【pir】update If_grad and while_grad op' stop gradient by yield_op (#70545)

xiaoguoguo626807 · web-flow · commit 27ca016e1ac0 · 2025-01-02T10:01:43.000+08:00
* generate_vjp optional output add check

* add if while grad op stopgradient update func

* modify bug

* fix while op update

* delete print
diff --git a/python/paddle/autograd/backward_utils.py b/python/paddle/autograd/backward_utils.py
@@ -691,3 +691,67 @@ def get_split_op(value):
 @lru_cache
 def warning_once(message: str):
     logging.warning(message)
+
+
+def update_if_output_stopgradient(if_op, true_yield_op, false_yield_op):
+    """
+    Update if_op's stop_gradient based on true_yield_op and false_yield_op.
+
+    Args:
+    true_yield_op: true block of if_op's last op.
+    false_yield_op: false block of if_op's last op.
+    if_op: update it's op_results()'s stop_gradient.
+    """
+    if (
+        true_yield_op.name() != 'cf.yield'
+        or false_yield_op.name() != 'cf.yield'
+    ):
+        raise ValueError("param isnot yield op")
+
+    # Check if operands_source sizes match
+    if len(true_yield_op.operands_source()) != len(
+        false_yield_op.operands_source()
+    ):
+        raise ValueError("Mismatched yield operands_source sizes")
+
+    # Check if op_results size matches operands_source
+    if len(if_op.results()) != len(true_yield_op.operands_source()):
+        raise ValueError(
+            "Mismatched if op_results size with yield operands_source"
+        )
+
+    # Update if_op's stop_gradient
+    for i in range(len(true_yield_op.operands_source())):
+        stop_grad1 = true_yield_op.operand_source(i).stop_gradient
+        stop_grad2 = false_yield_op.operand_source(i).stop_gradient
+
+        # Set to False if either stop_gradient is False
+        if not stop_grad1 or not stop_grad2:
+            if_op.result(i).stop_gradient = False
+
+
+def update_while_output_stopgradient(while_op, yield_op):
+    """
+    Update while_op's stop_gradient based on yield_op.
+
+    Args:
+    yield_op: The yield operation associated with the while loop.
+    while_op: The while operation whose op_results()'s stop_gradient needs to be updated.
+    """
+    # Check if yield_op is indeed a yield operation
+    if yield_op.name() != 'cf.yield':
+        raise ValueError("yield_op is not a yield operation")
+
+    # Check if operands_source size of yield_op matches op_results size of while_op
+    if len(while_op.results()) + 1 != len(yield_op.operands_source()):
+        raise ValueError(
+            f"Mismatched while op_results  size %d with yield operands_source %d. {len(while_op.results()) + 1, len(yield_op.operands_source())}"
+        )
+
+    # Update while_op's stop_gradient
+    for i in range(1, len(yield_op.operands_source())):
+        stop_grad = yield_op.operand_source(i).stop_gradient
+
+        # Set to False if stop_gradient is False
+        if not stop_grad:
+            while_op.result(i - 1).stop_gradient = False
diff --git a/python/paddle/autograd/ir_backward.py b/python/paddle/autograd/ir_backward.py
@@ -45,7 +45,9 @@
     return_map_value,
     return_map_value_list,
     some_in_set,
+    update_if_output_stopgradient,
     update_no_grad_set_by_stopgradient,
+    update_while_output_stopgradient,
     warning_once,
     while_prune_check,
 )
@@ -791,6 +793,11 @@ def append_yield(
                                     input_tuple[1]
                                 )
 
+                        update_if_output_stopgradient(
+                            grad_op,
+                            grad_op.as_if_op().true_block().ops[-1],
+                            grad_op.as_if_op().false_block().ops[-1],
+                        )
                         for input_tuple in inputs_used_by_other_op:
                             state.value_to_valuegrad[input_tuple[0]] = []
                         # update input_grad map
@@ -870,6 +877,10 @@ def append_yield(
                             sub_bwd_value_to_block_argument_map,
                             sub_control_flow_value_to_copyvalue_map,
                         )
+
+                        update_while_output_stopgradient(
+                            grad_op, while_grad_block.ops[-1]
+                        )
                         # update input_grad map
                         update_input_grad_map(op, input_grads, origin_inputs)
                     elif op.name() == "pd_op.pylayer":
