!5311 add performence compile option

Merge pull request !5311 from hangq/primitive

mindspore/lite/CMakeLists.txt

@@ -113,12 +113,11 @@ if (WIN32)
     add_compile_definitions(BUILDING_DLL)
 endif ()
 
-set(ANF_SRC
-        ${CMAKE_CURRENT_SOURCE_DIR}/../core/ir/meta_tensor.cc
+set(CORE_SRC
+        ${CORE_DIR}/ir/meta_tensor.cc
         ${CORE_DIR}/gvar/logging_level.cc
         ${CORE_DIR}/gvar/typeid_manager.cc
-        ${CMAKE_CURRENT_SOURCE_DIR}/../core/base/base.cc
-        ${CMAKE_CURRENT_SOURCE_DIR}/src/common/log_adapter.cc
+        ${CORE_DIR}/base/base.cc
         )
 if (BUILD_CONVERTER)
     if (PLATFORM_ARM64 OR PLATFORM_ARM32)

mindspore/lite/src/CMakeLists.txt

@@ -1,11 +1,21 @@
+if (PLATFORM_ARM32 OR PLATFORM_ARM64)
+  # for performance
+  if ("${CMAKE_BUILD_TYPE}" STREQUAL "Release")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fomit-frame-pointer -fstrict-aliasing -ffunction-sections -fdata-sections -ffast-math")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fomit-frame-pointer -fstrict-aliasing -ffunction-sections -fdata-sections -ffast-math")
+  endif()
+endif ()
+
 set(LITE_SRC
         ${CMAKE_CURRENT_SOURCE_DIR}/common/graph_util.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/common/ms_tensor_utils.cc
+        ${CMAKE_CURRENT_SOURCE_DIR}/common/log_adapter.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/runtime/allocator.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/runtime/runtime_api.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/runtime/thread_pool.c
         ${CMAKE_CURRENT_SOURCE_DIR}/runtime/workspace_pool.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/ir/tensor.cc
+        ${CMAKE_CURRENT_SOURCE_DIR}/ir/meta_tensor_extends.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/context.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/executor.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/kernel_registry.cc
@@ -28,24 +38,20 @@ if (SUPPORT_GPU)
       )
 endif ()
 
-set(ANF_SRC
-    ${ANF_SRC}
-    ${CMAKE_CURRENT_SOURCE_DIR}/ir/meta_tensor_extends.cc
-    )
 file(GLOB_RECURSE C_OPS_SRC ${CMAKE_CURRENT_SOURCE_DIR}/ops/*.cc)
 add_library(mindspore-lite SHARED ${LITE_SRC} ${C_OPS_SRC})
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-unused-private-field")
-add_library(anf OBJECT ${ANF_SRC})
+add_library(core_mid_ OBJECT ${CORE_SRC})
 if (SUPPORT_GPU)
   add_subdirectory(runtime/kernel/opencl)
   target_link_libraries(mindspore-lite
-      anf
+core_mid_
       cpu_kernel_mid_
       opencl_kernel_lib_
       )
 else ()
   target_link_libraries(mindspore-lite
-      anf
+      core_mid_
       cpu_kernel_mid_
       )
 endif ()

mindspore/lite/tools/converter/CMakeLists.txt

@@ -1,6 +1,6 @@
 add_definitions(-DPRIMITIVE_WRITEABLE)
-set(ANF_SRC
-        ${ANF_SRC}
+set(CORE_SRC
+        ${CORE_SRC}
         #core / abstract
         ${CMAKE_CURRENT_SOURCE_DIR}/../../../core/abstract/abstract_function.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/../../../core/abstract/analysis_context.cc
@@ -47,6 +47,8 @@ set(ANF_SRC
         ${CMAKE_CURRENT_SOURCE_DIR}/../../../core/utils/info.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/../../../core/utils/profile.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/../../../core/utils/ms_context.cc
+        )
+set(CCSRC_SRC
         ## ccsrc
         ${CCSRC_DIR}/debug/draw.cc
         ${CCSRC_DIR}/pybind_api/export_flags.cc
@@ -55,8 +57,7 @@ set(ANF_SRC
         ${CCSRC_DIR}/backend/optimizer/common/pattern_engine.cc
         ${CCSRC_DIR}/backend/optimizer/common/visit.cc
         ${CCSRC_DIR}/backend/optimizer/common/optimizer.cc
-        ${CMAKE_CURRENT_SOURCE_DIR}/../../src/common/graph_utils_extends.cc
-        )
+)
 
 if (WIN32)
     set(LITE_SRC
@@ -118,6 +119,8 @@ add_subdirectory(quantizer)
 
 set(SRC_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../../src)
 set(LITE_SRC
+        ${SRC_DIR}/common/graph_utils_extends.cc
+        ${SRC_DIR}/common/log_adapter.cc
         ${SRC_DIR}/common/graph_util.cc
         ${SRC_DIR}/common/ms_tensor_utils.cc
         ${SRC_DIR}/runtime/allocator.cc
@@ -163,7 +166,8 @@ endif()
 
 add_executable(converter_lite
         main.cc
-        ${ANF_SRC}
+        ${CORE_SRC}
+        ${CCSRC_SRC}
         ${CONVERTER_SRC}
         ${OPS_SRC}
         ${KERNEL_SRC}
@@ -172,7 +176,8 @@ add_executable(converter_lite
 
 if (WIN32)
     add_library(converter_mid OBJECT
-            ${ANF_SRC}
+            ${CORE_SRC}
+            ${CCSRC_SRC}
             ${CONVERTER_SRC}
             ${OPS_SRC}
             ${KERNEL_SRC}