Merge branch 'develop' into fc500110-bert_test

test=develop

Author: luotao1

CMakeLists.txt

@@ -276,9 +276,3 @@ add_subdirectory(paddle)
 if(WITH_PYTHON)
     add_subdirectory(python)
 endif()
-
-if(WITH_DOC)
-    find_package(Sphinx REQUIRED)
-    find_python_module(recommonmark REQUIRED)
-    add_subdirectory(doc)
-endif()

Dockerfile

@@ -11,12 +11,10 @@ RUN /bin/bash -c 'if [[ -n ${UBUNTU_MIRROR} ]]; then sed -i 's#http://archive.ub
 # ENV variables
 ARG WITH_GPU
 ARG WITH_AVX
-ARG WITH_DOC
 
 ENV WOBOQ OFF
 ENV WITH_GPU=${WITH_GPU:-ON}
 ENV WITH_AVX=${WITH_AVX:-ON}
-ENV WITH_DOC=${WITH_DOC:-OFF}
 
 ENV HOME /root
 # Add bash enhancements

cmake/FindSphinx.cmake

@@ -388,6 +388,7 @@ function(cc_test TARGET_NAME)
     endif()
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cpu_deterministic=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
+    set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_limit_of_tmp_allocation=4294967296) # 4G
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cudnn_deterministic=true)
     # No unit test should exceed 10 minutes.
     set_tests_properties(${TARGET_NAME} PROPERTIES TIMEOUT 600)
@@ -460,6 +461,7 @@ function(nv_test TARGET_NAME)
     endif()
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cpu_deterministic=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
+    set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_limit_of_tmp_allocation=4294967296) # 4G
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cudnn_deterministic=true)
   endif()
 endfunction(nv_test)
@@ -708,9 +710,10 @@ function(py_test TARGET_NAME)
     set(oneValueArgs "")
     set(multiValueArgs SRCS DEPS ARGS ENVS)
     cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+
     add_test(NAME ${TARGET_NAME}
              COMMAND ${CMAKE_COMMAND} -E env FLAGS_init_allocated_mem=true FLAGS_cudnn_deterministic=true
-             FLAGS_cpu_deterministic=true
+             FLAGS_cpu_deterministic=true FLAGS_limit_of_tmp_allocation=4294967296  # 4G
              PYTHONPATH=${PADDLE_BINARY_DIR}/python ${py_test_ENVS}
              ${PYTHON_EXECUTABLE} -u ${py_test_SRCS} ${py_test_ARGS}
              WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})

cmake/generic.cmake

@@ -67,6 +67,7 @@ paddle.fluid.initializer.BilinearInitializer.__init__ ArgSpec(args=['self'], var
 paddle.fluid.initializer.MSRAInitializer.__init__ ArgSpec(args=['self', 'uniform', 'fan_in', 'seed'], varargs=None, keywords=None, defaults=(True, None, 0))
 paddle.fluid.initializer.force_init_on_cpu ArgSpec(args=[], varargs=None, keywords=None, defaults=None)
 paddle.fluid.initializer.init_on_cpu ArgSpec(args=[], varargs='args', keywords='kwds', defaults=None)
+paddle.fluid.initializer.NumpyArrayInitializer.__init__ ArgSpec(args=['self', 'value'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.fc ArgSpec(args=['input', 'size', 'num_flatten_dims', 'param_attr', 'bias_attr', 'act', 'is_test', 'name'], varargs=None, keywords=None, defaults=(1, None, None, None, False, None))
 paddle.fluid.layers.embedding ArgSpec(args=['input', 'size', 'is_sparse', 'is_distributed', 'padding_idx', 'param_attr', 'dtype'], varargs=None, keywords=None, defaults=(False, False, None, None, 'float32'))
 paddle.fluid.layers.dynamic_lstm ArgSpec(args=['input', 'size', 'h_0', 'c_0', 'param_attr', 'bias_attr', 'use_peepholes', 'is_reverse', 'gate_activation', 'cell_activation', 'candidate_activation', 'dtype', 'name'], varargs=None, keywords=None, defaults=(None, None, None, None, True, False, 'sigmoid', 'tanh', 'tanh', 'float32', None))
@@ -121,7 +122,7 @@ paddle.fluid.layers.transpose ArgSpec(args=['x', 'perm', 'name'], varargs=None,
 paddle.fluid.layers.im2sequence ArgSpec(args=['input', 'filter_size', 'stride', 'padding', 'input_image_size', 'out_stride', 'name'], varargs=None, keywords=None, defaults=(1, 1, 0, None, 1, None))
 paddle.fluid.layers.nce ArgSpec(args=['input', 'label', 'num_total_classes', 'sample_weight', 'param_attr', 'bias_attr', 'num_neg_samples', 'name', 'sampler', 'custom_dist', 'seed', 'is_sparse'], varargs=None, keywords=None, defaults=(None, None, None, None, None, 'uniform', None, 0, False))
 paddle.fluid.layers.hsigmoid ArgSpec(args=['input', 'label', 'num_classes', 'param_attr', 'bias_attr', 'name', 'path_table', 'path_code', 'is_custom', 'is_sparse'], varargs=None, keywords=None, defaults=(None, None, None, None, None, False, False))
-paddle.fluid.layers.beam_search ArgSpec(args=['pre_ids', 'pre_scores', 'ids', 'scores', 'beam_size', 'end_id', 'level', 'name'], varargs=None, keywords=None, defaults=(0, None))
+paddle.fluid.layers.beam_search ArgSpec(args=['pre_ids', 'pre_scores', 'ids', 'scores', 'beam_size', 'end_id', 'level', 'is_accumulated', 'name'], varargs=None, keywords=None, defaults=(0, True, None))
 paddle.fluid.layers.row_conv ArgSpec(args=['input', 'future_context_size', 'param_attr', 'act'], varargs=None, keywords=None, defaults=(None, None))
 paddle.fluid.layers.multiplex ArgSpec(args=['inputs', 'index'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.layers.layer_norm ArgSpec(args=['input', 'scale', 'shift', 'begin_norm_axis', 'epsilon', 'param_attr', 'bias_attr', 'act', 'name'], varargs=None, keywords=None, defaults=(True, True, 1, 1e-05, None, None, None, None))
@@ -212,6 +213,7 @@ paddle.fluid.layers.bilinear_tensor_product ArgSpec(args=['x', 'y', 'size', 'act
 paddle.fluid.layers.merge_selected_rows ArgSpec(args=['x', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.get_tensor_from_selected_rows ArgSpec(args=['x', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.lstm ArgSpec(args=['input', 'init_h', 'init_c', 'max_len', 'hidden_size', 'num_layers', 'dropout_prob', 'is_bidirec', 'is_test', 'name', 'default_initializer', 'seed'], varargs=None, keywords=None, defaults=(0.0, False, False, None, None, -1))
+paddle.fluid.layers.shuffle_channel ArgSpec(args=['x', 'group', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.py_func ArgSpec(args=['func', 'x', 'out', 'backward_func', 'skip_vars_in_backward_input'], varargs=None, keywords=None, defaults=(None, None))
 paddle.fluid.layers.psroi_pool ArgSpec(args=['input', 'rois', 'output_channels', 'spatial_scale', 'pooled_height', 'pooled_width', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.teacher_student_sigmoid_loss ArgSpec(args=['input', 'label', 'soft_max_up_bound', 'soft_max_lower_bound'], varargs=None, keywords=None, defaults=(15.0, -15.0))
@@ -358,6 +360,7 @@ paddle.fluid.contrib.QuantizeTranspiler.__init__ ArgSpec(args=['self', 'weight_b
 paddle.fluid.contrib.QuantizeTranspiler.convert_to_int8 ArgSpec(args=['self', 'program', 'place', 'scope'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.contrib.QuantizeTranspiler.freeze_program ArgSpec(args=['self', 'program', 'place', 'fuse_bn', 'scope'], varargs=None, keywords=None, defaults=(False, None))
 paddle.fluid.contrib.QuantizeTranspiler.training_transpile ArgSpec(args=['self', 'program', 'startup_program'], varargs=None, keywords=None, defaults=(None, None))
+paddle.fluid.contrib.reader.ctr_reader.ctr_reader ArgSpec(args=['feed_dict', 'file_type', 'file_format', 'dense_slot_index', 'sparse_slot_index', 'capacity', 'thread_num', 'batch_size', 'file_list', 'slots', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.contrib.build_compressor ArgSpec(args=['place', 'data_reader', 'data_feeder', 'scope', 'metrics', 'epoch', 'config'], varargs=None, keywords=None, defaults=(None, None, None, None, None, None, None))
 paddle.fluid.contrib.CompressPass.__init__ ArgSpec(args=['self', 'place', 'data_reader', 'data_feeder', 'scope', 'metrics', 'epoch', 'program_exe'], varargs=None, keywords=None, defaults=(None, None, None, None, None, None, None))
 paddle.fluid.contrib.CompressPass.add_strategy ArgSpec(args=['self', 'strategy'], varargs=None, keywords=None, defaults=None)

