Opened 14 months ago

Last modified 7 months ago

#2482 assigned Task

MO GO8 for GPU

Reported by: andmirek Owned by: andmirek
Priority: low Milestone: Unscheduled
Component: MULTIPLE Version: 4.0-HEAD
Severity: minor Keywords:
Cc: andrew.porter@…, wgaudin@…, smueller Branch review: failed
MP ready?: no Task progress: Unspecified

Description (last modified by andmirek)

This ticket is to record code changes in Met Office? GO8 (p2.0) configuration (ORCA025) for code adaptation for GPU. This is continuation of the work documented in ticket #2197 for ORCA1.

Branches
ORCA1/ORCA025: svn+ssh://andmirek@forge.ipsl.jussieu.fr/ipsl/forge/projets/nemo/svn/NEMO/branches/UKMO/NEMO_4.0_mirror_SI3_GPU
ORCA012: svn+ssh://andmirek@forge.ipsl.jussieu.fr/ipsl/forge/projets/nemo/svn/NEMO/branches/UKMO/NEMO_4.0_mirror_SI3_GPU_O12

Commit History (19)

ChangesetAuthorTimeChangeLog
14258andmirek2021-01-04T16:19:17+01:00

Ticket #2482: few changes to improve code on GPU

13700andmirek2020-10-29T10:18:48+01:00

Ticket #2482: merge branch with the changes for ORCA012

13699andmirek2020-10-29T10:15:42+01:00

Ticket #2482: MO GO8 ORCA012 configuration

13548andmirek2020-10-01T09:47:52+02:00

Ticket #2482: update architecture file for GPU

13330andmirek2020-07-21T14:27:10+02:00

Ticket #2482: changes in arch file for cpp on Power/Intel? and additional library for profiling

13326andmirek2020-07-20T13:03:06+02:00

Ticket #2482: Update compiler file

13325andmirek2020-07-20T09:12:12+02:00

Ticket ##2482: Use explicit index in dynspg_ts.F90

13307andmirek2020-07-15T10:38:15+02:00

Ticket #2482: Few additional corrections

13306andmirek2020-07-15T10:32:38+02:00

Ticket #2482: Replace ifdef with if defined in iom.F90

13153andmirek2020-06-24T15:23:31+02:00

Ticket #2482: fixes after testing with ifort

13149andmirek2020-06-24T09:03:45+02:00

Ticket #2482: Dissable restart and use allocatable arrays

13092andmirek2020-06-10T16:58:06+02:00

Ticket #2482 remove print statement

13091andmirek2020-06-10T16:56:12+02:00

Ticket #2482: architecture file for Isambard (GPU with managed memory)

13090andmirek2020-06-10T16:44:43+02:00

Ticket #2482: calculate sums in the same way as in #2197, commit 11069

13052andmirek2020-06-05T10:18:24+02:00

Ticket #2482: New arch location

13051andmirek2020-06-05T10:17:33+02:00

Ticket #2482: arch file for Isambard CPU build

13050andmirek2020-06-05T10:11:46+02:00

Ticket #2482: new arch configurations

13049andmirek2020-06-05T09:53:58+02:00

Ticket #2482: fixes to build with PGI compiler

13048andmirek2020-06-05T09:08:20+02:00

Ticket #2482 development branch

Change History (26)

comment:1 Changed 14 months ago by andmirek

In 13048:

Ticket #2482 development branch

comment:2 Changed 14 months ago by andmirek

  • Description modified (diff)

comment:3 Changed 14 months ago by andmirek

In 13049:

Ticket #2482: fixes to build with PGI compiler

comment:4 Changed 14 months ago by andmirek

In 13050:

Ticket #2482: new arch configurations

comment:5 Changed 14 months ago by andmirek

In 13051:

Ticket #2482: arch file for Isambard CPU build

comment:6 Changed 14 months ago by andmirek

In 13052:

Ticket #2482: New arch location

comment:7 Changed 14 months ago by smueller

  • Cc smueller added

comment:8 Changed 14 months ago by andmirek

  • Description modified (diff)

comment:9 Changed 14 months ago by andmirek

Code runs on Isambard with PGI compiler on 36 processors (10 timesteps). There are warnings in job.output:

Warning: ieee_overflow is signaling
Warning: ieee_underflow is signaling
Warning: ieee_inexact is signaling

I wasn't able to track ieee_overflow. The point, where model stops when this exception is trapped, doesn't produce overflow but underflow - in tke_tke (zdftke.F90) in loop IF( nn_etau == 1 ) THEN !* penetration below the mixed layer (rn_efr fraction).

Intel compiler runs without any problems with fpe0 and solutions for both compilers are similar (based on run.stat).

Instruction how to build and set up the code on Isambard /home/mo-mandrej/GO8_ORCA025_GPU/orca25_gpu_build.

comment:10 Changed 14 months ago by andmirek

In 13090:

Ticket #2482: calculate sums in the same way as in #2197, commit 11069

comment:11 Changed 14 months ago by andmirek

In 13091:

Ticket #2482: architecture file for Isambard (GPU with managed memory)

comment:12 Changed 14 months ago by andmirek

In 13092:

Ticket #2482 remove print statement

comment:13 Changed 13 months ago by andmirek

In 13149:

Ticket #2482: Dissable restart and use allocatable arrays

comment:14 Changed 13 months ago by andmirek

In 13153:

Ticket #2482: fixes after testing with ifort

comment:15 Changed 13 months ago by andmirek

In 13306:

Ticket #2482: Replace ifdef with if defined in iom.F90

comment:16 Changed 13 months ago by andmirek

In 13307:

Ticket #2482: Few additional corrections

comment:17 Changed 12 months ago by andmirek

In 13325:

Ticket ##2482: Use explicit index in dynspg_ts.F90

comment:18 Changed 12 months ago by andmirek

In 13326:

Ticket #2482: Update compiler file

comment:19 Changed 12 months ago by andmirek

In 13330:

Ticket #2482: changes in arch file for cpp on Power/Intel? and additional library for profiling

comment:20 Changed 10 months ago by andmirek

In 13548:

Ticket #2482: update architecture file for GPU

comment:21 Changed 9 months ago by andmirek

In 13699:

Ticket #2482: MO GO8 ORCA012 configuration

comment:22 Changed 9 months ago by andmirek

In 13700:

Ticket #2482: merge branch with the changes for ORCA012

comment:23 Changed 9 months ago by andmirek

  • Description modified (diff)
  • Status changed from new to assigned
  • Type changed from Feature request to Task

comment:24 Changed 8 months ago by mathiot

Is this ticket at the right location ? Should it be in a UKMO trac system instead ?

comment:25 Changed 8 months ago by nemo

  • Version changed from release-4.0 to 4.0-HEAD

comment:26 Changed 7 months ago by andmirek

In 14258:

Ticket #2482: few changes to improve code on GPU

Note: See TracTickets for help on using tickets.