[mlir] Do not merge blocks during canonicalization by default (#95057)
This is a heavy process, and it can trigger a massive explosion in
adding block arguments. While potentially reducing the code size, the
resulting merged blocks with arguments are hiding some of the def-use
chain and can even hinder some further analyses/optimizations: a merge
block does not have it's own path-sensitive context, instead the context
is merged from all the predecessors.
Previous behavior can be restored by passing:
{test-convergence region-simplify=aggressive}
to the canonicalize pass.diff --git a/mlir/test/Transforms/test-canonicalize.mlir b/mlir/test/Transforms/test-canonicalize.mlir
index 4f0095e..0fc822b 100644
--- a/mlir/test/Transforms/test-canonicalize.mlir
+++ b/mlir/test/Transforms/test-canonicalize.mlir
@@ -1,5 +1,5 @@
// RUN: mlir-opt %s -pass-pipeline='builtin.module(func.func(canonicalize))' | FileCheck %s
-// RUN: mlir-opt %s -pass-pipeline='builtin.module(func.func(canonicalize{region-simplify=false}))' | FileCheck %s --check-prefixes=CHECK,NO-RS
+// RUN: mlir-opt %s -pass-pipeline='builtin.module(func.func(canonicalize{region-simplify=disabled}))' | FileCheck %s --check-prefixes=CHECK,NO-RS
// CHECK-LABEL: func @remove_op_with_inner_ops_pattern
func.func @remove_op_with_inner_ops_pattern() {
