[Mlir-commits] [mlir] [MLIR][OpenMP] Fix MLIR->LLVM value matching in privatization logic (PR #103718)

[Leandro Lupori]

================
@@ -1424,35 +1424,107 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder,
     }
   };
 
-  SmallVector<omp::PrivateClauseOp> privatizerClones;
-  SmallVector<llvm::Value *> privateVariables;
+  SmallVector<omp::PrivateClauseOp> mlirPrivatizerClones;
+  SmallVector<llvm::Value *> llvmPrivateVars;
 
   // TODO: Perform appropriate actions according to the data-sharing
   // attribute (shared, private, firstprivate, ...) of variables.
   // Currently shared and private are supported.
   auto privCB = [&](InsertPointTy allocaIP, InsertPointTy codeGenIP,
-                    llvm::Value &, llvm::Value &vPtr,
-                    llvm::Value *&replacementValue) -> InsertPointTy {
-    replacementValue = &vPtr;
+                    llvm::Value &, llvm::Value &llvmOmpRegionInputPtr,
----------------
luporl wrote:

This is a question only: in this context, should I understand `llvmOmpRegionInputPtr` as an input, that is an LLVM pointer, in the OpenMP region being translated, that must be replaced with `llvmReplacementValue`?
But if so, won't it point to the result of a load sometimes, which is not a pointer? Or should I interpret it as "argument that points to the input value that must be replaced"?

https://github.com/llvm/llvm-project/pull/103718