Compilation optimization (#44242)

* Compilation optimization

paddle/fluid/eager/api/generated/eager_generated/backwards/CMakeLists.txt

@@ -6,7 +6,7 @@ cc_library(
 if(NOT (NOT WITH_PYTHON AND ON_INFER))
   cc_library(
     final_dygraph_node
-    SRCS nodes.cc
-    DEPS ${eager_deps} ${eager_manual_nodes})
+    SRCS nodes.cc ${eager_manual_nodes}
+    DEPS ${eager_deps})
   add_dependencies(final_dygraph_node eager_final_state_codegen)
 endif()

paddle/fluid/eager/api/generated/eager_generated/forwards/CMakeLists.txt

@@ -6,7 +6,7 @@ cc_library(
 if(NOT (NOT WITH_PYTHON AND ON_INFER))
   cc_library(
     final_dygraph_function
-    SRCS dygraph_functions.cc
-    DEPS ${eager_deps} ${eager_manual_functions})
+    SRCS dygraph_functions.cc ${eager_manual_functions}
+    DEPS ${eager_deps})
   add_dependencies(final_dygraph_function eager_final_state_codegen)
 endif()

paddle/fluid/eager/api/manual/eager_manual/forwards/CMakeLists.txt

@@ -1,17 +1,4 @@
-cc_library(
-  add_n_fwd_func
-  SRCS add_n_fwd_func.cc
-  DEPS ${eager_deps} ${fluid_deps} ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS})
-
-add_dependencies(add_n_fwd_func eager_codegen)
-
-cc_library(
-  conv2d_fwd_function
-  SRCS conv2d_fwd_function.cc
-  DEPS ${eager_deps} ${fluid_deps} ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS})
-
-add_dependencies(conv2d_fwd_function eager_codegen)
-
 set(eager_manual_functions
-    conv2d_fwd_function add_n_fwd_func
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/eager_manual/forwards/add_n_fwd_func.cc
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/eager_manual/forwards/conv2d_fwd_function.cc
     PARENT_SCOPE)

paddle/fluid/eager/api/manual/eager_manual/nodes/CMakeLists.txt

@@ -1,13 +1,4 @@
-cc_library(
-  add_n_node
-  SRCS add_n_node.cc
-  DEPS ${eager_deps} ${fluid_deps})
-
-cc_library(
-  conv2d_nodes
-  SRCS conv2d_nodes.cc
-  DEPS ${eager_deps} ${fluid_deps})
-
 set(eager_manual_nodes
-    conv2d_nodes add_n_node
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/eager_manual/nodes/conv2d_nodes.cc
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/eager_manual/nodes/add_n_node.cc
     PARENT_SCOPE)

paddle/fluid/eager/api/manual/fluid_manual/forwards/CMakeLists.txt

@@ -1,28 +1,5 @@
-cc_library(
-  fused_gate_attention_fwd_func
-  SRCS fused_gate_attention_fwd_func.cc
-  DEPS ${eager_deps} ${fluid_deps} ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS})
-
-add_dependencies(fused_gate_attention_fwd_func eager_codegen
-                 copy_dygraph_forward_functions)
-
-cc_library(
-  fused_feedforward_fwd_func
-  SRCS fused_feedforward_fwd_func.cc
-  DEPS ${eager_deps} ${fluid_deps} ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS})
-
-add_dependencies(fused_feedforward_fwd_func eager_codegen
-                 copy_dygraph_forward_functions)
-
-cc_library(
-  fused_attention_fwd_func
-  SRCS fused_attention_fwd_func.cc
-  DEPS ${eager_deps} ${fluid_deps} ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS})
-
-add_dependencies(fused_attention_fwd_func eager_codegen
-                 copy_dygraph_forward_functions)
-
 set(fluid_manual_functions
-    fused_gate_attention_fwd_func fused_feedforward_fwd_func
-    fused_attention_fwd_func
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/fluid_manual/forwards/fused_gate_attention_fwd_func.cc
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/fluid_manual/forwards/fused_feedforward_fwd_func.cc
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/fluid_manual/forwards/fused_attention_fwd_func.cc
     PARENT_SCOPE)

