Combined Iterative and Model-driven Optimization in an Automatic Parallelization Framework

Pouchet, Louis-Noël; Bondhugula, Uday; Bastoul, Cédric; Cohen, Albert; Ramanujam, J.; Sadayappan, P.

doi:10.1109/sc.2010.14

“…However, the space of possible optimizations is dramatically enlarged, imposing challenges on the selection algorithms. In contrast to previous iterative approaches requiring the evaluation of up to hundreds of possible transformations [30], [32], we develop here a scheme that requires at most 5 candidate choices to be evaluated.…”

Section: Optimization Spacementioning

confidence: 99%

“…The best loop optimization sequence is often not only program-specific, but also depends on the target hardware. Pouchet et al illustrated this by showing the critical impact of tuning polyhedral optimizations for obtaining the best performance for a variety of numerical programs on different target processors [31], [30], [32].…”

Section: Introductionmentioning

confidence: 99%

“…The polyhedral framework enables the modeling of an arbitrarily complex sequence of loop transformations in a single optimization step. The downside of this expressiveness is the extreme difficulty of selecting an effective set of affine transformation coefficients that result in good performance, combining tiling, coarse and fine grain parallelization, together with fusion, distribution, interchange, skewing, permutation and shifting [18], [30], [32].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Predictive Modeling in a Polyhedral Optimization Space

Park

¹

,

Cavazos²,

Pouchet³

et al. 2013

Int J Parallel Prog

Self Cite

View full text Add to dashboard Cite

Abstract-Significant advances in compiler optimization have been made in recent years, enabling many transformations such as tiling, fusion, parallelization and vectorization on imperfectly nested loops. Nevertheless, the problem of finding the best combination of loop transformations remains a major challenge. Polyhedral models for compiler optimization have demonstrated strong potential for enhancing program performance, in particular for compute-intensive applications. But existing static cost models to optimize polyhedral transformations have significant limitations, and iterative compilation has become a very promising alternative to these models to find the most effective transformations. But since the number of polyhedral optimization alternatives can be enormous, it is often impractical to iterate over a significant fraction of the entire space of polyhedrally transformed variants. Recent research has focused on iterating over this search space either with manually-constructed heuristics or with automatic but very expensive search algorithms (e.g., genetic algorithms) that can eventually find good points in the polyhedral space.In this paper, we propose the use of machine learning to address the problem of selecting the best polyhedral optimizations. We show that these models can quickly find high-performance program variants in the polyhedral space, without resorting to extensive empirical search. We introduce models that take as input a characterization of a program based on its dynamic behavior, and predict the performance of aggressive high-level polyhedral transformations that includes tiling, parallelization and vectorization. We allow for a minimal empirical search on the target machine, discovering on average 83% of the searchspace-optimal combinations in at most 5 runs. Our end-to-end framework is validated using numerous benchmarks on two multi-core platforms.

show abstract

“…Since MLP is an indicator of intra-warp memory-level parallelism, we need to consider the overlap factor of multiple warps. ITMLP can be calculated using Equations (14) and (15).…”

Section: Calculating the Memory Access Cost Tmemmentioning

confidence: 99%

A performance analysis framework for identifying potential benefits in GPGPU applications

Sim

¹

,

Dasgupta

²

,

Kim

³

et al. 2012

Proceedings of the 17th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming

View full text Add to dashboard Cite

Tuning code for GPGPU and other emerging many-core platforms is a challenge because few models or tools can precisely pinpoint the root cause of performance bottlenecks. In this paper, we present a performance analysis framework that can help shed light on such bottlenecks for GPGPU applications. Although a handful of GPGPU profiling tools exist, most of the traditional tools, unfortunately, simply provide programmers with a variety of measurements and metrics obtained by running applications, and it is often difficult to map these metrics to understand the root causes of slowdowns, much less decide what next optimization step to take to alleviate the bottleneck. In our approach, we first develop an analytical performance model that can precisely predict performance and aims to provide programmer-interpretable metrics. Then, we apply static and dynamic profiling to instantiate our performance model for a particular input code and show how the model can predict the potential performance benefits. We demonstrate our framework on a suite of micro-benchmarks as well as a variety of computations extracted from real codes.

show abstract

“…Our work is also related to a rich body of work on optimizations and tuning of GPGPU applications [4,7,12,15,16,19]. Ryoo et al [16] introduced two metrics to prune optimization space by calculating the utilization and efficiency of GPU applications.…”

Section: Gpu Performance Modelingmentioning

confidence: 99%

A performance analysis framework for identifying potential benefits in GPGPU applications

Sim

¹

,

Dasgupta

²

,

Kim

³

et al. 2012

View full text Add to dashboard Cite

Tuning code for GPGPU and other emerging many-core platforms is a challenge because few models or tools can precisely pinpoint the root cause of performance bottlenecks. In this paper, we present a performance analysis framework that can help shed light on such bottlenecks for GPGPU applications. Although a handful of GPGPU profiling tools exist, most of the traditional tools, unfortunately, simply provide programmers with a variety of measurements and metrics obtained by running applications, and it is often difficult to map these metrics to understand the root causes of slowdowns, much less decide what next optimization step to take to alleviate the bottleneck. In our approach, we first develop an analytical performance model that can precisely predict performance and aims to provide programmer-interpretable metrics. Then, we apply static and dynamic profiling to instantiate our performance model for a particular input code and show how the model can predict the potential performance benefits. We demonstrate our framework on a suite of micro-benchmarks as well as a variety of computations extracted from real codes.

show abstract

Combined Iterative and Model-driven Optimization in an Automatic Parallelization Framework

Cited by 50 publications

References 31 publications

Predictive Modeling in a Polyhedral Optimization Space

Predictive Modeling in a Polyhedral Optimization Space

A performance analysis framework for identifying potential benefits in GPGPU applications

A performance analysis framework for identifying potential benefits in GPGPU applications

Contact Info

Product

Resources

About