fix double grad hang bug (#34023)

Author: sneaxiy

paddle/fluid/imperative/partial_grad_engine.cc

@@ -73,6 +73,7 @@ static void GetGraphInfoBetweenTargets(
     std::unordered_map<OpBase *, size_t> *op_deps_ptr,
     std::unordered_set<VariableWrapper *> *related_grad_vars_ptr,
     const std::unordered_set<VariableWrapper *> &no_grad_var_grad) {
+  VLOG(10) << "prune graph starts";
   /**
    * Step 1. Find the candidate startup grad ops, prepared for following BFS.
    */
@@ -117,6 +118,8 @@ static void GetGraphInfoBetweenTargets(
     auto *op = op_node_pair.first;
     auto *node = op_node_pair.second;
 
+    VLOG(10) << "Visit node " << node << " , visit op " << op->Type();
+
     for (auto &output_pair : op->GetOutsMap()) {
       if (!output_pair.second.IsGrad()) {
         VLOG(10) << "WARNING: " << op->Type() << " outputs a forward var";
@@ -135,6 +138,7 @@ static void GetGraphInfoBetweenTargets(
 
     for (auto &pending_node : node->GradPendingNodes()) {
       if (visited.count(pending_node.get()) == 0) {
+        visited.insert(pending_node.get());
         for (auto &pending_op : *pending_node) {
           preceding_ops[&pending_op].insert(op);
           q.emplace(&pending_op, pending_node.get());
@@ -143,6 +147,8 @@ static void GetGraphInfoBetweenTargets(
     }
   }
 
+  VLOG(10) << "Found endpoint op ends";
+
   /**
    * Step 3. Based on the found input_target_grads, BFS the graph in reverse
    * order. `target_vars` would record all grad vars in the graph, and
@@ -246,6 +252,8 @@ static void GetGraphInfoBetweenTargets(
     }
   }
 
+  VLOG(10) << "Found startup op ends";
+
   /**
    * Step 4. Prune output_targets which is not the input of startup_ops
    */

python/paddle/fluid/tests/unittests/test_imperative_double_grad.py

@@ -15,6 +15,7 @@
 import paddle.fluid as fluid
 import paddle
 from paddle.fluid.wrapped_decorator import wrap_decorator
+from paddle.vision.models import resnet50, resnet101
 import unittest
 from unittest import TestCase
 import numpy as np
@@ -228,8 +229,6 @@ def test_example_with_gradient_accumulation_and_create_graph(self):
             x_grad_expected = (i + 2) * (2.0 / float(numel) * (
                 x_np + dx_expected *
                 (x_np > 0) * 2 / float(numel))).astype('float32')
-            print(x_grad_actual)
-            print(x_grad_expected)
             self.assertTrue(np.allclose(x_grad_actual, x_grad_expected))
 
     @dygraph_guard
@@ -369,5 +368,38 @@ def test_raise(self):
         self.assertRaises(RuntimeError, self.raise_no_grad_op)
 
 
+class TestDoubleGradResNetBase(TestCase):
+    @dygraph_guard
+    def check_resnet(self):
+        data = np.random.rand(1, 3, 224, 224).astype(np.float32)
+        data = paddle.to_tensor(data)
+        data.stop_gradient = False
+        out = self.model(data)
+        preds = paddle.argmax(out, axis=1)
+        label_onehot = paddle.nn.functional.one_hot(
+            paddle.to_tensor(preds), num_classes=out.shape[1])
+        target = paddle.sum(out * label_onehot, axis=1)
+
+        g = paddle.grad(outputs=target, inputs=out)[0]
+        g_numpy = g.numpy()
+        self.assertEqual(list(g_numpy.shape), list(out.shape))
+
+
+class TestDoubleGradResNet50(TestDoubleGradResNetBase):
+    def setUp(self):
+        self.model = resnet50(pretrained=False)
+
+    def test_main(self):
+        self.check_resnet()
+
+
+class TestDoubleGradResNet101(TestDoubleGradResNetBase):
+    def setUp(self):
+        self.model = resnet101(pretrained=False)
+
+    def test_main(self):
+        self.check_resnet()
+
+
 if __name__ == '__main__':
     unittest.main()