paddle/fluid/eager/api/manual/fluid_manual/nodes/CMakeLists.txt

@@ -1,18 +1,5 @@
-cc_library(
-  fused_gate_attention_node
-  SRCS fused_gate_attention_node.cc
-  DEPS ${eager_deps} ${fluid_deps})
-
-cc_library(
-  fused_feedforward_node
-  SRCS fused_feedforward_node.cc
-  DEPS ${eager_deps} ${fluid_deps})
-
-cc_library(
-  fused_attention_node
-  SRCS fused_attention_node.cc
-  DEPS ${eager_deps} ${fluid_deps})
-
 set(fluid_manual_nodes
-    fused_gate_attention_node fused_feedforward_node fused_attention_node
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/fluid_manual/nodes/fused_gate_attention_node.cc
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/fluid_manual/nodes/fused_feedforward_node.cc
+    ${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/manual/fluid_manual/nodes/fused_attention_node.cc
     PARENT_SCOPE)

paddle/fluid/eager/auto_code_generator/eager_generator.cc

@@ -3083,27 +3083,44 @@ static std::string ConvertCoreOpsInfosToString(
   return core_ops_returns_info_init_str;
 }
 
-static std::string GenerateCoreOpsReturnsInfo() {
+static std::string GenerateCoreOpsArgsInfo() {
   const char* Core_Ops_Returns_MAP_TEMPLATE =
       "std::unordered_map<std::string, std::vector<std::string>> "
-      "core_ops_args_info = { %s };\n"
-      "std::unordered_map<std::string, std::vector<std::string>> "
-      "core_ops_args_type_info = { %s };\n"
-      "std::unordered_map<std::string, std::vector<std::string>> "
-      "core_ops_returns_info = { %s };\n";
+      "core_ops_args_info = { %s };\n";
 
   std::string core_ops_args_info_init_str =
       ConvertCoreOpsInfosToString(core_ops_args_info);
+
+  std::string core_ops_info_str = paddle::string::Sprintf(
+      Core_Ops_Returns_MAP_TEMPLATE, core_ops_args_info_init_str);
+
+  return core_ops_info_str;
+}
+
+static std::string GenerateCoreOpsArgsTypeInfo() {
+  const char* Core_Ops_Returns_MAP_TEMPLATE =
+      "std::unordered_map<std::string, std::vector<std::string>> "
+      "core_ops_args_type_info = { %s };\n";
+
   std::string core_ops_args_type_info_init_str =
       ConvertCoreOpsInfosToString(core_ops_args_type_info);
+
+  std::string core_ops_info_str = paddle::string::Sprintf(
+      Core_Ops_Returns_MAP_TEMPLATE, core_ops_args_type_info_init_str);
+
+  return core_ops_info_str;
+}
+
+static std::string GenerateCoreOpsReturnsInfo() {
+  const char* Core_Ops_Returns_MAP_TEMPLATE =
+      "std::unordered_map<std::string, std::vector<std::string>> "
+      "core_ops_returns_info = { %s };\n";
+
   std::string core_ops_returns_info_init_str =
       ConvertCoreOpsInfosToString(core_ops_returns_info);
 
-  std::string core_ops_info_str =
-      paddle::string::Sprintf(Core_Ops_Returns_MAP_TEMPLATE,
-                              core_ops_args_info_init_str,
-                              core_ops_args_type_info_init_str,
-                              core_ops_returns_info_init_str);
+  std::string core_ops_info_str = paddle::string::Sprintf(
+      Core_Ops_Returns_MAP_TEMPLATE, core_ops_returns_info_init_str);
 
   return core_ops_info_str;
 }
@@ -3252,6 +3269,12 @@ static void DygraphCodeGeneration(const std::string& output_dir,
 
   GenerateForwardDygraphFile(
       output_dir + "/forwards/dygraph_forward_functions_args_info.tmp.cc",
+      GenerateCoreOpsArgsInfo());
+  GenerateForwardDygraphFile(
+      output_dir + "/forwards/dygraph_forward_functions_args_type_info.tmp.cc",
+      GenerateCoreOpsArgsTypeInfo());
+  GenerateForwardDygraphFile(
+      output_dir + "/forwards/dygraph_forward_functions_returns_info.tmp.cc",
       GenerateCoreOpsReturnsInfo());
 
