[flang-commits] [flang] [llvm] [mlir] [Flang][OpenMP] Support mapping of zero-sized arrays (PR #208133)

[Michael Kruse via flang-commits]

================
@@ -6273,8 +6294,53 @@ static llvm::Value *getSizeInBytes(DataLayout &dl, const mlir::Type &type,
       // size, so we do some on the fly runtime math to get the size in
       // bytes from the extent (ub - lb) * sizeInBytes. NOTE: This may need
       // some adjustment for members with more complex types.
-      return builder.CreateMul(elementCount,
-                               builder.getInt64(underlyingTypeSzInBits / 8));
+      llvm::Value *sizeCalc = builder.CreateMul(
+          elementCount, builder.getInt64(underlyingTypeSzInBits / 8),
+          "element_count");
+
+      // This is a part of a "complicated" bit of size calculation logic that is
+      // in place to handle a couple of scenarios, one specific to Fortran and
+      // the other a more general OpenMP issue. The other piece of the
+      // calculation can be found as the final size calculation within the
+      // processIndividualMap function. Ideally we would move it here, but due
+      // to the complexity of calculating the final base address of some
+      // constructs (required for a nullary check), it's left as the final step.
+      // So, in the below 2 cases, the nullary check is in processIndividualMap
+      // and the size equality check is here. The cases this modifications help
+      // cover are:
+      //
+      // 1) If an argument has a null base pointer, then the size must be set to
+      // 0
----------------
Meinersbur wrote:

[nit] awkward line break

https://github.com/llvm/llvm-project/pull/208133