More consistent use of combineReduce, simpler and/or reductions

• update code base to reflect the consolidated list/map reductions
• replace old code use of returnReduce(list.size(), sumOp<label>()) in favour of returnReduceOr(list.size()) which is cleaner to write/understand and also marginally more efficient

General overview under https://develop.openfoam.com/Development/openfoam/-/wikis/coding/patterns/parallel

• broadcast
• reductions
• more reductions

changed milestone to %v2212

requested review from @kuti

assigned to @andy

@andy, @kuti, @Mattijs - although currently tagged as draft the MR should be complete to go. I have rebased onto the current develop, but most of the testing was already done in August.

However, given the number of lines being hit with the changes, an extra set of eyes (the more the better) is useful (ie, please review sooner, rather than later). Most of the changes are fairly trivial, but in a few places the logic has been untwisted slightly.

• saw returnReduceAnd. Maps to a native mpi call. Any knowledge of speedups? (since potentially no return trip necessary)
• I have to get used to these type of constructs

- @@ -188,11 +188,12 @@ void Foam::surfaceToCell::combine(topoSet& set, const bool add) const
         {
             // Find cell point is in. Linear search.
             label celli = queryMesh.findCell(outsidePoint, -1, false);
-            if (returnReduce(celli, maxOp<label>()) == -1)
+
+            if (returnReduceAnd(celli < 0))
             {
                 FatalErrorInFunction
                     << "outsidePoint " << outsidePoint

• Discouraging any further use of mapCombineAllGather, listCombineAllGather and combineAllGather in Pstream.H by deprecating them via FOAM_DEPRECATED_FOR?
• Should all scatter be replaced by broadcast incl. globalIndex? Example: metisLikeDecomp.C#L178.

@Mattijs - would you mind to double check the change below in snappyHexMesh.C#L419?

(old)

    forAllConstIters(patchSize, iter)
    {
        label sz = compactZoneID.size();
        compactZoneID.insert(iter.key(), sz);
    }

vs (new)

    if (Pstream::master())
    {
        forAllConstIters(patchSize, iter)
        {
            compactZoneID.insert(iter.key(), compactZoneID.size());
        }
    }

• Do we need to add the sort in parLagrangianDistributor.C#L132 (& #L93)? Changing the previous behaviour?

• linux64ClangDPInt32Opt (clang11)
• linux64GccDPInt32Opt
• linux64GccSPDPInt64Debug
• Alltest: No new error/No change in existing output

resolved all threads

approved this merge request

added 4 commits

• 12470219...3c214e99 - 2 commits from branch develop
• 9e180c98 - ENH: more consistent use of broadcast, combineReduce etc.
• 63f56f98 - ENH: use returnReduceAnd(), returnReduceOr() functions

Compare with previous version

marked this merge request as ready

added 27 commits

• 63f56f98...35aa6140 - 25 commits from branch develop
• 4507a3aa - ENH: more consistent use of broadcast, combineReduce etc.
• a98c7865 - ENH: use returnReduceAnd(), returnReduceOr() functions

Compare with previous version

added 3 commits

• 37893a5b - ENH: PatchTools::gatherAndMerge with recovery of the globalIndex
• dbd59548 - ENH: preserve globalIndex merge information within mergedSurf
• 2bfe085f - ENH: consolidate 'formatOptions' handling for coordSetWriter/surfaceWriter

Compare with previous version

resolved all threads

added 9 commits

• 2bfe085f...b9c15b85 - 4 commits from branch develop
• 473e1441 - ENH: more consistent use of broadcast, combineReduce etc.
• 70208a73 - ENH: use returnReduceAnd(), returnReduceOr() functions
• 799d2471 - ENH: PatchTools::gatherAndMerge with recovery of the globalIndex
• b7592c1e - ENH: preserve globalIndex merge information within mergedSurf
• 5b29ff0e - ENH: consolidate 'formatOptions' handling for coordSetWriter/surfaceWriter

Compare with previous version

mentioned in commit 7fa4f1ef

merged