Merge pull request #907 from SpecLad:master

3rdparty/libjasper/CMakeLists.txt

@@ -23,8 +23,8 @@ if(WIN32 AND NOT MINGW)
   add_definitions(-DJAS_WIN_MSVC_BUILD)
 endif(WIN32 AND NOT MINGW)
 
-ocv_warnings_disable(CMAKE_C_FLAGS -Wno-implicit-function-declaration -Wno-uninitialized
-                                   -Wmissing-prototypes -Wmissing-declarations -Wunused -Wshadow
+ocv_warnings_disable(CMAKE_C_FLAGS -Wno-implicit-function-declaration -Wno-uninitialized -Wmissing-prototypes
+                                   -Wno-unused-but-set-parameter -Wmissing-declarations -Wunused -Wshadow
                                    -Wsign-compare -Wstrict-overflow)
 ocv_warnings_disable(CMAKE_C_FLAGS -Wunused-parameter) # clang
 ocv_warnings_disable(CMAKE_C_FLAGS /wd4013 /wd4018 /wd4101 /wd4244 /wd4267 /wd4715) # vs2005
@@ -49,4 +49,3 @@ endif()
 if(NOT BUILD_SHARED_LIBS)
   install(TARGETS ${JASPER_LIBRARY} ARCHIVE DESTINATION ${OPENCV_3P_LIB_INSTALL_PATH} COMPONENT main)
 endif()
-

CMakeLists.txt

@@ -595,12 +595,15 @@ endif()
 status("")
 status("  GUI: ")
 
-if(HAVE_QT)
+if(HAVE_QT5)
+  status("    QT 5.x:"            HAVE_QT        THEN "YES (ver ${Qt5Core_VERSION_STRING})" ELSE NO)
+  status("    QT OpenGL support:" HAVE_QT_OPENGL THEN "YES (${Qt5OpenGL_LIBRARIES} ${Qt5OpenGL_VERSION_STRING})" ELSE NO)
+elseif(HAVE_QT)
   status("    QT 4.x:"            HAVE_QT        THEN "YES (ver ${QT_VERSION_MAJOR}.${QT_VERSION_MINOR}.${QT_VERSION_PATCH} ${QT_EDITION})" ELSE NO)
   status("    QT OpenGL support:" HAVE_QT_OPENGL THEN "YES (${QT_QTOPENGL_LIBRARY})" ELSE NO)
 else()
   if(DEFINED WITH_QT)
-    status("    QT 4.x:" NO)
+    status("    QT:" NO)
   endif()
   if(DEFINED WITH_WIN32UI)
     status("    Win32 UI:" HAVE_WIN32UI THEN YES ELSE NO)

cmake/OpenCVDetectAndroidSDK.cmake

@@ -176,7 +176,8 @@ macro(android_get_compatible_target VAR)
 endmacro()
 
 unset(__android_project_chain CACHE)
-#add_android_project(target_name ${path} NATIVE_DEPS opencv_core LIBRARY_DEPS ${OpenCV_BINARY_DIR} SDK_TARGET 11)
+
+# add_android_project(target_name ${path} NATIVE_DEPS opencv_core LIBRARY_DEPS ${OpenCV_BINARY_DIR} SDK_TARGET 11)
 macro(add_android_project target path)
   # parse arguments
   set(android_proj_arglist NATIVE_DEPS LIBRARY_DEPS SDK_TARGET IGNORE_JAVA IGNORE_MANIFEST)
@@ -212,6 +213,16 @@ macro(add_android_project target path)
     ocv_check_dependencies(${android_proj_NATIVE_DEPS} opencv_java)
   endif()
 
+  if(EXISTS "${path}/jni/Android.mk" )
+    # find if native_app_glue is used
+    file(STRINGS "${path}/jni/Android.mk" NATIVE_APP_GLUE REGEX ".*(call import-module,android/native_app_glue)" )
+    if(NATIVE_APP_GLUE)
+      if(ANDROID_NATIVE_API_LEVEL LESS 9 OR NOT EXISTS "${ANDROID_NDK}/sources/android/native_app_glue")
+        set(OCV_DEPENDENCIES_FOUND FALSE)
+      endif()
+    endif()
+  endif()
+
   if(OCV_DEPENDENCIES_FOUND AND android_proj_sdk_target AND ANDROID_EXECUTABLE AND ANT_EXECUTABLE AND ANDROID_TOOLS_Pkg_Revision GREATER 13 AND EXISTS "${path}/${ANDROID_MANIFEST_FILE}")
 
     project(${target})
@@ -268,9 +279,6 @@ macro(add_android_project target path)
       file(STRINGS "${path}/jni/Android.mk" JNI_LIB_NAME REGEX "LOCAL_MODULE[ ]*:=[ ]*.*" )
       string(REGEX REPLACE "LOCAL_MODULE[ ]*:=[ ]*([a-zA-Z_][a-zA-Z_0-9]*)[ ]*" "\\1" JNI_LIB_NAME "${JNI_LIB_NAME}")
 
-      # find using of native app glue to determine native activity
-      file(STRINGS "${path}/jni/Android.mk" NATIVE_APP_GLUE REGEX ".*(call import-module,android/native_app_glue)" )
-
       if(JNI_LIB_NAME)
         ocv_include_modules_recurse(${android_proj_NATIVE_DEPS})
         ocv_include_directories("${path}/jni")
@@ -291,9 +299,9 @@ macro(add_android_project target path)
             )
 
         get_target_property(android_proj_jni_location "${JNI_LIB_NAME}" LOCATION)
-    if (NOT (CMAKE_BUILD_TYPE MATCHES "debug"))
-        add_custom_command(TARGET ${JNI_LIB_NAME} POST_BUILD COMMAND ${CMAKE_STRIP} --strip-unneeded "${android_proj_jni_location}")
-    endif()
+        if (NOT (CMAKE_BUILD_TYPE MATCHES "debug"))
+            add_custom_command(TARGET ${JNI_LIB_NAME} POST_BUILD COMMAND ${CMAKE_STRIP} --strip-unneeded "${android_proj_jni_location}")
+        endif()
       endif()
     endif()
 

cmake/OpenCVDetectCXXCompiler.cmake

@@ -101,7 +101,7 @@ endif()
 
 if(MSVC64 OR MINGW64)
   set(X86_64 1)
-elseif(MSVC AND NOT CMAKE_CROSSCOMPILING)
+elseif(MINGW OR (MSVC AND NOT CMAKE_CROSSCOMPILING))
   set(X86 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*")
   set(X86_64 1)

cmake/OpenCVDetectOpenCL.cmake

@@ -20,12 +20,6 @@ else(APPLE)
               DOC "OpenCL include directory"
               NO_DEFAULT_PATH)
 
-    find_path(OPENCL_INCLUDE_DIR
-              NAMES OpenCL/cl.h CL/cl.h
-              HINTS ${OPENCL_ROOT_DIR}
-              PATH_SUFFIXES include include/nvidia-current
-              DOC "OpenCL include directory")
-
     if (X86_64)
       set(OPENCL_POSSIBLE_LIB_SUFFIXES lib/Win64 lib/x86_64 lib/x64)
     elseif (X86)
@@ -39,12 +33,6 @@ else(APPLE)
               DOC "OpenCL library"
               NO_DEFAULT_PATH)
 
-    find_library(OPENCL_LIBRARY
-              NAMES OpenCL
-              HINTS ${OPENCL_ROOT_DIR}
-              PATH_SUFFIXES ${OPENCL_POSSIBLE_LIB_SUFFIXES}
-              DOC "OpenCL library")
-
     mark_as_advanced(OPENCL_INCLUDE_DIR OPENCL_LIBRARY)
     include(FindPackageHandleStandardArgs)
     FIND_PACKAGE_HANDLE_STANDARD_ARGS(OPENCL DEFAULT_MSG OPENCL_LIBRARY OPENCL_INCLUDE_DIR )

cmake/OpenCVDetectPython.cmake

@@ -102,18 +102,12 @@ if(PYTHON_EXECUTABLE)
   if(BUILD_DOCS)
     find_host_program(SPHINX_BUILD sphinx-build)
     if(SPHINX_BUILD)
-        if(UNIX)
-            execute_process(COMMAND sh -c "${SPHINX_BUILD} -_ 2>&1 | sed -ne 1p"
-                             RESULT_VARIABLE SPHINX_PROCESS
-                             OUTPUT_VARIABLE SPHINX_VERSION
-                             OUTPUT_STRIP_TRAILING_WHITESPACE)
-        else()
-            execute_process(COMMAND ${PYTHON_EXECUTABLE} -c "import sphinx; print sphinx.__version__"
-                            RESULT_VARIABLE SPHINX_PROCESS
-                            OUTPUT_VARIABLE SPHINX_VERSION
-                            OUTPUT_STRIP_TRAILING_WHITESPACE)
-        endif()
-        if(SPHINX_PROCESS EQUAL 0)
+        execute_process(COMMAND "${SPHINX_BUILD}"
+                        OUTPUT_QUIET
+                        ERROR_VARIABLE SPHINX_OUTPUT
+                        OUTPUT_STRIP_TRAILING_WHITESPACE)
+        if(SPHINX_OUTPUT MATCHES "^Sphinx v([0-9][^ \n]*)")
+          set(SPHINX_VERSION "${CMAKE_MATCH_1}")
           set(HAVE_SPHINX 1)
           message(STATUS "Found Sphinx ${SPHINX_VERSION}: ${SPHINX_BUILD}")
         endif()

cmake/OpenCVFindLibsGUI.cmake

@@ -13,12 +13,31 @@ if(WITH_WIN32UI)
 endif(WITH_WIN32UI)
 
 # --- QT4 ---
-ocv_clear_vars(HAVE_QT)
+ocv_clear_vars(HAVE_QT HAVE_QT5)
 if(WITH_QT)
-  find_package(Qt4)
-  if(QT4_FOUND)
-    set(HAVE_QT TRUE)
-    add_definitions(-DHAVE_QT) # We need to define the macro this way, using cvconfig.h does not work
+  if(NOT CMAKE_VERSION VERSION_LESS 2.8.3 AND NOT WITH_QT EQUAL 4)
+    find_package(Qt5Core)
+    find_package(Qt5Gui)
+    find_package(Qt5Widgets)
+    find_package(Qt5Test)
+    find_package(Qt5Concurrent)
+    if(Qt5Core_FOUND AND Qt5Gui_FOUND AND Qt5Widgets_FOUND AND Qt5Test_FOUND AND Qt5Concurrent_FOUND)
+      set(HAVE_QT5 ON)
+      set(HAVE_QT  ON)
+      add_definitions(-DHAVE_QT)
+      find_package(Qt5OpenGL)
+      if(Qt5OpenGL_FOUND)
+        set(QT_QTOPENGL_FOUND ON)
+      endif()
+    endif()
+  endif()
+
+  if(NOT HAVE_QT)
+    find_package(Qt4)
+    if(QT4_FOUND)
+      set(HAVE_QT TRUE)
+      add_definitions(-DHAVE_QT) # We need to define the macro this way, using cvconfig.h does not work
+    endif()
   endif()
 endif()
 

modules/androidcamera/camera_wrapper/camera_wrapper.cpp

@@ -362,6 +362,9 @@ CameraHandler* CameraHandler::initCameraConnect(const CameraCallback& callback,
     typedef sp<Camera> (*Android23ConnectFuncType)(int);
     typedef sp<Camera> (*Android3DConnectFuncType)(int, int);
 
+    const int BACK_CAMERA_INDEX = 99;
+    const int FRONT_CAMERA_INDEX = 98;
+
     enum {
     CAMERA_SUPPORT_MODE_2D = 0x01, /* Camera Sensor supports 2D mode. */
     CAMERA_SUPPORT_MODE_3D = 0x02, /* Camera Sensor supports 3D mode. */
@@ -373,7 +376,51 @@ CameraHandler* CameraHandler::initCameraConnect(const CameraCallback& callback,
     const char Android23ConnectName[] = "_ZN7android6Camera7connectEi";
     const char Android3DConnectName[] = "_ZN7android6Camera7connectEii";
 
-    LOGD("CameraHandler::initCameraConnect(%p, %d, %p, %p)", callback, cameraId, userData, prevCameraParameters);
+    int localCameraIndex = cameraId;
+
+#if !defined(ANDROID_r2_2_0)
+    if (cameraId == BACK_CAMERA_INDEX)
+    {
+        LOGD("Back camera selected");
+        for (int i = 0; i < Camera::getNumberOfCameras(); i++)
+        {
+            CameraInfo info;
+            Camera::getCameraInfo(i, &info);
+            if (info.facing == CAMERA_FACING_BACK)
+            {
+                localCameraIndex = i;
+                break;
+            }
+        }
+    }
+    else if (cameraId == FRONT_CAMERA_INDEX)
+    {
+        LOGD("Front camera selected");
+        for (int i = 0; i < Camera::getNumberOfCameras(); i++)
+        {
+            CameraInfo info;
+            Camera::getCameraInfo(i, &info);
+            if (info.facing == CAMERA_FACING_FRONT)
+            {
+                localCameraIndex = i;
+                break;
+            }
+        }
+    }
+
+    if (localCameraIndex == BACK_CAMERA_INDEX)
+    {
+        LOGE("Back camera not found!");
+        return NULL;
+    }
+    else if (localCameraIndex == FRONT_CAMERA_INDEX)
+    {
+        LOGE("Front camera not found!");
+        return NULL;
+    }
+#endif
+
+    LOGD("CameraHandler::initCameraConnect(%p, %d, %p, %p)", callback, localCameraIndex, userData, prevCameraParameters);
 
     sp<Camera> camera = 0;
 
@@ -381,8 +428,8 @@ CameraHandler* CameraHandler::initCameraConnect(const CameraCallback& callback,
 
     if (!CameraHALHandle)
     {
-    LOGE("Cannot link to \"libcamera_client.so\"");
-    return NULL;
+        LOGE("Cannot link to \"libcamera_client.so\"");
+        return NULL;
     }
 
     // reset errors
@@ -390,24 +437,24 @@ CameraHandler* CameraHandler::initCameraConnect(const CameraCallback& callback,
 
     if (Android22ConnectFuncType Android22Connect = (Android22ConnectFuncType)dlsym(CameraHALHandle, Android22ConnectName))
     {
-    LOGD("Connecting to CameraService v 2.2");
-    camera = Android22Connect();
+        LOGD("Connecting to CameraService v 2.2");
+        camera = Android22Connect();
     }
     else if (Android23ConnectFuncType Android23Connect = (Android23ConnectFuncType)dlsym(CameraHALHandle, Android23ConnectName))
     {
-    LOGD("Connecting to CameraService v 2.3");
-    camera = Android23Connect(cameraId);
+        LOGD("Connecting to CameraService v 2.3");
+        camera = Android23Connect(localCameraIndex);
     }
     else if (Android3DConnectFuncType Android3DConnect = (Android3DConnectFuncType)dlsym(CameraHALHandle, Android3DConnectName))
     {
-    LOGD("Connecting to CameraService v 3D");
-    camera = Android3DConnect(cameraId, CAMERA_SUPPORT_MODE_2D);
+        LOGD("Connecting to CameraService v 3D");
+        camera = Android3DConnect(localCameraIndex, CAMERA_SUPPORT_MODE_2D);
     }
     else
     {
-    dlclose(CameraHALHandle);
-    LOGE("Cannot connect to CameraService. Connect method was not found!");
-    return NULL;
+        dlclose(CameraHALHandle);
+        LOGE("Cannot connect to CameraService. Connect method was not found!");
+        return NULL;
     }
 
     dlclose(CameraHALHandle);
@@ -422,7 +469,7 @@ CameraHandler* CameraHandler::initCameraConnect(const CameraCallback& callback,
     camera->setListener(handler);
 
     handler->camera = camera;
-    handler->cameraId = cameraId;
+    handler->cameraId = localCameraIndex;
 
     if (prevCameraParameters != 0)
     {

modules/calib3d/doc/camera_calibration_and_3d_reconstruction.rst

@@ -1486,6 +1486,6 @@ The function reconstructs 3-dimensional points (in homogeneous coordinates) by u
 
 .. [SteweniusCFS] Stewénius, H., Calibrated Fivepoint solver. http://www.vis.uky.edu/~stewe/FIVEPOINT/
 
-.. [Slabaugh] Slabaugh, G.G. Computing Euler angles from a rotation matrix. http://gregslabaugh.name/publications/euler.pdf
+.. [Slabaugh] Slabaugh, G.G. Computing Euler angles from a rotation matrix. http://www.soi.city.ac.uk/~sbbh653/publications/euler.pdf (verified: 2013-04-15)
 
 .. [Zhang2000] Z. Zhang. A Flexible New Technique for Camera Calibration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(11):1330-1334, 2000.

modules/core/src/matmul.cpp

@@ -2850,8 +2850,9 @@ PCA& PCA::operator()(InputArray _data, InputArray __mean, int flags, int maxComp
 
     if( _mean.data )
     {
-        CV_Assert( _mean.size() == mean_sz );
+        CV_Assert( _mean.size() == mean_sz );        
         _mean.convertTo(mean, ctype);
+        covar_flags |= CV_COVAR_USE_AVG; 
     }
 
     calcCovarMatrix( data, covar, mean, covar_flags, ctype );

modules/flann/include/opencv2/flann/ground_truth.h

@@ -42,7 +42,6 @@ template <typename Distance>
 void find_nearest(const Matrix<typename Distance::ElementType>& dataset, typename Distance::ElementType* query, int* matches, int nn,
                   int skip = 0, Distance distance = Distance())
 {
-    typedef typename Distance::ElementType ElementType;
     typedef typename Distance::ResultType DistanceType;
     int n = nn + skip;
 

modules/highgui/CMakeLists.txt

@@ -76,7 +76,26 @@ set(highgui_srcs
 
 file(GLOB highgui_ext_hdrs "include/opencv2/*.hpp" "include/opencv2/${name}/*.hpp" "include/opencv2/${name}/*.h")
 
-if(HAVE_QT)
+if(HAVE_QT5)
+  set(CMAKE_AUTOMOC ON)
+  set(CMAKE_INCLUDE_CURRENT_DIR ON)
+
+  QT5_ADD_RESOURCES(_RCC_OUTFILES src/window_QT.qrc)
+  list(APPEND highgui_srcs src/window_QT.cpp src/window_QT.h ${_RCC_OUTFILES})
+
+  foreach(dt5_dep Core Gui Widgets Test Concurrent)
+    add_definitions(${Qt5${dt5_dep}_DEFINITIONS})
+    include_directories(${Qt5${dt5_dep}_INCLUDE_DIRS})
+    list(APPEND HIGHGUI_LIBRARIES ${Qt5${dt5_dep}_LIBRARIES})
+  endforeach()
+
+  if(HAVE_QT_OPENGL)
+    add_definitions(${Qt5OpenGL_DEFINITIONS})
+    include_directories(${Qt5OpenGL_INCLUDE_DIRS})
+    list(APPEND HIGHGUI_LIBRARIES ${Qt5OpenGL_LIBRARIES})
+  endif()
+
+elseif(HAVE_QT)
   if (HAVE_QT_OPENGL)
     set(QT_USE_QTOPENGL TRUE)
   endif()

modules/highgui/include/opencv2/highgui/highgui_c.h

@@ -306,6 +306,8 @@ enum
     CV_CAP_OPENNI_ASUS =910,   // OpenNI (for Asus Xtion)
 
     CV_CAP_ANDROID  =1000,  // Android
+    CV_CAP_ANDROID_BACK =CV_CAP_ANDROID+99, // Android back camera
+    CV_CAP_ANDROID_FRONT =CV_CAP_ANDROID+98, // Android front camera
 
     CV_CAP_XIAPI    =1100,   // XIMEA Camera API
 

modules/highgui/src/cap_libv4l.cpp

@@ -1665,6 +1665,17 @@ static int icvSetPropertyCAM_V4L(CvCaptureCAM_V4L* capture, int property_id, dou
             width = height = 0;
         }
         break;
+    case CV_CAP_PROP_FPS:
+        struct v4l2_streamparm setfps;
+        memset (&setfps, 0, sizeof(struct v4l2_streamparm));
+        setfps.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
+        setfps.parm.capture.timeperframe.numerator = 1;
+        setfps.parm.capture.timeperframe.denominator = value;
+        if (xioctl (capture->deviceHandle, VIDIOC_S_PARM, &setfps) < 0){
+            fprintf(stderr, "HIGHGUI ERROR: V4L: Unable to set camera FPS\n");
+            retval=0;
+        }
+        break;
     default:
         retval = icvSetControl(capture, property_id, value);
     }

modules/highgui/src/grfmt_jpeg.cpp

@@ -52,6 +52,11 @@
 #include <stdio.h>
 #include <setjmp.h>
 
+// the following defines are a hack to avoid multiple problems with frame ponter handling and setjmp
+// see http://gcc.gnu.org/ml/gcc/2011-10/msg00324.html for some details
+#define mingw_getsp(...) 0
+#define __builtin_frame_address(...) 0
+
 #ifdef WIN32
 
 #define XMD_H // prevent redefinition of INT32

modules/highgui/src/grfmt_png.cpp

@@ -73,6 +73,11 @@
     #pragma warning( disable: 4611 )
 #endif
 
+// the following defines are a hack to avoid multiple problems with frame ponter handling and setjmp
+// see http://gcc.gnu.org/ml/gcc/2011-10/msg00324.html for some details
+#define mingw_getsp(...) 0
+#define __builtin_frame_address(...) 0
+
 namespace cv
 {
 

modules/highgui/src/window_QT.h

@@ -48,13 +48,13 @@
 #endif
 
 #include <QAbstractEventDispatcher>
-#include <QtGui/QApplication>
+#include <QApplication>
 #include <QFile>
 #include <QPushButton>
-#include <QtGui/QGraphicsView>
+#include <QGraphicsView>
 #include <QSizePolicy>
 #include <QInputDialog>
-#include <QtGui/QBoxLayout>
+#include <QBoxLayout>
 #include <QSettings>
 #include <qtimer.h>
 #include <QtConcurrentRun>
@@ -78,7 +78,7 @@
 #include <QRadioButton>
 #include <QButtonGroup>
 #include <QMenu>
-#include <QtTest/QTest>
+#include <QTest>
 
 //start private enum
 enum { CV_MODE_NORMAL = 0, CV_MODE_OPENGL = 1 };

modules/imgproc/doc/structural_analysis_and_shape_descriptors.rst

@@ -342,7 +342,7 @@ Finds the convex hull of a point set.
 
     :param hull_storage: Output memory storage in the old API (``cvConvexHull2`` returns a sequence containing the convex hull points or their indices).
 
-    :param clockwise: Orientation flag. If it is true, the output convex hull is oriented clockwise. Otherwise, it is oriented counter-clockwise. The usual screen coordinate system is assumed so that the origin is at the top-left corner, x axis is oriented to the right, and y axis is oriented downwards.
+    :param clockwise: Orientation flag. If it is true, the output convex hull is oriented clockwise. Otherwise, it is oriented counter-clockwise. The assumed coordinate system has its X axis pointing to the right, and its Y axis pointing upwards.
 
     :param orientation: Convex hull orientation parameter in the old API, ``CV_CLOCKWISE`` or ``CV_COUNTERCLOCKWISE``.
 

modules/imgproc/src/floodfill.cpp

@@ -127,7 +127,6 @@ floodFill_CnIR( Mat& image, Point seed,
                _Tp newVal, ConnectedComp* region, int flags,
                std::vector<FFillSegment>* buffer )
 {
-    typedef typename DataType<_Tp>::channel_type _CTp;
     _Tp* img = (_Tp*)(image.data + image.step * seed.y);
     Size roi = image.size();
     int i, L, R;
@@ -279,7 +278,6 @@ floodFillGrad_CnIR( Mat& image, Mat& msk,
                    Diff diff, ConnectedComp* region, int flags,
                    std::vector<FFillSegment>* buffer )
 {
-    typedef typename DataType<_Tp>::channel_type _CTp;
     int step = (int)image.step, maskStep = (int)msk.step;
     uchar* pImage = image.data;
     _Tp* img = (_Tp*)(pImage + step*seed.y);
@@ -610,7 +608,7 @@ int cv::floodFill( InputOutputArray _image, InputOutputArray _mask,
                 &comp, flags, &buffer);
     else
         CV_Error(CV_StsUnsupportedFormat, "");
-    
+
     if( rect )
         *rect = comp.rect;
     return comp.area;

modules/imgproc/src/imgwarp.cpp

@@ -1219,8 +1219,6 @@ static void resizeGeneric_( const Mat& src, Mat& dst,
                             const int* yofs, const void* _beta,
                             int xmin, int xmax, int ksize )
 {
-    typedef typename HResize::value_type T;
-    typedef typename HResize::buf_type WT;
     typedef typename HResize::alpha_type AT;
 
     const AT* beta = (const AT*)_beta;

modules/java/android_lib/lint.xml

@@ -1,5 +1,8 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <lint>
+    <issue id="InlinedApi">
+        <ignore path="src\org\opencv\android\JavaCameraView.java" />
+    </issue>
     <issue id="NewApi">
         <ignore path="src\org\opencv\android\JavaCameraView.java" />
     </issue>

modules/java/android_lib/res/values/attrs.xml

@@ -4,8 +4,8 @@
        <attr name="show_fps" format="boolean"/>
        <attr name="camera_id" format="integer" >
           <enum name="any" value="-1" />
-          <enum name="back" value="0" />
-          <enum name="front" value="1" />
+          <enum name="back" value="99" />
+          <enum name="front" value="98" />
        </attr>
     </declare-styleable>
 </resources>

modules/java/generator/src/java/android+CameraBridgeViewBase.java

@@ -47,10 +47,14 @@ public abstract class CameraBridgeViewBase extends SurfaceView implements Surfac
     protected int mMaxWidth;
     protected float mScale = 0;
     protected int mPreviewFormat = Highgui.CV_CAP_ANDROID_COLOR_FRAME_RGBA;
-    protected int mCameraIndex = -1;
+    protected int mCameraIndex = CAMERA_ID_ANY;
     protected boolean mEnabled;
     protected FpsMeter mFpsMeter = null;
 
+    public static final int CAMERA_ID_ANY   = -1;
+    public static final int CAMERA_ID_BACK  = 99;
+    public static final int CAMERA_ID_FRONT = 98;
+
     public CameraBridgeViewBase(Context context, int cameraId) {
         super(context);
         mCameraIndex = cameraId;
@@ -74,6 +78,7 @@ public abstract class CameraBridgeViewBase extends SurfaceView implements Surfac
         getHolder().addCallback(this);
         mMaxWidth = MAX_UNSPECIFIED;
         mMaxHeight = MAX_UNSPECIFIED;
+        styledAttrs.recycle();
     }
 
     public interface CvCameraViewListener {
@@ -155,8 +160,6 @@ public abstract class CameraBridgeViewBase extends SurfaceView implements Surfac
             mPreviewFormat = format;
         }
 
-        private CvCameraViewListenerAdapter() {}
-
         private int mPreviewFormat = Highgui.CV_CAP_ANDROID_COLOR_FRAME_RGBA;
         private CvCameraViewListener mOldStyleListener;
     };

modules/java/generator/src/java/android+JavaCameraView.java

@@ -6,6 +6,7 @@ import android.content.Context;
 import android.graphics.ImageFormat;
 import android.graphics.SurfaceTexture;
 import android.hardware.Camera;
+import android.hardware.Camera.CameraInfo;
 import android.hardware.Camera.PreviewCallback;
 import android.os.Build;
 import android.util.AttributeSet;
@@ -68,7 +69,7 @@ public class JavaCameraView extends CameraBridgeViewBase implements PreviewCallb
         synchronized (this) {
             mCamera = null;
 
-            if (mCameraIndex == -1) {
+            if (mCameraIndex == CAMERA_ID_ANY) {
                 Log.d(TAG, "Trying to open camera with old open()");
                 try {
                     mCamera = Camera.open();
@@ -92,11 +93,39 @@ public class JavaCameraView extends CameraBridgeViewBase implements PreviewCallb
                 }
             } else {
                 if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.GINGERBREAD) {
-                    Log.d(TAG, "Trying to open camera with new open(" + Integer.valueOf(mCameraIndex) + ")");
-                    try {
-                        mCamera = Camera.open(mCameraIndex);
-                    } catch (RuntimeException e) {
-                        Log.e(TAG, "Camera #" + mCameraIndex + "failed to open: " + e.getLocalizedMessage());
+                    int localCameraIndex = mCameraIndex;
+                    if (mCameraIndex == CAMERA_ID_BACK) {
+                        Log.i(TAG, "Trying to open back camera");
+                        Camera.CameraInfo cameraInfo = new Camera.CameraInfo();
+                        for (int camIdx = 0; camIdx < Camera.getNumberOfCameras(); ++camIdx) {
+                            Camera.getCameraInfo( camIdx, cameraInfo );
+                            if (cameraInfo.facing == Camera.CameraInfo.CAMERA_FACING_BACK) {
+                                localCameraIndex = camIdx;
+                                break;
+                            }
+                        }
+                    } else if (mCameraIndex == CAMERA_ID_FRONT) {
+                        Log.i(TAG, "Trying to open front camera");
+                        Camera.CameraInfo cameraInfo = new Camera.CameraInfo();
+                        for (int camIdx = 0; camIdx < Camera.getNumberOfCameras(); ++camIdx) {
+                            Camera.getCameraInfo( camIdx, cameraInfo );
+                            if (cameraInfo.facing == Camera.CameraInfo.CAMERA_FACING_FRONT) {
+                                localCameraIndex = camIdx;
+                                break;
+                            }
+                        }
+                    }
+                    if (localCameraIndex == CAMERA_ID_BACK) {
+                        Log.e(TAG, "Back camera not found!");
+                    } else if (localCameraIndex == CAMERA_ID_FRONT) {
+                        Log.e(TAG, "Front camera not found!");
+                    } else {
+                        Log.d(TAG, "Trying to open camera with new open(" + Integer.valueOf(localCameraIndex) + ")");
+                        try {
+                            mCamera = Camera.open(localCameraIndex);
+                        } catch (RuntimeException e) {
+                            Log.e(TAG, "Camera #" + localCameraIndex + "failed to open: " + e.getLocalizedMessage());
+                        }
                     }
                 }
             }
@@ -179,6 +208,8 @@ public class JavaCameraView extends CameraBridgeViewBase implements PreviewCallb
         synchronized (this) {
             if (mCamera != null) {
                 mCamera.stopPreview();
+                mCamera.setPreviewCallback(null);
+
                 mCamera.release();
             }
             mCamera = null;
@@ -267,9 +298,6 @@ public class JavaCameraView extends CameraBridgeViewBase implements PreviewCallb
             mRgba.release();
         }
 
