core: CPU target dispatcher update

- use suffixes like '.avx.cpp'
- added CMake-generated files for '.simd.hpp' optimization approach
- wrap HAL intrinsic headers into separate namespaces for different build flags
- automatic vzeroupper insertion (via CV_INSTRUMENT_REGION macro)

cmake/OpenCVCompilerOptimizations.cmake

@@ -275,6 +275,11 @@ set(CPU_BASELINE_FLAGS "")
 set(CPU_BASELINE_FINAL "")
 set(CPU_DISPATCH_FINAL "")
 
+if(CV_DISABLE_OPTIMIZATION)
+  set(CPU_DISPATCH "")
+  set(CPU_DISPATCH_REQUIRE "")
+endif()
+
 macro(ocv_check_compiler_optimization OPT)
   if(NOT DEFINED CPU_${OPT}_SUPPORTED)
     if((DEFINED CPU_${OPT}_FLAGS_ON AND NOT "x${CPU_${OPT}_FLAGS_ON}" STREQUAL "x") OR CPU_${OPT}_TEST_FILE)
@@ -319,7 +324,7 @@ macro(ocv_check_compiler_optimization OPT)
 endmacro()
 
 foreach(OPT ${CPU_KNOWN_OPTIMIZATIONS})
-  set(CPU_${OPT}_USAGE_COUNT 0 CACHE INTERNAL "" FORCE)
+  set(CPU_${OPT}_USAGE_COUNT 0 CACHE INTERNAL "")
   if(NOT DEFINED CPU_${OPT}_FORCE)
     set(CPU_${OPT}_FORCE "${CPU_${OPT}_IMPLIES}")
   endif()
@@ -515,15 +520,27 @@ macro(ocv_compiler_optimization_process_sources SOURCES_VAR_NAME LIBS_VAR_NAME T
   endforeach()
   foreach(fname ${${SOURCES_VAR_NAME}})
     string(TOLOWER "${fname}" fname_LOWER)
-    if(fname_LOWER MATCHES "[.]opt_.*[.]cpp$")
+    if(fname_LOWER MATCHES "\\.(.*)\\.cpp$")
-      if(CV_DISABLE_OPTIMIZATION OR NOT CV_ENABLE_INTRINSICS)
+      string(TOUPPER "${CMAKE_MATCH_1}" OPT_)
-        message(STATUS "Excluding from source files list: ${fname}")
+      if(OPT_ MATCHES "(CUDA.*|DISPATCH.*|OCL)") # don't touch files like filename.cuda.cpp
+        list(APPEND __result "${fname}")
+        #continue()
+      elseif(CV_DISABLE_OPTIMIZATION OR NOT CV_ENABLE_INTRINSICS)
+        message(STATUS "Excluding from source files list (optimization is disabled): ${fname}")
         #continue()
       else()
+        get_source_file_property(__definitions "${fname}" COMPILE_DEFINITIONS)
+        if(__definitions)
+          list(APPEND __definitions "CV_CPU_DISPATCH_MODE=${OPT_}")
+        else()
+          set(__definitions "CV_CPU_DISPATCH_MODE=${OPT_}")
+        endif()
+        set_source_files_properties("${fname}" PROPERTIES COMPILE_DEFINITIONS "${__definitions}")
+
         set(__opt_found 0)
         foreach(OPT ${CPU_BASELINE_FINAL})
           string(TOLOWER "${OPT}" OPT_LOWER)
-          if(fname_LOWER MATCHES "_${OPT_LOWER}[.]cpp$")
+          if(fname_LOWER MATCHES "\\.${OPT_LOWER}\\.cpp$")
 #message("${fname} BASELINE-${OPT}")
             set(__opt_found 1)
             list(APPEND __result "${fname}")
@@ -533,11 +550,11 @@ macro(ocv_compiler_optimization_process_sources SOURCES_VAR_NAME LIBS_VAR_NAME T
         foreach(OPT ${CPU_DISPATCH_FINAL})
           foreach(OPT2 ${CPU_DISPATCH_${OPT}_FORCED})
             string(TOLOWER "${OPT2}" OPT2_LOWER)
-            if(fname_LOWER MATCHES "_${OPT2_LOWER}[.]cpp$")
+            if(fname_LOWER MATCHES "\\.${OPT2_LOWER}\\.cpp$")
               list(APPEND __result_${OPT} "${fname}")
               math(EXPR CPU_${OPT}_USAGE_COUNT "${CPU_${OPT}_USAGE_COUNT}+1")
               set(CPU_${OPT}_USAGE_COUNT "${CPU_${OPT}_USAGE_COUNT}" CACHE INTERNAL "" FORCE)
-#message("${fname} ${OPT}")
+#message("(${CPU_${OPT}_USAGE_COUNT})${fname} ${OPT}")
 #message("    ${CPU_DISPATCH_${OPT}_INCLUDED}")
 #message("    ${CPU_DISPATCH_DEFINITIONS_${OPT}}")
 #message("    ${CPU_DISPATCH_FLAGS_${OPT}}")
@@ -573,7 +590,13 @@ macro(ocv_compiler_optimization_process_sources SOURCES_VAR_NAME LIBS_VAR_NAME T
         list(APPEND __result "$<TARGET_OBJECTS:${TARGET_BASE_NAME}_${OPT}>")
       else()
         foreach(fname ${__result_${OPT}})
-          set_source_files_properties("${fname}" PROPERTIES COMPILE_DEFINITIONS "${CPU_DISPATCH_DEFINITIONS_${OPT}}")
+          get_source_file_property(__definitions "${fname}" COMPILE_DEFINITIONS)
+          if(__definitions)
+            list(APPEND __definitions "${CPU_DISPATCH_DEFINITIONS_${OPT}}")
+          else()
+            set(__definitions "${CPU_DISPATCH_DEFINITIONS_${OPT}}")
+          endif()
+          set_source_files_properties("${fname}" PROPERTIES COMPILE_DEFINITIONS "${__definitions}")
           set_source_files_properties("${fname}" PROPERTIES COMPILE_FLAGS "${CPU_DISPATCH_FLAGS_${OPT}}")
         endforeach()
         list(APPEND __result ${__result_${OPT}})
@@ -620,18 +643,25 @@ macro(ocv_compiler_optimization_fill_cpu_config)
       set(OPENCV_CPU_CONTROL_DEFINITIONS_CONFIGMAKE "${OPENCV_CPU_CONTROL_DEFINITIONS_CONFIGMAKE}
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_${OPT}
 #  define CV_CPU_HAS_SUPPORT_${OPT} 1
-#  define CV_CPU_CALL_${OPT}(...) return __VA_ARGS__
+#  define CV_CPU_CALL_${OPT}(fn, args) return (opt_${OPT}::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_${OPT}
 #  define CV_CPU_HAS_SUPPORT_${OPT} (cv::checkHardwareSupport(CV_CPU_${OPT}))
-#  define CV_CPU_CALL_${OPT}(...) if (CV_CPU_HAS_SUPPORT_${OPT}) return __VA_ARGS__
+#  define CV_CPU_CALL_${OPT}(fn, args) if (CV_CPU_HAS_SUPPORT_${OPT}) return (opt_${OPT}::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_${OPT} 0
-#  define CV_CPU_CALL_${OPT}(...)
+#  define CV_CPU_CALL_${OPT}(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_${OPT}(fn, args, mode, ...)  CV_CPU_CALL_${OPT}(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 ")
     endif()
   endforeach()
 
