for feed_trainer api

f07724da · xiexionghang · e3c436c5 · f07724da · f07724da · f07724da
27 changed file
--- a/BCLOUD
+++ b/BCLOUD
 WORKROOT('../../../')
 COMPILER('gcc482')
+CPPFLAGS('-D_GNU_SOURCE -DNDEBUG')
+GLOBAL_CFLAGS_STR = '-g -O3 -pipe '
+CFLAGS(GLOBAL_CFLAGS_STR)
+GLOBAL_CXXFLAGS_STR = GLOBAL_CFLAGS_STR + ' -std=c++11 '
+CXXFLAGS(GLOBAL_CXXFLAGS_STR)
 INCPATHS('./')
 INCPATHS('$OUT/../')
 INCPATHS('../../third-party')
 INCPATHS('../../third-party/eigen')
 INCPATHS('$OUT_ROOT/baidu/third-party/python/output/include/python2.7')
-LDFLAGS('-lpthread -lcrypto -lrt -ldl -lssl -lz -lrt')
+LDFLAGS('-lpthread -lcrypto -lrt -ldl -lssl -lz -lrt -lgomp')
 CONFIGS('baidu/third-party/any@15595d8324be9e8a9a80d9ae442fdd12bd66df5d@git_branch')
 CONFIGS('baidu/third-party/boost@v1.41.0@git_branch')
 CONFIGS('baidu/third-party/c-ares@v1.13.0@git_branch')
@@ -29,6 +34,7 @@ CONFIGS('baidu/third-party/dlpack@v0.2@git_branch')
 CONFIGS('baidu/third-party/gzstream@master@git_branch')
 CONFIGS('baidu/third-party/pybind11@v2.2.4@git_branch')
 CONFIGS('baidu/third-party/python@gcc482output@git_branch')
+CONFIGS('baidu/third-party/yaml-cpp@yaml-cpp_0-6-2-0_GEN_PD_BL@git_tag')
 HEADERS('paddle/fluid/memory/*.h', '$INC/paddle/fluid/memory/')
 HEADERS('paddle/fluid/memory/detail/*.h', '$INC/paddle/fluid/memory/detail/')
@@ -51,6 +57,9 @@ HEADERS('paddle/fluid/inference/api/*.h', '$INC/paddle/fluid/inference/api/')
 HEADERS(GLOB_GEN_SRCS('paddle/fluid/framework/*pb.h'), '$INC/paddle/fluid/framework')
 HEADERS(GLOB_GEN_SRCS('paddle/fluid/platform/*pb.h'), '$INC/paddle/fluid/platform')
 PROTOC('../../third-party/protobuf/bin/protoc')
 paddle_fluid_avx_mklml_src = "paddle/fluid/memory/detail/memory_block.cc paddle/fluid/memory/detail/memory_block_desc.cc paddle/fluid/memory/detail/meta_cache.cc paddle/fluid/memory/detail/system_allocator.cc paddle/fluid/memory/detail/buddy_allocator.cc paddle/fluid/memory/allocation/allocator.cc paddle/fluid/memory/allocation/cpu_allocator.cc paddle/fluid/memory/allocation/locked_allocator.cc paddle/fluid/memory/allocation/buffered_allocator.cc paddle/fluid/memory/allocation/best_fit_allocator.cc paddle/fluid/memory/allocation/naive_best_fit_allocator.cc paddle/fluid/memory/allocation/retry_allocator.cc paddle/fluid/memory/allocation/aligned_allocator.cc paddle/fluid/memory/allocation/allocator_strategy.cc paddle/fluid/memory/allocation/allocator_facade.cc paddle/fluid/memory/allocation/auto_growth_best_fit_allocator.cc paddle/fluid/memory/malloc.cc paddle/fluid/memory/memcpy.cc paddle/fluid/platform/profiler.proto paddle/fluid/platform/enforce.cc paddle/fluid/platform/cpu_info.cc paddle/fluid/platform/place.cc paddle/fluid/platform/dynload/dynamic_loader.cc paddle/fluid/platform/dynload/warpctc.cc paddle/fluid/platform/dynload/mklml.cc paddle/fluid/platform/cpu_helper.cc paddle/fluid/platform/temporary_allocator.cc paddle/fluid/platform/device_context.cc paddle/fluid/platform/init.cc paddle/fluid/platform/timer.cc paddle/fluid/platform/lodtensor_printer.cc paddle/fluid/platform/device_tracer.cc paddle/fluid/platform/profiler.cc paddle/fluid/platform/device_memory_aligment.cc paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_optimizer_op_pass.cc paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_adam_op_pass.cc paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_sgd_op_pass.cc paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_momentum_op_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/op_graph_view.cc paddle/fluid/framework/ir/memory_optimize_pass/while_op_eager_deletion_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/recurrent_op_eager_deletion_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/reference_count_pass_helper.cc paddle/fluid/framework/ir/memory_optimize_pass/reference_count_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/memory_optimize_helper.cc paddle/fluid/framework/ir/memory_optimize_pass/memory_optimize_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/inplace_op_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/eager_deletion_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/record_skip_memory_opt_vars_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/memory_reuse_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/buffer_shared_inplace_op_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/modify_op_lock_and_record_event_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_print_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_check_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/sequential_execution_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/fuse_all_reduce_op_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/all_reduce_deps_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/backward_optimizer_op_deps_pass.cc paddle/fluid/framework/ir/node.cc paddle/fluid/framework/ir/graph.cc paddle/fluid/framework/ir/graph_helper.cc paddle/fluid/framework/ir/pass.cc paddle/fluid/framework/ir/graph_traits.cc paddle/fluid/framework/ir/graph_pattern_detector.cc paddle/fluid/framework/ir/fuse_pass_base.cc paddle/fluid/framework/ir/coalesce_grad_tensor_pass.cc paddle/fluid/framework/ir/graph_to_program_pass.cc paddle/fluid/framework/ir/graph_viz_pass.cc paddle/fluid/framework/ir/lock_free_optimize_pass.cc paddle/fluid/framework/ir/fc_fuse_pass.cc paddle/fluid/framework/ir/attention_lstm_fuse_pass.cc paddle/fluid/framework/ir/infer_clean_graph_pass.cc paddle/fluid/framework/ir/fc_lstm_fuse_pass.cc paddle/fluid/framework/ir/embedding_fc_lstm_fuse_pass.cc paddle/fluid/framework/ir/fc_gru_fuse_pass.cc paddle/fluid/framework/ir/seq_concat_fc_fuse_pass.cc paddle/fluid/framework/ir/multi_batch_merge_pass.cc paddle/fluid/framework/ir/conv_bn_fuse_pass.cc paddle/fluid/framework/ir/seqconv_eltadd_relu_fuse_pass.cc paddle/fluid/framework/ir/seqpool_concat_fuse_pass.cc paddle/fluid/framework/ir/repeated_fc_relu_fuse_pass.cc paddle/fluid/framework/ir/squared_mat_sub_fuse_pass.cc paddle/fluid/framework/ir/is_test_pass.cc paddle/fluid/framework/ir/conv_elementwise_add_act_fuse_pass.cc paddle/fluid/framework/ir/conv_elementwise_add2_act_fuse_pass.cc paddle/fluid/framework/ir/conv_elementwise_add_fuse_pass.cc paddle/fluid/framework/ir/conv_affine_channel_fuse_pass.cc paddle/fluid/framework/ir/transpose_flatten_concat_fuse_pass.cc paddle/fluid/framework/ir/identity_scale_op_clean_pass.cc paddle/fluid/framework/ir/sync_batch_norm_pass.cc paddle/fluid/framework/ir/runtime_context_cache_pass.cc paddle/fluid/framework/ir/quant_conv2d_dequant_fuse_pass.cc paddle/fluid/framework/ir/fillconstant_elementwisemul_fuse.cc paddle/fluid/framework/ir/shuffle_channel_detect_pass.cc paddle/fluid/framework/ir/delete_quant_dequant_op_pass.cc paddle/fluid/framework/ir/fuse_elewise_add_act_pass.cc paddle/fluid/framework/ir/fuse_relu_depthwise_conv_pass.cc paddle/fluid/framework/ir/pass_builder.cc paddle/fluid/framework/details/var_handle.cc paddle/fluid/framework/details/op_handle_base.cc paddle/fluid/framework/details/scale_loss_grad_op_handle.cc paddle/fluid/framework/details/fetch_op_handle.cc paddle/fluid/framework/details/computation_op_handle.cc paddle/fluid/framework/details/rpc_op_handle.cc paddle/fluid/framework/details/fetch_barrier_op_handle.cc paddle/fluid/framework/details/multi_devices_helper.cc paddle/fluid/framework/details/variable_visitor.cc paddle/fluid/framework/details/all_reduce_op_handle.cc paddle/fluid/framework/details/fused_all_reduce_op_handle.cc paddle/fluid/framework/details/reduce_op_handle.cc paddle/fluid/framework/details/broadcast_op_handle.cc paddle/fluid/framework/details/fused_broadcast_op_handle.cc paddle/fluid/framework/details/gather_op_handle.cc paddle/fluid/framework/details/eager_deletion_op_handle.cc paddle/fluid/framework/details/share_tensor_buffer_op_handle.cc paddle/fluid/framework/details/ssa_graph_executor.cc paddle/fluid/framework/details/threaded_ssa_graph_executor.cc paddle/fluid/framework/details/parallel_ssa_graph_executor.cc paddle/fluid/framework/details/async_ssa_graph_executor.cc paddle/fluid/framework/details/scope_buffered_ssa_graph_executor.cc paddle/fluid/framework/details/fast_threaded_ssa_graph_executor.cc paddle/fluid/framework/details/build_strategy.cc paddle/fluid/framework/fleet/fleet_wrapper.cc paddle/fluid/framework/fleet/nccl_wrapper.cc paddle/fluid/framework/io/fs.cc paddle/fluid/framework/io/shell.cc paddle/fluid/framework/framework.proto paddle/fluid/framework/data_feed.proto paddle/fluid/framework/trainer_desc.proto paddle/fluid/framework/ddim.cc paddle/fluid/framework/data_type.cc paddle/fluid/framework/tensor.cc paddle/fluid/framework/tensor_util.cc paddle/fluid/framework/lod_tensor.cc paddle/fluid/framework/garbage_collector.cc paddle/fluid/framework/reader.cc paddle/fluid/framework/threadpool.cc paddle/fluid/framework/var_type_traits.cc paddle/fluid/framework/scope.cc paddle/fluid/framework/scope_pool.cc paddle/fluid/framework/data_device_transform.cc paddle/fluid/framework/data_type_transform.cc paddle/fluid/framework/data_layout_transform.cc paddle/fluid/framework/data_transform.cc paddle/fluid/framework/attribute.cc paddle/fluid/framework/op_proto_maker.cc paddle/fluid/framework/op_info.cc paddle/fluid/framework/shape_inference.cc paddle/fluid/framework/transfer_scope_cache.cc paddle/fluid/framework/op_kernel_type.cc paddle/fluid/framework/operator.cc paddle/fluid/framework/version.cc paddle/fluid/framework/var_desc.cc paddle/fluid/framework/op_desc.cc paddle/fluid/framework/block_desc.cc paddle/fluid/framework/program_desc.cc paddle/fluid/framework/op_registry.cc paddle/fluid/framework/lod_rank_table.cc paddle/fluid/framework/feed_fetch_method.cc paddle/fluid/framework/variable_helper.cc paddle/fluid/framework/naive_executor.cc paddle/fluid/framework/executor_gc_helper.cc paddle/fluid/framework/executor.cc paddle/fluid/framework/multi_trainer.cc paddle/fluid/framework/pipeline_trainer.cc paddle/fluid/framework/dataset_factory.cc paddle/fluid/framework/dist_multi_trainer.cc paddle/fluid/framework/trainer_factory.cc paddle/fluid/framework/trainer.cc paddle/fluid/framework/data_feed_factory.cc paddle/fluid/framework/data_feed.cc paddle/fluid/framework/device_worker.cc paddle/fluid/framework/hogwild_worker.cc paddle/fluid/framework/downpour_worker.cc paddle/fluid/framework/pull_dense_worker.cc paddle/fluid/framework/section_worker.cc paddle/fluid/framework/device_worker_factory.cc paddle/fluid/framework/data_set.cc paddle/fluid/framework/parallel_executor.cc paddle/fluid/framework/prune.cc paddle/fluid/framework/selected_rows.cc paddle/fluid/framework/dlpack_tensor.cc paddle/fluid/imperative/flags.cc paddle/fluid/operators/math/detail/avx_functions.cc paddle/fluid/operators/math/concat_and_split.cc paddle/fluid/operators/math/context_project.cc paddle/fluid/operators/math/cross_entropy.cc paddle/fluid/operators/math/cos_sim_functor.cc paddle/fluid/operators/math/im2col.cc paddle/fluid/operators/math/sample_prob.cc paddle/fluid/operators/math/sampler.cc paddle/fluid/operators/math/gru_compute.cc paddle/fluid/operators/math/lstm_compute.cc paddle/fluid/operators/math/blas.cc paddle/fluid/operators/math/math_function.cc paddle/fluid/operators/math/maxouting.cc paddle/fluid/operators/math/pooling.cc paddle/fluid/operators/math/selected_rows_functor.cc paddle/fluid/operators/math/sequence2batch.cc paddle/fluid/operators/math/sequence_padding.cc paddle/fluid/operators/math/sequence_pooling.cc paddle/fluid/operators/math/sequence_scale.cc paddle/fluid/operators/math/softmax.cc paddle/fluid/operators/math/beam_search.cc paddle/fluid/operators/math/matrix_bit_code.cc paddle/fluid/operators/math/unpooling.cc paddle/fluid/operators/math/vol2col.cc paddle/fluid/operators/math/tree2col.cc paddle/fluid/operators/controlflow/feed_op.cc paddle/fluid/operators/controlflow/logical_op.cc paddle/fluid/operators/controlflow/while_op.cc paddle/fluid/operators/controlflow/get_places_op.cc paddle/fluid/operators/controlflow/fetch_op.cc paddle/fluid/operators/controlflow/compare_op.cc paddle/fluid/operators/controlflow/conditional_block_infer_op.cc paddle/fluid/operators/controlflow/conditional_block_op.cc paddle/fluid/operators/controlflow/tensor_array_read_write_op.cc paddle/fluid/operators/controlflow/op_variant.cc paddle/fluid/operators/controlflow/recurrent_op_helper.cc paddle/fluid/operators/controlflow/while_op_helper.cc paddle/fluid/operators/detection/bipartite_match_op.cc paddle/fluid/operators/detection/box_coder_op.cc paddle/fluid/operators/detection/iou_similarity_op.cc paddle/fluid/operators/detection/mine_hard_examples_op.cc paddle/fluid/operators/detection/multiclass_nms_op.cc paddle/fluid/operators/detection/poly_util.cc paddle/fluid/operators/detection/gpc.cc paddle/fluid/operators/detection/prior_box_op.cc paddle/fluid/operators/detection/density_prior_box_op.cc paddle/fluid/operators/detection/anchor_generator_op.cc paddle/fluid/operators/detection/target_assign_op.cc paddle/fluid/operators/detection/polygon_box_transform_op.cc paddle/fluid/operators/detection/rpn_target_assign_op.cc paddle/fluid/operators/detection/generate_proposal_labels_op.cc paddle/fluid/operators/detection/box_clip_op.cc paddle/fluid/operators/detection/yolov3_loss_op.cc paddle/fluid/operators/detection/yolo_box_op.cc paddle/fluid/operators/detection/box_decoder_and_assign_op.cc paddle/fluid/operators/detection/sigmoid_focal_loss_op.cc paddle/fluid/operators/detection/retinanet_detection_output_op.cc paddle/fluid/operators/detection/generate_proposals_op.cc paddle/fluid/operators/detection/distribute_fpn_proposals_op.cc paddle/fluid/operators/detection/collect_fpn_proposals_op.cc paddle/fluid/operators/detection/roi_perspective_transform_op.cc paddle/fluid/operators/detection/mask_util.cc paddle/fluid/operators/detection/generate_mask_labels_op.cc paddle/fluid/operators/elementwise/elementwise_mod_op.cc paddle/fluid/operators/elementwise/elementwise_floordiv_op.cc paddle/fluid/operators/elementwise/elementwise_max_op.cc paddle/fluid/operators/elementwise/elementwise_pow_op.cc paddle/fluid/operators/elementwise/elementwise_sub_op.cc paddle/fluid/operators/elementwise/elementwise_add_op.cc paddle/fluid/operators/elementwise/elementwise_min_op.cc paddle/fluid/operators/elementwise/elementwise_div_op.cc paddle/fluid/operators/elementwise/elementwise_mul_op.cc paddle/fluid/operators/fused/fusion_squared_mat_sub_op.cc paddle/fluid/operators/fused/fusion_seqpool_concat_op.cc paddle/fluid/operators/fused/fused_embedding_fc_lstm_op.cc paddle/fluid/operators/fused/fusion_seqexpand_concat_fc_op.cc paddle/fluid/operators/fused/fused_embedding_seq_pool_op.cc paddle/fluid/operators/fused/fusion_seqconv_eltadd_relu_op.cc paddle/fluid/operators/fused/fusion_gru_op.cc paddle/fluid/operators/fused/fusion_repeated_fc_relu_op.cc paddle/fluid/operators/fused/fusion_lstm_op.cc paddle/fluid/operators/fused/fused_elemwise_activation_op.cc paddle/fluid/operators/metrics/accuracy_op.cc paddle/fluid/operators/metrics/precision_recall_op.cc paddle/fluid/operators/metrics/auc_op.cc paddle/fluid/operators/optimizers/adamax_op.cc paddle/fluid/operators/optimizers/sgd_op.cc paddle/fluid/operators/optimizers/lars_momentum_op.cc paddle/fluid/operators/optimizers/adagrad_op.cc paddle/fluid/operators/optimizers/ftrl_op.cc paddle/fluid/operators/optimizers/momentum_op.cc paddle/fluid/operators/optimizers/adadelta_op.cc paddle/fluid/operators/optimizers/rmsprop_op.cc paddle/fluid/operators/optimizers/lamb_op.cc paddle/fluid/operators/optimizers/proximal_gd_op.cc paddle/fluid/operators/optimizers/proximal_adagrad_op.cc paddle/fluid/operators/optimizers/adam_op.cc paddle/fluid/operators/optimizers/decayed_adagrad_op.cc paddle/fluid/operators/reduce_ops/reduce_all_op.cc paddle/fluid/operators/reduce_ops/reduce_min_op.cc paddle/fluid/operators/reduce_ops/reduce_sum_op.cc paddle/fluid/operators/reduce_ops/reduce_any_op.cc paddle/fluid/operators/reduce_ops/reduce_max_op.cc paddle/fluid/operators/reduce_ops/reduce_mean_op.cc paddle/fluid/operators/reduce_ops/reduce_prod_op.cc paddle/fluid/operators/sequence_ops/sequence_erase_op.cc paddle/fluid/operators/sequence_ops/sequence_unpad_op.cc paddle/fluid/operators/sequence_ops/sequence_mask_op.cc paddle/fluid/operators/sequence_ops/sequence_expand_op.cc paddle/fluid/operators/sequence_ops/sequence_pad_op.cc paddle/fluid/operators/sequence_ops/sequence_enumerate_op.cc paddle/fluid/operators/sequence_ops/sequence_slice_op.cc paddle/fluid/operators/sequence_ops/sequence_softmax_op.cc paddle/fluid/operators/sequence_ops/sequence_expand_as_op.cc paddle/fluid/operators/sequence_ops/sequence_pool_op.cc paddle/fluid/operators/sequence_ops/sequence_reverse_op.cc paddle/fluid/operators/sequence_ops/sequence_scatter_op.cc paddle/fluid/operators/sequence_ops/sequence_conv_op.cc paddle/fluid/operators/sequence_ops/sequence_concat_op.cc paddle/fluid/operators/sequence_ops/sequence_reshape_op.cc paddle/fluid/operators/jit/helper.cc paddle/fluid/operators/jit/kernel_key.cc paddle/fluid/operators/jit/gen_base.cc paddle/fluid/operators/jit/kernel_pool.cc paddle/fluid/operators/jit/refer/refer.cc paddle/fluid/operators/jit/more/mkl/mkl.cc paddle/fluid/operators/jit/more/intrinsic/layer_norm.cc paddle/fluid/operators/jit/more/intrinsic/crf_decoding.cc paddle/fluid/operators/jit/more/mix/mix.cc paddle/fluid/operators/jit/gen/sgd.cc paddle/fluid/operators/jit/gen/hopv.cc paddle/fluid/operators/jit/gen/lstm.cc paddle/fluid/operators/jit/gen/gru.cc paddle/fluid/operators/jit/gen/vbroadcast.cc paddle/fluid/operators/jit/gen/matmul.cc paddle/fluid/operators/jit/gen/seqpool.cc paddle/fluid/operators/jit/gen/embseqpool.cc paddle/fluid/operators/jit/gen/act.cc paddle/fluid/operators/jit/gen/blas.cc paddle/fluid/operators/reader/reader_op_registry.cc paddle/fluid/operators/reader/py_reader.cc paddle/fluid/operators/reader/buffered_reader.cc paddle/fluid/operators/reader/open_files_op.cc paddle/fluid/operators/reader/create_random_data_generator_op.cc paddle/fluid/operators/reader/create_shuffle_reader_op.cc paddle/fluid/operators/reader/create_batch_reader_op.cc paddle/fluid/operators/reader/create_recordio_file_reader_op.cc paddle/fluid/operators/reader/create_double_buffer_reader_op.cc paddle/fluid/operators/reader/create_multi_pass_reader_op.cc paddle/fluid/operators/reader/create_custom_reader_op.cc paddle/fluid/operators/reader/create_py_reader_op.cc paddle/fluid/operators/reader/read_op.cc paddle/fluid/operators/increment_op.cc paddle/fluid/operators/stack_op.cc paddle/fluid/operators/fc_op.cc paddle/fluid/operators/assign_op.cc paddle/fluid/operators/load_op.cc paddle/fluid/operators/fill_op.cc paddle/fluid/operators/reorder_lod_tensor_by_rank_op.cc paddle/fluid/operators/conv_shift_op.cc paddle/fluid/operators/fill_zeros_like_op.cc paddle/fluid/operators/hash_op.cc paddle/fluid/operators/dequantize_op.cc paddle/fluid/operators/fake_quantize_op.cc paddle/fluid/operators/size_op.cc paddle/fluid/operators/scatter_op.cc paddle/fluid/operators/uniform_random_op.cc paddle/fluid/operators/beam_search_op.cc paddle/fluid/operators/beam_search_decode_op.cc paddle/fluid/operators/dropout_op.cc paddle/fluid/operators/interpolate_op.cc paddle/fluid/operators/sampling_id_op.cc paddle/fluid/operators/lstm_op.cc paddle/fluid/operators/modified_huber_loss_op.cc paddle/fluid/operators/temporal_shift_op.cc paddle/fluid/operators/sum_op.cc paddle/fluid/operators/arg_min_op.cc paddle/fluid/operators/psroi_pool_op.cc paddle/fluid/operators/uniform_random_batch_size_like_op.cc paddle/fluid/operators/rnn_memory_helper_op.cc paddle/fluid/operators/crf_decoding_op.cc paddle/fluid/operators/where_op.cc paddle/fluid/operators/fake_dequantize_op.cc paddle/fluid/operators/mean_iou_op.cc paddle/fluid/operators/roi_align_op.cc paddle/fluid/operators/range_op.cc paddle/fluid/operators/edit_distance_op.cc paddle/fluid/operators/multiplex_op.cc paddle/fluid/operators/clip_op.cc paddle/fluid/operators/gaussian_random_op.cc paddle/fluid/operators/norm_op.cc paddle/fluid/operators/rank_loss_op.cc paddle/fluid/operators/detection_map_op.cc paddle/fluid/operators/lstm_unit_op.cc paddle/fluid/operators/shard_index_op.cc paddle/fluid/operators/shape_op.cc paddle/fluid/operators/arg_max_op.cc paddle/fluid/operators/average_accumulates_op.cc paddle/fluid/operators/requantize_op.cc paddle/fluid/operators/conv_op.cc paddle/fluid/operators/add_position_encoding_op.cc paddle/fluid/operators/gru_unit_op.cc paddle/fluid/operators/batch_norm_op.cc paddle/fluid/operators/chunk_eval_op.cc paddle/fluid/operators/lod_rank_table_op.cc paddle/fluid/operators/unsqueeze_op.cc paddle/fluid/operators/positive_negative_pair_op.cc paddle/fluid/operators/im2sequence_op.cc paddle/fluid/operators/margin_rank_loss_op.cc paddle/fluid/operators/hinge_loss_op.cc paddle/fluid/operators/cvm_op.cc paddle/fluid/operators/huber_loss_op.cc paddle/fluid/operators/crop_op.cc paddle/fluid/operators/activation_op.cc paddle/fluid/operators/hierarchical_sigmoid_op.cc paddle/fluid/operators/unfold_op.cc paddle/fluid/operators/max_sequence_len_op.cc paddle/fluid/operators/mul_op.cc paddle/fluid/operators/attention_lstm_op.cc paddle/fluid/operators/top_k_op.cc paddle/fluid/operators/group_norm_op.cc paddle/fluid/operators/selu_op.cc paddle/fluid/operators/lstmp_op.cc paddle/fluid/operators/merge_lod_tensor_op.cc paddle/fluid/operators/truncated_gaussian_random_op.cc paddle/fluid/operators/label_smooth_op.cc paddle/fluid/operators/matmul_op.cc paddle/fluid/operators/spp_op.cc paddle/fluid/operators/unstack_op.cc paddle/fluid/operators/conv_transpose_op.cc paddle/fluid/operators/diag_op.cc paddle/fluid/operators/unpool_op.cc paddle/fluid/operators/lod_array_length_op.cc paddle/fluid/operators/affine_channel_op.cc paddle/fluid/operators/log_loss_op.cc paddle/fluid/operators/concat_op.cc paddle/fluid/operators/lod_tensor_to_array_op.cc paddle/fluid/operators/gru_op.cc paddle/fluid/operators/coalesce_tensor_op.cc paddle/fluid/operators/fsp_op.cc paddle/fluid/operators/linspace_op.cc paddle/fluid/operators/reverse_op.cc paddle/fluid/operators/recurrent_op.cc paddle/fluid/operators/split_selected_rows_op.cc paddle/fluid/operators/dgc_clip_by_norm_op.cc paddle/fluid/operators/scale_op.cc paddle/fluid/operators/save_op.cc paddle/fluid/operators/load_combine_op.cc paddle/fluid/operators/merge_selected_rows_op.cc paddle/fluid/operators/split_op.cc paddle/fluid/operators/cumsum_op.cc paddle/fluid/operators/deformable_psroi_pooling_op.cc paddle/fluid/operators/teacher_student_sigmoid_loss_op.cc paddle/fluid/operators/transpose_op.cc paddle/fluid/operators/fill_constant_batch_size_like_op.cc paddle/fluid/operators/sigmoid_cross_entropy_with_logits_op.cc paddle/fluid/operators/shuffle_channel_op.cc paddle/fluid/operators/affine_grid_op.cc paddle/fluid/operators/split_lod_tensor_op.cc paddle/fluid/operators/grid_sampler_op.cc paddle/fluid/operators/lookup_table_op.cc paddle/fluid/operators/cos_sim_op.cc paddle/fluid/operators/quantize_op.cc paddle/fluid/operators/spectral_norm_op.cc paddle/fluid/operators/cross_entropy_op.cc paddle/fluid/operators/print_op.cc paddle/fluid/operators/lrn_op.cc paddle/fluid/operators/nce_op.cc paddle/fluid/operators/similarity_focus_op.cc paddle/fluid/operators/get_tensor_from_selected_rows_op.cc paddle/fluid/operators/squared_l2_distance_op.cc paddle/fluid/operators/cudnn_lstm_op.cc paddle/fluid/operators/tree_conv_op.cc paddle/fluid/operators/one_hot_op.cc paddle/fluid/operators/lookup_sparse_table_op.cc paddle/fluid/operators/unique_op.cc paddle/fluid/operators/mean_op.cc paddle/fluid/operators/prelu_op.cc paddle/fluid/operators/delete_var_op.cc paddle/fluid/operators/ctc_align_op.cc paddle/fluid/operators/argsort_op.cc paddle/fluid/operators/data_norm_op.cc paddle/fluid/operators/minus_op.cc paddle/fluid/operators/shrink_rnn_memory_op.cc paddle/fluid/operators/lod_reset_op.cc paddle/fluid/operators/l1_norm_op.cc paddle/fluid/operators/gaussian_random_batch_size_like_op.cc paddle/fluid/operators/is_empty_op.cc paddle/fluid/operators/bilinear_tensor_product_op.cc paddle/fluid/operators/kldiv_loss_op.cc paddle/fluid/operators/squeeze_op.cc paddle/fluid/operators/softmax_op.cc paddle/fluid/operators/clip_by_norm_op.cc paddle/fluid/operators/pool_with_index_op.cc paddle/fluid/operators/linear_chain_crf_op.cc paddle/fluid/operators/reshape_op.cc paddle/fluid/operators/fill_constant_op.cc paddle/fluid/operators/space_to_depth_op.cc paddle/fluid/operators/gather_op.cc paddle/fluid/operators/softmax_with_cross_entropy_op.cc paddle/fluid/operators/slice_op.cc paddle/fluid/operators/sign_op.cc paddle/fluid/operators/expand_op.cc paddle/fluid/operators/smooth_l1_loss_op.cc paddle/fluid/operators/tensor_array_to_tensor_op.cc paddle/fluid/operators/row_conv_op.cc paddle/fluid/operators/pad2d_op.cc paddle/fluid/operators/pixel_shuffle_op.cc paddle/fluid/operators/assign_value_op.cc paddle/fluid/operators/random_crop_op.cc paddle/fluid/operators/squared_l2_norm_op.cc paddle/fluid/operators/save_combine_op.cc paddle/fluid/operators/pool_op.cc paddle/fluid/operators/cast_op.cc paddle/fluid/operators/array_to_lod_tensor_op.cc paddle/fluid/operators/fill_any_like_op.cc paddle/fluid/operators/flatten_op.cc paddle/fluid/operators/sample_logits_op.cc paddle/fluid/operators/pad_op.cc paddle/fluid/operators/bpr_loss_op.cc paddle/fluid/operators/roi_pool_op.cc paddle/fluid/operators/pad_constant_like_op.cc paddle/fluid/operators/isfinite_op.cc paddle/fluid/operators/layer_norm_op.cc paddle/fluid/operators/maxout_op.cc paddle/fluid/operators/warpctc_op.cc paddle/fluid/string/piece.cc paddle/fluid/string/pretty_log.cc paddle/fluid/string/string_helper.cc paddle/fluid/recordio/header.cc paddle/fluid/recordio/chunk.cc paddle/fluid/recordio/writer.cc paddle/fluid/recordio/scanner.cc paddle/fluid/inference/io.cc paddle/fluid/inference/analysis/ir_passes/subgraph_detector.cc paddle/fluid/inference/analysis/ir_passes/subgraph_util.cc paddle/fluid/inference/analysis/passes/ir_graph_build_pass.cc paddle/fluid/inference/analysis/passes/ir_analysis_pass.cc paddle/fluid/inference/analysis/passes/memory_optimize_pass.cc paddle/fluid/inference/analysis/passes/ir_params_sync_among_devices_pass.cc paddle/fluid/inference/analysis/passes/ir_graph_to_program_pass.cc paddle/fluid/inference/analysis/passes/adjust_cudnn_workspace_size_pass.cc paddle/fluid/inference/analysis/passes/inference_op_replace_pass.cc paddle/fluid/inference/analysis/passes/passes.cc paddle/fluid/inference/analysis/helper.cc paddle/fluid/inference/analysis/ir_pass_manager.cc paddle/fluid/inference/analysis/argument.cc paddle/fluid/inference/analysis/analysis_pass.cc paddle/fluid/inference/analysis/analyzer.cc paddle/fluid/inference/utils/benchmark.cc paddle/fluid/inference/api/api.cc paddle/fluid/inference/api/api_impl.cc paddle/fluid/inference/api/helper.cc paddle/fluid/inference/api/analysis_predictor.cc paddle/fluid/inference/api/details/zero_copy_tensor.cc paddle/fluid/inference/api/details/reset_tensor_array.cc paddle/fluid/inference/api/analysis_config.cc paddle/fluid/inference/api/paddle_pass_builder.cc"
 paddle_fluid_avx_mklml_src += ' paddle/fluid/framework/revision.cc'
@@ -58,8 +67,23 @@ paddle_fluid_avx_mklml_src += ' paddle/fluid/framework/revision.cc'
 StaticLibrary('paddle_fluid_avx_mklml', Sources(paddle_fluid_avx_mklml_src, CppFlags('-DHPPL_STUB_FUNC -DLAPACK_FOUND -DPADDLE_DISABLE_PROFILER -DPADDLE_NO_PYTHON -DPADDLE_ON_INFERENCE -DPADDLE_USE_DSO -DPADDLE_USE_PTHREAD_BARRIER -DPADDLE_USE_PTHREAD_SPINLOCK -DPADDLE_VERSION=0.0.0 -DPADDLE_WITH_AVX -DPADDLE_WITH_MKLML -DPADDLE_WITH_XBYAK -DXBYAK64 -DXBYAK_NO_OP_NAMES -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -DPYBIND_AVX_MKLML' + r" -DPADDLE_REVISION=\"%s@%s@%s\"" % (REPO_URL(), REPO_BRANCH(), REPO_REVISION())), CFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -fopenmp -mavx -O3 -DNDEBUG '), CxxFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -fopenmp -mavx -O3 -DNDEBUG ')))
 SharedLibrary('paddle_fluid_avx_mklml', Sources(paddle_fluid_avx_mklml_src,  CppFlags('-DHPPL_STUB_FUNC -DLAPACK_FOUND -DPADDLE_DISABLE_PROFILER -DPADDLE_NO_PYTHON -DPADDLE_ON_INFERENCE -DPADDLE_USE_DSO -DPADDLE_USE_PTHREAD_BARRIER -DPADDLE_USE_PTHREAD_SPINLOCK -DPADDLE_VERSION=0.0.0 -DPADDLE_WITH_AVX -DPADDLE_WITH_MKLML -DPADDLE_WITH_XBYAK -DXBYAK64 -DXBYAK_NO_OP_NAMES -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -DPYBIND_AVX_MKLML' + r" -DPADDLE_REVISION=\"%s@%s@%s\"" % (REPO_URL(), REPO_BRANCH(), REPO_REVISION())), CFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -fopenmp -mavx -O3 -DNDEBUG '), CxxFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -fopenmp -mavx -O3 -DNDEBUG ')), LinkDeps(True))
+#feed
+HEADERS('paddle/fluid/train/custom_trainer/feed/*.h', '$INC/paddle/fluid/train/custom_trainer/feed/')
+HEADERS('paddle/fluid/train/custom_trainer/feed/common/*.h', '$INC/paddle/fluid/train/custom_trainer/feed/common/')
+HEADERS('paddle/fluid/train/custom_trainer/feed/dataset/*.h', '$INC/paddle/fluid/train/custom_trainer/feed/dataset/')
+HEADERS('paddle/fluid/train/custom_trainer/feed/process/*.h', '$INC/paddle/fluid/train/custom_trainer/feed/process/')
+HEADERS('paddle/fluid/train/custom_trainer/feed/shuffler/*.h', '$INC/paddle/fluid/train/custom_trainer/feed/shuffler/')
+HEADERS('paddle/fluid/train/custom_trainer/feed/params_accessor/*.h', '$INC/paddle/fluid/train/custom_trainer/feed/params_accessor/')
+NEED_OUTPUT("baidu/third-party/mklml")
+OUTPUT('paddle/fluid/train/custom_trainer/feed/conf', '$OUT')
+OUTPUT('paddle/fluid/train/custom_trainer/feed/scripts', '$OUT')
+custom_trainer_src = GLOB('paddle/fluid/train/custom_trainer/feed/*/*.cc')
+Application('feed_trainer', Sources('paddle/fluid/train/custom_trainer/feed/main.cc', custom_trainer_src), CppFlags('-DHPPL_STUB_FUNC -DLAPACK_FOUND -DPADDLE_DISABLE_PROFILER -DPADDLE_NO_PYTHON -DCUSTOM_TRAINER -DPADDLE_ON_INFERENCE -DPADDLE_USE_DSO -DPADDLE_USE_PTHREAD_BARRIER -DPADDLE_USE_PTHREAD_SPINLOCK -DPADDLE_VERSION=0.0.0 -DPADDLE_WITH_AVX -DPADDLE_WITH_MKLML -DPADDLE_WITH_XBYAK -DXBYAK64 -DXBYAK_NO_OP_NAMES -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -DPYBIND_AVX_MKLML' + r" -DPADDLE_REVISION=\"%s@%s@%s\"" % (REPO_URL(), REPO_BRANCH(), REPO_REVISION())), CFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -fopenmp -mavx -O3 -DNDEBUG '), CxxFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -fopenmp -mavx -O3 -DNDEBUG '), Libraries('$OUT/lib/libpaddle_fluid_avx_mklml.a'))
 paddle_fluid_noavx_openblas_src = "paddle/fluid/memory/detail/memory_block.cc paddle/fluid/memory/detail/memory_block_desc.cc paddle/fluid/memory/detail/meta_cache.cc paddle/fluid/memory/detail/system_allocator.cc paddle/fluid/memory/detail/buddy_allocator.cc paddle/fluid/memory/allocation/allocator.cc paddle/fluid/memory/allocation/cpu_allocator.cc paddle/fluid/memory/allocation/locked_allocator.cc paddle/fluid/memory/allocation/buffered_allocator.cc paddle/fluid/memory/allocation/best_fit_allocator.cc paddle/fluid/memory/allocation/naive_best_fit_allocator.cc paddle/fluid/memory/allocation/retry_allocator.cc paddle/fluid/memory/allocation/aligned_allocator.cc paddle/fluid/memory/allocation/allocator_strategy.cc paddle/fluid/memory/allocation/allocator_facade.cc paddle/fluid/memory/allocation/auto_growth_best_fit_allocator.cc paddle/fluid/memory/malloc.cc paddle/fluid/memory/memcpy.cc paddle/fluid/platform/profiler.proto paddle/fluid/platform/enforce.cc paddle/fluid/platform/cpu_info.cc paddle/fluid/platform/place.cc paddle/fluid/platform/dynload/dynamic_loader.cc paddle/fluid/platform/dynload/warpctc.cc paddle/fluid/platform/cpu_helper.cc paddle/fluid/platform/temporary_allocator.cc paddle/fluid/platform/device_context.cc paddle/fluid/platform/init.cc paddle/fluid/platform/timer.cc paddle/fluid/platform/lodtensor_printer.cc paddle/fluid/platform/device_tracer.cc paddle/fluid/platform/profiler.cc paddle/fluid/platform/device_memory_aligment.cc paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_optimizer_op_pass.cc paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_adam_op_pass.cc paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_sgd_op_pass.cc paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_momentum_op_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/op_graph_view.cc paddle/fluid/framework/ir/memory_optimize_pass/while_op_eager_deletion_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/recurrent_op_eager_deletion_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/reference_count_pass_helper.cc paddle/fluid/framework/ir/memory_optimize_pass/reference_count_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/memory_optimize_helper.cc paddle/fluid/framework/ir/memory_optimize_pass/memory_optimize_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/inplace_op_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/eager_deletion_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/record_skip_memory_opt_vars_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/memory_reuse_pass.cc paddle/fluid/framework/ir/memory_optimize_pass/buffer_shared_inplace_op_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/modify_op_lock_and_record_event_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_print_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_check_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/sequential_execution_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/fuse_all_reduce_op_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/all_reduce_deps_pass.cc paddle/fluid/framework/ir/multi_devices_graph_pass/backward_optimizer_op_deps_pass.cc paddle/fluid/framework/ir/node.cc paddle/fluid/framework/ir/graph.cc paddle/fluid/framework/ir/graph_helper.cc paddle/fluid/framework/ir/pass.cc paddle/fluid/framework/ir/graph_traits.cc paddle/fluid/framework/ir/graph_pattern_detector.cc paddle/fluid/framework/ir/fuse_pass_base.cc paddle/fluid/framework/ir/coalesce_grad_tensor_pass.cc paddle/fluid/framework/ir/graph_to_program_pass.cc paddle/fluid/framework/ir/graph_viz_pass.cc paddle/fluid/framework/ir/lock_free_optimize_pass.cc paddle/fluid/framework/ir/fc_fuse_pass.cc paddle/fluid/framework/ir/attention_lstm_fuse_pass.cc paddle/fluid/framework/ir/infer_clean_graph_pass.cc paddle/fluid/framework/ir/fc_lstm_fuse_pass.cc paddle/fluid/framework/ir/embedding_fc_lstm_fuse_pass.cc paddle/fluid/framework/ir/fc_gru_fuse_pass.cc paddle/fluid/framework/ir/seq_concat_fc_fuse_pass.cc paddle/fluid/framework/ir/multi_batch_merge_pass.cc paddle/fluid/framework/ir/conv_bn_fuse_pass.cc paddle/fluid/framework/ir/seqconv_eltadd_relu_fuse_pass.cc paddle/fluid/framework/ir/seqpool_concat_fuse_pass.cc paddle/fluid/framework/ir/repeated_fc_relu_fuse_pass.cc paddle/fluid/framework/ir/squared_mat_sub_fuse_pass.cc paddle/fluid/framework/ir/is_test_pass.cc paddle/fluid/framework/ir/conv_elementwise_add_act_fuse_pass.cc paddle/fluid/framework/ir/conv_elementwise_add2_act_fuse_pass.cc paddle/fluid/framework/ir/conv_elementwise_add_fuse_pass.cc paddle/fluid/framework/ir/conv_affine_channel_fuse_pass.cc paddle/fluid/framework/ir/transpose_flatten_concat_fuse_pass.cc paddle/fluid/framework/ir/identity_scale_op_clean_pass.cc paddle/fluid/framework/ir/sync_batch_norm_pass.cc paddle/fluid/framework/ir/runtime_context_cache_pass.cc paddle/fluid/framework/ir/quant_conv2d_dequant_fuse_pass.cc paddle/fluid/framework/ir/fillconstant_elementwisemul_fuse.cc paddle/fluid/framework/ir/shuffle_channel_detect_pass.cc paddle/fluid/framework/ir/delete_quant_dequant_op_pass.cc paddle/fluid/framework/ir/fuse_elewise_add_act_pass.cc paddle/fluid/framework/ir/fuse_relu_depthwise_conv_pass.cc paddle/fluid/framework/ir/pass_builder.cc paddle/fluid/framework/details/var_handle.cc paddle/fluid/framework/details/op_handle_base.cc paddle/fluid/framework/details/scale_loss_grad_op_handle.cc paddle/fluid/framework/details/fetch_op_handle.cc paddle/fluid/framework/details/computation_op_handle.cc paddle/fluid/framework/details/rpc_op_handle.cc paddle/fluid/framework/details/fetch_barrier_op_handle.cc paddle/fluid/framework/details/multi_devices_helper.cc paddle/fluid/framework/details/variable_visitor.cc paddle/fluid/framework/details/all_reduce_op_handle.cc paddle/fluid/framework/details/fused_all_reduce_op_handle.cc paddle/fluid/framework/details/reduce_op_handle.cc paddle/fluid/framework/details/broadcast_op_handle.cc paddle/fluid/framework/details/fused_broadcast_op_handle.cc paddle/fluid/framework/details/gather_op_handle.cc paddle/fluid/framework/details/eager_deletion_op_handle.cc paddle/fluid/framework/details/share_tensor_buffer_op_handle.cc paddle/fluid/framework/details/ssa_graph_executor.cc paddle/fluid/framework/details/threaded_ssa_graph_executor.cc paddle/fluid/framework/details/parallel_ssa_graph_executor.cc paddle/fluid/framework/details/async_ssa_graph_executor.cc paddle/fluid/framework/details/scope_buffered_ssa_graph_executor.cc paddle/fluid/framework/details/fast_threaded_ssa_graph_executor.cc paddle/fluid/framework/details/build_strategy.cc paddle/fluid/framework/fleet/fleet_wrapper.cc paddle/fluid/framework/fleet/nccl_wrapper.cc paddle/fluid/framework/io/fs.cc paddle/fluid/framework/io/shell.cc paddle/fluid/framework/framework.proto paddle/fluid/framework/data_feed.proto paddle/fluid/framework/trainer_desc.proto paddle/fluid/framework/ddim.cc paddle/fluid/framework/data_type.cc paddle/fluid/framework/tensor.cc paddle/fluid/framework/tensor_util.cc paddle/fluid/framework/lod_tensor.cc paddle/fluid/framework/garbage_collector.cc paddle/fluid/framework/reader.cc paddle/fluid/framework/threadpool.cc paddle/fluid/framework/var_type_traits.cc paddle/fluid/framework/scope.cc paddle/fluid/framework/scope_pool.cc paddle/fluid/framework/data_device_transform.cc paddle/fluid/framework/data_type_transform.cc paddle/fluid/framework/data_layout_transform.cc paddle/fluid/framework/data_transform.cc paddle/fluid/framework/attribute.cc paddle/fluid/framework/op_proto_maker.cc paddle/fluid/framework/op_info.cc paddle/fluid/framework/shape_inference.cc paddle/fluid/framework/transfer_scope_cache.cc paddle/fluid/framework/op_kernel_type.cc paddle/fluid/framework/operator.cc paddle/fluid/framework/version.cc paddle/fluid/framework/var_desc.cc paddle/fluid/framework/op_desc.cc paddle/fluid/framework/block_desc.cc paddle/fluid/framework/program_desc.cc paddle/fluid/framework/op_registry.cc paddle/fluid/framework/lod_rank_table.cc paddle/fluid/framework/feed_fetch_method.cc paddle/fluid/framework/variable_helper.cc paddle/fluid/framework/naive_executor.cc paddle/fluid/framework/executor_gc_helper.cc paddle/fluid/framework/executor.cc paddle/fluid/framework/multi_trainer.cc paddle/fluid/framework/pipeline_trainer.cc paddle/fluid/framework/dataset_factory.cc paddle/fluid/framework/dist_multi_trainer.cc paddle/fluid/framework/trainer_factory.cc paddle/fluid/framework/trainer.cc paddle/fluid/framework/data_feed_factory.cc paddle/fluid/framework/data_feed.cc paddle/fluid/framework/device_worker.cc paddle/fluid/framework/hogwild_worker.cc paddle/fluid/framework/downpour_worker.cc paddle/fluid/framework/pull_dense_worker.cc paddle/fluid/framework/section_worker.cc paddle/fluid/framework/device_worker_factory.cc paddle/fluid/framework/data_set.cc paddle/fluid/framework/parallel_executor.cc paddle/fluid/framework/prune.cc paddle/fluid/framework/selected_rows.cc paddle/fluid/framework/dlpack_tensor.cc paddle/fluid/imperative/flags.cc paddle/fluid/operators/math/detail/avx_functions.cc paddle/fluid/operators/math/concat_and_split.cc paddle/fluid/operators/math/context_project.cc paddle/fluid/operators/math/cross_entropy.cc paddle/fluid/operators/math/cos_sim_functor.cc paddle/fluid/operators/math/im2col.cc paddle/fluid/operators/math/sample_prob.cc paddle/fluid/operators/math/sampler.cc paddle/fluid/operators/math/gru_compute.cc paddle/fluid/operators/math/lstm_compute.cc paddle/fluid/operators/math/blas.cc paddle/fluid/operators/math/math_function.cc paddle/fluid/operators/math/maxouting.cc paddle/fluid/operators/math/pooling.cc paddle/fluid/operators/math/selected_rows_functor.cc paddle/fluid/operators/math/sequence2batch.cc paddle/fluid/operators/math/sequence_padding.cc paddle/fluid/operators/math/sequence_pooling.cc paddle/fluid/operators/math/sequence_scale.cc paddle/fluid/operators/math/softmax.cc paddle/fluid/operators/math/beam_search.cc paddle/fluid/operators/math/matrix_bit_code.cc paddle/fluid/operators/math/unpooling.cc paddle/fluid/operators/math/vol2col.cc paddle/fluid/operators/math/tree2col.cc paddle/fluid/operators/controlflow/feed_op.cc paddle/fluid/operators/controlflow/logical_op.cc paddle/fluid/operators/controlflow/while_op.cc paddle/fluid/operators/controlflow/get_places_op.cc paddle/fluid/operators/controlflow/fetch_op.cc paddle/fluid/operators/controlflow/compare_op.cc paddle/fluid/operators/controlflow/conditional_block_infer_op.cc paddle/fluid/operators/controlflow/conditional_block_op.cc paddle/fluid/operators/controlflow/tensor_array_read_write_op.cc paddle/fluid/operators/controlflow/op_variant.cc paddle/fluid/operators/controlflow/recurrent_op_helper.cc paddle/fluid/operators/controlflow/while_op_helper.cc paddle/fluid/operators/detection/bipartite_match_op.cc paddle/fluid/operators/detection/box_coder_op.cc paddle/fluid/operators/detection/iou_similarity_op.cc paddle/fluid/operators/detection/mine_hard_examples_op.cc paddle/fluid/operators/detection/multiclass_nms_op.cc paddle/fluid/operators/detection/poly_util.cc paddle/fluid/operators/detection/gpc.cc paddle/fluid/operators/detection/prior_box_op.cc paddle/fluid/operators/detection/density_prior_box_op.cc paddle/fluid/operators/detection/anchor_generator_op.cc paddle/fluid/operators/detection/target_assign_op.cc paddle/fluid/operators/detection/polygon_box_transform_op.cc paddle/fluid/operators/detection/rpn_target_assign_op.cc paddle/fluid/operators/detection/generate_proposal_labels_op.cc paddle/fluid/operators/detection/box_clip_op.cc paddle/fluid/operators/detection/yolov3_loss_op.cc paddle/fluid/operators/detection/yolo_box_op.cc paddle/fluid/operators/detection/box_decoder_and_assign_op.cc paddle/fluid/operators/detection/sigmoid_focal_loss_op.cc paddle/fluid/operators/detection/retinanet_detection_output_op.cc paddle/fluid/operators/detection/generate_proposals_op.cc paddle/fluid/operators/detection/distribute_fpn_proposals_op.cc paddle/fluid/operators/detection/collect_fpn_proposals_op.cc paddle/fluid/operators/detection/roi_perspective_transform_op.cc paddle/fluid/operators/detection/mask_util.cc paddle/fluid/operators/detection/generate_mask_labels_op.cc paddle/fluid/operators/elementwise/elementwise_mod_op.cc paddle/fluid/operators/elementwise/elementwise_floordiv_op.cc paddle/fluid/operators/elementwise/elementwise_max_op.cc paddle/fluid/operators/elementwise/elementwise_pow_op.cc paddle/fluid/operators/elementwise/elementwise_sub_op.cc paddle/fluid/operators/elementwise/elementwise_add_op.cc paddle/fluid/operators/elementwise/elementwise_min_op.cc paddle/fluid/operators/elementwise/elementwise_div_op.cc paddle/fluid/operators/elementwise/elementwise_mul_op.cc paddle/fluid/operators/fused/fusion_squared_mat_sub_op.cc paddle/fluid/operators/fused/fusion_seqpool_concat_op.cc paddle/fluid/operators/fused/fused_embedding_fc_lstm_op.cc paddle/fluid/operators/fused/fusion_seqexpand_concat_fc_op.cc paddle/fluid/operators/fused/fused_embedding_seq_pool_op.cc paddle/fluid/operators/fused/fusion_seqconv_eltadd_relu_op.cc paddle/fluid/operators/fused/fusion_gru_op.cc paddle/fluid/operators/fused/fusion_repeated_fc_relu_op.cc paddle/fluid/operators/fused/fusion_lstm_op.cc paddle/fluid/operators/fused/fused_elemwise_activation_op.cc paddle/fluid/operators/metrics/accuracy_op.cc paddle/fluid/operators/metrics/precision_recall_op.cc paddle/fluid/operators/metrics/auc_op.cc paddle/fluid/operators/optimizers/adamax_op.cc paddle/fluid/operators/optimizers/sgd_op.cc paddle/fluid/operators/optimizers/lars_momentum_op.cc paddle/fluid/operators/optimizers/adagrad_op.cc paddle/fluid/operators/optimizers/ftrl_op.cc paddle/fluid/operators/optimizers/momentum_op.cc paddle/fluid/operators/optimizers/adadelta_op.cc paddle/fluid/operators/optimizers/rmsprop_op.cc paddle/fluid/operators/optimizers/lamb_op.cc paddle/fluid/operators/optimizers/proximal_gd_op.cc paddle/fluid/operators/optimizers/proximal_adagrad_op.cc paddle/fluid/operators/optimizers/adam_op.cc paddle/fluid/operators/optimizers/decayed_adagrad_op.cc paddle/fluid/operators/reduce_ops/reduce_all_op.cc paddle/fluid/operators/reduce_ops/reduce_min_op.cc paddle/fluid/operators/reduce_ops/reduce_sum_op.cc paddle/fluid/operators/reduce_ops/reduce_any_op.cc paddle/fluid/operators/reduce_ops/reduce_max_op.cc paddle/fluid/operators/reduce_ops/reduce_mean_op.cc paddle/fluid/operators/reduce_ops/reduce_prod_op.cc paddle/fluid/operators/sequence_ops/sequence_erase_op.cc paddle/fluid/operators/sequence_ops/sequence_unpad_op.cc paddle/fluid/operators/sequence_ops/sequence_mask_op.cc paddle/fluid/operators/sequence_ops/sequence_expand_op.cc paddle/fluid/operators/sequence_ops/sequence_pad_op.cc paddle/fluid/operators/sequence_ops/sequence_enumerate_op.cc paddle/fluid/operators/sequence_ops/sequence_slice_op.cc paddle/fluid/operators/sequence_ops/sequence_softmax_op.cc paddle/fluid/operators/sequence_ops/sequence_expand_as_op.cc paddle/fluid/operators/sequence_ops/sequence_pool_op.cc paddle/fluid/operators/sequence_ops/sequence_reverse_op.cc paddle/fluid/operators/sequence_ops/sequence_scatter_op.cc paddle/fluid/operators/sequence_ops/sequence_conv_op.cc paddle/fluid/operators/sequence_ops/sequence_concat_op.cc paddle/fluid/operators/sequence_ops/sequence_reshape_op.cc paddle/fluid/operators/jit/helper.cc paddle/fluid/operators/jit/kernel_key.cc paddle/fluid/operators/jit/gen_base.cc paddle/fluid/operators/jit/kernel_pool.cc paddle/fluid/operators/jit/refer/refer.cc paddle/fluid/operators/jit/more/mix/mix.cc paddle/fluid/operators/jit/gen/sgd.cc paddle/fluid/operators/jit/gen/hopv.cc paddle/fluid/operators/jit/gen/lstm.cc paddle/fluid/operators/jit/gen/gru.cc paddle/fluid/operators/jit/gen/vbroadcast.cc paddle/fluid/operators/jit/gen/matmul.cc paddle/fluid/operators/jit/gen/seqpool.cc paddle/fluid/operators/jit/gen/embseqpool.cc paddle/fluid/operators/jit/gen/act.cc paddle/fluid/operators/jit/gen/blas.cc paddle/fluid/operators/reader/reader_op_registry.cc paddle/fluid/operators/reader/py_reader.cc paddle/fluid/operators/reader/buffered_reader.cc paddle/fluid/operators/reader/open_files_op.cc paddle/fluid/operators/reader/create_random_data_generator_op.cc paddle/fluid/operators/reader/create_shuffle_reader_op.cc paddle/fluid/operators/reader/create_batch_reader_op.cc paddle/fluid/operators/reader/create_recordio_file_reader_op.cc paddle/fluid/operators/reader/create_double_buffer_reader_op.cc paddle/fluid/operators/reader/create_multi_pass_reader_op.cc paddle/fluid/operators/reader/create_custom_reader_op.cc paddle/fluid/operators/reader/create_py_reader_op.cc paddle/fluid/operators/reader/read_op.cc paddle/fluid/operators/increment_op.cc paddle/fluid/operators/stack_op.cc paddle/fluid/operators/fc_op.cc paddle/fluid/operators/assign_op.cc paddle/fluid/operators/load_op.cc paddle/fluid/operators/fill_op.cc paddle/fluid/operators/reorder_lod_tensor_by_rank_op.cc paddle/fluid/operators/conv_shift_op.cc paddle/fluid/operators/fill_zeros_like_op.cc paddle/fluid/operators/hash_op.cc paddle/fluid/operators/dequantize_op.cc paddle/fluid/operators/fake_quantize_op.cc paddle/fluid/operators/size_op.cc paddle/fluid/operators/scatter_op.cc paddle/fluid/operators/uniform_random_op.cc paddle/fluid/operators/beam_search_op.cc paddle/fluid/operators/beam_search_decode_op.cc paddle/fluid/operators/dropout_op.cc paddle/fluid/operators/interpolate_op.cc paddle/fluid/operators/sampling_id_op.cc paddle/fluid/operators/lstm_op.cc paddle/fluid/operators/modified_huber_loss_op.cc paddle/fluid/operators/temporal_shift_op.cc paddle/fluid/operators/sum_op.cc paddle/fluid/operators/arg_min_op.cc paddle/fluid/operators/psroi_pool_op.cc paddle/fluid/operators/uniform_random_batch_size_like_op.cc paddle/fluid/operators/rnn_memory_helper_op.cc paddle/fluid/operators/crf_decoding_op.cc paddle/fluid/operators/where_op.cc paddle/fluid/operators/fake_dequantize_op.cc paddle/fluid/operators/mean_iou_op.cc paddle/fluid/operators/roi_align_op.cc paddle/fluid/operators/range_op.cc paddle/fluid/operators/edit_distance_op.cc paddle/fluid/operators/multiplex_op.cc paddle/fluid/operators/clip_op.cc paddle/fluid/operators/gaussian_random_op.cc paddle/fluid/operators/norm_op.cc paddle/fluid/operators/rank_loss_op.cc paddle/fluid/operators/detection_map_op.cc paddle/fluid/operators/lstm_unit_op.cc paddle/fluid/operators/shard_index_op.cc paddle/fluid/operators/shape_op.cc paddle/fluid/operators/arg_max_op.cc paddle/fluid/operators/average_accumulates_op.cc paddle/fluid/operators/requantize_op.cc paddle/fluid/operators/conv_op.cc paddle/fluid/operators/add_position_encoding_op.cc paddle/fluid/operators/gru_unit_op.cc paddle/fluid/operators/batch_norm_op.cc paddle/fluid/operators/chunk_eval_op.cc paddle/fluid/operators/lod_rank_table_op.cc paddle/fluid/operators/unsqueeze_op.cc paddle/fluid/operators/positive_negative_pair_op.cc paddle/fluid/operators/im2sequence_op.cc paddle/fluid/operators/margin_rank_loss_op.cc paddle/fluid/operators/hinge_loss_op.cc paddle/fluid/operators/cvm_op.cc paddle/fluid/operators/huber_loss_op.cc paddle/fluid/operators/crop_op.cc paddle/fluid/operators/activation_op.cc paddle/fluid/operators/hierarchical_sigmoid_op.cc paddle/fluid/operators/unfold_op.cc paddle/fluid/operators/max_sequence_len_op.cc paddle/fluid/operators/mul_op.cc paddle/fluid/operators/attention_lstm_op.cc paddle/fluid/operators/top_k_op.cc paddle/fluid/operators/group_norm_op.cc paddle/fluid/operators/selu_op.cc paddle/fluid/operators/lstmp_op.cc paddle/fluid/operators/merge_lod_tensor_op.cc paddle/fluid/operators/truncated_gaussian_random_op.cc paddle/fluid/operators/label_smooth_op.cc paddle/fluid/operators/matmul_op.cc paddle/fluid/operators/spp_op.cc paddle/fluid/operators/unstack_op.cc paddle/fluid/operators/conv_transpose_op.cc paddle/fluid/operators/diag_op.cc paddle/fluid/operators/unpool_op.cc paddle/fluid/operators/lod_array_length_op.cc paddle/fluid/operators/affine_channel_op.cc paddle/fluid/operators/log_loss_op.cc paddle/fluid/operators/concat_op.cc paddle/fluid/operators/lod_tensor_to_array_op.cc paddle/fluid/operators/gru_op.cc paddle/fluid/operators/coalesce_tensor_op.cc paddle/fluid/operators/fsp_op.cc paddle/fluid/operators/linspace_op.cc paddle/fluid/operators/reverse_op.cc paddle/fluid/operators/recurrent_op.cc paddle/fluid/operators/split_selected_rows_op.cc paddle/fluid/operators/dgc_clip_by_norm_op.cc paddle/fluid/operators/scale_op.cc paddle/fluid/operators/save_op.cc paddle/fluid/operators/load_combine_op.cc paddle/fluid/operators/merge_selected_rows_op.cc paddle/fluid/operators/split_op.cc paddle/fluid/operators/cumsum_op.cc paddle/fluid/operators/deformable_psroi_pooling_op.cc paddle/fluid/operators/teacher_student_sigmoid_loss_op.cc paddle/fluid/operators/transpose_op.cc paddle/fluid/operators/fill_constant_batch_size_like_op.cc paddle/fluid/operators/sigmoid_cross_entropy_with_logits_op.cc paddle/fluid/operators/shuffle_channel_op.cc paddle/fluid/operators/affine_grid_op.cc paddle/fluid/operators/split_lod_tensor_op.cc paddle/fluid/operators/grid_sampler_op.cc paddle/fluid/operators/lookup_table_op.cc paddle/fluid/operators/cos_sim_op.cc paddle/fluid/operators/quantize_op.cc paddle/fluid/operators/spectral_norm_op.cc paddle/fluid/operators/cross_entropy_op.cc paddle/fluid/operators/print_op.cc paddle/fluid/operators/lrn_op.cc paddle/fluid/operators/nce_op.cc paddle/fluid/operators/similarity_focus_op.cc paddle/fluid/operators/get_tensor_from_selected_rows_op.cc paddle/fluid/operators/squared_l2_distance_op.cc paddle/fluid/operators/cudnn_lstm_op.cc paddle/fluid/operators/tree_conv_op.cc paddle/fluid/operators/one_hot_op.cc paddle/fluid/operators/lookup_sparse_table_op.cc paddle/fluid/operators/unique_op.cc paddle/fluid/operators/mean_op.cc paddle/fluid/operators/prelu_op.cc paddle/fluid/operators/delete_var_op.cc paddle/fluid/operators/ctc_align_op.cc paddle/fluid/operators/argsort_op.cc paddle/fluid/operators/data_norm_op.cc paddle/fluid/operators/minus_op.cc paddle/fluid/operators/shrink_rnn_memory_op.cc paddle/fluid/operators/lod_reset_op.cc paddle/fluid/operators/l1_norm_op.cc paddle/fluid/operators/gaussian_random_batch_size_like_op.cc paddle/fluid/operators/is_empty_op.cc paddle/fluid/operators/bilinear_tensor_product_op.cc paddle/fluid/operators/kldiv_loss_op.cc paddle/fluid/operators/squeeze_op.cc paddle/fluid/operators/softmax_op.cc paddle/fluid/operators/clip_by_norm_op.cc paddle/fluid/operators/pool_with_index_op.cc paddle/fluid/operators/linear_chain_crf_op.cc paddle/fluid/operators/reshape_op.cc paddle/fluid/operators/fill_constant_op.cc paddle/fluid/operators/space_to_depth_op.cc paddle/fluid/operators/gather_op.cc paddle/fluid/operators/softmax_with_cross_entropy_op.cc paddle/fluid/operators/slice_op.cc paddle/fluid/operators/sign_op.cc paddle/fluid/operators/expand_op.cc paddle/fluid/operators/smooth_l1_loss_op.cc paddle/fluid/operators/tensor_array_to_tensor_op.cc paddle/fluid/operators/row_conv_op.cc paddle/fluid/operators/pad2d_op.cc paddle/fluid/operators/pixel_shuffle_op.cc paddle/fluid/operators/assign_value_op.cc paddle/fluid/operators/random_crop_op.cc paddle/fluid/operators/squared_l2_norm_op.cc paddle/fluid/operators/save_combine_op.cc paddle/fluid/operators/pool_op.cc paddle/fluid/operators/cast_op.cc paddle/fluid/operators/array_to_lod_tensor_op.cc paddle/fluid/operators/fill_any_like_op.cc paddle/fluid/operators/flatten_op.cc paddle/fluid/operators/sample_logits_op.cc paddle/fluid/operators/pad_op.cc paddle/fluid/operators/bpr_loss_op.cc paddle/fluid/operators/roi_pool_op.cc paddle/fluid/operators/pad_constant_like_op.cc paddle/fluid/operators/isfinite_op.cc paddle/fluid/operators/layer_norm_op.cc paddle/fluid/operators/maxout_op.cc paddle/fluid/operators/warpctc_op.cc paddle/fluid/string/piece.cc paddle/fluid/string/pretty_log.cc paddle/fluid/string/string_helper.cc paddle/fluid/recordio/header.cc paddle/fluid/recordio/chunk.cc paddle/fluid/recordio/writer.cc paddle/fluid/recordio/scanner.cc paddle/fluid/inference/io.cc paddle/fluid/inference/analysis/ir_passes/subgraph_detector.cc paddle/fluid/inference/analysis/ir_passes/subgraph_util.cc paddle/fluid/inference/analysis/passes/ir_graph_build_pass.cc paddle/fluid/inference/analysis/passes/ir_analysis_pass.cc paddle/fluid/inference/analysis/passes/memory_optimize_pass.cc paddle/fluid/inference/analysis/passes/ir_params_sync_among_devices_pass.cc paddle/fluid/inference/analysis/passes/ir_graph_to_program_pass.cc paddle/fluid/inference/analysis/passes/adjust_cudnn_workspace_size_pass.cc paddle/fluid/inference/analysis/passes/inference_op_replace_pass.cc paddle/fluid/inference/analysis/passes/passes.cc paddle/fluid/inference/analysis/helper.cc paddle/fluid/inference/analysis/ir_pass_manager.cc paddle/fluid/inference/analysis/argument.cc paddle/fluid/inference/analysis/analysis_pass.cc paddle/fluid/inference/analysis/analyzer.cc paddle/fluid/inference/utils/benchmark.cc paddle/fluid/inference/api/api.cc paddle/fluid/inference/api/api_impl.cc paddle/fluid/inference/api/helper.cc paddle/fluid/inference/api/analysis_predictor.cc paddle/fluid/inference/api/details/zero_copy_tensor.cc paddle/fluid/inference/api/details/reset_tensor_array.cc paddle/fluid/inference/api/analysis_config.cc paddle/fluid/inference/api/paddle_pass_builder.cc"
 paddle_fluid_noavx_openblas_src += ' paddle/fluid/framework/revision.cc'
 StaticLibrary('paddle_fluid_noavx_openblas', Sources(paddle_fluid_noavx_openblas_src, CppFlags('-DHPPL_STUB_FUNC -DPADDLE_DISABLE_PROFILER -DPADDLE_NO_PYTHON -DPADDLE_ON_INFERENCE -DPADDLE_USE_DSO -DPADDLE_USE_OPENBLAS -DPADDLE_USE_PTHREAD_BARRIER -DPADDLE_USE_PTHREAD_SPINLOCK -DPADDLE_VERSION=0.0.0 -DPADDLE_WITH_XBYAK -DXBYAK64 -DXBYAK_NO_OP_NAMES -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -DPYBIND_NOAVX_OPENBLAS' + r" -DPADDLE_REVISION=\"%s@%s@%s\"" % (REPO_URL(), REPO_BRANCH(), REPO_REVISION())), CFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -msse3 -O3 -DNDEBUG '), CxxFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -msse3 -O3 -DNDEBUG ')))
 SharedLibrary('paddle_fluid_noavx_openblas', Sources(paddle_fluid_noavx_openblas_src, CppFlags('-DHPPL_STUB_FUNC -DPADDLE_DISABLE_PROFILER -DPADDLE_NO_PYTHON -DPADDLE_ON_INFERENCE -DPADDLE_USE_DSO -DPADDLE_USE_OPENBLAS -DPADDLE_USE_PTHREAD_BARRIER -DPADDLE_USE_PTHREAD_SPINLOCK -DPADDLE_VERSION=0.0.0 -DPADDLE_WITH_XBYAK -DXBYAK64 -DXBYAK_NO_OP_NAMES -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -DPYBIND_NOAVX_OPENBLAS' + r" -DPADDLE_REVISION=\"%s@%s@%s\"" % (REPO_URL(), REPO_BRANCH(), REPO_REVISION())), CFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -msse3 -O3 -DNDEBUG '), CxxFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -msse3 -O3 -DNDEBUG ')), LinkDeps(True))
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -54,6 +54,7 @@ option(WITH_AVX         "Compile PaddlePaddle with AVX intrinsics"      ${AVX_FO
 option(WITH_PYTHON      "Compile PaddlePaddle with python interpreter"  ON)
 option(WITH_TESTING     "Compile PaddlePaddle with unit testing"        OFF)
 option(WITH_MKL         "Compile PaddlePaddle with MKL support."        ${AVX_FOUND})
+option(WITH_CUSTOM_TRAINER  "Turn on trainer implement by custom"       OFF)
 option(WITH_SYSTEM_BLAS   "Use system blas library"           OFF)
 option(WITH_DISTRIBUTE  "Compile with distributed support"              OFF)
 option(WITH_BRPC_RDMA     "Use brpc rdma as the rpc protocal"           OFF)
@@ -146,10 +147,12 @@ include(external/dlpack)
 include(external/snappy)    # download snappy
 include(external/snappystream) # download snappystream
 include(external/warpctc)   # download, build, install warpctc
+include(external/yaml-cpp)    # download yaml
 if (NOT WIN32)
 # there is no official support of nccl, cupti in windows
 include(cupti)
+include(external/gzstream)
 endif (NOT WIN32)
 if(WITH_PSLIB)

--- a/cmake/external/yaml-cpp.cmake
+++ b/cmake/external/yaml-cpp.cmake
+# 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.
+include (ExternalProject)
+IF(NOT ${WITH_CUSTOM_TRAINER})
+  return()
+ENDIF(NOT ${WITH_CUSTOM_TRAINER})
+set(YAML_SOURCES_DIR ${THIRD_PARTY_PATH}/yaml-cpp)
+set(YAML_INSTALL_DIR ${THIRD_PARTY_PATH}/install/yaml-cpp)
+set(YAML_INCLUDE_DIR "${YAML_INSTALL_DIR}/include" CACHE PATH "yaml include directory." FORCE)
+SET(YAML_CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS})
+ExternalProject_Add(
+    extern_yaml
+    GIT_REPOSITORY "https://github.com/jbeder/yaml-cpp"
+    GIT_TAG "yaml-cpp-0.6.2"
+    PREFIX          ${YAML_SOURCES_DIR}
+    UPDATE_COMMAND  ""
+    CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
+                    -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+                    -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
+                    -DCMAKE_C_FLAGS_DEBUG=${CMAKE_C_FLAGS_DEBUG}
+                    -DCMAKE_C_FLAGS_RELEASE=${CMAKE_C_FLAGS_RELEASE}
+                    -DCMAKE_CXX_FLAGS=${YAML_CMAKE_CXX_FLAGS}
+                    -DCMAKE_CXX_FLAGS_RELEASE=${CMAKE_CXX_FLAGS_RELEASE}
+                    -DCMAKE_CXX_FLAGS_DEBUG=${CMAKE_CXX_FLAGS_DEBUG}
+                    -DCMAKE_INSTALL_PREFIX=${YAML_INSTALL_DIR}
+                    -DCMAKE_INSTALL_LIBDIR=${YAML_INSTALL_DIR}/lib
+                    -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+                    -DBUILD_TESTING=OFF
+                    -DYAML_BUILD_TESTS:BOOL=OFF
+                    -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE}
+                    ${EXTERNAL_OPTIONAL_ARGS}
+    CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${YAML_INSTALL_DIR}
+                     -DCMAKE_INSTALL_LIBDIR:PATH=${YAML_INSTALL_DIR}/lib
+                     -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
+                     -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
+)
+set(YAML_LIBRARIES "${YAML_INSTALL_DIR}/lib/libyaml-cpp.a")
+add_library(yaml-cpp STATIC IMPORTED GLOBAL)
+set_property(TARGET yaml-cpp PROPERTY IMPORTED_LOCATION ${YAML_LIBRARIES})
+include_directories(${YAML_INCLUDE_DIR})
+add_dependencies(yaml-cpp extern_yaml)
--- a/paddle/fluid/platform/dynload/CMakeLists.txt
+++ b/paddle/fluid/platform/dynload/CMakeLists.txt
@@ -21,4 +21,7 @@ cc_library(dynload_warpctc SRCS warpctc.cc DEPS dynamic_loader warpctc)
 if (WITH_MKLML)
    cc_library(dynload_mklml SRCS mklml.cc DEPS dynamic_loader mklml)
 endif()
+if (WITH_CUSTOM_TRAINER)
+    cc_library(dynload_custom_trainer SRCS custom_trainer.cc DEPS dynamic_loader)
+endif()
 # TODO(TJ): add iomp, mkldnn?
--- a/paddle/fluid/platform/dynload/dynamic_loader.cc
+++ b/paddle/fluid/platform/dynload/dynamic_loader.cc
@@ -48,6 +48,7 @@ DEFINE_string(
    "Specify path for loading tensorrt library, such as libnvinfer.so.");
 DEFINE_string(mklml_dir, "", "Specify path for loading libmklml_intel.so.");
+DEFINE_string(custom_trainer_dir, "", "Specify path for loading custom_trainer.so.");
 namespace paddle {
 namespace platform {
@@ -253,6 +254,10 @@ void* GetMKLMLDsoHandle() {
 #endif
 }
+void* GetCustomTrainerDsoHandle() {
+  return GetDsoHandleFromSearchPath(FLAGS_custom_trainer_dir, "custom_trainer.so");
+}
 }  // namespace dynload
 }  // namespace platform
 }  // namespace paddle
--- a/paddle/fluid/platform/dynload/dynamic_loader.h
+++ b/paddle/fluid/platform/dynload/dynamic_loader.h
@@ -32,6 +32,7 @@ void* GetWarpCTCDsoHandle();
 void* GetNCCLDsoHandle();
 void* GetTensorRtDsoHandle();
 void* GetMKLMLDsoHandle();
+void* GetCustomTrainerDsoHandle();
 }  // namespace dynload
 }  // namespace platform

--- a/paddle/fluid/train/CMakeLists.txt
+++ b/paddle/fluid/train/CMakeLists.txt
@@ -29,3 +29,7 @@ endfunction(train_test)
 if(WITH_TESTING)
  train_test(recognize_digits ARGS mlp conv)
 endif()
+if(WITH_CUSTOM_TRAINER)
+   add_subdirectory(custom_trainer)
+endif()
--- a/paddle/fluid/train/custom_trainer/CMakeLists.txt
+++ b/paddle/fluid/train/custom_trainer/CMakeLists.txt
+add_subdirectory(feed)
--- a/paddle/fluid/train/custom_trainer/feed/CMakeLists.txt
+++ b/paddle/fluid/train/custom_trainer/feed/CMakeLists.txt
+add_subdirectory(common)
+add_subdirectory(process)
+cc_library(custom_trainer_main SRCS main.cc DEPS custom_trainer_process custom_trainer_common)
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+set(ARCHIVE_START "-Wl,--whole-archive")
+set(ARCHIVE_END "-Wl,--no-whole-archive")
+set(EXTERNAL_LIB "-lrt -ldl -lpthread")
+add_executable(feed_trainer main.cc)
+target_link_libraries(feed_trainer
+        ${MACOS_LD_FLAGS}
+        ${ARCHIVE_START}
+        ${ARCHIVE_END}
+        glog gflags protobuf snappystream snappy z xxhash yaml-cpp 
+        paddle_fluid custom_trainer_common custom_trainer_process
+        ${EXTERNAL_LIB})
--- a/paddle/fluid/train/custom_trainer/feed/common/CMakeLists.txt
+++ b/paddle/fluid/train/custom_trainer/feed/common/CMakeLists.txt
+cc_library(custom_trainer_common SRCS registerer.cc DEPS memory)
--- a/paddle/fluid/train/custom_trainer/feed/common/registerer.cc
+++ b/paddle/fluid/train/custom_trainer/feed/common/registerer.cc
+#include "paddle/fluid/train/custom_trainer/feed/common/registerer.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+BaseClassMap& global_factory_map() {
+    static BaseClassMap *base_class = new BaseClassMap();
+    return *base_class;
+}
+BaseClassMap& global_factory_map_cpp() {
+    return global_factory_map();
+}
+}// feed
+}// namespace custom_trainer
+}// namespace paddle
+/* vim: set expandtab ts=4 sw=4 sts=4 tw=100: */
--- a/paddle/fluid/train/custom_trainer/feed/common/registerer.h
+++ b/paddle/fluid/train/custom_trainer/feed/common/registerer.h
+#pragma once
+#include <map>
+#include <string>
+#include <iostream>
+#include <vector>
+#include <glog/logging.h>
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+class Any {
+public:
+    Any() : content_(NULL) {}
+    template<typename ValueType>
+        Any(const ValueType &value) : content_(new Holder<ValueType>(value)) {}
+    Any(const Any &other) : content_(other.content_ ? other.content_->clone() : NULL) {}
+    ~Any() {
+        delete content_;
+    }
+    template<typename ValueType> ValueType *any_cast() {
+        return content_ ? &static_cast<Holder<ValueType> *>(content_)->held_ : NULL;
+    }
+private:
+    class PlaceHolder {
+    public:
+        virtual ~PlaceHolder() {}
+        virtual PlaceHolder *clone() const = 0;
+    };
+    template<typename ValueType>
+        class Holder : public PlaceHolder {
+        public:
+            explicit Holder(const ValueType &value) : held_(value) {}
+            virtual PlaceHolder *clone() const {
+                return new Holder(held_);
+            }
+            ValueType held_;
+        };
+    PlaceHolder *content_;
+};
+class ObjectFactory {
+public:
+    ObjectFactory() {}
+    virtual ~ObjectFactory() {}
+    virtual Any NewInstance() {
+        return Any();
+    }
+private:
+};
+typedef std::map<std::string, ObjectFactory*> FactoryMap;
+typedef std::map<std::string, FactoryMap> BaseClassMap;
+#ifdef __cplusplus
+extern "C" {
+#endif
+BaseClassMap& global_factory_map();
+#ifdef __cplusplus
+}
+#endif
+BaseClassMap& global_factory_map_cpp();
+#define REGISTER_REGISTERER(base_class) \
+    class base_class ## Registerer { \
+        public: \
+            static base_class *CreateInstanceByName(const ::std::string &name) { \
+                if (global_factory_map_cpp().find(#base_class) \
+                        == global_factory_map_cpp().end()) { \
+                    LOG(ERROR) << "Can't Find BaseClass For CreateClass with:" << #base_class; \
+                    return NULL; \
+                } \
+                FactoryMap &map = global_factory_map_cpp()[#base_class]; \
+                FactoryMap::iterator iter = map.find(name); \
+                if (iter == map.end()) { \
+                    LOG(ERROR) << "Can't Find Class For Create with:" << name; \
+                    return NULL; \
+                } \
+                Any object = iter->second->NewInstance(); \
+                return *(object.any_cast<base_class*>()); \
+            } \
+    };
+#define REGISTER_CLASS(clazz, name) \
+    class ObjectFactory##name : public ObjectFactory { \
+        public: \
+            Any NewInstance() { \
+                return Any(new name()); \
+            } \
+    }; \
+    void register_factory_##name() { \
+        FactoryMap &map = global_factory_map_cpp()[#clazz]; \
+            if (map.find(#name) == map.end()) { \
+                map[#name] = new ObjectFactory##name(); \
+            } \
+    } \
+    void register_factory_##name() __attribute__((constructor)); 
+#define CREATE_CLASS(base_class, name) \
+    base_class##Registerer::CreateInstanceByName(name);
+}//namespace feed
+}//namespace custom_trainer
+}//namespace paddle
+/* vim: set expandtab ts=4 sw=4 sts=4 tw=100: */
--- a/paddle/fluid/train/custom_trainer/feed/common/runtime_environment.h
+++ b/paddle/fluid/train/custom_trainer/feed/common/runtime_environment.h
+/*
+ *Author: xiexionghang
+ *运行环境，屏蔽MPI or Local环境的运行差异
+ *为了兼容不同环境的底层实现，Env的接口调用条件严格于sum(limit(env[n]))
+ *如：MPI环境下，写接口只允许单线程调用，那么默认对所有Env保证此调用限制
+ */
+#pragma once
+#include "paddle/fluid/string/string_helper.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+class RuntimeEnvironment {
+public:
+    RuntimeEnvironment() {}
+    virtual ~RuntimeEnvironment() {}
+    //配置初始化
+    virtual int initialize(YAML::Node& config) = 0;
+    //环境初始化，会在所有依赖模块initialize后调用
+    virtual int wireup() = 0;
+    //多线程可调用接口  Start
+    //当前环境rank_idx
+    virtual uint32_t rank_idx() = 0;
+    //环境定制化log
+    template<class... ARGS>
+    void log(int log_type, const char* fmt, ARGS && ... args) {
+        print_log(log_type, paddle::string::format_string(fmt, args...));
+    }
+    //多线程可调用接口      End
+    //接口只允许在主线程调用   Start
+    //barrier
+    virtual void barrier_all() = 0;
+    //接口只允许在主线程调用   End
+protected:
+    virtual void print_log(int log_type, const std::string& log_str) = 0;
+};
+class MPIRuntimeEnvironment : public RuntimeEnvironment {
+public:
+    MPIRuntimeEnvironment() {}
+    virtual ~MPIRuntimeEnvironment() {}
+    //配置初始化
+    virtual int initialize(YAML::Node& config) = 0;
+    //环境初始化，会在所有依赖模块initialize后调用
+    virtual int wireup() = 0;
+    //当前环境rank_idx
+    virtual uint32_t rank_idx() = 0;
+};
+}  // namespace feed
+}  // namespace custom_trainer
+}  // namespace paddle
--- a/paddle/fluid/train/custom_trainer/feed/dataset/dataset_container.cc
+++ b/paddle/fluid/train/custom_trainer/feed/dataset/dataset_container.cc
+/* DatasetContainer
+ * 保存一个数据源的样本，并驱动样本的异步加载
+ */
+#include <map>
+#include <string>
+#include <vector>
+#include <memory>
+#include <yaml-cpp/yaml.h>
+#include "paddle/fluid/framework/io/shell.h"
+#include "paddle/fluid/string/string_helper.h"
+#include "paddle/fluid/train/custom_trainer/feed/dataset/dataset_container.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+paddle::framework::Channel<DataItem> DatasetContainer::fetch(int epoch_id) {
+    paddle::framework::Channel<DataItem> result;
+    if (_ready_epoch_id < epoch_id) {
+        return result;
+    }
+    _current_epoch_id = epoch_id;
+    _current_dataset_idx = epoch_id % _prefetch_num;
+    //result = _dataset_list[_current_dataset_idx].fetch();
+    //_dataset_list[_current_dataset_idx].reset((decltype(result.get())*)NULL);
+    return result;
+}  
+void DatasetContainer::async_download_data() {
+    while (true) {
+            //do download
+        sleep(30);
+    }
+}
+}//namespace feed
+}//namespace custom_trainer
+}//namespace paddle
--- a/paddle/fluid/train/custom_trainer/feed/dataset/dataset_container.h
+++ b/paddle/fluid/train/custom_trainer/feed/dataset/dataset_container.h
+/* DatasetContainer
+ * 保存一个数据源的样本，并驱动样本的异步加载
+ */
+#pragma once
+#include <map>
+#include <string>
+#include <vector>
+#include <memory>
+#include <yaml-cpp/yaml.h>
+#include "paddle/fluid/framework/channel.h"
+#include "paddle/fluid/framework/data_set.h"
+#include "paddle/fluid/train/custom_trainer/feed/common/runtime_environment.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+//单条样本的原始数据
+class DataItem {
+public:
+    DataItem() {}
+    virtual ~DataItem() {}
+    std::string id;  //样本id标识，可用于shuffle
+    std::string data;//样本完整数据
+};
+class DatasetContainer {
+public:
+    DatasetContainer() {}
+    virtual ~DatasetContainer() {}
+    virtual int initialize(const YAML::Node& config) {
+        _dataset_config = config;
+        _prefetch_num = config["prefetch_num"].as<int>();
+        _data_root_path = config["root_path"].as<std::string>();
+        _data_path_generater = config["_data_path_generater"].as<std::string>();
+        return 0;
+    }  
+    virtual void run();
+    //获取特定epoch_i样本，如果数据未ready，Channel内为空指针
+    virtual ::paddle::framework::Channel<DataItem> fetch(int epoch_id);
+    //触发可预取的数据判断
+    virtual void pre_detect_data(RuntimeEnvironment* env);
+protected:
+    //异步样本download
+    virtual void async_download_data();
+    virtual void download(int epoch_id, const std::vector<std::string>& paths);
+    int _prefetch_num = 0;
+    YAML::Node _dataset_config;
+    std::string _data_root_path;
+    std::string _data_path_generater;
+    uint32_t _current_dataset_idx;             //当前样本数据idx
+    int _current_epoch_id = -1;  
+    int _ready_epoch_id = -1; //已下载完成的epoch_id
+    std::vector<std::shared_ptr<::paddle::framework::Dataset>> _dataset_list;
+};
+}//namespace feed
+}//namespace custom_trainer
+}//namespace paddle
--- a/paddle/fluid/train/custom_trainer/feed/main.cc
+++ b/paddle/fluid/train/custom_trainer/feed/main.cc
+#include <time.h>
+#include <fstream>
+#include <yaml-cpp/yaml.h>
+#include "paddle/fluid/platform/init.h"
+#include "paddle/fluid/train/custom_trainer/feed/trainer_context.h"
+#include "paddle/fluid/train/custom_trainer/feed/process/process.h"
+#include "paddle/fluid/train/custom_trainer/feed/process/init_env_process.h"
+using namespace paddle::custom_trainer::feed;
+DEFINE_string(feed_trainer_conf_path, "./conf/trainer.yaml", "path of trainer conf");
+int main(int argc, char* argv[]) {
+    //gflags
+    google::ParseCommandLineFlags(&argc, &argv, true);
+    std::string gflag_conf = "./conf/gflags.conf";
+    google::SetCommandLineOption("flagfile", gflag_conf.c_str()); 
+    //load trainer config
+    auto trainer_context_ptr = std::make_shared<TrainerContext>();
+    trainer_context_ptr->trainer_config = YAML::LoadFile(FLAGS_feed_trainer_conf_path);    
+    std::vector<std::string> process_name_list = {
+        "InitEnvProcess"
+    };
+    InitEnvProcess init_process;
+    init_process.run();
+    for (const auto& process_name : process_name_list) {
+        Process* process = CREATE_CLASS(Process, process_name);
+        if (process == NULL) {
+            VLOG(1) << "Process:" << process_name << " does not exist"; 
+            return -1;
+        }
+        if (process->initialize(trainer_context_ptr) != 0) {
+            VLOG(1) << "Process:" << process_name << " initialize failed"; 
+            return -1;
+        }
+        trainer_context_ptr->process_list.push_back(std::shared_ptr<Process>(process));
+    } 
+    for (auto& process : trainer_context_ptr->process_list) {
+        process->run();
+    }
+    return 0;
+}
--- a/paddle/fluid/train/custom_trainer/feed/process/CMakeLists.txt
+++ b/paddle/fluid/train/custom_trainer/feed/process/CMakeLists.txt
+cc_library(custom_trainer_process SRCS process.cc init_env_process.cc DEPS memory)
--- a/paddle/fluid/train/custom_trainer/feed/process/data_set_process.h
+++ b/paddle/fluid/train/custom_trainer/feed/process/data_set_process.h
+/*
+ *Author: xiexionghang
+ *组织训练样本的读取工作
+ */
+#pragma once
+#include "paddle/fluid/train/custom_trainer/feed/process/process.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+class DatasetProcess : public Process {
+public:
+    DatasetProcess() {}
+    virtual ~DatasetProcess() {}
+    virtual int initialize(std::shared_ptr<TrainerContext> context_ptr);
+private:
+    std::map<std::string, DatasetContainer> _dataset_map;
+};
+}  // namespace feed
+}  // namespace custom_trainer
+}  // namespace paddle
--- a/paddle/fluid/train/custom_trainer/feed/process/init_env_process.cc
+++ b/paddle/fluid/train/custom_trainer/feed/process/init_env_process.cc
+/*
+ *Author: xiexionghang
+ *用于训练环境的整体配置读取、环境初始化工作
+ */
+#include "paddle/fluid/platform/init.h"
+#include "paddle/fluid/platform/place.h"
+#include "paddle/fluid/framework/executor.h"
+#include "paddle/fluid/train/custom_trainer/feed/process/init_env_process.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+int InitEnvProcess::initialize(std::shared_ptr<TrainerContext> context_ptr) {
+    paddle::framework::InitDevices(false);
+    context_ptr->cpu_place = paddle::platform::CPUPlace();
+    VLOG(3) << "Env initialize success"; 
+    return 0;
+}
+}  // namespace feed
+}  // namespace custom_trainer
+}  // namespace paddle
--- a/paddle/fluid/train/custom_trainer/feed/process/init_env_process.h
+++ b/paddle/fluid/train/custom_trainer/feed/process/init_env_process.h
+/*
+ *Author: xiexionghang
+ *用于训练环境的整体配置读取、环境初始化工作
+ */
+#pragma once
+#include "paddle/fluid/train/custom_trainer/feed/process/process.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+class InitEnvProcess : public Process {
+public:
+    InitEnvProcess() {}
+    virtual ~InitEnvProcess() {}
+    virtual int initialize(std::shared_ptr<TrainerContext> context_ptr);
+};
+}  // namespace feed
+}  // namespace custom_trainer
+}  // namespace paddle
--- a/paddle/fluid/train/custom_trainer/feed/process/process.cc
+++ b/paddle/fluid/train/custom_trainer/feed/process/process.cc
+#include "paddle/fluid/train/custom_trainer/feed/process/process.h"
+#include "paddle/fluid/train/custom_trainer/feed/process/init_env_process.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+REGISTER_CLASS(Process, InitEnvProcess);
+int Process::run() {
+    return 0;
+}
+}  // namespace feed
+}  // namespace custom_trainer
+}  // namespace paddle
--- a/paddle/fluid/train/custom_trainer/feed/process/process.h
+++ b/paddle/fluid/train/custom_trainer/feed/process/process.h
+#pragma once
+#include "paddle/fluid/train/custom_trainer/feed/common/registerer.h"
+#include "paddle/fluid/train/custom_trainer/feed/trainer_context.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+class Process {
+public:
+    Process() {}
+    virtual ~Process() {}
+    virtual int initialize(std::shared_ptr<TrainerContext> context_ptr) = 0;
+    virtual int run();
+};
+REGISTER_REGISTERER(Process);
+}  // namespace feed
+}  // namespace custom_trainer
+}  // namespace paddle
--- a/paddle/fluid/train/custom_trainer/feed/scripts/start_feed_trainer.sh
+++ b/paddle/fluid/train/custom_trainer/feed/scripts/start_feed_trainer.sh
+#!bash
+export LD_LIBRARY_PATH=LD_LIBRARY_PATH:./so
+./bin/feed_trainer
--- a/paddle/fluid/train/custom_trainer/feed/temp/feed_trainer.cpp
+++ b/paddle/fluid/train/custom_trainer/feed/temp/feed_trainer.cpp
+//   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.
+#include <time.h>
+#include <fstream>
+#include "paddle/fluid/framework/executor.h"
+#include "paddle/fluid/framework/op_registry.h"
+#include "paddle/fluid/framework/program_desc.h"
+#include "paddle/fluid/framework/tensor_util.h"
+#include "paddle/fluid/platform/device_context.h"
+#include "paddle/fluid/platform/init.h"
+#include "paddle/fluid/platform/place.h"
+#include "paddle/fluid/platform/profiler.h"
+namespace paddle {
+namespace train {
+void ReadBinaryFile(const std::string& filename, std::string* contents) {
+  std::ifstream fin(filename, std::ios::in | std::ios::binary);
+  PADDLE_ENFORCE(static_cast<bool>(fin), "Cannot open file %s", filename);
+  fin.seekg(0, std::ios::end);
+  contents->clear();
+  contents->resize(fin.tellg());
+  fin.seekg(0, std::ios::beg);
+  fin.read(&(contents->at(0)), contents->size());
+  fin.close();
+}
+std::unique_ptr<paddle::framework::ProgramDesc> Load(
+    paddle::framework::Executor* executor, const std::string& model_filename) {
+  VLOG(3) << "loading model from " << model_filename;
+  std::string program_desc_str;
+  ReadBinaryFile(model_filename, &program_desc_str);
+  std::unique_ptr<paddle::framework::ProgramDesc> main_program(
+      new paddle::framework::ProgramDesc(program_desc_str));
+  return main_program;
+}
+}  // namespace train
+}  // namespace paddle
+int main() {
+  paddle::framework::InitDevices(false);
+  const auto cpu_place = paddle::platform::CPUPlace();
+  paddle::framework::Executor executor(cpu_place);
+  paddle::framework::Scope scope;
+  auto startup_program = paddle::train::Load(&executor, "startup_program");
+  auto train_program = paddle::train::Load(&executor, "main_program");
+  std::string loss_name = "";
+  for (auto op_desc : train_program->Block(0).AllOps()) {
+    if (op_desc->Type() == "mean") {
+      loss_name = op_desc->Output("Out")[0];
+      break;
+    }
+  }
+  PADDLE_ENFORCE_NE(loss_name, "", "loss not found");
+  // init all parameters
+  executor.Run(*startup_program, &scope, 0);
+  // prepare data
+  auto x_var = scope.Var("x");
+  auto x_tensor = x_var->GetMutable<paddle::framework::LoDTensor>();
+  x_tensor->Resize({2, 13});
+  auto x_data = x_tensor->mutable_data<float>(cpu_place);
+  for (int i = 0; i < 2 * 13; ++i) {
+    x_data[i] = static_cast<float>(i);
+  }
+  auto y_var = scope.Var("y");
+  auto y_tensor = y_var->GetMutable<paddle::framework::LoDTensor>();
+  y_tensor->Resize({2, 1});
+  auto y_data = y_tensor->mutable_data<float>(cpu_place);
+  for (int i = 0; i < 2 * 1; ++i) {
+    y_data[i] = static_cast<float>(i);
+  }
+  auto loss_var = scope.Var(loss_name);
+  paddle::platform::ProfilerState pf_state;
+  pf_state = paddle::platform::ProfilerState::kCPU;
+  paddle::platform::EnableProfiler(pf_state);
+  clock_t t1 = clock();
+  for (int i = 0; i < 10; ++i) {
+    executor.Run(*train_program, &scope, 0, false, true);
+    std::cout << "step: " << i << " loss: "
+              << loss_var->Get<paddle::framework::LoDTensor>().data<float>()[0]
+              << std::endl;
+  }
+  clock_t t2 = clock();
+  paddle::platform::DisableProfiler(paddle::platform::EventSortingKey::kTotal,
+                                    "run_paddle_op_profiler");
+  std::cout << "run_time = " << t2 - t1 << std::endl;
+  return 0;
+}
--- a/paddle/fluid/train/custom_trainer/feed/trainer_context.h
+++ b/paddle/fluid/train/custom_trainer/feed/trainer_context.h
+#pragma once
+#include <string>
+#include <memory>
+#include <vector>
+#include <yaml-cpp/yaml.h>
+#include "paddle/fluid/platform/place.h"
+#include "paddle/fluid/train/custom_trainer/feed/common/runtime_environment.h"
+namespace paddle {
+namespace custom_trainer {
+namespace feed {
+class Process;
+class TrainerContext {
+public:
+YAML::Node trainer_config;
+paddle::platform::CPUPlace cpu_place;
+std::shared_ptr<RuntimeEnvironment> environment;
+std::vector<std::shared_ptr<Process>> process_list;
+};
+}  // namespace feed
+}  // namespace custom_trainer
+}  // namespace paddle
--- a/publish_include.sh
+++ b/publish_include.sh
+#!bash
+OUTPUT_PATH=../../../bc_out/baidu/feed-mlarch/paddle-trainer/output/include/
+INCLUDE_DIR=paddle/fluid/train/custom_trainer/feed/
+SUB_DIR_LIST=(common dataset params_accessor process shuffler)
+rm -rf ${OUTPUT_PATH}/${INCLUDE_DIR}/*
+cp ${INCLUDE_DIR}/*.h ${OUTPUT_PATH}/${INCLUDE_DIR}/
+for sub_name in "${SUB_DIR_LIST[@]}"
+do
+    mkdir ${OUTPUT_PATH}/${INCLUDE_DIR}/${sub_name}
+    cp ${INCLUDE_DIR}/${sub_name}/*.h ${OUTPUT_PATH}/${INCLUDE_DIR}/${sub_name}/
+done
--- a/release.bcloud
+++ b/release.bcloud
+#!/bin/bash
+mkdir -p so
+cp baidu_third-party_mklml/so/* so
+rm -rf baidu_third-party_mklml