-        private JavaCameraFrame(CvCameraViewFrame obj) {
-        }
-
         private Mat mYuvFrameData;
         private Mat mRgba;
         private int mWidth;

modules/java/generator/src/java/android+NativeCameraView.java

@@ -53,14 +53,16 @@ public class NativeCameraView extends CameraBridgeViewBase {
         /* 1. We need to stop thread which updating the frames
          * 2. Stop camera and release it
          */
-        try {
-            mStopThread = true;
-            mThread.join();
-        } catch (InterruptedException e) {
-            e.printStackTrace();
-        } finally {
-            mThread =  null;
-            mStopThread = false;
+        if (mThread != null) {
+            try {
+                mStopThread = true;
+                mThread.join();
+            } catch (InterruptedException e) {
+                e.printStackTrace();
+            } finally {
+                mThread =  null;
+                mStopThread = false;
+            }
         }
 
         /* Now release camera */
@@ -131,17 +133,17 @@ public class NativeCameraView extends CameraBridgeViewBase {
         }
     }
 
-    private class NativeCameraFrame implements CvCameraViewFrame {
+    private static class NativeCameraFrame implements CvCameraViewFrame {
 
         @Override
         public Mat rgba() {
-            mCamera.retrieve(mRgba, Highgui.CV_CAP_ANDROID_COLOR_FRAME_RGBA);
+            mCapture.retrieve(mRgba, Highgui.CV_CAP_ANDROID_COLOR_FRAME_RGBA);
             return mRgba;
         }
 
         @Override
         public Mat gray() {
-            mCamera.retrieve(mGray, Highgui.CV_CAP_ANDROID_GREY_FRAME);
+            mCapture.retrieve(mGray, Highgui.CV_CAP_ANDROID_GREY_FRAME);
             return mGray;
         }
 
@@ -158,9 +160,6 @@ public class NativeCameraView extends CameraBridgeViewBase {
 
     private class CameraWorker implements Runnable {
 
-        private Mat mRgba = new Mat();
-        private Mat mGray = new Mat();
-
         public void run() {
             do {
                 if (!mCamera.grab()) {

modules/legacy/src/blobtrackgenyml.cpp

@@ -61,7 +61,7 @@ protected:
     {
         int     ObjNum = m_TrackList.GetBlobNum();
         int     i;
-        char    video_name[1024];
+        char    video_name[1024+1];
         char*   struct_name = NULL;
         CvFileStorage* storage = cvOpenFileStorage(m_pFileName,NULL,CV_STORAGE_WRITE_TEXT);
 

modules/legacy/src/kdtree.cpp

@@ -117,10 +117,10 @@ class CvKDTreeWrap : public CvFeatureTree {
            CvMat* results) {
     int rn = results->rows * results->cols;
     std::vector<int> inbounds;
-    dispatch_cvtype(mat, ((__treetype*)data)->
-        find_ortho_range((typename __treetype::scalar_type*)bounds_min->data.ptr,
+    assert(CV_MAT_DEPTH(mat->type) == CV_32F || CV_MAT_DEPTH(mat->type) == CV_64F);
+    ((__treetype*)data)->find_ortho_range((typename __treetype::scalar_type*)bounds_min->data.ptr,
              (typename __treetype::scalar_type*)bounds_max->data.ptr,
-             inbounds));
+             inbounds);
     std::copy(inbounds.begin(),
         inbounds.begin() + std::min((int)inbounds.size(), rn),
         (int*) results->data.ptr);

modules/objdetect/src/cascadedetect.cpp

@@ -1140,7 +1140,7 @@ void CascadeClassifier::detectMultiScale( const Mat& image, std::vector<Rect>& o
 
         Size windowSize( cvRound(originalWindowSize.width*factor), cvRound(originalWindowSize.height*factor) );
         Size scaledImageSize( cvRound( grayImage.cols/factor ), cvRound( grayImage.rows/factor ) );
-        Size processingRectSize( scaledImageSize.width - originalWindowSize.width + 1, scaledImageSize.height - originalWindowSize.height + 1 );
+        Size processingRectSize( scaledImageSize.width - originalWindowSize.width, scaledImageSize.height - originalWindowSize.height );
 
         if( processingRectSize.width <= 0 || processingRectSize.height <= 0 )
             break;

modules/ocl/include/opencv2/ocl.hpp

@@ -151,7 +151,7 @@ namespace cv
             static Context *getContext();
             static void setContext(Info &oclinfo);
 
-            enum {CL_DOUBLE, CL_UNIFIED_MEM};
+            enum {CL_DOUBLE, CL_UNIFIED_MEM, CL_VER_1_2};
             bool supportsFeature(int ftype);
             size_t computeUnits();
             size_t maxWorkGroupSize();
@@ -264,9 +264,9 @@ namespace cv
             void create(Size size, int type);
 
             //! allocates new oclMatrix with specified device memory type.
-            void createEx(int rows, int cols, int type, 
+            void createEx(int rows, int cols, int type,
                           DevMemRW rw_type, DevMemType mem_type, void* hptr = 0);
-            void createEx(Size size, int type, DevMemRW rw_type, 
+            void createEx(Size size, int type, DevMemRW rw_type,
                           DevMemType mem_type, void* hptr = 0);
 
             //! decreases reference counter;
@@ -406,6 +406,9 @@ namespace cv
         //! computes element-wise product of the two arrays (c = a * b)
         // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
         CV_EXPORTS void multiply(const oclMat &a, const oclMat &b, oclMat &c, double scale = 1);
+        //! multiplies matrix to a number (dst = scalar * src)
+        // supports CV_32FC1 only
+        CV_EXPORTS void multiply(double scalar, const oclMat &src, oclMat &dst);
         //! computes element-wise quotient of the two arrays (c = a / b)
         // supports all types except CV_8SC1,CV_8SC2,CV8SC3 and CV_8SC4
         CV_EXPORTS void divide(const oclMat &a, const oclMat &b, oclMat &c, double scale = 1);
@@ -823,7 +826,44 @@ namespace cv
         };
 #endif
 
+        class CV_EXPORTS OclCascadeClassifierBuf : public  cv::CascadeClassifier
+        {
+        public:
+            OclCascadeClassifierBuf() :
+                m_flags(0), initialized(false), m_scaleFactor(0), buffers(NULL) {}
 
+            ~OclCascadeClassifierBuf() {}
+
+            void detectMultiScale(oclMat &image, CV_OUT std::vector<cv::Rect>& faces,
+                                  double scaleFactor = 1.1, int minNeighbors = 3, int flags = 0,
+                                  Size minSize = Size(), Size maxSize = Size());
+            void release();
+
+        private:
+            void Init(const int rows, const int cols, double scaleFactor, int flags,
+                      const int outputsz, const size_t localThreads[],
+                      Size minSize, Size maxSize);
+            void CreateBaseBufs(const int datasize, const int totalclassifier, const int flags, const int outputsz);
+            void CreateFactorRelatedBufs(const int rows, const int cols, const int flags,
+                                         const double scaleFactor, const size_t localThreads[],
+                                         Size minSize, Size maxSize);
+            void GenResult(CV_OUT std::vector<cv::Rect>& faces, const std::vector<cv::Rect> &rectList, const std::vector<int> &rweights);
+
+            int m_rows;
+            int m_cols;
+            int m_flags;
+            int m_loopcount;
+            int m_nodenum;
+            bool findBiggestObject;
+            bool initialized;
+            double m_scaleFactor;
+            Size m_minSize;
+            Size m_maxSize;
+            std::vector<Size> sizev;
+            std::vector<float> scalev;
+            oclMat gimg1, gsum, gsqsum;
+            void * buffers;
+        };
 
         /////////////////////////////// Pyramid /////////////////////////////////////
         CV_EXPORTS void pyrDown(const oclMat &src, oclMat &dst);
@@ -849,7 +889,6 @@ namespace cv
             std::vector<oclMat> image_sqsums;
         };
 
-
         //! computes the proximity map for the raster template and the image where the template is searched for
         // Supports TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED for type 8UC1 and 8UC4
         // Supports TM_SQDIFF, TM_CCORR for type 32FC1 and 32FC4
@@ -1093,13 +1132,11 @@ namespace cv
         /****************************************************************************************\
         *                                      Distance                                          *
         \****************************************************************************************/
-
         template<typename T>
         struct CV_EXPORTS Accumulator
         {
             typedef T Type;
         };
-
         template<> struct Accumulator<unsigned char>
         {
             typedef float Type;
@@ -1173,469 +1210,244 @@ namespace cv
         {
         public:
             enum DistType {L1Dist = 0, L2Dist, HammingDist};
-
             explicit BruteForceMatcher_OCL_base(DistType distType = L2Dist);
 
-
-
             // Add descriptors to train descriptor collection
-
             void add(const std::vector<oclMat> &descCollection);
 
-
-
             // Get train descriptors collection
-
             const std::vector<oclMat> &getTrainDescriptors() const;
 
-
-
             // Clear train descriptors collection
-
             void clear();
 
-
-
             // Return true if there are not train descriptors in collection
-
             bool empty() const;
 
-
-
             // Return true if the matcher supports mask in match methods
-
             bool isMaskSupported() const;
 
-
-
             // Find one best match for each query descriptor
-
             void matchSingle(const oclMat &query, const oclMat &train,
-
                              oclMat &trainIdx, oclMat &distance,
-
                              const oclMat &mask = oclMat());
 
-
-
             // Download trainIdx and distance and convert it to CPU vector with DMatch
-
             static void matchDownload(const oclMat &trainIdx, const oclMat &distance, std::vector<DMatch> &matches);
-
             // Convert trainIdx and distance to vector with DMatch
-
             static void matchConvert(const Mat &trainIdx, const Mat &distance, std::vector<DMatch> &matches);
 
-
-
             // Find one best match for each query descriptor
-
             void match(const oclMat &query, const oclMat &train, std::vector<DMatch> &matches, const oclMat &mask = oclMat());
 
-
-
             // Make gpu collection of trains and masks in suitable format for matchCollection function
-
             void makeGpuCollection(oclMat &trainCollection, oclMat &maskCollection, const std::vector<oclMat> &masks = std::vector<oclMat>());
 
-
-
             // Find one best match from train collection for each query descriptor
-
             void matchCollection(const oclMat &query, const oclMat &trainCollection,
-
                                  oclMat &trainIdx, oclMat &imgIdx, oclMat &distance,
-
                                  const oclMat &masks = oclMat());
 
-
-
             // Download trainIdx, imgIdx and distance and convert it to vector with DMatch
-
             static void matchDownload(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance, std::vector<DMatch> &matches);
-
             // Convert trainIdx, imgIdx and distance to vector with DMatch
-
             static void matchConvert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance, std::vector<DMatch> &matches);
 
-
-
             // Find one best match from train collection for each query descriptor.
-
             void match(const oclMat &query, std::vector<DMatch> &matches, const std::vector<oclMat> &masks = std::vector<oclMat>());
 
-
-
             // Find k best matches for each query descriptor (in increasing order of distances)
-
             void knnMatchSingle(const oclMat &query, const oclMat &train,
-
                                 oclMat &trainIdx, oclMat &distance, oclMat &allDist, int k,
-
                                 const oclMat &mask = oclMat());
 
-
-
             // Download trainIdx and distance and convert it to vector with DMatch
-
             // compactResult is used when mask is not empty. If compactResult is false matches
-
             // vector will have the same size as queryDescriptors rows. If compactResult is true
-
             // matches vector will not contain matches for fully masked out query descriptors.
-
             static void knnMatchDownload(const oclMat &trainIdx, const oclMat &distance,
-
                                          std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
-
             // Convert trainIdx and distance to vector with DMatch
-
             static void knnMatchConvert(const Mat &trainIdx, const Mat &distance,
-
                                         std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
 
-
-
             // Find k best matches for each query descriptor (in increasing order of distances).
-
             // compactResult is used when mask is not empty. If compactResult is false matches
-
             // vector will have the same size as queryDescriptors rows. If compactResult is true
-
             // matches vector will not contain matches for fully masked out query descriptors.
-
             void knnMatch(const oclMat &query, const oclMat &train,
-
                           std::vector< std::vector<DMatch> > &matches, int k, const oclMat &mask = oclMat(),
-
                           bool compactResult = false);
 
-
-
             // Find k best matches from train collection for each query descriptor (in increasing order of distances)
-
             void knnMatch2Collection(const oclMat &query, const oclMat &trainCollection,
-
                                      oclMat &trainIdx, oclMat &imgIdx, oclMat &distance,
-
                                      const oclMat &maskCollection = oclMat());
 
-
-
             // Download trainIdx and distance and convert it to vector with DMatch
-
             // compactResult is used when mask is not empty. If compactResult is false matches
-
             // vector will have the same size as queryDescriptors rows. If compactResult is true
-
             // matches vector will not contain matches for fully masked out query descriptors.
-
             static void knnMatch2Download(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance,
-
                                           std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
-
             // Convert trainIdx and distance to vector with DMatch
-
             static void knnMatch2Convert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance,
-
                                          std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
 
-
-
             // Find k best matches  for each query descriptor (in increasing order of distances).
-
             // compactResult is used when mask is not empty. If compactResult is false matches
-
             // vector will have the same size as queryDescriptors rows. If compactResult is true
-
             // matches vector will not contain matches for fully masked out query descriptors.
-
             void knnMatch(const oclMat &query, std::vector< std::vector<DMatch> > &matches, int k,
-
                           const std::vector<oclMat> &masks = std::vector<oclMat>(), bool compactResult = false);
 
-
-
             // Find best matches for each query descriptor which have distance less than maxDistance.
-
             // nMatches.at<int>(0, queryIdx) will contain matches count for queryIdx.
-
             // carefully nMatches can be greater than trainIdx.cols - it means that matcher didn't find all matches,
-
             // because it didn't have enough memory.
-
             // If trainIdx is empty, then trainIdx and distance will be created with size nQuery x max((nTrain / 100), 10),
-
             // otherwize user can pass own allocated trainIdx and distance with size nQuery x nMaxMatches
-
             // Matches doesn't sorted.
-
             void radiusMatchSingle(const oclMat &query, const oclMat &train,
-
                                    oclMat &trainIdx, oclMat &distance, oclMat &nMatches, float maxDistance,
-
                                    const oclMat &mask = oclMat());
 
-
-
             // Download trainIdx, nMatches and distance and convert it to vector with DMatch.
-
             // matches will be sorted in increasing order of distances.
-
             // compactResult is used when mask is not empty. If compactResult is false matches
-
             // vector will have the same size as queryDescriptors rows. If compactResult is true
-
             // matches vector will not contain matches for fully masked out query descriptors.
-
             static void radiusMatchDownload(const oclMat &trainIdx, const oclMat &distance, const oclMat &nMatches,
-
                                             std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
-
             // Convert trainIdx, nMatches and distance to vector with DMatch.
-
             static void radiusMatchConvert(const Mat &trainIdx, const Mat &distance, const Mat &nMatches,
-
                                            std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
 
-
-
             // Find best matches for each query descriptor which have distance less than maxDistance
-
             // in increasing order of distances).
-
             void radiusMatch(const oclMat &query, const oclMat &train,
-
                              std::vector< std::vector<DMatch> > &matches, float maxDistance,
-
                              const oclMat &mask = oclMat(), bool compactResult = false);
 
-
-
             // Find best matches for each query descriptor which have distance less than maxDistance.
-
             // If trainIdx is empty, then trainIdx and distance will be created with size nQuery x max((nQuery / 100), 10),
-
             // otherwize user can pass own allocated trainIdx and distance with size nQuery x nMaxMatches
-
             // Matches doesn't sorted.
-
             void radiusMatchCollection(const oclMat &query, oclMat &trainIdx, oclMat &imgIdx, oclMat &distance, oclMat &nMatches, float maxDistance,
-
                                        const std::vector<oclMat> &masks = std::vector<oclMat>());
 
-
-
             // Download trainIdx, imgIdx, nMatches and distance and convert it to vector with DMatch.
-
             // matches will be sorted in increasing order of distances.
-
             // compactResult is used when mask is not empty. If compactResult is false matches
-
             // vector will have the same size as queryDescriptors rows. If compactResult is true
-
             // matches vector will not contain matches for fully masked out query descriptors.
-
             static void radiusMatchDownload(const oclMat &trainIdx, const oclMat &imgIdx, const oclMat &distance, const oclMat &nMatches,
-
                                             std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
-
             // Convert trainIdx, nMatches and distance to vector with DMatch.
-
             static void radiusMatchConvert(const Mat &trainIdx, const Mat &imgIdx, const Mat &distance, const Mat &nMatches,
-
                                            std::vector< std::vector<DMatch> > &matches, bool compactResult = false);
 
-
-
             // Find best matches from train collection for each query descriptor which have distance less than
-
             // maxDistance (in increasing order of distances).
-
             void radiusMatch(const oclMat &query, std::vector< std::vector<DMatch> > &matches, float maxDistance,
-
                              const std::vector<oclMat> &masks = std::vector<oclMat>(), bool compactResult = false);
 
-
-
             DistType distType;
 
-
-
         private:
-
             std::vector<oclMat> trainDescCollection;
-
         };
 
-
-
         template <class Distance>
-
         class CV_EXPORTS BruteForceMatcher_OCL;
 
-
-
         template <typename T>
-
         class CV_EXPORTS BruteForceMatcher_OCL< L1<T> > : public BruteForceMatcher_OCL_base
-
         {
-
         public:
-
             explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(L1Dist) {}
-
             explicit BruteForceMatcher_OCL(L1<T> /*d*/) : BruteForceMatcher_OCL_base(L1Dist) {}
-
         };
-
         template <typename T>
-
         class CV_EXPORTS BruteForceMatcher_OCL< L2<T> > : public BruteForceMatcher_OCL_base
-
         {
-
         public:
-
             explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(L2Dist) {}
-
             explicit BruteForceMatcher_OCL(L2<T> /*d*/) : BruteForceMatcher_OCL_base(L2Dist) {}
-
         };
-
         template <> class CV_EXPORTS BruteForceMatcher_OCL< Hamming > : public BruteForceMatcher_OCL_base
-
         {
-
         public:
-
             explicit BruteForceMatcher_OCL() : BruteForceMatcher_OCL_base(HammingDist) {}
-
             explicit BruteForceMatcher_OCL(Hamming /*d*/) : BruteForceMatcher_OCL_base(HammingDist) {}
-
         };
 
-
+        class CV_EXPORTS BFMatcher_OCL : public BruteForceMatcher_OCL_base
+        {
+        public:
+            explicit BFMatcher_OCL(int norm = NORM_L2) : BruteForceMatcher_OCL_base(norm == NORM_L1 ? L1Dist : norm == NORM_L2 ? L2Dist : HammingDist) {}
+        };
 
         /////////////////////////////// PyrLKOpticalFlow /////////////////////////////////////
 
         class CV_EXPORTS PyrLKOpticalFlow
-
         {
-
         public:
-
             PyrLKOpticalFlow()
-
             {
-
                 winSize = Size(21, 21);
-
                 maxLevel = 3;
-
                 iters = 30;
-
                 derivLambda = 0.5;
-
                 useInitialFlow = false;
-
                 minEigThreshold = 1e-4f;
-
                 getMinEigenVals = false;
-
                 isDeviceArch11_ = false;
-
             }
 
-
-
             void sparse(const oclMat &prevImg, const oclMat &nextImg, const oclMat &prevPts, oclMat &nextPts,
-
                         oclMat &status, oclMat *err = 0);
 
-
-
             void dense(const oclMat &prevImg, const oclMat &nextImg, oclMat &u, oclMat &v, oclMat *err = 0);
 
-
-
             Size winSize;
-
             int maxLevel;
-
             int iters;
-
             double derivLambda;
-
             bool useInitialFlow;
-
             float minEigThreshold;
-
             bool getMinEigenVals;
 
-
-
             void releaseMemory()
-
             {
-
                 dx_calcBuf_.release();
-
                 dy_calcBuf_.release();
 
-
-
                 prevPyr_.clear();
-
                 nextPyr_.clear();
 
-
-
                 dx_buf_.release();
-
                 dy_buf_.release();
-
             }
 
-
-
         private:
-
             void calcSharrDeriv(const oclMat &src, oclMat &dx, oclMat &dy);
 
-
-
             void buildImagePyramid(const oclMat &img0, std::vector<oclMat> &pyr, bool withBorder);
 
-
-
             oclMat dx_calcBuf_;
-
             oclMat dy_calcBuf_;
 
-
-
             std::vector<oclMat> prevPyr_;
-
             std::vector<oclMat> nextPyr_;
 
-
-
             oclMat dx_buf_;
-
             oclMat dy_buf_;
 
-
-
             oclMat uPyr_[2];
-
             oclMat vPyr_[2];
 
-
-
             bool isDeviceArch11_;
-
         };
         //////////////// build warping maps ////////////////////
         //! builds plane warping maps
@@ -1706,6 +1518,7 @@ namespace cv
         private:
             oclMat minSSD, leBuf, riBuf;
         };
+
         class CV_EXPORTS StereoBeliefPropagation
         {
         public:
@@ -1736,6 +1549,133 @@ namespace cv
             std::vector<oclMat> datas;
             oclMat out;
         };
+
+        class CV_EXPORTS StereoConstantSpaceBP
+        {
+        public:
+            enum { DEFAULT_NDISP    = 128 };
+            enum { DEFAULT_ITERS    = 8   };
+            enum { DEFAULT_LEVELS   = 4   };
+            enum { DEFAULT_NR_PLANE = 4   };
+            static void estimateRecommendedParams(int width, int height, int &ndisp, int &iters, int &levels, int &nr_plane);
+            explicit StereoConstantSpaceBP(
+                int ndisp    = DEFAULT_NDISP,
+                int iters    = DEFAULT_ITERS,
+                int levels   = DEFAULT_LEVELS,
+                int nr_plane = DEFAULT_NR_PLANE,
+                int msg_type = CV_32F);
+            StereoConstantSpaceBP(int ndisp, int iters, int levels, int nr_plane,
+                float max_data_term, float data_weight, float max_disc_term, float disc_single_jump,
+                int min_disp_th = 0,
+                int msg_type = CV_32F);
+            void operator()(const oclMat &left, const oclMat &right, oclMat &disparity);
+            int ndisp;
+            int iters;
+            int levels;
+            int nr_plane;
+            float max_data_term;
+            float data_weight;
+            float max_disc_term;
+            float disc_single_jump;
+            int min_disp_th;
+            int msg_type;
+            bool use_local_init_data_cost;
+        private:
+            oclMat u[2], d[2], l[2], r[2];
+            oclMat disp_selected_pyr[2];
+            oclMat data_cost;
+            oclMat data_cost_selected;
+            oclMat temp;
+            oclMat out;
+        };
+
+        // Implementation of the Zach, Pock and Bischof Dual TV-L1 Optical Flow method
+        //
+        // see reference:
+        //   [1] C. Zach, T. Pock and H. Bischof, "A Duality Based Approach for Realtime TV-L1 Optical Flow".
+        //   [2] Javier Sanchez, Enric Meinhardt-Llopis and Gabriele Facciolo. "TV-L1 Optical Flow Estimation".
+        class CV_EXPORTS OpticalFlowDual_TVL1_OCL
+        {
+        public:
+            OpticalFlowDual_TVL1_OCL();
+
+            void operator ()(const oclMat& I0, const oclMat& I1, oclMat& flowx, oclMat& flowy);
+
+            void collectGarbage();
+
+            /**
+            * Time step of the numerical scheme.
+            */
+            double tau;
+
+            /**
+            * Weight parameter for the data term, attachment parameter.
+            * This is the most relevant parameter, which determines the smoothness of the output.
+            * The smaller this parameter is, the smoother the solutions we obtain.
+            * It depends on the range of motions of the images, so its value should be adapted to each image sequence.
+            */
+            double lambda;
+
+            /**
+            * Weight parameter for (u - v)^2, tightness parameter.
+            * It serves as a link between the attachment and the regularization terms.
+            * In theory, it should have a small value in order to maintain both parts in correspondence.
+            * The method is stable for a large range of values of this parameter.
+            */
+            double theta;
+
+            /**
+            * Number of scales used to create the pyramid of images.
+            */
+            int nscales;
+
+            /**
+            * Number of warpings per scale.
+            * Represents the number of times that I1(x+u0) and grad( I1(x+u0) ) are computed per scale.
+            * This is a parameter that assures the stability of the method.
+            * It also affects the running time, so it is a compromise between speed and accuracy.
+            */
+            int warps;
+
+            /**
+            * Stopping criterion threshold used in the numerical scheme, which is a trade-off between precision and running time.
+            * A small value will yield more accurate solutions at the expense of a slower convergence.
+            */
+            double epsilon;
+
+            /**
+            * Stopping criterion iterations number used in the numerical scheme.
+            */
+            int iterations;
+
+            bool useInitialFlow;
+
+        private:
+            void procOneScale(const oclMat& I0, const oclMat& I1, oclMat& u1, oclMat& u2);
+
+            std::vector<oclMat> I0s;
+            std::vector<oclMat> I1s;
+            std::vector<oclMat> u1s;
+            std::vector<oclMat> u2s;
+
+            oclMat I1x_buf;
+            oclMat I1y_buf;
+
+            oclMat I1w_buf;
+            oclMat I1wx_buf;
+            oclMat I1wy_buf;
+
+            oclMat grad_buf;
+            oclMat rho_c_buf;
+
+            oclMat p11_buf;
+            oclMat p12_buf;
+            oclMat p21_buf;
+            oclMat p22_buf;
+
+            oclMat diff_buf;
+            oclMat norm_buf;
+        };
     }
 }
 #if defined _MSC_VER && _MSC_VER >= 1200

modules/ocl/include/opencv2/ocl/ocl.hpp

@@ -45,4 +45,4 @@
 #error this is a compatibility header which should not be used inside the OpenCV library
 #endif
 
-#include "opencv2/ocl.hpp"
+#include "opencv2/ocl.hpp"

modules/ocl/src/arithm.cpp

@@ -22,6 +22,7 @@
 //    Jiang Liyuan, jlyuan001.good@163.com
 //    Rock Li, Rock.Li@amd.com
 //    Zailong Wu, bullet@yeah.net
+//    Peng Xiao, pengxiao@outlook.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -61,8 +62,6 @@ namespace cv
     namespace ocl
     {
         ////////////////////////////////OpenCL kernel strings/////////////////////
-        extern const char *bitwise;
-        extern const char *bitwiseM;
         extern const char *transpose_kernel;
         extern const char *arithm_nonzero;
         extern const char *arithm_sum;
@@ -76,19 +75,11 @@ namespace cv
         extern const char *arithm_add;
         extern const char *arithm_add_scalar;
         extern const char *arithm_add_scalar_mask;
+        extern const char *arithm_bitwise_binary;
+        extern const char *arithm_bitwise_binary_mask;
+        extern const char *arithm_bitwise_binary_scalar;
+        extern const char *arithm_bitwise_binary_scalar_mask;
         extern const char *arithm_bitwise_not;
-        extern const char *arithm_bitwise_and;
-        extern const char *arithm_bitwise_and_mask;
-        extern const char *arithm_bitwise_and_scalar;
-        extern const char *arithm_bitwise_and_scalar_mask;
-        extern const char *arithm_bitwise_or;
-        extern const char *arithm_bitwise_or_mask;
-        extern const char *arithm_bitwise_or_scalar;
-        extern const char *arithm_bitwise_or_scalar_mask;
-        extern const char *arithm_bitwise_xor;
-        extern const char *arithm_bitwise_xor_mask;
-        extern const char *arithm_bitwise_xor_scalar;
-        extern const char *arithm_bitwise_xor_scalar_mask;
         extern const char *arithm_compare_eq;
         extern const char *arithm_compare_ne;
         extern const char *arithm_mul;
@@ -126,7 +117,7 @@ inline int divUp(int total, int grain)
 /////////////////////// add subtract multiply divide /////////////////////////
 //////////////////////////////////////////////////////////////////////////////
 template<typename T>
-void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, 
+void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
                     String kernelName, const char **kernelString, void *_scalar, int op_type = 0)
 {
     if(!src1.clCxt->supportsFeature(Context::CL_DOUBLE) && src1.type() == CV_64F)
@@ -195,12 +186,12 @@ void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
             openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, -1, depth);
     }
 }
-static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, 
+static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
                            String kernelName, const char **kernelString, int op_type = 0)
 {
     arithmetic_run<char>(src1, src2, dst, kernelName, kernelString, (void *)NULL, op_type);
 }
-static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask, 
+static void arithmetic_run(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask,
                            String kernelName, const char **kernelString, int op_type = 0)
 {
     if(!src1.clCxt->supportsFeature(Context::CL_DOUBLE) && src1.type() == CV_64F)
@@ -295,6 +286,7 @@ void cv::ocl::multiply(const oclMat &src1, const oclMat &src2, oclMat &dst, doub
     else
         arithmetic_run<float>(src1, src2, dst, "arithm_mul", &arithm_mul, (void *)(&scalar));
 }
+
 void cv::ocl::divide(const oclMat &src1, const oclMat &src2, oclMat &dst, double scalar)
 {
 
