[LGR] Write complete restart records for local-grid runs by arturcastiel · Pull Request #5246 · OPM/opm-common

arturcastiel · 2026-07-22T14:37:18Z

What this does

Corrects the per-grid header content that OPM flow writes for models with
local grid refinements (CARFIN): a local grid's INTEHEAD well- and
group-allocation items are now per-grid quantities, and the header records
the model's refinement count. Previously these items carried field-wide
values or were left at zero.

As requested in review, the additional header items
(lgrRestartHeaderItems), the solution-term count, and the local-grid
solution-section resequencing have been removed from this PR. They are not
needed for restarting our own simulations: flow currently rejects
refined-grid restarts at initialization (FlowProblemBlackoil.hpp), the
restart loader reads INTEHEAD exclusively through the named items of
VectorItems/intehead.hpp (none of the removed items had a named entry),
and solution arrays are looked up by name, so section order is immaterial
to the loader. With the resequencing gone, RestartIO.cpp is no longer
touched by this PR.

The changes

Per-grid NWMAXZ (CreateInteHead): a local grid's INTEHEAD
well-allocation size is now max(wells-in-that-grid, 1) instead of the
field-global WELLDIMS(1) — the slot is a per-grid quantity. Follows the
sibling per-grid fix precedent in the same overload.
Per-grid group accounting (CreateInteHead): a local grid's
actual-group and maximum-group header items now hold the number of
distinct groups of the wells inside that grid (minimum one), while the
maximum-wells-per-group item keeps the field-wide capacity; the two
conventions coincide on single-group models. This resolves the
long-standing placeholder in the local-grid group accounting.
INTEHEAD[NLGR]: the model-total count of local grid refinements,
written to every grid's header. The item identifies the file as a
local-grid restart; it was previously left at zero.
Tests: test_HeadersLGR extended with per-grid NWMAXZ and NLGR
assertions across the existing LGR fixtures.

Scope guarantee

Every addition is gated on the model actually containing local grid
refinements — restart output for non-LGR models is unchanged.

Testing performed

Unit battery on this branch: the header and restart suites
(test_HeadersLGR, test_InteHEAD, test_Restart, test_RestartLGR,
test_RestartFileView) and the full opm-common test suite pass.

arturcastiel · 2026-07-22T14:37:31Z

jenkins build this please

arturcastiel · 2026-07-22T15:21:48Z

jenkins build this please

arturcastiel · 2026-07-22T17:50:32Z

jenkins build this please

arturcastiel · 2026-07-23T08:56:08Z

jenkins build this please

arturcastiel · 2026-07-23T12:37:47Z

jenkins build this please

arturcastiel · 2026-08-17T08:41:10Z

@bska

bska · 2026-08-18T09:46:58Z

This PR is doing too much work in my opinion and I'd really appreciate it if it were split into more parts. In particular, I'd like the tracer bits to be a separate, follow-on PR instead of included here.

arturcastiel · 2026-08-18T09:48:45Z

This PR is doing too much work in my opinion and I'd really appreciate it if it were split into more parts. In particular, I'd like the tracer bits to be a separate, follow-on PR instead of included here.

I will split into more pieces.

arturcastiel · 2026-08-18T12:47:36Z

jenkins build this please

arturcastiel · 2026-08-26T09:32:30Z

jenkins build this please

arturcastiel · 2026-08-26T13:13:17Z

@bska

arturcastiel · 2026-08-26T13:55:15Z

This PR is doing too much work in my opinion and I'd really appreciate it if it were split into more parts. In particular, I'd like the tracer bits to be a separate, follow-on PR instead of included here.

The split is done. This PR is now the six-commit core on current master:
the four header commits (per-grid NWMAXZ, refinement count, local-grid
group items, additional header items), the solution-layout commit, and
the test pins. It merges cleanly against master.

The tracer-counting change is out of this PR entirely and will follow as
a separate PR, re-authored on the reduced Aggregate* signatures rather
than rebased mechanically.

After the rebase we re-ran the full validation set on this head: the
complete unit suite plus restart-coverage runs across the local-grid
model matrix (single- and multi-refinement, wells-in-LGR, and amalgamated
cases). Results are at par with the pre-rebase branch across the matrix;
the amalgamated-LGR case now completes cleanly on current master. The
tracer cases are covered by the follow-up PR.

bska

Thanks a lot for splitting up the original PR. There is still too much happening here, and some of the INTEHEAD values are too obtuse. Unless we need them for restarting our own simulations, I suggest we leave them out for now. If we do need them, then they should be properly documented. The weird 3D array ordering in the restart file should similarly not be added unless it's needed four our own simulation restart or postprocessors like ResInsight.

bska · 2026-08-26T15:48:05Z

+    // Solution-term count as recomputed by restarting runs: active
+    // phases plus two, plus one when a dissolution or vaporisation
+    // transfer is active.
+    const auto& phasePred = rspec.phases();
+    const int   numActivePhases =
+        static_cast<int>(phasePred.active(Phase::OIL))
+        + static_cast<int>(phasePred.active(Phase::WATER))
+        + static_cast<int>(phasePred.active(Phase::GAS));
+    const int solutionTermCount = numActivePhases + 2
+        + ((simConfig.hasDISGAS() || simConfig.hasVAPOIL()) ? 1 : 0);


We don't this solutionTermCount for our simulation restart so please remove it.

bska · 2026-08-26T15:53:10Z

+    // fluid-in-place arrays leading the section, or placed after the
+    // saturation-pressure terms, is rejected while being read back.
+    template <typename OutputVector, typename OutputVectorInt>
+    void writeLgrOrderedSolutionVectors(const RestartValue& value,


We're not doing this at this point. Please continue to use the existing order for everything.

bska · 2026-08-26T16:00:54Z

+        InteHEAD& lgrRestartHeaderItems(const bool modelHasLgr,
+                                          const int  solutionTermCount);


Do we need the lgrRestartHeaderItems() for restarting our own simulations? If not, I'd really prefer that we not define them.

