Paper Info
Title
Efficient hierarchical online-autotuning: a case study on polyhedral accelerator mapping
Abstract
Identifying the (near) optimal program variants an optimizing and parallelizing compiler should generate is known to be difficult. Autotuning is the best solution to navigate the often high-dimensional space of possible options. However, to be practical an autotuner should (a) have high convergence speed and (b) be robust in face of varying inputs. Current techniques for offline tuning, where convergence speed is less important, provide solutions only for known inputs, whereas online tuning can be input sensitive but currently lacks in convergence speed. In this paper, we present hierarchical online-autotuning, a novel technique to exploit structure in the search space and the underlying tuning problem to increase convergence speed during online tuning. By modeling symmetries and redundancies in configurations and by exploiting domain knowledge to predict performance we reduce the search space size by orders of magnitudes. Combining our tuner with a polyhedral parallelizing compiler for GPUs, we show that the performance of a GEMM GPU kernel generated with default parameters is increased by 6× and that the convergence speed of the tuning process is increased by a factor of up to 1.7 compared to OpenTuner. With hierarchical tuning we make the deployment of always-on online-autotuning practical.
Year
DOI
Venue
2019
10.1145/3330345.3330377
Proceedings of the ACM International Conference on Supercomputing
Keywords
Field
DocType
GPGPU, online-autotuning, performance optimization, polyhedral compilation
Kernel (linear algebra),Convergence (routing),Software deployment,Domain knowledge,Computer science,Parallel computing,Compiler,Exploit,General-purpose computing on graphics processing units,Tuner
Conference
ISBN
Citations 
PageRank 
978-1-4503-6079-1
0
0.34
References 
Authors
0
3
Name
Order
Citations
PageRank
Philip Pfaffe111.03
Tobias Grosser227116.04
Martin Tillmann361.45