Eliminating conflict misses for high performance architectures
- 13 July 1998
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 353-360
- https://doi.org/10.1145/277830.277917
Abstract
Many cache misses in scientific programs are due to conflictscaused by limited set associativity. Two data-layout transformations,inter- and intra-variable padding, can eliminatemany conflict misses at compile time. We present GroupPad,an inter-variable padding heuristic to preserve groupreuse in stencil computations frequently found in scientificcomputations. We show padding can also improve performancein parallel programs. Our optimizations have beenimplemented and tested on a...Keywords
This publication has 21 references indexed in Scilit:
- Fusion of loops for parallelism and localityIEEE Transactions on Parallel and Distributed Systems, 1997
- Non-singular data transformationsPublished by Association for Computing Machinery (ACM) ,1997
- A compiler algorithm for optimizing locality in loop nestsPublished by Association for Computing Machinery (ACM) ,1997
- Eliminating cache conflict misses through XOR-based placement functionsPublished by Association for Computing Machinery (ACM) ,1997
- Improving data locality with loop transformationsACM Transactions on Programming Languages and Systems, 1996
- Reducing false sharing on shared memory multiprocessors through compile time data transformationsPublished by Association for Computing Machinery (ACM) ,1995
- Data and computation transformations for multiprocessorsPublished by Association for Computing Machinery (ACM) ,1995
- SUIFACM SIGPLAN Notices, 1994
- The cache performance and optimizations of blocked algorithmsPublished by Association for Computing Machinery (ACM) ,1991
- Strategies for cache and local memory management by global program transformationJournal of Parallel and Distributed Computing, 1988