@@ -479,6 +471,11 @@ void cv::ocl::subtract(const Scalar &src2, const oclMat &src1, oclMat &dst, cons
 
     arithmetic_scalar( src1, src2, dst, mask, kernelName, kernelString, -1);
 }
+void cv::ocl::multiply(double scalar, const oclMat &src, oclMat &dst)
+{
+    String kernelName = "arithm_muls";
+    arithmetic_scalar_run( src, dst, kernelName, &arithm_mul, scalar);
+}
 void cv::ocl::divide(double scalar, const oclMat &src,  oclMat &dst)
 {
     if(!src.clCxt->supportsFeature(Context::CL_DOUBLE))
@@ -1647,7 +1644,8 @@ static void bitwise_run(const oclMat &src1, oclMat &dst, String kernelName, cons
 
 
 template<typename T>
-void bitwise_run(const oclMat &src1, const oclMat &src2, oclMat &dst, String kernelName, const char **kernelString, void *_scalar)
+void bitwise_run(const oclMat &src1, const oclMat &src2, oclMat &dst, String kernelName,
+ const char **kernelString, void *_scalar, const char* _opt = NULL)
 {
     dst.create(src1.size(), src1.type());
     CV_Assert(src1.cols == src2.cols && src2.cols == dst.cols &&
@@ -1697,13 +1695,15 @@ void bitwise_run(const oclMat &src1, const oclMat &src2, oclMat &dst, String ker
         args.push_back( std::make_pair( sizeof(T), (void *)&scalar ));
     }
 
-    openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, -1, depth);
+    openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, -1, depth, _opt);
 }
-static void bitwise_run(const oclMat &src1, const oclMat &src2, oclMat &dst, String kernelName, const char **kernelString)
+static void bitwise_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
+ String kernelName, const char **kernelString, const char* _opt = NULL)
 {
-    bitwise_run<char>(src1, src2, dst, kernelName, kernelString, (void *)NULL);
+    bitwise_run<char>(src1, src2, dst, kernelName, kernelString, (void *)NULL, _opt);
 }
-static void bitwise_run(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask, String kernelName, const char **kernelString)
+static void bitwise_run(const oclMat &src1, const oclMat &src2, oclMat &dst,
+ const oclMat &mask, String kernelName, const char **kernelString, const char* _opt = NULL)
 {
     dst.create(src1.size(), src1.type());
     CV_Assert(src1.cols == src2.cols && src2.cols == dst.cols &&
@@ -1751,12 +1751,13 @@ static void bitwise_run(const oclMat &src1, const oclMat &src2, oclMat &dst, con
     args.push_back( std::make_pair( sizeof(cl_int), (void *)&cols ));
     args.push_back( std::make_pair( sizeof(cl_int), (void *)&dst_step1 ));
 
-    openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth);
+    openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth, _opt);
 }
 
 
 template <typename WT , typename CL_WT>
-void bitwise_scalar_run(const oclMat &src1, const Scalar &src2, oclMat &dst, const oclMat &mask, String kernelName, const char **kernelString, int isMatSubScalar)
+void bitwise_scalar_run(const oclMat &src1, const Scalar &src2, oclMat &dst,
+ const oclMat &mask, String kernelName, const char **kernelString, int isMatSubScalar, const char* opt = NULL)
 {
     dst.create(src1.size(), src1.type());
 
@@ -1818,14 +1819,16 @@ void bitwise_scalar_run(const oclMat &src1, const Scalar &src2, oclMat &dst, con
         args.push_back( std::make_pair( sizeof(cl_int) , (void *)&isMatSubScalar));
     }
 
-    openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth);
+    openCLExecuteKernel(clCxt, kernelString, kernelName, globalThreads, localThreads, args, channels, depth, opt);
 }
 
 
-typedef void (*BitwiseFuncS)(const oclMat &src1, const Scalar &src2, oclMat &dst, const oclMat &mask, String kernelName, const char **kernelString, int isMatSubScalar);
+typedef void (*BitwiseFuncS)(const oclMat &src1, const Scalar &src2, oclMat &dst,
+ const oclMat &mask, String kernelName, const char **kernelString, int isMatSubScalar, const char* opt);
 
 
-static void bitwise_scalar(const oclMat &src1, const Scalar &src2, oclMat &dst, const oclMat &mask, String kernelName, const char **kernelString, int isMatSubScalar)
+static void bitwise_scalar(const oclMat &src1, const Scalar &src2, oclMat &dst,
+ const oclMat &mask, String kernelName, const char **kernelString, int isMatSubScalar, const char* opt)
 {
     static BitwiseFuncS tab[8] =
     {
@@ -1853,11 +1856,12 @@ static void bitwise_scalar(const oclMat &src1, const Scalar &src2, oclMat &dst,
     BitwiseFuncS func = tab[src1.depth()];
     if(func == 0)
         cv::error(Error::StsBadArg, "Unsupported arithmetic operation", "", __FILE__, __LINE__);
-    func(src1, src2, dst, mask, kernelName, kernelString, isMatSubScalar);
+    func(src1, src2, dst, mask, kernelName, kernelString, isMatSubScalar, opt);
 }
-static void bitwise_scalar(const oclMat &src1, const Scalar &src2, oclMat &dst, const oclMat &mask, String kernelName, const char **kernelString)
+static void bitwise_scalar(const oclMat &src1, const Scalar &src2, oclMat &dst,
+ const oclMat &mask, String kernelName, const char **kernelString, const char * opt = NULL)
 {
-    bitwise_scalar(src1, src2, dst, mask, kernelName, kernelString, 0);
+    bitwise_scalar(src1, src2, dst, mask, kernelName, kernelString, 0, opt);
 }
 
 void cv::ocl::bitwise_not(const oclMat &src, oclMat &dst)
@@ -1880,12 +1884,13 @@ void cv::ocl::bitwise_or(const oclMat &src1, const oclMat &src2, oclMat &dst, co
         std::cout << "Selected device do not support double" << std::endl;
         return;
     }
-    oclMat emptyMat;
-    String kernelName = mask.empty() ? "arithm_bitwise_or" : "arithm_bitwise_or_with_mask";
+
+    String kernelName = mask.empty() ? "arithm_bitwise_binary" : "arithm_bitwise_binary_with_mask";
+    static const char opt [] = "-D OP_BINARY=|";
     if (mask.empty())
-        bitwise_run(src1, src2, dst, kernelName, &arithm_bitwise_or);
+        bitwise_run(src1, src2, dst, kernelName, &arithm_bitwise_binary, opt);
     else
-        bitwise_run(src1, src2, dst, mask, kernelName, &arithm_bitwise_or_mask);
+        bitwise_run(src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_mask, opt);
 }
 
 
@@ -1896,11 +1901,12 @@ void cv::ocl::bitwise_or(const oclMat &src1, const Scalar &src2, oclMat &dst, co
         std::cout << "Selected device do not support double" << std::endl;
         return;
     }
-    String kernelName = mask.data ? "arithm_s_bitwise_or_with_mask" : "arithm_s_bitwise_or";
+    static const char opt [] = "-D OP_BINARY=|";
+    String kernelName = mask.data ? "arithm_s_bitwise_binary_with_mask" : "arithm_s_bitwise_binary";
     if (mask.data)
-        bitwise_scalar( src1, src2, dst, mask, kernelName, &arithm_bitwise_or_scalar_mask);
+        bitwise_scalar( src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_scalar_mask, opt);
     else
-        bitwise_scalar( src1, src2, dst, mask, kernelName, &arithm_bitwise_or_scalar);
+        bitwise_scalar( src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_scalar, opt);
 }
 
 void cv::ocl::bitwise_and(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask)
@@ -1913,12 +1919,13 @@ void cv::ocl::bitwise_and(const oclMat &src1, const oclMat &src2, oclMat &dst, c
     }
     oclMat emptyMat;
 
-    String kernelName = mask.empty() ? "arithm_bitwise_and" : "arithm_bitwise_and_with_mask";
+    String kernelName = mask.empty() ? "arithm_bitwise_binary" : "arithm_bitwise_binary_with_mask";
 
+    static const char opt [] = "-D OP_BINARY=&";
     if (mask.empty())
-        bitwise_run(src1, src2, dst, kernelName, &arithm_bitwise_and);
+        bitwise_run(src1, src2, dst, kernelName, &arithm_bitwise_binary, opt);
     else
-        bitwise_run(src1, src2, dst, mask, kernelName, &arithm_bitwise_and_mask);
+        bitwise_run(src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_mask, opt);
 }
 
 void cv::ocl::bitwise_and(const oclMat &src1, const Scalar &src2, oclMat &dst, const oclMat &mask)
@@ -1928,11 +1935,12 @@ void cv::ocl::bitwise_and(const oclMat &src1, const Scalar &src2, oclMat &dst, c
         std::cout << "Selected device do not support double" << std::endl;
         return;
     }
-    String kernelName = mask.data ? "arithm_s_bitwise_and_with_mask" : "arithm_s_bitwise_and";
+    static const char opt [] = "-D OP_BINARY=&";
+    String kernelName = mask.data ? "arithm_s_bitwise_binary_with_mask" : "arithm_s_bitwise_binary";
     if (mask.data)
-        bitwise_scalar(src1, src2, dst, mask, kernelName, &arithm_bitwise_and_scalar_mask);
+        bitwise_scalar(src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_scalar_mask, opt);
     else
-        bitwise_scalar(src1, src2, dst, mask, kernelName, &arithm_bitwise_and_scalar);
+        bitwise_scalar(src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_scalar, opt);
 }
 
 void cv::ocl::bitwise_xor(const oclMat &src1, const oclMat &src2, oclMat &dst, const oclMat &mask)
@@ -1942,14 +1950,14 @@ void cv::ocl::bitwise_xor(const oclMat &src1, const oclMat &src2, oclMat &dst, c
         std::cout << "Selected device do not support double" << std::endl;
         return;
     }
-    oclMat emptyMat;
-    String kernelName = mask.empty() ? "arithm_bitwise_xor" : "arithm_bitwise_xor_with_mask";
+    String kernelName = mask.empty() ? "arithm_bitwise_binary" : "arithm_bitwise_binary_with_mask";
 
+    static const char opt [] = "-D OP_BINARY=^";
 
     if (mask.empty())
-        bitwise_run(src1, src2, dst, kernelName, &arithm_bitwise_xor);
+        bitwise_run(src1, src2, dst, kernelName, &arithm_bitwise_binary, opt);
     else
-        bitwise_run(src1, src2, dst, mask, kernelName, &arithm_bitwise_xor_mask);
+        bitwise_run(src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_mask, opt);
 }
 
 
@@ -1961,11 +1969,12 @@ void cv::ocl::bitwise_xor(const oclMat &src1, const Scalar &src2, oclMat &dst, c
         std::cout << "Selected device do not support double" << std::endl;
         return;
     }
-    String kernelName = mask.data ? "arithm_s_bitwise_xor_with_mask" : "arithm_s_bitwise_xor";
+    String kernelName = mask.data ? "arithm_s_bitwise_binary_with_mask" : "arithm_s_bitwise_binary";
+    static const char opt [] = "-D OP_BINARY=^";
     if (mask.data)
-        bitwise_scalar( src1, src2, dst, mask, kernelName, &arithm_bitwise_xor_scalar_mask);
+        bitwise_scalar( src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_scalar_mask, opt);
     else
-        bitwise_scalar( src1, src2, dst, mask, kernelName, &arithm_bitwise_xor_scalar);
+        bitwise_scalar( src1, src2, dst, mask, kernelName, &arithm_bitwise_binary_scalar, opt);
 }
 
 oclMat cv::ocl::operator ~ (const oclMat &src)

modules/ocl/src/brute_force_matcher.cpp

@@ -844,8 +844,8 @@ void cv::ocl::BruteForceMatcher_OCL_base::knnMatch2Collection(const oclMat &quer
     if (query.empty() || trainCollection.empty())
         return;
 
-    typedef void (*caller_t)(const oclMat & query, const oclMat & trains, const oclMat & masks,
-                             const oclMat & trainIdx, const oclMat & imgIdx, const oclMat & distance);
+    // typedef void (*caller_t)(const oclMat & query, const oclMat & trains, const oclMat & masks,
+    //                          const oclMat & trainIdx, const oclMat & imgIdx, const oclMat & distance);
 
     CV_Assert(query.channels() == 1 && query.depth() < CV_64F);
 
@@ -992,7 +992,7 @@ void cv::ocl::BruteForceMatcher_OCL_base::knnMatch(const oclMat &query, std::vec
 
 // radiusMatchSingle
 void cv::ocl::BruteForceMatcher_OCL_base::radiusMatchSingle(const oclMat &query, const oclMat &train,
-        oclMat &trainIdx, oclMat &distance, oclMat &nMatches, float maxDistance, const oclMat &mask)
+        oclMat &trainIdx,   oclMat &distance, oclMat &nMatches, float maxDistance, const oclMat &mask)
 {
     if (query.empty() || train.empty())
         return;
@@ -1094,9 +1094,9 @@ void cv::ocl::BruteForceMatcher_OCL_base::radiusMatchCollection(const oclMat &qu
     if (query.empty() || empty())
         return;
 
+#if 0
     typedef void (*caller_t)(const oclMat & query, const oclMat * trains, int n, float maxDistance, const oclMat * masks,
                              const oclMat & trainIdx, const oclMat & imgIdx, const oclMat & distance, const oclMat & nMatches);
-#if 0
     static const caller_t callers[3][6] =
     {
         {

modules/ocl/src/gemm.cpp

@@ -60,7 +60,7 @@ void cv::ocl::gemm(const oclMat &src1, const oclMat &src2, double alpha,
                    const oclMat &src3, double beta, oclMat &dst, int flags)
 {
     CV_Assert(src1.cols == src2.rows &&
-              (src3.empty() || src1.rows == src3.rows && src2.cols == src3.cols));
+              (src3.empty() || (src1.rows == src3.rows && src2.cols == src3.cols)));
     CV_Assert(!(cv::GEMM_3_T & flags)); // cv::GEMM_3_T is not supported
     if(!src3.empty())
     {

modules/ocl/src/haar.cpp

@@ -20,6 +20,7 @@
 //    Jia Haipeng, jiahaipeng95@gmail.com
 //    Wu Xinglong, wxl370@126.com
 //    Wang Yao, bitwangyaoyao@gmail.com
+//    Sen Liu, swjtuls1987@126.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -144,7 +145,8 @@ typedef struct
     int imgoff;
     float factor;
 } detect_piramid_info;
-#ifdef WIN32
+
+#if defined WIN32 && !defined __MINGW__ && !defined __MINGW32__
 #define _ALIGNED_ON(_ALIGNMENT) __declspec(align(_ALIGNMENT))
 typedef _ALIGNED_ON(128) struct  GpuHidHaarFeature
 {
@@ -841,15 +843,13 @@ static void gpuSetHaarClassifierCascade( CvHaarClassifierCascade *_cascade
         } /* j */
     }
 }
+
 CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemStorage *storage, double scaleFactor,
         int minNeighbors, int flags, CvSize minSize, CvSize maxSize)
 {
     CvHaarClassifierCascade *cascade = oldCascade;
 
-    //double alltime = (double)cvGetTickCount();
-    //double t = (double)cvGetTickCount();
     const double GROUP_EPS = 0.2;
-    oclMat gtemp, gsum1, gtilted1, gsqsum1, gnormImg, gsumcanny;
     CvSeq *result_seq = 0;
     cv::Ptr<CvMemStorage> temp_storage;
 
@@ -860,7 +860,6 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
     int datasize=0;
     int totalclassifier=0;
 
-    //void *out;
     GpuHidHaarClassifierCascade *gcascade;
     GpuHidHaarStageClassifier    *stage;
     GpuHidHaarClassifier         *classifier;
@@ -869,11 +868,8 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
     int *candidate;
     cl_int status;
 
-    //    bool doCannyPruning = (flags & CV_HAAR_DO_CANNY_PRUNING) != 0;
     bool findBiggestObject = (flags & CV_HAAR_FIND_BIGGEST_OBJECT) != 0;
-    //    bool roughSearch = (flags & CV_HAAR_DO_ROUGH_SEARCH) != 0;
 
-    //double t = 0;
     if( maxSize.height == 0 || maxSize.width == 0 )
     {
         maxSize.height = gimg.rows;
@@ -895,27 +891,20 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
     if( findBiggestObject )
         flags &= ~CV_HAAR_SCALE_IMAGE;
 
-    //gtemp = oclMat( gimg.rows, gimg.cols, CV_8UC1);
-    //gsum1 =  oclMat( gimg.rows + 1, gimg.cols + 1, CV_32SC1 );
-    //gsqsum1 = oclMat( gimg.rows + 1, gimg.cols + 1, CV_32FC1 );
-
     if( !cascade->hid_cascade )
-        /*out = (void *)*/gpuCreateHidHaarClassifierCascade(cascade, &datasize, &totalclassifier);
-    if( cascade->hid_cascade->has_tilted_features )
-        gtilted1 = oclMat( gimg.rows + 1, gimg.cols + 1, CV_32SC1 );
+        gpuCreateHidHaarClassifierCascade(cascade, &datasize, &totalclassifier);
 
     result_seq = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvAvgComp), storage );
 
     if( CV_MAT_CN(gimg.type()) > 1 )
     {
+        oclMat gtemp;
         cvtColor( gimg, gtemp, COLOR_BGR2GRAY );
         gimg = gtemp;
     }
 
     if( findBiggestObject )
         flags &= ~(CV_HAAR_SCALE_IMAGE | CV_HAAR_DO_CANNY_PRUNING);
-    //t = (double)cvGetTickCount() - t;
-    //printf( "before if time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
 
     if( gimg.cols < minSize.width || gimg.rows < minSize.height )
         CV_Error(CV_StsError, "Image too small");
@@ -923,12 +912,9 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
     if( (flags & CV_HAAR_SCALE_IMAGE) )
     {
         CvSize winSize0 = cascade->orig_window_size;
-        //float scalefactor = 1.1f;
-        //float factor = 1.f;
         int totalheight = 0;
         int indexy = 0;
         CvSize sz;
-        //t = (double)cvGetTickCount();
         std::vector<CvSize> sizev;
         std::vector<float> scalev;
         for(factor = 1.f;; factor *= scaleFactor)
@@ -949,20 +935,15 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
             sizev.push_back(sz);
             scalev.push_back(factor);
         }
-        //int flag = 0;
 
         oclMat gimg1(gimg.rows, gimg.cols, CV_8UC1);
         oclMat gsum(totalheight + 4, gimg.cols + 1, CV_32SC1);
         oclMat gsqsum(totalheight + 4, gimg.cols + 1, CV_32FC1);
 
-        //cl_mem cascadebuffer;
         cl_mem stagebuffer;
-        //cl_mem classifierbuffer;
         cl_mem nodebuffer;
         cl_mem candidatebuffer;
         cl_mem scaleinfobuffer;
-        //cl_kernel kernel;
-        //kernel = openCLGetKernelFromSource(gimg.clCxt, &haarobjectdetect, "gpuRunHaarClassifierCascade");
         cv::Rect roi, roi2;
         cv::Mat imgroi, imgroisq;
         cv::ocl::oclMat resizeroi, gimgroi, gimgroisq;
@@ -970,18 +951,13 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
 
         size_t blocksize = 8;
         size_t localThreads[3] = { blocksize, blocksize , 1 };
-        size_t globalThreads[3] = { grp_per_CU *((gsum.clCxt)->computeUnits()) *localThreads[0],
+        size_t globalThreads[3] = { grp_per_CU * gsum.clCxt->computeUnits() *localThreads[0],
                                     localThreads[1], 1
                                   };
         int outputsz = 256 * globalThreads[0] / localThreads[0];
         int loopcount = sizev.size();
         detect_piramid_info *scaleinfo = (detect_piramid_info *)malloc(sizeof(detect_piramid_info) * loopcount);
 
-        //t = (double)cvGetTickCount() - t;
-        // printf( "pre time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
-        //int *it =scaleinfo;
-        // t = (double)cvGetTickCount();
-
         for( int i = 0; i < loopcount; i++ )
         {
             sz = sizev[i];
@@ -991,7 +967,6 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
             resizeroi = gimg1(roi2);
             gimgroi = gsum(roi);
             gimgroisq = gsqsum(roi);
-            //scaleinfo[i].rows = gimgroi.rows;
             int width = gimgroi.cols - 1 - cascade->orig_window_size.width;
             int height = gimgroi.rows - 1 - cascade->orig_window_size.height;
             scaleinfo[i].width_height = (width << 16) | height;
@@ -999,76 +974,40 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
 
             int grpnumperline = (width + localThreads[0] - 1) / localThreads[0];
             int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline;
-            //outputsz +=width*height;
 
             scaleinfo[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp;
             scaleinfo[i].imgoff = gimgroi.offset >> 2;
             scaleinfo[i].factor = factor;
-            //printf("rows = %d,ystep = %d,width = %d,height = %d,grpnumperline = %d,totalgrp = %d,imgoff = %d,factor = %f\n",
-            //	scaleinfo[i].rows,scaleinfo[i].ystep,scaleinfo[i].width,scaleinfo[i].height,scaleinfo[i].grpnumperline,
-            //	scaleinfo[i].totalgrp,scaleinfo[i].imgoff,scaleinfo[i].factor);
             cv::ocl::resize(gimg, resizeroi, Size(sz.width - 1, sz.height - 1), 0, 0, INTER_LINEAR);
-            //cv::imwrite("D:\\1.jpg",gimg1);
             cv::ocl::integral(resizeroi, gimgroi, gimgroisq);
-            //cv::ocl::oclMat chk(sz.height,sz.width,CV_32SC1),chksq(sz.height,sz.width,CV_32FC1);
-            //cv::ocl::integral(gimg1, chk, chksq);
-            //double r = cv::norm(chk,gimgroi,NORM_INF);
-            //if(r > std::numeric_limits<double>::epsilon())
-            //{
-            //	printf("failed");
-            //}
             indexy += sz.height;
         }
-        //int ystep = factor > 2 ? 1 : 2;
-        // t = (double)cvGetTickCount() - t;
-        //printf( "resize integral time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
-        //t = (double)cvGetTickCount();
+
         gcascade   = (GpuHidHaarClassifierCascade *)cascade->hid_cascade;
         stage      = (GpuHidHaarStageClassifier *)(gcascade + 1);
         classifier = (GpuHidHaarClassifier *)(stage + gcascade->count);
         node       = (GpuHidHaarTreeNode *)(classifier->node);
 
-        //int m,n;
-        //m = (gsum.cols - 1 - cascade->orig_window_size.width  + ystep - 1)/ystep;
-        //n = (gsum.rows - 1 - cascade->orig_window_size.height + ystep - 1)/ystep;
-        //int counter = m*n;
-
         int nodenum = (datasize - sizeof(GpuHidHaarClassifierCascade) -
                        sizeof(GpuHidHaarStageClassifier) * gcascade->count - sizeof(GpuHidHaarClassifier) * totalclassifier) / sizeof(GpuHidHaarTreeNode);
-        //if(flag == 0){
+
         candidate = (int *)malloc(4 * sizeof(int) * outputsz);
-        //memset((char*)candidate,0,4*sizeof(int)*outputsz);
-        gpuSetImagesForHaarClassifierCascade( cascade,/* &sum1, &sqsum1, _tilted,*/ 1., gsum.step / 4 );
 
-        //cascadebuffer = clCreateBuffer(gsum.clCxt->clContext,CL_MEM_READ_ONLY,sizeof(GpuHidHaarClassifierCascade),NULL,&status);
-        //openCLVerifyCall(status);
-        //openCLSafeCall(clEnqueueWriteBuffer(gsum.clCxt->clCmdQueue,cascadebuffer,1,0,sizeof(GpuHidHaarClassifierCascade),gcascade,0,NULL,NULL));
+        gpuSetImagesForHaarClassifierCascade( cascade, 1., gsum.step / 4 );
 
         stagebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(GpuHidHaarStageClassifier) * gcascade->count);
-        //openCLVerifyCall(status);
-        openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
-
-        //classifierbuffer = clCreateBuffer(gsum.clCxt->clContext,CL_MEM_READ_ONLY,sizeof(GpuHidHaarClassifier)*totalclassifier,NULL,&status);
-        //status = clEnqueueWriteBuffer(gsum.clCxt->clCmdQueue,classifierbuffer,1,0,sizeof(GpuHidHaarClassifier)*totalclassifier,classifier,0,NULL,NULL);
+        cl_command_queue qu = (cl_command_queue)gsum.clCxt->oclCommandQueue();
+        openCLSafeCall(clEnqueueWriteBuffer(qu, stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
 
         nodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, nodenum * sizeof(GpuHidHaarTreeNode));
-        //openCLVerifyCall(status);
-        openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), nodebuffer, 1, 0,
-                                            nodenum * sizeof(GpuHidHaarTreeNode),
+
+        openCLSafeCall(clEnqueueWriteBuffer(qu, nodebuffer, 1, 0, nodenum * sizeof(GpuHidHaarTreeNode),
                                             node, 0, NULL, NULL));
         candidatebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_WRITE_ONLY, 4 * sizeof(int) * outputsz);
-        //openCLVerifyCall(status);
-        scaleinfobuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount);
-        //openCLVerifyCall(status);
-        openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), scaleinfobuffer, 1, 0, sizeof(detect_piramid_info)*loopcount, scaleinfo, 0, NULL, NULL));
-        //flag  = 1;
-        //}
 
-        //t = (double)cvGetTickCount() - t;
-        //printf( "update time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
+        scaleinfobuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount);
+        openCLSafeCall(clEnqueueWriteBuffer(qu, scaleinfobuffer, 1, 0, sizeof(detect_piramid_info)*loopcount, scaleinfo, 0, NULL, NULL));
 
-        //size_t globalThreads[3] = { counter+blocksize*blocksize-counter%(blocksize*blocksize),1,1};
-        //t = (double)cvGetTickCount();
         int startstage = 0;
         int endstage = gcascade->count;
         int startnode = 0;
@@ -1086,11 +1025,6 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
         pq.s[3] = gcascade->pq3;
         float correction = gcascade->inv_window_area;
 
