Value Function Based Performance Optimization of Deep Learning Workloads
Benoit Steiner, Cummins, Chris, Horace He et al. · arXiv (Cornell University) · 2020
As machine learning techniques become ubiquitous, the efficiency of neural network implementations is becoming correspondingly paramount. Frameworks, such as Halide and TVM, separate out the algorithmic representation of the network from the schedule that determines its implementation. Finding good schedules, however, remains extremely challenging. We model this scheduling problem as a sequence of optimization choices, and present a new technique to accurately predict the expected performance of a partial schedule. By leveraging these predictions we can make these optimization decisions greedily and rapidly identify an efficient schedule. This enables us to find schedules that improve the throughput of deep neural networks by 2.6x over Halide and 1.5x over TVM. Moreover, our technique is two to three orders of magnitude faster than that of these tools, and completes in seconds instead of hours.
Cited by 2 works. doi:10.48550/arxiv.2011.14486 Open-access PDF
Authors
References
- Reinforcement Learning: An Introduction
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- TVM: An Automated End-to-End Optimizing Compiler for Deep Learning
- Decoupling algorithms from schedules for easy optimization of image processing pipelines
- Learning to optimize halide with tree search and random programs
- Learning to Optimize Tensor Programs
- FlexTensor
- Ansor: Generating High-Performance Tensor Programs for Deep Learning
- Loop transformations leveraging hardware prefetching