Merge pull request #9746 from typhoonzero/multigpumultinode

[Feature] Enable multi gpu distributed training of fluid

Author: typhoonzero

paddle/fluid/framework/details/CMakeLists.txt

@@ -5,6 +5,7 @@ cc_library(fetch_op_handle SRCS fetch_op_handle.cc DEPS op_handle_base scope lod
 nv_library(nccl_all_reduce_op_handle SRCS nccl_all_reduce_op_handle.cc DEPS op_handle_base scope lod_tensor ddim memory
         dynload_cuda)
 cc_library(computation_op_handle SRCS computation_op_handle.cc DEPS framework_proto scope place operator op_registry)
+cc_library(send_op_handle SRCS send_op_handle.cc DEPS framework_proto scope place operator op_registry)
 
 cc_library(ssa_graph SRCS ssa_graph.cc DEPS var_handle op_handle_base)
 cc_library(ssa_graph_builder SRCS ssa_graph_builder.cc DEPS ssa_graph)
@@ -15,7 +16,7 @@ else()
     set(multi_devices_graph_builder_deps)
 endif()
 cc_library(multi_devices_graph_builder SRCS multi_devices_graph_builder.cc DEPS ssa_graph_builder computation_op_handle
-            scale_loss_grad_op_handle ${multi_devices_graph_builder_deps})
+            scale_loss_grad_op_handle send_op_handle ${multi_devices_graph_builder_deps})
 cc_library(ssa_graph_executor SRCS ssa_graph_executor.cc DEPS ssa_graph framework_proto)
 cc_library(threaded_ssa_graph_executor SRCS threaded_ssa_graph_executor.cc DEPS fetch_op_handle ssa_graph_executor scope
         simple_threadpool device_context)

paddle/fluid/framework/details/multi_devices_graph_builder.cc

@@ -15,6 +15,7 @@
 #include "paddle/fluid/framework/details/multi_devices_graph_builder.h"
 #include "paddle/fluid/framework/details/computation_op_handle.h"
 #include "paddle/fluid/framework/details/scale_loss_grad_op_handle.h"
+#include "paddle/fluid/framework/details/send_op_handle.h"
 #include "paddle/fluid/framework/scope.h"
 
 #ifdef PADDLE_WITH_CUDA
@@ -54,6 +55,27 @@ MultiDevSSAGraphBuilder::MultiDevSSAGraphBuilder(
   }
 }
 
+void MultiDevSSAGraphBuilder::CreateOpHandleIOs(SSAGraph *result, OpDesc *op,
+                                                const platform::Place &p,
+                                                const size_t &i) const {
+  auto *op_handle = result->ops_.back().get();
+  op_handle->dev_ctxes_[p] = const_cast<platform::DeviceContext *>(
+      platform::DeviceContextPool::Instance().Get(p));
+
+  auto var_names = op->InputArgumentNames();
+
+  for (auto &each_var_name : var_names) {
+    VarHandle *var = CreateOrGetLatestVarHandle(result, each_var_name, p, i);
+    op_handle->AddInput(var);
+  }
+
+  var_names = op->OutputArgumentNames();
+
+  for (auto &each_var_name : var_names) {
+    CreateOpOutput(result, op_handle, each_var_name, p, i);
+  }
+}
+
 std::unique_ptr<SSAGraph> MultiDevSSAGraphBuilder::Build(
     const ProgramDesc &program) const {
   auto graph = new SSAGraph();
@@ -76,27 +98,28 @@ std::unique_ptr<SSAGraph> MultiDevSSAGraphBuilder::Build(
       }
     }
 
+    // append send op if program is distributed trainer main program.
+    // always use the first device
+    if (!is_forwarding && op->Type() == "send") {
+      auto &p = places_[0];
+      auto *s = local_scopes_[0];
+      // FIXME(wuyi): send op always copy from GPU 0
+      result.ops_.emplace_back(new SendOpHandle(*op, s, p));
+      // Create inputs for output on original place and no ssa output
+      // is created for send op.
+      CreateOpHandleIOs(&result, op, p, 0);
+      continue;
+    }
+
     for (size_t i = 0; i < places_.size(); ++i) {
       auto &p = places_[i];
       auto *s = local_scopes_[i];
 
       result.ops_.emplace_back(new ComputationOpHandle(*op, s, p));
       auto *op_handle = result.ops_.back().get();
-      op_handle->dev_ctxes_[p] = const_cast<platform::DeviceContext *>(
-          platform::DeviceContextPool::Instance().Get(p));
+      CreateOpHandleIOs(&result, op, p, i);
 
-      auto var_names = op->InputArgumentNames();
-
-      for (auto &each_var_name : var_names) {
-        VarHandle *var =
-            CreateOrGetLatestVarHandle(&result, each_var_name, p, i);
-        op_handle->AddInput(var);
-      }
-      var_names = op->OutputArgumentNames();
-
-      for (auto &each_var_name : var_names) {
-        CreateOpOutput(&result, op_handle, each_var_name, p, i);
-      }
+      auto var_names = op->OutputArgumentNames();
 
       if (is_forwarding) {
         if (var_names.size() == 1 && var_names[0] == loss_var_name_) {

paddle/fluid/framework/details/multi_devices_graph_builder.h

@@ -14,6 +14,9 @@
 
 #pragma once
 
+#include <string>
+#include <vector>
+
 #include "paddle/fluid/framework/details/ssa_graph_builder.h"
 
 namespace paddle {
@@ -41,6 +44,10 @@ class MultiDevSSAGraphBuilder : public SSAGraphBuilder {
 
   std::unique_ptr<SSAGraph> Build(const ProgramDesc &program) const override;
 
