Run product state functions inplace to avoid copies where possible (#…

…6396)

* Run product state merges inplace to avoid copies

* rename to merged_state

* comment

cirq-core/cirq/sim/simulation_product_state.py

@@ -63,12 +63,19 @@ def split_untangled_states(self) -> bool:
         return self._split_untangled_states
 
     def create_merged_state(self) -> TSimulationState:
+        merged_state = self.sim_states[None]
         if not self.split_untangled_states:
-            return self.sim_states[None]
-        final_args = self.sim_states[None]
-        for args in set([self.sim_states[k] for k in self.sim_states.keys() if k is not None]):
-            final_args = final_args.kronecker_product(args)
-        return final_args.transpose_to_qubit_order(self.qubits)
+            return merged_state
+        extra_states = set([self.sim_states[k] for k in self.sim_states.keys() if k is not None])
+        if not extra_states:
+            return merged_state
+
+        # This comes from a member variable so we need to copy it if we're going to modify inplace
+        # before returning. We're not running a step currently, so no need to copy buffers.
+        merged_state = merged_state.copy(deep_copy_buffers=False)
+        for state in extra_states:
+            merged_state.kronecker_product(state, inplace=True)
+        return merged_state.transpose_to_qubit_order(self.qubits, inplace=True)
 
     def _act_on_fallback_(
         self, action: Any, qubits: Sequence['cirq.Qid'], allow_decompose: bool = True
@@ -106,7 +113,7 @@ def _act_on_fallback_(
             if op_args_opt is None:
                 op_args_opt = self.sim_states[q]
             elif q not in op_args_opt.qubits:
-                op_args_opt = op_args_opt.kronecker_product(self.sim_states[q])
+                op_args_opt.kronecker_product(self.sim_states[q], inplace=True)
         op_args = op_args_opt or self.sim_states[None]
 
         # (Backfill the args map with the new value)
@@ -123,7 +130,7 @@ def _act_on_fallback_(
         ):
             for q in qubits:
                 if op_args.allows_factoring and len(op_args.qubits) > 1:
-                    q_args, op_args = op_args.factor((q,), validate=False)
+                    q_args, _ = op_args.factor((q,), validate=False, inplace=True)
                     self._sim_states[q] = q_args
 
             # (Backfill the args map with the new value)