Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

OpenACC directives #98

Merged
merged 296 commits into from
May 28, 2024
Merged

OpenACC directives #98

merged 296 commits into from
May 28, 2024

Conversation

esclapez
Copy link
Collaborator

  • Move OpenACC porting into dev.

CasparJungbacker and others added 30 commits June 13, 2023 16:16
@CasparJungbacker
Clean up coriolis
4f69897
@CasparJungbacker
Get rid of old loop indices
4498501
@CasparJungbacker
Offload forces
c35a920
@CasparJungbacker
Offload Coriolis
6823fa6
@CasparJungbacker
Offload lstend
1092e8c
@CasparJungbacker
Data movement
53657fc
@CasparJungbacker
Offload gravity wave damping
1fff0ff
@fjansson @CasparJungbacker
modthermodynamics: faster icethermo
a4aae98 
New namelist option in PHYSICS: lfast_thermo. Default is .false. for
the old icethermo. Set to .true. for the new version.

New idea: always use two iterations to determine ql, for better
vectorization.  When iterating, don't calculate numeric derivatives
but use Clausius-Clapeyron. Using tabulated esl and esi as in the
original icethermo, but read from a table combining them, interpolated
with ilratio.

Future optimization possibilities: don't always store esl, qvsl, qvsi,
perhaps also qsat and tmp0. Instead calculate them as needed.
@CasparJungbacker
Merge pull request #5 from CasparJungbacker/OpenACC-forces
fefe973 
Offload forces and gravity wave damping
@CasparJungbacker
Move data directives of base fields
10a6c71 
- Allocate GPU fields in modfields.f90
- Create modgpu.f90/initgpu for updating GPU fields after startup
  routine
- Removed data directive from modadvection.f90
@CasparJungbacker
default(present) all regions
bef5ea4
@CasparJungbacker
Call modgpu from program
3145f52
@CasparJungbacker
Move data clauses for grid thickness arrays
1308946
@CasparJungbacker
Move data clauses for pressure gradients
f926b37
@CasparJungbacker
Remove all data clauses from modforces.f90
e694ce2
@CasparJungbacker
Move more diagnostics
faa5bc0
@CasparJungbacker
Separate loop for large scale tendencies of scalars
57e1d55
@CasparJungbacker
Move data clauses of large scale forcings
1157d26
@CasparJungbacker
Move data clauses for slab averages + fix missing statements
ae67b3a
@CasparJungbacker
Move damping coefficients to gpu
a11c4f6
@CasparJungbacker
Remove remaining exit clauses
699d6d7
@CasparJungbacker
Only launch kernel when there are scalar fields
0f33e6c
@CasparJungbacker
Move surface and subgrid data
caef041
@CasparJungbacker
Offload prescribed radiation
0d6d8e1
@CasparJungbacker
Move radiation data to gpu
119e98c
@CasparJungbacker
Make module-wide arrays for Courant
124b441
@CasparJungbacker
Offload timestep update
8c23732
@CasparJungbacker
Add temporary data copy
958a8e7
@CasparJungbacker
Offload time step integration
1569c74
@CasparJungbacker
Update halo exchange
7f94632
@CasparJungbacker
Copy link
Collaborator

Hmmm, tests are failing in modstartup.f90/readinitfiles with message:

Program received signal SIGFPE: Floating-point exception - erroneous arithmetic operation.

@fjansson looks like the same NetCDF/HDF5 error we saw previously in debug builds?

@fjansson
Copy link
Contributor

It looks like the crash now is in icethermo and not from HDF5 this time. That crash would happen when opening a netcdf file.

@CasparJungbacker
Copy link
Collaborator

I'm having the same issue on my local machine and Snellius. Release build works though.

@esclapez
Copy link
Collaborator Author

Okay, got it. A couple of things missing from the refactoring of the surface schemes.

@fjansson
Copy link
Contributor

One more single-precision error. Perhaps the 5.0 needs a _field_r

src/modthermodynamics.f90:842:58:

  842 |           PrDiff_min = min(PrDiff_min, presf(k) - esat_tab(thl0(i,j,k)*exnf(k) + 5.0))
      |                                                          1
Error: Type mismatch in argument ‘t’ at (1); passed REAL(8) to REAL(4)

@fjansson fjansson merged commit 2e69c31 into dev May 28, 2024
50 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@CasparJungbacker CasparJungbacker CasparJungbacker approved these changes

@fjansson fjansson Awaiting requested review from fjansson

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@esclapez @CasparJungbacker @fjansson @lsoucasse