+  set(OPENCV_CPU_CONTROL_DEFINITIONS_CONFIGMAKE "${OPENCV_CPU_CONTROL_DEFINITIONS_CONFIGMAKE}
+#define CV_CPU_CALL_BASELINE(fn, args) return (cpu_baseline::fn args)
+#define __CV_CPU_DISPATCH_CHAIN_BASELINE(fn, args, mode, ...)  CV_CPU_CALL_BASELINE(fn, args) /* last in sequence */
+")
+
+
   set(__file "${CMAKE_SOURCE_DIR}/modules/core/include/opencv2/core/cv_cpu_helper.h")
   if(EXISTS "${__file}")
     file(READ "${__file}" __content)
@@ -644,6 +674,57 @@ macro(ocv_compiler_optimization_fill_cpu_config)
   endif()
 endmacro()
 
+macro(ocv_add_dispatched_file filename)
+  if(NOT OPENCV_INITIAL_PASS)
+    set(__codestr "
+#include \"precomp.hpp\"
+#include \"${filename}.simd.hpp\"
+")
+
+    set(__declarations_str "#define CV_CPU_SIMD_FILENAME \"${filename}.simd.hpp\"")
+    set(__dispatch_modes "BASELINE")
+
+    set(__optimizations "${ARGN}")
+    if(CV_DISABLE_OPTIMIZATION OR NOT CV_ENABLE_INTRINSICS)
+      set(__optimizations "")
+    endif()
+
+    foreach(OPT ${__optimizations})
+      string(TOLOWER "${OPT}" OPT_LOWER)
+      set(__file "${CMAKE_CURRENT_BINARY_DIR}/${filename}.${OPT_LOWER}.cpp")
+      if(EXISTS "${__file}")
+        file(READ "${__file}" __content)
+      endif()
+      if(__content STREQUAL __codestr)
+        #message(STATUS "${__file} contains up-to-date content")
+      else()
+        file(WRITE "${__file}" "${__codestr}")
+      endif()
+      list(APPEND OPENCV_MODULE_${the_module}_SOURCES_DISPATCHED "${__file}")
+
+      set(__declarations_str "${__declarations_str}
+#define CV_CPU_DISPATCH_MODE ${OPT}
+#include \"opencv2/core/private/cv_cpu_include_simd_declarations.hpp\"
+")
+      set(__dispatch_modes "${OPT}, ${__dispatch_modes}")
+    endforeach()
+
+    set(__declarations_str "${__declarations_str}
+#define CV_CPU_DISPATCH_MODES_ALL ${__dispatch_modes}
+")
+
+    set(__file "${CMAKE_CURRENT_BINARY_DIR}/${filename}.simd_declarations.hpp")
+    if(EXISTS "${__file}")
+      file(READ "${__file}" __content)
+    endif()
+    if(__content STREQUAL __declarations_str)
+      #message(STATUS "${__file} contains up-to-date content")
+    else()
+      file(WRITE "${__file}" "${__declarations_str}")
+    endif()
+  endif()
+endmacro()
+
 if(CV_DISABLE_OPTIMIZATION OR CV_ICC)
   ocv_update(CV_ENABLE_UNROLLED 0)
 else()

cmake/OpenCVModule.cmake

@@ -314,6 +314,7 @@ macro(ocv_glob_modules)
   set(OPENCV_INITIAL_PASS OFF)
   if(${BUILD_opencv_world})
     foreach(m ${OPENCV_MODULES_BUILD})
+      set(the_module "${m}")
       if("${m}" STREQUAL opencv_world)
         add_subdirectory("${OPENCV_MODULE_opencv_world_LOCATION}" "${CMAKE_CURRENT_BINARY_DIR}/world")
       elseif(NOT OPENCV_MODULE_${m}_IS_PART_OF_WORLD AND NOT ${m} STREQUAL opencv_world)
@@ -329,6 +330,7 @@ macro(ocv_glob_modules)
     endforeach()
   else()
     foreach(m ${OPENCV_MODULES_BUILD})
+      set(the_module "${m}")
       if(m MATCHES "^opencv_")
         string(REGEX REPLACE "^opencv_" "" __shortname "${m}")
         add_subdirectory("${OPENCV_MODULE_${m}_LOCATION}" "${CMAKE_CURRENT_BINARY_DIR}/${__shortname}")
@@ -646,11 +648,13 @@ macro(ocv_set_module_sources)
     ocv_get_module_external_sources()
   endif()
 
+  if(OPENCV_MODULE_${the_module}_SOURCES_DISPATCHED)
+    list(APPEND OPENCV_MODULE_${the_module}_SOURCES ${OPENCV_MODULE_${the_module}_SOURCES_DISPATCHED})
+  endif()
+
   # use full paths for module to be independent from the module location
   ocv_convert_to_full_paths(OPENCV_MODULE_${the_module}_HEADERS)
 
-  ocv_compiler_optimization_process_sources(OPENCV_MODULE_${the_module}_SOURCES OPENCV_MODULE_${the_module}_DEPS_EXT ${the_module})
-
   set(OPENCV_MODULE_${the_module}_HEADERS ${OPENCV_MODULE_${the_module}_HEADERS} CACHE INTERNAL "List of header files for ${the_module}")
   set(OPENCV_MODULE_${the_module}_SOURCES ${OPENCV_MODULE_${the_module}_SOURCES} CACHE INTERNAL "List of source files for ${the_module}")
 endmacro()
@@ -766,6 +770,11 @@ macro(ocv_create_module)
 endmacro()
 
 macro(_ocv_create_module)
+
+  ocv_compiler_optimization_process_sources(OPENCV_MODULE_${the_module}_SOURCES OPENCV_MODULE_${the_module}_DEPS_EXT ${the_module})
+  set(OPENCV_MODULE_${the_module}_HEADERS ${OPENCV_MODULE_${the_module}_HEADERS} CACHE INTERNAL "List of header files for ${the_module}")
+  set(OPENCV_MODULE_${the_module}_SOURCES ${OPENCV_MODULE_${the_module}_SOURCES} CACHE INTERNAL "List of source files for ${the_module}")
+
   # The condition we ought to be testing here is whether ocv_add_precompiled_headers will
   # be called at some point in the future. We can't look into the future, though,
   # so this will have to do.

