Merge pull request #8441 from alalek:dispatch_mathfuncs_core

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(CV_DISABLE_OPTIMIZATION OR NOT CV_ENABLE_INTRINSICS)
-        message(STATUS "Excluding from source files list: ${fname}")
+    if(fname_LOWER MATCHES "\\.(.*)\\.cpp$")
+      string(TOUPPER "${CMAKE_MATCH_1}" OPT_)
+      if(OPT_ MATCHES "(CUDA.*|DISPATCH.*|OCL)") # don't touch files like filename.cuda.cpp
+        list(APPEND __result "${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/CMakeLists.txt

 set(the_description "The Core Functionality")
+
+ocv_add_dispatched_file(mathfuncs_core SSE2 AVX AVX2)
+
 ocv_add_module(core
                "${OPENCV_HAL_LINKER_LIBS}"
                OPTIONAL opencv_cudev

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

+// 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/core/src/mathfuncs_core.dispatch.cpp

+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+#include "precomp.hpp"
+
+#include "mathfuncs_core.simd.hpp"
+#include "mathfuncs_core.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
+
+namespace cv { namespace hal {
+
+///////////////////////////////////// ATAN2 ////////////////////////////////////
+
+void fastAtan32f(const float *Y, const float *X, float *angle, int len, bool angleInDegrees )
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(fastAtan32f, cv_hal_fastAtan32f, Y, X, angle, len, angleInDegrees);
+
+    CV_CPU_DISPATCH(fastAtan32f, (Y, X, angle, len, angleInDegrees),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+void fastAtan64f(const double *Y, const double *X, double *angle, int len, bool angleInDegrees)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(fastAtan64f, cv_hal_fastAtan64f, Y, X, angle, len, angleInDegrees);
+
+    CV_CPU_DISPATCH(fastAtan64f, (Y, X, angle, len, angleInDegrees),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+// deprecated
+void fastAtan2(const float *Y, const float *X, float *angle, int len, bool angleInDegrees )
+{
+    CV_INSTRUMENT_REGION()
+
+    fastAtan32f(Y, X, angle, len, angleInDegrees);
+}
+
+void magnitude32f(const float* x, const float* y, float* mag, int len)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(magnitude32f, cv_hal_magnitude32f, x, y, mag, len);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsMagnitude_32f, x, y, mag, len) >= 0);
+
+    CV_CPU_DISPATCH(magnitude32f, (x, y, mag, len),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+void magnitude64f(const double* x, const double* y, double* mag, int len)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(magnitude64f, cv_hal_magnitude64f, x, y, mag, len);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsMagnitude_64f, x, y, mag, len) >= 0);
+
+    CV_CPU_DISPATCH(magnitude64f, (x, y, mag, len),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+
+void invSqrt32f(const float* src, float* dst, int len)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(invSqrt32f, cv_hal_invSqrt32f, src, dst, len);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsInvSqrt_32f_A21, src, dst, len) >= 0);
+
+    CV_CPU_DISPATCH(invSqrt32f, (src, dst, len),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+
+void invSqrt64f(const double* src, double* dst, int len)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(invSqrt64f, cv_hal_invSqrt64f, src, dst, len);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsInvSqrt_64f_A50, src, dst, len) >= 0);
+
+    CV_CPU_DISPATCH(invSqrt64f, (src, dst, len),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+
+void sqrt32f(const float* src, float* dst, int len)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(sqrt32f, cv_hal_sqrt32f, src, dst, len);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsSqrt_32f_A21, src, dst, len) >= 0);
+
+    CV_CPU_DISPATCH(sqrt32f, (src, dst, len),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+
+void sqrt64f(const double* src, double* dst, int len)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(sqrt64f, cv_hal_sqrt64f, src, dst, len);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsSqrt_64f_A50, src, dst, len) >= 0);
+
+    CV_CPU_DISPATCH(sqrt64f, (src, dst, len),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+void exp32f(const float *src, float *dst, int n)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(exp32f, cv_hal_exp32f, src, dst, n);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsExp_32f_A21, src, dst, n) >= 0);
+
+    CV_CPU_DISPATCH(exp32f, (src, dst, n),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+void exp64f(const double *src, double *dst, int n)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(exp64f, cv_hal_exp64f, src, dst, n);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsExp_64f_A50, src, dst, n) >= 0);
+
+    CV_CPU_DISPATCH(exp64f, (src, dst, n),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+void log32f(const float *src, float *dst, int n)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(log32f, cv_hal_log32f, src, dst, n);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsLn_32f_A21, src, dst, n) >= 0);
+
+    CV_CPU_DISPATCH(log32f, (src, dst, n),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+void log64f(const double *src, double *dst, int n)
+{
+    CV_INSTRUMENT_REGION()
+
+    CALL_HAL(log64f, cv_hal_log64f, src, dst, n);
+    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsLn_64f_A50, src, dst, n) >= 0);
+
+    CV_CPU_DISPATCH(log64f, (src, dst, n),
+        CV_CPU_DISPATCH_MODES_ALL);
+}
+
+//=============================================================================
+// for compatibility with 3.0
+
+void exp(const float* src, float* dst, int n)
+{
+    exp32f(src, dst, n);
+}
+
+void exp(const double* src, double* dst, int n)
+{
+    exp64f(src, dst, n);
+}
+
+void log(const float* src, float* dst, int n)
+{
+    log32f(src, dst, n);
+}
+
+void log(const double* src, double* dst, int n)
+{
+    log64f(src, dst, n);
+}
+
+void magnitude(const float* x, const float* y, float* dst, int n)
+{
+    magnitude32f(x, y, dst, n);
+}
+
+void magnitude(const double* x, const double* y, double* dst, int n)
+{
+    magnitude64f(x, y, dst, n);
+}
+
+void sqrt(const float* src, float* dst, int len)
+{
+    sqrt32f(src, dst, len);
+}
+
+void sqrt(const double* src, double* dst, int len)
+{
+    sqrt64f(src, dst, len);
+}
+
+void invSqrt(const float* src, float* dst, int len)
+{
+    invSqrt32f(src, dst, len);
+}
+
+void invSqrt(const double* src, double* dst, int len)
+{
+    invSqrt64f(src, dst, len);
+}
+
+}} // namespace cv::hal::
+
+float cv::fastAtan2( float y, float x )
+{
+    using namespace cv::hal;
+    CV_CPU_CALL_BASELINE(fastAtan2, (y, x));
+}

