[flang-commits] [flang] [fir] Support promoting `fir.do_loop` with results to `affine.for`. (PR #137790)

[via flang-commits]

================
@@ -461,14 +475,28 @@ class AffineLoopConversion : public mlir::OpRewritePattern<fir::DoLoopOp> {
     LLVM_ATTRIBUTE_UNUSED auto loopAnalysis =
         functionAnalysis.getChildLoopAnalysis(loop);
     auto &loopOps = loop.getBody()->getOperations();
+    auto resultOp = cast<fir::ResultOp>(loop.getBody()->getTerminator());
----------------
NexMing wrote:

Yes, it can have mlutiple results. I can add tests if needed.
This optimization is indeed experimental and requires more testing to uncover issues.

https://github.com/llvm/llvm-project/pull/137790