paddle/fluid/API.spec

@@ -1,4 +1,3 @@
-
 #windows treat symbolic file as a real file, which is different with unix
 #We create a hidden file and compile it instead of origin source file.
 function(windows_symbolic TARGET)
@@ -129,12 +128,6 @@ cc_test(version_test SRCS version_test.cc DEPS version)
 
 cc_library(proto_desc SRCS var_desc.cc op_desc.cc block_desc.cc program_desc.cc DEPS shape_inference op_info operator glog version)
 
-if(WITH_NGRAPH)
-  cc_library(ngraph_bridge SRCS ngraph_bridge.cc DEPS operator framework_proto ngraph)
-  cc_library(ngraph_operator SRCS ngraph_operator.cc DEPS ngraph_bridge operator op_info device_context tensor scope glog
-             shape_inference data_transform lod_tensor profiler)
-endif(WITH_NGRAPH)
-
 cc_library(op_registry SRCS op_registry.cc DEPS op_proto_maker op_info operator glog proto_desc)
 nv_test(op_registry_test SRCS op_registry_test.cc DEPS op_registry)
 
@@ -171,13 +164,12 @@ if(WITH_DISTRIBUTE)
 
    set(DISTRIBUTE_COMPILE_FLAGS "-Wno-non-virtual-dtor -Wno-error=non-virtual-dtor -Wno-error=delete-non-virtual-dtor")
    set_source_files_properties(executor.cc PROPERTIES COMPILE_FLAGS ${DISTRIBUTE_COMPILE_FLAGS})
-
 else()
-  if(WITH_NGRAPH)
-    cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto glog lod_rank_table feed_fetch_method graph_to_program_pass ngraph_operator variable_helper)
-  else(WITH_NGRAPH)
+  if (WITH_NGRAPH)
+    cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto glog lod_rank_table feed_fetch_method graph_to_program_pass variable_helper ngraph_engine)
+  else ()
     cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto glog lod_rank_table feed_fetch_method graph_to_program_pass variable_helper)
-  endif(WITH_NGRAPH)
+  endif()
   cc_test(test_naive_executor SRCS naive_executor_test.cc DEPS naive_executor elementwise_add_op)
 endif()
 
@@ -214,3 +206,24 @@ endif (NOT WIN32)
 
 cc_library(dlpack_tensor SRCS dlpack_tensor.cc DEPS tensor dlpack)
 cc_test(dlpack_tensor_test SRCS dlpack_tensor_test.cc DEPS dlpack_tensor glog)
+
+# Get the current working branch
+execute_process(
+  COMMAND git rev-parse --abbrev-ref HEAD
+  WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}
+  OUTPUT_VARIABLE PADDLE_BRANCH
+  OUTPUT_STRIP_TRAILING_WHITESPACE
+)
+
+# Get the latest abbreviated commit hash of the working branch
+execute_process(
+  COMMAND git log -1 --format=%h
+  WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}
+  OUTPUT_VARIABLE PADDLE_COMMIT
+  OUTPUT_STRIP_TRAILING_WHITESPACE
+)
+
+message(STATUS "commit: ${PADDLE_COMMIT}")
+message(STATUS "branch: ${PADDLE_BRANCH}")
+
+configure_file(commit.h.in commit.h)