-        //int grpnumperline = ((m + localThreads[0] - 1) / localThreads[0]);
-        //int totalgrp = ((n + localThreads[1] - 1) / localThreads[1])*grpnumperline;
-        //   openCLVerifyKernel(gsum.clCxt, kernel, &blocksize, globalThreads, localThreads);
-        //openCLSafeCall(clSetKernelArg(kernel,argcount++,sizeof(cl_mem),(void*)&cascadebuffer));
-
         std::vector<std::pair<size_t, const void *> > args;
         args.push_back ( std::make_pair(sizeof(cl_mem) , (void *)&stagebuffer ));
         args.push_back ( std::make_pair(sizeof(cl_mem) , (void *)&scaleinfobuffer ));
@@ -1110,28 +1044,20 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
         args.push_back ( std::make_pair(sizeof(cl_float) , (void *)&correction ));
 
         openCLExecuteKernel(gsum.clCxt, &haarobjectdetect, "gpuRunHaarClassifierCascade", globalThreads, localThreads, args, -1, -1);
-        //t = (double)cvGetTickCount() - t;
-        //printf( "detection time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
-        //t = (double)cvGetTickCount();
-        //openCLSafeCall(clEnqueueReadBuffer(gsum.clCxt->impl->clCmdQueue, candidatebuffer, 1, 0, 4 * sizeof(int)*outputsz, candidate, 0, NULL, NULL));
+
         openCLReadBuffer( gsum.clCxt, candidatebuffer, candidate, 4 * sizeof(int)*outputsz );
 
         for(int i = 0; i < outputsz; i++)
             if(candidate[4 * i + 2] != 0)
-                allCandidates.push_back(Rect(candidate[4 * i], candidate[4 * i + 1], candidate[4 * i + 2], candidate[4 * i + 3]));
-        // t = (double)cvGetTickCount() - t;
-        //printf( "post time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
-        //t = (double)cvGetTickCount();
+                allCandidates.push_back(Rect(candidate[4 * i], candidate[4 * i + 1],
+                candidate[4 * i + 2], candidate[4 * i + 3]));
+
         free(scaleinfo);
         free(candidate);
-        //openCLSafeCall(clReleaseMemObject(cascadebuffer));
         openCLSafeCall(clReleaseMemObject(stagebuffer));
         openCLSafeCall(clReleaseMemObject(scaleinfobuffer));
         openCLSafeCall(clReleaseMemObject(nodebuffer));
         openCLSafeCall(clReleaseMemObject(candidatebuffer));
-        // openCLSafeCall(clReleaseKernel(kernel));
-        //t = (double)cvGetTickCount() - t;
-        //printf( "release time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
     }
     else
     {
@@ -1149,7 +1075,6 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
         classifier = (GpuHidHaarClassifier *)(stage + gcascade->count);
         node       = (GpuHidHaarTreeNode *)(classifier->node);
         cl_mem stagebuffer;
-        //cl_mem classifierbuffer;
         cl_mem nodebuffer;
         cl_mem candidatebuffer;
         cl_mem scaleinfobuffer;
@@ -1184,24 +1109,20 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
         size_t blocksize = 8;
         size_t localThreads[3] = { blocksize, blocksize , 1 };
         size_t globalThreads[3] = { grp_per_CU *gsum.clCxt->computeUnits() *localThreads[0],
-                                    localThreads[1], 1
-                                  };
+                                    localThreads[1], 1 };
         int outputsz = 256 * globalThreads[0] / localThreads[0];
         int nodenum = (datasize - sizeof(GpuHidHaarClassifierCascade) -
                        sizeof(GpuHidHaarStageClassifier) * gcascade->count - sizeof(GpuHidHaarClassifier) * totalclassifier) / sizeof(GpuHidHaarTreeNode);
         nodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY,
                                         nodenum * sizeof(GpuHidHaarTreeNode));
-        //openCLVerifyCall(status);
-        openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), nodebuffer, 1, 0,
+        cl_command_queue qu = (cl_command_queue)gsum.clCxt->oclCommandQueue();
+        openCLSafeCall(clEnqueueWriteBuffer(qu, nodebuffer, 1, 0,
                                             nodenum * sizeof(GpuHidHaarTreeNode),
                                             node, 0, NULL, NULL));
         cl_mem newnodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_WRITE,
                                loopcount * nodenum * sizeof(GpuHidHaarTreeNode));
         int startstage = 0;
         int endstage = gcascade->count;
-        //cl_kernel kernel;
-        //kernel = openCLGetKernelFromSource(gsum.clCxt, &haarobjectdetect_scaled2, "gpuRunHaarClassifierCascade_scaled2");
-        //cl_kernel kernel2 = openCLGetKernelFromSource(gimg.clCxt, &haarobjectdetect_scaled2, "gpuscaleclassifier");
         for(int i = 0; i < loopcount; i++)
         {
             sz = sizev[i];
@@ -1220,7 +1141,7 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
             int height = (gsum.rows - 1 - sz.height + ystep - 1) / ystep;
             int grpnumperline = (width + localThreads[0] - 1) / localThreads[0];
             int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline;
-            //outputsz +=width*height;
+
             scaleinfo[i].width_height = (width << 16) | height;
             scaleinfo[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp;
             scaleinfo[i].imgoff = 0;
@@ -1238,28 +1159,20 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
             size_t globalThreads2[3] = {nodenum, 1, 1};
 
             openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuscaleclassifier", globalThreads2, NULL/*localThreads2*/, args1, -1, -1);
-
-            //clEnqueueNDRangeKernel(gsum.clCxt->impl->clCmdQueue, kernel2, 1, NULL, globalThreads2, 0, 0, NULL, NULL);
-            //clFinish(gsum.clCxt->impl->clCmdQueue);
         }
-        //clReleaseKernel(kernel2);
+
         int step = gsum.step / 4;
         int startnode = 0;
         int splitstage = 3;
-        int splitnode = stage[0].count + stage[1].count + stage[2].count;
         stagebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(GpuHidHaarStageClassifier) * gcascade->count);
-        //openCLVerifyCall(status);
-        openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
+        openCLSafeCall(clEnqueueWriteBuffer(qu, stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
         candidatebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR, 4 * sizeof(int) * outputsz);
-        //openCLVerifyCall(status);
         scaleinfobuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount);
-        //openCLVerifyCall(status);
-        openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), scaleinfobuffer, 1, 0, sizeof(detect_piramid_info)*loopcount, scaleinfo, 0, NULL, NULL));
+        openCLSafeCall(clEnqueueWriteBuffer(qu, scaleinfobuffer, 1, 0, sizeof(detect_piramid_info)*loopcount, scaleinfo, 0, NULL, NULL));
         pbuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(cl_int4) * loopcount);
-        openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), pbuffer, 1, 0, sizeof(cl_int4)*loopcount, p, 0, NULL, NULL));
+        openCLSafeCall(clEnqueueWriteBuffer(qu, pbuffer, 1, 0, sizeof(cl_int4)*loopcount, p, 0, NULL, NULL));
         correctionbuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(cl_float) * loopcount);
-        openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), correctionbuffer, 1, 0, sizeof(cl_float)*loopcount, correction, 0, NULL, NULL));
-        //int argcount = 0;
+        openCLSafeCall(clEnqueueWriteBuffer(qu, correctionbuffer, 1, 0, sizeof(cl_float)*loopcount, correction, 0, NULL, NULL));
 
         std::vector<std::pair<size_t, const void *> > args;
         args.push_back ( std::make_pair(sizeof(cl_mem) , (void *)&stagebuffer ));
@@ -1268,22 +1181,21 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
         args.push_back ( std::make_pair(sizeof(cl_mem) , (void *)&gsum.data ));
         args.push_back ( std::make_pair(sizeof(cl_mem) , (void *)&gsqsum.data ));
         args.push_back ( std::make_pair(sizeof(cl_mem) , (void *)&candidatebuffer ));
+        args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&gsum.rows ));
+        args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&gsum.cols ));
         args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&step ));
         args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&loopcount ));
         args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&startstage ));
         args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&splitstage ));
         args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&endstage ));
         args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&startnode ));
-        args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&splitnode ));
         args.push_back ( std::make_pair(sizeof(cl_mem) , (void *)&pbuffer ));
         args.push_back ( std::make_pair(sizeof(cl_mem) , (void *)&correctionbuffer ));
         args.push_back ( std::make_pair(sizeof(cl_int) , (void *)&nodenum ));
 
-
         openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuRunHaarClassifierCascade_scaled2", globalThreads, localThreads, args, -1, -1);
 
-        //openCLSafeCall(clEnqueueReadBuffer(gsum.clCxt->clCmdQueue,candidatebuffer,1,0,4*sizeof(int)*outputsz,candidate,0,NULL,NULL));
-        candidate = (int *)clEnqueueMapBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), candidatebuffer, 1, CL_MAP_READ, 0, 4 * sizeof(int), 0, 0, 0, &status);
+        candidate = (int *)clEnqueueMapBuffer(qu, candidatebuffer, 1, CL_MAP_READ, 0, 4 * sizeof(int) * outputsz, 0, 0, 0, &status);
 
         for(int i = 0; i < outputsz; i++)
         {
@@ -1294,7 +1206,7 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
         free(scaleinfo);
         free(p);
         free(correction);
-        clEnqueueUnmapMemObject((cl_command_queue)gsum.clCxt->oclCommandQueue(), candidatebuffer, candidate, 0, 0, 0);
+        clEnqueueUnmapMemObject(qu, candidatebuffer, candidate, 0, 0, 0);
         openCLSafeCall(clReleaseMemObject(stagebuffer));
         openCLSafeCall(clReleaseMemObject(scaleinfobuffer));
         openCLSafeCall(clReleaseMemObject(nodebuffer));
@@ -1303,20 +1215,547 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
         openCLSafeCall(clReleaseMemObject(pbuffer));
         openCLSafeCall(clReleaseMemObject(correctionbuffer));
     }
-    //t = (double)cvGetTickCount() ;
+
     cvFree(&cascade->hid_cascade);
-    //    printf("%d\n",globalcounter);
     rectList.resize(allCandidates.size());
     if(!allCandidates.empty())
         std::copy(allCandidates.begin(), allCandidates.end(), rectList.begin());
 
-    //cout << "count = " << rectList.size()<< endl;
+    if( minNeighbors != 0 || findBiggestObject )
+        groupRectangles(rectList, rweights, std::max(minNeighbors, 1), GROUP_EPS);
+    else
+        rweights.resize(rectList.size(), 0);
+
+    if( findBiggestObject && rectList.size() )
+    {
+        CvAvgComp result_comp = {{0, 0, 0, 0}, 0};
+
+        for( size_t i = 0; i < rectList.size(); i++ )
+        {
+            cv::Rect r = rectList[i];
+            if( r.area() > cv::Rect(result_comp.rect).area() )
+            {
+                result_comp.rect = r;
+                result_comp.neighbors = rweights[i];
+            }
+        }
+        cvSeqPush( result_seq, &result_comp );
+    }
+    else
+    {
+        for( size_t i = 0; i < rectList.size(); i++ )
+        {
+            CvAvgComp c;
+            c.rect = rectList[i];
+            c.neighbors = rweights[i];
+            cvSeqPush( result_seq, &c );
+        }
+    }
+
+    return result_seq;
+}
+
+struct OclBuffers
+{
+    cl_mem stagebuffer;
+    cl_mem nodebuffer;
+    cl_mem candidatebuffer;
+    cl_mem scaleinfobuffer;
+    cl_mem pbuffer;
+    cl_mem correctionbuffer;
+    cl_mem newnodebuffer;
+};
+
+struct getRect
+{
+    Rect operator()(const CvAvgComp &e) const
+    {
+        return e.rect;
+    }
+};
+
+void cv::ocl::OclCascadeClassifierBuf::detectMultiScale(oclMat &gimg, CV_OUT std::vector<cv::Rect>& faces,
+                                                        double scaleFactor, int minNeighbors, int flags,
+                                                        Size minSize, Size maxSize)
+{
+    int blocksize = 8;
+    int grp_per_CU = 12;
+    size_t localThreads[3] = { blocksize, blocksize, 1 };
+    size_t globalThreads[3] = { grp_per_CU * Context::getContext()->computeUnits() * localThreads[0],
+        localThreads[1],
+        1 };
+    int outputsz = 256 * globalThreads[0] / localThreads[0];
+
+    Init(gimg.rows, gimg.cols, scaleFactor, flags, outputsz, localThreads, minSize, maxSize);
+
+    const double GROUP_EPS = 0.2;
+
+    cv::ConcurrentRectVector allCandidates;
+    std::vector<cv::Rect> rectList;
+    std::vector<int> rweights;
+
+    CvHaarClassifierCascade      *cascade = oldCascade;
+    GpuHidHaarClassifierCascade  *gcascade;
+    GpuHidHaarStageClassifier    *stage;
+    GpuHidHaarClassifier         *classifier;
+    GpuHidHaarTreeNode           *node;
+
+    if( CV_MAT_DEPTH(gimg.type()) != CV_8U )
+        CV_Error( CV_StsUnsupportedFormat, "Only 8-bit images are supported" );
+
+    if( CV_MAT_CN(gimg.type()) > 1 )
+    {
+        oclMat gtemp;
+        cvtColor( gimg, gtemp, CV_BGR2GRAY );
+        gimg = gtemp;
+    }
+
+    int *candidate;
+
+    if( (flags & CV_HAAR_SCALE_IMAGE) )
+    {
+        int indexy = 0;
+        CvSize sz;
+
+        cv::Rect roi, roi2;
+        cv::Mat imgroi, imgroisq;
+        cv::ocl::oclMat resizeroi, gimgroi, gimgroisq;
+
+        for( int i = 0; i < m_loopcount; i++ )
+        {
+            sz = sizev[i];
+            roi = Rect(0, indexy, sz.width, sz.height);
+            roi2 = Rect(0, 0, sz.width - 1, sz.height - 1);
+            resizeroi = gimg1(roi2);
+            gimgroi = gsum(roi);
+            gimgroisq = gsqsum(roi);
+
+            cv::ocl::resize(gimg, resizeroi, Size(sz.width - 1, sz.height - 1), 0, 0, INTER_LINEAR);
+            cv::ocl::integral(resizeroi, gimgroi, gimgroisq);
+            indexy += sz.height;
+        }
+
+        gcascade   = (GpuHidHaarClassifierCascade *)(cascade->hid_cascade);
+        stage      = (GpuHidHaarStageClassifier *)(gcascade + 1);
+        classifier = (GpuHidHaarClassifier *)(stage + gcascade->count);
+        node       = (GpuHidHaarTreeNode *)(classifier->node);
+
+        gpuSetImagesForHaarClassifierCascade( cascade, 1., gsum.step / 4 );
+
+        cl_command_queue qu = (cl_command_queue)gsum.clCxt->oclCommandQueue();
+        openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->stagebuffer, 1, 0,
+                                            sizeof(GpuHidHaarStageClassifier) * gcascade->count,
+                                            stage, 0, NULL, NULL));
+
+        openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->nodebuffer, 1, 0,
+                                            m_nodenum * sizeof(GpuHidHaarTreeNode),
+                                            node, 0, NULL, NULL));
+
+        int startstage = 0;
+        int endstage = gcascade->count;
+        int startnode = 0;
+        int pixelstep = gsum.step / 4;
+        int splitstage = 3;
+        int splitnode = stage[0].count + stage[1].count + stage[2].count;
+        cl_int4 p, pq;
+        p.s[0] = gcascade->p0;
+        p.s[1] = gcascade->p1;
+        p.s[2] = gcascade->p2;
+        p.s[3] = gcascade->p3;
+        pq.s[0] = gcascade->pq0;
+        pq.s[1] = gcascade->pq1;
+        pq.s[2] = gcascade->pq2;
+        pq.s[3] = gcascade->pq3;
+        float correction = gcascade->inv_window_area;
+
+        vector<pair<size_t, const void *> > args;
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->stagebuffer ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->scaleinfobuffer ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->nodebuffer ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsum.data ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsqsum.data ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->candidatebuffer ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&pixelstep ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&m_loopcount ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&startstage ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitstage ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&endstage ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&startnode ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitnode ));
+        args.push_back ( make_pair(sizeof(cl_int4) , (void *)&p ));
+        args.push_back ( make_pair(sizeof(cl_int4) , (void *)&pq ));
+        args.push_back ( make_pair(sizeof(cl_float) , (void *)&correction ));
+
+        openCLExecuteKernel(gsum.clCxt, &haarobjectdetect, "gpuRunHaarClassifierCascade", globalThreads, localThreads, args, -1, -1);
+
+        candidate = (int *)malloc(4 * sizeof(int) * outputsz);
+        memset(candidate, 0, 4 * sizeof(int) * outputsz);
+        openCLReadBuffer( gsum.clCxt, ((OclBuffers *)buffers)->candidatebuffer, candidate, 4 * sizeof(int)*outputsz );
+
+        for(int i = 0; i < outputsz; i++)
+            if(candidate[4 * i + 2] != 0)
+                allCandidates.push_back(Rect(candidate[4 * i], candidate[4 * i + 1],
+                candidate[4 * i + 2], candidate[4 * i + 3]));
+
+        free((void *)candidate);
+        candidate = NULL;
+    }
+    else
+    {
+        cv::ocl::integral(gimg, gsum, gsqsum);
+
+        gpuSetHaarClassifierCascade(cascade);
+
+        gcascade   = (GpuHidHaarClassifierCascade *)cascade->hid_cascade;
+        stage      = (GpuHidHaarStageClassifier *)(gcascade + 1);
+        classifier = (GpuHidHaarClassifier *)(stage + gcascade->count);
+        node       = (GpuHidHaarTreeNode *)(classifier->node);
+
+        cl_command_queue qu = (cl_command_queue)gsum.clCxt->oclCommandQueue();
+        openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->nodebuffer, 1, 0,
+                                            m_nodenum * sizeof(GpuHidHaarTreeNode),
+                                            node, 0, NULL, NULL));
+
+        cl_int4 *p = (cl_int4 *)malloc(sizeof(cl_int4) * m_loopcount);
+        float *correction = (float *)malloc(sizeof(float) * m_loopcount);
+        int startstage = 0;
+        int endstage = gcascade->count;
+        double factor;
+        for(int i = 0; i < m_loopcount; i++)
+        {
+            factor = scalev[i];
+            int equRect_x = (int)(factor * gcascade->p0 + 0.5);
+            int equRect_y = (int)(factor * gcascade->p1 + 0.5);
+            int equRect_w = (int)(factor * gcascade->p3 + 0.5);
+            int equRect_h = (int)(factor * gcascade->p2 + 0.5);
+            p[i].s[0] = equRect_x;
+            p[i].s[1] = equRect_y;
+            p[i].s[2] = equRect_x + equRect_w;
+            p[i].s[3] = equRect_y + equRect_h;
+            correction[i] = 1. / (equRect_w * equRect_h);
+            int startnodenum = m_nodenum * i;
+            float factor2 = (float)factor;
+
+            vector<pair<size_t, const void *> > args1;
+            args1.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->nodebuffer ));
+            args1.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->newnodebuffer ));
+            args1.push_back ( make_pair(sizeof(cl_float) , (void *)&factor2 ));
+            args1.push_back ( make_pair(sizeof(cl_float) , (void *)&correction[i] ));
+            args1.push_back ( make_pair(sizeof(cl_int) , (void *)&startnodenum ));
+
+            size_t globalThreads2[3] = {m_nodenum, 1, 1};
+
+            openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuscaleclassifier", globalThreads2, NULL/*localThreads2*/, args1, -1, -1);
+        }
+
+        int step = gsum.step / 4;
+        int startnode = 0;
+        int splitstage = 3;
+        openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
+        openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->pbuffer, 1, 0, sizeof(cl_int4)*m_loopcount, p, 0, NULL, NULL));
+        openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->correctionbuffer, 1, 0, sizeof(cl_float)*m_loopcount, correction, 0, NULL, NULL));
+
+        vector<pair<size_t, const void *> > args;
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->stagebuffer ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->scaleinfobuffer ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->newnodebuffer ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsum.data ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsqsum.data ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->candidatebuffer ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&gsum.rows ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&gsum.cols ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&step ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&m_loopcount ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&startstage ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitstage ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&endstage ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&startnode ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->pbuffer ));
+        args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->correctionbuffer ));
+        args.push_back ( make_pair(sizeof(cl_int) , (void *)&m_nodenum ));
+
+        openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuRunHaarClassifierCascade_scaled2", globalThreads, localThreads, args, -1, -1);
+
+        candidate = (int *)clEnqueueMapBuffer(qu, ((OclBuffers *)buffers)->candidatebuffer, 1, CL_MAP_READ, 0, 4 * sizeof(int) * outputsz, 0, 0, 0, NULL);
+
+        for(int i = 0; i < outputsz; i++)
+        {
+            if(candidate[4 * i + 2] != 0)
+                allCandidates.push_back(Rect(candidate[4 * i], candidate[4 * i + 1],
+                candidate[4 * i + 2], candidate[4 * i + 3]));
+        }
+
+        free(p);
+        free(correction);
+        clEnqueueUnmapMemObject(qu, ((OclBuffers *)buffers)->candidatebuffer, candidate, 0, 0, 0);
+    }
+
+    rectList.resize(allCandidates.size());
+    if(!allCandidates.empty())
+        std::copy(allCandidates.begin(), allCandidates.end(), rectList.begin());
 
     if( minNeighbors != 0 || findBiggestObject )
         groupRectangles(rectList, rweights, std::max(minNeighbors, 1), GROUP_EPS);
     else
         rweights.resize(rectList.size(), 0);
 
+    GenResult(faces, rectList, rweights);
+}
+
+void cv::ocl::OclCascadeClassifierBuf::Init(const int rows, const int cols,
+    double scaleFactor, int flags,
+    const int outputsz, const size_t localThreads[],
+    CvSize minSize, CvSize maxSize)
+{
+    CvHaarClassifierCascade      *cascade = oldCascade;
+
+    if( !CV_IS_HAAR_CLASSIFIER(cascade) )
+        CV_Error( !cascade ? CV_StsNullPtr : CV_StsBadArg, "Invalid classifier cascade" );
+
+    if( scaleFactor <= 1 )
+        CV_Error( CV_StsOutOfRange, "scale factor must be > 1" );
+
+    if( cols < minSize.width || rows < minSize.height )
+        CV_Error(CV_StsError, "Image too small");
+
+    int datasize=0;
+    int totalclassifier=0;
+
+    if( !cascade->hid_cascade )
+        gpuCreateHidHaarClassifierCascade(cascade, &datasize, &totalclassifier);
+
+    if( maxSize.height == 0 || maxSize.width == 0 )
+    {
+        maxSize.height = rows;
+        maxSize.width = cols;
+    }
+
+    findBiggestObject = (flags & CV_HAAR_FIND_BIGGEST_OBJECT) != 0;
+    if( findBiggestObject )
+        flags &= ~(CV_HAAR_SCALE_IMAGE | CV_HAAR_DO_CANNY_PRUNING);
+
+    CreateBaseBufs(datasize, totalclassifier, flags, outputsz);
+    CreateFactorRelatedBufs(rows, cols, flags, scaleFactor, localThreads, minSize, maxSize);
+
+    m_scaleFactor = scaleFactor;
+    m_rows = rows;
+    m_cols = cols;
+    m_flags = flags;
+    m_minSize = minSize;
+    m_maxSize = maxSize;
+
+    initialized = true;
+}
+
+void cv::ocl::OclCascadeClassifierBuf::CreateBaseBufs(const int datasize, const int totalclassifier,
+                                                      const int flags, const int outputsz)
+{
+    if (!initialized)
+    {
+        buffers = malloc(sizeof(OclBuffers));
+
+        size_t tempSize =
+            sizeof(GpuHidHaarStageClassifier) * ((GpuHidHaarClassifierCascade *)oldCascade->hid_cascade)->count;
+        m_nodenum = (datasize - sizeof(GpuHidHaarClassifierCascade) - tempSize - sizeof(GpuHidHaarClassifier) * totalclassifier)
+            / sizeof(GpuHidHaarTreeNode);
+
+        ((OclBuffers *)buffers)->stagebuffer     = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY,  tempSize);
+        ((OclBuffers *)buffers)->nodebuffer      = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY,  m_nodenum * sizeof(GpuHidHaarTreeNode));
+    }
+
+    if (initialized
+        && ((m_flags & CV_HAAR_SCALE_IMAGE) ^ (flags & CV_HAAR_SCALE_IMAGE)))
+    {
+        openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->candidatebuffer));
+    }
+
+    if (flags & CV_HAAR_SCALE_IMAGE)
+    {
+        ((OclBuffers *)buffers)->candidatebuffer = openCLCreateBuffer(cv::ocl::Context::getContext(),
+                                                        CL_MEM_WRITE_ONLY,
+                                                        4 * sizeof(int) * outputsz);
+    }
+    else
+    {
+        ((OclBuffers *)buffers)->candidatebuffer = openCLCreateBuffer(cv::ocl::Context::getContext(),
+                                                        CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR,
+                                                        4 * sizeof(int) * outputsz);
+    }
+}
+
+void cv::ocl::OclCascadeClassifierBuf::CreateFactorRelatedBufs(
+    const int rows, const int cols, const int flags,
+    const double scaleFactor, const size_t localThreads[],
+    CvSize minSize, CvSize maxSize)
+{
+    if (initialized)
+    {
+        if ((m_flags & CV_HAAR_SCALE_IMAGE) && !(flags & CV_HAAR_SCALE_IMAGE))
+        {
+            gimg1.release();
+            gsum.release();
+            gsqsum.release();
+        }
+        else if (!(m_flags & CV_HAAR_SCALE_IMAGE) && (flags & CV_HAAR_SCALE_IMAGE))
+        {
+            openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->newnodebuffer));
+            openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->correctionbuffer));
+            openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->pbuffer));
+        }
+        else if ((m_flags & CV_HAAR_SCALE_IMAGE) && (flags & CV_HAAR_SCALE_IMAGE))
+        {
+            if (fabs(m_scaleFactor - scaleFactor) < 1e-6
+                && (rows == m_rows && cols == m_cols)
+                && (minSize.width == m_minSize.width)
+                && (minSize.height == m_minSize.height)
+                && (maxSize.width == m_maxSize.width)
+                && (maxSize.height == m_maxSize.height))
+            {
+                return;
+            }
+        }
+        else
+        {
+            if (fabs(m_scaleFactor - scaleFactor) < 1e-6
+                && (rows == m_rows && cols == m_cols)
+                && (minSize.width == m_minSize.width)
+                && (minSize.height == m_minSize.height)
+                && (maxSize.width == m_maxSize.width)
+                && (maxSize.height == m_maxSize.height))
+            {
+                return;
+            }
+            else
+            {
+                openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->newnodebuffer));
+                openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->correctionbuffer));
+                openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->pbuffer));
+            }
+        }
+    }
+
+    int loopcount;
+    int indexy = 0;
+    int totalheight = 0;
+    double factor;
+    Rect roi;
+    CvSize sz;
+    CvSize winSize0 = oldCascade->orig_window_size;
+    detect_piramid_info *scaleinfo;
+    if (flags & CV_HAAR_SCALE_IMAGE)
+    {
+        for(factor = 1.f;; factor *= scaleFactor)
+        {
+            CvSize winSize = { cvRound(winSize0.width * factor), cvRound(winSize0.height * factor) };
+            sz.width     = cvRound( cols / factor ) + 1;
+            sz.height    = cvRound( rows / factor ) + 1;
+            CvSize sz1     = { sz.width - winSize0.width - 1,      sz.height - winSize0.height - 1 };
+
+            if( sz1.width <= 0 || sz1.height <= 0 )
+                break;
+            if( winSize.width > maxSize.width || winSize.height > maxSize.height )
+                break;
+            if( winSize.width < minSize.width || winSize.height < minSize.height )
+                continue;
+
+            totalheight += sz.height;
+            sizev.push_back(sz);
+            scalev.push_back(static_cast<float>(factor));
+        }
+
+        loopcount = sizev.size();
+        gimg1.create(rows, cols, CV_8UC1);
+        gsum.create(totalheight + 4, cols + 1, CV_32SC1);
+        gsqsum.create(totalheight + 4, cols + 1, CV_32FC1);
+
+        scaleinfo = (detect_piramid_info *)malloc(sizeof(detect_piramid_info) * loopcount);
+        for( int i = 0; i < loopcount; i++ )
+        {
+            sz = sizev[i];
+            roi = Rect(0, indexy, sz.width, sz.height);
+            int width = sz.width - 1 - oldCascade->orig_window_size.width;
+            int height = sz.height - 1 - oldCascade->orig_window_size.height;
+            int grpnumperline = (width + localThreads[0] - 1) / localThreads[0];
+            int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline;
+
+            ((detect_piramid_info *)scaleinfo)[i].width_height = (width << 16) | height;
+            ((detect_piramid_info *)scaleinfo)[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp;
+            ((detect_piramid_info *)scaleinfo)[i].imgoff = gsum(roi).offset >> 2;
+            ((detect_piramid_info *)scaleinfo)[i].factor = scalev[i];
+
+            indexy += sz.height;
+        }
+    }
+    else
+    {
+        for(factor = 1;
+            cvRound(factor * winSize0.width) < cols - 10 && cvRound(factor * winSize0.height) < rows - 10;
+            factor *= scaleFactor)
+        {
+            CvSize winSize = { cvRound( winSize0.width * factor ), cvRound( winSize0.height * factor ) };
+            if( winSize.width < minSize.width || winSize.height < minSize.height )
+            {
+                continue;
+            }
+            sizev.push_back(winSize);
+            scalev.push_back(factor);
+        }
+
+        loopcount = scalev.size();
+        if(loopcount == 0)
+        {
+            loopcount = 1;
+            sizev.push_back(minSize);
+            scalev.push_back( min(cvRound(minSize.width / winSize0.width), cvRound(minSize.height / winSize0.height)) );
+        }
+
+        ((OclBuffers *)buffers)->pbuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY,
+            sizeof(cl_int4) * loopcount);
+        ((OclBuffers *)buffers)->correctionbuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY,
+            sizeof(cl_float) * loopcount);
+        ((OclBuffers *)buffers)->newnodebuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_WRITE,
+            loopcount * m_nodenum * sizeof(GpuHidHaarTreeNode));
+
+        scaleinfo = (detect_piramid_info *)malloc(sizeof(detect_piramid_info) * loopcount);
+        for( int i = 0; i < loopcount; i++ )
+        {
+            sz = sizev[i];
+            factor = scalev[i];
+            int ystep = cvRound(std::max(2., factor));
+            int width = (cols - 1 - sz.width  + ystep - 1) / ystep;
+            int height = (rows - 1 - sz.height + ystep - 1) / ystep;
+            int grpnumperline = (width + localThreads[0] - 1) / localThreads[0];
+            int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline;
+
+            ((detect_piramid_info *)scaleinfo)[i].width_height = (width << 16) | height;
+            ((detect_piramid_info *)scaleinfo)[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp;
+            ((detect_piramid_info *)scaleinfo)[i].imgoff = 0;
+            ((detect_piramid_info *)scaleinfo)[i].factor = factor;
+        }
+    }
+
+    if (loopcount != m_loopcount)
+    {
+        if (initialized)
+        {
+            openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->scaleinfobuffer));
+        }
+        ((OclBuffers *)buffers)->scaleinfobuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount);
+    }
+
+    openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)cv::ocl::Context::getContext()->oclCommandQueue(), ((OclBuffers *)buffers)->scaleinfobuffer, 1, 0,
+        sizeof(detect_piramid_info)*loopcount,
+        scaleinfo, 0, NULL, NULL));
+    free(scaleinfo);
+
+    m_loopcount = loopcount;
+}
+
+void cv::ocl::OclCascadeClassifierBuf::GenResult(CV_OUT std::vector<cv::Rect>& faces,
+                                                 const std::vector<cv::Rect> &rectList,
+                                                 const std::vector<int> &rweights)
+{
+    CvSeq *result_seq = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvAvgComp), cvCreateMemStorage(0) );
 
     if( findBiggestObject && rectList.size() )
     {
@@ -1343,13 +1782,34 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
             cvSeqPush( result_seq, &c );
         }
     }
-    //t = (double)cvGetTickCount() - t;
-    //printf( "get face time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
-    //alltime = (double)cvGetTickCount() - alltime;
-    //printf( "all time = %g ms\n", alltime/((double)cvGetTickFrequency()*1000.) );
-    return result_seq;
+
+    vector<CvAvgComp> vecAvgComp;
+    Seq<CvAvgComp>(result_seq).copyTo(vecAvgComp);
+    faces.resize(vecAvgComp.size());
+    std::transform(vecAvgComp.begin(), vecAvgComp.end(), faces.begin(), getRect());
 }
 