cmake/OpenCVPCHSupport.cmake

@@ -288,11 +288,12 @@ MACRO(ADD_PRECOMPILED_HEADER _targetName _input)
     foreach(src ${_sources})
       if(NOT "${src}" MATCHES "\\.mm$")
         get_source_file_property(oldProps "${src}" COMPILE_FLAGS)
-        if(NOT oldProps)
+        get_source_file_property(oldProps2 "${src}" COMPILE_DEFINITIONS)
+        if(NOT oldProps AND NOT oldProps2)
           set(newProperties "-include \"${CMAKE_CURRENT_BINARY_DIR}/${_name}\"")
           set_source_files_properties("${src}" PROPERTIES COMPILE_FLAGS "${newProperties}")
         else()
-          ocv_debug_message("Skip PCH, flags: ${oldProps} , file: ${src}")
+          ocv_debug_message("Skip PCH, flags: ${oldProps} defines: ${oldProps2}, file: ${src}")
         endif()
       endif()
     endforeach()
@@ -339,11 +340,12 @@ MACRO(ADD_NATIVE_PRECOMPILED_HEADER _targetName _input)
                AND NOT "${src}" MATCHES "^\$" # CMake generator expressions
           )
             get_source_file_property(oldProps "${src}" COMPILE_FLAGS)
-            if(NOT oldProps)
+            get_source_file_property(oldProps2 "${src}" COMPILE_DEFINITIONS)
+            if(NOT oldProps AND NOT oldProps2)
               set(newProperties "/Yu\"${_input}\" /FI\"${_input}\"")
               set_source_files_properties("${src}" PROPERTIES COMPILE_FLAGS "${newProperties}")
             else()
-              ocv_debug_message("Skip PCH, flags: ${oldProps} , file: ${src}")
+              ocv_debug_message("Skip PCH, flags: ${oldProps} defines: ${oldProps2}, file: ${src}")
             endif()
           endif()
         endforeach()

modules/core/include/opencv2/core/cv_cpu_dispatch.h

@@ -7,6 +7,23 @@
 #include "cv_cpu_config.h"
 #include "cv_cpu_helper.h"
 
+#ifdef CV_CPU_DISPATCH_MODE
+#define CV_CPU_OPTIMIZATION_NAMESPACE __CV_CAT(opt_, CV_CPU_DISPATCH_MODE)
+#define CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN namespace __CV_CAT(opt_, CV_CPU_DISPATCH_MODE) {
+#define CV_CPU_OPTIMIZATION_NAMESPACE_END }
+#else
+#define CV_CPU_OPTIMIZATION_NAMESPACE cpu_baseline
+#define CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN namespace cpu_baseline {
+#define CV_CPU_OPTIMIZATION_NAMESPACE_END }
+#endif
+
+
+#define __CV_CPU_DISPATCH_CHAIN_END(fn, args, mode, ...)  /* done */
+#define __CV_CPU_DISPATCH(fn, args, mode, ...) __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
+#define __CV_CPU_DISPATCH_EXPAND(fn, args, ...) __CV_EXPAND(__CV_CPU_DISPATCH(fn, args, __VA_ARGS__))
+#define CV_CPU_DISPATCH(fn, args, ...) __CV_CPU_DISPATCH_EXPAND(fn, args, __VA_ARGS__, END) // expand macros
+
+
 #if defined CV_ENABLE_INTRINSICS \
     && !defined CV_DISABLE_OPTIMIZATION \
     && !defined __CUDACC__ /* do not include SSE/AVX/NEON headers for NVCC compiler */ \
@@ -76,6 +93,16 @@
 
 #endif // CV_ENABLE_INTRINSICS && !CV_DISABLE_OPTIMIZATION && !__CUDACC__
 
+#if defined CV_CPU_COMPILE_AVX && !defined CV_CPU_BASELINE_COMPILE_AVX
+struct VZeroUpperGuard {
+#ifdef __GNUC__
+    __attribute__((always_inline))
+#endif
+    inline ~VZeroUpperGuard() { _mm256_zeroupper(); }
+};
+#define __CV_AVX_GUARD VZeroUpperGuard __vzeroupper_guard;
+#endif
+
 #endif // __OPENCV_BUILD
 
 

