[Cherry-pick] Optimize log (#45783)

paddle/fluid/eager/accumulation/accumulation_node.cc

@@ -16,6 +16,7 @@
 
 #include "glog/logging.h"
 #include "paddle/fluid/eager/eager_tensor.h"
+#include "paddle/fluid/eager/utils.h"
 #include "paddle/fluid/imperative/gradient_accumulator.h"
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
@@ -89,7 +90,7 @@ GradNodeAccumulation::operator()(
                          kSlotSmallVectorSize>& grads,  // NOLINT
     bool create_graph,
     bool is_new_grad) {
-  VLOG(3) << "Running Eager Backward Node: GradNodeAccumulation";
+  VLOG(3) << "Running AD API Grad: GradNodeAccumulation";
   PADDLE_ENFORCE(grads.size() == 1,
                  paddle::platform::errors::Fatal(
                      "GradNodeAccumulation should take exactly 1 grad tensor"
@@ -122,7 +123,22 @@ GradNodeAccumulation::operator()(
   if (ReduceHooksRegistered()) {
     ApplyReduceHooks();
   }
+  VLOG(3) << "Finish AD API Grad: GradNodeAccumulation";
+  if (VLOG_IS_ON(4)) {
+    const char* INPUT_PRINT_TEMPLATE = "{ Input: [%s], Output: [%s] } ";
 
+    std::string input_str = "";
+    std::string output_str = "";
+    const char* TENSOR_OUT_GRAD_TEMPLATE = "(grads[0][0], [%s]), ";
+    std::string input_out_grad_str = paddle::string::Sprintf(
+        TENSOR_OUT_GRAD_TEMPLATE, egr::EagerUtils::TensorStr(grads[0][0]));
+    const char* TENSOR_X_GRAD_TEMPLATE = "(grad_out, [%s]), ";
+    std::string output_x_grad_str = paddle::string::Sprintf(
+        TENSOR_X_GRAD_TEMPLATE, egr::EagerUtils::TensorStr(grad_out));
+    output_str += output_x_grad_str;
+    VLOG(4) << paddle::string::Sprintf(
+        INPUT_PRINT_TEMPLATE, input_str, output_str);
+  }
   return {{grad_out}};
 }
 

paddle/fluid/eager/accumulation/accumulation_node.h

@@ -24,7 +24,7 @@ class GradNodeAccumulation : public GradNodeBase {
  public:
   // Constructor: configure fwd input tensors to grad node
   explicit GradNodeAccumulation(AutogradMeta* meta) : GradNodeBase(1, 1) {
-    VLOG(6) << "Construct GradNodeAccumulation";
+    VLOG(5) << "Construct GradNodeAccumulation";
     if (meta) {
       weak_grad_ = meta->WeakGrad();
     }
@@ -33,7 +33,7 @@ class GradNodeAccumulation : public GradNodeBase {
   }
 
   ~GradNodeAccumulation() override {
-    VLOG(6) << "Destruct GradNodeAccumulation";
+    VLOG(5) << "Destruct GradNodeAccumulation";
   }
 
   // Functor: perform backward computations
@@ -44,7 +44,7 @@ class GradNodeAccumulation : public GradNodeBase {
              bool create_graph = false,
              bool is_new_grad = false) override;
 
-  void ClearTensorWrappers() override { VLOG(6) << "Do nothing here now"; }
+  void ClearTensorWrappers() override { VLOG(5) << "Do nothing here now"; }
 
   std::string name() { return "GradNodeAccumulation"; }
 

paddle/fluid/eager/api/manual/eager_manual/dygraph_forward_api.h

@@ -16,10 +16,10 @@
 
 #include "paddle/phi/api/include/tensor.h"
 
-paddle::experimental::Tensor add_n_dygraph_function(
+paddle::experimental::Tensor add_n_ad_func(
     const std::vector<paddle::experimental::Tensor>& x);
 
-paddle::experimental::Tensor conv2d_dygraph_function(
+paddle::experimental::Tensor conv2d_ad_func(
     const paddle::experimental::Tensor& input,
     const paddle::experimental::Tensor& filter,
     std::vector<int> strides,

paddle/fluid/eager/api/manual/eager_manual/forwards/add_n_fwd_func.cc

@@ -23,7 +23,7 @@
 #pragma GCC diagnostic ignored "-Wunused-variable"
 DECLARE_bool(check_nan_inf);
 
-paddle::experimental::Tensor add_n_dygraph_function(
+paddle::experimental::Tensor add_n_ad_func(
     const std::vector<paddle::experimental::Tensor>& x) {
   // Dygraph Record Event
   paddle::platform::RecordEvent dygraph_entrance_record_event(
@@ -46,7 +46,7 @@ paddle::experimental::Tensor add_n_dygraph_function(
       paddle::imperative::AutoCastGuard guard(
           egr::Controller::Instance().GetCurrentTracer(),
           paddle::imperative::AmpLevel::O0);
-      return add_n_dygraph_function(NEW_x);
+      return add_n_ad_func(NEW_x);
     }
   }
 