+void cv::ocl::OclCascadeClassifierBuf::release()
+{
+    openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->stagebuffer));
+    openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->scaleinfobuffer));
+    openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->nodebuffer));
+    openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->candidatebuffer));
+
+    if( (m_flags & CV_HAAR_SCALE_IMAGE) )
+    {
+        cvFree(&oldCascade->hid_cascade);
+    }
+    else
+    {
+        openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->newnodebuffer));
+        openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->correctionbuffer));
+        openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->pbuffer));
+    }
+
+    free(buffers);
+    buffers = NULL;
+}
 
 #ifndef _MAX_PATH
 #define _MAX_PATH 1024

modules/ocl/src/imgproc.cpp

@@ -1012,10 +1012,8 @@ namespace cv
             warpPerspective_gpu(src, dst, coeffs, interpolation);
         }
 
-
         ////////////////////////////////////////////////////////////////////////
         // integral
-
         void integral(const oclMat &src, oclMat &sum, oclMat &sqsum)
         {
             CV_Assert(src.type() == CV_8UC1);
@@ -1029,13 +1027,24 @@ namespace cv
             int vcols = (pre_invalid + src.cols + vlen - 1) / vlen;
 
             oclMat t_sum , t_sqsum;
-            t_sum.create(src.cols, src.rows, CV_32SC1);
-            t_sqsum.create(src.cols, src.rows, CV_32FC1);
-
             int w = src.cols + 1, h = src.rows + 1;
-            sum.create(h, w, CV_32SC1);
+            int depth;
+            if( src.cols * src.rows <= 2901 * 2901 ) //2901 is the maximum size for int when all values are 255
+            {
+                t_sum.create(src.cols, src.rows, CV_32SC1);
+                sum.create(h, w, CV_32SC1);
+            }
+            else
+            {
+                 //Use float to prevent overflow
+                t_sum.create(src.cols, src.rows, CV_32FC1);
+                sum.create(h, w, CV_32FC1);
+            }
+            t_sqsum.create(src.cols, src.rows, CV_32FC1);
             sqsum.create(h, w, CV_32FC1);
-            int sum_offset = sum.offset / vlen, sqsum_offset = sqsum.offset / vlen;
+            depth = sum.depth();
+            int sum_offset = sum.offset / vlen;
+            int sqsum_offset = sqsum.offset / vlen;
 
             std::vector<std::pair<size_t , const void *> > args;
             args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&src.data ));
@@ -1048,7 +1057,7 @@ namespace cv
             args.push_back( std::make_pair( sizeof(cl_int) , (void *)&src.step ));
             args.push_back( std::make_pair( sizeof(cl_int) , (void *)&t_sum.step));
             size_t gt[3] = {((vcols + 1) / 2) * 256, 1, 1}, lt[3] = {256, 1, 1};
-            openCLExecuteKernel(src.clCxt, &imgproc_integral, "integral_cols", gt, lt, args, -1, -1);
+            openCLExecuteKernel(src.clCxt, &imgproc_integral, "integral_cols", gt, lt, args, -1, depth);
             args.clear();
             args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&t_sum.data ));
             args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&t_sqsum.data ));
@@ -1062,9 +1071,9 @@ namespace cv
             args.push_back( std::make_pair( sizeof(cl_int) , (void *)&sum_offset));
             args.push_back( std::make_pair( sizeof(cl_int) , (void *)&sqsum_offset));
             size_t gt2[3] = {t_sum.cols  * 32, 1, 1}, lt2[3] = {256, 1, 1};
-            openCLExecuteKernel(src.clCxt, &imgproc_integral, "integral_rows", gt2, lt2, args, -1, -1);
-            //std::cout << "tested" << std::endl;
+            openCLExecuteKernel(src.clCxt, &imgproc_integral, "integral_rows", gt2, lt2, args, -1, depth);
         }
+
         void integral(const oclMat &src, oclMat &sum)
         {
             CV_Assert(src.type() == CV_8UC1);
@@ -1074,10 +1083,18 @@ namespace cv
             int vcols = (pre_invalid + src.cols + vlen - 1) / vlen;
 
             oclMat t_sum;
-            t_sum.create(src.cols, src.rows, CV_32SC1);
-
             int w = src.cols + 1, h = src.rows + 1;
-            sum.create(h, w, CV_32SC1);
+            int depth;
+            if(src.cols * src.rows <= 2901 * 2901)
+            {
+                t_sum.create(src.cols, src.rows, CV_32SC1);
+                sum.create(h, w, CV_32SC1);
+            }else
+            {
+                 t_sum.create(src.cols, src.rows, CV_32FC1);
+                 sum.create(h, w, CV_32FC1);
+            }
+            depth = sum.depth();
             int sum_offset = sum.offset / vlen;
 
             std::vector<std::pair<size_t , const void *> > args;
@@ -1090,7 +1107,7 @@ namespace cv
             args.push_back( std::make_pair( sizeof(cl_int) , (void *)&src.step ));
             args.push_back( std::make_pair( sizeof(cl_int) , (void *)&t_sum.step));
             size_t gt[3] = {((vcols + 1) / 2) * 256, 1, 1}, lt[3] = {256, 1, 1};
-            openCLExecuteKernel(src.clCxt, &imgproc_integral_sum, "integral_sum_cols", gt, lt, args, -1, -1);
+            openCLExecuteKernel(src.clCxt, &imgproc_integral_sum, "integral_sum_cols", gt, lt, args, -1, depth);
             args.clear();
             args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&t_sum.data ));
             args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&sum.data ));
@@ -1100,7 +1117,7 @@ namespace cv
             args.push_back( std::make_pair( sizeof(cl_int) , (void *)&sum.step));
             args.push_back( std::make_pair( sizeof(cl_int) , (void *)&sum_offset));
             size_t gt2[3] = {t_sum.cols  * 32, 1, 1}, lt2[3] = {256, 1, 1};
-            openCLExecuteKernel(src.clCxt, &imgproc_integral_sum, "integral_sum_rows", gt2, lt2, args, -1, -1);
+            openCLExecuteKernel(src.clCxt, &imgproc_integral_sum, "integral_sum_rows", gt2, lt2, args, -1, depth);
             //std::cout << "tested" << std::endl;
         }
 

modules/ocl/src/initialization.cpp

@@ -128,6 +128,8 @@ namespace cv
             std::vector<cl_device_id> devices;
             std::vector<String> devName;
             String platName;
+            String clVersion;
+
             cl_context oclcontext;
             cl_command_queue clCmdQueue;
             int devnum;
@@ -260,7 +262,7 @@ namespace cv
 
         int setDevMemType(DevMemRW rw_type, DevMemType mem_type)
         {
-            if( (mem_type == DEVICE_MEM_PM && 
+            if( (mem_type == DEVICE_MEM_PM &&
                  Context::getContext()->impl->unified_memory == 0) )
                 return -1;
             gDeviceMemRW = rw_type;
@@ -303,6 +305,7 @@ namespace cv
             const static int max_name_length = 256;
             char deviceName[max_name_length];
             char plfmName[max_name_length];
+            char clVersion[256];
             for (unsigned i = 0; i < numPlatforms; ++i)
             {
 
@@ -322,6 +325,8 @@ namespace cv
                     ocltmpinfo.PlatformName = String(plfmName);
                     ocltmpinfo.impl->platName = String(plfmName);
                     ocltmpinfo.impl->oclplatform = platforms[i];
+                    openCLSafeCall(clGetPlatformInfo(platforms[i], CL_PLATFORM_VERSION, sizeof(clVersion), clVersion, NULL));
+                    ocltmpinfo.impl->clVersion = clVersion;
                     for(unsigned j = 0; j < numsdev; ++j)
                     {
                         ocltmpinfo.impl->devices.push_back(devices[j]);
@@ -424,13 +429,13 @@ namespace cv
         }
 
         void openCLMallocPitchEx(Context *clCxt, void **dev_ptr, size_t *pitch,
-                                 size_t widthInBytes, size_t height, 
+                                 size_t widthInBytes, size_t height,
                                  DevMemRW rw_type, DevMemType mem_type, void* hptr)
         {
             cl_int status;
             if(hptr && (mem_type==DEVICE_MEM_UHP || mem_type==DEVICE_MEM_CHP))
-                *dev_ptr = clCreateBuffer(clCxt->impl->oclcontext, 
-                                          gDevMemRWValueMap[rw_type]|gDevMemTypeValueMap[mem_type], 
+                *dev_ptr = clCreateBuffer(clCxt->impl->oclcontext,
+                                          gDevMemRWValueMap[rw_type]|gDevMemTypeValueMap[mem_type],
                                           widthInBytes * height, hptr, &status);
             else
                 *dev_ptr = clCreateBuffer(clCxt->impl->oclcontext, gDevMemRWValueMap[rw_type]|gDevMemTypeValueMap[mem_type],
@@ -985,6 +990,8 @@ namespace cv
                 return impl->double_support == 1;
             case CL_UNIFIED_MEM:
                 return impl->unified_memory == 1;
+            case CL_VER_1_2:
+                return impl->clVersion.find("OpenCL 1.2") != String::npos;
             default:
                 return false;
             }

modules/ocl/src/matrix_operations.cpp

@@ -196,7 +196,7 @@ void cv::ocl::oclMat::upload(const Mat &m)
         // try to use host ptr
         createEx(wholeSize, m.type(), gDeviceMemRW, gDeviceMemType, m.datastart);
         if(gDeviceMemType!=DEVICE_MEM_UHP && gDeviceMemType!=DEVICE_MEM_CHP)
-            openCLMemcpy2D(clCxt, data, step, m.datastart, m.step, 
+            openCLMemcpy2D(clCxt, data, step, m.datastart, m.step,
                            wholeSize.width * elemSize(), wholeSize.height, clMemcpyHostToDevice);
     }
 
@@ -571,11 +571,16 @@ static void set_to_withoutmask_run(const oclMat &dst, const Scalar &scalar, Stri
         CV_Error(Error::StsUnsupportedFormat, "unknown depth");
     }
 #ifdef CL_VERSION_1_2
-    if(dst.offset == 0 && dst.cols == dst.wholecols)
+    //this enables backwards portability to
+    //run on OpenCL 1.1 platform if library binaries are compiled with OpenCL 1.2 support
+    if(Context::getContext()->supportsFeature(Context::CL_VER_1_2) &&
+        dst.offset == 0 && dst.cols == dst.wholecols)
     {
-        clEnqueueFillBuffer((cl_command_queue)dst.clCxt->oclCommandQueue(), (cl_mem)dst.data, args[0].second, args[0].first, 0, dst.step * dst.rows, 0, NULL, NULL);
+        clEnqueueFillBuffer((cl_command_queue)dst.clCxt->oclCommandQueue(),
+            (cl_mem)dst.data, args[0].second, args[0].first, 0, dst.step * dst.rows, 0, NULL, NULL);
     }
     else
+#endif
     {
         args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&dst.data ));
         args.push_back( std::make_pair( sizeof(cl_int) , (void *)&dst.cols ));
@@ -583,17 +588,8 @@ static void set_to_withoutmask_run(const oclMat &dst, const Scalar &scalar, Stri
         args.push_back( std::make_pair( sizeof(cl_int) , (void *)&step_in_pixel ));
         args.push_back( std::make_pair( sizeof(cl_int) , (void *)&offset_in_pixel));
         openCLExecuteKernel(dst.clCxt , &operator_setTo, kernelName, globalThreads,
-                            localThreads, args, -1, -1, compile_option);
+            localThreads, args, -1, -1, compile_option);
     }
-#else
-    args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&dst.data ));
-    args.push_back( std::make_pair( sizeof(cl_int) , (void *)&dst.cols ));
-    args.push_back( std::make_pair( sizeof(cl_int) , (void *)&dst.rows ));
-    args.push_back( std::make_pair( sizeof(cl_int) , (void *)&step_in_pixel ));
-    args.push_back( std::make_pair( sizeof(cl_int) , (void *)&offset_in_pixel));
-    openCLExecuteKernel(dst.clCxt , &operator_setTo, kernelName, globalThreads,
-                        localThreads, args, -1, -1, compile_option);
-#endif
 }
 
 static void set_to_withmask_run(const oclMat &dst, const Scalar &scalar, const oclMat &mask, String kernelName)
@@ -887,7 +883,7 @@ oclMat cv::ocl::oclMat::reshape(int new_cn, int new_rows) const
 
 }
 
-void cv::ocl::oclMat::createEx(Size size, int type, 
+void cv::ocl::oclMat::createEx(Size size, int type,
                                DevMemRW rw_type, DevMemType mem_type, void* hptr)
 {
     createEx(size.height, size.width, type, rw_type, mem_type, hptr);
@@ -898,7 +894,7 @@ void cv::ocl::oclMat::create(int _rows, int _cols, int _type)
     createEx(_rows, _cols, _type, gDeviceMemRW, gDeviceMemType);
 }
 
-void cv::ocl::oclMat::createEx(int _rows, int _cols, int _type, 
+void cv::ocl::oclMat::createEx(int _rows, int _cols, int _type,
                                DevMemRW rw_type, DevMemType mem_type, void* hptr)
 {
     clCxt = Context::getContext();
@@ -919,7 +915,7 @@ void cv::ocl::oclMat::createEx(int _rows, int _cols, int _type,
         size_t esz = elemSize();
 
         void *dev_ptr;
-        openCLMallocPitchEx(clCxt, &dev_ptr, &step, GPU_MATRIX_MALLOC_STEP(esz * cols), 
+        openCLMallocPitchEx(clCxt, &dev_ptr, &step, GPU_MATRIX_MALLOC_STEP(esz * cols),
                             rows, rw_type, mem_type, hptr);
 
         if (esz * cols == step)

modules/ocl/src/mcwutil.cpp

@@ -43,11 +43,10 @@
 //
 //M*/
 
+#define CL_USE_DEPRECATED_OPENCL_1_1_APIS
 #include "precomp.hpp"
 
-#ifndef CL_VERSION_1_2
-#define CL_VERSION_1_2 0
-#endif
+using namespace std;
 
 namespace cv
 {
@@ -160,30 +159,44 @@ namespace cv
                 CV_Error(-1, "Image forma is not supported");
                 break;
             }
-#if CL_VERSION_1_2
-            cl_image_desc desc;
-            desc.image_type       = CL_MEM_OBJECT_IMAGE2D;
-            desc.image_width      = mat.cols;
-            desc.image_height     = mat.rows;
-            desc.image_depth      = 0;
-            desc.image_array_size = 1;
-            desc.image_row_pitch  = 0;
-            desc.image_slice_pitch = 0;
-            desc.buffer           = NULL;
-            desc.num_mip_levels   = 0;
-            desc.num_samples      = 0;
-            texture = clCreateImage((cl_context)mat.clCxt->oclContext(), CL_MEM_READ_WRITE, &format, &desc, NULL, &err);
-#else
-            texture = clCreateImage2D(
-                (cl_context)mat.clCxt->oclContext(),
-                CL_MEM_READ_WRITE,
-                &format,
-                mat.cols,
-                mat.rows,
-                0,
-                NULL,
-                &err);
+#ifdef CL_VERSION_1_2
+            //this enables backwards portability to
+            //run on OpenCL 1.1 platform if library binaries are compiled with OpenCL 1.2 support
+            if(Context::getContext()->supportsFeature(Context::CL_VER_1_2))
+            {
+                cl_image_desc desc;
+                desc.image_type       = CL_MEM_OBJECT_IMAGE2D;
+                desc.image_width      = mat.cols;
+                desc.image_height     = mat.rows;
+                desc.image_depth      = 0;
+                desc.image_array_size = 1;
+                desc.image_row_pitch  = 0;
+                desc.image_slice_pitch = 0;
+                desc.buffer           = NULL;
+                desc.num_mip_levels   = 0;
+                desc.num_samples      = 0;
+                texture = clCreateImage((cl_context)mat.clCxt->oclContext(), CL_MEM_READ_WRITE, &format, &desc, NULL, &err);
+            }
+            else
+#endif
+            {
+#ifdef __GNUC__
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
+#endif
+                texture = clCreateImage2D(
+                    (cl_context)mat.clCxt->oclContext(),
+                    CL_MEM_READ_WRITE,
+                    &format,
+                    mat.cols,
+                    mat.rows,
+                    0,
+                    NULL,
+                    &err);
+#ifdef __GNUC__
+#pragma GCC diagnostic pop
 #endif
+            }
             size_t origin[] = { 0, 0, 0 };
             size_t region[] = { mat.cols, mat.rows, 1 };
 
@@ -196,7 +209,7 @@ namespace cv
                 clEnqueueCopyBufferRect((cl_command_queue)mat.clCxt->oclCommandQueue(), (cl_mem)mat.data, devData, origin, origin,
                     regin, mat.step, 0, mat.cols * mat.elemSize(), 0, 0, NULL, NULL);
                 clFlush((cl_command_queue)mat.clCxt->oclCommandQueue());
-           }
+            }
             else
             {
                 devData = (cl_mem)mat.data;
@@ -212,7 +225,6 @@ namespace cv
             openCLSafeCall(err);
             return texture;
         }
-
         void releaseTexture(cl_mem& texture)
         {
             openCLFree(texture);

modules/ocl/src/moments.cpp

@@ -330,7 +330,7 @@ static void ocl_cvMoments( const void* array, CvMoments* mom, int binary )
     mom->m12 = dstsum[8];
     mom->m03 = dstsum[9];
     delete [] dstsum;
-
+    openCLSafeCall(clReleaseMemObject(sum));
     icvCompleteMomentState( mom );
 }
 

modules/ocl/src/opencl/arithm_bitwise_and_scalar.cl

@@ -1,966 +0,0 @@
-////////////////////////////////////////////////////////////////////////////////////////
-//
-//  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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
-// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
-// Third party copyrights are property of their respective owners.
-//
-// @Authors
-//    Jiang Liyuan, jlyuan001.good@163.com
-//
-// 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 oclMaterials 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.
-//
-//
-#if defined (DOUBLE_SUPPORT)
-#ifdef cl_khr_fp64
-#pragma OPENCL EXTENSION cl_khr_fp64:enable
-#elif defined (cl_amd_fp64)
-#pragma OPENCL EXTENSION cl_amd_fp64:enable
-#endif
-#endif
-
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////BITWISE_AND////////////////////////////////////////////////////
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-/**************************************and with scalar without mask**************************************/
-__kernel void arithm_s_bitwise_and_C1_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (dst_offset & 3)
-        int src1_index = mad24(y, src1_step, x + src1_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + x & (int)0xfffffffc);
-
-        uchar4 src1_data = vload4(0, src1 + src1_index);
-        uchar4 src2_data = (uchar4)(src2.x, src2.x, src2.x, src2.x);
-
-        uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data & src2_data;
-
-        data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
-        data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
-        data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : data.z;
-        data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : data.w;
-
-        *((__global uchar4 *)(dst + dst_index)) = data;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_and_C1_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (dst_offset & 3)
-        int src1_index = mad24(y, src1_step, x + src1_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + x & (int)0xfffffffc);
-
-        char4 src1_data = vload4(0, src1 + src1_index);
-        char4 src2_data = (char4)(src2.x, src2.x, src2.x, src2.x);
-
-        char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data & src2_data;
-
-        data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
-        data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
-        data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : data.z;
-        data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : data.w;
-
-        *((__global char4 *)(dst + dst_index)) = data;
-    }
-}
-
-__kernel void arithm_s_bitwise_and_C1_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 1)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffffc);
-
-        ushort2 src1_data = vload2(0, (__global ushort *)((__global char *)src1 + src1_index));
-        ushort2 src2_data = (ushort2)(src2.x, src2.x);
-
-        ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index));
-        ushort2 tmp_data = src1_data & src2_data;
-
-        data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x;
-        data.y = (dst_index + 2 <  dst_end  ) ? tmp_data.y : data.y;
-
-        *((__global ushort2 *)((__global uchar *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C1_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 1)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffffc);
-
-        short2 src1_data = vload2(0, (__global short *)((__global char *)src1 + src1_index));
-        short2 src2_data = (short2)(src2.x, src2.x);
-        short2 data = *((__global short2 *)((__global uchar *)dst + dst_index));
-
-        short2 tmp_data = src1_data & src2_data;
-
-        data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x;
-        data.y = (dst_index + 2 <  dst_end  ) ? tmp_data.y : data.y;
-
-        *((__global short2 *)((__global uchar *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C1_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        int src_data1 = *((__global int *)((__global char *)src1 + src1_index));
-        int src_data2 = src2.x;
-
-        int data = src_data1 & src_data2;
-
-        *((__global int *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C1_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        char4 src1_data = *((__global char4 *)((__global char *)src1 + src1_index));
-        char4 src2_data = (char4)(src2.s0, src2.s1, src2.s2, src2.s3);
-
-        char4 data  = *((__global char4 *)((__global char *)dst  + dst_index));
-        char4 tmp_data = src1_data & src2_data;
-
-        data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
-        data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
-        data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : data.z;
-        data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : data.w;
-
-        *((__global char4 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_and_C1_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        short4 src1_data = *((__global short4 *)((__global char *)src1 + src1_index));
-        short4 src2_data = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
-
-        short4 tmp_data = src1_data & src2_data;
-
-        *((__global short4 *)((__global char *)dst + dst_index)) = tmp_data;
-    }
-}
-#endif
-__kernel void arithm_s_bitwise_and_C2_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 1)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffffc);
-
-        uchar4 src1_data = vload4(0, src1 + src1_index);
-        uchar4 src2_data = (uchar4)(src2.x, src2.y, src2.x, src2.y);
-
-        uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data & src2_data;
-
-
-        data.xy = (dst_index + 0 >= dst_start) ? tmp_data.xy : data.xy;
-        data.zw = (dst_index + 2 <  dst_end  ) ? tmp_data.zw : data.zw;
-
-        *((__global uchar4 *)(dst + dst_index)) = data;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_and_C2_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 1)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffffc);
-
-        char4 src1_data = vload4(0, src1 + src1_index);
-        char4 src2_data = (char4)(src2.x, src2.y, src2.x, src2.y);
-
-        char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data & src2_data;
-
-        data.xy = (dst_index + 0 >= dst_start) ? tmp_data.xy : data.xy;
-        data.zw = (dst_index + 2 <  dst_end  ) ? tmp_data.zw : data.zw;
-
-        *((__global char4 *)(dst + dst_index)) = data;
-    }
-}
-
-__kernel void arithm_s_bitwise_and_C2_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        ushort2 src_data1 = *((__global ushort2 *)((__global char *)src1 + src1_index));
-        ushort2 src_data2 = (ushort2)(src2.x, src2.y);
-
-        ushort2 data = src_data1 & src_data2;
-
-        *((__global ushort2 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C2_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        short2 src_data1 = *((__global short2 *)((__global char *)src1 + src1_index));
-        short2 src_data2 = (short2)(src2.x, src2.y);
-
-        short2 data = src_data1 & src_data2;
-
-        *((__global short2 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C2_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        int2 src_data1 = *((__global int2 *)((__global char *)src1 + src1_index));
-        int2 src_data2 = (int2)(src2.x, src2.y);
-
-        int2 data = src_data1 & src_data2;
-        *((__global int2 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C2_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        char8 src1_data = *((__global char8 *)((__global char *)src1 + src1_index));
-        char8 src2_data = (char8)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7);
-
-        char8 tmp_data = src1_data & src2_data;
-
-        *((__global char8 *)((__global char *)dst + dst_index)) = tmp_data;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_and_C2_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 4) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 4) + dst_offset);
-
-        short8 src1_data = *((__global short8 *)((__global char *)src1 + src1_index));
-        short8 src2_data = (short8)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7);
-
-        short8 tmp_data = src1_data & src2_data;
-
-        *((__global short8 *)((__global char *)dst + dst_index)) = tmp_data;
-    }
-}
-#endif
-__kernel void arithm_s_bitwise_and_C3_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        uchar4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        uchar4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        uchar4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        uchar4 src2_data_0 = (uchar4)(src2.x, src2.y, src2.z, src2.x);
-        uchar4 src2_data_1 = (uchar4)(src2.y, src2.z, src2.x, src2.y);
-        uchar4 src2_data_2 = (uchar4)(src2.z, src2.x, src2.y, src2.z);
-
-        uchar4 data_0 = *((__global uchar4 *)(dst + dst_index + 0));
-        uchar4 data_1 = *((__global uchar4 *)(dst + dst_index + 4));
-        uchar4 data_2 = *((__global uchar4 *)(dst + dst_index + 8));
-
-        uchar4 tmp_data_0 = src1_data_0 & src2_data_0;
-        uchar4 tmp_data_1 = src1_data_1 & src2_data_1;
-        uchar4 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0.xyz = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global uchar4 *)(dst + dst_index + 0)) = data_0;
-        *((__global uchar4 *)(dst + dst_index + 4)) = data_1;
-        *((__global uchar4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_and_C3_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        char4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        char4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        char4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        char4 src2_data_0 = (char4)(src2.x, src2.y, src2.z, src2.x);
-        char4 src2_data_1 = (char4)(src2.y, src2.z, src2.x, src2.y);
-        char4 src2_data_2 = (char4)(src2.z, src2.x, src2.y, src2.z);
-
-        char4 data_0 = *((__global char4 *)(dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)(dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)(dst + dst_index + 8));
-
-        char4 tmp_data_0 = convert_char4_sat(convert_uchar4_sat(src1_data_0) & convert_uchar4_sat(src2_data_0));
-        char4 tmp_data_1 = convert_char4_sat(convert_uchar4_sat(src1_data_1) & convert_uchar4_sat(src2_data_1));
-        char4 tmp_data_2 = convert_char4_sat(convert_uchar4_sat(src1_data_2) & convert_uchar4_sat(src2_data_2));
-
-        data_0.xyz = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global char4 *)(dst + dst_index + 0)) = data_0;
-        *((__global char4 *)(dst + dst_index + 4)) = data_1;
-        *((__global char4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-__kernel void arithm_s_bitwise_and_C3_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        ushort2 src1_data_0 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 0));
-        ushort2 src1_data_1 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 4));
-        ushort2 src1_data_2 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 8));
-
-        ushort2 src2_data_0 = (ushort2)(src2.x, src2.y);
-        ushort2 src2_data_1 = (ushort2)(src2.z, src2.x);
-        ushort2 src2_data_2 = (ushort2)(src2.y, src2.z);
-
-        ushort2 data_0 = *((__global ushort2 *)((__global char *)dst + dst_index + 0));
-        ushort2 data_1 = *((__global ushort2 *)((__global char *)dst + dst_index + 4));
-        ushort2 data_2 = *((__global ushort2 *)((__global char *)dst + dst_index + 8));
-
-        ushort2 tmp_data_0 = src1_data_0 & src2_data_0;
-        ushort2 tmp_data_1 = src1_data_1 & src2_data_1;
-        ushort2 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0.xy = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global ushort2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_and_C3_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        short2 src1_data_0 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 0));
-        short2 src1_data_1 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 4));
-        short2 src1_data_2 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 8));
-
-        short2 src2_data_0 = (short2)(src2.x, src2.y);
-        short2 src2_data_1 = (short2)(src2.z, src2.x);
-        short2 src2_data_2 = (short2)(src2.y, src2.z);
-
-        short2 data_0 = *((__global short2 *)((__global char *)dst + dst_index + 0));
-        short2 data_1 = *((__global short2 *)((__global char *)dst + dst_index + 4));
-        short2 data_2 = *((__global short2 *)((__global char *)dst + dst_index + 8));
-
-        short2 tmp_data_0 = src1_data_0 & src2_data_0;
-        short2 tmp_data_1 = src1_data_1 & src2_data_1;
-        short2 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0.xy = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global short2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global short2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global short2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_and_C3_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        int src1_data_0 = *((__global int *)((__global char *)src1 + src1_index + 0));
-        int src1_data_1 = *((__global int *)((__global char *)src1 + src1_index + 4));
-        int src1_data_2 = *((__global int *)((__global char *)src1 + src1_index + 8));
-
-        int src2_data_0 = src2.x;
-        int src2_data_1 = src2.y;
-        int src2_data_2 = src2.z;
-
-        int data_0 = *((__global int *)((__global char *)dst + dst_index + 0));
-        int data_1 = *((__global int *)((__global char *)dst + dst_index + 4));
-        int data_2 = *((__global int *)((__global char *)dst + dst_index + 8));
-
-        int tmp_data_0 = src1_data_0 & src2_data_0;
-        int tmp_data_1 = src1_data_1 & src2_data_1;
-        int tmp_data_2 = src1_data_2 & src2_data_2;
-
-        *((__global int *)((__global char *)dst + dst_index + 0))= tmp_data_0;
-        *((__global int *)((__global char *)dst + dst_index + 4))= tmp_data_1;
-        *((__global int *)((__global char *)dst + dst_index + 8))= tmp_data_2;
-    }
-}
-__kernel void arithm_s_bitwise_and_C3_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        char4 src1_data_0 = *((__global char4 *)((__global char *)src1 + src1_index + 0));
-        char4 src1_data_1 = *((__global char4 *)((__global char *)src1 + src1_index + 4));
-        char4 src1_data_2 = *((__global char4 *)((__global char *)src1 + src1_index + 8));
-
-        char4 src2_data_0 = (char4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        char4 src2_data_1 = (char4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        char4 src2_data_2 = (char4)(src2.s8, src2.s9, src2.sA, src2.sB);
-
-        char4 data_0 = *((__global char4 *)((__global char *)dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)((__global char *)dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)((__global char *)dst + dst_index + 8));
-
-        char4 tmp_data_0 = src1_data_0 & src2_data_0;
-        char4 tmp_data_1 = src1_data_1 & src2_data_1;
-        char4 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        *((__global char4 *)((__global char *)dst + dst_index + 0))= tmp_data_0;
-        *((__global char4 *)((__global char *)dst + dst_index + 4))= tmp_data_1;
-        *((__global char4 *)((__global char *)dst + dst_index + 8))= tmp_data_2;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_and_C3_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 24) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 24));
-
-        short4 src1_data_0 = *((__global short4 *)((__global char *)src1 + src1_index + 0 ));
-        short4 src1_data_1 = *((__global short4 *)((__global char *)src1 + src1_index + 8 ));
-        short4 src1_data_2 = *((__global short4 *)((__global char *)src1 + src1_index + 16));
-
-        short4 src2_data_0 = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        short4 src2_data_1 = (short4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        short4 src2_data_2 = (short4)(src2.s8, src2.s9, src2.sa, src2.sb);
-
-        short4 data_0 = *((__global short4 *)((__global char *)dst + dst_index + 0 ));
-        short4 data_1 = *((__global short4 *)((__global char *)dst + dst_index + 8 ));
-        short4 data_2 = *((__global short4 *)((__global char *)dst + dst_index + 16));
-
-        short4 tmp_data_0 = src1_data_0 & src2_data_0;
-        short4 tmp_data_1 = src1_data_1 & src2_data_1;
-        short4 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        *((__global short4 *)((__global char *)dst + dst_index + 0 ))= tmp_data_0;
-        *((__global short4 *)((__global char *)dst + dst_index + 8 ))= tmp_data_1;
-        *((__global short4 *)((__global char *)dst + dst_index + 16))= tmp_data_2;
-    }
-}
-#endif
-__kernel void arithm_s_bitwise_and_C4_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        uchar4 src_data1 = *((__global uchar4 *)(src1 + src1_index));
-
-        uchar4 data = src_data1 & src2;
-
-        *((__global uchar4 *)(dst + dst_index)) = data;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_and_C4_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        char4 src_data1 = *((__global char4 *)(src1 + src1_index));
-
-        char4 data = src_data1 & src2;
-
-        *((__global char4 *)(dst + dst_index)) = data;
-    }
-}
-
-__kernel void arithm_s_bitwise_and_C4_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        ushort4 src_data1 = *((__global ushort4 *)((__global char *)src1 + src1_index));
-
-        ushort4 data = src_data1 & src2;
-
-        *((__global ushort4 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C4_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        short4 src_data1 = *((__global short4 *)((__global char *)src1 + src1_index));
-
-        short4 data = src_data1 & src2;
-
-        *((__global short4 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C4_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 4) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 4) + dst_offset);
-
-        int4 src_data1 = *((__global int4 *)((__global char *)src1 + src1_index));
-
-        int4 data = src_data1 & src2;
-
-        *((__global int4 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_and_C4_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 4) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 4) + dst_offset);
-
-        char16 src1_data = *((__global char16 *)((__global char *)src1 + src1_index));
-        char16 src2_data = (char16)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7,
-                                    src2.s8, src2.s9, src2.sa, src2.sb, src2.sc, src2.sd, src2.se, src2.sf);
-
-        char16 tmp_data = src1_data & src2_data;
-
-        *((__global char16 *)((__global char *)dst + dst_index)) = tmp_data;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_and_C4_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 5) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 5) + dst_offset);
-
-        short4 src1_data_0 = *((__global short4 *)((__global char *)src1 + src1_index + 0));
-        short4 src1_data_1 = *((__global short4 *)((__global char *)src1 + src1_index + 8));
-        short4 src1_data_2 = *((__global short4 *)((__global char *)src1 + src1_index + 16));
-        short4 src1_data_3 = *((__global short4 *)((__global char *)src1 + src1_index + 24));
-
-        short4 src2_data_0 = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        short4 src2_data_1 = (short4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        short4 src2_data_2 = (short4)(src2.s8, src2.s9, src2.sa, src2.sb);
-        short4 src2_data_3 = (short4)(src2.sc, src2.sd, src2.se, src2.sf);
-
-        short4 tmp_data_0 = src1_data_0 & src2_data_0;
-        short4 tmp_data_1 = src1_data_1 & src2_data_1;
-        short4 tmp_data_2 = src1_data_2 & src2_data_2;
-        short4 tmp_data_3 = src1_data_3 & src2_data_3;
-
-        *((__global short4 *)((__global char *)dst + dst_index + 0 ))= tmp_data_0;
-        *((__global short4 *)((__global char *)dst + dst_index + 8 ))= tmp_data_1;
-        *((__global short4 *)((__global char *)dst + dst_index + 16))= tmp_data_2;
-        *((__global short4 *)((__global char *)dst + dst_index + 24))= tmp_data_3;
-
-    }
-}
-#endif

