Toward adjoinable MPI
Automatic differentiation is the primary means of obtaining analytic derivatives from a numerical model given as a computer program. Therefore, it is an essential productivity tool in numerous computational science and engineering domains. Computing gradients with the adjoint (also called reverse) m...
| Main Authors: | , , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | en_US |
| Published: |
Institute of Electrical and Electronics Engineers
2010
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/1721.1/58833 |
| _version_ | 1826217371827699712 |
|---|---|
| author | Hill, Christopher N. Utke, Jean Hascoet, Laurent Heimbach, Patrick Hovland, Paul Naumann, Uwe |
| author2 | Massachusetts Institute of Technology. Department of Earth, Atmospheric, and Planetary Sciences |
| author_facet | Massachusetts Institute of Technology. Department of Earth, Atmospheric, and Planetary Sciences Hill, Christopher N. Utke, Jean Hascoet, Laurent Heimbach, Patrick Hovland, Paul Naumann, Uwe |
| author_sort | Hill, Christopher N. |
| collection | MIT |
| description | Automatic differentiation is the primary means of obtaining analytic derivatives from a numerical model given as a computer program. Therefore, it is an essential productivity tool in numerous computational science and engineering domains. Computing gradients with the adjoint (also called reverse) mode via source transformation is a particularly beneficial but also challenging use of automatic differentiation. To date only ad hoc solutions for adjoint differentiation of MPI programs have been available, forcing automatic differentiation tool users to reason about parallel communication dataflow and dependencies and manually develop adjoint communication code. Using the communication graph as a model we characterize the principal problems of adjoining the most frequently used communication idioms. We propose solutions to cover these idioms and consider the consequences for the MPI implementation, the MPI user and MPI-aware program analysis. The MIT general circulation model serves as a use case to illustrate the viability of our approach. |
| first_indexed | 2024-09-23T17:02:37Z |
| format | Article |
| id | mit-1721.1/58833 |
| institution | Massachusetts Institute of Technology |
| language | en_US |
| last_indexed | 2024-09-23T17:02:37Z |
| publishDate | 2010 |
| publisher | Institute of Electrical and Electronics Engineers |
| record_format | dspace |
| spelling | mit-1721.1/588332022-09-29T23:15:26Z Toward adjoinable MPI Hill, Christopher N. Utke, Jean Hascoet, Laurent Heimbach, Patrick Hovland, Paul Naumann, Uwe Massachusetts Institute of Technology. Department of Earth, Atmospheric, and Planetary Sciences Hill, Christopher N. Hill, Christopher N. MPI automatic differentiation reverse mode source transformation Automatic differentiation is the primary means of obtaining analytic derivatives from a numerical model given as a computer program. Therefore, it is an essential productivity tool in numerous computational science and engineering domains. Computing gradients with the adjoint (also called reverse) mode via source transformation is a particularly beneficial but also challenging use of automatic differentiation. To date only ad hoc solutions for adjoint differentiation of MPI programs have been available, forcing automatic differentiation tool users to reason about parallel communication dataflow and dependencies and manually develop adjoint communication code. Using the communication graph as a model we characterize the principal problems of adjoining the most frequently used communication idioms. We propose solutions to cover these idioms and consider the consequences for the MPI implementation, the MPI user and MPI-aware program analysis. The MIT general circulation model serves as a use case to illustrate the viability of our approach. United States. Dept. of Energy (Contract DEAC02- 06CH11357) United States. National Aeronautics and Space Administration. Modeling Analysis and Prediction Program 2010-10-01T18:30:13Z 2010-10-01T18:30:13Z 2009-07 2009-05 Article http://purl.org/eprint/type/JournalArticle 978-1-4244-3751-1 1530-2075 INSPEC Accession Number: 10761833 http://hdl.handle.net/1721.1/58833 Utke, J. et al. “Toward adjoinable MPI.” Parallel & Distributed Processing, 2009. IPDPS 2009. IEEE International Symposium on. 2009. 1-8. ©2009 Institute of Electrical and Electronics Engineers. en_US http://dx.doi.org/10.1109/IPDPS.2009.5161165 IEEE International Symposium on Parallel & Distributed Processing, 2009. IPDPS 2009 Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Institute of Electrical and Electronics Engineers IEEE |
| spellingShingle | MPI automatic differentiation reverse mode source transformation Hill, Christopher N. Utke, Jean Hascoet, Laurent Heimbach, Patrick Hovland, Paul Naumann, Uwe Toward adjoinable MPI |
| title | Toward adjoinable MPI |
| title_full | Toward adjoinable MPI |
| title_fullStr | Toward adjoinable MPI |
| title_full_unstemmed | Toward adjoinable MPI |
| title_short | Toward adjoinable MPI |
| title_sort | toward adjoinable mpi |
| topic | MPI automatic differentiation reverse mode source transformation |
| url | http://hdl.handle.net/1721.1/58833 |
| work_keys_str_mv | AT hillchristophern towardadjoinablempi AT utkejean towardadjoinablempi AT hascoetlaurent towardadjoinablempi AT heimbachpatrick towardadjoinablempi AT hovlandpaul towardadjoinablempi AT naumannuwe towardadjoinablempi |