Open
Conversation
github-actions
bot
added
the
cla-required
github-actions
bot
added
cla-signed
Author
|
A selection of results comparing to stable
|
tommbendall
requested changes
Jan 27, 2026
Contributor
tommbendall
left a comment
tommbendall
left a comment
There was a problem hiding this comment.
This is a really good speed-up. I'm happy that this is scientifically equivalent to main and the KGOs are only changing from bit-wise differences in the kernels.
I am generally happy -- the new mixed solver kernels have been literally taken from the existing kernel and split into two parts, making this easy to review.
My main thoughts that aren't captured through my comments on the code:
- Do you prefer to keep the old
apply_mixed_operator_kernel_mod.F90file, which I think is now unused? Or could it be removed? - For the
assemble_w2h_from_w2hb_kernel, there is already a very similar kernel in the core repo: https://github.com/MetOffice/lfric_core/blob/main/components/science/source/kernel/inter_function_space/sci_average_w2b_to_w2_kernel_mod.F90 . I think it could be better to modify the existing kernel to work for both W2H and W2, rather than add a new kernel here? (appreciating that you'd rather not make this a linked ticket, so feel free to refuse) - The timers will conflict with the changes from #68 and #176 -- so I'm suggesting reverting these changes
science/gungho/unit-test/kernel/solver/apply_mixed_wp_operator_kernel_mod_test.pf
Outdated
Show resolved
Hide resolved
science/gungho/unit-test/kernel/solver/apply_mixed_u_operator_kernel_mod_test.pf
Outdated
Show resolved
Hide resolved
| @@ -0,0 +1,86 @@ | |||
| !----------------------------------------------------------------------------- | |||
| ! Copyright (c) 2017, Met Office, on behalf of HMSO and Queen's Printer | |||
Contributor
There was a problem hiding this comment.
This looks like the old copyright statement
| if ( LPROF ) call start_timing( id, 'mixed_operator' ) | ||
| type(r_solver_field_type) :: y_uv_broken | ||
|
|
||
| if ( subroutine_timers ) call timer('mixed_solver.operator') |
Contributor
| @@ -418,7 +414,7 @@ contains | |||
| call invoke( inc_X_times_Y(rhs, h_diag) ) | |||
| end if | |||
|
|
|||
| if ( LPROF ) call stop_timing( id, 'mixed_schur_rhs' ) | |||
| if ( subroutine_timers ) call timer('schur_precon.rhs') | |||
Contributor
|
|
||
| if ( LPROF ) call start_timing( id, 'schur_back_substitute' ) | ||
| type(r_solver_field_type), target :: uvw_norm | ||
| if ( subroutine_timers ) call timer('schur_precon.back_sub') |
Contributor
| @@ -507,7 +504,7 @@ contains | |||
| state_p => state%get_field_from_position(isol_p) | |||
| call invoke( setval_X(state_p, exner_inc) ) | |||
|
|
|||
| if ( LPROF ) call stop_timing( id, 'schur_back_substitute' ) | |||
| if ( subroutine_timers ) call timer('schur_precon.back_sub') | |||
Contributor
|
|
||
| if ( LPROF ) call start_timing( id, 'helmholtz_lhs' ) | ||
| if ( subroutine_timers ) call timer('pressure_solver.helmholtz_lhs') |
Contributor
| @@ -242,7 +241,7 @@ contains | |||
| nullify( w3_mask, w2_mask ) | |||
| end if | |||
| nullify( x_vec, y_vec ) | |||
| if ( LPROF ) call stop_timing( id, 'helmholtz_lhs' ) | |||
| if ( subroutine_timers ) call timer('pressure_solver.helmholtz_lhs') | |||
Contributor
Co-authored-by: Thomas Bendall <14180399+tommbendall@users.noreply.github.com>
Reverting back to stable version
…_kernel_mod_test.pf Co-authored-by: Thomas Bendall <14180399+tommbendall@users.noreply.github.com>
…kernel_mod_test.pf Co-authored-by: Thomas Bendall <14180399+tommbendall@users.noreply.github.com>
Author
|
tommbendall
approved these changes
Jan 28, 2026
Contributor
tommbendall
left a comment
tommbendall
left a comment
There was a problem hiding this comment.
Thanks for the reply. I'm happy with the changes in that case, so this passes science review.
This is now ready for code review @mo-rickywong
mo-rickywong
requested changes
Feb 6, 2026
Contributor
mo-rickywong
left a comment
mo-rickywong
left a comment
There was a problem hiding this comment.
Large number of KGO changes (to be expected).
- Minor copyright issues
- Main issue is the calculated expected answers in the unit tests. Known output for a known input. Generally correct known answers should be hardcoded, not calculated.
General comment on the short variable names which are vague to decipher.
Comment on lines
+266
to
+267
| ! halo exchange. | ||
| ! The second step computes the vertical wind (yvec_w) & pressure (yvec_p) lhs, since these |
Contributor
There was a problem hiding this comment.
Suggested change
| ! halo exchange. | |
| ! The second step computes the vertical wind (yvec_w) & pressure (yvec_p) lhs, since these | |
| ! halo exchange. | |
| ! | |
| ! The second step computes the vertical wind (yvec_w) & pressure (yvec_p) lhs, since these |
Comment on lines
+268
to
+271
| ! are horizontally discontinuous fields then there are no further halo exchanges required. | ||
| ! Alternatively this code could be computed in a single kernel | ||
| ! (apply_mixed_operator_kernel_type), however this appears to have poorer cache usage, | ||
| ! resulting in an increased computation time |
Contributor
There was a problem hiding this comment.
