerical Simulations in the SILC Matrix Computation Framework....Pages 175-192 Exploring Tuning Strategies for Quantum Chemistry Computations....Pages 193-208 Automatic Tuning of CUDA Execution Parameters for Stencil Processing....Pages 209-228 Static Task Cluster Size Determination in Homogeneous Distributed Systems....Pages 229-252 Front Matter....Pages 253-253 Algorithmic Parameter Optimization of the DFO Method with the OPAL Framework....Pages 255-274 A Bayesian Method of Online Automatic Tuning....Pages 275-293 ABCLibScript: A Computer Language for Automatic Performance Tuning....Pages 295-313 Automatically Tuning Task-Based Programs for Multicore Processors....Pages 315-334 Efficient Program Compilation Through Machine Learning Techniques....Pages 335-351 Autotuning and Specialization: Speeding up Matrix Multiply for Small Matrices with Compiler Technology....Pages 353-370 Back Matter....Pages 371-377 Automatic Performance Tuning is a new software paradigm which enables software to be high performance in any computing environment. Its methodologies have been developed over the past decade, and it is now rapidly growing in terms of its scope and applicability, as well as in its scientific knowledge and technological methods. Software developers and researchers in the area of scientific and technical computing, high performance database systems, optimized compilers, high performance systems software, and low-power computing will find this book to be an invaluable reference to this powerful new paradigm.