modules/core/src/mathfuncs_core.cpp → modules/core/src/mathfuncs_core.simd.hpp

-/*M///////////////////////////////////////////////////////////////////////////////////////
-//
-//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
-//
-//  By downloading, copying, installing or using the software you agree to this license.
-//  If you do not agree to this license, do not download, install,
-//  copy or use the software.
-//
-//
-//                           License Agreement
-//                For Open Source Computer Vision Library
-//
-// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
-// Copyright (C) 2009-2011, Willow Garage Inc., all rights reserved.
-// Third party copyrights are property of their respective owners.
-//
-// Redistribution and use in source and binary forms, with or without modification,
-// are permitted provided that the following conditions are met:
-//
-//   * Redistribution's of source code must retain the above copyright notice,
-//     this list of conditions and the following disclaimer.
-//
-//   * Redistribution's in binary form must reproduce the above copyright notice,
-//     this list of conditions and the following disclaimer in the documentation
-//     and/or other materials provided with the distribution.
-//
-//   * The name of the copyright holders may not be used to endorse or promote products
-//     derived from this software without specific prior written permission.
-//
-// This software is provided by the copyright holders and contributors "as is" and
-// any express or implied warranties, including, but not limited to, the implied
-// warranties of merchantability and fitness for a particular purpose are disclaimed.
-// In no event shall the Intel Corporation or contributors be liable for any direct,
-// indirect, incidental, special, exemplary, or consequential damages
-// (including, but not limited to, procurement of substitute goods or services;
-// loss of use, data, or profits; or business interruption) however caused
-// and on any theory of liability, whether in contract, strict liability,
-// or tort (including negligence or otherwise) arising in any way out of
-// the use of this software, even if advised of the possibility of such damage.
-//
-//M*/
-
-#include "precomp.hpp"
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+namespace cv { namespace hal {
+
+CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
+
+// forward declarations
+void fastAtan32f(const float *Y, const float *X, float *angle, int len, bool angleInDegrees);
+void fastAtan64f(const double *Y, const double *X, double *angle, int len, bool angleInDegrees);
+void fastAtan2(const float *Y, const float *X, float *angle, int len, bool angleInDegrees);
+void magnitude32f(const float* x, const float* y, float* mag, int len);
+void magnitude64f(const double* x, const double* y, double* mag, int len);
+void invSqrt32f(const float* src, float* dst, int len);
+void invSqrt64f(const double* src, double* dst, int len);
+void sqrt32f(const float* src, float* dst, int len);
+void sqrt64f(const double* src, double* dst, int len);
+void exp32f(const float *src, float *dst, int n);
+void exp64f(const double *src, double *dst, int n);
+void log32f(const float *src, float *dst, int n);
+void log64f(const double *src, double *dst, int n);
+float fastAtan2(float y, float x);
+
+
+#ifndef CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
 
 using namespace std;
 