modules/core/include/opencv2/core/cv_cpu_helper.h

@@ -2,132 +2,147 @@
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_SSE
 #  define CV_CPU_HAS_SUPPORT_SSE 1
-#  define CV_CPU_CALL_SSE(...) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE(fn, args) return (opt_SSE::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_SSE
 #  define CV_CPU_HAS_SUPPORT_SSE (cv::checkHardwareSupport(CV_CPU_SSE))
-#  define CV_CPU_CALL_SSE(...) if (CV_CPU_HAS_SUPPORT_SSE) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE(fn, args) if (CV_CPU_HAS_SUPPORT_SSE) return (opt_SSE::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_SSE 0
-#  define CV_CPU_CALL_SSE(...)
+#  define CV_CPU_CALL_SSE(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_SSE(fn, args, mode, ...)  CV_CPU_CALL_SSE(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_SSE2
 #  define CV_CPU_HAS_SUPPORT_SSE2 1
-#  define CV_CPU_CALL_SSE2(...) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE2(fn, args) return (opt_SSE2::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_SSE2
 #  define CV_CPU_HAS_SUPPORT_SSE2 (cv::checkHardwareSupport(CV_CPU_SSE2))
-#  define CV_CPU_CALL_SSE2(...) if (CV_CPU_HAS_SUPPORT_SSE2) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE2(fn, args) if (CV_CPU_HAS_SUPPORT_SSE2) return (opt_SSE2::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_SSE2 0
-#  define CV_CPU_CALL_SSE2(...)
+#  define CV_CPU_CALL_SSE2(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_SSE2(fn, args, mode, ...)  CV_CPU_CALL_SSE2(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_SSE3
 #  define CV_CPU_HAS_SUPPORT_SSE3 1
-#  define CV_CPU_CALL_SSE3(...) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE3(fn, args) return (opt_SSE3::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_SSE3
 #  define CV_CPU_HAS_SUPPORT_SSE3 (cv::checkHardwareSupport(CV_CPU_SSE3))
-#  define CV_CPU_CALL_SSE3(...) if (CV_CPU_HAS_SUPPORT_SSE3) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE3(fn, args) if (CV_CPU_HAS_SUPPORT_SSE3) return (opt_SSE3::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_SSE3 0
-#  define CV_CPU_CALL_SSE3(...)
+#  define CV_CPU_CALL_SSE3(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_SSE3(fn, args, mode, ...)  CV_CPU_CALL_SSE3(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_SSSE3
 #  define CV_CPU_HAS_SUPPORT_SSSE3 1
-#  define CV_CPU_CALL_SSSE3(...) return __VA_ARGS__
+#  define CV_CPU_CALL_SSSE3(fn, args) return (opt_SSSE3::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_SSSE3
 #  define CV_CPU_HAS_SUPPORT_SSSE3 (cv::checkHardwareSupport(CV_CPU_SSSE3))
-#  define CV_CPU_CALL_SSSE3(...) if (CV_CPU_HAS_SUPPORT_SSSE3) return __VA_ARGS__
+#  define CV_CPU_CALL_SSSE3(fn, args) if (CV_CPU_HAS_SUPPORT_SSSE3) return (opt_SSSE3::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_SSSE3 0
-#  define CV_CPU_CALL_SSSE3(...)
+#  define CV_CPU_CALL_SSSE3(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_SSSE3(fn, args, mode, ...)  CV_CPU_CALL_SSSE3(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_SSE4_1
 #  define CV_CPU_HAS_SUPPORT_SSE4_1 1
-#  define CV_CPU_CALL_SSE4_1(...) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE4_1(fn, args) return (opt_SSE4_1::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_SSE4_1
 #  define CV_CPU_HAS_SUPPORT_SSE4_1 (cv::checkHardwareSupport(CV_CPU_SSE4_1))
-#  define CV_CPU_CALL_SSE4_1(...) if (CV_CPU_HAS_SUPPORT_SSE4_1) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE4_1(fn, args) if (CV_CPU_HAS_SUPPORT_SSE4_1) return (opt_SSE4_1::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_SSE4_1 0
-#  define CV_CPU_CALL_SSE4_1(...)
+#  define CV_CPU_CALL_SSE4_1(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_SSE4_1(fn, args, mode, ...)  CV_CPU_CALL_SSE4_1(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_SSE4_2
 #  define CV_CPU_HAS_SUPPORT_SSE4_2 1
-#  define CV_CPU_CALL_SSE4_2(...) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE4_2(fn, args) return (opt_SSE4_2::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_SSE4_2
