update

paddle/pybind/protobuf.cc

@@ -157,6 +157,7 @@ void BindBlockDesc(py::module &m) {
       .def_property_readonly("parent", &BlockDescBind::Parent)
       .def("append_op", &BlockDescBind::AppendOp,
            py::return_value_policy::reference)
+      .def("append_allocated_op", &BlockDescBind::AppendAllocatedOp)
       .def("prepend_op", &BlockDescBind::PrependOp,
            py::return_value_policy::reference)
       .def("var",

python/paddle/v2/fluid/backward.py

@@ -15,7 +15,11 @@ def rename_arg(op_desc_list, old_name, new_name, begin_idx=None, end_idx=None):
         op_desc_list[i].rename_output(old_name, new_name)
 
 
-def backward_impl(block, target_block, no_grad_set, callback=None):
+def backward_impl(block,
+                  target_block,
+                  no_grad_set,
+                  grad_info_map,
+                  callback=None):
     grad_op_descs = []
     grad_to_var = {}
     program = block.program
@@ -25,7 +29,8 @@ def backward_impl(block, target_block, no_grad_set, callback=None):
             sub_block_idx = each_op.block_attr("sub_block")
             sub_block = program.block(sub_block_idx)
             grad_sub_block = program.create_block(parent_idx=sub_block_idx)
-            backward_impl(sub_block, grad_sub_block, no_grad_set, callback)
+            backward_impl(sub_block, grad_sub_block, no_grad_set, grad_info_map,
+                          callback)
             grad_sub_block_list.append(grad_sub_block)
         grad_op_desc = core.get_grad_op_desc(each_op.desc,
                                              no_grad_set[block.idx],
@@ -71,17 +76,28 @@ def backward_impl(block, target_block, no_grad_set, callback=None):
             pending_sum_ops.append((core.OpDesc(
                 type="sum_op", inputs=inputs, outputs=var_name, attrs={}),
                                     len(grad_op_descs)))
+    # TODO: remove op in no grad set
+
     # 根据append的顺序可以看出pending_sum_ops一定是根据sum_op的插入位置排序的
     for p in reversed(pending_sum_ops):
         grad_op_descs.insert(p[1], p[0])
-    # create new gradient variables in the target block
+    # create new gradient variables in the target block desc
     for op_desc in grad_op_descs:
         for grad_var_name in op_desc.output_arg_names():
-            if target_block.has_var(
+            if target_block.desc.has_var(
                     grad_var_name) or grad_var_name == core.get_empty_var_name(
                     ):
                 continue
-            target_block.var(grad_var_name)
+            target_block.desc.var(grad_var_name)
+            if not grad_to_var.has_key(grad_var_name):
+                continue
+            grad_info_map[grad_to_var[grad_var_name]] = (grad_var_name,
+                                                         target_block)
+    # insert backward operators to target_block
+    for op_desc in grad_op_descs:
+        target_block.desc.append_allocated_op(op_desc)
+
+    target_block.sync_with_cpp()
 
 
 def append_backward_ops(loss, parameter_list=None, no_grad_set=None):