The local-grid overload of getWellTableDims() dimensioned NWMAXZ with the field-wide WELLDIMS item 1 value. NWMAXZ is a per-grid allocation size — the number of wells located in that grid, with a minimum of one — so the local-grid header now carries the local grid's own well count. Co-authored-by: AXON <axon@arturcastiel.github.io>

Write the model-total count of local grid refinements to INTEHEAD item 95 of every grid's header. The item identifies the file as a local-grid restart and was previously left at zero. Co-authored-by: AXON <axon@arturcastiel.github.io>

A local grid's actual-group and maximum-group header items are per-grid quantities: the number of distinct groups of the wells inside that grid, with a minimum of one. The local-grid path copied the field-wide values into all three group items, which coincides on single-group models only. The maximum-wells-per-group item keeps the field-wide capacity. This resolves the long-standing placeholder in the local-grid group accounting, which is removed. Co-authored-by: AXON <axon@arturcastiel.github.io>

Co-authored-by: AXON <axon@arturcastiel.github.io>

arturcastiel · 2026-08-27T08:37:52Z

All three points taken — done:

lgrRestartHeaderItems() and the solutionTermCount computation are
removed.
The solution-section resequencing is removed as well; RestartIO.cpp is
no longer touched by this PR, so record order is the existing one
everywhere.

The branch is now the four-commit core: per-grid NWMAXZ, the refinement
count, per-grid group items, and the test pins. Body updated to match.

arturcastiel · 2026-08-27T08:47:45Z

jenkins build this please

arturcastiel added the manual:enhancement This is an enhancement/improvent that needs to be documented in the manual label Jul 22, 2026

arturcastiel force-pushed the lgr-restart-nwmaxz branch from ae27beb to 3230816 Compare July 23, 2026 12:36

arturcastiel marked this pull request as ready for review July 23, 2026 12:41

arturcastiel requested a review from bska July 23, 2026 12:41

arturcastiel changed the title ~~[WIP] [LGR] Flow Run + Ref Simulator Restart Enabled~~ [LGR] Flow Run + Ref Simulator Restart Enabled Jul 23, 2026

arturcastiel force-pushed the lgr-restart-nwmaxz branch from 3230816 to 1ef3359 Compare August 18, 2026 12:46

arturcastiel changed the title ~~[LGR] Flow Run + Ref Simulator Restart Enabled~~ [LGR] Write complete restart records for local-grid runs Aug 18, 2026

arturcastiel force-pushed the lgr-restart-nwmaxz branch from 87e08c6 to e3607a0 Compare August 26, 2026 09:31

bska requested changes Aug 26, 2026

View reviewed changes

arturcastiel and others added 4 commits August 27, 2026 08:17

test_HeadersLGR: pin per-grid NWMAXZ and refinement-count header items

e999392

Co-authored-by: AXON <axon@arturcastiel.github.io>

arturcastiel force-pushed the lgr-restart-nwmaxz branch from e3607a0 to e999392 Compare August 27, 2026 08:42

arturcastiel mentioned this pull request Aug 27, 2026

[LGR] Restart records follow-up: count tracer components by solution-tracer support #5323

Draft

		InteHEAD& lgrRestartHeaderItems(const bool modelHasLgr,
		const int solutionTermCount);

Conversation

arturcastiel commented Jul 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this does

The changes

Scope guarantee

Testing performed

Uh oh!

arturcastiel commented Jul 22, 2026

Uh oh!

arturcastiel commented Jul 22, 2026

Uh oh!

arturcastiel commented Jul 22, 2026

Uh oh!

arturcastiel commented Jul 23, 2026

Uh oh!

arturcastiel commented Jul 23, 2026

Uh oh!

arturcastiel commented Aug 17, 2026

Uh oh!

bska commented Aug 18, 2026

Uh oh!

arturcastiel commented Aug 18, 2026

Uh oh!

arturcastiel commented Aug 18, 2026

Uh oh!

arturcastiel commented Aug 26, 2026

Uh oh!

arturcastiel commented Aug 26, 2026

Uh oh!

arturcastiel commented Aug 26, 2026

Uh oh!

bska left a comment

Choose a reason for hiding this comment

Uh oh!

bska Aug 26, 2026

Choose a reason for hiding this comment

Uh oh!

bska Aug 26, 2026

Choose a reason for hiding this comment

Uh oh!

bska Aug 26, 2026

Choose a reason for hiding this comment

Uh oh!

arturcastiel commented Aug 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

arturcastiel commented Aug 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

arturcastiel commented Jul 22, 2026 •

edited

Loading

arturcastiel commented Aug 27, 2026 •

edited

Loading