 #  define CV_CPU_HAS_SUPPORT_SSE4_2 (cv::checkHardwareSupport(CV_CPU_SSE4_2))
-#  define CV_CPU_CALL_SSE4_2(...) if (CV_CPU_HAS_SUPPORT_SSE4_2) return __VA_ARGS__
+#  define CV_CPU_CALL_SSE4_2(fn, args) if (CV_CPU_HAS_SUPPORT_SSE4_2) return (opt_SSE4_2::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_SSE4_2 0
-#  define CV_CPU_CALL_SSE4_2(...)
+#  define CV_CPU_CALL_SSE4_2(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_SSE4_2(fn, args, mode, ...)  CV_CPU_CALL_SSE4_2(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_POPCNT
 #  define CV_CPU_HAS_SUPPORT_POPCNT 1
-#  define CV_CPU_CALL_POPCNT(...) return __VA_ARGS__
+#  define CV_CPU_CALL_POPCNT(fn, args) return (opt_POPCNT::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_POPCNT
 #  define CV_CPU_HAS_SUPPORT_POPCNT (cv::checkHardwareSupport(CV_CPU_POPCNT))
-#  define CV_CPU_CALL_POPCNT(...) if (CV_CPU_HAS_SUPPORT_POPCNT) return __VA_ARGS__
+#  define CV_CPU_CALL_POPCNT(fn, args) if (CV_CPU_HAS_SUPPORT_POPCNT) return (opt_POPCNT::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_POPCNT 0
-#  define CV_CPU_CALL_POPCNT(...)
+#  define CV_CPU_CALL_POPCNT(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_POPCNT(fn, args, mode, ...)  CV_CPU_CALL_POPCNT(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_AVX
 #  define CV_CPU_HAS_SUPPORT_AVX 1
-#  define CV_CPU_CALL_AVX(...) return __VA_ARGS__
+#  define CV_CPU_CALL_AVX(fn, args) return (opt_AVX::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_AVX
 #  define CV_CPU_HAS_SUPPORT_AVX (cv::checkHardwareSupport(CV_CPU_AVX))
-#  define CV_CPU_CALL_AVX(...) if (CV_CPU_HAS_SUPPORT_AVX) return __VA_ARGS__
+#  define CV_CPU_CALL_AVX(fn, args) if (CV_CPU_HAS_SUPPORT_AVX) return (opt_AVX::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_AVX 0
-#  define CV_CPU_CALL_AVX(...)
+#  define CV_CPU_CALL_AVX(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_AVX(fn, args, mode, ...)  CV_CPU_CALL_AVX(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_FP16
 #  define CV_CPU_HAS_SUPPORT_FP16 1
-#  define CV_CPU_CALL_FP16(...) return __VA_ARGS__
+#  define CV_CPU_CALL_FP16(fn, args) return (opt_FP16::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_FP16
 #  define CV_CPU_HAS_SUPPORT_FP16 (cv::checkHardwareSupport(CV_CPU_FP16))
-#  define CV_CPU_CALL_FP16(...) if (CV_CPU_HAS_SUPPORT_FP16) return __VA_ARGS__
+#  define CV_CPU_CALL_FP16(fn, args) if (CV_CPU_HAS_SUPPORT_FP16) return (opt_FP16::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_FP16 0
-#  define CV_CPU_CALL_FP16(...)
+#  define CV_CPU_CALL_FP16(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_FP16(fn, args, mode, ...)  CV_CPU_CALL_FP16(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_AVX2
 #  define CV_CPU_HAS_SUPPORT_AVX2 1
-#  define CV_CPU_CALL_AVX2(...) return __VA_ARGS__
+#  define CV_CPU_CALL_AVX2(fn, args) return (opt_AVX2::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_AVX2
 #  define CV_CPU_HAS_SUPPORT_AVX2 (cv::checkHardwareSupport(CV_CPU_AVX2))
-#  define CV_CPU_CALL_AVX2(...) if (CV_CPU_HAS_SUPPORT_AVX2) return __VA_ARGS__
+#  define CV_CPU_CALL_AVX2(fn, args) if (CV_CPU_HAS_SUPPORT_AVX2) return (opt_AVX2::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_AVX2 0
-#  define CV_CPU_CALL_AVX2(...)
+#  define CV_CPU_CALL_AVX2(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_AVX2(fn, args, mode, ...)  CV_CPU_CALL_AVX2(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_FMA3
 #  define CV_CPU_HAS_SUPPORT_FMA3 1
-#  define CV_CPU_CALL_FMA3(...) return __VA_ARGS__
+#  define CV_CPU_CALL_FMA3(fn, args) return (opt_FMA3::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_FMA3
 #  define CV_CPU_HAS_SUPPORT_FMA3 (cv::checkHardwareSupport(CV_CPU_FMA3))
-#  define CV_CPU_CALL_FMA3(...) if (CV_CPU_HAS_SUPPORT_FMA3) return __VA_ARGS__
+#  define CV_CPU_CALL_FMA3(fn, args) if (CV_CPU_HAS_SUPPORT_FMA3) return (opt_FMA3::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_FMA3 0
-#  define CV_CPU_CALL_FMA3(...)
+#  define CV_CPU_CALL_FMA3(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_FMA3(fn, args, mode, ...)  CV_CPU_CALL_FMA3(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
 
