follow comments

python/paddle/v2/fluid/distribute_transpiler.py

@@ -385,7 +385,7 @@ class DistributeTranspiler:
                 # param is already created on global program
                 param_block = None
                 for p in self.param_grad_ep_mapping[endpoint]["params"]:
-                    if same_or_split_var(p.name, opt_op.input(key)):
+                    if same_or_split_var(p.name, opt_op.input(key)[0]):
                         param_block = p
                         break
                 if not param_block:
@@ -403,7 +403,7 @@ class DistributeTranspiler:
                 continue
             # update accumulator variable shape
             param_shape = new_inputs["Param"].shape
-            var = program.global_block().vars[opt_op.input(key)]
+            var = program.global_block().vars[opt_op.input(key)[0]]
             new_shape = self._get_optimizer_input_shape(opt_op.type, key,
                                                         var.shape, param_shape)
             tmpvar = program.global_block().create_var(
@@ -440,20 +440,18 @@ class DistributeTranspiler:
             else:
                 varlist = [var]
             for var in varlist:
+                # TODO(typhoonzero): will remove below line later.
                 program.global_block().create_var(
                     name=var.name,
                     persistable=var.persistable,
                     dtype=var.dtype,
                     shape=var.shape)
-                try:
+                if not pserver_program.global_block().vars.has_key(var.name):
                     pserver_program.global_block().create_var(
                         name=var.name,
                         persistable=var.persistable,
                         dtype=var.dtype,
                         shape=var.shape)
-                except ValueError:
-                    # create var if not created yet.
-                    pass
 
         outputs = self._get_output_map_from_op(self.program.global_block().vars,
                                                opt_op)