#2197 closed Task (wontfix)
HPC-09_Mireck_OpenACC
Reported by: | mathiot | Owned by: | andmirek |
---|---|---|---|
Priority: | high | Milestone: | 2019 WP |
Component: | TRA | Version: | trunk |
Severity: | minor | Keywords: | HPC, |
Cc: |
Description (last modified by andmirek)
BE CAREFUL !!! Due to dynamic behaviour of this ticket creation page, it is highly recommend to set first all other fields before writing the ticket description below.
If you have lost your draft after an unwanted reload, you can click on the link 'Restore Form' in the contextual menu upper right to recover it.
Remove these lines after reading.
Summary
Action | HPC-09_Mireck_OpenACC |
---|---|
PI(S) | Mirek |
Digest | Investigation on OpenACC |
Dependencies | |
Branch | NEMO/branches/$YEAR/dev_r{REV}_{ACTION_NAME} |
Previewer(s) | |
Reviewer(s) | |
Wiki | wiki:2019WP/HPC-09_Mireck_OpenACC |
Abstract
Investigation on Open ACC
Description
Investigate NEMO on GPU. This ticket is to document development of NEMO code to be ready to run on GPU. The aproach taken is based on automatic insertion of compiler directives into the pre-proccessed by compiler code with PSYclone. PSYclone works on a target computer so the commits into the branch are the changes in the code to efficiently run code on GPU after directives are inserted.
Implementation
How exactly the directives will be implemented is not decided yet. This is an exploratory work to see what performance improvement can be gained on GPU. This work involves software developer(s) from NVIDIA and STFC.
Reference manual and web pages updates
No changes in manual will be needed.
Once the PI has completed this section, he should send a mail to the previewer(s) asking them to preview the work within two weeks.
Commit History (20)
Changeset | Author | Time | ChangeLog |
---|---|---|---|
13090 | andmirek | 2020-06-10T16:44:43+02:00 | Ticket #2482: calculate sums in the same way as in #2197, commit 11069 |
11888 | andmirek | 2019-11-12T10:09:51+01:00 | Ticket #2197, small changes to improve code structure |
11467 | andmirek | 2019-08-22T11:49:08+02:00 | Ticket #2197 allocate arrays at the beggining of the run |
11069 | andmirek | 2019-06-03T12:23:43+02:00 | ticket #2197 glob sum for parallel architecture |
10853 | andmirek | 2019-04-09T11:04:31+02:00 | ticket #2197 bugfix and change in NaN detection |
10851 | andmirek | 2019-04-08T15:39:54+02:00 | ticket #2197 dissable diagnostics without key_iomput |
10849 | andmirek | 2019-04-08T13:35:31+02:00 | ticket #2197 for OCE ICE build without key_iomput |
10848 | andmirek | 2019-04-08T11:44:42+02:00 | ticket #2197 add build keys |
10847 | andmirek | 2019-04-05T16:51:46+02:00 | ticket #2197 merge dev_r10037_shlat2d: for use for ORCA1 |
10846 | andmirek | 2019-04-05T16:48:35+02:00 | ticket #2197 merge dev_r9950_old_tidal_mixing r10040: for use for ORCA25 |
10845 | andmirek | 2019-04-05T16:43:17+02:00 | |
10844 | andmirek | 2019-04-05T16:07:13+02:00 | |
10843 | andmirek | 2019-04-05T16:01:32+02:00 | ticket #2197 merge with dev_r9950_GO8_package at 10320 |
10842 | andmirek | 2019-04-05T15:53:48+02:00 | ticket #2197 branch for tests with OPENACC |
10838 | wayne_gaudin | 2019-04-04T08:36:27+02:00 | Ticket #2197 - extracted versions added |
10837 | wayne_gaudin | 2019-04-04T08:34:28+02:00 | Ticket #2197 - added version extraction to the Makefile |
10834 | wayne_gaudin | 2019-04-03T16:10:09+02:00 | Ticket #2197 - added namelist input to the Master directory |
10831 | wayne_gaudin | 2019-04-03T15:44:58+02:00 | Ticket #2197 - added the Master directory with code and makefile |
10828 | wayne_gaudin | 2019-04-03T08:44:12+02:00 | Updated comment ticket #2197 |
10827 | wayne_gaudin | 2019-04-03T08:33:53+02:00 | Initial comment, ticket #2197 |
Change History (27)
comment:1 Changed 5 years ago by nemo
- Priority changed from low to high
comment:2 Changed 5 years ago by nicolasmartin
- Owner set to andmirek
- Status changed from new to assigned
comment:3 Changed 5 years ago by wayne_gaudin
comment:4 Changed 5 years ago by wayne_gaudin
In 10828:
comment:5 Changed 5 years ago by wayne_gaudin
In 10831:
comment:6 Changed 5 years ago by wayne_gaudin
In 10834:
comment:7 Changed 5 years ago by wayne_gaudin
In 10837:
comment:8 Changed 5 years ago by wayne_gaudin
In 10838:
comment:9 Changed 5 years ago by andmirek
In 10842:
comment:10 Changed 5 years ago by andmirek
In 10843:
comment:11 Changed 5 years ago by andmirek
In 10844:
comment:12 Changed 5 years ago by andmirek
In 10845:
comment:13 Changed 5 years ago by andmirek
In 10846:
comment:14 Changed 5 years ago by andmirek
In 10847:
comment:15 Changed 5 years ago by andmirek
In 10848:
comment:16 Changed 5 years ago by andmirek
In 10849:
comment:17 Changed 5 years ago by andmirek
In 10851:
comment:18 Changed 5 years ago by andmirek
In 10853:
comment:19 Changed 5 years ago by andmirek
In 11069:
comment:20 Changed 5 years ago by andmirek
In 11467:
comment:21 Changed 5 years ago by andmirek
Last commit has changes to the way arrays are used in the code. Instead using allocatable and automatic arrays in the code, during initialization large number of arrays are created and later used in the code. This change is to investigate the impact of these change on performance of NEMO on GPU. With the changes model produces the same solution after 992 timesteps for GO8 configuration (ORCA1). The model seems to run slightly slower with new approach (1007.370s vs. 978.629s) and uses approximately the same amount of RAM 11.6G with changes, 11.8G reference.
comment:22 Changed 5 years ago by andmirek
- Description modified (diff)
- Milestone changed from 2019 WP to Unscheduled
comment:23 Changed 5 years ago by andmirek
- wp_comment set to This is an exploratory work at this moment. No plans to have it in the trunk. There is a version of the code (processed by PSYclone) running on GPU. The performance is under investigation and improvement in implementation of directives in the code continues.
comment:24 Changed 5 years ago by andmirek
In 11888:
comment:25 Changed 5 years ago by andmirek
To be continued in 2020
comment:26 Changed 4 years ago by mathiot
- Milestone changed from Unscheduled to 2019 WP
- Resolution set to wontfix
- Status changed from assigned to closed
- wp_comment changed from This is an exploratory work at this moment. No plans to have it in the trunk. There is a version of the code (processed by PSYclone) running on GPU. The performance is under investigation and improvement in implementation of directives in the code continues. to This is an exploratory work at this moment. No plans to have it in the trunk. There is a version of the code (processed by PSYclone) running on GPU. The performance is under investigation and improvement in implementation of directives in the code continues. Work postpone in 2020 as part of HPC-01_daley_GPU
postpone in 2020 as part of HPC-01_daley_GPU
comment:27 Changed 4 years ago by andmirek
In 13090:
In 10827: