remove flags_enable_parallel_graph (#51375)

paddle/fluid/framework/parallel_executor.cc

@@ -59,10 +59,6 @@ PADDLE_DEFINE_EXPORTED_string(
     "",
     "Profiler filename for PE, which generated by gperftools."
     "Only valid when compiled `WITH_PRIFILER=ON`. Empty if disable.");
-PADDLE_DEFINE_EXPORTED_bool(
-    enable_parallel_graph,
-    false,
-    "Force disable parallel graph execution mode if set false.");
 
 namespace paddle {
 namespace framework {
@@ -1276,9 +1272,7 @@ bool ParallelExecutor::EnableParallelGraphExecution(
     const ir::Graph &graph,
     const ExecutionStrategy &exec_strategy,
     const BuildStrategy &build_strategy) const {
-  if (!FLAGS_enable_parallel_graph) {
   return false;
-  }
 
   bool enable_parallel_graph = true;
 

python/paddle/fluid/executor.py

@@ -1689,21 +1689,7 @@ class Executor:
                     )
                     return False
 
-                # Unsupported case 2: parallel graph
-                if core.globals()['FLAGS_enable_parallel_graph'] in [
-                    1,
-                    '1',
-                    True,
-                    'True',
-                    'true',
-                ]:
-                    warnings.warn(
-                        "Standalone executor is not used for parallel graph",
-                        UserWarning,
-                    )
-                    return False
-
-                # Unsupported case 3: inference
+                # Unsupported case 2: inference
                 if compiled_program._is_inference:
                     warnings.warn(
                         "Standalone executor is not used for inference",
@@ -1711,7 +1697,7 @@ class Executor:
                     )
                     return False
 
-                # Unsupported case 4: async mode
+                # Unsupported case 3: async mode
                 if (
                     compiled_program._build_strategy is not None
                     and compiled_program._build_strategy.async_mode
@@ -1722,7 +1708,7 @@ class Executor:
                     )
                     return False
 
-                # Unsupported case 5: CUDA Graph
+                # Unsupported case 4: CUDA Graph
                 if (
                     compiled_program._build_strategy is not None
                     and compiled_program._build_strategy.allow_cuda_graph_capture

python/paddle/fluid/tests/unittests/CMakeLists.txt

@@ -886,12 +886,8 @@ add_subdirectory(standalone_executor)
 
 set_tests_properties(
   test_buffer_shared_memory_reuse_pass_and_fuse_optimization_op_pass
-  test_data_norm_op
-  test_dataloader_keep_order
-  test_dataloader_unkeep_order
-  test_parallel_ssa_graph_inference_feed_partial_data
-  test_buffer_shared_memory_reuse_pass
-  PROPERTIES LABELS "RUN_TYPE=DIST")
+  test_data_norm_op test_dataloader_keep_order test_dataloader_unkeep_order
+  test_buffer_shared_memory_reuse_pass PROPERTIES LABELS "RUN_TYPE=DIST")
 set_tests_properties(
   test_sync_batch_norm_op
   test_inplace_abn_op
@@ -1018,8 +1014,6 @@ set_tests_properties(test_imperative_selected_rows_to_lod_tensor
                      PROPERTIES TIMEOUT 200)
 set_tests_properties(test_index_select_op PROPERTIES TIMEOUT 120)
 set_tests_properties(test_index_add_op PROPERTIES TIMEOUT 120)
-set_tests_properties(test_parallel_ssa_graph_inference_feed_partial_data
-                     PROPERTIES TIMEOUT 120)
 set_tests_properties(test_tensordot PROPERTIES TIMEOUT 200)
 set_tests_properties(test_partial_eager_deletion_transformer PROPERTIES TIMEOUT
                                                                         120)

python/paddle/fluid/tests/unittests/asp/test_fleet_with_asp_sharding.py

@@ -40,7 +40,6 @@ class TestFleetWithASPSharding(unittest.TestCase):
         os.environ["PADDLE_TRAINERS_NUM"] = "1"
         os.environ["PADDLE_TRAINER_ID"] = "0"
 
-        os.environ['FLAGS_enable_parallel_graph'] = "0"
         os.environ['FLAGS_fraction_of_gpu_memory_to_use'] = "0.1"
         os.environ['FLAGS_sync_nccl_allreduce'] = "1"
         os.environ['FLAGS_eager_delete_tensor_gb'] = "0"

