Barotropic weight vectorization #202
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Codecov Report
@@ Coverage Diff @@
## dev/gfdl #202 +/- ##
============================================
- Coverage 37.13% 37.01% -0.12%
============================================
Files 262 261 -1
Lines 72706 71361 -1345
Branches 13583 13354 -229
============================================
- Hits 27000 26416 -584
+ Misses 40702 40041 -661
+ Partials 5004 4904 -100
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
Hallberg-NOAA
approved these changes
Sep 12, 2022
This patch modifies the calculation of barotropic weight functions wt_[uv] to a more vectorizable form. The if-blocks used to evaluate the piecewise function was replaced with a sequence of min/max evaluations, which are vectorizable. A nondimensional subroundoff term is also introducted to address the division by zero in the evaluation of 1 - 0.5 * Instep / visc_rem. When visc_rem is zero, this term will never be selected by the max() operator, since it is negatively asymptotic. When visc_rem is sufficiently large to accept this term, the value of subroundoff is too small to modify the significand of the floating point value. In our runs, the overall time of btstep was reduced by 6%. Experiment: benchmark Platform: AMD Ryzen 5 2600 Compiler: gfortran 12.2 Flags: -O3 -mavx Timer: (Ocean BT pre-calcs only) Before: - 1.781952 - 1.787916 - 1.790447 After: - 1.544491 - 1.553111 - 1.544491 Timer: (Ocean barotropic mode stepping) Before: - 4.042281 - 4.050225 - 4.051413 After: - 3.785838 - 3.822393 - 3.835511 Speedup: ~13% of "(Ocean BT pre-calc step)" ~6% speedup of btstep
Hallberg-NOAA
force-pushed
the
btstep_wt_vec
branch
from
9323934
to
974e1ea
Compare
|
This PR has passed pipeline testing at https://gitlab.gfdl.noaa.gov/ogrp/MOM6/-/pipelines/16753. |
chzhangudel
pushed a commit
to chzhangudel/MOM6
that referenced
this pull request
Sep 29, 2022
Support for KPP Nonlocal term beyond T & S
When APPLY_NONLOCAL_TRANSPORT=True, the KPP Nonlocal term will be applied to
the tracers in the MOM_CFC_cap module and the tracer in the pseudo_salt_tracer
module (if they are active in the run).
To support this, some of the KPP diagnostics were moved from the KPP_CS type to
the tracer_type data structure and the KPP_NonLocalTransport_temp() and
KPP_NonLocalTransport_saln() functions were refactored to put common code in
KPP_NonLocalTransport() (which can be called from any of the tracer modules).
Additionally, scaling terms for the diagnostics available from the
KPP_NonLocalTransport() function are encoded via the conversion argument of the
register_diag_field() call in register_tracer_diagnostics(). This results in a
round-off level change to the KPP_NLT_temp_budget and KPP_NLT_saln_budget
diagnostics.
Lastly, two structural changes to the code:
1. To avoid a circular dependency between MOM_variables.F90 and
MOM_tracer_registry.F90, tracer_type and tracer_registry_type have been
moved to a new module named MOM_tracer_types.F90. Both of those types remain
public in MOM_tracer_registry.F90, so they are still available through "use
MOM_tracer_registry" statements.
2. Some code in MOM_CFC_cap.F90 was refactored to replace nearly-repeated lines
(first for CFC11 and then for CFC12) with loops over an array of the new
CFC_tracer_metadata type.
This PR does not change answers for MOM6-examples when compared against MOM6/main.
There is a minor change in MOM_parameter_doc.all for the following tests because the description of APPLY_NONLOCAL_TRANSPORT has been updated:
```
ocean_only/CVmix_SCM_tests/cooling_only/KPP/
ocean_only/CVmix_SCM_tests/mech_only/KPP/
ocean_only/CVmix_SCM_tests/skin_warming_wind/KPP/
ocean_only/CVmix_SCM_tests/wind_only/KPP/
ocean_only/SCM_idealized_hurricane/
ocean_only/SCM_idealized_hurricane/
ocean_only/single_column/EPBL/
ocean_only/single_column/KPP/
```
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This patch modifies the calculation of barotropic weight functions
wt_[uv] to a more vectorizable form. The if-blocks used to evaluate the
piecewise function was replaced with a sequence of min/max evaluations,
which are vectorizable.
A nondimensional subroundoff term is also introducted to address the
division by zero in the evaluation of 1 - 0.5 * Instep / visc_rem.
When visc_rem is zero, this term will never be selected by the max()
operator, since it is negatively asymptotic. When visc_rem is
sufficiently large to accept this term, the value of subroundoff is too
small to modify the significand of the floating point value.
In our runs, the overall time of btstep was reduced by 6%.
Experiment: benchmark
Platform: AMD Ryzen 5 2600
Compiler: gfortran 12.2
Flags: -O3 -mavx
Timer: (Ocean BT pre-calcs only)
Before:
After:
Timer: (Ocean barotropic mode stepping)
Before:
After:
Speedup:
~13% of "(Ocean BT pre-calc step)"
~6% speedup of btstep