!4316 fix debugger d-chip watchpoints and ensure GPU updated tensor values

Merge pull request !4316 from john_tzanakakis/master_ms1_grpc

mindspore/ccsrc/debug/debugger/debugger.cc

@@ -172,10 +172,15 @@ void Debugger::PostExecute() {
     return;
   }
   if (debugger_enabled_ && !is_dataset_graph_) {
+    if (device_target_ != kGPUDevice) {
+      num_step_++;
       MS_LOG(INFO) << "Debugger suspend at end of step; number of steps executed: " << num_step_;
+      SendWatchpointsAndSuspend(CheckWatchpoints());
+    } else {
       CommandLoop();
     }
   }
+}
 
 bool Debugger::ReadNodeDataRequired() {
   if (debugger_enabled_ && !is_dataset_graph_) {

mindspore/ccsrc/debug/tensor_load.h

@@ -46,7 +46,7 @@ class TensorLoader {
       }
     }
     tensor_list.push_back(tensor);
-    tensor_list_map.insert({tensor->GetName(), tensor});
+    tensor_list_map[tensor->GetName()] = tensor;  // use [] instead of insert to ensure latest value
     auto node_name = tensor->GetName();
     node_name = node_name.substr(0, node_name.find_first_of(":"));
     node_tensor_map.insert({node_name, tensor});

mindspore/ccsrc/runtime/device/gpu/gpu_kernel_runtime.cc

@@ -241,7 +241,6 @@ void LoadKernelData(Debugger *debugger, const CNodePtr &kernel,
   }
 
   // get inputs
-  if (!dump_enabled) {
   auto input_size = AnfAlgo::GetInputTensorNum(kernel);
   for (size_t j = 0; j < input_size; ++j) {
     auto input_kernel = kernel->input(j + 1);
@@ -261,7 +260,6 @@ void LoadKernelData(Debugger *debugger, const CNodePtr &kernel,
                     << ", tensor_name:" << input_tensor_name << ", host_format:" << format << ".!";
     }
   }
-  }
 
   // get outputs
   auto output_size = AnfAlgo::GetOutputTensorNum(kernel);