To update FV3 dynamic core to the GFDL 201912 released version#58
Conversation
… in atmos_cubed_sphere
…N=Y (DEFAULT).Startinging with FMS 2019.01.01, quad-precision arithmetic is enabled it the ENABLE_QUAD_PRECISION macro. This is used to determine whether certain grid metric terms code within FMS/FV3-dycore use quad-precision arithmetic.
climbfuji
left a comment
climbfuji
left a comment
There was a problem hiding this comment.
A few comments, no show stoppers, but I would like to hear back if possible. If any changes/updates are needed those can be done on a follow-up PR.
| endif() | ||
|
|
||
| if(QUAD_PRECISION) | ||
| add_definitions(-DENABLE_QUAD_PRECISION) |
There was a problem hiding this comment.
Why is there no
message("QUAD_PRECISION is ENABLED")
?
There was a problem hiding this comment.
I see it further down, why not make the same change as for GNU and PGI?
| option(32BIT "Enable 32BIT (single precision arithmetic in dycore)" OFF) | ||
| option(OPENMP "Enable OpenMP threading" ON) | ||
| option(AVX2 "Enable AVX2 instruction set" OFF) | ||
| option(QUAD_PRECISION "Enable QUAD_PRECISION (for certain grid metric terms in dycore)" ON) |
There was a problem hiding this comment.
Has anyone tested this new FMS release / quad precision flag with GNU (on cheyenne, or generic linux, or generic macOS)?
There was a problem hiding this comment.
I didn't run this version with GNU compiler. If there are issues, we must fix them in the next PR.
|
|
||
| &fv_core_nml | ||
| layout = 24,20 | ||
| layout = 18,24 |
There was a problem hiding this comment.
Why do we need to change this? Because we need to take processors away for the write component?
The layout = 24,20 in the original code guarantees uniform block size. Is this the case for 18,24 as well?
There was a problem hiding this comment.
Yes, I change it to use same number of the processors while using write component.
There was a problem hiding this comment.
Does this still give uniform block sizes? That is important. You can look at the out and/or err files for this job, there is a warning if non-uniform block sizes are used.
There was a problem hiding this comment.
Do you mean "the non-divisible blocksize"? There is no "the non-divisible blocksize" warning as the write component has issues with it.
There was a problem hiding this comment.
Each block has the uniform size.
| atmos_4xdaily.tile5.nc \ | ||
| atmos_4xdaily.tile6.nc \ | ||
| atmos_4xdaily.nest02.tile7.nc \ | ||
| atmos_4xdaily.nest02.nc \ |
There was a problem hiding this comment.
Note that this test is not exercised in rt.conf.
| VERBOSE = | ||
| OPENMP = Y | ||
| HYDRO = N | ||
| QUAD_PRECISION = Y |
There was a problem hiding this comment.
wcoss_phase1/2 is not supported any more. We will plan to remove wcoss phase1/2 related files.
| AVX2 = Y | ||
| HYDRO = N | ||
| CCPP = N | ||
| QUAD_PRECISION = Y |
There was a problem hiding this comment.
With this setting QUAD_PRECISION is set to Y, will all the RT test results be changed on cray?
There was a problem hiding this comment.
No. Only atmos_4xdaily.* files are changed due to change in netcdf file format in FMS.
|
|
||
| if [[ $MACHINE_ID = hera.* ]] || [[ $MACHINE_ID = orion.* ]] || [[ $MACHINE_ID = cheyenne.* ]]; then | ||
| RTPWD=${RTPWD:-$DISKNM/NEMSfv3gfs/develop-20200603/${COMPILER^^}} | ||
| RTPWD=${RTPWD:-$DISKNM/NEMSfv3gfs/develop-20200608/${COMPILER^^}} |
There was a problem hiding this comment.
May I ask what regression test results are changed with this commit from 20200603 to 20200608 on platform hera, dell, cray and orion?
There was a problem hiding this comment.
All nesting and regional tests will have different results. also atmos_4xdaily.tile?.nc has identical fields but the head differ.
There was a problem hiding this comment.
All atmos_4xdaily.* files are changed in all tests that write out those files. Then in regional_768 test we switched from using diag_manager to write component.
junwang-noaa
left a comment
junwang-noaa
left a comment
There was a problem hiding this comment.
I have some questions on the impact of results. The code change itself looks good.
|
So only atmos_4xdaily files are different with the FMS version update from
2019.0.1 to 2020.02 on all the platforms (orion/hera/dell/cray)?
…On Wed, Jun 10, 2020 at 1:23 PM Dusan Jovic ***@***.***> wrote:
***@***.**** commented on this pull request.
------------------------------
In conf/configure.fv3.wcoss_cray
<#58 (comment)>
:
> @@ -21,6 +21,7 @@ OPENMP = Y
AVX2 = Y
HYDRO = N
CCPP = N
+QUAD_PRECISION = Y
No. Only atmos_4xdaily.* files are changed due to change in netcdf file
format in FMS.
—
You are receiving this because your review was requested.
Reply to this email directly, view it on GitHub
<#58 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AI7D6TM6JQZ2JNQHLIWMCNTRV66QZANCNFSM4KU2RPWQ>
.
|
|
*@climbfuji* Yes, I checked with dimension (1728, 1440), the layout gives
uniform block size.
…On Wed, Jun 10, 2020 at 1:52 PM XiaqiongZhou-NOAA ***@***.***> wrote:
***@***.**** commented on this pull request.
------------------------------
In parm/ccpp_regional_c768.nml.IN
<#58 (comment)>
:
> @@ -36,7 +36,7 @@
/
&fv_core_nml
- layout = 24,20
+ layout = 18,24
Do you mean "the non-divisible blocksize"? There is no "the non-divisible
blocksize" warning as the write component has issues with it.
—
You are receiving this because your review was requested.
Reply to this email directly, view it on GitHub
<#58 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AI7D6TO737QN7J3N5YBVEH3RV7B7PANCNFSM4KU2RPWQ>
.
|
…ointer_fv3atm Update submodule pointer fv3atm 2020/07/01
…ase/public-v2 Revert "merge ufs-release/public-v2 -> dev/emc"
4 participants
To update FV3 dynamic core to the GFDL 201912 released version,
the namelist files in ~parm should be revised due to the namelist changes for nesting
Related pull requests:
atmos_cubed_sphere #15
fv3atm #65
ccpp-physics #397