add program cache for executor

python/paddle/fluid/executor.py

@@ -177,6 +177,7 @@ class Executor(object):
         # TODO(dzhwinter) : only use the first place
         self.executor = core.Executor(act_places[0])
         self.places = places
+        self.program_caches = dict()
 
     def aslodtensor(self, data):
         def accumulate(data):
@@ -240,8 +241,14 @@ class Executor(object):
         if scope is None:
             scope = global_scope()
 
-        program = program.clone()
-        global_block = program.global_block()
+        program_cache_key = str(feed.keys() + fetch_list)
+        program_cache = self.program_caches.get(program_cache_key, None)
+
+        if program_cache is None:
+            program_cache = program.clone()
+            self.program_caches[program_cache_key] = program_cache
+
+            global_block = program_cache.global_block()
 
             if feed_var_name in global_block.vars:
                 feed_var = global_block.var(feed_var_name)
@@ -279,7 +286,8 @@ class Executor(object):
                 else:
                     break
 
-        if not has_fetch_operators(global_block, fetch_list, fetch_var_name):
+            if not has_fetch_operators(global_block, fetch_list,
+                                       fetch_var_name):
                 for i, var in enumerate(fetch_list):
                     assert isinstance(var, Variable) or isinstance(var, str), (
                         "Wrong type for fetch_list[%s]: %s" % (i, type(var)))
@@ -289,7 +297,7 @@ class Executor(object):
                         outputs={'Out': [fetch_var]},
                         attrs={'col': i})
 
-        self.executor.run(program.desc, scope, 0, True, True)
+        self.executor.run(program_cache.desc, scope, 0, True, True)
         outs = [
             core.get_fetch_variable(scope, fetch_var_name, i)
             for i in xrange(len(fetch_list))