   VLOG(6) << "-------- GenerateNodeCCFile -------";

paddle/fluid/eager/auto_code_generator/generate_file_structures.py

@@ -96,6 +96,11 @@ def GenerateFileStructureForIntermediateDygraph(eager_dir, split_count):
                                         "nodes" + str(i + 1) + ".cc"))
     empty_files.append(
         os.path.join(forwards_dir, "dygraph_forward_functions_args_info.cc"))
+    empty_files.append(
+        os.path.join(forwards_dir,
+                     "dygraph_forward_functions_args_type_info.cc"))
+    empty_files.append(
+        os.path.join(forwards_dir, "dygraph_forward_functions_returns_info.cc"))
     for path in empty_files:
         if not os.path.exists(path):
             open(path, 'a').close()
@@ -125,7 +130,7 @@ def GenerateFileStructureForIntermediateDygraph(eager_dir, split_count):
         f.write("cc_library(dygraph_node SRCS ")
         for i in range(split_count):
             f.write("nodes" + str(i + 1) + ".cc ")
-        f.write("DEPS ${eager_deps} ${fluid_deps} ${fluid_manual_nodes})\n")
+        f.write("${fluid_manual_nodes} DEPS ${eager_deps} ${fluid_deps})\n")
         f.write("add_dependencies(dygraph_node copy_dygraph_node)")
 
     with open(forwards_level_cmakelist_path, "w") as f:
@@ -143,15 +148,23 @@ def GenerateFileStructureForIntermediateDygraph(eager_dir, split_count):
         f.write(
             "  COMMAND ${CMAKE_COMMAND} -E copy_if_different \"${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/generated/fluid_generated/forwards/dygraph_forward_functions_args_info.tmp.cc\" \"${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/generated/fluid_generated/forwards/dygraph_forward_functions_args_info.cc\"\n"
         )
+        f.write(
+            "  COMMAND ${CMAKE_COMMAND} -E copy_if_different \"${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/generated/fluid_generated/forwards/dygraph_forward_functions_args_type_info.tmp.cc\" \"${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/generated/fluid_generated/forwards/dygraph_forward_functions_args_type_info.cc\"\n"
+        )
+        f.write(
+            "  COMMAND ${CMAKE_COMMAND} -E copy_if_different \"${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/generated/fluid_generated/forwards/dygraph_forward_functions_returns_info.tmp.cc\" \"${PADDLE_SOURCE_DIR}/paddle/fluid/eager/api/generated/fluid_generated/forwards/dygraph_forward_functions_returns_info.cc\"\n"
+        )
         f.write("  DEPENDS eager_codegen\n")
         f.write("  VERBATIM)\n")
 
         f.write("cc_library(dygraph_function SRCS ")
         for i in range(split_count):
             f.write("dygraph_forward_functions" + str(i + 1) + ".cc ")
         f.write("dygraph_forward_functions_args_info.cc ")
+        f.write("dygraph_forward_functions_args_type_info.cc ")
+        f.write("dygraph_forward_functions_returns_info.cc ")
         f.write(
-            "DEPS ${eager_deps} ${fluid_deps} ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS} ${fluid_manual_functions})\n"
+            "${fluid_manual_functions} DEPS ${eager_deps} ${fluid_deps} ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS})\n"
         )
         f.write(
             "add_dependencies(dygraph_function copy_dygraph_forward_functions)")

paddle/fluid/framework/new_executor/CMakeLists.txt

@@ -1,4 +1,16 @@
-set(INTERPRETERCORE_DEPS
+add_subdirectory(workqueue)
+add_subdirectory(garbage_collector)
+
+set(STANDALONE_EXECUTOR_SRCS
+    data_transfer.cc
+    new_executor_defs.cc
+    interpretercore_util.cc
+    event_manager.cc
+    stream_analyzer.cc
+    interpretercore.cc
+    standalone_executor.cc)
+
+set(STANDALONE_EXECUTOR_DEPS
     op_registry
     device_context
     scope
@@ -20,62 +32,33 @@ set(INTERPRETERCORE_DEPS
     variable_helper
     timer
     monitor
-    nan_inf_utils)
-
-add_subdirectory(workqueue)
-add_subdirectory(garbage_collector)
-
-cc_library(
-  data_transfer
-  SRCS data_transfer.cc
-  DEPS enforce scope glog)
-cc_library(
-  new_executor_defs
-  SRCS new_executor_defs.cc
-  DEPS enforce glog scope)
-cc_library(
-  interpretercore_util
-  SRCS interpretercore_util.cc
-  DEPS ${INTERPRETERCORE_DEPS} workqueue new_executor_defs data_transfer)
-cc_library(
-  event_manager
-  SRCS event_manager.cc
-  DEPS ${DEVICE_EVENT_LIBS} glog new_executor_defs)
-cc_library(
-  stream_analyzer
-  SRCS stream_analyzer.cc
-  DEPS ${DEVICE_EVENT_LIBS} glog device_context new_executor_defs)
+    nan_inf_utils
+    enforce
+    scope
+    glog
+    enforce
+    glog
+    scope
+    workqueue
+    interpretercore_event_garbage_collector
+    ${DEVICE_EVENT_LIBS}
+    glog)
 
 if(WITH_GPU OR WITH_ROCM)
-  cc_library(
-    interpretercore
-    SRCS interpretercore.cc
-    DEPS workqueue
-         ${DEVICE_EVENT_LIBS}
-         interpretercore_util
-         interpretercore_event_garbage_collector
-         interpretercore_fast_garbage_collector
-         stream_analyzer
-         event_manager)
-else()
-  cc_library(
-    interpretercore
-    SRCS interpretercore.cc
-    DEPS workqueue ${DEVICE_EVENT_LIBS} interpretercore_util
-         interpretercore_event_garbage_collector stream_analyzer event_manager)
+  set(STANDALONE_EXECUTOR_DEPS ${STANDALONE_EXECUTOR_DEPS}
+                               interpretercore_fast_garbage_collector)
 endif()
 
 cc_library(
   standalone_executor
-  SRCS standalone_executor.cc
-  DEPS interpretercore)
+  SRCS ${STANDALONE_EXECUTOR_SRCS}
+  DEPS ${STANDALONE_EXECUTOR_DEPS})
 
 cc_library(
   staticgraph_executor_statistics
   SRCS executor_statistics.cc
   DEPS enforce glog os_info)
 
-# cc_binary(standalone_executor_test SRCS standalone_executor_test.cc DEPS interpretercore standalone_executor operator op_registry executor ${GLOB_OP_LIB} ${GLOB_OPERATOR_DEPS} profiler)
 # skip win32 since wget is not installed by default on windows machine.
 if(WITH_GPU
    AND WITH_TESTING
@@ -120,13 +103,7 @@ if(WITH_GPU
   cc_test(
     standalone_executor_test
     SRCS standalone_executor_test.cc
-    DEPS interpretercore
-         standalone_executor
-         operator
-         op_registry
-         executor
-         ${OPS}
-         ${OP_DEPS})
+    DEPS standalone_executor operator op_registry executor ${OPS} ${OP_DEPS})
   set_tests_properties(standalone_executor_test PROPERTIES TIMEOUT 100)
 
   add_dependencies(standalone_executor_test download_program)

paddle/fluid/imperative/CMakeLists.txt

@@ -5,7 +5,7 @@ cc_library(
 cc_library(
   var_helper
   SRCS var_helper.cc
-  DEPS tensor phi_api)
+  DEPS tensor selected_rows)
 if(WITH_XPU)
   cc_library(
     prepared_operator
@@ -20,8 +20,8 @@ if(WITH_XPU)
          op_kernel_type
          data_transform
          nan_inf_utils
-         phi_api
-         phi_utils
+         scalar
+         int_array
          var_helper
          profiler)
 else()
@@ -37,21 +37,16 @@ else()
          op_kernel_type
          data_transform
          nan_inf_utils
-         phi_api
-         phi_utils
+         scalar
+         int_array
          var_helper
          profiler)
 endif()
 cc_library(
   layer
   SRCS layer.cc
-  DEPS prepared_operator
-       math_function
-       imperative_flag
-       variable_helper
-       op_registry
-       var_helper
-       phi_api)
+  DEPS prepared_operator math_function imperative_flag variable_helper
+       op_registry var_helper)
 add_subdirectory(jit)
 if(WITH_GPU)
   cc_library(

paddle/fluid/operators/CMakeLists.txt

@@ -101,7 +101,7 @@ else()
     cc_library(gather_scatter_kernel SRCS gather_scatter_kernel.cc gather_scatter_kernel.cu DEPS tensor)
 endif()
 
-set(OP_HEADER_DEPS ${OP_HEADER_DEPS} phi phi_api_utils gather_scatter_kernel)
+set(OP_HEADER_DEPS ${OP_HEADER_DEPS} phi phi_api_utils gather_scatter_kernel backward_infermeta)
 
 register_operators(EXCLUDES py_layer_op py_func_op warpctc_op dgc_op load_combine_op lstm_op run_program_op eye_op quantize_linear_op
         recurrent_op save_combine_op sparse_attention_op sync_batch_norm_op spectral_op ${OP_MKL_DEPS} DEPS ${OP_HEADER_DEPS})

paddle/fluid/platform/device/gpu/cuda/CMakeLists.txt

@@ -10,4 +10,4 @@ nv_library(
 nv_test(
   cudnn_helper_test
   SRCS cudnn_helper_test.cc
-  DEPS dynload_cuda phi)
+  DEPS dynload_cuda)