Roofline
Communications of the ACMPublished 1 April 2009Open access
Samuel Williams, Andrew Waterman, David A. Patterson
Citations2,356
Generate an AI Snapshot to get a quick, structured summary of this paper.
Study Snapshot
ObjectiveStudy objective
MethodsResearch methodology
PopulationPopulation studied
Sample sizeSample sizes
OutcomesStudy outcomes here
ResultsStudy results comes here
LimitationsResearch study limitations comes here
A concise AI-generated summary of the paper will appear here once you click Generate AI Snapshot.
Abstract
The Roofline model offers insight on how to improve the performance of software and hardware.
Keywords
Computer Science
Computer Architecture: A Quantitative Approach
9,536 Citations1989John L. Hennessy, David A. Patterson
This best-selling title, considered for over a decade to be essential reading for every serious student and practitioner of computer design, has been updated throughout to address the most important trends facing computer designers today.
Proceedings of the IEEEThe Design and Implementation of FFTW3
5,156 Citations2005Matteo Frigo, Steven G. Johnson
It is shown that such an approach can yield an implementation of the discrete Fourier transform that is competitive with hand-optimized libraries, and the software structure that makes the current FFTW3 version flexible and adaptive is described.
Validity of the single processor approach to achieving large scale computing capabilities
4,200 Citations1967Gene M. Amdahl
The PARSEC benchmark suite
3,400 Citations2008Christian Bienia, Sanjeev Kumar +2 more
This paper presents and characterizes the Princeton Application Repository for Shared-Memory Computers (PARSEC), a benchmark suite for studies of Chip-Multiprocessors (CMPs), and shows that the benchmark suite covers a wide spectrum of working sets, locality, data sharing, synchronization and off-chip traffic.
Operations ResearchA Proof for the Queuing Formula: <i>L</i> = λ<i>W</i>
2,738 Citations1961John D. C. Little
The Landscape of Parallel Computing Research: A View from Berkeley
1,993 Citations2006Krste Asanović, Ras Bodik +9 more
The parallel landscape is frame with seven questions, and the following are recommended to explore the design space rapidly: • The overarching goal should be to make it easy to write programs that execute efficiently on highly parallel computing systems • The target should be 1000s of cores per chip, as these chips are built from processing elements that are the most efficient in MIPS (Million Instructions per Second) per watt, MIPS per area of silicon, and MIPS each development dollar.
ComputerAmdahl's Law in the Multicore Era
1,229 Citations2008Mark D. Hill, Michael R. Marty
Augmenting Amdahl's law with a corollary for multicore hardware makes it relevant to future generations of chips with multiple processor cores.
IEEE Transactions on ComputersEvaluating associativity in CPU caches
600 Citations1989Mark D. Hill, Alan Jay Smith
All-associativity simulation is theoretically less efficient than forest simulation or stack simulation (a commonly used simulation algorithm), in practice it is not much slower and allows the simulation of many more caches with a single pass through an address trace.
IEEE International Conference on High Performance Computing, Data, and AnalyticsStencil computation optimization and auto-tuning on state-of-the-art multicore architectures
464 Citations2008Kaushik Datta, Mark Murphy +7 more
Journal of Hebei Institute of TechnologyA proof of the queueing formula: L=λW
354 Citations2001Shang Zhi
It is about the arrival rate multiplied by average waiting time equals average queue length, and the result is that the number of people queued up at the entrance to the terminal increases with the length of the queue.
Communications of the ACMLatency lags bandwith
239 Citations2004David A. Patterson
As I review performance trends, I am struck by a consistent theme across many technologies: bandwidth improves much more quickly than latency.
Proceedings of the IEEESelf-Adapting Linear Algebra Algorithms and Software
205 Citations2005James Demmel, Jack Dongarra +6 more
The generation of dense and sparse Basic Linear Algebra Subprograms (BLAS) kernels and the selection of linear solver algorithms are described.
Mapping computational concepts to GPUs
190 Citations2005Mark Harris
This chapter presents intuitive mappings of standard computational concepts onto the special-purpose features of GPUs and introduces a simple GPU programming framework and demonstrates the use of the framework in a short sample program.
ACM Transactions on Programming Languages and SystemsImproving the ratio of memory operations to floating-point operations in loops
154 Citations1994Steve Carr, Ken Kennedy
This paper develops and evaluates techniques that automatically restructure program loops to achieve high performance on specific target architectures and attempts to balance computation and memory accesses and seek to eliminate or reduce pipeline interlock.
Performance Optimizations and Bounds for Sparse Matrix-Vector Multiply
115 Citations2002Richard Vuduc, James Demmel +4 more
IEEE Transactions on ComputersAnalytic Queueing Network Models for Parallel Processing of Task Systems
104 Citations1986Thomasian, Bay
An efficient algorithm to determine the mean completion time and related performance measures for a task system: a set of tasks with precedence relationships in their execution sequence, such that the resulting graph is acyclic.
Auto-tuning performance on multicore computers
76 Citations2008David A. Patterson, Samuel Williams
It is shown that auto-tuning consistently delivers speedups in excess of 3× across all multicore computers except the memory-bound Intel Clovertown, where the benefit was as little as 1.5×.
A genetic algorithms approach to modeling the performance of memory-bound computations
66 Citations2007Mustafa M. Tikir, Laura Carrington +2 more
A Genetic Algorithm approach is used to "learn" bandwidth as a function of cache hit rates per machine with MultiMAPS as the fitness test and 56 individual performance predictions are made.
IEEE Transactions on Software EngineeringPerformance of Synchronized Iterative Processes in Multiprocessor Systems
58 Citations1982Michel Dubois, Fayé A. Briggs
A general methodology for studying the degree of matching between an architecture and an algorithm is introduced and applied to the case of synchronized iterative algorithms in MIMD machines.
Minds at UW (University of Wisconsin)Analyzing the behavior and performance of parallel programs
43 Citations1993Vikram Adve
A conceptually simple deterministic model for parallel program performance prediction, using deterministic values to represent mean task times including communication, and (if necessary) shared-resource contention computed from a separate, stochastic model is developed.
A Hierarchical Approach to Modeling and Improving the Performance of Scientific Applications on the KSR1
23 Citations1994Eric Boyd, W. Azeem +4 more
This work delineates a comprehensive approach to modeling and improving application performance on the KSR1, and proposes a workload characterization, and derive upper bounds on the performance of specific machine-workload pairs.