 #if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_NEON
 #  define CV_CPU_HAS_SUPPORT_NEON 1
-#  define CV_CPU_CALL_NEON(...) return __VA_ARGS__
+#  define CV_CPU_CALL_NEON(fn, args) return (opt_NEON::fn args)
 #elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_NEON
 #  define CV_CPU_HAS_SUPPORT_NEON (cv::checkHardwareSupport(CV_CPU_NEON))
-#  define CV_CPU_CALL_NEON(...) if (CV_CPU_HAS_SUPPORT_NEON) return __VA_ARGS__
+#  define CV_CPU_CALL_NEON(fn, args) if (CV_CPU_HAS_SUPPORT_NEON) return (opt_NEON::fn args)
 #else
 #  define CV_CPU_HAS_SUPPORT_NEON 0
-#  define CV_CPU_CALL_NEON(...)
+#  define CV_CPU_CALL_NEON(fn, args)
 #endif
+#define __CV_CPU_DISPATCH_CHAIN_NEON(fn, args, mode, ...)  CV_CPU_CALL_NEON(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
+
+#define CV_CPU_CALL_BASELINE(fn, args) return (cpu_baseline::fn args)
+#define __CV_CPU_DISPATCH_CHAIN_BASELINE(fn, args, mode, ...)  CV_CPU_CALL_BASELINE(fn, args) /* last in sequence */

