Fix AVX and other SIMD support detection

CMakeLists.txt

@@ -159,7 +159,6 @@ OCV_OPTION(BUILD_FAT_JAVA_LIB       "Create fat java wrapper containing the whol
 OCV_OPTION(BUILD_ANDROID_SERVICE    "Build OpenCV Manager for Google Play" OFF IF ANDROID AND ANDROID_USE_STLPORT AND ANDROID_SOURCE_TREE )
 OCV_OPTION(BUILD_ANDROID_PACKAGE    "Build platform-specific package for Google Play" OFF IF ANDROID )
 
-
 # 3rd party libs
 OCV_OPTION(BUILD_ZLIB               "Build zlib from source"         WIN32 OR APPLE )
 OCV_OPTION(BUILD_TIFF               "Build libtiff from source"      WIN32 OR ANDROID OR APPLE )
@@ -168,6 +167,7 @@ OCV_OPTION(BUILD_JPEG               "Build libjpeg from source"      WIN32 OR AN
 OCV_OPTION(BUILD_PNG                "Build libpng from source"       WIN32 OR ANDROID OR APPLE )
 OCV_OPTION(BUILD_OPENEXR            "Build openexr from source"      WIN32 OR ANDROID OR APPLE )
 
+
 # OpenCV installation options
 # ===================================================
 OCV_OPTION(INSTALL_C_EXAMPLES       "Install C examples"        OFF )
@@ -176,16 +176,10 @@ OCV_OPTION(INSTALL_ANDROID_EXAMPLES "Install Android examples"  OFF  IF ANDROID
 OCV_OPTION(INSTALL_TO_MANGLED_PATHS "Enables mangled install paths, that help with side by side installs." OFF IF (UNIX AND NOT ANDROID AND NOT IOS AND BUILD_SHARED_LIBS) )
 
 
-if(MSVC_IDE OR CMAKE_GENERATOR MATCHES Xcode)
-  set(ENABLE_SOLUTION_FOLDERS0 ON)
-else()
-  set(ENABLE_SOLUTION_FOLDERS0 OFF)
-endif()
-
 # OpenCV build options
 # ===================================================
 OCV_OPTION(ENABLE_PRECOMPILED_HEADERS "Use precompiled headers"                                  ON   IF (NOT IOS) )
-OCV_OPTION(ENABLE_SOLUTION_FOLDERS    "Solution folder in Visual Studio or in other IDEs"        ${ENABLE_SOLUTION_FOLDERS0} IF (CMAKE_VERSION VERSION_GREATER "2.8.0") )
+OCV_OPTION(ENABLE_SOLUTION_FOLDERS    "Solution folder in Visual Studio or in other IDEs"        (MSVC_IDE OR CMAKE_GENERATOR MATCHES Xcode) IF (CMAKE_VERSION VERSION_GREATER "2.8.0") )
 OCV_OPTION(ENABLE_PROFILING           "Enable profiling in the GCC compiler (Add flags: -g -pg)" OFF  IF CMAKE_COMPILER_IS_GNUCXX )
 OCV_OPTION(ENABLE_OMIT_FRAME_POINTER  "Enable -fomit-frame-pointer for GCC"                      ON   IF CMAKE_COMPILER_IS_GNUCXX )
 OCV_OPTION(ENABLE_POWERPC             "Enable PowerPC for GCC"                                   ON   IF (CMAKE_COMPILER_IS_GNUCXX AND CMAKE_SYSTEM_PROCESSOR MATCHES powerpc.*) )
@@ -196,6 +190,7 @@ OCV_OPTION(ENABLE_SSE3                "Enable SSE3 instructions"
 OCV_OPTION(ENABLE_SSSE3               "Enable SSSE3 instructions"                                OFF  IF (CMAKE_COMPILER_IS_GNUCXX AND (X86 OR X86_64)) )
 OCV_OPTION(ENABLE_SSE41               "Enable SSE4.1 instructions"                               OFF  IF ((CV_ICC OR CMAKE_COMPILER_IS_GNUCXX) AND (X86 OR X86_64)) )
 OCV_OPTION(ENABLE_SSE42               "Enable SSE4.2 instructions"                               OFF  IF (CMAKE_COMPILER_IS_GNUCXX AND (X86 OR X86_64)) )
+OCV_OPTION(ENABLE_AVX                 "Enable AVX instructions"                                  OFF  IF ((MSVC OR CMAKE_COMPILER_IS_GNUCXX) AND (X86 OR X86_64)) )
 OCV_OPTION(ENABLE_NOISY_WARNINGS      "Show all warnings even if they are too noisy"             OFF )
 OCV_OPTION(OPENCV_WARNINGS_ARE_ERRORS "Treat warnings as errors"                                 OFF )
 
