Abstract
We present a technique for introducing and op-timizing the use of memory in a functional array language, aimed at GPU execution, that supports correct-by-construction parallelism. Using linear memory access descriptors as building blocks, we define a notion of memory in the compiler IR that enables cost-free change-of-layout transformations (e.g., slicing, transposition), whose results can even be carried across control flow such as ifs/loops without manifestation in memory. The memory notion allows a graceful transition to an unsafe IR that is automatically optimized (1) to mix reads and writes to the same array inside a parallel construct, and (2) to map semantically different arrays to the same memory block. The result is code similar to what imperative users would write. Our evaluation shows that our optimizations have significant impact (1.1 x -2 x) and result in performance competitive to hand-written code from challenging benchmarks, such as Rodinia's NW, LUD, Hotspot.
Originalsprog | Engelsk |
---|---|
Titel | Proceedings of SC 2022 : International Conference for High Performance Computing, Networking, Storage and Analysis |
Forlag | IEEE Computer Society Press |
Publikationsdato | 2022 |
Sider | 1-15 |
Artikelnummer | 31 |
ISBN (Elektronisk) | 9781665454445 |
DOI | |
Status | Udgivet - 2022 |
Begivenhed | 2022 International Conference for High Performance Computing, Networking, Storage and Analysis, SC 2022 - Dallas, USA Varighed: 13 nov. 2022 → 18 nov. 2022 |
Konference
Konference | 2022 International Conference for High Performance Computing, Networking, Storage and Analysis, SC 2022 |
---|---|
Land/Område | USA |
By | Dallas |
Periode | 13/11/2022 → 18/11/2022 |
Sponsor | ACM's Special Interest Group on High Performance Computing (SIGHPC), Association for Computing Machinery, IEEE Computer Society, IEEE's Technical Committee on High Performance Computing (TCHPC) |
Bibliografisk note
Funding Information:ACKNOWLEDGMENTS We are grateful to Niels G. W. Serup for his initial prototyping work related to memory optimizations. This work has been supported by the Independent Research Fund Denmark (DFF) under the grants FUTHARK: Functional Technology for High-performance Architectures and Monitoring Changes in Big Satellite Data via Massively Parallel AI, and by the UCPH Data-Plus grant: High-Performance Land Change Assessment.
Publisher Copyright:
© 2022 IEEE.