Learning Memory Access Patterns

Milad Hashemi, Kevin Swersky, Jamie Smith, Grant Ayers, Heiner Litz, Jichuan Chang, Christos Kozyrakis, Parthasarathy Ranganathan
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:1919-1928, 2018.

Abstract

The explosion in workload complexity and the recent slow-down in Moore’s law scaling call for new approaches towards efficient computing. Researchers are now beginning to use recent advances in machine learning in software optimizations; augmenting or replacing traditional heuristics and data structures. However, the space of machine learning for computer hardware architecture is only lightly explored. In this paper, we demonstrate the potential of deep learning to address the von Neumann bottleneck of memory performance. We focus on the critical problem of learning memory access patterns, with the goal of constructing accurate and efficient memory prefetchers. We relate contemporary prefetching strategies to n-gram models in natural language processing, and show how recurrent neural networks can serve as a drop-in replacement. On a suite of challenging benchmark datasets, we find that neural networks consistently demonstrate superior performance in terms of precision and recall. This work represents the first step towards practical neural-network based prefetching, and opens a wide range of exciting directions for machine learning in computer architecture research.

Cite this Paper


BibTeX
@InProceedings{pmlr-v80-hashemi18a, title = {Learning Memory Access Patterns}, author = {Hashemi, Milad and Swersky, Kevin and Smith, Jamie and Ayers, Grant and Litz, Heiner and Chang, Jichuan and Kozyrakis, Christos and Ranganathan, Parthasarathy}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {1919--1928}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/hashemi18a/hashemi18a.pdf}, url = {http://proceedings.mlr.press/v80/hashemi18a.html}, abstract = {The explosion in workload complexity and the recent slow-down in Moore’s law scaling call for new approaches towards efficient computing. Researchers are now beginning to use recent advances in machine learning in software optimizations; augmenting or replacing traditional heuristics and data structures. However, the space of machine learning for computer hardware architecture is only lightly explored. In this paper, we demonstrate the potential of deep learning to address the von Neumann bottleneck of memory performance. We focus on the critical problem of learning memory access patterns, with the goal of constructing accurate and efficient memory prefetchers. We relate contemporary prefetching strategies to n-gram models in natural language processing, and show how recurrent neural networks can serve as a drop-in replacement. On a suite of challenging benchmark datasets, we find that neural networks consistently demonstrate superior performance in terms of precision and recall. This work represents the first step towards practical neural-network based prefetching, and opens a wide range of exciting directions for machine learning in computer architecture research.} }
Endnote
%0 Conference Paper %T Learning Memory Access Patterns %A Milad Hashemi %A Kevin Swersky %A Jamie Smith %A Grant Ayers %A Heiner Litz %A Jichuan Chang %A Christos Kozyrakis %A Parthasarathy Ranganathan %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-hashemi18a %I PMLR %P 1919--1928 %U http://proceedings.mlr.press/v80/hashemi18a.html %V 80 %X The explosion in workload complexity and the recent slow-down in Moore’s law scaling call for new approaches towards efficient computing. Researchers are now beginning to use recent advances in machine learning in software optimizations; augmenting or replacing traditional heuristics and data structures. However, the space of machine learning for computer hardware architecture is only lightly explored. In this paper, we demonstrate the potential of deep learning to address the von Neumann bottleneck of memory performance. We focus on the critical problem of learning memory access patterns, with the goal of constructing accurate and efficient memory prefetchers. We relate contemporary prefetching strategies to n-gram models in natural language processing, and show how recurrent neural networks can serve as a drop-in replacement. On a suite of challenging benchmark datasets, we find that neural networks consistently demonstrate superior performance in terms of precision and recall. This work represents the first step towards practical neural-network based prefetching, and opens a wide range of exciting directions for machine learning in computer architecture research.
APA
Hashemi, M., Swersky, K., Smith, J., Ayers, G., Litz, H., Chang, J., Kozyrakis, C. & Ranganathan, P.. (2018). Learning Memory Access Patterns. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:1919-1928 Available from http://proceedings.mlr.press/v80/hashemi18a.html.

Related Material