@@ -762,7 +757,7 @@ status("    Use Clp:"   HAVE_CLP   THEN YES ELSE NO)
 
 if(HAVE_CUDA)
   status("")
-  status("  NVIDIA CUDA:"            "(ver ${CUDA_VERSION_STRING})")
+  status("  NVIDIA CUDA")
 
   status("    Use CUFFT:"            HAVE_CUFFT  THEN YES ELSE NO)
   status("    Use CUBLAS:"           HAVE_CUBLAS THEN YES ELSE NO)

cmake/OpenCVCompilerOptions.cmake

@@ -124,35 +124,33 @@ if(CMAKE_COMPILER_IS_GNUCXX)
 
   # SSE3 and further should be disabled under MingW because it generates compiler errors
   if(NOT MINGW)
+    if(ENABLE_AVX)
+      add_extra_compiler_option(-mavx)
+    endif()
+
+    # GCC depresses SSEx instructions when -mavx is used. Instead, it generates new AVX instructions or AVX equivalence for all SSEx instructions when needed.
+    if(NOT OPENCV_EXTRA_CXX_FLAGS MATCHES "-mavx")
       if(ENABLE_SSE3)
         add_extra_compiler_option(-msse3)
       endif()
 
-    if(${CMAKE_OPENCV_GCC_VERSION_NUM} GREATER 402)
-      set(HAVE_GCC43_OR_NEWER 1)
-    endif()
-    if(${CMAKE_OPENCV_GCC_VERSION_NUM} GREATER 401)
-      set(HAVE_GCC42_OR_NEWER 1)
-    endif()
-
-    if(HAVE_GCC42_OR_NEWER OR APPLE)
       if(ENABLE_SSSE3)
         add_extra_compiler_option(-mssse3)
       endif()
-      if(HAVE_GCC43_OR_NEWER OR APPLE)
+
       if(ENABLE_SSE41)
         add_extra_compiler_option(-msse4.1)
       endif()
+
       if(ENABLE_SSE42)
         add_extra_compiler_option(-msse4.2)
       endif()
     endif()
-    endif()
   endif(NOT MINGW)
 
   if(X86 OR X86_64)
     if(NOT APPLE AND CMAKE_SIZEOF_VOID_P EQUAL 4)
-      if(ENABLE_SSE2)
+      if(OPENCV_EXTRA_CXX_FLAGS MATCHES "-m(sse2|avx)")
         add_extra_compiler_option(-mfpmath=sse)# !! important - be on the same wave with x64 compilers
       else()
         add_extra_compiler_option(-mfpmath=387)
@@ -201,30 +199,35 @@ if(MSVC)
     set(OPENCV_EXTRA_FLAGS_RELEASE "${OPENCV_EXTRA_FLAGS_RELEASE} /Zi")
   endif()
 
-  if(NOT MSVC64)
+  if(ENABLE_AVX AND NOT MSVC_VERSION LESS 1600)
-    # 64-bit MSVC compiler uses SSE/SSE2 by default
+    set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:AVX")
-    if(ENABLE_SSE)
-      set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:SSE")
-    endif()
-    if(ENABLE_SSE2)
-      set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:SSE2")
-    endif()
   endif()
 
-  if(ENABLE_SSE3)
+  if(ENABLE_SSE4_1 AND CV_ICC AND NOT OPENCV_EXTRA_FLAGS MATCHES "/arch:")
-    set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:SSE3")
-  endif()
-  if(ENABLE_SSE4_1)
     set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:SSE4.1")
   endif()
 
-  if(ENABLE_SSE OR ENABLE_SSE2 OR ENABLE_SSE3 OR ENABLE_SSE4_1)
+  if(ENABLE_SSE3 AND CV_ICC AND NOT OPENCV_EXTRA_FLAGS MATCHES "/arch:")
+    set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:SSE3")
+  endif()
+
+  if(NOT MSVC64)
+    # 64-bit MSVC compiler uses SSE/SSE2 by default
+    if(ENABLE_SSE2 AND NOT OPENCV_EXTRA_FLAGS MATCHES "/arch:")
+      set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:SSE2")
+    endif()
+    if(ENABLE_SSE AND NOT OPENCV_EXTRA_FLAGS MATCHES "/arch:")
+      set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /arch:SSE")
+    endif()
+  endif()
+
+  if(ENABLE_SSE OR ENABLE_SSE2 OR ENABLE_SSE3 OR ENABLE_SSE4_1 OR ENABLE_AVX)
     set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /Oi")
   endif()
 
   if(X86 OR X86_64)
     if(CMAKE_SIZEOF_VOID_P EQUAL 4 AND ENABLE_SSE2)
-      set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /fp:fast")# !! important - be on the same wave with x64 compilers
+      set(OPENCV_EXTRA_FLAGS "${OPENCV_EXTRA_FLAGS} /fp:fast") # !! important - be on the same wave with x64 compilers
     endif()
   endif()
 endif()

cmake/OpenCVDetectCXXCompiler.cmake

@@ -89,8 +89,8 @@ if(CMAKE_COMPILER_IS_GNUCXX)
     endif()
 endif()
 
