[flang-commits] [flang] [flang] Add reduction semantics to fir.do_loop (PR #93934)

[via flang-commits]

================
@@ -2238,6 +2285,17 @@ void fir::DoLoopOp::print(mlir::OpAsmPrinter &p) {
     << getUpperBound() << " step " << getStep();
   if (getUnordered())
     p << " unordered";
+  if (hasReduceOperands()) {
+    p << " reduce(";
+    auto attrs = getReduceAttrsAttr();
+    auto operands = getReduceOperands();
+    llvm::interleaveComma(llvm::zip(attrs, operands), p, [&](auto it) {
+      p << std::get<0>(it) << " -> " << std::get<1>(it) << " : "
+        << std::get<1>(it).getType();
+    });
+    p << ')';
+    printBlockTerminators = true;
+  }
----------------
khaki3 wrote:

`iter_args` affects the result types (e.g. `iter_args(%arg1 = %7) -> (index, i32)`). While we do not expect `reduce` to be coupled with `iter_args`, that is still acceptable. Which one should we choose?

- `iter_args(%arg1 = %7) reduce(...) -> (index, i32)` 
- `reduce(...) iter_args(%arg1 = %7) -> (index, i32)` 


https://github.com/llvm/llvm-project/pull/93934