modules/core/include/opencv2/core/cvdef.h

@@ -52,6 +52,17 @@
 #include "cvconfig.h"
 #endif
 
+#ifndef __CV_EXPAND
+#define __CV_EXPAND(x) x
+#endif
+
+#ifndef __CV_CAT
+#define __CV_CAT__(x, y) x ## y
+#define __CV_CAT_(x, y) __CV_CAT__(x, y)
+#define __CV_CAT(x, y) __CV_CAT_(x, y)
+#endif
+
+
 #if !defined _CRT_SECURE_NO_DEPRECATE && defined _MSC_VER && _MSC_VER > 1300
 #  define _CRT_SECURE_NO_DEPRECATE /* to avoid multiple Visual Studio warnings */
 #endif

modules/core/include/opencv2/core/hal/intrin.hpp

@@ -60,6 +60,25 @@
 // access from within opencv code more accessible
 namespace cv {
 
+#ifndef CV_DOXYGEN
+
+#ifdef CV_CPU_DISPATCH_MODE
+#define CV_CPU_OPTIMIZATION_HAL_NAMESPACE __CV_CAT(hal_, CV_CPU_DISPATCH_MODE)
+#define CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN namespace __CV_CAT(hal_, CV_CPU_DISPATCH_MODE) {
+#define CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END }
+#else
+#define CV_CPU_OPTIMIZATION_HAL_NAMESPACE hal_baseline
+#define CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN namespace hal_baseline {
+#define CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END }
+#endif
+
+
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
+using namespace CV_CPU_OPTIMIZATION_HAL_NAMESPACE;
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
+#endif
+
 //! @addtogroup core_hal_intrin
 //! @{
 
@@ -281,6 +300,9 @@ template <typename T> struct V_SIMD128Traits
 
 //! @}
 
+#ifndef CV_DOXYGEN
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
+#endif
 }
 
 #ifdef CV_DOXYGEN
@@ -323,6 +345,10 @@ template <typename T> struct V_SIMD128Traits
 
 namespace cv {
 
+#ifndef CV_DOXYGEN
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
+#endif
+
 template <typename R> struct V_RegTrait128;
 
 template <> struct V_RegTrait128<uchar> {
@@ -407,6 +433,10 @@ template <> struct V_RegTrait128<double> {
 };
 #endif
 
+#ifndef CV_DOXYGEN
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
+#endif
+
 } // cv::
 
 //! @endcond

modules/core/include/opencv2/core/hal/intrin_cpp.hpp

@@ -53,6 +53,10 @@
 namespace cv
 {
 
+#ifndef CV_DOXYGEN
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
+#endif
+
 /** @addtogroup core_hal_intrin
 
 "Universal intrinsics" is a types and functions set intended to simplify vectorization of code on
@@ -1827,7 +1831,9 @@ static inline bool hasSIMD128()
 
 //! @}
 
-
+#ifndef CV_DOXYGEN
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
+#endif
 }
 
 #endif

modules/core/include/opencv2/core/hal/intrin_neon.hpp

@@ -53,6 +53,8 @@ namespace cv
 
 //! @cond IGNORED
 
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
+
 #define CV_SIMD128 1
 #if defined(__aarch64__)
 #define CV_SIMD128_64F 1
@@ -1238,11 +1240,13 @@ inline v_float16x4 v_cvt_f16(const v_float32x4& a)
 //! @brief Check CPU capability of SIMD operation
 static inline bool hasSIMD128()
 {
-    return checkHardwareSupport(CV_CPU_NEON);
+    return (CV_CPU_HAS_SUPPORT_NEON) ? true : false;
 }
 