@@ -197,23 +180,17 @@ static inline void atanImpl(const T *Y, const T *X, T *angle, int len, bool angl
 
 } // anonymous::
 
-namespace cv { namespace hal {
-
 ///////////////////////////////////// ATAN2 ////////////////////////////////////
 
 void fastAtan32f(const float *Y, const float *X, float *angle, int len, bool angleInDegrees )
 {
     CV_INSTRUMENT_REGION()
-
-    CALL_HAL(fastAtan32f, cv_hal_fastAtan32f, Y, X, angle, len, angleInDegrees);
     atanImpl<float>(Y, X, angle, len, angleInDegrees);
 }
 
 void fastAtan64f(const double *Y, const double *X, double *angle, int len, bool angleInDegrees)
 {
     CV_INSTRUMENT_REGION()
-
-    CALL_HAL(fastAtan64f, cv_hal_fastAtan64f, Y, X, angle, len, angleInDegrees);
     atanImpl<double>(Y, X, angle, len, angleInDegrees);
 }
 
@@ -221,7 +198,6 @@ void fastAtan64f(const double *Y, const double *X, double *angle, int len, bool
 void fastAtan2(const float *Y, const float *X, float *angle, int len, bool angleInDegrees )
 {
     CV_INSTRUMENT_REGION()
-
     fastAtan32f(Y, X, angle, len, angleInDegrees);
 }
 
@@ -229,9 +205,6 @@ void magnitude32f(const float* x, const float* y, float* mag, int len)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(magnitude32f, cv_hal_magnitude32f, x, y, mag, len);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsMagnitude_32f, x, y, mag, len) >= 0);
-
     int i = 0;
 
 #if CV_SIMD128
@@ -257,9 +230,6 @@ void magnitude64f(const double* x, const double* y, double* mag, int len)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(magnitude64f, cv_hal_magnitude64f, x, y, mag, len);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsMagnitude_64f, x, y, mag, len) >= 0);
-
     int i = 0;
 
 #if CV_SIMD128_64F
@@ -286,9 +256,6 @@ void invSqrt32f(const float* src, float* dst, int len)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(invSqrt32f, cv_hal_invSqrt32f, src, dst, len);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsInvSqrt_32f_A21, src, dst, len) >= 0);
-
     int i = 0;
 
 #if CV_SIMD128
@@ -310,9 +277,6 @@ void invSqrt64f(const double* src, double* dst, int len)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(invSqrt64f, cv_hal_invSqrt64f, src, dst, len);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsInvSqrt_64f_A50, src, dst, len) >= 0);
-
     int i = 0;
 
 #if CV_SSE2
@@ -330,9 +294,6 @@ void sqrt32f(const float* src, float* dst, int len)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(sqrt32f, cv_hal_sqrt32f, src, dst, len);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsSqrt_32f_A21, src, dst, len) >= 0);
-
     int i = 0;
 
 #if CV_SIMD128
@@ -354,9 +315,6 @@ void sqrt64f(const double* src, double* dst, int len)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(sqrt64f, cv_hal_sqrt64f, src, dst, len);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsSqrt_64f_A50, src, dst, len) >= 0);
-
     int i = 0;
 
 #if CV_SIMD128_64F
@@ -381,9 +339,6 @@ void exp32f(const float *src, float *dst, int n)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(exp32f, cv_hal_exp32f, src, dst, n);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsExp_32f_A21, src, dst, n) >= 0);
-
     for (int i = 0; i < n; i++)
     {
         dst[i] = std::exp(src[i]);
@@ -394,9 +349,6 @@ void exp64f(const double *src, double *dst, int n)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(exp64f, cv_hal_exp64f, src, dst, n);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsExp_64f_A50, src, dst, n) >= 0);
-
     for (int i = 0; i < n; i++)
     {
         dst[i] = std::exp(src[i]);
@@ -407,9 +359,6 @@ void log32f(const float *src, float *dst, int n)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(log32f, cv_hal_log32f, src, dst, n);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsLn_32f_A21, src, dst, n) >= 0);
-
     for (int i = 0; i < n; i++)
     {
         dst[i] = std::log(src[i]);
@@ -419,9 +368,6 @@ void log64f(const double *src, double *dst, int n)
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(log64f, cv_hal_log64f, src, dst, n);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsLn_64f_A50, src, dst, n) >= 0);
-
     for (int i = 0; i < n; i++)
     {
         dst[i] = std::log(src[i]);
@@ -534,9 +480,6 @@ void exp32f( const float *_x, float *y, int n )
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(exp32f, cv_hal_exp32f, _x, y, n);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsExp_32f_A21, _x, y, n) >= 0);
-
     static const float
     A4 = (float)(1.000000000000002438532970795181890933776 / EXPPOLY_32F_A0),
     A3 = (float)(.6931471805521448196800669615864773144641 / EXPPOLY_32F_A0),
@@ -551,7 +494,90 @@ void exp32f( const float *_x, float *y, int n )
     const Cv32suf* x = (const Cv32suf*)_x;
     Cv32suf buf[4];
 