Suggested change
| ! are horizontally discontinuous fields then there are no further halo exchanges required. | |
| ! Alternatively this code could be computed in a single kernel | |
| ! (apply_mixed_operator_kernel_type), however this appears to have poorer cache usage, | |
| ! resulting in an increased computation time | |
| ! are horizontally discontinuous fields then there are no further halo exchanges required. |
Remove last comment, Its a development comment/suggestion which shouldn't be on main. Comments should be on the state of the code on trunk.
| ! resulting in an increased computation time | ||
|
|
||
| ! Create broken y_uv | ||
| w2hb_fs => function_space_collection%get_fs(mesh, 0, 0, W2Hbroken) |
Contributor
There was a problem hiding this comment.
Hardcoded Magic numbers
science/gungho/source/algorithm/solver/mixed_schur_preconditioner_alg_mod.x90
Outdated
Show resolved
Hide resolved
| @@ -0,0 +1,156 @@ | |||
| !----------------------------------------------------------------------------- | |||
| ! (C) Crown copyright 2026 Met Office. All rights reserved. | |||
Contributor
There was a problem hiding this comment.
Suggested change
| ! (C) Crown copyright 2026 Met Office. All rights reserved. | |
| ! (C) Crown copyright Met Office. All rights reserved. |
Comment on lines
+128
to
+145
| t(:) = 0.0_r_solver | ||
| do k = 1, nlayers | ||
| t(k) = t(k) - Pt2(k,1,3)*w(k) - Pt2(k,1,4)*w(k+1) | ||
| t(k+1) = t(k+1) - Pt2(k,2,3)*w(k) - Pt2(k,2,4)*w(k+1) | ||
| end do | ||
| t(:) = invMt(:)*t(:) | ||
|
|
||
| answer(:) = 0.0_r_solver | ||
| do k = 0, nlayers - 1 | ||
| do dfv = 1,ndf_w2v | ||
| u = (/ uv(1+k), uv(4+k), w(1+k), w(2+k) /) | ||
| df = ndf_w2h+dfv | ||
| answer(map_w2v(dfv)+k) = answer(map_w2v(dfv)+k) & | ||
| + Nu(map_w2(df)+k)*(sum(Mu(k+1,df,:)*u) & | ||
| - sum(P2t(k+1,df,:)*t(k+1:k+2)) & | ||
| - Grad(k+1,df,1)*p(1+k)) | ||
| end do | ||
| end do |
Contributor
There was a problem hiding this comment.
Answers shouldn't be calculated, hardcode expected output.
Comment on lines
+150
to
+151
| u = (/ uv(1), uv(4), w(1), w(2) /) | ||
| answer(1) = M3p(1,1,1)*p(1) - sum(P3t(1,1,:)*t(1:2)) + sum(Q32(1,1,:)*u(1:4)) |
Contributor
There was a problem hiding this comment.
Answers shouldn't be calculated, hardcode expected output.
Comment on lines
+95
to
+98
| do df = 1,ndf_w2h | ||
| answer = Nu(df)*(Mu(df,df,1)*uv(df) - Grad(df,1,1)*p(1)) | ||
| @assertEqual(answer, Luv(df), tol) | ||
| end do |
Contributor
There was a problem hiding this comment.
Correct answers shouldn't be calculated, should be hardcoded
Comment on lines
95
to
98
| do df = 1, ndf_w2h | ||
| answer = mask(df)*( 0.5_r_solver*rhs(df) + y(df)*z(df)*grad(1,df,1)*x(1)) | ||
| answer = mask(df)*( 0.5_r_solver*rhs(df) + y(df)*grad(1,df,1)*x(1)) | ||
| @assertEqual(answer, Luv(df), tol) | ||
| end do |
Contributor
There was a problem hiding this comment.
As before known answers shouldn't be calculated, that's what the kernel under test is for.
Co-authored-by: Ricky Wong <141156427+mo-rickywong@users.noreply.github.com>
…ner_alg_mod.x90 Co-authored-by: Ricky Wong <141156427+mo-rickywong@users.noreply.github.com>
…2hb_kernel_mod.F90 Co-authored-by: Ricky Wong <141156427+mo-rickywong@users.noreply.github.com>
…2hb_kernel_mod.F90 Co-authored-by: Ricky Wong <141156427+mo-rickywong@users.noreply.github.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
PR Summary
Sci/Tech Reviewer: @tommbendall
Code Reviewer: @mo-rickywong
Add in a number of solver optimisations.
The principle performance improvement in this pull request is to split the application of the mixed operator into two seperate new kernels.
These changes result in a performance improvement for 2 main reasons
The C224 & C896 lfric atm tests in the test suite were run with these changes giving the following solver times
Code Quality Checklist
Testing
trac.log
These results are from before the KGO update. The failure in the lfric_inputs appears not to be due to this pull request as none of the code changes should be used and is likely one of the occasional lfric_inputs failures that we see
Test Suite Results - lfric_apps - solver_improvements/run6
Suite Information
Task Information
❌ failed tasks - 150
⌛ waiting tasks - 2
Security Considerations
Performance Impact
AI Assistance and Attribution
Documentation
PSyclone Approval
Sci/Tech Review
(Please alert the code reviewer via a tag when you have approved the SR)
Code Review