-if(CMAKE_SYSTEM_PROCESSOR MATCHES amd64.*|x86_64.* OR CMAKE_GENERATOR MATCHES "Visual Studio.*Win64")
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*" OR CMAKE_GENERATOR MATCHES "Visual Studio.*Win64")
     set(X86_64 1)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES i686.*|i386.*|x86.*)
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "i686.*|i386.*|x86.*|amd64.*|AMD64.*")
     set(X86 1)
 endif()

modules/core/include/opencv2/core/internal.hpp

@@ -104,35 +104,33 @@ CV_INLINE IppiSize ippiSize(int width, int height)
 }
 #endif
 
-#if defined __SSE2__ || (defined _MSC_VER && _MSC_VER >= 1300)
+#ifndef IPPI_CALL
+#  define IPPI_CALL(func) CV_Assert((func) >= 0)
+#endif
+
+#if defined __SSE2__ || defined _M_X64  || (defined _M_IX86_FP && _M_IX86_FP >= 2)
 #  include "emmintrin.h"
 #  define CV_SSE 1
 #  define CV_SSE2 1
 #  if defined __SSE3__ || (defined _MSC_VER && _MSC_VER >= 1500)
 #    include "pmmintrin.h"
 #    define CV_SSE3 1
-#  else
-#    define CV_SSE3 0
 #  endif
-#  if defined __SSSE3__
+#  if defined __SSSE3__  || (defined _MSC_VER && _MSC_VER >= 1500)
 #    include "tmmintrin.h"
 #    define CV_SSSE3 1
-#  else
-#    define CV_SSSE3 0
 #  endif
 #  if defined __SSE4_1__ || (defined _MSC_VER && _MSC_VER >= 1500)
 #    include <smmintrin.h>
 #    define CV_SSE4_1 1
-#  else
-#    define CV_SSE4_1 0
 #  endif
 #  if defined __SSE4_2__ || (defined _MSC_VER && _MSC_VER >= 1500)
 #    include <nmmintrin.h>
 #    define CV_SSE4_2 1
-#  else
-#    define CV_SSE4_2 0
 #  endif
 #  if defined __AVX__ || (defined _MSC_FULL_VER && _MSC_FULL_VER >= 160040219)
+// MS Visual Studio 2010 (2012?) has no macro pre-defined to identify the use of /arch:AVX
+// See: http://connect.microsoft.com/VisualStudio/feedback/details/605858/arch-avx-should-define-a-predefined-macro-in-x64-and-set-a-unique-value-for-m-ix86-fp-in-win32
 #    include <immintrin.h>
 #    define CV_AVX 1
 #    if defined(_XCR_XFEATURE_ENABLED_MASK)
@@ -140,33 +138,38 @@ CV_INLINE IppiSize ippiSize(int width, int height)
 #    else
 #      define __xgetbv() 0
 #    endif
-#  else
-#    define CV_AVX 0
 #  endif
-#  else
-#  define CV_SSE 0
-#  define CV_SSE2 0
-#  define CV_SSE3 0
-#  define CV_SSSE3 0
-#  define CV_SSE4_1 0
-#  define CV_SSE4_2 0
-#  define CV_AVX 0
-#  endif
-
-#if defined ANDROID && defined __ARM_NEON__
-#  include "arm_neon.h"
-#  define CV_NEON 1
-
-#  define CPU_HAS_NEON_FEATURE (true)
-//TODO: make real check using stuff from "cpu-features.h"
-//((bool)android_getCpuFeatures() & ANDROID_CPU_ARM_FEATURE_NEON)
-#else
-#  define CV_NEON 0
-#  define CPU_HAS_NEON_FEATURE (false)
 #endif
 
-#ifndef IPPI_CALL
+#ifdef __ARM_NEON__
-#  define IPPI_CALL(func) CV_Assert((func) >= 0)
+#  include <arm_neon.h>
+#  define CV_NEON 1
+#  define CPU_HAS_NEON_FEATURE (true)
+#endif
+
+#ifndef CV_SSE
+#  define CV_SSE 0
+#endif
+#ifndef CV_SSE2
+#  define CV_SSE2 0
+#endif
+#ifndef CV_SSE3
+#  define CV_SSE3 0
+#endif
+#ifndef CV_SSSE3
+#  define CV_SSSE3 0
+#endif
+#ifndef CV_SSE4_1
+#  define CV_SSE4_1 0
+#endif
+#ifndef CV_SSE4_2
+#  define CV_SSE4_2 0
+#endif
+#ifndef CV_AVX
+#  define CV_AVX 0
+#endif
+#ifndef CV_NEON
+#  define CV_NEON 0
 #endif
 
 #ifdef HAVE_TBB