[llvm-branch-commits] [mlir] [mlir][Transforms] Add 1:N `matchAndRewrite` overload (PR #116470)

[Matthias Springer via llvm-branch-commits]

================
@@ -1538,10 +1607,9 @@ void ConversionPatternRewriter::replaceOp(Operation *op, ValueRange newValues) {
     impl->logger.startLine()
         << "** Replace : '" << op->getName() << "'(" << op << ")\n";
   });
-  SmallVector<ReplacementValues> newVals(newValues.size());
-  for (auto [index, val] : llvm::enumerate(newValues))
-    if (val)
-      newVals[index].push_back(val);
+  SmallVector<ValueRange> newVals;
+  for (int i = 0; i < newValues.size(); ++i)
+    newVals.push_back(newValues.slice(i, 1));
----------------
matthias-springer wrote:

This is a tricky one. I had the code that you suggested in an earlier version, but I got ASAN errors. (I think it was stack-use-after-free.) I think the problem is that `ValueRange` stores a pointer to the `Value` object, but the `Value` goes out of scope. (

https://github.com/llvm/llvm-project/pull/116470