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

-# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import os
-import unittest
-
-import numpy as np
-
-os.environ['FLAGS_enable_parallel_graph'] = str(1)
-import os
-
-from parallel_executor_test_base import DeviceType, TestParallelExecutorBase
-from simple_nets import init_data, simple_fc_net
-
-import paddle.fluid.core as core
-
-
-class TestMNIST(TestParallelExecutorBase):
-    @classmethod
-    def setUpClass(cls):
-        os.environ['CPU_NUM'] = str(4)
-
-    # simple_fc
-    def check_simple_fc_convergence(self, use_device, use_reduce=False):
-        if use_device == DeviceType.CUDA and not core.is_compiled_with_cuda():
-            return
-
-        img, label = init_data()
-        self.check_network_convergence(
-            simple_fc_net,
-            feed_dict={"image": img, "label": label},
-            use_device=use_device,
-            use_reduce=use_reduce,
-        )
-
-    def test_simple_fc(self):
-        # use_device
-        self.check_simple_fc_convergence(True)
-
-    def check_simple_fc_parallel_accuracy(self, use_device):
-        if use_device and not core.is_compiled_with_cuda():
-            return
-
-        img, label = init_data()
-        single_first_loss, single_last_loss, _ = self.check_network_convergence(
-            method=simple_fc_net,
-            feed_dict={"image": img, "label": label},
-            use_device=use_device,
-            use_parallel_executor=False,
-        )
-        (
-            parallel_first_loss,
-            parallel_last_loss,
-            _,
-        ) = self.check_network_convergence(
-            method=simple_fc_net,
-            feed_dict={"image": img, "label": label},
-            use_device=use_device,
-            use_parallel_executor=True,
-        )
-
-        self.assertAlmostEqual(
-            np.mean(parallel_first_loss),
-            single_first_loss,
-            delta=1e-6,
-        )
-        self.assertAlmostEqual(
-            np.mean(parallel_last_loss), single_last_loss, delta=1e-6
-        )
-
-    def test_simple_fc_parallel_accuracy(self):
-        self.check_simple_fc_parallel_accuracy(DeviceType.CUDA)
-
-
-if __name__ == '__main__':
-    unittest.main()

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

-# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import unittest
-
-import paddle.fluid as fluid
-
-fluid.core.globals()['FLAGS_enable_parallel_graph'] = 1
-
-if __name__ == '__main__':
-    unittest.main()

tools/parallel_UT_rule.py

@@ -1492,7 +1492,6 @@ FOURTH_HIGH_PARALLEL_JOB_NEW = [
     'float16_gpu_test',
     'test_dict',
     'test_bilinear_tensor_product_op',
-    'test_parallel_executor_pg',
     'test_assert',
     'test_smooth_l1_loss_op',
     'sequence_padding_test',
@@ -1508,7 +1507,6 @@ FOURTH_HIGH_PARALLEL_JOB_NEW = [
     'test_inplace_abn_op',
     'test_parallel_executor_seresnext_base_gpu',
     'test_parallel_executor_seresnext_with_fuse_all_reduce_gpu',
-    'test_parallel_ssa_graph_inference_feed_partial_data',
     'test_parallel_executor_seresnext_with_reduce_gpu',
     'test_data_norm_op',
     'test_install_check',
@@ -2351,7 +2349,6 @@ TETRAD_PARALLEL_JOB = [
     'test_transforms',
     'test_sum_op',
     'test_scatter_op',
-    'test_parallel_executor_pg',
     'test_mix_precision_all_reduce_fuse',
     'test_tensorrt_engine_op',
     'test_zeropad2d',
@@ -2657,7 +2654,6 @@ TWO_PARALLEL_JOB = [
     'test_sigmoid_cross_entropy_with_logits_op',
     'test_regularizer_api',
     'test_lrn_op',
-    'test_parallel_ssa_graph_inference_feed_partial_data',
     'test_lod_reset_op',
     'test_install_check',
     'test_anchor_generator_op',

tools/static_mode_white_list.py

@@ -370,7 +370,6 @@ STATIC_MODE_TESTING_LIST = [
     'test_pairwise_distance',
     'test_parallel_executor_drop_scope',
     'test_parallel_executor_run_load_infer_program',
-    'test_parallel_ssa_graph_inference_feed_partial_data',
     'test_parameter',
     'test_partial_concat_op',
     'test_partial_eager_deletion_transformer',
@@ -650,7 +649,6 @@ STATIC_MODE_TESTING_LIST = [
     'test_fused_multihead_matmul_op',
     'test_ir_inplace_pass',
     'test_mix_precision_all_reduce_fuse',
-    'test_parallel_executor_pg',
     'test_rank_attention_op',
     'test_fleet_base',
     'test_fleet_graph_executor',

tools/windows/run_unittests.sh

@@ -32,7 +32,6 @@ disable_wingpu_test="^test_model$|\
 ^test_multiprocess_dataloader_iterable_dataset_dynamic$|\
 ^test_parallel_executor_feed_persistable_var$|\
 ^test_parallel_executor_inference_feed_partial_data$|\
-^test_parallel_ssa_graph_inference_feed_partial_data$|\
 ^test_py_reader_combination$|\
 ^test_py_reader_pin_memory$|\
 ^test_py_reader_push_pop$|\
@@ -145,7 +144,6 @@ disable_win_inference_test="^trt_quant_int8_yolov3_r50_test$|\
 ^test_py_reader_push_pop$|\
 ^test_parallel_executor_feed_persistable_var$|\
 ^test_parallel_executor_inference_feed_partial_data$|\
-^test_parallel_ssa_graph_inference_feed_partial_data$|\
 ^test_reader_reset$|\
 ^test_parallel_executor_seresnext_base_gpu$|\
 ^test_py_reader_pin_memory$|\