Reduce memory increase from PRs 1567 and 1616

[weiwangncar]

Reduce memory used by the model from two previous PRs

TYPE: bug fix

KEYWORDS: qnbca, wif_input_opt, package

SOURCE: internal

DESCRIPTION OF CHANGES:
Problem:
PR1567 changed default wif_input_opt from 0 to 1, which increased memory use by about 12%. PR1616 added new variable qnbca, but it is not packaged.

Solution:
Change default wif_input_opt option back to 0, and add qnbca to package for mp_physics=28 and its dfi counterpart. Testing the latest release 4.4 code shows a memory reduction (when mp_physics=28 is not used) is down by 11%.

LIST OF MODIFIED FILES:
M Registry/Registry.EM_COMMON
M Registry/registry.new3d_wif

TESTS CONDUCTED:

1. Do mods fix problem? The fix reduces memory use in the model.
2. The Jenkins tests are all passing.

RELEASE NOTE:

[dudhia]

Need to update PR message if this is the final version.

[weiwangncar]

The Jenkins tests have passed (with adding wif_input_opt = 1 to two of the test namelists):

Test Type              | Expected  | Received |  Failed
= = = = = = = = = = = = = = = = = = = = = = = =  = = = =
Number of Tests        : 23           24
Number of Builds       : 60           58
Number of Simulations  : 158           156        0
Number of Comparisons  : 95           92        0

Failed Simulations are: 
None
Which comparisons are not bit-for-bit: 
None

[weiwangncar]

@dudhia @kkeene44 @twjuliano This PR is ready for review.

[twjuliano]

@weiwangncar Thanks for looking into this memory issue. The only concern I have is that we added the BC aerosols to be included in mp=28 only when wif_input_opt=2 (we kept wif_input_opt=1 the same so that a user can run a simulation without BC). Thus, I added qnbca (but not dfi_qnbca) in the use_wif_input_bc package in the new3d_wif registry. Just want to make sure that it is OK to have qnbca included in the general Thompson package even when wif_input_opt=1 (i.e., BC not included).

[dudhia]

Was #1567 after the last release or will this change the behavior of a release?

[weiwangncar]

@dudhia PR1567 is for v4.4. Is that what you're asking?

[dudhia]

Was it after the last minor release?

…

On Thu, Apr 21, 2022 at 10:18 AM weiwangncar ***@***.***> wrote: @dudhia <https://github.com/dudhia> PR1567 is for v4.4. Is that what you're asking? — Reply to this email directly, view it on GitHub <#1722 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEIZ77B4LODIGDOOF4WJL6LVGF5UXANCNFSM5T4NN2QA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[weiwangncar]

@twjuliano I don't think this matters to your concern. Without this PR, qnbca array is available ALL the time - that is why it increases the memory used by the model whether you are using mp_physics = 28 or not. With this change, the array is made available only to this microphysics option. Does it make sense? It wouldn't hurt if you could run a test to verify that.

[dudhia]

I notice that the default is still wif_input_option = 1 in the registry, and also the description there needs updating to include the 2 option. My view is that it is OK to leave this in the Thompson package even if not used.

…

On Thu, Apr 21, 2022 at 10:06 AM Timothy Juliano ***@***.***> wrote: @weiwangncar <https://github.com/weiwangncar> Thanks for looking into this memory issue. The only concern I have is that we added the BC aerosols to be included in mp=28 only when wif_input_opt=2 (we kept wif_input_opt=1 the same so that a user can run a simulation without BC). Thus, I added qnbca (but not dfi_qnbca) in the use_wif_input_bc package in the new3d_wif registry. Just want to make sure that it is OK to have qnbca included in the general Thompson package even when wif_input_opt=1 (i.e., BC not included). — Reply to this email directly, view it on GitHub <#1722 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEIZ77DRF4LBKRCYDC4V6FLVGF4ITANCNFSM5T4NN2QA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[weiwangncar]

@dudhia Yes. It has been on the develop branch all this time, until the cut-over to release-v4.4.

[weiwangncar]

@dudhia That will change with this PR - this PR sets wif_input_opt = 0.

[dudhia]

OK, need to make sure that if a user forgets to set that to 1 or 2, there is a reasonable message from real and probably a stop for option 28.

[weiwangncar]

@dudhia The default value for wif_input_opt is 0 in previous releases.

[twjuliano]

@twjuliano I don't think this matters to your concern. Without this PR, qnbca array is available ALL the time - that is why it increases the memory used by the model whether you are using mp_physics = 28 or not. With this change, the array is made available only to this microphysics option. Does it make sense? It wouldn't hurt if you could run a test to verify that.

@weiwangncar I see, that makes sense, thanks. I can run the test later today if that isn't too late.

[weiwangncar]

@twjuliano No, it's not too late. I will wait for your test results.

[twjuliano]

@weiwangncar I just ran a test: this works as intended, thanks!

[weiwangncar]

@twjuliano Thanks for running the test!
@kkeene44 Can you review this and if it is ok, approve it?