Workshop on Automatic Tuning for Petascale Systems
Held July 9-12, 2007, Snowbird Ski & Summer Resort, Snowbird, Utah, USA
Kathy Yelick (University of California at Berkeley), "yelick" AT "cs.berkeley.edu"
Keith Cooper (Rice University), "keith" AT "cs.rice.edu"
Over the past several years, processors have grown in complexity due to the inclusion of multiple cores, processor accelerators, and various techniques to mask the ever-growing gap between on-chip computation performance and off-chip memory access performance. Each of these performance features requires a significant effort in tuning of applications and libraries. As a result, compiler and library developers have turned to automating the process of software tuning, using large amounts of computation time to explore a space of different variants of the program and running each variant on the target architecture.
The goal of this workshop was to bring together researchers and practitioners in automatic tuning, compiler code generation, and architecture design to identify some of the opportunities and challenges of using automatic tuning on future petascale systems. The emphasis was on optimizations for multicore processors, accelerators, and novel computational and memory structures that will make up the compute nodes of petascale systems.
Day 1 - Monday, July 9
- Workshop Overview
Kathy Yelick, U.C. Berkeley and LBNL
- Register Allocation in Kernel Generators
Matteo Frigo, Cilk Arts
- SPIRAL: Past, Present and Future
Jeremy Johnson, Drexel University
- Parallelism in Spiral
- GPU Computing with CUDA
- Cray Math Software: Current and Future Developments
- Tuning Sparse Matrix Vector Multiplication for multicore processors
Day 2 - Tuesday, July 10
- Paolo Bientenisi
- Nikos Pitsianis
- Targeting Multicore Systems in Linear Algebra Applications
- Clint Whaley
- Search-based Compilation: Lessons from a Decade of Experiments
- Compilers and Runtime Systems for Dynamically Adaptive Applications
- Compiler-Assisted Performance Tuning
- Jackie Chame
Day 3 - Wednesday, July 11
- OSKI: Autotuned Sparse Matrix Kernels
- Tuning 3D Stencil Codes
- Jim Demmel (Optimal, ScaLAPACK)
- Library experiences (short talks/discussion)
- Michael Frank
Day 4 - Thursday, July 12
- POET: Parameterized Optimizations for Empirical Tuning
- Guojing Cong
- Hardware Performance Monitoring with PAPI
- Yaoqing Gao
- Wrap-up Discussion
Community building Engaging applications
Informal discussions with specific subgroups
This workshop was sponsored by the Center for Scalable Application Development Software, with funding from the DOE Office of Science through the Scientific Discovery through Advanced Computing (SciDAC) program.