[flang-commits] [flang] [flang] Lower REDUCE intrinsic with no DIM argument and rank 1 (PR #94652)

[via flang-commits]

================
@@ -5705,7 +5705,63 @@ void IntrinsicLibrary::genRandomSeed(llvm::ArrayRef<fir::ExtendedValue> args) {
 fir::ExtendedValue
 IntrinsicLibrary::genReduce(mlir::Type resultType,
                             llvm::ArrayRef<fir::ExtendedValue> args) {
-  TODO(loc, "intrinsic: reduce");
+  assert(args.size() == 6);
+
+  fir::BoxValue arrayTmp = builder.createBox(loc, args[0]);
+  mlir::Value array = fir::getBase(arrayTmp);
+  mlir::Value operation = fir::getBase(args[1]);
+  int rank = arrayTmp.rank();
+  assert(rank >= 1);
+
+  mlir::Type ty = array.getType();
+  mlir::Type arrTy = fir::dyn_cast_ptrOrBoxEleTy(ty);
+  mlir::Type eleTy = mlir::cast<fir::SequenceType>(arrTy).getEleTy();
+
+  // Handle optional mask argument
+  bool absentDim = isStaticallyAbsent(args[1]);
----------------
jeanPerier wrote:

Isn't DIM `args[2]` ?

https://github.com/llvm/llvm-project/pull/94652