modules/ocl/src/opencl/arithm_bitwise_binary.cl

@@ -16,6 +16,7 @@
 //
 // @Authors
 //    Jiang Liyuan, jlyuan001.good@163.com
+//    Peng Xiao,    pengxiao@outlook.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -50,11 +51,17 @@
 #endif
 #endif
 
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////BITWISE_AND////////////////////////////////////////////////////
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-/**************************************bitwise_and without mask**************************************/
-__kernel void arithm_bitwise_and_D0 (__global uchar *src1, int src1_step, int src1_offset,
+//bitwise_binary without mask for and, or, xor operators
+
+/////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////bitwise_binary///////////////////////////////////////////
+/////////////////////////////////////////////////////////////////////////////////////////////////////
+
+#ifndef OP_BINARY
+#define OP_BINARY &
+#endif
+
+__kernel void arithm_bitwise_binary_D0 (__global uchar *src1, int src1_step, int src1_offset,
                                      __global uchar *src2, int src2_step, int src2_offset,
                                      __global uchar *dst,  int dst_step,  int dst_offset,
                                      int rows, int cols, int dst_step1)
@@ -95,7 +102,7 @@ __kernel void arithm_bitwise_and_D0 (__global uchar *src1, int src1_step, int sr
         }
 
         uchar4 dst_data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data & src2_data;
+        uchar4 tmp_data = src1_data OP_BINARY src2_data;
 
         dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
         dst_data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : dst_data.y;
@@ -107,7 +114,7 @@ __kernel void arithm_bitwise_and_D0 (__global uchar *src1, int src1_step, int sr
 }
 
 
-__kernel void arithm_bitwise_and_D1 (__global char *src1, int src1_step, int src1_offset,
+__kernel void arithm_bitwise_binary_D1 (__global char *src1, int src1_step, int src1_offset,
                                      __global char *src2, int src2_step, int src2_offset,
                                      __global char *dst,  int dst_step,  int dst_offset,
                                      int rows, int cols, int dst_step1)
@@ -148,7 +155,7 @@ __kernel void arithm_bitwise_and_D1 (__global char *src1, int src1_step, int src
             src2_data.xyzw = (src2_index == -1) ? src2_data.wxyz:tmp.xyzw;
         }
         char4 dst_data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data & src2_data;
+        char4 tmp_data = src1_data OP_BINARY src2_data;
 
         dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
         dst_data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : dst_data.y;
@@ -160,7 +167,7 @@ __kernel void arithm_bitwise_and_D1 (__global char *src1, int src1_step, int src
 }
 
 
-__kernel void arithm_bitwise_and_D2 (__global ushort *src1, int src1_step, int src1_offset,
+__kernel void arithm_bitwise_binary_D2 (__global ushort *src1, int src1_step, int src1_offset,
                                      __global ushort *src2, int src2_step, int src2_offset,
                                      __global ushort *dst,  int dst_step,  int dst_offset,
                                      int rows, int cols, int dst_step1)
@@ -202,7 +209,7 @@ __kernel void arithm_bitwise_and_D2 (__global ushort *src1, int src1_step, int s
             src2_data.xyzw = (src2_index == -1) ? src2_data.wxyz:tmp.xyzw;
         }
         ushort4 dst_data = *((__global ushort4 *)((__global char *)dst + dst_index));
-        ushort4 tmp_data = src1_data & src2_data;
+        ushort4 tmp_data = src1_data OP_BINARY src2_data;
 
         dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
         dst_data.y = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : dst_data.y;
@@ -215,7 +222,7 @@ __kernel void arithm_bitwise_and_D2 (__global ushort *src1, int src1_step, int s
 
 
 
-__kernel void arithm_bitwise_and_D3 (__global short *src1, int src1_step, int src1_offset,
+__kernel void arithm_bitwise_binary_D3 (__global short *src1, int src1_step, int src1_offset,
                                      __global short *src2, int src2_step, int src2_offset,
                                      __global short *dst,  int dst_step,  int dst_offset,
                                      int rows, int cols, int dst_step1)
@@ -257,7 +264,7 @@ __kernel void arithm_bitwise_and_D3 (__global short *src1, int src1_step, int sr
             src2_data.xyzw = (src2_index == -1) ? src2_data.wxyz:tmp.xyzw;
         }
         short4 dst_data = *((__global short4 *)((__global char *)dst + dst_index));
-        short4 tmp_data = src1_data & src2_data;
+        short4 tmp_data = src1_data OP_BINARY src2_data;
 
         dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
         dst_data.y = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : dst_data.y;
@@ -270,7 +277,7 @@ __kernel void arithm_bitwise_and_D3 (__global short *src1, int src1_step, int sr
 
 
 
-__kernel void arithm_bitwise_and_D4 (__global int *src1, int src1_step, int src1_offset,
+__kernel void arithm_bitwise_binary_D4 (__global int *src1, int src1_step, int src1_offset,
                                      __global int *src2, int src2_step, int src2_offset,
                                      __global int *dst,  int dst_step,  int dst_offset,
                                      int rows, int cols, int dst_step1)
@@ -286,13 +293,13 @@ __kernel void arithm_bitwise_and_D4 (__global int *src1, int src1_step, int src1
 
         int data1 = *((__global int *)((__global char *)src1 + src1_index));
         int data2 = *((__global int *)((__global char *)src2 + src2_index));
-        int tmp  = data1 & data2;
+        int tmp  = data1 OP_BINARY data2;
 
         *((__global int *)((__global char *)dst + dst_index)) = tmp;
     }
 }
 
-__kernel void arithm_bitwise_and_D5 (__global char *src1, int src1_step, int src1_offset,
+__kernel void arithm_bitwise_binary_D5 (__global char *src1, int src1_step, int src1_offset,
                                      __global char *src2, int src2_step, int src2_offset,
                                      __global char *dst,  int dst_step,  int dst_offset,
                                      int rows, int cols, int dst_step1)
@@ -308,14 +315,14 @@ __kernel void arithm_bitwise_and_D5 (__global char *src1, int src1_step, int src
 
         char4 data1 = *((__global char4 *)((__global char *)src1 + src1_index));
         char4 data2 = *((__global char4 *)((__global char *)src2 + src2_index));
-        char4 tmp = data1 & data2;
+        char4 tmp = data1 OP_BINARY data2;
 
         *((__global char4 *)((__global char *)dst + dst_index)) = tmp;
     }
 }
 
 #if defined (DOUBLE_SUPPORT)
-__kernel void arithm_bitwise_and_D6 (__global char *src1, int src1_step, int src1_offset,
+__kernel void arithm_bitwise_binary_D6 (__global char *src1, int src1_step, int src1_offset,
                                      __global char *src2, int src2_step, int src2_offset,
                                      __global char *dst,  int dst_step,  int dst_offset,
                                      int rows, int cols, int dst_step1)
@@ -332,7 +339,7 @@ __kernel void arithm_bitwise_and_D6 (__global char *src1, int src1_step, int src
         char8 data1 = *((__global char8 *)((__global char *)src1 + src1_index));
         char8 data2 = *((__global char8 *)((__global char *)src2 + src2_index));
 
-        *((__global char8 *)((__global char *)dst + dst_index)) = data1 & data2;
+        *((__global char8 *)((__global char *)dst + dst_index)) = data1 OP_BINARY data2;
     }
 }
 #endif

modules/ocl/src/opencl/arithm_bitwise_binary_mask.cl

@@ -16,6 +16,7 @@
 //
 // @Authors
 //    Jiang Liyuan, jlyuan001.good@163.com
+//    Peng Xiao,    pengxiao@outlook.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -49,11 +50,16 @@
 #pragma OPENCL EXTENSION cl_amd_fp64:enable
 #endif
 #endif
+
+#ifndef OP_BINARY
+#define OP_BINARY &
+#endif
+
+//////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////bitwise_binary////////////////////////////////////////////
 //////////////////////////////////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////BITWISE_AND////////////////////////////////////////////////////
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-/**************************************bitwise_and with mask**************************************/
-__kernel void arithm_bitwise_and_with_mask_C1_D0 (
+/**************************************bitwise_binary with mask**************************************/
+__kernel void arithm_bitwise_binary_with_mask_C1_D0 (
         __global uchar *src1, int src1_step, int src1_offset,
         __global uchar *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -85,7 +91,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D0 (
         uchar4 mask_data = vload4(0, mask + mask_index);
 
         uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data & src2_data;
+        uchar4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = ((mask_data.y) && (dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
@@ -98,7 +104,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D0 (
 
 
 
-__kernel void arithm_bitwise_and_with_mask_C1_D1 (
+__kernel void arithm_bitwise_binary_with_mask_C1_D1 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -130,7 +136,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D1 (
         uchar4 mask_data = vload4(0, mask + mask_index);
 
         char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data & src2_data;
+        char4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = convert_char((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = convert_char((mask_data.y) && (dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
@@ -143,7 +149,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D1 (
 
 
 
-__kernel void arithm_bitwise_and_with_mask_C1_D2 (
+__kernel void arithm_bitwise_binary_with_mask_C1_D2 (
         __global ushort *src1, int src1_step, int src1_offset,
         __global ushort *src2, int src2_step, int src2_offset,
         __global uchar  *mask, int mask_step, int mask_offset,
@@ -161,7 +167,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D2 (
 #ifdef dst_align
 #undef dst_align
 #endif
-#define dst_align ((dst_offset >> 1) & 1)
+#define dst_align ((dst_offset / 2) & 1)
         int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
         int src2_index = mad24(y, src2_step, (x << 1) + src2_offset - (dst_align << 1));
         int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
@@ -175,7 +181,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D2 (
         uchar2  mask_data = vload2(0, mask + mask_index);
 
         ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index));
-        ushort2 tmp_data = src1_data & src2_data;
+        ushort2 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = convert_ushort((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = convert_ushort((mask_data.y) && (dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : data.y;
@@ -186,7 +192,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D2 (
 
 
 
-__kernel void arithm_bitwise_and_with_mask_C1_D3 (
+__kernel void arithm_bitwise_binary_with_mask_C1_D3 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -204,7 +210,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D3 (
 #ifdef dst_align
 #undef dst_align
 #endif
-#define dst_align ((dst_offset >> 1) & 1)
+#define dst_align ((dst_offset / 2) & 1)
         int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
         int src2_index = mad24(y, src2_step, (x << 1) + src2_offset - (dst_align << 1));
         int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
@@ -218,7 +224,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D3 (
         uchar2  mask_data = vload2(0, mask + mask_index);
 
         short2 data = *((__global short2 *)((__global uchar *)dst + dst_index));
-        short2 tmp_data = src1_data & src2_data;
+        short2 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = convert_short((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = convert_short((mask_data.y) && (dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : data.y;
@@ -229,7 +235,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D3 (
 
 
 
-__kernel void arithm_bitwise_and_with_mask_C1_D4 (
+__kernel void arithm_bitwise_binary_with_mask_C1_D4 (
         __global int   *src1, int src1_step, int src1_offset,
         __global int   *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -253,7 +259,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D4 (
         int src_data2 = *((__global int *)((__global char *)src2 + src2_index));
         int dst_data  = *((__global int *)((__global char *)dst  + dst_index));
 
-        int data = src_data1 & src_data2;
+        int data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global int *)((__global char *)dst + dst_index)) = data;
@@ -262,7 +268,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D4 (
 
 
 
-__kernel void arithm_bitwise_and_with_mask_C1_D5 (
+__kernel void arithm_bitwise_binary_with_mask_C1_D5 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -286,7 +292,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D5 (
         char4 src_data2 = *((__global char4 *)((__global char *)src2 + src2_index));
         char4 dst_data  = *((__global char4 *)((__global char *)dst  + dst_index));
 
-        char4 data = src_data1 & src_data2;
+        char4 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global char4 *)((__global char *)dst + dst_index)) = data;
@@ -295,7 +301,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D5 (
 
 
 
-__kernel void arithm_bitwise_and_with_mask_C1_D6 (
+__kernel void arithm_bitwise_binary_with_mask_C1_D6 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -319,7 +325,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D6 (
         char8 src_data2 = *((__global char8 *)((__global char *)src2 + src2_index));
         char8 dst_data  = *((__global char8 *)((__global char *)dst  + dst_index));
 
-        char8 data = src_data1 & src_data2;
+        char8 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global char8 *)((__global char *)dst + dst_index)) = data;
@@ -329,7 +335,7 @@ __kernel void arithm_bitwise_and_with_mask_C1_D6 (
 
 
 
-__kernel void arithm_bitwise_and_with_mask_C2_D0 (
+__kernel void arithm_bitwise_binary_with_mask_C2_D0 (
         __global uchar *src1, int src1_step, int src1_offset,
         __global uchar *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -347,7 +353,7 @@ __kernel void arithm_bitwise_and_with_mask_C2_D0 (
 #ifdef dst_align
 #undef dst_align
 #endif
-#define dst_align ((dst_offset >> 1) & 1)
+#define dst_align ((dst_offset / 2) & 1)
         int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
         int src2_index = mad24(y, src2_step, (x << 1) + src2_offset - (dst_align << 1));
         int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
@@ -361,7 +367,7 @@ __kernel void arithm_bitwise_and_with_mask_C2_D0 (
         uchar2 mask_data = vload2(0, mask + mask_index);
 
         uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data & src2_data;
+        uchar4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.xy : data.xy;
         data.zw = ((mask_data.y) && (dst_index + 2 <  dst_end  )) ? tmp_data.zw : data.zw;
@@ -371,7 +377,7 @@ __kernel void arithm_bitwise_and_with_mask_C2_D0 (
 }
 
 
-__kernel void arithm_bitwise_and_with_mask_C2_D1 (
+__kernel void arithm_bitwise_binary_with_mask_C2_D1 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -389,7 +395,7 @@ __kernel void arithm_bitwise_and_with_mask_C2_D1 (
 #ifdef dst_align
 #undef dst_align
 #endif
-#define dst_align ((dst_offset >> 1) & 1)
+#define dst_align ((dst_offset / 2) & 1)
         int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
         int src2_index = mad24(y, src2_step, (x << 1) + src2_offset - (dst_align << 1));
         int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
@@ -403,7 +409,7 @@ __kernel void arithm_bitwise_and_with_mask_C2_D1 (
         uchar2 mask_data = vload2(0, mask + mask_index);
 
         char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data & src2_data;
+        char4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.xy : data.xy;
         data.zw = ((mask_data.y) && (dst_index + 2 <  dst_end  )) ? tmp_data.zw : data.zw;
@@ -412,7 +418,7 @@ __kernel void arithm_bitwise_and_with_mask_C2_D1 (
     }
 }
 
-__kernel void arithm_bitwise_and_with_mask_C2_D2 (
+__kernel void arithm_bitwise_binary_with_mask_C2_D2 (
         __global ushort *src1, int src1_step, int src1_offset,
         __global ushort *src2, int src2_step, int src2_offset,
         __global uchar  *mask, int mask_step, int mask_offset,
@@ -436,13 +442,13 @@ __kernel void arithm_bitwise_and_with_mask_C2_D2 (
         ushort2 src_data2 = *((__global ushort2 *)((__global char *)src2 + src2_index));
         ushort2 dst_data  = *((__global ushort2 *)((__global char *)dst  + dst_index));
 
-        ushort2 data = src_data1 & src_data2;
+        ushort2 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global ushort2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_bitwise_and_with_mask_C2_D3 (
+__kernel void arithm_bitwise_binary_with_mask_C2_D3 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -466,13 +472,13 @@ __kernel void arithm_bitwise_and_with_mask_C2_D3 (
         short2 src_data2 = *((__global short2 *)((__global char *)src2 + src2_index));
         short2 dst_data  = *((__global short2 *)((__global char *)dst  + dst_index));
 
-        short2 data = src_data1 & src_data2;
+        short2 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global short2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_bitwise_and_with_mask_C2_D4 (
+__kernel void arithm_bitwise_binary_with_mask_C2_D4 (
         __global int   *src1, int src1_step, int src1_offset,
         __global int   *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -496,13 +502,13 @@ __kernel void arithm_bitwise_and_with_mask_C2_D4 (
         int2 src_data2 = *((__global int2 *)((__global char *)src2 + src2_index));
         int2 dst_data  = *((__global int2 *)((__global char *)dst  + dst_index));
 
-        int2 data = src_data1 & src_data2;
+        int2 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global int2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_bitwise_and_with_mask_C2_D5 (
+__kernel void arithm_bitwise_binary_with_mask_C2_D5 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -526,14 +532,14 @@ __kernel void arithm_bitwise_and_with_mask_C2_D5 (
         char8 src_data2 = *((__global char8 *)((__global char *)src2 + src2_index));
         char8 dst_data  = *((__global char8 *)((__global char *)dst  + dst_index));
 
-        char8 data = src_data1 & src_data2;
+        char8 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global char8 *)((__global char *)dst + dst_index)) = data;
     }
 }
 
-__kernel void arithm_bitwise_and_with_mask_C2_D6 (
+__kernel void arithm_bitwise_binary_with_mask_C2_D6 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -557,7 +563,7 @@ __kernel void arithm_bitwise_and_with_mask_C2_D6 (
         char16 src_data2 = *((__global char16 *)((__global char *)src2 + src2_index));
         char16 dst_data  = *((__global char16 *)((__global char *)dst  + dst_index));
 
-        char16 data = src_data1 & src_data2;
+        char16 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global char16 *)((__global char *)dst + dst_index)) = data;
@@ -565,398 +571,7 @@ __kernel void arithm_bitwise_and_with_mask_C2_D6 (
 }
 