-#if CV_SSE2
+#if CV_AVX2
+    if( n >= 8 )
+    {
+        static const __m256d prescale4 = _mm256_set1_pd(exp_prescale);
+        static const __m256 postscale8 = _mm256_set1_ps((float)exp_postscale);
+        static const __m128 maxval4 = _mm_set1_ps((float)(exp_max_val/exp_prescale));
+        static const __m128 minval4 = _mm_set1_ps((float)(-exp_max_val/exp_prescale));
+
+        static const __m256 mA1 = _mm256_set1_ps(A1);
+        static const __m256 mA2 = _mm256_set1_ps(A2);
+        static const __m256 mA3 = _mm256_set1_ps(A3);
+        static const __m256 mA4 = _mm256_set1_ps(A4);
+        bool y_aligned = (size_t)(void*)y % 32 == 0;
+
+        ushort CV_DECL_ALIGNED(32) tab_idx[16];
+
+        for( ; i <= n - 8; i += 8 )
+        {
+            __m128i xi0, xi1;
+
+            __m256d xd0 = _mm256_cvtps_pd(_mm_min_ps(_mm_max_ps(_mm_loadu_ps(&x[i].f), minval4), maxval4));
+            __m256d xd1 = _mm256_cvtps_pd(_mm_min_ps(_mm_max_ps(_mm_loadu_ps(&x[i+4].f), minval4), maxval4));
+
+            xd0 = _mm256_mul_pd(xd0, prescale4);
+            xd1 = _mm256_mul_pd(xd1, prescale4);
+
+            xi0 = _mm256_cvtpd_epi32(xd0);
+            xi1 = _mm256_cvtpd_epi32(xd1);
+
+            xd0 = _mm256_sub_pd(xd0, _mm256_cvtepi32_pd(xi0));
+            xd1 = _mm256_sub_pd(xd1, _mm256_cvtepi32_pd(xi1));
+
+            // gcc does not support _mm256_set_m128
+            //xf = _mm256_set_m128(_mm256_cvtpd_ps(xd1), _mm256_cvtpd_ps(xd0));
+            __m256 xf = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm256_cvtpd_ps(xd0)), _mm256_cvtpd_ps(xd1), 1);
+
+            xf = _mm256_mul_ps(xf, postscale8);
+
+            xi0 = _mm_packs_epi32(xi0, xi1);
+
+            _mm_store_si128((__m128i*)tab_idx, _mm_and_si128(xi0, _mm_set1_epi16(EXPTAB_MASK)));
+
+            xi0 = _mm_add_epi16(_mm_srai_epi16(xi0, EXPTAB_SCALE), _mm_set1_epi16(127));
+            xi0 = _mm_max_epi16(xi0, _mm_setzero_si128());
+            xi0 = _mm_min_epi16(xi0, _mm_set1_epi16(255));
+            xi1 = _mm_unpackhi_epi16(xi0, _mm_setzero_si128());
+            xi0 = _mm_unpacklo_epi16(xi0, _mm_setzero_si128());
+
+            __m256d yd0 = _mm256_set_pd(expTab[tab_idx[3]], expTab[tab_idx[2]], expTab[tab_idx[1]], expTab[tab_idx[0]]);
+            __m256d yd1 = _mm256_set_pd(expTab[tab_idx[7]], expTab[tab_idx[6]], expTab[tab_idx[5]], expTab[tab_idx[4]]);
+
+            // gcc does not support _mm256_set_m128
+            //__m256 yf = _mm256_set_m128(_mm256_cvtpd_ps(yd1), _mm256_cvtpd_ps(yd0));
+            __m256 yf = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm256_cvtpd_ps(yd0)), _mm256_cvtpd_ps(yd1), 1);
+
+            //_mm256_set_m128i(xi1, xi0)
+            __m256i temp = (__m256i)_mm256_insertf128_ps(_mm256_castps128_ps256((__m128)xi0), (__m128)xi1, 1);
+
+            yf = _mm256_mul_ps(yf, _mm256_castsi256_ps(_mm256_slli_epi32(temp, 23)));
+
+            __m256 zf = _mm256_add_ps(xf, mA1);
+
+#if CV_FMA3
+            zf = _mm256_fmadd_ps(zf, xf, mA2);
+            zf = _mm256_fmadd_ps(zf, xf, mA3);
+            zf = _mm256_fmadd_ps(zf, xf, mA4);
+#else
+            zf = _mm256_add_ps(_mm256_mul_ps(zf, xf), mA2);
+            zf = _mm256_add_ps(_mm256_mul_ps(zf, xf), mA3);
+            zf = _mm256_add_ps(_mm256_mul_ps(zf, xf), mA4);
+#endif
+            zf = _mm256_mul_ps(zf, yf);
+
+            if( y_aligned )
+            {
+                _mm256_store_ps(y + i, zf);
+            }
+            else
+            {
+                _mm256_storeu_ps(y + i, zf);
+            }
+        }
+    }
+#elif CV_SSE2
     if( n >= 8 )
     {
         static const __m128d prescale2 = _mm_set1_pd(exp_prescale);
@@ -738,9 +764,6 @@ void exp64f( const double *_x, double *y, int n )
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(exp64f, cv_hal_exp64f, _x, y, n);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsExp_64f_A50, _x, y, n) >= 0);
-
     static const double
     A5 = .99999999999999999998285227504999 / EXPPOLY_32F_A0,
     A4 = .69314718055994546743029643825322 / EXPPOLY_32F_A0,
@@ -1187,9 +1210,6 @@ void log32f( const float *_x, float *y, int n )
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(log32f, cv_hal_log32f, _x, y, n);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsLn_32f_A21, _x, y, n) >= 0);
-
     static const float shift[] = { 0, -1.f/512 };
     static const float
     A0 = 0.3333333333333333333333333f,
@@ -1336,9 +1356,6 @@ void log64f( const double *x, double *y, int n )
 {
     CV_INSTRUMENT_REGION()
 