paddle/fluid/framework/CMakeLists.txt

@@ -0,0 +1,21 @@
+#pragma once
+
+#include <string>
+
+namespace paddle {
+namespace framework {
+
+static std::string paddle_commit() {
+  return "@PADDLE_COMMIT@";
+}
+
+static std::string paddle_compile_branch() {
+  return "@PADDLE_BRANCH@";
+}
+
+static std::string paddle_version() {
+  return "@PADDLE_VERSION@";
+}
+
+}  // namespace framework
+}  // namespace paddle

paddle/fluid/framework/commit.h.in

@@ -91,7 +91,7 @@ struct BuildStrategy {
   int num_trainers_{1};
   int trainer_id_{0};
   std::vector<std::string> trainers_endpoints_;
-  bool remove_unnecessary_lock_{false};
+  bool remove_unnecessary_lock_{true};
 
   // NOTE:
   // Before you add new options, think if it's a general strategy that works

paddle/fluid/framework/details/build_strategy.h

@@ -25,6 +25,9 @@ struct ExecutionStrategy {
   size_t num_threads_{0};
   bool use_cuda_{true};
   bool allow_op_delay_{false};
+  // If we set this to 1, we will delete all variables when finish a batch. and
+  // this will loss 15%+ performance.
+  // Please be aware about this parameters.
   size_t num_iteration_per_drop_scope_{1};
   ExecutorType type_{kDefault};
   bool dry_run_{false};

paddle/fluid/framework/details/execution_strategy.h

@@ -27,7 +27,7 @@ limitations under the License. */
 #include "paddle/fluid/platform/profiler.h"
 
 #ifdef PADDLE_WITH_NGRAPH
-#include "paddle/fluid/framework/ngraph_operator.h"
+#include "paddle/fluid/operators/ngraph/ngraph_engine.h"
 #endif
 
 DECLARE_bool(benchmark);
@@ -133,24 +133,6 @@ static void DeleteUnusedTensors(
   }
 }
 
-static void EnableFusedOp(ExecutorPrepareContext* ctx) {
-#ifdef PADDLE_WITH_NGRAPH
-  VLOG(3) << "use_ngraph=True";
-  auto intervals = NgraphOperator::NgraphOpIntervals(&ctx->ops_);
-  for (auto& interval : intervals) {
-    auto* ng_op = new NgraphOperator(ctx->prog_, ctx->block_id_, interval.at(0),
-                                     interval.at(1));
-    *interval[0] = std::unique_ptr<OperatorBase>(ng_op);
-  }
-  for (auto it = intervals.rbegin(); it != intervals.rend(); ++it) {
-    ctx->ops_.erase(it->at(0) + 1, it->at(1));
-  }
-#else
-  LOG(WARNING)
-      << "'NGRAPH' is not supported, Please re-compile with WITH_NGRAPH option";
-#endif
-}
-
 Executor::Executor(const platform::Place& place) : place_(place) {}
 
 void Executor::Close() {
@@ -204,6 +186,9 @@ void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
                    bool create_local_scope, bool create_vars) {
   platform::RecordBlock b(block_id);
   if (FLAGS_use_mkldnn) EnableMKLDNN(pdesc);
+#ifdef PADDLE_WITH_NGRAPH
+  if (FLAGS_use_ngraph) operators::NgraphEngine::EnableNgraph(pdesc);
+#endif
   auto ctx = Prepare(pdesc, block_id);
   RunPreparedContext(ctx.get(), scope, create_local_scope, create_vars);
 }
@@ -379,7 +364,6 @@ std::unique_ptr<ExecutorPrepareContext> Executor::Prepare(
   for (auto& op_desc : block.AllOps()) {
     ctx->ops_.push_back(OpRegistry::CreateOp(*op_desc));
   }
-  if (FLAGS_use_ngraph) EnableFusedOp(ctx.get());
   return ctx;
 }