 
-
-__kernel void arithm_bitwise_and_with_mask_C3_D0 (
-        __global uchar *src1, int src1_step, int src1_offset,
-        __global uchar *src2, int src2_step, int src2_offset,
-        __global uchar *mask, int mask_step, int mask_offset,
-        __global uchar *dst,  int dst_step,  int dst_offset,
-        int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-        int src2_index = mad24(y, src2_step, (x * 3) + src2_offset - (dst_align * 3));
-        int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        uchar4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        uchar4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        uchar4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        uchar4 src2_data_0 = vload4(0, src2 + src2_index + 0);
-        uchar4 src2_data_1 = vload4(0, src2 + src2_index + 4);
-        uchar4 src2_data_2 = vload4(0, src2 + src2_index + 8);
-
-        uchar4 mask_data = vload4(0, mask + mask_index);
-
-        uchar4 data_0 = *((__global uchar4 *)(dst + dst_index + 0));
-        uchar4 data_1 = *((__global uchar4 *)(dst + dst_index + 4));
-        uchar4 data_2 = *((__global uchar4 *)(dst + dst_index + 8));
-
-        uchar4 tmp_data_0 =  src1_data_0 & src2_data_0;
-        uchar4 tmp_data_1 =  src1_data_1 & src2_data_1;
-        uchar4 tmp_data_2 =  src1_data_2 & src2_data_2;
-
-        data_0.xyz = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((mask_data.w) && (dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global uchar4 *)(dst + dst_index + 0)) = data_0;
-        *((__global uchar4 *)(dst + dst_index + 4)) = data_1;
-        *((__global uchar4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-
-__kernel void arithm_bitwise_and_with_mask_C3_D1 (
-        __global char *src1, int src1_step, int src1_offset,
-        __global char *src2, int src2_step, int src2_offset,
-        __global uchar *mask, int mask_step, int mask_offset,
-        __global char *dst,  int dst_step,  int dst_offset,
-        int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-        int src2_index = mad24(y, src2_step, (x * 3) + src2_offset - (dst_align * 3));
-        int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        char4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        char4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        char4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        char4 src2_data_0 = vload4(0, src2 + src2_index + 0);
-        char4 src2_data_1 = vload4(0, src2 + src2_index + 4);
-        char4 src2_data_2 = vload4(0, src2 + src2_index + 8);
-
-        uchar4 mask_data = vload4(0, mask + mask_index);
-
-        char4 data_0 = *((__global char4 *)(dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)(dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)(dst + dst_index + 8));
-
-        char4 tmp_data_0 =  src1_data_0 & src2_data_0;
-        char4 tmp_data_1 =  src1_data_1 & src2_data_1;
-        char4 tmp_data_2 =  src1_data_2 & src2_data_2;
-
-        data_0.xyz = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((mask_data.w) && (dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global char4 *)(dst + dst_index + 0)) = data_0;
-        *((__global char4 *)(dst + dst_index + 4)) = data_1;
-        *((__global char4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-__kernel void arithm_bitwise_and_with_mask_C3_D2 (
-        __global ushort *src1, int src1_step, int src1_offset,
-        __global ushort *src2, int src2_step, int src2_offset,
-        __global uchar  *mask, int mask_step, int mask_offset,
-        __global ushort *dst,  int dst_step,  int dst_offset,
-        int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-        int src2_index = mad24(y, src2_step, (x * 6) + src2_offset - (dst_align * 6));
-        int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        ushort2 src1_data_0 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 0));
-        ushort2 src1_data_1 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 4));
-        ushort2 src1_data_2 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 8));
-
-        ushort2 src2_data_0 = vload2(0, (__global ushort *)((__global char *)src2 + src2_index + 0));
-        ushort2 src2_data_1 = vload2(0, (__global ushort *)((__global char *)src2 + src2_index + 4));
-        ushort2 src2_data_2 = vload2(0, (__global ushort *)((__global char *)src2 + src2_index + 8));
-
-        uchar2 mask_data = vload2(0, mask + mask_index);
-
-        ushort2 data_0 = *((__global ushort2 *)((__global char *)dst + dst_index + 0));
-        ushort2 data_1 = *((__global ushort2 *)((__global char *)dst + dst_index + 4));
-        ushort2 data_2 = *((__global ushort2 *)((__global char *)dst + dst_index + 8));
-
-        ushort2 tmp_data_0 =  src1_data_0 & src2_data_0 ;
-        ushort2 tmp_data_1 =  src1_data_1 & src2_data_1 ;
-        ushort2 tmp_data_2 =  src1_data_2 & src2_data_2 ;
-
-        data_0.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global ushort2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_bitwise_and_with_mask_C3_D3 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *src2, int src2_step, int src2_offset,
-        __global uchar  *mask, int mask_step, int mask_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-        int src2_index = mad24(y, src2_step, (x * 6) + src2_offset - (dst_align * 6));
-        int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        short2 src1_data_0 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 0));
-        short2 src1_data_1 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 4));
-        short2 src1_data_2 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 8));
-
-        short2 src2_data_0 = vload2(0, (__global short *)((__global char *)src2 + src2_index + 0));
-        short2 src2_data_1 = vload2(0, (__global short *)((__global char *)src2 + src2_index + 4));
-        short2 src2_data_2 = vload2(0, (__global short *)((__global char *)src2 + src2_index + 8));
-
-        uchar2 mask_data = vload2(0, mask + mask_index);
-
-        short2 data_0 = *((__global short2 *)((__global char *)dst + dst_index + 0));
-        short2 data_1 = *((__global short2 *)((__global char *)dst + dst_index + 4));
-        short2 data_2 = *((__global short2 *)((__global char *)dst + dst_index + 8));
-
-        short2 tmp_data_0 =  src1_data_0 & src2_data_0 ;
-        short2 tmp_data_1 =  src1_data_1 & src2_data_1 ;
-        short2 tmp_data_2 =  src1_data_2 & src2_data_2 ;
-
-        data_0.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global short2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global short2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global short2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_bitwise_and_with_mask_C3_D4 (
-        __global int   *src1, int src1_step, int src1_offset,
-        __global int   *src2, int src2_step, int src2_offset,
-        __global uchar *mask, int mask_step, int mask_offset,
-        __global int   *dst,  int dst_step,  int dst_offset,
-        int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x * 12) + src2_offset);
-        int mask_index = mad24(y, mask_step, x + mask_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        int src1_data_0 = *((__global int *)((__global char *)src1 + src1_index + 0));
-        int src1_data_1 = *((__global int *)((__global char *)src1 + src1_index + 4));
-        int src1_data_2 = *((__global int *)((__global char *)src1 + src1_index + 8));
-
-        int src2_data_0 = *((__global int *)((__global char *)src2 + src2_index + 0));
-        int src2_data_1 = *((__global int *)((__global char *)src2 + src2_index + 4));
-        int src2_data_2 = *((__global int *)((__global char *)src2 + src2_index + 8));
-
-        uchar mask_data = * (mask + mask_index);
-
-        int data_0 = *((__global int *)((__global char *)dst + dst_index + 0));
-        int data_1 = *((__global int *)((__global char *)dst + dst_index + 4));
-        int data_2 = *((__global int *)((__global char *)dst + dst_index + 8));
-
-        int tmp_data_0 =  src1_data_0 &  src2_data_0 ;
-        int tmp_data_1 =  src1_data_1 &  src2_data_1 ;
-        int tmp_data_2 =  src1_data_2 &  src2_data_2 ;
-
-        data_0 = mask_data ? tmp_data_0 : data_0;
-        data_1 = mask_data ? tmp_data_1 : data_1;
-        data_2 = mask_data ? tmp_data_2 : data_2;
-
-        *((__global int *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global int *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global int *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_bitwise_and_with_mask_C3_D5 (
-        __global char *src1, int src1_step, int src1_offset,
-        __global char *src2, int src2_step, int src2_offset,
-        __global uchar *mask, int mask_step, int mask_offset,
-        __global char *dst,  int dst_step,  int dst_offset,
-        int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x * 12) + src2_offset);
-        int mask_index = mad24(y, mask_step, x + mask_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        char4 src1_data_0 = *((__global char4 *)((__global char *)src1 + src1_index + 0));
-        char4 src1_data_1 = *((__global char4 *)((__global char *)src1 + src1_index + 4));
-        char4 src1_data_2 = *((__global char4 *)((__global char *)src1 + src1_index + 8));
-
-        char4 src2_data_0 = *((__global char4 *)((__global char *)src2 + src2_index + 0));
-        char4 src2_data_1 = *((__global char4 *)((__global char *)src2 + src2_index + 4));
-        char4 src2_data_2 = *((__global char4 *)((__global char *)src2 + src2_index + 8));
-
-        uchar mask_data = * (mask + mask_index);
-
-        char4 data_0 = *((__global char4 *)((__global char *)dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)((__global char *)dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)((__global char *)dst + dst_index + 8));
-
-        char4 tmp_data_0 = src1_data_0 & src2_data_0;
-        char4 tmp_data_1 = src1_data_1 & src2_data_1;
-        char4 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0 = mask_data ? tmp_data_0 : data_0;
-        data_1 = mask_data ? tmp_data_1 : data_1;
-        data_2 = mask_data ? tmp_data_2 : data_2;
-
-        *((__global char4 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global char4 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global char4 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_bitwise_and_with_mask_C3_D6 (
-        __global char *src1, int src1_step, int src1_offset,
-        __global char *src2, int src2_step, int src2_offset,
-        __global uchar  *mask, int mask_step, int mask_offset,
-        __global char *dst,  int dst_step,  int dst_offset,
-        int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 24) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x * 24) + src2_offset);
-        int mask_index = mad24(y, mask_step, x + mask_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 24));
-
-        char8 src1_data_0 = *((__global char8 *)((__global char *)src1 + src1_index + 0 ));
-        char8 src1_data_1 = *((__global char8 *)((__global char *)src1 + src1_index + 8 ));
-        char8 src1_data_2 = *((__global char8 *)((__global char *)src1 + src1_index + 16));
-
-        char8 src2_data_0 = *((__global char8 *)((__global char *)src2 + src2_index + 0 ));
-        char8 src2_data_1 = *((__global char8 *)((__global char *)src2 + src2_index + 8 ));
-        char8 src2_data_2 = *((__global char8 *)((__global char *)src2 + src2_index + 16));
-
-        uchar mask_data = * (mask + mask_index);
-
-        char8 data_0 = *((__global char8 *)((__global char *)dst + dst_index + 0 ));
-        char8 data_1 = *((__global char8 *)((__global char *)dst + dst_index + 8 ));
-        char8 data_2 = *((__global char8 *)((__global char *)dst + dst_index + 16));
-
-        char8 tmp_data_0 = src1_data_0 & src2_data_0;
-        char8 tmp_data_1 = src1_data_1 & src2_data_1;
-        char8 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0 = mask_data ? tmp_data_0 : data_0;
-        data_1 = mask_data ? tmp_data_1 : data_1;
-        data_2 = mask_data ? tmp_data_2 : data_2;
-
-        *((__global char8 *)((__global char *)dst + dst_index + 0 ))= data_0;
-        *((__global char8 *)((__global char *)dst + dst_index + 8 ))= data_1;
-        *((__global char8 *)((__global char *)dst + dst_index + 16))= data_2;
-    }
-}
-#endif
-
-
-__kernel void arithm_bitwise_and_with_mask_C4_D0 (
+__kernel void arithm_bitwise_binary_with_mask_C4_D0 (
         __global uchar *src1, int src1_step, int src1_offset,
         __global uchar *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -980,7 +595,7 @@ __kernel void arithm_bitwise_and_with_mask_C4_D0 (
         uchar4 src_data2 = *((__global uchar4 *)(src2 + src2_index));
         uchar4 dst_data  = *((__global uchar4 *)(dst  + dst_index));
 
-        uchar4 data = src_data1 & src_data2;
+        uchar4 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global uchar4 *)(dst + dst_index)) = data;
@@ -988,7 +603,7 @@ __kernel void arithm_bitwise_and_with_mask_C4_D0 (
 }
 
 
-__kernel void arithm_bitwise_and_with_mask_C4_D1 (
+__kernel void arithm_bitwise_binary_with_mask_C4_D1 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -1012,14 +627,14 @@ __kernel void arithm_bitwise_and_with_mask_C4_D1 (
         char4 src_data2 = *((__global char4 *)(src2 + src2_index));
         char4 dst_data  = *((__global char4 *)(dst  + dst_index));
 
-        char4 data = src_data1 & src_data2;
+        char4 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global char4 *)(dst + dst_index)) = data;
     }
 }
 
-__kernel void arithm_bitwise_and_with_mask_C4_D2 (
+__kernel void arithm_bitwise_binary_with_mask_C4_D2 (
         __global ushort *src1, int src1_step, int src1_offset,
         __global ushort *src2, int src2_step, int src2_offset,
         __global uchar  *mask, int mask_step, int mask_offset,
@@ -1043,13 +658,13 @@ __kernel void arithm_bitwise_and_with_mask_C4_D2 (
         ushort4 src_data2 = *((__global ushort4 *)((__global char *)src2 + src2_index));
         ushort4 dst_data  = *((__global ushort4 *)((__global char *)dst  + dst_index));
 
-        ushort4 data = src_data1 & src_data2;
+        ushort4 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global ushort4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_bitwise_and_with_mask_C4_D3 (
+__kernel void arithm_bitwise_binary_with_mask_C4_D3 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -1073,13 +688,13 @@ __kernel void arithm_bitwise_and_with_mask_C4_D3 (
         short4 src_data2 = *((__global short4 *)((__global char *)src2 + src2_index));
         short4 dst_data  = *((__global short4 *)((__global char *)dst  + dst_index));
 
-        short4 data = src_data1 & src_data2;
+        short4 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global short4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_bitwise_and_with_mask_C4_D4 (
+__kernel void arithm_bitwise_binary_with_mask_C4_D4 (
         __global int   *src1, int src1_step, int src1_offset,
         __global int   *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -1103,13 +718,13 @@ __kernel void arithm_bitwise_and_with_mask_C4_D4 (
         int4 src_data2 = *((__global int4 *)((__global char *)src2 + src2_index));
         int4 dst_data  = *((__global int4 *)((__global char *)dst  + dst_index));
 
-        int4 data = src_data1 & src_data2;
+        int4 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global int4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_bitwise_and_with_mask_C4_D5 (
+__kernel void arithm_bitwise_binary_with_mask_C4_D5 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -1133,14 +748,14 @@ __kernel void arithm_bitwise_and_with_mask_C4_D5 (
         char16 src_data2 = *((__global char16 *)((__global char *)src2 + src2_index));
         char16 dst_data  = *((__global char16 *)((__global char *)dst  + dst_index));
 
-        char16 data = src_data1 & src_data2;
+        char16 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global char16 *)((__global char *)dst + dst_index)) = data;
     }
 }
 #if defined (DOUBLE_SUPPORT)
-__kernel void arithm_bitwise_and_with_mask_C4_D6 (
+__kernel void arithm_bitwise_binary_with_mask_C4_D6 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *src2, int src2_step, int src2_offset,
         __global uchar  *mask, int mask_step, int mask_offset,
@@ -1175,10 +790,10 @@ __kernel void arithm_bitwise_and_with_mask_C4_D6 (
         char8 dst_data_2  = *((__global char8 *)((__global char *)dst  + dst_index + 16));
         char8 dst_data_3  = *((__global char8 *)((__global char *)dst  + dst_index + 24));
 
-        char8 data_0 = src_data1_0 & src_data2_0;
-        char8 data_1 = src_data1_1 & src_data2_1;
-        char8 data_2 = src_data1_2 & src_data2_2;
-        char8 data_3 = src_data1_3 & src_data2_3;
+        char8 data_0 = src_data1_0 OP_BINARY src_data2_0;
+        char8 data_1 = src_data1_1 OP_BINARY src_data2_1;
+        char8 data_2 = src_data1_2 OP_BINARY src_data2_2;
+        char8 data_3 = src_data1_3 OP_BINARY src_data2_3;
 
         data_0 = mask_data ? data_0 : dst_data_0;
         data_1 = mask_data ? data_1 : dst_data_1;

modules/ocl/src/opencl/arithm_bitwise_binary_scalar.cl

@@ -16,6 +16,7 @@
 //
 // @Authors
 //    Jiang Liyuan, jlyuan001.good@163.com
+//    Peng Xiao,    pengxiao@outlook.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -49,11 +50,16 @@
 #pragma OPENCL EXTENSION cl_amd_fp64:enable
 #endif
 #endif
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////BITWISE_XOR////////////////////////////////////////////////////
+
+#ifndef OP_BINARY
+#define OP_BINARY &
+#endif
+
 ///////////////////////////////////////////////////////////////////////////////////////////////////////
-/**************************************xor with scalar without mask**************************************/
-__kernel void arithm_s_bitwise_xor_C1_D0 (
+////////////////////////////////////////////bitwise_binary/////////////////////////////////////////////
+///////////////////////////////////////////////////////////////////////////////////////////////////////
+/******************************bitwise binary with scalar without mask********************************/
+__kernel void arithm_s_bitwise_binary_C1_D0 (
         __global   uchar *src1, int src1_step, int src1_offset,
         __global   uchar *dst,  int dst_step,  int dst_offset,
         uchar4 src2, int rows, int cols, int dst_step1)
@@ -79,7 +85,7 @@ __kernel void arithm_s_bitwise_xor_C1_D0 (
         uchar4 src2_data = (uchar4)(src2.x, src2.x, src2.x, src2.x);
 
         uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data ^ src2_data;
+        uchar4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
@@ -91,7 +97,7 @@ __kernel void arithm_s_bitwise_xor_C1_D0 (
 }
 
 
-__kernel void arithm_s_bitwise_xor_C1_D1 (
+__kernel void arithm_s_bitwise_binary_C1_D1 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         char4 src2, int rows, int cols, int dst_step1)
@@ -117,7 +123,7 @@ __kernel void arithm_s_bitwise_xor_C1_D1 (
         char4 src2_data = (char4)(src2.x, src2.x, src2.x, src2.x);
 
         char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data ^ src2_data;
+        char4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
@@ -128,7 +134,7 @@ __kernel void arithm_s_bitwise_xor_C1_D1 (
     }
 }
 
-__kernel void arithm_s_bitwise_xor_C1_D2 (
+__kernel void arithm_s_bitwise_binary_C1_D2 (
         __global   ushort *src1, int src1_step, int src1_offset,
         __global   ushort *dst,  int dst_step,  int dst_offset,
         ushort4 src2, int rows, int cols, int dst_step1)
@@ -155,7 +161,7 @@ __kernel void arithm_s_bitwise_xor_C1_D2 (
         ushort2 src2_data = (ushort2)(src2.x, src2.x);
 
         ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index));
-        ushort2 tmp_data = src1_data ^ src2_data;
+        ushort2 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x;
         data.y = (dst_index + 2 <  dst_end  ) ? tmp_data.y : data.y;
@@ -163,7 +169,7 @@ __kernel void arithm_s_bitwise_xor_C1_D2 (
         *((__global ushort2 *)((__global uchar *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C1_D3 (
+__kernel void arithm_s_bitwise_binary_C1_D3 (
         __global   short *src1, int src1_step, int src1_offset,
         __global   short *dst,  int dst_step,  int dst_offset,
         short4 src2, int rows, int cols, int dst_step1)
@@ -190,7 +196,7 @@ __kernel void arithm_s_bitwise_xor_C1_D3 (
         short2 src2_data = (short2)(src2.x, src2.x);
         short2 data = *((__global short2 *)((__global uchar *)dst + dst_index));
 
-        short2 tmp_data = src1_data ^ src2_data;
+        short2 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x;
         data.y = (dst_index + 2 <  dst_end  ) ? tmp_data.y : data.y;
@@ -198,7 +204,7 @@ __kernel void arithm_s_bitwise_xor_C1_D3 (
         *((__global short2 *)((__global uchar *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C1_D4 (
+__kernel void arithm_s_bitwise_binary_C1_D4 (
         __global   int *src1, int src1_step, int src1_offset,
         __global   int *dst,  int dst_step,  int dst_offset,
         int4 src2, int rows, int cols, int dst_step1)
@@ -215,12 +221,12 @@ __kernel void arithm_s_bitwise_xor_C1_D4 (
         int src_data1 = *((__global int *)((__global char *)src1 + src1_index));
         int src_data2 = src2.x;
 
-        int data = src_data1 ^ src_data2;
+        int data = src_data1 OP_BINARY src_data2;
 
         *((__global int *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C1_D5 (
+__kernel void arithm_s_bitwise_binary_C1_D5 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         char16 src2, int rows, int cols, int dst_step1)
@@ -241,7 +247,7 @@ __kernel void arithm_s_bitwise_xor_C1_D5 (
         char4 src2_data = (char4)(src2.s0, src2.s1, src2.s2, src2.s3);
 
         char4 data  = *((__global char4 *)((__global char *)dst  + dst_index));
-        char4 tmp_data = src1_data ^ src2_data;
+        char4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
@@ -251,9 +257,8 @@ __kernel void arithm_s_bitwise_xor_C1_D5 (
         *((__global char4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-
 #if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_xor_C1_D6 (
+__kernel void arithm_s_bitwise_binary_C1_D6 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *dst,  int dst_step,  int dst_offset,
         short16 src2, int rows, int cols, int dst_step1)
@@ -270,13 +275,13 @@ __kernel void arithm_s_bitwise_xor_C1_D6 (
         short4 src1_data = *((__global short4 *)((__global char *)src1 + src1_index));
         short4 src2_data = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
 
-        short4 tmp_data = src1_data ^ src2_data;
+        short4 tmp_data = src1_data OP_BINARY src2_data;
 
         *((__global short4 *)((__global char *)dst + dst_index)) = tmp_data;
     }
 }
 #endif
-__kernel void arithm_s_bitwise_xor_C2_D0 (
+__kernel void arithm_s_bitwise_binary_C2_D0 (
         __global   uchar *src1, int src1_step, int src1_offset,
         __global   uchar *dst,  int dst_step,  int dst_offset,
         uchar4 src2, int rows, int cols, int dst_step1)
@@ -303,7 +308,7 @@ __kernel void arithm_s_bitwise_xor_C2_D0 (
         uchar4 src2_data = (uchar4)(src2.x, src2.y, src2.x, src2.y);
 
         uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data ^ src2_data;
+        uchar4 tmp_data = src1_data OP_BINARY src2_data;
 
 
         data.xy = (dst_index + 0 >= dst_start) ? tmp_data.xy : data.xy;
@@ -314,7 +319,7 @@ __kernel void arithm_s_bitwise_xor_C2_D0 (
 }
 
 
-__kernel void arithm_s_bitwise_xor_C2_D1 (
+__kernel void arithm_s_bitwise_binary_C2_D1 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         char4 src2, int rows, int cols, int dst_step1)
@@ -341,7 +346,7 @@ __kernel void arithm_s_bitwise_xor_C2_D1 (
         char4 src2_data = (char4)(src2.x, src2.y, src2.x, src2.y);
 
         char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data ^ src2_data;
+        char4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.xy = (dst_index + 0 >= dst_start) ? tmp_data.xy : data.xy;
         data.zw = (dst_index + 2 <  dst_end  ) ? tmp_data.zw : data.zw;
@@ -350,7 +355,7 @@ __kernel void arithm_s_bitwise_xor_C2_D1 (
     }
 }
 
-__kernel void arithm_s_bitwise_xor_C2_D2 (
+__kernel void arithm_s_bitwise_binary_C2_D2 (
         __global   ushort *src1, int src1_step, int src1_offset,
         __global   ushort *dst,  int dst_step,  int dst_offset,
         ushort4 src2, int rows, int cols, int dst_step1)
@@ -367,12 +372,12 @@ __kernel void arithm_s_bitwise_xor_C2_D2 (
         ushort2 src_data1 = *((__global ushort2 *)((__global char *)src1 + src1_index));
         ushort2 src_data2 = (ushort2)(src2.x, src2.y);
 
-        ushort2 data = src_data1 ^ src_data2;
+        ushort2 data = src_data1 OP_BINARY src_data2;
 
         *((__global ushort2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C2_D3 (
+__kernel void arithm_s_bitwise_binary_C2_D3 (
         __global   short *src1, int src1_step, int src1_offset,
         __global   short *dst,  int dst_step,  int dst_offset,
         short4 src2, int rows, int cols, int dst_step1)
@@ -389,12 +394,12 @@ __kernel void arithm_s_bitwise_xor_C2_D3 (
         short2 src_data1 = *((__global short2 *)((__global char *)src1 + src1_index));
         short2 src_data2 = (short2)(src2.x, src2.y);
 
-        short2 data = src_data1 ^ src_data2;
+        short2 data = src_data1 OP_BINARY src_data2;
 
         *((__global short2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C2_D4 (
+__kernel void arithm_s_bitwise_binary_C2_D4 (
         __global   int *src1, int src1_step, int src1_offset,
         __global   int *dst,  int dst_step,  int dst_offset,
         int4 src2, int rows, int cols, int dst_step1)
@@ -411,11 +416,11 @@ __kernel void arithm_s_bitwise_xor_C2_D4 (
         int2 src_data1 = *((__global int2 *)((__global char *)src1 + src1_index));
         int2 src_data2 = (int2)(src2.x, src2.y);
 
-        int2 data = src_data1 ^ src_data2;
+        int2 data = src_data1 OP_BINARY src_data2;
         *((__global int2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C2_D5 (
+__kernel void arithm_s_bitwise_binary_C2_D5 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         char16 src2, int rows, int cols, int dst_step1)
@@ -432,13 +437,13 @@ __kernel void arithm_s_bitwise_xor_C2_D5 (
         char8 src1_data = *((__global char8 *)((__global char *)src1 + src1_index));
         char8 src2_data = (char8)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7);
 
-        char8 tmp_data = src1_data ^ src2_data;
+        char8 tmp_data = src1_data OP_BINARY src2_data;
 
         *((__global char8 *)((__global char *)dst + dst_index)) = tmp_data;
     }
 }
 #if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_xor_C2_D6 (
+__kernel void arithm_s_bitwise_binary_C2_D6 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *dst,  int dst_step,  int dst_offset,
         short16 src2, int rows, int cols, int dst_step1)
@@ -455,347 +460,14 @@ __kernel void arithm_s_bitwise_xor_C2_D6 (
         short8 src1_data = *((__global short8 *)((__global char *)src1 + src1_index));
         short8 src2_data = (short8)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7);
 
-        short8 tmp_data = src1_data ^ src2_data;
+        short8 tmp_data = src1_data OP_BINARY src2_data;
 
         *((__global short8 *)((__global char *)dst + dst_index)) = tmp_data;
     }
 }
 #endif
-__kernel void arithm_s_bitwise_xor_C3_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        uchar4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        uchar4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        uchar4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        uchar4 src2_data_0 = (uchar4)(src2.x, src2.y, src2.z, src2.x);
-        uchar4 src2_data_1 = (uchar4)(src2.y, src2.z, src2.x, src2.y);
-        uchar4 src2_data_2 = (uchar4)(src2.z, src2.x, src2.y, src2.z);
-
-        uchar4 data_0 = *((__global uchar4 *)(dst + dst_index + 0));
-        uchar4 data_1 = *((__global uchar4 *)(dst + dst_index + 4));
-        uchar4 data_2 = *((__global uchar4 *)(dst + dst_index + 8));
-
-        uchar4 tmp_data_0 = src1_data_0 ^ src2_data_0;
-        uchar4 tmp_data_1 = src1_data_1 ^ src2_data_1;
-        uchar4 tmp_data_2 = src1_data_2 ^ src2_data_2;
-
-        data_0.xyz = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global uchar4 *)(dst + dst_index + 0)) = data_0;
-        *((__global uchar4 *)(dst + dst_index + 4)) = data_1;
-        *((__global uchar4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_xor_C3_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        char4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        char4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        char4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        char4 src2_data_0 = (char4)(src2.x, src2.y, src2.z, src2.x);
-        char4 src2_data_1 = (char4)(src2.y, src2.z, src2.x, src2.y);
-        char4 src2_data_2 = (char4)(src2.z, src2.x, src2.y, src2.z);
-
-        char4 data_0 = *((__global char4 *)(dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)(dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)(dst + dst_index + 8));
-
-        char4 tmp_data_0 = src1_data_0 ^ src2_data_0;
-        char4 tmp_data_1 = src1_data_1 ^ src2_data_1;
-        char4 tmp_data_2 = src1_data_2 ^ src2_data_2;
-
-        data_0.xyz = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global char4 *)(dst + dst_index + 0)) = data_0;
-        *((__global char4 *)(dst + dst_index + 4)) = data_1;
-        *((__global char4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-__kernel void arithm_s_bitwise_xor_C3_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        ushort2 src1_data_0 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 0));
-        ushort2 src1_data_1 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 4));
-        ushort2 src1_data_2 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 8));
-
-        ushort2 src2_data_0 = (ushort2)(src2.x, src2.y);
-        ushort2 src2_data_1 = (ushort2)(src2.z, src2.x);
-        ushort2 src2_data_2 = (ushort2)(src2.y, src2.z);
-
-        ushort2 data_0 = *((__global ushort2 *)((__global char *)dst + dst_index + 0));
-        ushort2 data_1 = *((__global ushort2 *)((__global char *)dst + dst_index + 4));
-        ushort2 data_2 = *((__global ushort2 *)((__global char *)dst + dst_index + 8));
-
-        ushort2 tmp_data_0 = src1_data_0 ^ src2_data_0;
-        ushort2 tmp_data_1 = src1_data_1 ^ src2_data_1;
-        ushort2 tmp_data_2 = src1_data_2 ^ src2_data_2;
-
-        data_0.xy = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global ushort2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_xor_C3_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        short2 src1_data_0 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 0));
-        short2 src1_data_1 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 4));
-        short2 src1_data_2 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 8));
-
-        short2 src2_data_0 = (short2)(src2.x, src2.y);
-        short2 src2_data_1 = (short2)(src2.z, src2.x);
-        short2 src2_data_2 = (short2)(src2.y, src2.z);
-
-        short2 data_0 = *((__global short2 *)((__global char *)dst + dst_index + 0));
-        short2 data_1 = *((__global short2 *)((__global char *)dst + dst_index + 4));
-        short2 data_2 = *((__global short2 *)((__global char *)dst + dst_index + 8));
-
-        short2 tmp_data_0 = src1_data_0 ^ src2_data_0;
-        short2 tmp_data_1 = src1_data_1 ^ src2_data_1;
-        short2 tmp_data_2 = src1_data_2 ^ src2_data_2;
-
-        data_0.xy = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global short2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global short2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global short2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_xor_C3_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        int src1_data_0 = *((__global int *)((__global char *)src1 + src1_index + 0));
-        int src1_data_1 = *((__global int *)((__global char *)src1 + src1_index + 4));
-        int src1_data_2 = *((__global int *)((__global char *)src1 + src1_index + 8));
-
-        int src2_data_0 = src2.x;
-        int src2_data_1 = src2.y;
-        int src2_data_2 = src2.z;
-
-        int data_0 = *((__global int *)((__global char *)dst + dst_index + 0));
-        int data_1 = *((__global int *)((__global char *)dst + dst_index + 4));
-        int data_2 = *((__global int *)((__global char *)dst + dst_index + 8));
-
-        int tmp_data_0 = src1_data_0 ^ src2_data_0;
-        int tmp_data_1 = src1_data_1 ^ src2_data_1;
-        int tmp_data_2 = src1_data_2 ^ src2_data_2;
-
-        *((__global int *)((__global char *)dst + dst_index + 0))= tmp_data_0;
-        *((__global int *)((__global char *)dst + dst_index + 4))= tmp_data_1;
-        *((__global int *)((__global char *)dst + dst_index + 8))= tmp_data_2;
-    }
-}
-__kernel void arithm_s_bitwise_xor_C3_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
 
-        char4 src1_data_0 = *((__global char4 *)((__global char *)src1 + src1_index + 0));
-        char4 src1_data_1 = *((__global char4 *)((__global char *)src1 + src1_index + 4));
-        char4 src1_data_2 = *((__global char4 *)((__global char *)src1 + src1_index + 8));
-
-        char4 src2_data_0 = (char4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        char4 src2_data_1 = (char4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        char4 src2_data_2 = (char4)(src2.s8, src2.s9, src2.sA, src2.sB);
-
-        char4 data_0 = *((__global char4 *)((__global char *)dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)((__global char *)dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)((__global char *)dst + dst_index + 8));
-
-        char4 tmp_data_0 = src1_data_0 ^ src2_data_0;
-        char4 tmp_data_1 = src1_data_1 ^ src2_data_1;
-        char4 tmp_data_2 = src1_data_2 ^ src2_data_2;
-
-        *((__global char4 *)((__global char *)dst + dst_index + 0))= tmp_data_0;
-        *((__global char4 *)((__global char *)dst + dst_index + 4))= tmp_data_1;
-        *((__global char4 *)((__global char *)dst + dst_index + 8))= tmp_data_2;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_xor_C3_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 24) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 24));
-
-        short4 src1_data_0 = *((__global short4 *)((__global char *)src1 + src1_index + 0 ));
-        short4 src1_data_1 = *((__global short4 *)((__global char *)src1 + src1_index + 8 ));
-        short4 src1_data_2 = *((__global short4 *)((__global char *)src1 + src1_index + 16));
-
-        short4 src2_data_0 = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        short4 src2_data_1 = (short4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        short4 src2_data_2 = (short4)(src2.s8, src2.s9, src2.sa, src2.sb);
-
-        short4 data_0 = *((__global short4 *)((__global char *)dst + dst_index + 0 ));
-        short4 data_1 = *((__global short4 *)((__global char *)dst + dst_index + 8 ));
-        short4 data_2 = *((__global short4 *)((__global char *)dst + dst_index + 16));
-
-        short4 tmp_data_0 = src1_data_0 ^ src2_data_0;
-        short4 tmp_data_1 = src1_data_1 ^ src2_data_1;
-        short4 tmp_data_2 = src1_data_2 ^ src2_data_2;
-
-        *((__global short4 *)((__global char *)dst + dst_index + 0 ))= tmp_data_0;
-        *((__global short4 *)((__global char *)dst + dst_index + 8 ))= tmp_data_1;
-        *((__global short4 *)((__global char *)dst + dst_index + 16))= tmp_data_2;
-    }
-}
-#endif
-__kernel void arithm_s_bitwise_xor_C4_D0 (
+__kernel void arithm_s_bitwise_binary_C4_D0 (
         __global   uchar *src1, int src1_step, int src1_offset,
         __global   uchar *dst,  int dst_step,  int dst_offset,
         uchar4 src2, int rows, int cols, int dst_step1)
@@ -811,14 +483,14 @@ __kernel void arithm_s_bitwise_xor_C4_D0 (
 
         uchar4 src_data1 = *((__global uchar4 *)(src1 + src1_index));
 