-    CALL_HAL(log64f, cv_hal_log64f, x, y, n);
-    CV_IPP_RUN_FAST(CV_INSTRUMENT_FUN_IPP(ippsLn_64f_A50, x, y, n) >= 0);
-
     static const double shift[] = { 0, -1./512 };
     static const double
     A7 = 1.0,
@@ -1524,64 +1541,13 @@ void log64f( const double *x, double *y, int n )
 
 #endif // issue 7795
 
-//=============================================================================
-// for compatibility with 3.0
-
-void exp(const float* src, float* dst, int n)
-{
-    exp32f(src, dst, n);
-}
-
-void exp(const double* src, double* dst, int n)
-{
-    exp64f(src, dst, n);
-}
-
-void log(const float* src, float* dst, int n)
+float fastAtan2( float y, float x )
 {
-    log32f(src, dst, n);
-}
-
-void log(const double* src, double* dst, int n)
-{
-    log64f(src, dst, n);
-}
-
-void magnitude(const float* x, const float* y, float* dst, int n)
-{
-    magnitude32f(x, y, dst, n);
-}
-
-void magnitude(const double* x, const double* y, double* dst, int n)
-{
-    magnitude64f(x, y, dst, n);
-}
-
-void sqrt(const float* src, float* dst, int len)
-{
-    sqrt32f(src, dst, len);
-}
-
-void sqrt(const double* src, double* dst, int len)
-{
-    sqrt64f(src, dst, len);
-}
-
-void invSqrt(const float* src, float* dst, int len)
-{
-    invSqrt32f(src, dst, len);
-}
-
-void invSqrt(const double* src, double* dst, int len)
-{
-    invSqrt64f(src, dst, len);
+    return atanImpl<float>(y, x);
 }
 
+#endif // CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
 
-} // cv::hal::
-} // cv::
+CV_CPU_OPTIMIZATION_NAMESPACE_END
 
-float cv::fastAtan2( float y, float x )
-{
-    return atanImpl<float>(y, x);
-}
+}} // namespace cv::hal

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}")