[Mlir-commits] [mlir] [MLIR][LLVM] Support for indirectbr (PR #135092)

[Tobias Gysi]

================
@@ -2240,24 +2240,23 @@ static LogicalResult verifyComdat(Operation *op,
 
 static LogicalResult verifyBlockTags(LLVMFuncOp funcOp) {
   llvm::DenseSet<BlockTagAttr> blockTags;
-  BlockTagOp badBlockTagOp;
-  if (funcOp
-          .walk([&](BlockTagOp blockTagOp) {
-            if (blockTags.contains(blockTagOp.getTag())) {
-              badBlockTagOp = blockTagOp;
-              return WalkResult::interrupt();
-            }
-            blockTags.insert(blockTagOp.getTag());
-            return WalkResult::advance();
-          })
-          .wasInterrupted()) {
-    badBlockTagOp.emitError()
-        << "duplicate block tag '" << badBlockTagOp.getTag().getId()
-        << "' in the same function: ";
-    return failure();
-  }
+  LogicalResult r = success();
+  // Note that presence of `BlockTagOp`s currently can't prevent an unrecheable
+  // block to be removed by canonicalizer's region simplify pass, which needs to
+  // be dialect aware to allow extra constraints to be described.
+  funcOp.walk([&](BlockTagOp blockTagOp) {
----------------
gysit wrote:

Note that the walk can return a walk result, i.e. you could write:

`WalkResult res = funcOp.walk...`

And then further down you can return failure or success based on the walk result:

`return failure(res.wasInterrupted());`

That is a bit more idiomatic than setting the result in the lambda. 

https://github.com/llvm/llvm-project/pull/135092