[flang-commits] [flang] [flang][OpenMP] Lower `target .. private(..)` to `omp.private` ops (PR #94195)

[Kiran Chandramohan via flang-commits]

================
@@ -758,15 +758,33 @@ genBodyOfTargetOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
                   llvm::ArrayRef<const semantics::Symbol *> mapSyms,
                   llvm::ArrayRef<mlir::Location> mapSymLocs,
                   llvm::ArrayRef<mlir::Type> mapSymTypes,
+                  DataSharingProcessor &dsp,
                   const mlir::Location &currentLocation,
                   const ConstructQueue &queue, ConstructQueue::iterator item) {
   assert(mapSymTypes.size() == mapSymLocs.size());
 
   fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
   mlir::Region &region = targetOp.getRegion();
-
-  auto *regionBlock =
-      firOpBuilder.createBlock(&region, {}, mapSymTypes, mapSymLocs);
+  mlir::OperandRange privateVars = targetOp.getPrivateVars();
+
+  llvm::SmallVector<mlir::Type> allRegionArgTypes;
+  allRegionArgTypes.reserve(mapSymTypes.size() +
+                            targetOp.getPrivateVars().size());
+  llvm::transform(mapSymTypes, std::back_inserter(allRegionArgTypes),
+                  [](mlir::Type t) { return t; });
+  llvm::transform(privateVars, std::back_inserter(allRegionArgTypes),
+                  [](mlir::Value v) { return v.getType(); });
+
+  llvm::SmallVector<mlir::Location> allRegionArgLocs;
+  allRegionArgTypes.reserve(mapSymTypes.size() +
+                            targetOp.getPrivateVars().size());
+  llvm::transform(mapSymLocs, std::back_inserter(allRegionArgLocs),
+                  [](mlir::Location l) { return l; });
+  llvm::transform(privateVars, std::back_inserter(allRegionArgLocs),
+                  [](mlir::Value v) { return v.getLoc(); });
+
+  auto *regionBlock = firOpBuilder.createBlock(&region, {}, allRegionArgTypes,
+                                               allRegionArgLocs);
----------------
kiranchandramohan wrote:

Add a comment for this block of code. Is it possible to share any of it with the parallel operation? Or is it not possible due to the isolated from above nature of target?

https://github.com/llvm/llvm-project/pull/94195