[Mlir-commits] [mlir] [MLIR][CF] Avoid collapsing blocks which participate in cycles (PR #160783)

[llvmlistbot at llvm.org]

================
@@ -122,6 +122,18 @@ static LogicalResult collapseBranch(Block *&successor,
   Block *successorDest = successorBranch.getDest();
   if (successorDest == successor)
     return failure();
+  // Don't try to collapse branches which participate in a cycle.
+  Block *currBlock = successorDest;
+  BranchOp nextBranch = dyn_cast<BranchOp>(currBlock->getTerminator());
+  while (nextBranch) {
----------------
benwu25 wrote:

I see your argument, but weirdly enough, I have already written this testcase (`@delayed_3_cycle`), and somehow it terminates. I think that `A -> B -> C` is first canonicalized as just `C` by the `simplifyBrToBlockWithSinglePred`, so then when we collapse on `C` it somehow works out.

However, if `B` has another predecessor, maybe the loop will be broken.

https://github.com/llvm/llvm-project/pull/160783