@@ -56,7 +56,7 @@ paddle::experimental::Tensor add_n_dygraph_function(
   std::vector<egr::AutogradMeta*>* x_autograd_meta = &x_autograd_meta_vec;
   // Forward API Call
   VLOG(3) << "Final State Running: "
-          << "add_n_dygraph_function";
+          << "add_n_ad_func";
   auto api_result = paddle::experimental::add_n(x);
   // Check NaN and Inf if needed
   if (FLAGS_check_nan_inf) {

paddle/fluid/eager/api/manual/eager_manual/forwards/conv2d_fwd_function.cc

@@ -24,7 +24,7 @@
 #pragma GCC diagnostic ignored "-Wunused-variable"
 DECLARE_bool(check_nan_inf);
 
-paddle::experimental::Tensor conv2d_dygraph_function(
+paddle::experimental::Tensor conv2d_ad_func(
     const paddle::experimental::Tensor& input,
     const paddle::experimental::Tensor& filter,
     std::vector<int> strides,
@@ -60,17 +60,17 @@ paddle::experimental::Tensor conv2d_dygraph_function(
       paddle::imperative::AutoCastGuard guard(
           egr::Controller::Instance().GetCurrentTracer(),
           paddle::imperative::AmpLevel::O0);
-      return conv2d_dygraph_function(NEW_input,
-                                     NEW_filter,
-                                     strides,
-                                     paddings,
-                                     paddding_algorithm,
-                                     groups,
-                                     dilations,
-                                     data_format,
-                                     use_addto,
-                                     workspace_size_MB,
-                                     exhaustive_search);
+      return conv2d_ad_func(NEW_input,
+                            NEW_filter,
+                            strides,
+                            paddings,
+                            paddding_algorithm,
+                            groups,
+                            dilations,
+                            data_format,
+                            use_addto,
+                            workspace_size_MB,
+                            exhaustive_search);
     }
   }
 
@@ -89,17 +89,17 @@ paddle::experimental::Tensor conv2d_dygraph_function(
     bool is_enable_tune =
         paddle::imperative::LayoutAutoTune::Instance().UseLayoutAutoTune();
     paddle::imperative::LayoutAutoTune::Instance().DisableLayoutAutoTune();
-    auto out = conv2d_dygraph_function(NEW_input,
-                                       filter,
-                                       strides,
-                                       paddings,
-                                       paddding_algorithm,
-                                       groups,
-                                       dilations,
-                                       data_format,
-                                       use_addto,
-                                       workspace_size_MB,
-                                       exhaustive_search);
+    auto out = conv2d_ad_func(NEW_input,
+                              filter,
+                              strides,
+                              paddings,
+                              paddding_algorithm,
+                              groups,
+                              dilations,
+                              data_format,
+                              use_addto,
+                              workspace_size_MB,
+                              exhaustive_search);
     transformer->SetOutTensorLayout(&out);
     if (is_enable_tune) {
       paddle::imperative::LayoutAutoTune::Instance().EnableLayoutAutoTune();
@@ -115,7 +115,7 @@ paddle::experimental::Tensor conv2d_dygraph_function(
       egr::EagerUtils::nullable_autograd_meta(filter);
   // Forward API Call
   VLOG(3) << "Final State Running: "
-          << "conv2d_dygraph_function";
+          << "conv2d_ad_func";
   auto api_result = paddle::experimental::conv2d(input,
                                                  filter,
                                                  strides,

paddle/fluid/eager/api/manual/eager_manual/nodes/add_n_node.cc

@@ -64,8 +64,7 @@ AddNGradNodeFinal::operator()(
 
   // dygraph function
   for (size_t i = 0; i < returns[0].size(); i++) {
-    returns[0][i] =
-        ::scale_dygraph_function(out_grad, phi::Scalar(1.0), 0.0, true);
+    returns[0][i] = ::scale_ad_func(out_grad, phi::Scalar(1.0), 0.0, true);
   }
 
   // Check NaN and Inf id needed

paddle/fluid/eager/api/manual/fluid_manual/forwards/fused_attention_fwd_func.cc

@@ -531,7 +531,6 @@ fused_attention_dygraph_function(
       egr::EagerUtils::SetHistory(p_autograd_Y, grad_node);
       grad_node->SetGradInMeta(Y, 19);
       egr::EagerUtils::CheckAndRetainGrad(Y);
-
       auto QKVOut_accumulation_node =
           std::make_shared<egr::GradNodeAccumulation>(p_autograd_QKVOut);
       egr::EagerUtils::SetOutRankWithSlot(p_autograd_QKVOut, 0);

paddle/fluid/eager/auto_code_generator/generator/codegen_utils.py

@@ -161,11 +161,24 @@ def str2Hump(text):
     string = str2Hump(string)
     if string.rfind("Grad") == (len(string) - 4):
         string = string[:-4]
-    return f"{string}GradNodeFinal"
+    return f"{string}GradNode"
 
 
 def GetDygraphForwardFunctionName(string):
-    return f"{string}_dygraph_function"
+    return f"{string}_ad_func"
+
+
+def GetDygraphLogName(string):
+
+    def str2Hump(text):
+        arr = filter(None, text.split('_'))
+        res = ''
+        for i in arr:
+            res = res + i[0].upper() + i[1:]
+        return res
+
+    string = str2Hump(string)
+    return string
 
 
 def GetIntermediateAPIFunctionName(string):
@@ -198,7 +211,7 @@ def GetInplacedFunctionName(function_name):
 
 
 def GetForwardFunctionName(string):
-    return f"{string}_dygraph_function"
+    return f"{string}_ad_func"
 
 
 def GetIndent(num):