Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into add-async-ssa-graph-executor

test=develop

paddle/fluid/API.spec

@@ -427,7 +427,7 @@ paddle.fluid.optimizer.MomentumOptimizer.__init__ ArgSpec(args=['self', 'learnin
 paddle.fluid.optimizer.MomentumOptimizer.apply_gradients ArgSpec(args=['self', 'params_grads'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.optimizer.MomentumOptimizer.backward ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set', 'callbacks'], varargs=None, keywords=None, defaults=(None, None, None, None))
 paddle.fluid.optimizer.MomentumOptimizer.minimize ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None, None))
-paddle.fluid.optimizer.AdagradOptimizer.__init__ ArgSpec(args=['self', 'learning_rate', 'epsilon', 'regularization', 'name'], varargs=None, keywords=None, defaults=(1e-06, None, None))
+paddle.fluid.optimizer.AdagradOptimizer.__init__ ArgSpec(args=['self', 'learning_rate', 'epsilon', 'regularization', 'name', 'initial_accumulator_value'], varargs=None, keywords=None, defaults=(1e-06, None, None, 0.0))
 paddle.fluid.optimizer.AdagradOptimizer.apply_gradients ArgSpec(args=['self', 'params_grads'], varargs=None, keywords=None, defaults=None)
 paddle.fluid.optimizer.AdagradOptimizer.backward ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set', 'callbacks'], varargs=None, keywords=None, defaults=(None, None, None, None))
 paddle.fluid.optimizer.AdagradOptimizer.minimize ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None, None))

paddle/fluid/framework/details/all_reduce_deps_pass.cc

@@ -30,8 +30,6 @@ namespace paddle {
 namespace framework {
 namespace details {
 
-static constexpr char kAllOpDescs[] = "all_op_descs";
-
 VarHandle* GetValidInput(const OpHandleBase* a) {
   for (auto p : a->Inputs()) {
     VarHandle* b = dynamic_cast<VarHandle*>(p);

paddle/fluid/framework/details/all_reduce_op_handle.cc

@@ -53,7 +53,7 @@ AllReduceOpHandle::AllReduceOpHandle(ir::Node *node,
 #endif
 
 void AllReduceOpHandle::RunImpl() {
-  platform::RecordEvent record_event(Name(), dev_ctxes_.cbegin()->second);
+  platform::RecordEvent record_event(Name());
 
   WaitInputVarGenerated();
   auto in_var_handles = DynamicCast<VarHandle>(this->Inputs());

paddle/fluid/framework/details/broadcast_op_handle.cc

@@ -22,7 +22,7 @@ namespace framework {
 namespace details {
 
 void BroadcastOpHandle::RunImpl() {
-  platform::RecordEvent record_event(Name(), dev_ctxes_.begin()->second);
+  platform::RecordEvent record_event(Name());
 
   if (places_.size() == 1) return;
 

paddle/fluid/framework/details/build_strategy.cc

@@ -34,9 +34,11 @@ namespace details {
 static inline bool SeqOnlyAllReduceOps(const BuildStrategy &strategy) {
   // Should fix the allreduce op order if scheduling
   // them in multiple threads or processes to avoid hang.
+  // NOTE: ParallelGraph would execute this pass on each graph, so
+  // don't need to append it here.
   return (!strategy.enable_sequential_execution_ &&
-          strategy.num_trainers_ > 1) ||
-         strategy.enable_parallel_graph_;
+          strategy.num_trainers_ > 1) &&
+         !strategy.enable_parallel_graph_;
 }
 
 class ParallelExecutorPassBuilder : public ir::PassBuilder {

paddle/fluid/framework/details/fused_broadcast_op_handle.cc

@@ -22,7 +22,7 @@ namespace framework {
 namespace details {
 
 void FusedBroadcastOpHandle::RunImpl() {
-  platform::RecordEvent record_event(Name(), dev_ctxes_.begin()->second);
+  platform::RecordEvent record_event(Name());
 
   if (places_.size() == 1UL) return;
 

paddle/fluid/framework/details/memory_optimize_helper.h

@@ -29,8 +29,6 @@ namespace paddle {
 namespace framework {
 namespace details {
 
-constexpr char kAllOpDescs[] = "all_op_descs";
-
 std::vector<ir::Node*> SortOpLikeDescOrder(const ir::Graph& graph);
 
 // NOTE(dzh): A ordered set for node reuse in memory optimize.

paddle/fluid/framework/details/multi_devices_graph_pass.cc

@@ -412,20 +412,32 @@ void MultiDevSSAGraphBuilderBase::CreateComputationalOp(ir::Graph *result,
 
 void MultiDevSSAGraphBuilderBase::CreateAllReduceOp(
     ir::Graph *result, const std::string &og) const {
+  OpHandleBase *op_handle = nullptr;
+
+  auto append_allreduce_op = [&](
+      const std::vector<Scope *> &scopes,
+      const std::vector<platform::Place> &places) -> OpHandleBase * {
 #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
-  result->Get<GraphOps>(kGraphOps).emplace_back(new AllReduceOpHandle(
-      result->CreateEmptyNode("allreduce", ir::Node::Type::kOperation),
-      local_scopes_, places_, nccl_ctxs_));
+    result->Get<GraphOps>(kGraphOps).emplace_back(new AllReduceOpHandle(
+        result->CreateEmptyNode("allreduce", ir::Node::Type::kOperation),
+        scopes, places, nccl_ctxs_));
 #else
-  result->Get<GraphOps>(kGraphOps).emplace_back(new AllReduceOpHandle(
-      result->CreateEmptyNode("allreduce", ir::Node::Type::kOperation),
-      local_scopes_, places_));
+    result->Get<GraphOps>(kGraphOps).emplace_back(new AllReduceOpHandle(
+        result->CreateEmptyNode("allreduce", ir::Node::Type::kOperation),
+        scopes, places));
 #endif
-  auto *op_handle = result->Get<GraphOps>(kGraphOps).back();
+    return result->Get<GraphOps>(kGraphOps).back();
+  };
+
+  if (!strategy_.enable_parallel_graph_)
+    op_handle = append_allreduce_op(local_scopes_, places_);
 
   for (size_t i = 0; i < places_.size(); ++i) {
-    auto &p = places_[i];
-    SetCommunicationContext(op_handle, p);
+    if (strategy_.enable_parallel_graph_) {
+      op_handle = append_allreduce_op({local_scopes_[i]}, {places_[i]});
+    }
+
+    SetCommunicationContext(op_handle, places_[i]);
     auto &vars = result->Get<GraphVars>(kGraphVars)[i][og];
     PADDLE_ENFORCE(!vars.empty());
     auto &prev_grad = vars.back();
@@ -433,7 +445,7 @@ void MultiDevSSAGraphBuilderBase::CreateAllReduceOp(
 
     auto var =
         new VarHandle(result->CreateEmptyNode(og, ir::Node::Type::kVariable),
-                      vars.size(), i, og, p);
+                      vars.size(), i, og, places_[i]);
     vars.emplace_back(var);
     op_handle->AddOutput(var);
   }

paddle/fluid/framework/details/multi_devices_helper.h

@@ -36,13 +36,14 @@ namespace details {
 // map from variable name to variables. The variables, who have the same name,
 // will have a differsent version. The offset in the
 // `std::vector<VarHandle*>` is the version of varaibles.
-typedef std::vector<std::unordered_map<std::string, std::vector<VarHandle*>>>
+typedef std::vector<std::unordered_map<std::string, std::vector<VarHandle *>>>
     GraphVars;
 const char kGraphVars[] = "vars";
 
 // aux variables to represent dependency. Useful to resolve data hazard.
-typedef std::unordered_set<VarHandleBase*> GraphDepVars;
+typedef std::unordered_set<VarHandleBase *> GraphDepVars;
 const char kGraphDepVars[] = "dep_vars";
+
 }  // namespace details
 }  // namespace framework
 }  // namespace paddle

paddle/fluid/framework/details/op_handle_base.h

@@ -70,6 +70,9 @@ class OpHandleBase {
     auto it = dev_ctxes_.find(place);
     return it != dev_ctxes_.end() ? it->second : nullptr;
   }
+  const std::map<platform::Place, platform::DeviceContext *> &DeviceContext() {
+    return dev_ctxes_;
+  }
 
   void SetDeviceContext(platform::Place place, platform::DeviceContext *ctx_) {
     dev_ctxes_[place] = ctx_;

paddle/fluid/framework/details/parallel_ssa_graph_executor.cc

@@ -13,22 +13,92 @@
 // limitations under the License.
 
 #include "paddle/fluid/framework/details/parallel_ssa_graph_executor.h"
+#include "paddle/fluid/framework/ir/graph_helper.h"
 
 namespace paddle {
 namespace framework {
 namespace details {
 
+std::vector<std::unique_ptr<ir::Graph>>
+ParallelSSAGraphExecutor::SeparateMultiDevicesGraph(
+    std::unique_ptr<ir::Graph> &&graph) {
+  std::vector<std::unique_ptr<ir::Graph>> graphs;
+  graphs.reserve(places_.size());
+  for (size_t i = 0; i < places_.size(); ++i) {
+    ProgramDesc empty;
+    graphs.emplace_back(std::unique_ptr<ir::Graph>(new ir::Graph(empty)));
+    auto &g = graphs.back();
+    g->Set(kGraphVars, new GraphVars(1UL));
+    g->Set(kGraphDepVars, new GraphDepVars);
+  }
+  auto op_handles = ir::FilterByNodeWrapper<OpHandleBase>(*graph);
+
+  for (auto &op : op_handles) {
+    auto &dev_ctx = op->DeviceContext();
+    auto &p = dev_ctx.begin()->first;
+    int dev_id = boost::get<platform::CUDAPlace>(p).device;
+    auto &dev_dummys = graphs[dev_id]->Get<GraphDepVars>(kGraphDepVars);
+    graphs[dev_id]->AddNode(graph->RemoveNode(op->Node()).release());
+
+    for (auto &var : op->Inputs()) {
+      auto dummy_ptr = dynamic_cast<DummyVarHandle *>(var);
+      if (dummy_ptr) {
+        dev_dummys.insert(var);
+        if (graph->Nodes().count(var->Node()))
+          graphs[dev_id]->AddNode(graph->RemoveNode(var->Node()).release());
+      }
+    }
+    for (auto &var : op->Outputs()) {
+      auto dummy_ptr = dynamic_cast<DummyVarHandle *>(var);
+      if (dummy_ptr) {
+        dev_dummys.insert(var);
+        if (graph->Nodes().count(var->Node()))
+          graphs[dev_id]->AddNode(graph->RemoveNode(var->Node()).release());
+      }
+    }
+  }
+
+  for (size_t dev_id = 0; dev_id < places_.size(); ++dev_id) {
+    auto &dev_vars = graphs[dev_id]->Get<GraphVars>(kGraphVars)[0];
+    auto &origin_vars = graph->Get<GraphVars>(kGraphVars)[dev_id];
+    for (auto &name_pair : origin_vars) {
+      dev_vars.emplace(name_pair.first, name_pair.second);
+      for (auto &version_pair : name_pair.second) {
+        if (graph->Nodes().count(version_pair->Node())) {
+          graphs[dev_id]->AddNode(
+              graph->RemoveNode(version_pair->Node()).release());
+        }
+      }
+    }
+  }
+
+  return graphs;
+}
+
 ParallelSSAGraphExecutor::ParallelSSAGraphExecutor(
     const ExecutionStrategy &strategy, const std::vector<Scope *> &local_scopes,
     const std::vector<platform::Place> &places,
-    std::vector<std::unique_ptr<ir::Graph>> &&graphs)
+    const framework::ProgramDesc &main_prog, std::unique_ptr<ir::Graph> &&graph)
     : strategy_(std::move(strategy)),
       local_scopes_(std::move(local_scopes)),
       pool_(places.size() >= 2 ? new ::ThreadPool(places.size()) : nullptr),
       places_(std::move(places)),
-      graphs_(std::move(graphs)) {
+      main_prog_(main_prog),
+      // TODO(Yancey1989): Copying graphs is not safely since it deleted the
+      // attrs.
+      graphs_(SeparateMultiDevicesGraph(std::move(graph))) {
   PADDLE_ENFORCE_EQ(places_.size(), local_scopes_.size());
 
+  auto seq_allreduce_pass =
+      ir::PassRegistry::Instance().Get("all_reduce_deps_pass");
+  seq_allreduce_pass->Erase(details::kAllOpDescs);
+  seq_allreduce_pass->Set<const std::vector<OpDesc *>>(
+      details::kAllOpDescs,
+      new std::vector<OpDesc *>(main_prog_.Block(0).AllOps()));
+  for (size_t i = 0; i < graphs_.size(); ++i) {
+    graphs_[i] = seq_allreduce_pass->Apply(std::move(graphs_[i]));
+  }
+
   // set the correct size of thread pool to each device.
   strategy_.num_threads_ = strategy_.num_threads_ < places_.size()
                                ? 1UL
@@ -37,7 +107,7 @@ ParallelSSAGraphExecutor::ParallelSSAGraphExecutor(
           << " to run the operators of the graph on each device.";
   for (size_t i = 0; i < places.size(); ++i) {
     executors_.emplace_back(new details::ThreadedSSAGraphExecutor(
-        strategy_, {local_scopes_[i]}, {places_[i]}, std::move(graphs_[i])));
+        strategy_, local_scopes_, {places_[i]}, std::move(graphs_.at(i))));
   }
 }
 

paddle/fluid/framework/details/parallel_ssa_graph_executor.h

@@ -18,7 +18,9 @@
 #include <vector>
 
 #include "ThreadPool.h"
+#include "paddle/fluid/framework/details/multi_devices_helper.h"
 #include "paddle/fluid/framework/details/threaded_ssa_graph_executor.h"
+#include "paddle/fluid/framework/ir/graph.h"
 
 namespace paddle {
 namespace framework {
@@ -29,17 +31,23 @@ class ParallelSSAGraphExecutor : public SSAGraphExecutor {
   ParallelSSAGraphExecutor(const ExecutionStrategy &strategy,
                            const std::vector<Scope *> &local_scopes,
                            const std::vector<platform::Place> &places,
-                           std::vector<std::unique_ptr<ir::Graph>> &&graphs);
+                           const framework::ProgramDesc &main_prog,
+                           std::unique_ptr<ir::Graph> &&graph);
   ~ParallelSSAGraphExecutor() final = default;
+
   const ir::Graph &Graph() const override { return *graphs_[0]; }
 
   FeedFetchList Run(const std::vector<std::string> &fetch_tensors) override;
 
  private:
+  std::vector<std::unique_ptr<ir::Graph>> SeparateMultiDevicesGraph(
+      std::unique_ptr<ir::Graph> &&graph);
+
   ExecutionStrategy strategy_;
   std::vector<Scope *> local_scopes_;
   std::unique_ptr<::ThreadPool> pool_{nullptr};
   std::vector<platform::Place> places_;
+  framework::ProgramDesc main_prog_;
   std::vector<std::unique_ptr<ir::Graph>> graphs_;
 
   std::vector<std::unique_ptr<details::ThreadedSSAGraphExecutor>> executors_;

paddle/fluid/framework/details/reduce_op_handle.cc

@@ -139,7 +139,7 @@ void ReduceOpHandle::GatherSelectedRows(
 #endif
 
 void ReduceOpHandle::RunImpl() {
-  platform::RecordEvent record_event(Name(), dev_ctxes_.cbegin()->second);
+  platform::RecordEvent record_event(Name());
 
   if (places_.size() == 1) return;
   // the input and output may have dummy var.

paddle/fluid/framework/details/scope_buffered_ssa_graph_executor.cc

@@ -63,7 +63,7 @@ FeedFetchList ScopeBufferedSSAGraphExecutor::Run(
     eptr = std::current_exception();
   }
 
-  platform::RecordEvent e("ScopeBufferedSSAGraphExecutorAfterRun", nullptr);
+  platform::RecordEvent e("ScopeBufferedSSAGraphExecutorAfterRun");
   ++drop_scope_counter_;
 
   bool stream_end = false;

paddle/fluid/framework/details/threaded_ssa_graph_executor.cc

@@ -50,7 +50,7 @@ ThreadedSSAGraphExecutor::ThreadedSSAGraphExecutor(
 inline FeedFetchList ThreadedSSAGraphExecutor::RunImpl(
     const std::vector<std::string> &fetch_tensors) {
   std::unique_ptr<platform::RecordEvent> event(
-      new platform::RecordEvent("ThreadedSSAGraphExecutorPrepare", nullptr));
+      new platform::RecordEvent("ThreadedSSAGraphExecutorPrepare"));
   std::unordered_map<OpHandleBase *, size_t> pending_ops;
   std::unordered_set<VarHandleBase *> pending_vars;
   auto ready_vars = std::make_shared<BlockingQueue<VarHandleBase *>>();
@@ -240,7 +240,7 @@ void ThreadedSSAGraphExecutor::RunOp(
       VLOG(10) << op << " " << op->Name() << " Done ";
       running_ops_--;
       ready_var_q->Extend(op->Outputs());
-      VLOG(10) << op << " " << op->Name() << "Signal posted";
+      VLOG(10) << op << " " << op->Name() << " Signal posted";
     } catch (...) {
       exception_holder_.Catch(std::current_exception());
     }

paddle/fluid/framework/ir/graph.h

@@ -26,6 +26,14 @@ limitations under the License. */
 
 namespace paddle {
 namespace framework {
+
+namespace details {
+
+// This attr is not recommended, because the graph should not dependence
+// the program once it is built.
+constexpr char kAllOpDescs[] = "all_op_descs";
+}  //  namespace details
+
 namespace ir {
 
 /*
@@ -168,10 +176,13 @@ class Graph {
     return ret;
   }
 
-  void RemoveNode(ir::Node *node) {
+  std::unique_ptr<ir::Node> RemoveNode(ir::Node *node) {
     PADDLE_ENFORCE(node_set_.find(node) != node_set_.end());
-    node_set_.erase(node);
+    std::unique_ptr<ir::Node> ret;
+    ret.reset(nodes_.at(node).release());
     nodes_.erase(node);
+    node_set_.erase(node);
+    return ret;
   }
 
   // NOTE low performance, but simple and secure.
@@ -184,13 +195,6 @@ class Graph {
     return nullptr;
   }
 
-  void ResolveHazard(
-      const std::map<std::string, std::vector<ir::Node *>> &var_nodes);
-
- private:
-  std::map<std::string, std::vector<ir::Node *>> InitFromProgram(
-      const ProgramDesc &program);
-
   // This method takes ownership of `node`.
   ir::Node *AddNode(ir::Node *node) {
     PADDLE_ENFORCE(node_set_.find(node) == node_set_.end());
@@ -199,6 +203,13 @@ class Graph {
     return node;
   }
 
+  void ResolveHazard(
+      const std::map<std::string, std::vector<ir::Node *>> &var_nodes);
+
+ private:
+  std::map<std::string, std::vector<ir::Node *>> InitFromProgram(
+      const ProgramDesc &program);
+
   // NOTE: program_ shouldn't be exposed to user.
   const ProgramDesc program_;
   std::map<std::string, boost::any> attrs_;

paddle/fluid/framework/operator.cc

@@ -177,9 +177,7 @@ void OperatorBase::Run(const Scope& scope, const platform::Place& place) {
     // in concurrency scenerio. Here use an `if` to fix this issue.
     // Please not remove the `if`, ask @Superjomn if there are any concern.
     if (platform::IsProfileEnabled()) {
-      platform::DeviceContextPool& pool =
-          platform::DeviceContextPool::Instance();
-      platform::RecordEvent record_event(Type(), pool.Get(place));
+      platform::RecordEvent record_event(Type());
       RunImpl(scope, place);
     } else {
       RunImpl(scope, place);

paddle/fluid/framework/parallel_executor.cc

@@ -22,6 +22,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/ir/graph.h"
 
 #include "paddle/fluid/framework/details/async_ssa_graph_executor.h"
+#include "paddle/fluid/framework/details/all_reduce_deps_pass.h"
 #include "paddle/fluid/framework/details/fast_threaded_ssa_graph_executor.h"
 #include "paddle/fluid/framework/details/multi_devices_helper.h"
 #include "paddle/fluid/framework/details/parallel_ssa_graph_executor.h"
@@ -194,7 +195,6 @@ ParallelExecutor::ParallelExecutor(
   member_->use_all_reduce_ =
       build_strategy.reduce_ == BuildStrategy::ReduceStrategy::kAllReduce;
   member_->nranks_ = build_strategy.num_trainers_ * places.size();
-
   if (!member_->use_all_reduce_) {
     PADDLE_ENFORCE(places.size() > 1,
                    "If you set build_strategy.reduce with 'Reduce',"
@@ -222,9 +222,10 @@ ParallelExecutor::ParallelExecutor(
   // choice the execution strategy.
   build_strategy.enable_parallel_graph_ =
       EnableParallelGraphExecution(main_program, exec_strategy, build_strategy);
-
-  VLOG(1) << "Enable ParallelGraph Execution: "
-          << build_strategy.enable_parallel_graph_;
+  if (build_strategy.enable_parallel_graph_)
+    VLOG(0) << "The Executor would execute the graph by ParallelGraph "
+               "Execution which can get better performance,"
+            << "you can force it off by env FLAGS_enable_parallel_graph=0";
 
   if (member_->use_cuda_) {
 // Bcast Parameters to all GPUs
@@ -258,22 +259,11 @@ ParallelExecutor::ParallelExecutor(
 
   // Step 2. Convert main_program to SSA form and dependency graph. Also, insert
   // ncclOp
-  std::vector<std::unique_ptr<ir::Graph>> graphs;
+  std::unique_ptr<ir::Graph> graph;
 #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
-  if (build_strategy.enable_parallel_graph_) {
-    for (size_t i = 0; i < member_->places_.size(); ++i) {
-      std::unique_ptr<ir::Graph> graph = build_strategy.Apply(
-          main_program, {member_->places_[i]}, loss_var_name,
-          {member_->local_scopes_[i]}, member_->nranks_, member_->use_cuda_,
-          member_->nccl_ctxs_.get());
-      graphs.push_back(std::move(graph));
-    }
-  } else {
-    std::unique_ptr<ir::Graph> graph = build_strategy.Apply(
-        main_program, member_->places_, loss_var_name, member_->local_scopes_,
-        member_->nranks_, member_->use_cuda_, member_->nccl_ctxs_.get());
-    graphs.push_back(std::move(graph));
-  }
+  graph = build_strategy.Apply(main_program, member_->places_, loss_var_name,
+                               member_->local_scopes_, member_->nranks_,
+                               member_->use_cuda_, member_->nccl_ctxs_.get());
 #else
   if (build_strategy.async_mode_ && !build_strategy.is_distribution_) {
     for (size_t i = 0; i < member_->places_.size(); ++i) {
@@ -293,40 +283,37 @@ ParallelExecutor::ParallelExecutor(
   VLOG(10) << "Eager Deletion Threshold "
            << static_cast<float>(max_memory_size) / (1 << 30);
   if (max_memory_size >= 0) {
-    for (size_t i = 0; i < graphs.size(); ++i) {
-      graphs[i] = member_->PrepareGCAndRefCnts(
-          std::move(graphs[i]), static_cast<size_t>(max_memory_size));
-    }
+    graph = member_->PrepareGCAndRefCnts(std::move(graph),
+                                         static_cast<size_t>(max_memory_size));
   }
 
   // Step 3. Create vars in each scope. Passes may also create new vars.
   //         skip control vars and empty vars
   std::vector<details::VariableInfo> var_infos;
-  for (auto &graph : graphs) {
-    for (auto &node : graph->Nodes()) {
-      if (node->IsVar() && !node->IsCtrlVar() && node->Var()) {
-        var_infos.emplace_back();
-        var_infos.back().name_ = node->Var()->Name();
-        var_infos.back().type_ = node->Var()->GetType();
-        var_infos.back().persistable_ = node->Var()->Persistable();
-      }
+  for (auto &node : graph->Nodes()) {
+    if (node->IsVar() && !node->IsCtrlVar() && node->Var()) {
+      var_infos.emplace_back();
+      var_infos.back().name_ = node->Var()->Name();
+      var_infos.back().type_ = node->Var()->GetType();
+      var_infos.back().persistable_ = node->Var()->Persistable();
     }
   }
 
   // If the loss_var_name is given, the number of graph should be only one.
   if (loss_var_name.size()) {
-    size_t graph_num = ir::GraphNum(*graphs[0]);
+    size_t graph_num = ir::GraphNum(*graph);
     if (graph_num > 1) {
       LOG(WARNING)
           << "The number of graph should be only one, "
              "but the current graph has "
-          << ir::GraphNum(*graphs[0])
+          << ir::GraphNum(*graph)
           << " sub_graphs. If you want to see the nodes of the "
              "sub_graphs, you should use 'FLAGS_print_sub_graph_dir' "
              "to specify the output dir. NOTES: if you not do training, "
              "please don't pass loss_var_name.";
     }
   }
+
   if (build_strategy.async_mode_ && !build_strategy.is_distribution_) {
     VLOG(3) << "use AsyncSSAGraphExecutor";
     member_->executor_.reset(new details::AsyncSSAGraphExecutor(
@@ -334,20 +321,27 @@ ParallelExecutor::ParallelExecutor(
         std::move(graphs)));
   } else if (build_strategy.enable_parallel_graph_) {
     VLOG(3) << "use ParallelSSAGraphExecutor";
+#ifdef PADDLE_WITH_CUDA
+    // TODO(Yancey1989): Remove passing in the main_program when
+    // allreduce_seq_pass doesn't need it as the attr.
     member_->executor_.reset(new details::ParallelSSAGraphExecutor(
-        exec_strategy, member_->local_scopes_, member_->places_,
-        std::move(graphs)));
+        exec_strategy, member_->local_scopes_, member_->places_, main_program,
+        std::move(graph)));
+#else
+    PADDLE_THROW(
+        "Paddle should be compiled with CUDA for ParallelGraph Execution.");
+#endif
   } else {
     if (exec_strategy.type_ == ExecutionStrategy::kDefault) {
       VLOG(3) << "use ThreadedSSAGraphExecutor";
       member_->executor_.reset(new details::ThreadedSSAGraphExecutor(
           exec_strategy, member_->local_scopes_, member_->places_,
-          std::move(graphs[0])));
+          std::move(graph)));
     } else {
       VLOG(3) << "use FastThreadedSSAGraphExecutor";
       member_->executor_.reset(new details::FastThreadedSSAGraphExecutor(
           exec_strategy, member_->local_scopes_, member_->places_,
-          std::move(graphs[0])));
+          std::move(graph)));
     }
   }
 
@@ -509,11 +503,10 @@ bool ParallelExecutor::EnableParallelGraphExecution(
   }
 
   if (!member_->use_all_reduce_ || !member_->use_cuda_)
-    enable_parallel_graph = false;
 
-  if (build_strategy.enable_sequential_execution_ ||
-      exec_strategy.type_ == ExecutionStrategy::ExecutorType::kExperimental)
-    enable_parallel_graph = false;
+    if (build_strategy.enable_sequential_execution_ ||
+        exec_strategy.type_ == ExecutionStrategy::ExecutorType::kExperimental)
+      enable_parallel_graph = false;
   return enable_parallel_graph;
 }
 

paddle/fluid/inference/tests/test_helper.h

@@ -171,9 +171,7 @@ void TestInference(const std::string& dirname,
   // Enable the profiler
   paddle::platform::EnableProfiler(state);
   {
-    paddle::platform::RecordEvent record_event(
-        "init_program",
-        paddle::platform::DeviceContextPool::Instance().Get(place));
+    paddle::platform::RecordEvent record_event("init_program");
     inference_program = InitProgram(&executor, scope, dirname, is_combined);
   }
 
@@ -230,9 +228,7 @@ void TestInference(const std::string& dirname,
 
     // Run repeat times to profile the performance
     for (int i = 0; i < repeat; ++i) {
-      paddle::platform::RecordEvent record_event(
-          "run_inference",
-          paddle::platform::DeviceContextPool::Instance().Get(place));
+      paddle::platform::RecordEvent record_event("run_inference");
 
       if (PrepareContext) {
         // Note: if you change the inference_program, you need to call

paddle/fluid/memory/allocation/legacy_allocator.cc

@@ -356,7 +356,7 @@ void MemInfo::Minus(const size_t &size) {
   usage_ -= size;
 }
 
-uint64_t MemInfo::GetPeakUsage() { return peak_usage_; }
+uint64_t MemInfo::GetPeakUsage() const { return peak_usage_; }
 
 LegacyMemMonitor::~LegacyMemMonitor() {
   for (auto &item : gpu_mem_info_) delete item.second;
@@ -380,10 +380,10 @@ void LegacyMemMonitor::Minus(const int &device, const size_t &size) {
   gpu_mem_info_[device]->Minus(size);
 }
 
-uint64_t LegacyMemMonitor::GetMemUsage(const int &device) {
+uint64_t LegacyMemMonitor::GetMemUsage(const int &device) const {
   return gpu_mem_info_.find(device) == gpu_mem_info_.end()
              ? 0
-             : gpu_mem_info_[device]->GetPeakUsage();
+             : gpu_mem_info_.at(device)->GetPeakUsage();
 }
 
 void LegacyMemMonitor::PrintMemUsage() {

paddle/fluid/memory/allocation/legacy_allocator.h

@@ -27,20 +27,20 @@ namespace allocation {
 class MemInfo {
  public:
   MemInfo() : usage_(0), peak_usage_(0) {}
-  MemInfo(const MemInfo &) = delete;
-  MemInfo &operator=(const MemInfo &) = delete;
 
   // return a flag to indicate current operation will create a peak point or not
   bool Add(const size_t &);
   void Minus(const size_t &);
 
-  uint64_t GetPeakUsage();
+  uint64_t GetPeakUsage() const;
 
  private:
   /* current memory usage*/
   uint64_t usage_;
   uint64_t peak_usage_;
   std::mutex mutex_;
+
+  DISABLE_COPY_AND_ASSIGN(MemInfo);
 };
 
 class LegacyMemMonitor {
@@ -56,11 +56,11 @@ class LegacyMemMonitor {
   void Add(const int &, const size_t &);
   void Minus(const int &, const size_t &);
 
-  uint64_t GetMemUsage(const int &);
+  uint64_t GetMemUsage(const int &) const;
 
   void PrintMemUsage();
 
- protected:
+ private:
   MemUsage gpu_mem_info_;
 };
 

paddle/fluid/operators/distributed/brpc/brpc_client.cc

@@ -80,7 +80,7 @@ VarHandlePtr BRPCClient::AsyncSendVar(const std::string& ep,
     google::protobuf::Closure* done = brpc::NewCallback(
         &HandleSendResponse, cntl, response, var_h, ch_ptr, ch_ctx, this);
 
-    platform::RecordRPCEvent record_event(method, p_ctx);
+    platform::RecordRPCEvent record_event(method);
 
     ch_ctx->stub->SendVariable(cntl, &request, response, done);
 
@@ -184,7 +184,7 @@ VarHandlePtr BRPCClient::_AsyncGetVar(const std::string& ep,
     google::protobuf::Closure* done = brpc::NewCallback(
         &HandleGetResponse, cntl, response, var_h, ch_ptr, ch_ctx, this);
 
-    platform::RecordRPCEvent record_event(method, p_ctx);
+    platform::RecordRPCEvent record_event(method);
 
     if (method_name == kGetMonomerRPC) {
       ch_ctx->stub->GetMonomerVariable(cntl, &req, response, done);
@@ -272,7 +272,7 @@ VarHandlePtr BRPCClient::AsyncPrefetchVar(const std::string& ep,
                                   &cntl->request_attachment(), out_var_name_val,
                                   false, 0, table_name_val);
 
-    platform::RecordRPCEvent record_event(method, p_ctx);
+    platform::RecordRPCEvent record_event(method);
 
     google::protobuf::Closure* done = brpc::NewCallback(
         &HandleGetResponse, cntl, response, var_h, ch_ptr, ch_ctx, this);
@@ -311,7 +311,7 @@ VarHandlePtr BRPCClient::AsyncSendFetchBarrier(const std::string& ep,
   VarHandlePtr var_h(
       new VarHandle(ep, method, FETCH_BARRIER_MESSAGE, nullptr, nullptr));
 
-  platform::RecordRPCEvent record_event(method, nullptr);
+  platform::RecordRPCEvent record_event(method);
 
   google::protobuf::Closure* done = brpc::NewCallback(
       &HandleFetchBarrierResponse, cntl, response, var_h, ch_ptr, ch_ctx, this);
@@ -406,7 +406,7 @@ VarHandlePtr BRPCClient::AsyncSendVarMessage(
   sendrecv::VoidMessage* response = new sendrecv::VoidMessage();
   cntl->set_timeout_ms(time_out);
 
-  platform::RecordRPCEvent record_event(method_name, nullptr);
+  platform::RecordRPCEvent record_event(method_name);
 
   VarHandlePtr var_h(
       new VarHandle(ep, method_name, req.varname(), nullptr, nullptr));

paddle/fluid/operators/distributed/grpc/grpc_client.cc

@@ -89,7 +89,7 @@ VarHandlePtr GRPCClient::AsyncSendVar(const std::string& ep,
     // stub context
     s->response_call_back_ = nullptr;
 
-    platform::RecordRPCEvent record_event(method, p_ctx);
+    platform::RecordRPCEvent record_event(method);
 
     auto call = s->stub_g_.PrepareUnaryCall(
         s->context_.get(), "/sendrecv.SendRecvService/SendVariable", req, &cq_);
@@ -184,7 +184,7 @@ VarHandlePtr GRPCClient::_AsyncGetVar(
         // stub context
         s->response_call_back_ = ProcGetResponse;
 
-        platform::RecordRPCEvent record_event(method, p_ctx);
+        platform::RecordRPCEvent record_event(method);
 
         auto call =
             s->stub_g_.PrepareUnaryCall(s->context_.get(), rpc_path, buf, &cq_);
@@ -235,7 +235,7 @@ VarHandlePtr GRPCClient::AsyncPrefetchVar(const std::string& ep,
     // stub context
     s->response_call_back_ = ProcGetResponse;
 
-    platform::RecordRPCEvent record_event(method, p_ctx);
+    platform::RecordRPCEvent record_event(method);
 
     auto call = s->stub_g_.PrepareUnaryCall(
         s->context_.get(), "/sendrecv.SendRecvService/PrefetchVariable", req,
@@ -265,7 +265,7 @@ VarHandlePtr GRPCClient::AsyncSendBatchBarrier(const std::string& ep,
   sendrecv::VariableMessage req;
   req.set_varname(BATCH_BARRIER_MESSAGE);
 
-  platform::RecordRPCEvent record_event(method, nullptr);
+  platform::RecordRPCEvent record_event(method);
 
   auto rpc = s->stub_->AsyncSendVariable(s->context_.get(), req, &cq_);
   rpc->Finish(&s->reply_, &s->status_, reinterpret_cast<void*>(s));
@@ -290,7 +290,7 @@ VarHandlePtr GRPCClient::AsyncSendFetchBarrier(const std::string& ep,
   sendrecv::VariableMessage req;
   req.set_varname(FETCH_BARRIER_MESSAGE);
 
-  platform::RecordRPCEvent record_event(method, nullptr);
+  platform::RecordRPCEvent record_event(method);
 
   auto rpc = s->stub_->AsyncGetVariable(s->context_.get(), req, &cq_);
   rpc->Finish(&s->reply_, &s->status_, reinterpret_cast<void*>(s));
@@ -317,7 +317,7 @@ VarHandlePtr GRPCClient::AsyncGetMonomerBarrier(const std::string& ep,
   sendrecv::VariableMessage req;
   req.set_varname(var_name);
 
-  platform::RecordRPCEvent record_event(method, nullptr);
+  platform::RecordRPCEvent record_event(method);
 
   auto rpc = s->stub_->AsyncGetMonomerBarrier(s->context_.get(), req, &cq_);
   rpc->Finish(&s->reply_, &s->status_, reinterpret_cast<void*>(s));
@@ -342,7 +342,7 @@ VarHandlePtr GRPCClient::AsyncSendComplete(const std::string& ep,
   sendrecv::VariableMessage req;
   req.set_varname(COMPLETE_MESSAGE);
 
-  platform::RecordRPCEvent record_event(method, nullptr);
+  platform::RecordRPCEvent record_event(method);
 
   auto rpc = s->stub_->AsyncSendVariable(s->context_.get(), req, &cq_);
   rpc->Finish(&s->reply_, &s->status_, reinterpret_cast<void*>(s));
@@ -372,7 +372,7 @@ VarHandlePtr GRPCClient::AsyncCheckpointNotify(const std::string& ep,
   req.set_varname(CHECKPOINT_SAVE_MESSAGE);
   req.set_out_varname(dir);
 
-  platform::RecordRPCEvent record_event(method, nullptr);
+  platform::RecordRPCEvent record_event(method);
 
   auto rpc = s->stub_->AsyncCheckpointNotify(s->context_.get(), req, &cq_);
   rpc->Finish(&s->reply_, &s->status_, reinterpret_cast<void*>(s));

paddle/fluid/operators/distributed/grpc/grpc_serde.cc

@@ -38,7 +38,7 @@ void SerializeToByteBuffer(const std::string& name, framework::Variable* var,
                            ::grpc::ByteBuffer* msg, const std::string& out_name,
                            const int trainer_id,
                            const std::string& table_name) {
-  platform::RecordRPCEvent record_event("serial", &ctx);
+  platform::RecordRPCEvent record_event("serial");
   VarMsg request;
   TensorPayload* payload = nullptr;
 
@@ -147,7 +147,7 @@ void DeserializeFromByteBuffer(const ::grpc::ByteBuffer& msg,
                                const platform::DeviceContext& ctx,
                                const framework::Scope* scope,
                                framework::Variable** var, int* trainer_id) {
-  platform::RecordRPCEvent record_event("deserial", &ctx);
+  platform::RecordRPCEvent record_event("deserial");
   operators::distributed::GRPCVariableResponse resp(scope, &ctx);
   PADDLE_ENFORCE(resp.Parse(msg) == 0, "parse bytebuffer to tensor error!");
   *var = resp.GetVar();

paddle/fluid/operators/reader/read_op.cc

@@ -85,9 +85,7 @@ class ReadOp : public framework::OperatorBase {
     std::vector<framework::LoDTensor> ins;
 
     // For profiling
-    platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
-    auto& ctx = *pool.Get(dev_place);
-    platform::RecordEvent record_event(Type(), &ctx);
+    platform::RecordEvent record_event(Type());
 
     reader->ReadNext(&ins);
     if (ins.empty()) {

paddle/fluid/platform/CMakeLists.txt

@@ -88,7 +88,11 @@ cc_library(timer SRCS timer.cc)
 cc_test(timer_test SRCS timer_test.cc DEPS timer)
 
 cc_library(device_tracer SRCS device_tracer.cc DEPS boost profiler_proto framework_proto ${GPU_CTX_DEPS})
-cc_library(profiler SRCS profiler.cc DEPS device_context device_tracer)
+if(WITH_GPU)
+    nv_library(profiler SRCS profiler.cc profiler.cu DEPS device_context device_tracer)
+else()
+    cc_library(profiler SRCS profiler.cc DEPS device_context device_tracer)
+endif()
 cc_test(profiler_test SRCS profiler_test.cc DEPS profiler)
 
 nv_test(float16_gpu_test SRCS float16_test.cu DEPS lod_tensor)

paddle/fluid/platform/device_tracer.h

@@ -32,6 +32,8 @@ inline uint64_t PosixInNsec() {
   return 1000 * (static_cast<uint64_t>(tv.tv_sec) * 1000000 + tv.tv_usec);
 }
 
+class Event;
+
 // DeviceTracer performs the following tasks:
 // 1. Register cuda callbacks for various events: kernel, memcpy, etc.
 // 2. Collect cuda statistics: start/end ts, memory, etc.
@@ -68,11 +70,13 @@ class DeviceTracer {
   virtual void Enable() = 0;
   // Needs to be called once after use.
   virtual void Disable() = 0;
+  // Needs to be called once before reuse.
+  virtual void Reset() = 0;
 
   // Add a pair to correlate internal cuda id with high level
-  // annotation (string). So cuda statistics can be represented by
+  // annotation event(with string). So cuda statistics can be represented by
   // human-readable annotations.
-  virtual void AddAnnotation(uint64_t id, const std::string& anno) = 0;
+  virtual void AddAnnotation(uint32_t id, Event* event) = 0;
 
   virtual void AddMemRecords(const std::string& name, uint64_t start_ns,
                              uint64_t end_ns, int64_t device_id,
@@ -92,6 +96,9 @@ class DeviceTracer {
   // Generate a proto after done (Disabled).
   virtual proto::Profile GenProfile(const std::string& profile_path) = 0;
 
+  // generate kernel elapsed time into Event
+  virtual void GenEventKernelCudaElapsedTime() = 0;
+
   virtual bool IsEnabled() = 0;
 };
 
@@ -99,14 +106,19 @@ class DeviceTracer {
 DeviceTracer* GetDeviceTracer();
 
 // Set a name for the cuda kernel operation being launched by the thread.
-void SetCurAnnotation(const std::string& anno);
+void SetCurAnnotation(Event* event);
 // Clear the name after the operation is done.
 void ClearCurAnnotation();
 // Current name of the operation being run in the thread.
-std::string CurAnnotation();
+std::string CurAnnotationName();
+Event* CurAnnotation();
 
 void SetCurBlock(int block_id);
 void ClearCurBlock();
 int BlockDepth();
+
+// Set current thread id, so we can map the system thread id to thread id.
+void RecoreCurThreadId(int32_t id);
+int32_t GetThreadIdFromSystemThreadId(uint32_t id);
 }  // namespace platform
 }  // namespace paddle

paddle/fluid/platform/init.cc

@@ -22,6 +22,7 @@ limitations under the License. */
 #include "paddle/fluid/string/split.h"
 #ifdef PADDLE_WITH_CUDA
 #include "paddle/fluid/platform/cuda_device_guard.h"
+#include "paddle/fluid/platform/dynload/cupti.h"
 #endif
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/init.h"
@@ -30,6 +31,9 @@ limitations under the License. */
 
 DEFINE_int32(paddle_num_threads, 1,
              "Number of threads for each paddle instance.");
+DEFINE_int32(multiple_of_cupti_buffer_size, 1,
+             "Multiple of the CUPTI device buffer size. If the timestamps have "
+             "been dropped when you are profiling, try increasing this value.");
 
 namespace paddle {
 namespace framework {
@@ -78,7 +82,32 @@ void InitP2P(std::vector<int> devices) {
 #endif
 }
 
+void InitCupti() {
+#ifdef PADDLE_WITH_CUPTI
+  if (FLAGS_multiple_of_cupti_buffer_size == 1) return;
+  size_t attrValue = 0, attrValueSize = sizeof(size_t);
+#define MULTIPLY_ATTR_VALUE(attr)                                 \
+  {                                                               \
+    PADDLE_ENFORCE(!platform::dynload::cuptiActivityGetAttribute( \
+        attr, &attrValueSize, &attrValue));                       \
+    attrValue *= FLAGS_multiple_of_cupti_buffer_size;             \
+    LOG(WARNING) << "Set " #attr " " << attrValue << " byte";     \
+    PADDLE_ENFORCE(!platform::dynload::cuptiActivitySetAttribute( \
+        attr, &attrValueSize, &attrValue));                       \
+  }
+  MULTIPLY_ATTR_VALUE(CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_SIZE);
+  MULTIPLY_ATTR_VALUE(CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_SIZE_CDP);
+#if CUDA_VERSION >= 9000
+  MULTIPLY_ATTR_VALUE(CUPTI_ACTIVITY_ATTR_PROFILING_SEMAPHORE_POOL_SIZE);
+#endif
+#undef MULTIPLY_ATTR_VALUE
+#endif
+}
+
 void InitDevices(bool init_p2p) {
+  // CUPTI attribute should be set before any CUDA context is created (see CUPTI
+  // documentation about CUpti_ActivityAttribute).
+  InitCupti();
   /*Init all available devices by default */
   std::vector<int> devices;
 #ifdef PADDLE_WITH_CUDA

paddle/fluid/platform/profiler.cc

@@ -12,6 +12,8 @@ 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 "paddle/fluid/platform/profiler.h"
+
 #include <algorithm>
 #include <iomanip>
 #include <limits>
@@ -27,7 +29,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/block_desc.h"
 #include "paddle/fluid/platform/device_tracer.h"
 #include "paddle/fluid/platform/port.h"
-#include "paddle/fluid/platform/profiler.h"
 #include "paddle/fluid/string/printf.h"
 
 DEFINE_bool(enable_rpc_profiler, false, "Enable rpc profiler or not.");
@@ -66,12 +67,13 @@ struct EventList {
       ((kEventSize + kEventAlign - 1) / kEventAlign * kEventAlign);
 
   template <typename... Args>
-  void Record(Args&&... args) {
+  Event* Record(Args&&... args) {
     if (event_blocks.empty() || event_blocks.front().size() == kNumBlock) {
       event_blocks.emplace_front();
       event_blocks.front().reserve(kNumBlock);
     }
     event_blocks.front().emplace_back(std::forward<Args>(args)...);
+    return &event_blocks.front().back();
   }
 
   std::vector<Event> Reduce() {
@@ -98,21 +100,8 @@ inline uint64_t GetTimeInNsec() {
       .count();
 }
 
-Event::Event(EventType type, std::string name, uint32_t thread_id,
-             const DeviceContext* dev_ctx)
-    : type_(type), name_(name), thread_id_(thread_id), has_cuda_(false) {
-#ifdef PADDLE_WITH_CUDA
-  has_cuda_ = dev_ctx ? platform::is_gpu_place(dev_ctx->GetPlace()) : false;
-  if (has_cuda_) {
-    auto* cuda_dev_ctx = static_cast<const CUDADeviceContext*>(dev_ctx);
-    PADDLE_ENFORCE(cudaSetDevice(
-        boost::get<platform::CUDAPlace>(cuda_dev_ctx->GetPlace()).device));
-    PADDLE_ENFORCE(cudaGetDevice(&device_));
-    PADDLE_ENFORCE(cudaEventCreate(&event_));
-    auto stream = cuda_dev_ctx->stream();
-    PADDLE_ENFORCE(cudaEventRecord(event_, stream));
-  }
-#endif
+Event::Event(EventType type, std::string name, uint32_t thread_id)
+    : type_(type), name_(name), thread_id_(thread_id) {
   cpu_ns_ = GetTimeInNsec();
 }
 
@@ -124,88 +113,70 @@ double Event::CpuElapsedMs(const Event& e) const {
 
 double Event::CudaElapsedMs(const Event& e) const {
 #ifdef PADDLE_WITH_CUDA
-  if (!has_cuda_) return 0.0;
-  PADDLE_ENFORCE(e.has_cuda() && has_cuda());
-  PADDLE_ENFORCE(e.device() == device());
-  PADDLE_ENFORCE(cudaEventSynchronize(event_));
-  PADDLE_ENFORCE(cudaEventSynchronize(e.event()));
-  float ms;
-  PADDLE_ENFORCE(cudaEventElapsedTime(&ms, event_, e.event()));
-  return ms;
+#ifdef PADDLE_WITH_CUPTI
+  return gpu_ns_ / 1000000.0;
+#endif
 #else
   PADDLE_THROW("CUDA is not enabled");
 #endif
 }
 
-#ifdef PADDLE_WITH_CUDA
-static void ForEachDevice(std::function<void(int)> func) {
-  auto original_device = GetCurrentDeviceId();
-  int count = GetCUDADeviceCount();
-  for (int i = 0; i < count; i++) {
-    SetDeviceId(i);
-    func(i);
-  }
-  SetDeviceId(original_device);
-}
-#endif
-
 inline EventList& GetEventList() {
   if (!g_event_list) {
     std::lock_guard<std::mutex> guard(g_all_event_lists_mutex);
     g_event_list = std::make_shared<EventList>();
     g_thread_id = g_next_thread_id++;
     g_all_event_lists.emplace_front(g_event_list);
+    RecoreCurThreadId(g_thread_id);
   }
   return *g_event_list;
 }
 
-void Mark(const std::string& name, const DeviceContext* dev_ctx) {
-  GetEventList().Record(EventType::kMark, name, g_thread_id, dev_ctx);
+void Mark(const std::string& name) {
+  GetEventList().Record(EventType::kMark, name, g_thread_id);
 }
 
-void PushEvent(const std::string& name, const DeviceContext* dev_ctx) {
-  GetEventList().Record(EventType::kPushRange, name, g_thread_id, dev_ctx);
+Event* PushEvent(const std::string& name) {
+  return GetEventList().Record(EventType::kPushRange, name, g_thread_id);
 }
 
-void PopEvent(const std::string& name, const DeviceContext* dev_ctx) {
-  GetEventList().Record(EventType::kPopRange, name, g_thread_id, dev_ctx);
+void PopEvent(const std::string& name) {
+  GetEventList().Record(EventType::kPopRange, name, g_thread_id);
 }
 
-RecordEvent::RecordEvent(const std::string& name, const DeviceContext* dev_ctx)
+RecordEvent::RecordEvent(const std::string& name)
     : is_enabled_(false), start_ns_(PosixInNsec()) {
   if (g_state == ProfilerState::kDisabled) return;
-  std::lock_guard<std::mutex> l(profiler_mu);
+  // lock is not needed, the code below is thread-safe
 
   is_enabled_ = true;
-  dev_ctx_ = dev_ctx;
   name_ = name;
-  PushEvent(name_, dev_ctx_);
+  Event* e = PushEvent(name_);
   // Maybe need the same push/pop behavior.
-  SetCurAnnotation(name_);
+  SetCurAnnotation(e);
 }
 
 RecordEvent::~RecordEvent() {
   if (g_state == ProfilerState::kDisabled || !is_enabled_) return;
-  std::lock_guard<std::mutex> l(profiler_mu);
+  // lock is not needed, the code below is thread-safe
   DeviceTracer* tracer = GetDeviceTracer();
   if (tracer) {
-    tracer->AddCPURecords(CurAnnotation(), start_ns_, PosixInNsec(),
+    tracer->AddCPURecords(CurAnnotationName(), start_ns_, PosixInNsec(),
                           BlockDepth(), g_thread_id);
   }
   ClearCurAnnotation();
-  PopEvent(name_, dev_ctx_);
+  PopEvent(name_);
 }
 
-RecordRPCEvent::RecordRPCEvent(const std::string& name,
-                               const DeviceContext* dev_ctx) {
+RecordRPCEvent::RecordRPCEvent(const std::string& name) {
   if (FLAGS_enable_rpc_profiler) {
-    event_.reset(new platform::RecordEvent(name, dev_ctx));
+    event_.reset(new platform::RecordEvent(name));
   }
 }
 
 RecordBlock::RecordBlock(int block_id)
     : is_enabled_(false), start_ns_(PosixInNsec()) {
-  std::lock_guard<std::mutex> l(profiler_mu);
+  // lock is not needed, the code below is thread-safe
   if (g_state == ProfilerState::kDisabled) return;
   is_enabled_ = true;
   SetCurBlock(block_id);
@@ -213,7 +184,7 @@ RecordBlock::RecordBlock(int block_id)
 }
 
 RecordBlock::~RecordBlock() {
-  std::lock_guard<std::mutex> l(profiler_mu);
+  // lock is not needed, the code below is thread-safe
   if (g_state == ProfilerState::kDisabled || !is_enabled_) return;
   DeviceTracer* tracer = GetDeviceTracer();
   if (tracer) {
@@ -225,11 +196,21 @@ RecordBlock::~RecordBlock() {
   ClearCurBlock();
 }
 
+void SynchronizeAllDevice() {
+#ifdef PADDLE_WITH_CUDA
+  int count = GetCUDADeviceCount();
+  for (int i = 0; i < count; i++) {
+    SetDeviceId(i);
+    PADDLE_ENFORCE(cudaDeviceSynchronize());
+  }
+#endif
+}
+
 void EnableProfiler(ProfilerState state) {
   PADDLE_ENFORCE(state != ProfilerState::kDisabled,
                  "Can't enable profiling, since the input state is ",
                  "ProfilerState::kDisabled");
-
+  SynchronizeAllDevice();
   std::lock_guard<std::mutex> l(profiler_mu);
   if (state == g_state) {
     return;
@@ -238,23 +219,20 @@ void EnableProfiler(ProfilerState state) {
   should_send_profile_state = true;
   GetDeviceTracer()->Enable();
 #ifdef PADDLE_WITH_CUDA
-  if (g_state == ProfilerState::kCUDA) {
+  if (g_state == ProfilerState::kCUDA || g_state == ProfilerState::kAll ||
+      g_state == ProfilerState::kCPU) {
     // Generate some dummy events first to reduce the startup overhead.
-    for (int i = 0; i < 5; i++) {
-      ForEachDevice([](int d) {
-        DeviceContext* dev_ctx = new CUDADeviceContext(CUDAPlace(d));
-        Mark("_cuda_startup_", dev_ctx);
-        dev_ctx->Wait();
-        delete dev_ctx;
-      });
-    }
+    DummyKernelAndEvent();
+    GetDeviceTracer()->Reset();
   }
 #endif
   // Mark the profiling start.
-  Mark("_start_profiler_", nullptr);
+  Mark("_start_profiler_");
 }
 
 void ResetProfiler() {
+  SynchronizeAllDevice();
+  GetDeviceTracer()->Reset();
   std::lock_guard<std::mutex> guard(g_all_event_lists_mutex);
   for (auto it = g_all_event_lists.begin(); it != g_all_event_lists.end();
        ++it) {
@@ -481,20 +459,23 @@ void ParseEvents(const std::vector<std::vector<Event>>& events,
 
 void DisableProfiler(EventSortingKey sorted_key,
                      const std::string& profile_path) {
+  SynchronizeAllDevice();
   std::lock_guard<std::mutex> l(profiler_mu);
   if (g_state == ProfilerState::kDisabled) return;
   // Mark the profiling stop.
-  Mark("_stop_profiler_", nullptr);
+  Mark("_stop_profiler_");
 
-  std::vector<std::vector<Event>> all_events = GetAllEvents();
-  ParseEvents(all_events, true, sorted_key);
-  ParseEvents(all_events, false, sorted_key);
-  ResetProfiler();
   DeviceTracer* tracer = GetDeviceTracer();
   if (tracer->IsEnabled()) {
     tracer->Disable();
     tracer->GenProfile(profile_path);
+    tracer->GenEventKernelCudaElapsedTime();
   }
+
+  std::vector<std::vector<Event>> all_events = GetAllEvents();
+  ParseEvents(all_events, true, sorted_key);
+  ParseEvents(all_events, false, sorted_key);
+  ResetProfiler();
   g_state = ProfilerState::kDisabled;
   should_send_profile_state = true;
 }

paddle/fluid/platform/profiler.cu

+/* Copyright (c) 2019 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 "paddle/fluid/platform/profiler.h"
+
+#include <cuda.h>
+
+namespace paddle {
+namespace platform {
+
+__global__ void DummyKernel(int *a) { a[0] = 0; }
+
+static void ForEachDevice(std::function<void(int)> func) {
+  auto original_device = GetCurrentDeviceId();
+  int count = GetCUDADeviceCount();
+  for (int i = 0; i < count; i++) {
+    SetDeviceId(i);
+    func(i);
+  }
+  SetDeviceId(original_device);
+}
+
+void DummyKernelAndEvent() {
+  for (int i = 0; i < 5; i++) {
+    ForEachDevice([](int d) {
+      CUDADeviceContext *dev_ctx = new CUDADeviceContext(CUDAPlace(d));
+      Mark("_cuda_startup_");
+      int *ptr;
+      PADDLE_ENFORCE(cudaMalloc(&ptr, sizeof(int)));
+      DummyKernel<<<1, 1, 0, dev_ctx->stream()>>>(ptr);
+      dev_ctx->Wait();
+      PADDLE_ENFORCE(cudaFree(ptr));
+      delete dev_ctx;
+    });
+  }
+}
+
+}  // namespace platform
+}  // namespace paddle

paddle/fluid/platform/profiler.h

@@ -28,17 +28,17 @@ class Event {
  public:
   // The DeviceContext is used to get the cuda stream.
   // If CPU profiling mode, can pass nullptr.
-  Event(EventType type, std::string name, uint32_t thread_id,
-        const DeviceContext* dev_ctx);
+  Event(EventType type, std::string name, uint32_t thread_id);
 
   const EventType& type() const;
   std::string name() const { return name_; }
   uint32_t thread_id() const { return thread_id_; }
-  bool has_cuda() const { return has_cuda_; }
 
 #ifdef PADDLE_WITH_CUDA
+#ifndef PADDLE_WITH_CUPTI
   cudaEvent_t event() const { return event_; }
   int device() const { return device_; }
+#endif
 #endif
 
   double CpuElapsedMs(const Event& e) const;
@@ -49,11 +49,21 @@ class Event {
   std::string name_;
   uint32_t thread_id_;
   int64_t cpu_ns_;
-  bool has_cuda_;
 #ifdef PADDLE_WITH_CUDA
+#ifdef PADDLE_WITH_CUPTI
+  int64_t gpu_ns_ = 0;
+
+ public:
+  void AddCudaElapsedTime(int64_t start_ns, int64_t end_ns) {
+    gpu_ns_ += end_ns - start_ns;
+  }
+
+ private:
+#else
   cudaEvent_t event_ = nullptr;
   int device_ = -1;
 #endif
+#endif
 };
 
 enum ProfilerState {
@@ -63,22 +73,19 @@ enum ProfilerState {
   kAll,       // Profile both CPU and GPU. (Currently experimental).
 };
 
-void Mark(const std::string& name, const DeviceContext* dev_ctx);
+void Mark(const std::string& name);
 
-void PushEvent(const std::string& name, const DeviceContext* dev_ctx);
+Event* PushEvent(const std::string& name);
 
-void PopEvent(const std::string& name, const DeviceContext* dev_ctx);
+void PopEvent(const std::string& name);
 
 struct RecordEvent {
-  // dev_ctx can be set to nullptr if device is cpu.
-  RecordEvent(const std::string& name, const DeviceContext* dev_ctx);
+  explicit RecordEvent(const std::string& name);
 
   ~RecordEvent();
 
   bool is_enabled_;
   uint64_t start_ns_;
-  // The device context is used by Event to get the current cuda stream.
-  const DeviceContext* dev_ctx_;
   // Event name
   std::string name_;
   // Need to distinguish name by op type, block_id, program_id and perhaps
@@ -88,8 +95,7 @@ struct RecordEvent {
 
 class RecordRPCEvent {
  public:
-  // dev_ctx can be set to nullptr if device is cpu.
-  RecordRPCEvent(const std::string& name, const DeviceContext* dev_ctx);
+  explicit RecordRPCEvent(const std::string& name);
   ~RecordRPCEvent() {}
 
  private:
@@ -132,5 +138,9 @@ bool ShouldSendProfileState();
 void SetProfileListener();
 int64_t ListenerId();
 
+#ifdef PADDLE_WITH_CUDA
+void DummyKernelAndEvent();
+#endif
+
 }  // namespace platform
 }  // namespace paddle

paddle/fluid/platform/profiler.proto

@@ -31,6 +31,7 @@ message Event {
   optional int64 sub_device_id = 6;
 
   optional MemCopy memcopy = 7;
+  optional string detail_info = 9;
 }
 
 message Profile {

paddle/fluid/platform/profiler_test.cc

@@ -23,76 +23,49 @@ TEST(Event, CpuElapsedTime) {
   using paddle::platform::Event;
   using paddle::platform::EventType;
 
-  Event start_event(EventType::kPushRange, "test", 0, nullptr);
-  EXPECT_TRUE(start_event.has_cuda() == false);
+  Event start_event(EventType::kPushRange, "test", 0);
   int counter = 0;
   while (counter != 1000) {
     counter++;
   }
-  Event stop_event(EventType::kPopRange, "test", 0, nullptr);
+  Event stop_event(EventType::kPopRange, "test", 0);
   EXPECT_GT(start_event.CpuElapsedMs(stop_event), 0);
 }
 
-#ifdef PADDLE_WITH_CUDA
-TEST(Event, CudaElapsedTime) {
-  using paddle::platform::DeviceContext;
-  using paddle::platform::CUDADeviceContext;
-  using paddle::platform::CUDAPlace;
-  using paddle::platform::Event;
-  using paddle::platform::EventType;
-
-  DeviceContext* dev_ctx = new CUDADeviceContext(CUDAPlace(0));
-  Event start_event(EventType::kPushRange, "test", 0, dev_ctx);
-  EXPECT_TRUE(start_event.has_cuda() == true);
-  int counter = 0;
-  while (counter != 1000) {
-    counter++;
-  }
-  Event stop_event(EventType::kPopRange, "test", 0, dev_ctx);
-  EXPECT_GT(start_event.CudaElapsedMs(stop_event), 0);
-}
-#endif
-
 TEST(RecordEvent, RecordEvent) {
   using paddle::platform::DeviceContext;
   using paddle::platform::Event;
   using paddle::platform::EventType;
   using paddle::platform::RecordEvent;
+  using paddle::platform::PushEvent;
+  using paddle::platform::PopEvent;
   using paddle::platform::ProfilerState;
   using paddle::platform::EventSortingKey;
 
   ProfilerState state = ProfilerState::kCPU;
-  DeviceContext* dev_ctx = nullptr;
-#ifdef PADDLE_WITH_CUDA
-  using paddle::platform::CUDADeviceContext;
-  using paddle::platform::CUDAPlace;
-  state = ProfilerState::kCUDA;
-  dev_ctx =
-      new paddle::platform::CUDADeviceContext(paddle::platform::CUDAPlace(0));
-#endif
   EnableProfiler(state);
 
   /* Usage 1:
-  *  PushEvent(evt_name, dev_ctx);
+  *  PushEvent(evt_name);
   *  ...
   *  code to be analyzed
   *  ...
-  * PopEvent(evt_name, dev_ctx);
+  * PopEvent(evt_name);
   */
   LOG(INFO) << "Usage 1: PushEvent & PopEvent";
   for (int loop = 0; loop < 3; ++loop) {
     for (int i = 1; i < 5; ++i) {
       std::string name = "op_" + std::to_string(i);
-      PushEvent(name, dev_ctx);
+      PushEvent(name);
       int counter = 1;
       while (counter != i * 1000) counter++;
-      PopEvent(name, dev_ctx);
+      PopEvent(name);
     }
   }
 
   /* Usage 2:
    * {
-   *   RecordEvent record_event(name, dev_ctx);
+   *   RecordEvent record_event(name);
    *   ...
    *   code to be analyzed
    *   ...
@@ -101,7 +74,7 @@ TEST(RecordEvent, RecordEvent) {
   LOG(INFO) << "Usage 2: RecordEvent";
   for (int i = 1; i < 5; ++i) {
     std::string name = "evs_op_" + std::to_string(i);
-    RecordEvent record_event(name, dev_ctx);
+    RecordEvent record_event(name);
     int counter = 1;
     while (counter != i * 1000) counter++;
   }
@@ -123,20 +96,20 @@ TEST(RecordEvent, RecordEvent) {
   LOG(INFO) << "Usage 3: nested RecordEvent";
   for (int i = 1; i < 5; ++i) {
     std::string name = "ano_evs_op_" + std::to_string(i);
-    RecordEvent record_event(name, dev_ctx);
+    RecordEvent record_event(name);
     int counter = 1;
     while (counter != i * 100) counter++;
     {
       std::string nested_name = "nested_ano_evs_op_" + std::to_string(i);
-      RecordEvent nested_record_event(nested_name, dev_ctx);
+      RecordEvent nested_record_event(nested_name);
       int nested_counter = 1;
       while (nested_counter != i * 100) nested_counter++;
     }
   }
 
   // Bad Usage:
-  PushEvent("event_without_pop", dev_ctx);
-  PopEvent("event_without_push", dev_ctx);
+  PushEvent("event_without_pop");
+  PopEvent("event_without_push");
   std::vector<std::vector<Event>> events = paddle::platform::GetAllEvents();
 
   int cuda_startup_count = 0;

python/paddle/fluid/__init__.py

@@ -131,7 +131,8 @@ def __bootstrap__():
         'eager_delete_tensor_gb', 'fast_eager_deletion_mode',
         'allocator_strategy', 'reader_queue_speed_test_mode',
         'print_sub_graph_dir', 'pe_profile_fname', 'warpctc_dir',
-        'inner_op_parallelism', 'enable_parallel_graph'
+        'inner_op_parallelism', 'enable_parallel_graph',
+        'multiple_of_cupti_buffer_size'
     ]
     if 'Darwin' not in sysstr:
         read_env_flags.append('use_pinned_memory')

python/paddle/fluid/optimizer.py

@@ -649,6 +649,7 @@ class AdagradOptimizer(Optimizer):
         regularization: A Regularizer, such as
                         fluid.regularizer.L2DecayRegularizer.
         name: A optional name prefix.
+        initial_accumulator_value (float): Initial value for moment accumulator.
 
     Examples:
         .. code-block:: python
@@ -662,7 +663,8 @@ class AdagradOptimizer(Optimizer):
                  learning_rate,
                  epsilon=1.0e-6,
                  regularization=None,
-                 name=None):
+                 name=None,
+                 initial_accumulator_value=0.0):
         assert learning_rate is not None
         assert epsilon is not None
         super(AdagradOptimizer, self).__init__(
@@ -671,6 +673,7 @@ class AdagradOptimizer(Optimizer):
             name=name)
         self.type = "adagrad"
         self._epsilon = epsilon
+        self.initial_accumulator_value = initial_accumulator_value
 
     def _create_accumulators(self, block, parameters):
         assert isinstance(block, framework.Block)
@@ -683,6 +686,16 @@ class AdagradOptimizer(Optimizer):
 
         moment_acc = self._get_accumulator(self._moment_acc_str,
                                            param_and_grad[0])
+        startup_block = framework.default_startup_program().global_block()
+        startup_block.append_op(
+            type='fill_constant',
+            inputs={},
+            outputs={'Out': [moment_acc]},
+            attrs={
+                'dtype': moment_acc.dtype,
+                'value': self.initial_accumulator_value,
+                'shape': moment_acc.shape,
+            })
 
         # Create the adagrad optimizer op
         adagrad_op = block.append_op(

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

@@ -274,7 +274,7 @@ class TestAdagradOptimizer(unittest.TestCase):
 
         # Check init_program
         init_ops = init_program.global_block().ops
-        self.assertEqual(len(init_ops), 2)
+        self.assertEqual(len(init_ops), 3)
         self.assertEqual(init_ops[0].type, "fill_constant")
         self.assertAlmostEqual(init_ops[0].attr('value'), learning_rate)
         self.assertEqual(init_ops[1].type, "fill_constant")

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

+# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import unittest
+
+import numpy as np
+import os
+os.environ['FLAGS_enable_parallel_graph'] = str(1)
+import paddle.fluid.core as core
+import os
+import paddle.fluid as fluid
+from parallel_executor_test_base import TestParallelExecutorBase
+
+
+def simple_fc_net(use_feed):
+    img = fluid.layers.data(name='image', shape=[784], dtype='float32')
+    label = fluid.layers.data(name='label', shape=[1], dtype='int64')
+    hidden = img
+    for _ in range(4):
+        hidden = fluid.layers.fc(
+            hidden,
+            size=200,
+            act='tanh',
+            bias_attr=fluid.ParamAttr(
+                initializer=fluid.initializer.Constant(value=1.0)))
+    prediction = fluid.layers.fc(hidden, size=10, act='softmax')
+    loss = fluid.layers.cross_entropy(input=prediction, label=label)
+    loss = fluid.layers.mean(loss)
+    return loss
+
+
+class TestMNIST(TestParallelExecutorBase):
+    @classmethod
+    def setUpClass(cls):
+        os.environ['CPU_NUM'] = str(4)
+
+    def _init_data(self):
+        np.random.seed(5)
+        img = np.random.random(size=[32, 784]).astype(np.float32)
+        label = np.ones(shape=[32, 1], dtype='int64')
+        return img, label
+
+    # simple_fc
+    def check_simple_fc_convergence(self, use_cuda, use_reduce=False):
+        if use_cuda and not core.is_compiled_with_cuda():
+            return
+
+        img, label = self._init_data()
+
+        self.check_network_convergence(
+            simple_fc_net,
+            feed_dict={"image": img,
+                       "label": label},
+            use_cuda=use_cuda,
+            use_reduce=use_reduce)
+
+    def test_simple_fc(self):
+        # use_cuda
+        self.check_simple_fc_convergence(True)
+
+    def check_simple_fc_parallel_accuracy(self, use_cuda):
+        if use_cuda and not core.is_compiled_with_cuda():
+            return
+
+        img, label = self._init_data()
+
+        single_first_loss, single_last_loss = self.check_network_convergence(
+            method=simple_fc_net,
+            seed=1,
+            feed_dict={"image": img,
+                       "label": label},
+            use_cuda=use_cuda,
+            use_parallel_executor=False)
+        parallel_first_loss, parallel_last_loss = self.check_network_convergence(
+            method=simple_fc_net,
+            seed=1,
+            feed_dict={"image": img,
+                       "label": label},
+            use_cuda=use_cuda,
+            use_parallel_executor=True)
+
+        self.assertAlmostEquals(
+            np.mean(parallel_first_loss),
+            single_first_loss,
+            delta=1e-6, )
+        self.assertAlmostEquals(
+            np.mean(parallel_last_loss), single_last_loss, delta=1e-6)
+
+    def test_simple_fc_parallel_accuracy(self):
+        self.check_simple_fc_parallel_accuracy(True)
+
+
+if __name__ == '__main__':
+    unittest.main()

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

@@ -16,15 +16,19 @@ from __future__ import print_function
 
 import unittest
 import os
+import tempfile
 import numpy as np
 import paddle.fluid as fluid
 import paddle.fluid.profiler as profiler
 import paddle.fluid.layers as layers
 import paddle.fluid.core as core
+import paddle.fluid.proto.profiler.profiler_pb2 as profiler_pb2
 
 
 class TestProfiler(unittest.TestCase):
-    def net_profiler(self, state, profile_path='/tmp/profile'):
+    def net_profiler(self, state, use_parallel_executor=False):
+        profile_path = os.path.join(tempfile.gettempdir(), "profile")
+        open(profile_path, "w").write("")
         startup_program = fluid.Program()
         main_program = fluid.Program()
 
@@ -60,6 +64,11 @@ class TestProfiler(unittest.TestCase):
         place = fluid.CPUPlace() if state == 'CPU' else fluid.CUDAPlace(0)
         exe = fluid.Executor(place)
         exe.run(startup_program)
+        if use_parallel_executor:
+            pe = fluid.ParallelExecutor(
+                state != 'CPU',
+                loss_name=avg_cost.name,
+                main_program=main_program)
 
         pass_acc_calculator = fluid.average.WeightedAverage()
         with profiler.profiler(state, 'total', profile_path) as prof:
@@ -69,6 +78,9 @@ class TestProfiler(unittest.TestCase):
                 x = np.random.random((32, 784)).astype("float32")
                 y = np.random.randint(0, 10, (32, 1)).astype("int64")
 
+                if use_parallel_executor:
+                    pe.run(feed={'x': x, 'y': y}, fetch_list=[avg_cost.name])
+                    continue
                 outs = exe.run(main_program,
                                feed={'x': x,
                                      'y': y},
@@ -77,21 +89,37 @@ class TestProfiler(unittest.TestCase):
                 b_size = np.array(outs[2])
                 pass_acc_calculator.add(value=acc, weight=b_size)
                 pass_acc = pass_acc_calculator.eval()
+        data = open(profile_path, 'rb').read()
+        self.assertGreater(len(data), 0)
+        profile_pb = profiler_pb2.Profile()
+        profile_pb.ParseFromString(data)
+        self.assertGreater(len(profile_pb.events), 0)
+        for event in profile_pb.events:
+            if event.type == profiler_pb2.Event.GPUKernel:
+                if not event.detail_info and not event.name.startswith("MEM"):
+                    raise Exception(
+                        "Kernel %s missing event. Has this kernel been recorded by RecordEvent?"
+                        % event.name)
+            elif event.type == profiler_pb2.Event.CPU and (
+                    event.name.startswith("Driver API") or
+                    event.name.startswith("Runtime API")):
+                print("Warning: unregister", event.name)
 
     def test_cpu_profiler(self):
         self.net_profiler('CPU')
+        self.net_profiler('CPU', use_parallel_executor=True)
 
     @unittest.skipIf(not core.is_compiled_with_cuda(),
                      "profiler is enabled only with GPU")
     def test_cuda_profiler(self):
         self.net_profiler('GPU')
+        self.net_profiler('GPU', use_parallel_executor=True)
 
     @unittest.skipIf(not core.is_compiled_with_cuda(),
                      "profiler is enabled only with GPU")
     def test_all_profiler(self):
-        self.net_profiler('All', '/tmp/profile_out')
-        with open('/tmp/profile_out', 'rb') as f:
-            self.assertGreater(len(f.read()), 0)
+        self.net_profiler('All')
+        self.net_profiler('All', use_parallel_executor=True)
 
 
 if __name__ == '__main__':

tools/manylinux1/build_all.sh

@@ -24,3 +24,8 @@ sed 's/<baseimg>/9.0-cudnn7-devel-centos6/g' Dockerfile.x64 | \
 sed 's/<NCCL_MAKE_OPTS>/NVCC_GENCODE="-gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_52,code=sm_52 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_60,code=compute_60 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_62,code=sm_62 -gencode=arch=compute_70,code=sm_70"/g'> Dockerfile.tmp
 docker build -t ${REPO}/paddle_manylinux_devel:cuda9.0_cudnn7 -f Dockerfile.tmp .
 docker push ${REPO}/paddle_manylinux_devel:cuda9.0_cudnn7
+
+sed 's/<baseimg>/10.0-devel-centos6/g' Dockerfile.x64 | \
+sed 's/<NCCL_MAKE_OPTS>/NVCC_GENCODE="-gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_52,code=sm_52 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_60,code=compute_60 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_62,code=sm_62 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75"/g'> Dockerfile.tmp
+docker build -t ${REPO}/paddle_manylinux_devel:cuda10.0_cudnn7 -f Dockerfile.tmp .
+docker push ${REPO}/paddle_manylinux_devel:cuda10.0_cudnn7

tools/manylinux1/build_scripts/build.sh

@@ -107,11 +107,13 @@ curl-config --features
 rm -rf /usr/local/ssl
 
 # Install patchelf (latest with unreleased bug fixes)
-curl -sLO https://nixos.org/releases/patchelf/patchelf-0.9/patchelf-0.9.tar.gz
-check_sha256sum patchelf-0.9.tar.gz $PATCHELF_HASH
-tar -xzf patchelf-0.9.tar.gz
-(cd patchelf-0.9 && ./configure && make && make install)
-rm -rf patchelf-0.9.tar.gz patchelf-0.9
+# FIXME(typhoonzero): restore this when the link is fixed.
+# curl -sLO http://nipy.bic.berkeley.edu/manylinux/patchelf-0.9njs2.tar.gz
+# check_sha256sum patchelf-0.9njs2.tar.gz $PATCHELF_HASH
+# tar -xzf patchelf-0.9njs2.tar.gz
+# (cd patchelf-0.9njs2 && ./configure && make && make install)
+# rm -rf patchelf-0.9njs2.tar.gz patchelf-0.9njs2
+yum install -y patchelf
 
 # Install latest pypi release of auditwheel
 LD_LIBRARY_PATH="${ORIGINAL_LD_LIBRARY_PATH}:$(dirname ${PY35_BIN})/lib" $PY35_BIN/pip install auditwheel

tools/manylinux1/build_scripts/build_utils.sh

@@ -87,6 +87,8 @@ function do_cpython_build {
     # NOTE Make libpython shared library visible to python calls below
     LD_LIBRARY_PATH="${prefix}/lib" ${prefix}/bin/python get-pip.py
     LD_LIBRARY_PATH="${prefix}/lib" ${prefix}/bin/pip install wheel
+    cd /
+    ls ${MY_DIR}
     local abi_tag=$(LD_LIBRARY_PATH="${prefix}/lib" ${prefix}/bin/python ${MY_DIR}/python-tag-abi-tag.py)
     ln -s ${prefix} /opt/python/${abi_tag}
 }

tools/timeline.py

@@ -131,8 +131,12 @@ class Timeline(object):
                     if (k, event.device_id, "CPU") not in self._devices:
                         pid = self._allocate_pid()
                         self._devices[(k, event.device_id, "CPU")] = pid
-                        self._chrome_trace.emit_pid("%s:cpu:block:%d" %
-                                                    (k, event.device_id), pid)
+                        # -1 device id represents CUDA api call
+                        if event.device_id == -1:
+                            self._chrome_trace.emit_pid("%s:cuda_api" % k, pid)
+                        else:
+                            self._chrome_trace.emit_pid(
+                                "%s:cpu:block:%d" % (k, event.device_id), pid)
                 elif event.type == profiler_pb2.Event.GPUKernel:
                     if (k, event.device_id, "GPUKernel") not in self._devices:
                         pid = self._allocate_pid()
@@ -150,7 +154,9 @@ class Timeline(object):
                 pid = self._devices[(k, event.device_id, type)]
                 args = {'name': event.name}
                 if event.memcopy.bytes > 0:
-                    args = {'mem_bytes': event.memcopy.bytes}
+                    args['mem_bytes'] = event.memcopy.bytes
+                if event.detail_info:
+                    args['detail_info'] = event.detail_info
                 # TODO(panyx0718): Chrome tracing only handles ms. However, some
                 # ops takes micro-seconds. Hence, we keep the ns here.
                 self._chrome_trace.emit_region(
@@ -173,7 +179,7 @@ if args.timeline_path:
 profile_paths = profile_path.split(',')
 profile_dict = dict()
 if len(profile_paths) == 1:
-    with open(profile_path, 'r') as f:
+    with open(profile_path, 'rb') as f:
         profile_s = f.read()
         profile_pb = profiler_pb2.Profile()
         profile_pb.ParseFromString(profile_s)
@@ -181,7 +187,7 @@ if len(profile_paths) == 1:
 else:
     for profile_path in profile_paths:
         k, v = profile_path.split('=')
-        with open(v, 'r') as f:
+        with open(v, 'rb') as f:
             profile_s = f.read()
             profile_pb = profiler_pb2.Profile()
             profile_pb.ParseFromString(profile_s)