Minor speedup for model lowering: Skip redundant run_decompositions when no ops match decomp table (#18496)

Summary:
Pull Request resolved: #18496

Adds an early-exit check to _gen_edge_manager_for_partitioners: before
calling program.run_decompositions(table), scan the graph for ops that
appear in the decomposition table. If none are found, skip the call
entirely.

Each run_decompositions call performs a full re-export of the program
via make_fx(), re-tracing every node through FakeTensor dispatch.
On the EDGE_DO_NOT_DECOMP path this function is called up to 3 times;
the early-exit eliminates at least one redundant call where the previous
pass already decomposed all matching ops.

The check recursively walks control flow submodules (cond/map/scan) to
avoid incorrectly skipping when decomposable ops are nested.

## Benchmark

Model: small CNN feature extractor (~50K params, 9 conv layers with
LayerNorm, targeting Ethos-U55 via the ARM/TOSA lowering pipeline).
Graph: ~1200 nodes.

  lower() before:  82 s
  lower() after:   71 s
  Delta:          -11 s  (-13 %)

Differential Revision: D96489903

Author: Andrew Pullin

exir/program/_program.py

@@ -1100,7 +1100,33 @@ def _can_skip_using_EDGE_DO_NOT_DECOMP(
     return check_op_support is None
 
 
-def _gen_edge_manager_for_partitioners(
+def _has_decomposable_ops(
+    program: "ExportedProgram",
+    decomp_table: dict,
+) -> bool:
+    """Check if any ops in the graph match the decomposition table.
+
+    Returns True if the graph contains at least one op that appears in the
+    decomposition table, meaning run_decompositions would actually decompose
+    something.  Returns True for empty tables (functionalization-only path)
+    since we can't cheaply determine if the graph needs functionalization.
+    """
+    if not decomp_table:
+        return True  # empty table = functionalize, can't skip cheaply
+
+    def _graph_has_match(gm: torch.fx.GraphModule) -> bool:
+        for node in gm.graph.nodes:
+            if node.op == "call_function" and node.target in decomp_table:
+                return True
+        for _, submod, _ in get_control_flow_submodules(gm):
+            if _graph_has_match(submod):
+                return True
+        return False
+
+    return _graph_has_match(program.graph_module)
+
+
+def _gen_edge_manager_for_partitioners(  # noqa: C901
     partitioner: Dict[str, List[Partitioner]],
     aten_programs: Dict[str, ExportedProgram],
     config: EdgeCompileConfig,
@@ -1135,7 +1161,8 @@ def _gen_edge_manager_for_partitioners(
                 table = _default_decomposition_table()
                 for op in config.preserve_ops:
                     table.pop(op, None)
-                program = program.run_decompositions(table)
+                if _has_decomposable_ops(program, table):
+                    program = program.run_decompositions(table)
 
             # Process each partitioner individually using their specific requirements
             for curr_partitioner in partitioners_for_program:
@@ -1155,7 +1182,8 @@ def _gen_edge_manager_for_partitioners(
                         if table.pop(op, None) is not None:
                             ops_needing_preservation.append(op)
 
-                    program = program.run_decompositions(table)
+                    if _has_decomposable_ops(program, table):
+                        program = program.run_decompositions(table)
                     final_ops_to_preserve.update(ops_needing_preservation)
                 else:
                     # EDGE_DO_NOT_DECOMP path for the partitioner
@@ -1169,7 +1197,8 @@ def _gen_edge_manager_for_partitioners(
                         table.pop(op, None)
 
                     # First pass of decompositions with this partitioner's preserved ops
-                    program = program.run_decompositions(table)
+                    if _has_decomposable_ops(program, table):
+                        program = program.run_decompositions(table)
 
                     # Filter ops using EDGE_DO_NOT_DECOMP
                     temp_partitioner_dict = {name: [curr_partitioner]}
@@ -1182,7 +1211,9 @@ def _gen_edge_manager_for_partitioners(
                     final_ops_to_preserve.update(preserved_ops)
 
                     # Second pass of decompositions with this partitioner's preserved ops after filtering
-                    program = program.run_decompositions(_default_decomposition_table())
+                    full_table = _default_decomposition_table()
+                    if _has_decomposable_ops(program, full_table):
+                        program = program.run_decompositions(full_table)
 
                     # Restore ops from edge_no_decomp_namespace to aten ops
                     _restore_transformed_ops_to_aten_ops(program)