 //! @}
 
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
+
 //! @endcond
 
 }

modules/core/include/opencv2/core/hal/intrin_sse.hpp

@@ -56,6 +56,8 @@ namespace cv
 
 //! @cond IGNORED
 
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
+
 struct v_uint8x16
 {
     typedef uchar lane_type;
@@ -1791,11 +1793,13 @@ inline v_float16x4 v_cvt_f16(const v_float32x4& a)
 //! @brief Check CPU capability of SIMD operation
 static inline bool hasSIMD128()
 {
-    return checkHardwareSupport(CV_CPU_SSE2);
+    return (CV_CPU_HAS_SUPPORT_SSE2) ? true : false;
 }
 
 //! @}
 
+CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
+
 //! @endcond
 
 }

modules/core/include/opencv2/core/private.hpp

@@ -540,7 +540,7 @@ CV_EXPORTS InstrNode*   getCurrentNode();
 
 ///// General instrumentation
 // General OpenCV region instrumentation macro
-#define CV_INSTRUMENT_REGION()              CV_INSTRUMENT_REGION_META(__FUNCTION__, false, ::cv::instr::TYPE_GENERAL, ::cv::instr::IMPL_PLAIN)
+#define CV_INSTRUMENT_REGION_()             CV_INSTRUMENT_REGION_META(__FUNCTION__, false, ::cv::instr::TYPE_GENERAL, ::cv::instr::IMPL_PLAIN)
 // Custom OpenCV region instrumentation macro
 #define CV_INSTRUMENT_REGION_NAME(NAME)     CV_INSTRUMENT_REGION_CUSTOM_META(NAME,  false, ::cv::instr::TYPE_GENERAL, ::cv::instr::IMPL_PLAIN)
 // Instrumentation for parallel_for_ or other regions which forks and gathers threads
@@ -566,7 +566,7 @@ CV_EXPORTS InstrNode*   getCurrentNode();
 #else
 #define CV_INSTRUMENT_REGION_META(...)
 
-#define CV_INSTRUMENT_REGION()
+#define CV_INSTRUMENT_REGION_()
 #define CV_INSTRUMENT_REGION_NAME(...)
 #define CV_INSTRUMENT_REGION_MT_FORK()
 
@@ -580,6 +580,12 @@ CV_EXPORTS InstrNode*   getCurrentNode();
 #define CV_INSTRUMENT_MARK_OPENCL(...)
 #endif
 
+#ifdef __CV_AVX_GUARD
+#define CV_INSTRUMENT_REGION() __CV_AVX_GUARD CV_INSTRUMENT_REGION_()
+#else
+#define CV_INSTRUMENT_REGION() CV_INSTRUMENT_REGION_()
+#endif
+
 //! @endcond
 
 #endif // OPENCV_CORE_PRIVATE_HPP

modules/core/include/opencv2/core/private/cv_cpu_include_simd_declarations.hpp

@@ -0,0 +1,30 @@
+// Helper file to include dispatched functions declaration:
+//
+// Usage:
+//     #define CV_CPU_SIMD_FILENAME "<filename>.simd.hpp"
+//     #define CV_CPU_DISPATCH_MODE AVX2
+//     #include "opencv2/core/private/cv_cpu_include_simd_declarations.hpp"
+//     #define CV_CPU_DISPATCH_MODE SSE2
+//     #include "opencv2/core/private/cv_cpu_include_simd_declarations.hpp"
+
+#ifndef CV_DISABLE_OPTIMIZATION
+#ifdef _MSC_VER
+#pragma warning(disable: 4702) // unreachable code
+#endif
+#endif
+
+#ifndef CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
+#define CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
+#endif
+
+#undef CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
+#undef CV_CPU_OPTIMIZATION_NAMESPACE_END
+
+#define CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN namespace __CV_CAT(opt_, CV_CPU_DISPATCH_MODE) {
+#define CV_CPU_OPTIMIZATION_NAMESPACE_END }
+
+#include CV_CPU_SIMD_FILENAME
+
+#undef CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
+#undef CV_CPU_OPTIMIZATION_NAMESPACE_END
+#undef CV_CPU_DISPATCH_MODE

modules/world/CMakeLists.txt

@@ -24,6 +24,7 @@ if(NOT OPENCV_INITIAL_PASS)
 
   message(STATUS "Processing WORLD modules...")
   foreach(m ${OPENCV_MODULES_BUILD})
+    set(the_module ${m})
     if(OPENCV_MODULE_${m}_IS_PART_OF_WORLD)
       message(STATUS "    module ${m}...")
       set(CMAKE_CURRENT_SOURCE_DIR "${OPENCV_MODULE_${m}_LOCATION}")