+ private:
+  void CreateOpHandleIOs(SSAGraph *result, OpDesc *op, const platform::Place &p,
+                         const size_t &i) const;
+
  private:
   std::string loss_var_name_;
   const std::vector<platform::Place> &places_;

paddle/fluid/framework/details/send_op_handle.cc

@@ -0,0 +1,43 @@
+//   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 "paddle/fluid/framework/details/send_op_handle.h"
+
+namespace paddle {
+namespace framework {
+namespace details {
+
+SendOpHandle::SendOpHandle(const framework::OpDesc &op_desc,
+                           const Scope *local_scope,
+                           const platform::Place &place)
+    : op_(framework::OpRegistry::CreateOp(op_desc)),
+      local_scope_(local_scope),
+      place_(place) {}
+
+void SendOpHandle::RunImpl() {
+  // Wait input done
+  for (auto *in : inputs_) {
+    auto &p = static_cast<VarHandle *>(in)->place_;
+    if (in->DebugString() == "dummy") {  // HACK
+      continue;
+    }
+    in->generated_op_->Wait(dev_ctxes_[p]);
+  }
+  op_->Run(*local_scope_, place_);
+}
+
+std::string SendOpHandle::Name() const { return "send"; }
+}  // namespace details
+}  // namespace framework
+}  // namespace paddle

paddle/fluid/framework/details/send_op_handle.h

@@ -0,0 +1,50 @@
+//   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.
+
+#pragma once
+
+#include <string>
+#include <vector>
+
+#include "paddle/fluid/framework/details/op_handle_base.h"
+#include "paddle/fluid/framework/lod_tensor.h"
+#include "paddle/fluid/framework/op_registry.h"
+#include "paddle/fluid/framework/operator.h"
+#include "paddle/fluid/framework/scope.h"
+
+namespace paddle {
+namespace framework {
+namespace details {
+
+struct SendOpHandle : public OpHandleBase {
+  std::unique_ptr<OperatorBase> op_;
+  const Scope* local_scope_;
+  const platform::Place& place_;
+
+  SendOpHandle(const framework::OpDesc& op_desc, const Scope* local_scope,
+               const platform::Place& place);
+
+  std::string Name() const override;
+
+  // Delay and buffer nccl_all_reduce together can significantly increase
+  // performance. Disable this feature by returning false.
+  bool IsMultiDeviceTransfer() override { return false; };
+
+ protected:
+  void RunImpl() override;
+};
+
+}  // namespace details
+}  // namespace framework
+}  // namespace paddle

paddle/fluid/framework/parallel_executor.h

@@ -48,13 +48,13 @@ class ParallelExecutor {
            const std::string& fetched_var_name,
            const std::unordered_map<std::string, LoDTensor>& feed_tensors);
 
+  void BCastParamsToGPUs(const std::unordered_set<std::string>& vars) const;
+
  private:
   void SplitTensorToPlaces(
       const std::unordered_map<std::string, LoDTensor>& feed_tensors);
 
   ParallelExecutorPrivate* member_;
-
-  void BCastParamsToGPUs(const std::unordered_set<std::string>& vars) const;
 };
 
 }  // namespace framework

paddle/fluid/operators/detail/grpc_client.cc

@@ -65,9 +65,8 @@ bool RPCClient::AsyncSendVariable(const std::string& ep,
 }
 
 void ProcGetResponse(const VarHandle& var_h,
-                     // const sendrecv::VariableMessage& ret_msg) {
                      const ::grpc::ByteBuffer& ret_msg) {
-  framework::Variable* outvar = NULL;
+  framework::Variable* outvar = nullptr;
   DeserializeFromByteBuffer(ret_msg, *var_h.ctx, var_h.scope, &outvar);
 }
 

paddle/fluid/operators/detail/serde_test.cc

@@ -107,7 +107,7 @@ void RunSerdeTestSelectedRows(platform::Place place) {
   for (int i = 0; i < tensor_numel; ++i) {
     EXPECT_FLOAT_EQ(tensor_data2[i], 32.7);
   }
-  for (int64_t i = 0; i < rows2->size(); ++i) {
+  for (size_t i = 0; i < rows2->size(); ++i) {
     EXPECT_EQ(rows_data2[i], i);
   }
   EXPECT_EQ(slr2->height(), 1000);

paddle/fluid/pybind/pybind.cc

@@ -553,6 +553,7 @@ All parameter, weight, gradient are variables in Paddle.
                                   bcast_vars, main_program, loss_var_name,
                                   scope, local_scopes, allow_op_delay);
            })
+      .def("bcast_params", &ParallelExecutor::BCastParamsToGPUs)
       .def("local_scopes",
            [](ParallelExecutor &self) -> std::vector<Scope *> * {
              return &self.GetLocalScopes();

python/paddle/fluid/distribute_transpiler.py

@@ -255,6 +255,7 @@ def transpile(self,
     def get_trainer_program(self):
         # remove optimize ops and add a send op to main_program
         self.program.global_block().delete_ops(self.optimize_ops)
+        self.program.sync_with_cpp()
         # FIXME(typhoonzero): serialize once will fix error occurs when clone.
         self.program.__str__()
         return self.program