[Mlir-commits] [mlir] [mlir] Fix non-termination when a block becomes unreachable during rewriting (PR #221152)

[Matthias Springer]

================
@@ -482,6 +482,15 @@ bool GreedyPatternRewriteDriver::processWorklist() {
       }
     });
 
+    // Folding an operation that uses its own result can loop forever, so
+    // skip it. Such operations are legal in unreachable code.
----------------
matthias-springer wrote:

Two potential directions come to mind:

1. `%0 = arith.addi %0, %c1 : i32` looks invalid to me. We allow it inside of a graph region, but what's the semantics of such an op? Makes me wonder if ops should explicitly have to opt in to being placed inside of a graph region...

2. Patch the canonicalization pattern / folder for `arith.addi`, so that it does not apply when operand and result are the same. But many of these patterns are defined in `ArithCanonicalization.td` and I'm not sure how difficult it is to add such checks.


https://github.com/llvm/llvm-project/pull/221152