cleanup

PaddlePaddle · Mar 21, 2018 · e25a270 · e25a270
1 parent 735a5c0
commit e25a270
Show file tree

Hide file tree

Showing 4 changed files with 9 additions and 68 deletions.
diff --git a/benchmark/cluster/vgg16/vgg16_fluid.py b/benchmark/cluster/vgg16/vgg16_fluid.py
@@ -205,10 +205,6 @@ def train_loop(exe, trainer_prog):
                        len(data) / (time.time() - ts))
                 )  # The accuracy is the accumulation of batches, but not the current batch.
 
-                if batch_id >= 20:
-                    print("test complete after batch_id >=20")
-                    break
-
             pass_elapsed = time.time() - start_time
             pass_train_acc = train_pass_acc.eval()
             pass_test_acc = test(exe)

diff --git a/paddle/fluid/operators/detail/grpc_client.h b/paddle/fluid/operators/detail/grpc_client.h
@@ -58,10 +58,7 @@ void ProcGetResponse(const VarHandle& var_h, const grpc::ByteBuffer& msg);
 
 class BaseProcessor {
  public:
-  explicit BaseProcessor(std::shared_ptr<grpc::Channel> ch) {
-    stub_ = sendrecv::SendRecvService::NewStub(ch);
-    context_ = NULL;
-  }
+  explicit BaseProcessor(std::shared_ptr<grpc::Channel> ch) { context_ = NULL; }
 
   virtual ~BaseProcessor() {}
 
@@ -86,7 +83,6 @@ class BaseProcessor {
 
   virtual void Process() = 0;
 
-  std::unique_ptr<sendrecv::SendRecvService::Stub> stub_;
   std::unique_ptr<grpc::ClientContext> context_;
   grpc::Status status_;
   VarHandle var_h_;
@@ -137,23 +133,29 @@ class GetProcessor : public BaseProcessor {
 class BatchBarrierProcessor : public BaseProcessor {
  public:
   explicit BatchBarrierProcessor(std::shared_ptr<grpc::Channel> ch)
-      : BaseProcessor(ch) {}
+      : BaseProcessor(ch) {
+    stub_ = sendrecv::SendRecvService::NewStub(ch);
+  }
 
   virtual ~BatchBarrierProcessor() {}
 
   virtual void Process() {}
   sendrecv::VoidMessage reply_;
+  std::unique_ptr<sendrecv::SendRecvService::Stub> stub_;
 };
 
 class FetchBarrierProcessor : public BaseProcessor {
  public:
   explicit FetchBarrierProcessor(std::shared_ptr<grpc::Channel> ch)
-      : BaseProcessor(ch) {}
+      : BaseProcessor(ch) {
+    stub_ = sendrecv::SendRecvService::NewStub(ch);
+  }
 
   virtual ~FetchBarrierProcessor() {}
 
   virtual void Process() {}
   sendrecv::VariableMessage reply_;
+  std::unique_ptr<sendrecv::SendRecvService::Stub> stub_;
 };
 
 class RPCClient {

diff --git a/paddle/fluid/operators/detail/sendrecvop_utils.cc b/paddle/fluid/operators/detail/sendrecvop_utils.cc
@@ -75,29 +75,13 @@ void SerializeToByteBuffer(const std::string& name, framework::Variable* var,
       }
       if (platform::is_gpu_place(ctx.GetPlace())) {
 #ifdef PADDLE_WITH_CUDA
-        /*
-      struct timeval t0_wait, t1_wait;
-      gettimeofday(&t0_wait, 0);
-      std::thread::id this_id = std::this_thread::get_id();
-      */
-
         PADDLE_ENFORCE(platform::is_gpu_place(tensor.place()));
         platform::CPUPlace cpu;
         auto& gpu_dev_ctx =
             static_cast<const platform::CUDADeviceContext&>(ctx);
         auto copy_size = tensor.memory_size();
         payload = memory::Alloc(cpu, copy_size);
 
-        /*
-        gettimeofday(&t1_wait, 0);
-        double t_wait = double((t1_wait.tv_sec - t0_wait.tv_sec) * 1000.0 +
-                               (t1_wait.tv_usec - t0_wait.tv_usec) / 1000.0);
-        std::stringstream ss;
-        ss << "se malloc var_name:" << name << ", dims: " << tensor.dims()
-           << ", time:" << t_wait << "ms, thread_id:" << this_id;
-        std::cout << ss.str() << '\n';
-        */
-
         memory::Copy(cpu, payload,
                      boost::get<platform::CUDAPlace>(tensor.place()),
                      reinterpret_cast<const void*>(tensor.data<void>()),
@@ -108,30 +92,6 @@ void SerializeToByteBuffer(const std::string& name, framework::Variable* var,
           memory::Free(cpu, backing);
         };
 
-        {
-          PrintDetail(name, tensor, ctx, "memcopy gpu");
-          /*
-          std::stringstream ss;
-          ss << "se memcpy gpu var_name:" << name
-             << ", dims: " << tensor.dims();
-          std::cout << ss.str() << ", data:";
-
-          float* data = reinterpret_cast<float*>(payload);
-          for (int i = 0; i < 10; i++) {
-            printf("%.f ", data[i]);
-          }
-          printf("\n");
-          */
-        }
-/*
-gettimeofday(&t1_wait, 0);
-t_wait = double((t1_wait.tv_sec - t0_wait.tv_sec) * 1000.0 +
-                (t1_wait.tv_usec - t0_wait.tv_usec) / 1000.0);
-std::stringstream ss2;
-ss2 << "se memcpy gpu var_name:" << name << ", dims: " << tensor.dims()
-    << ", time:" << t_wait << "ms, thread_id:" << this_id;
-std::cout << ss2.str() << '\n';
-*/
 #endif
       } else {
         payload = tensor.data<void>();

diff --git a/paddle/fluid/operators/detail/sendrecvop_utils.h b/paddle/fluid/operators/detail/sendrecvop_utils.h
@@ -63,23 +63,6 @@ inline std::type_index ToTypeIndex(sendrecv::VariableMessage::Type type) {
   }
 }
 
-inline void PrintDetail(std::string name, const framework::Tensor& src,
-                        const platform::DeviceContext& ctx,
-                        std::string post_fix = "") {
-  if (name != "batch_norm_11.w_1@GRAD.trainer_0") {
-    return;
-  }
-
-  std::vector<float> vec;
-  framework::TensorToVector<float>(src, ctx, &vec);
-
-  printf("%s name:%s, data:", post_fix.c_str(), name.c_str());
-  for (uint32_t i = 0; i < vec.size(); i++) {
-    printf("%.f ", vec[i]);
-  }
-  printf("\n");
-}
-
 }  // namespace detail
 }  // namespace operators
 }  // namespace paddle