[llvm-branch-commits] [flang] [MLIR][OpenMP] Add Lowering support for OpenMP Declare Mapper directive (PR #117046)

[Kiran Chandramohan via llvm-branch-commits]

================
@@ -2701,7 +2701,42 @@ static void
 genOMP(lower::AbstractConverter &converter, lower::SymMap &symTable,
        semantics::SemanticsContext &semaCtx, lower::pft::Evaluation &eval,
        const parser::OpenMPDeclareMapperConstruct &declareMapperConstruct) {
-  TODO(converter.getCurrentLocation(), "OpenMPDeclareMapperConstruct");
+  fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+  lower::StatementContext stmtCtx;
+  const auto &spec =
+      std::get<parser::OmpDeclareMapperSpecifier>(declareMapperConstruct.t);
+  const auto &mapperName{std::get<std::optional<parser::Name>>(spec.t)};
+  const auto &varType{std::get<parser::TypeSpec>(spec.t)};
+  const auto &varName{std::get<parser::Name>(spec.t)};
+  assert(varType.declTypeSpec->category() ==
+             semantics::DeclTypeSpec::Category::TypeDerived &&
+         "Expected derived type");
+
+  std::string mapperNameStr;
+  if (mapperName.has_value())
+    mapperNameStr = mapperName->ToString();
+  else
+    mapperNameStr =
+        "default_" + varType.declTypeSpec->derivedTypeSpec().name().ToString();
+
+  mlir::OpBuilder::InsertPoint insPt = firOpBuilder.saveInsertionPoint();
+  firOpBuilder.setInsertionPointToStart(converter.getModuleOp().getBody());
+  auto mlirType = converter.genType(varType.declTypeSpec->derivedTypeSpec());
+  auto varVal = firOpBuilder.createTemporaryAlloc(
+      converter.getCurrentLocation(), mlirType, varName.ToString());
----------------
kiranchandramohan wrote:

> Also, I'm not sure what we stand to gain from this approach. 

In the approach that you are proposing, there is a variable and map.info that is created in the function scope with an alloca. This is not correct since the existence of a declare mapper does not cause the creation of a variable or a mapping.

Worst case, 
-> this can appear to be aliasing with other memory locations and reduce performance. 
-> increase live-ins of openmp regions when outlined

> Every other instance of where a directive has a mapClause associated with it, does so by creating the map related Ops just above it. Why are we nesting them in case of declMapperOp?

Because it is a declaration. It will only have an effect when it is used either via its usage in implicit/default mapping of the derived type for which it was created or when it is directly referred to by a map(mapper...) clause.

I think you have a reasonable point here because of the following:
```
1. If a DECLARE MAPPER directive is not specified for a type DT, a predefined mapper exists for type DT as if the type DT had appeared in the directive as follows:

!$OMP DECLARE MAPPER (DT :: var) MAP (TOFROM: var)
2. If a variable is not a scalar then it is treated as if it had appeared in a map clause with a map-type of tofrom. Which is effectively equivalent to the following and extending declare mapper for non-derived types:
!$OMP DECLARE MAPPER (T :: var) MAP (TOFROM: var)
```
We can thus argue that all mapping should also generate declarations that are ultimately all handled in a separate pass in flang/mlir. May be 1 and 2 are trivial and that is why it is not done that way.

https://github.com/llvm/llvm-project/pull/117046