Manage OpenMP GPU Data Environment Under Unified Address Space
Abstract
OpenMP has supported the offload of computations to accelerators such as GPUs since version 4.0. A crucial aspect in OpenMP offloading is to manage the accelerator data environment. Currently, this has to be explicitly programmed by users, which is non-trival and often results in suboptimal performance. The unified memory feature available in recent GPU architectures introduces another option, implicit management. However, our experiments show that it incurs several performance issues, especially under GPU memory oversubscription. In this paper, we propose a compiler and runtime collaborative approach to manage OpenMP GPU data under unified memory. In our framework, the compiler performs data reuse analysis to assist runtime data management. The runtime combines static and dynamic information to make optimized data management decisions.We have implement the proposed technology in the LLVM framework. The evaluation shows our method can achieve significant performance improvement for OpenMP GPU offloading.
- Authors:
- Publication Date:
- Research Org.:
- Oak Ridge National Lab. (ORNL), Oak Ridge, TN (United States). Oak Ridge Leadership Computing Facility (OLCF); Brookhaven National Lab. (BNL), Upton, NY (United States)
- Sponsoring Org.:
- USDOE Office of Science (SC), Advanced Scientific Computing Research (SC-21)
- OSTI Identifier:
- 1484438
- Report Number(s):
- BNL-209639-2018-COPA
Journal ID: ISSN 0302-9743
- DOE Contract Number:
- SC0012704
- Resource Type:
- Conference
- Resource Relation:
- Journal Volume: 11128; Conference: International Workshop on OpenMP 2018, Barcelona, Spain, 9/26/2018 - 9/28/2018
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 97 MATHEMATICS AND COMPUTING; Data management; Unified memory; OpenMP offloading; Compiler; Runtime; LLVM
Citation Formats
Li, Lingda. Manage OpenMP GPU Data Environment Under Unified Address Space. United States: N. p., 2018.
Web. doi:10.1007/978-3-319-98521-3_5.
Li, Lingda. Manage OpenMP GPU Data Environment Under Unified Address Space. United States. doi:10.1007/978-3-319-98521-3_5.
Li, Lingda. Wed .
"Manage OpenMP GPU Data Environment Under Unified Address Space". United States. doi:10.1007/978-3-319-98521-3_5. https://www.osti.gov/servlets/purl/1484438.
@article{osti_1484438,
title = {Manage OpenMP GPU Data Environment Under Unified Address Space},
author = {Li, Lingda},
abstractNote = {OpenMP has supported the offload of computations to accelerators such as GPUs since version 4.0. A crucial aspect in OpenMP offloading is to manage the accelerator data environment. Currently, this has to be explicitly programmed by users, which is non-trival and often results in suboptimal performance. The unified memory feature available in recent GPU architectures introduces another option, implicit management. However, our experiments show that it incurs several performance issues, especially under GPU memory oversubscription. In this paper, we propose a compiler and runtime collaborative approach to manage OpenMP GPU data under unified memory. In our framework, the compiler performs data reuse analysis to assist runtime data management. The runtime combines static and dynamic information to make optimized data management decisions.We have implement the proposed technology in the LLVM framework. The evaluation shows our method can achieve significant performance improvement for OpenMP GPU offloading.},
doi = {10.1007/978-3-319-98521-3_5},
journal = {},
issn = {0302-9743},
number = ,
volume = 11128,
place = {United States},
year = {2018},
month = {9}
}
Works referenced in this record:
OpenCL: A Parallel Programming Standard for Heterogeneous Computing Systems
journal, May 2010
- Stone, John E.; Gohara, David; Shi, Guochun
- Computing in Science & Engineering, Vol. 12, Issue 3, p. 66-73