[flang-commits] [flang] [llvm] [flang][OpenMP] Decompose compound constructs, do recursive lowering (PR #90098)

[Sergio Afonso via flang-commits]

================
@@ -1447,43 +1397,47 @@ genParallelOp(Fortran::lower::AbstractConverter &converter,
 
   // TODO Merge with the reduction CB.
   genInfo.setGenRegionEntryCb(genRegionEntryCB).setDataSharingProcessor(&dsp);
-  return genOpWithBody<mlir::omp::ParallelOp>(genInfo, clauseOps);
+  return genOpWithBody<mlir::omp::ParallelOp>(genInfo, queue, item, clauseOps);
 }
 
 static mlir::omp::SectionOp
 genSectionOp(Fortran::lower::AbstractConverter &converter,
              Fortran::lower::SymMap &symTable,
              Fortran::semantics::SemanticsContext &semaCtx,
-             Fortran::lower::pft::Evaluation &eval, bool genNested,
-             mlir::Location loc, const List<Clause> &clauses) {
+             Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
+             const List<Clause> &clauses, const ConstructQueue &queue,
+             ConstructQueue::iterator item) {
   // Currently only private/firstprivate clause is handled, and
   // all privatization is done within `omp.section` operations.
   return genOpWithBody<mlir::omp::SectionOp>(
       OpWithBodyGenInfo(converter, symTable, semaCtx, loc, eval,
                         llvm::omp::Directive::OMPD_section)
-          .setGenNested(genNested)
-          .setClauses(&clauses));
+          .setClauses(&clauses),
+      queue, item);
 }
 
 static mlir::omp::SectionsOp
 genSectionsOp(Fortran::lower::AbstractConverter &converter,
               Fortran::lower::SymMap &symTable,
               Fortran::semantics::SemanticsContext &semaCtx,
               Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
-              const mlir::omp::SectionsClauseOps &clauseOps) {
+              const List<Clause> &clauses, const ConstructQueue &queue,
+              ConstructQueue::iterator item) {
+  mlir::omp::SectionsClauseOps clauseOps;
+  genSectionsClauses(converter, semaCtx, clauses, loc, clauseOps);
----------------
skatrak wrote:

The previous implementation processed these clauses before the `omp.parallel` operation in the case of PARALLEL SECTIONS, but now it looks like they would be processed afterwards. This means any new operations created as a result would be created inside of the `omp.parallel` region rather than outside, as it used to be the case.

Is this change intended? If so, does the spec say something to support this lowering rather than what was previously implemented?

https://github.com/llvm/llvm-project/pull/90098