-        uchar4 data = src_data1 ^ src2;
+        uchar4 data = src_data1 OP_BINARY src2;
 
         *((__global uchar4 *)(dst + dst_index)) = data;
     }
 }
 
 
-__kernel void arithm_s_bitwise_xor_C4_D1 (
+__kernel void arithm_s_bitwise_binary_C4_D1 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         char4 src2, int rows, int cols, int dst_step1)
@@ -834,13 +506,13 @@ __kernel void arithm_s_bitwise_xor_C4_D1 (
 
         char4 src_data1 = *((__global char4 *)(src1 + src1_index));
 
-        char4 data = src_data1 ^ src2;
+        char4 data = src_data1 OP_BINARY src2;
 
         *((__global char4 *)(dst + dst_index)) = data;
     }
 }
 
-__kernel void arithm_s_bitwise_xor_C4_D2 (
+__kernel void arithm_s_bitwise_binary_C4_D2 (
         __global   ushort *src1, int src1_step, int src1_offset,
         __global   ushort *dst,  int dst_step,  int dst_offset,
         ushort4 src2, int rows, int cols, int dst_step1)
@@ -856,12 +528,12 @@ __kernel void arithm_s_bitwise_xor_C4_D2 (
 
         ushort4 src_data1 = *((__global ushort4 *)((__global char *)src1 + src1_index));
 
-        ushort4 data = src_data1 ^ src2;
+        ushort4 data = src_data1 OP_BINARY src2;
 
         *((__global ushort4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C4_D3 (
+__kernel void arithm_s_bitwise_binary_C4_D3 (
         __global   short *src1, int src1_step, int src1_offset,
         __global   short *dst,  int dst_step,  int dst_offset,
         short4 src2, int rows, int cols, int dst_step1)
@@ -877,12 +549,12 @@ __kernel void arithm_s_bitwise_xor_C4_D3 (
 
         short4 src_data1 = *((__global short4 *)((__global char *)src1 + src1_index));
 
-        short4 data = src_data1 ^ src2;
+        short4 data = src_data1 OP_BINARY src2;
 
         *((__global short4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C4_D4 (
+__kernel void arithm_s_bitwise_binary_C4_D4 (
         __global   int *src1, int src1_step, int src1_offset,
         __global   int *dst,  int dst_step,  int dst_offset,
         int4 src2, int rows, int cols, int dst_step1)
@@ -898,12 +570,12 @@ __kernel void arithm_s_bitwise_xor_C4_D4 (
 
         int4 src_data1 = *((__global int4 *)((__global char *)src1 + src1_index));
 
-        int4 data = src_data1 ^ src2;
+        int4 data = src_data1 OP_BINARY src2;
 
         *((__global int4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_xor_C4_D5 (
+__kernel void arithm_s_bitwise_binary_C4_D5 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         char16 src2, int rows, int cols, int dst_step1)
@@ -921,13 +593,13 @@ __kernel void arithm_s_bitwise_xor_C4_D5 (
         char16 src2_data = (char16)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7,
                                     src2.s8, src2.s9, src2.sa, src2.sb, src2.sc, src2.sd, src2.se, src2.sf);
 
-        char16 tmp_data = src1_data ^ src2_data;
+        char16 tmp_data = src1_data OP_BINARY src2_data;
 
         *((__global char16 *)((__global char *)dst + dst_index)) = tmp_data;
     }
 }
 #if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_xor_C4_D6 (
+__kernel void arithm_s_bitwise_binary_C4_D6 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *dst,  int dst_step,  int dst_offset,
         short16 src2, int rows, int cols, int dst_step1)
@@ -951,10 +623,10 @@ __kernel void arithm_s_bitwise_xor_C4_D6 (
         short4 src2_data_2 = (short4)(src2.s8, src2.s9, src2.sa, src2.sb);
         short4 src2_data_3 = (short4)(src2.sc, src2.sd, src2.se, src2.sf);
 
-        short4 tmp_data_0 = src1_data_0 ^ src2_data_0;
-        short4 tmp_data_1 = src1_data_1 ^ src2_data_1;
-        short4 tmp_data_2 = src1_data_2 ^ src2_data_2;
-        short4 tmp_data_3 = src1_data_3 ^ src2_data_3;
+        short4 tmp_data_0 = src1_data_0 OP_BINARY src2_data_0;
+        short4 tmp_data_1 = src1_data_1 OP_BINARY src2_data_1;
+        short4 tmp_data_2 = src1_data_2 OP_BINARY src2_data_2;
+        short4 tmp_data_3 = src1_data_3 OP_BINARY src2_data_3;
 
         *((__global short4 *)((__global char *)dst + dst_index + 0 ))= tmp_data_0;
         *((__global short4 *)((__global char *)dst + dst_index + 8 ))= tmp_data_1;
@@ -963,4 +635,4 @@ __kernel void arithm_s_bitwise_xor_C4_D6 (
 
     }
 }
-#endif
+#endif

modules/ocl/src/opencl/arithm_bitwise_binary_scalar_mask.cl

@@ -49,11 +49,16 @@
 #pragma OPENCL EXTENSION cl_amd_fp64:enable
 #endif
 #endif
+
+#ifndef OP_BINARY
+#define OP_BINARY &
+#endif
+
+//////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////bitwise_binary////////////////////////////////////////////
 //////////////////////////////////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////BITWISE_AND////////////////////////////////////////////////////
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-/**************************************bitwise_and with scalar with mask**************************************/
-__kernel void arithm_s_bitwise_and_with_mask_C1_D0 (
+/**************************************bitwise_binary with scalar with mask**************************************/
+__kernel void arithm_s_bitwise_binary_with_mask_C1_D0 (
         __global   uchar *src1, int src1_step, int src1_offset,
         __global   uchar *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -83,7 +88,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D0 (
         uchar4 mask_data = vload4(0, mask + mask_index);
 
         uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data & src2_data;
+        uchar4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = ((mask_data.y) && (dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
@@ -95,7 +100,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D0 (
 }
 
 
-__kernel void arithm_s_bitwise_and_with_mask_C1_D1 (
+__kernel void arithm_s_bitwise_binary_with_mask_C1_D1 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -125,7 +130,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D1 (
         uchar4 mask_data = vload4(0, mask + mask_index);
 
         char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data & src2_data;
+        char4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
         data.y = ((mask_data.y) && (dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
@@ -136,7 +141,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D1 (
     }
 }
 
-__kernel void arithm_s_bitwise_and_with_mask_C1_D2 (
+__kernel void arithm_s_bitwise_binary_with_mask_C1_D2 (
         __global   ushort *src1, int src1_step, int src1_offset,
         __global   ushort *dst,  int dst_step,  int dst_offset,
         __global   uchar  *mask, int mask_step, int mask_offset,
@@ -153,7 +158,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D2 (
 #ifdef dst_align
 #undef dst_align
 #endif
-#define dst_align ((dst_offset >> 1) & 1)
+#define dst_align ((dst_offset / 2) & 1)
         int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
         int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
 
@@ -166,7 +171,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D2 (
         uchar2  mask_data = vload2(0, mask + mask_index);
 
         ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index));
-        ushort2 tmp_data = src1_data & src2_data;
+        ushort2 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.x : data.x;
         data.y = ((mask_data.y) && (dst_index + 2 <  dst_end  )) ? tmp_data.y : data.y;
@@ -174,7 +179,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D2 (
         *((__global ushort2 *)((__global uchar *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_and_with_mask_C1_D3 (
+__kernel void arithm_s_bitwise_binary_with_mask_C1_D3 (
         __global   short *src1, int src1_step, int src1_offset,
         __global   short *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -191,7 +196,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D3 (
 #ifdef dst_align
 #undef dst_align
 #endif
-#define dst_align ((dst_offset >> 1) & 1)
+#define dst_align ((dst_offset / 2) & 1)
         int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
         int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
 
@@ -204,7 +209,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D3 (
         uchar2  mask_data = vload2(0, mask + mask_index);
 
         short2 data = *((__global short2 *)((__global uchar *)dst + dst_index));
-        short2 tmp_data = src1_data & src2_data;
+        short2 tmp_data = src1_data OP_BINARY src2_data;
 
         data.x = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.x : data.x;
         data.y = ((mask_data.y) && (dst_index + 2 <  dst_end  )) ? tmp_data.y : data.y;
@@ -212,7 +217,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D3 (
         *((__global short2 *)((__global uchar *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_and_with_mask_C1_D4 (
+__kernel void arithm_s_bitwise_binary_with_mask_C1_D4 (
         __global   int   *src1, int src1_step, int src1_offset,
         __global   int   *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -234,14 +239,14 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D4 (
         int src_data2 = src2.x;
         int dst_data  = *((__global int *)((__global char *)dst  + dst_index));
 
-        int data = src_data1 & src_data2;
+        int data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global int *)((__global char *)dst + dst_index)) = data;
     }
 }
 
-__kernel void arithm_s_bitwise_and_with_mask_C1_D5 (
+__kernel void arithm_s_bitwise_binary_with_mask_C1_D5 (
         __global char *src1, int src1_step, int src1_offset,
         __global char *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -263,7 +268,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D5 (
         char4 src2_data = (char4)(src2.s0, src2.s1, src2.s2, src2.s3);
         char4 dst_data  = *((__global char4 *)((__global char *)dst  + dst_index));
 
-        char4 data = src1_data & src2_data;
+        char4 data = src1_data OP_BINARY src2_data;
         data = mask_data ? data : dst_data;
 
         *((__global char4 *)((__global char *)dst + dst_index)) = data;
@@ -271,7 +276,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D5 (
 }
 
 #if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_and_with_mask_C1_D6 (
+__kernel void arithm_s_bitwise_binary_with_mask_C1_D6 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -293,14 +298,14 @@ __kernel void arithm_s_bitwise_and_with_mask_C1_D6 (
         short4 src2_data = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
         short4 dst_data  = *((__global short4 *)((__global char *)dst  + dst_index));
 
-        short4 data = src1_data & src2_data;
+        short4 data = src1_data OP_BINARY src2_data;
         data = mask_data ? data : dst_data;
 
         *((__global short4 *)((__global char *)dst + dst_index)) = data;
     }
 }
 #endif
-__kernel void arithm_s_bitwise_and_with_mask_C2_D0 (
+__kernel void arithm_s_bitwise_binary_with_mask_C2_D0 (
         __global   uchar *src1, int src1_step, int src1_offset,
         __global   uchar *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -317,7 +322,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D0 (
 #ifdef dst_align
 #undef dst_align
 #endif
-#define dst_align ((dst_offset >> 1) & 1)
+#define dst_align ((dst_offset / 2) & 1)
         int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
         int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
 
@@ -330,7 +335,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D0 (
         uchar2 mask_data = vload2(0, mask + mask_index);
 
         uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data & src2_data;
+        uchar4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.xy : data.xy;
         data.zw = ((mask_data.y) && (dst_index + 2 <  dst_end  )) ? tmp_data.zw : data.zw;
@@ -340,7 +345,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D0 (
 }
 
 
-__kernel void arithm_s_bitwise_and_with_mask_C2_D1 (
+__kernel void arithm_s_bitwise_binary_with_mask_C2_D1 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -357,7 +362,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D1 (
 #ifdef dst_align
 #undef dst_align
 #endif
-#define dst_align ((dst_offset >> 1) & 1)
+#define dst_align ((dst_offset / 2) & 1)
         int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
         int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
 
@@ -370,7 +375,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D1 (
         uchar2 mask_data = vload2(0, mask + mask_index);
 
         char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data & src2_data;
+        char4 tmp_data = src1_data OP_BINARY src2_data;
 
         data.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data.xy : data.xy;
         data.zw = ((mask_data.y) && (dst_index + 2 <  dst_end  )) ? tmp_data.zw : data.zw;
@@ -379,7 +384,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D1 (
     }
 }
 
-__kernel void arithm_s_bitwise_and_with_mask_C2_D2 (
+__kernel void arithm_s_bitwise_binary_with_mask_C2_D2 (
         __global   ushort *src1, int src1_step, int src1_offset,
         __global   ushort *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -401,13 +406,13 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D2 (
         ushort2 src_data2 = (ushort2)(src2.x, src2.y);
         ushort2 dst_data  = *((__global ushort2 *)((__global char *)dst  + dst_index));
 
-        ushort2 data = src_data1 & src_data2;
+        ushort2 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global ushort2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_and_with_mask_C2_D3 (
+__kernel void arithm_s_bitwise_binary_with_mask_C2_D3 (
         __global   short *src1, int src1_step, int src1_offset,
         __global   short *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -429,13 +434,13 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D3 (
         short2 src_data2 = (short2)(src2.x, src2.y);
         short2 dst_data  = *((__global short2 *)((__global char *)dst  + dst_index));
 
-        short2 data = src_data1 & src_data2;
+        short2 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global short2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_and_with_mask_C2_D4 (
+__kernel void arithm_s_bitwise_binary_with_mask_C2_D4 (
         __global   int *src1, int src1_step, int src1_offset,
         __global   int *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -457,13 +462,13 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D4 (
         int2 src_data2 = (int2)(src2.x, src2.y);
         int2 dst_data  = *((__global int2 *)((__global char *)dst  + dst_index));
 
-        int2 data = src_data1 & src_data2;
+        int2 data = src_data1 OP_BINARY src_data2;
         data = mask_data ? data : dst_data;
 
         *((__global int2 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_and_with_mask_C2_D5 (
+__kernel void arithm_s_bitwise_binary_with_mask_C2_D5 (
         __global   char *src1, int src1_step, int src1_offset,
         __global  char *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -485,7 +490,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D5 (
         char8 src2_data = (char8)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7);
         char8 dst_data = *((__global char8 *)((__global char *)dst  + dst_index));
 
-        char8 data = src1_data & src2_data;
+        char8 data = src1_data OP_BINARY src2_data;
 
         data = mask_data ? data : dst_data;
 
@@ -493,7 +498,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D5 (
     }
 }
 #if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_and_with_mask_C2_D6 (
+__kernel void arithm_s_bitwise_binary_with_mask_C2_D6 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *dst,  int dst_step,  int dst_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -515,388 +520,14 @@ __kernel void arithm_s_bitwise_and_with_mask_C2_D6 (
         short8 src2_data = (short8)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7);
         short8 dst_data = *((__global short8 *)((__global char *)dst  + dst_index));
 
-        short8 data = src1_data & src2_data;
+        short8 data = src1_data OP_BINARY src2_data;
         data = mask_data ? data : dst_data;
 
         *((__global short8 *)((__global char *)dst + dst_index)) = data;
     }
 }
 #endif
-__kernel void arithm_s_bitwise_and_with_mask_C3_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        __global   uchar *mask, int mask_step, int mask_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-        int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        uchar4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        uchar4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        uchar4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        uchar4 src2_data_0 = (uchar4)(src2.x, src2.y, src2.z, src2.x);
-        uchar4 src2_data_1 = (uchar4)(src2.y, src2.z, src2.x, src2.y);
-        uchar4 src2_data_2 = (uchar4)(src2.z, src2.x, src2.y, src2.z);
-
-        uchar4 mask_data = vload4(0, mask + mask_index);
-
-        uchar4 data_0 = *((__global uchar4 *)(dst + dst_index + 0));
-        uchar4 data_1 = *((__global uchar4 *)(dst + dst_index + 4));
-        uchar4 data_2 = *((__global uchar4 *)(dst + dst_index + 8));
-
-        uchar4 tmp_data_0 = src1_data_0 & src2_data_0;
-        uchar4 tmp_data_1 = src1_data_1 & src2_data_1;
-        uchar4 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0.xyz = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((mask_data.w) && (dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global uchar4 *)(dst + dst_index + 0)) = data_0;
-        *((__global uchar4 *)(dst + dst_index + 4)) = data_1;
-        *((__global uchar4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_and_with_mask_C3_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        __global   uchar *mask, int mask_step, int mask_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-        int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        char4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        char4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        char4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        char4 src2_data_0 = (char4)(src2.x, src2.y, src2.z, src2.x);
-        char4 src2_data_1 = (char4)(src2.y, src2.z, src2.x, src2.y);
-        char4 src2_data_2 = (char4)(src2.z, src2.x, src2.y, src2.z);
-
-        uchar4 mask_data = vload4(0, mask + mask_index);
-
-        char4 data_0 = *((__global char4 *)(dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)(dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)(dst + dst_index + 8));
-
-        char4 tmp_data_0 = src1_data_0 & src2_data_0;
-        char4 tmp_data_1 = src1_data_1 & src2_data_1;
-        char4 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0.xyz = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((mask_data.y) && (dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((mask_data.z) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((mask_data.w) && (dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global char4 *)(dst + dst_index + 0)) = data_0;
-        *((__global char4 *)(dst + dst_index + 4)) = data_1;
-        *((__global char4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-__kernel void arithm_s_bitwise_and_with_mask_C3_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        __global   uchar *mask, int mask_step, int mask_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-        int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        ushort2 src1_data_0 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 0));
-        ushort2 src1_data_1 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 4));
-        ushort2 src1_data_2 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 8));
-
-        ushort2 src2_data_0 = (ushort2)(src2.x, src2.y);
-        ushort2 src2_data_1 = (ushort2)(src2.z, src2.x);
-        ushort2 src2_data_2 = (ushort2)(src2.y, src2.z);
-
-        uchar2 mask_data = vload2(0, mask + mask_index);
-
-        ushort2 data_0 = *((__global ushort2 *)((__global char *)dst + dst_index + 0));
-        ushort2 data_1 = *((__global ushort2 *)((__global char *)dst + dst_index + 4));
-        ushort2 data_2 = *((__global ushort2 *)((__global char *)dst + dst_index + 8));
-
-        ushort2 tmp_data_0 = src1_data_0 & src2_data_0;
-        ushort2 tmp_data_1 = src1_data_1 & src2_data_1;
-        ushort2 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global ushort2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_and_with_mask_C3_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        __global   uchar *mask, int mask_step, int mask_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-        int mask_index = mad24(y, mask_step, x + mask_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        short2 src1_data_0 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 0));
-        short2 src1_data_1 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 4));
-        short2 src1_data_2 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 8));
-
-        short2 src2_data_0 = (short2)(src2.x, src2.y);
-        short2 src2_data_1 = (short2)(src2.z, src2.x);
-        short2 src2_data_2 = (short2)(src2.y, src2.z);
-
-        uchar2 mask_data = vload2(0, mask + mask_index);
-
-        short2 data_0 = *((__global short2 *)((__global char *)dst + dst_index + 0));
-        short2 data_1 = *((__global short2 *)((__global char *)dst + dst_index + 4));
-        short2 data_2 = *((__global short2 *)((__global char *)dst + dst_index + 8));
-
-        short2 tmp_data_0 = src1_data_0 & src2_data_0;
-        short2 tmp_data_1 = src1_data_1 & src2_data_1;
-        short2 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0.xy = ((mask_data.x) && (dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((mask_data.x) && (dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((mask_data.y) && (dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global short2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global short2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global short2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_and_with_mask_C3_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        __global   uchar *mask, int mask_step, int mask_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int mask_index = mad24(y, mask_step, x + mask_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        int src1_data_0 = *((__global int *)((__global char *)src1 + src1_index + 0));
-        int src1_data_1 = *((__global int *)((__global char *)src1 + src1_index + 4));
-        int src1_data_2 = *((__global int *)((__global char *)src1 + src1_index + 8));
-
-        int src2_data_0 = src2.x;
-        int src2_data_1 = src2.y;
-        int src2_data_2 = src2.z;
-
-        uchar mask_data = * (mask + mask_index);
-
-        int data_0 = *((__global int *)((__global char *)dst + dst_index + 0));
-        int data_1 = *((__global int *)((__global char *)dst + dst_index + 4));
-        int data_2 = *((__global int *)((__global char *)dst + dst_index + 8));
-
-        int tmp_data_0 = src1_data_0 & src2_data_0;
-        int tmp_data_1 = src1_data_1 & src2_data_1;
-        int tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0 = mask_data ? tmp_data_0 : data_0;
-        data_1 = mask_data ? tmp_data_1 : data_1;
-        data_2 = mask_data ? tmp_data_2 : data_2;
-
-        *((__global int *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global int *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global int *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_and_with_mask_C3_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        __global   uchar *mask, int mask_step, int mask_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int mask_index = mad24(y, mask_step, x + mask_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        char4 src1_data_0 = *((__global char4 *)((__global char *)src1 + src1_index + 0));
-        char4 src1_data_1 = *((__global char4 *)((__global char *)src1 + src1_index + 4));
-        char4 src1_data_2 = *((__global char4 *)((__global char *)src1 + src1_index + 8));
-
-        char4 src2_data_0 = (char4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        char4 src2_data_1 = (char4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        char4 src2_data_2 = (char4)(src2.s8, src2.s9, src2.sA, src2.sB);
-
-        uchar mask_data = * (mask + mask_index);
-
-        char4 data_0 = *((__global char4 *)((__global char *)dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)((__global char *)dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)((__global char *)dst + dst_index + 8));
-
-        char4 tmp_data_0 = src1_data_0 & src2_data_0;
-        char4 tmp_data_1 = src1_data_1 & src2_data_1;
-        char4 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0 = mask_data ? tmp_data_0 : data_0;
-        data_1 = mask_data ? tmp_data_1 : data_1;
-        data_2 = mask_data ? tmp_data_2 : data_2;
-
-        *((__global char4 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global char4 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global char4 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_and_with_mask_C3_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        __global uchar  *mask, int mask_step, int mask_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 24) + src1_offset);
-        int mask_index = mad24(y, mask_step, x + mask_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 24));
-
-        short4 src1_data_0 = *((__global short4 *)((__global char *)src1 + src1_index + 0 ));
-        short4 src1_data_1 = *((__global short4 *)((__global char *)src1 + src1_index + 8 ));
-        short4 src1_data_2 = *((__global short4 *)((__global char *)src1 + src1_index + 16));
-
-        short4 src2_data_0 = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        short4 src2_data_1 = (short4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        short4 src2_data_2 = (short4)(src2.s8, src2.s9, src2.sa, src2.sb);
-
-        uchar mask_data = * (mask + mask_index);
-
-        short4 data_0 = *((__global short4 *)((__global char *)dst + dst_index + 0 ));
-        short4 data_1 = *((__global short4 *)((__global char *)dst + dst_index + 8 ));
-        short4 data_2 = *((__global short4 *)((__global char *)dst + dst_index + 16));
-
-        short4 tmp_data_0 = src1_data_0 & src2_data_0;
-        short4 tmp_data_1 = src1_data_1 & src2_data_1;
-        short4 tmp_data_2 = src1_data_2 & src2_data_2;
-
-        data_0 = mask_data ? tmp_data_0 : data_0;
-        data_1 = mask_data ? tmp_data_1 : data_1;
-        data_2 = mask_data ? tmp_data_2 : data_2;
-
-        *((__global short4 *)((__global char *)dst + dst_index + 0 ))= data_0;
-        *((__global short4 *)((__global char *)dst + dst_index + 8 ))= data_1;
-        *((__global short4 *)((__global char *)dst + dst_index + 16))= data_2;
-    }
-}
-#endif
-__kernel void arithm_s_bitwise_and_with_mask_C4_D0 (
+__kernel void arithm_s_bitwise_binary_with_mask_C4_D0 (
         __global   uchar *src1, int src1_step, int src1_offset,
         __global   uchar *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -917,7 +548,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C4_D0 (
         uchar4 src_data1 = *((__global uchar4 *)(src1 + src1_index));
         uchar4 dst_data  = *((__global uchar4 *)(dst  + dst_index));
 
-        uchar4 data = src_data1 & src2;
+        uchar4 data = src_data1 OP_BINARY src2;
         data = mask_data ? data : dst_data;
 
         *((__global uchar4 *)(dst + dst_index)) = data;
@@ -925,7 +556,7 @@ __kernel void arithm_s_bitwise_and_with_mask_C4_D0 (
 }
 
 
-__kernel void arithm_s_bitwise_and_with_mask_C4_D1 (
+__kernel void arithm_s_bitwise_binary_with_mask_C4_D1 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -946,14 +577,14 @@ __kernel void arithm_s_bitwise_and_with_mask_C4_D1 (
         char4 src_data1 = *((__global char4 *)(src1 + src1_index));
         char4 dst_data  = *((__global char4 *)(dst  + dst_index));
 
-        char4 data = src_data1 & src2;
+        char4 data = src_data1 OP_BINARY src2;
         data = mask_data ? data : dst_data;
 
         *((__global char4 *)(dst + dst_index)) = data;
     }
 }
 
-__kernel void arithm_s_bitwise_and_with_mask_C4_D2 (
+__kernel void arithm_s_bitwise_binary_with_mask_C4_D2 (
         __global   ushort *src1, int src1_step, int src1_offset,
         __global   ushort *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -974,13 +605,13 @@ __kernel void arithm_s_bitwise_and_with_mask_C4_D2 (
         ushort4 src_data1 = *((__global ushort4 *)((__global char *)src1 + src1_index));
         ushort4 dst_data  = *((__global ushort4 *)((__global char *)dst  + dst_index));
 
-        ushort4 data = src_data1 & src2;
+        ushort4 data = src_data1 OP_BINARY src2;
         data = mask_data ? data : dst_data;
 
         *((__global ushort4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_and_with_mask_C4_D3 (
+__kernel void arithm_s_bitwise_binary_with_mask_C4_D3 (
         __global   short *src1, int src1_step, int src1_offset,
         __global   short *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -1001,13 +632,13 @@ __kernel void arithm_s_bitwise_and_with_mask_C4_D3 (
         short4 src_data1 = *((__global short4 *)((__global char *)src1 + src1_index));
         short4 dst_data  = *((__global short4 *)((__global char *)dst  + dst_index));
 
-        short4 data = src_data1 & src2;
+        short4 data = src_data1 OP_BINARY src2;
         data = mask_data ? data : dst_data;
 
         *((__global short4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_and_with_mask_C4_D4 (
+__kernel void arithm_s_bitwise_binary_with_mask_C4_D4 (
         __global   int *src1, int src1_step, int src1_offset,
         __global   int *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -1028,13 +659,13 @@ __kernel void arithm_s_bitwise_and_with_mask_C4_D4 (
         int4 src_data1 = *((__global int4 *)((__global char *)src1 + src1_index));
         int4 dst_data  = *((__global int4 *)((__global char *)dst  + dst_index));
 
-        int4 data = src_data1 & src2;
+        int4 data = src_data1 OP_BINARY src2;
         data = mask_data ? data : dst_data;
 
         *((__global int4 *)((__global char *)dst + dst_index)) = data;
     }
 }
-__kernel void arithm_s_bitwise_and_with_mask_C4_D5 (
+__kernel void arithm_s_bitwise_binary_with_mask_C4_D5 (
         __global   char *src1, int src1_step, int src1_offset,
         __global   char *dst,  int dst_step,  int dst_offset,
         __global   uchar *mask, int mask_step, int mask_offset,
@@ -1057,14 +688,14 @@ __kernel void arithm_s_bitwise_and_with_mask_C4_D5 (
                                     src2.s8, src2.s9, src2.sa, src2.sb, src2.sc, src2.sd, src2.se, src2.sf);
         char16 dst_data  = *((__global char16 *)((__global char *)dst  + dst_index));
 
-        char16 data = src1_data & src2_data;
+        char16 data = src1_data OP_BINARY src2_data;
         data = mask_data ? data : dst_data;
 
         *((__global char16 *)((__global char *)dst + dst_index)) = data;
     }
 }
 #if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_and_with_mask_C4_D6 (
+__kernel void arithm_s_bitwise_binary_with_mask_C4_D6 (
         __global short *src1, int src1_step, int src1_offset,
         __global short *dst,  int dst_step,  int dst_offset,
         __global uchar *mask, int mask_step, int mask_offset,
@@ -1097,10 +728,10 @@ __kernel void arithm_s_bitwise_and_with_mask_C4_D6 (
         short4 dst_data_2  = *((__global short4 *)((__global char *)dst  + dst_index + 16));
         short4 dst_data_3  = *((__global short4 *)((__global char *)dst  + dst_index + 24));
 
-        short4 data_0 = src1_data_0 & src2_data_0;
-        short4 data_1 = src1_data_1 & src2_data_1;
-        short4 data_2 = src1_data_2 & src2_data_2;
-        short4 data_3 = src1_data_3 & src2_data_3;
+        short4 data_0 = src1_data_0 OP_BINARY src2_data_0;
+        short4 data_1 = src1_data_1 OP_BINARY src2_data_1;
+        short4 data_2 = src1_data_2 OP_BINARY src2_data_2;
+        short4 data_3 = src1_data_3 OP_BINARY src2_data_3;
 
         data_0 = mask_data ? data_0 : dst_data_0;
         data_1 = mask_data ? data_1 : dst_data_1;

modules/ocl/src/opencl/arithm_bitwise_or.cl

@@ -1,294 +0,0 @@
-/*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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
-// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
-// Third party copyrights are property of their respective owners.
-//
-// @Authors
-//    Jiang Liyuan, jlyuan001.good@163.com
-//
-// 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 oclMaterials 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*/
-#if defined (DOUBLE_SUPPORT)
-#ifdef cl_khr_fp64
-#pragma OPENCL EXTENSION cl_khr_fp64:enable
-#elif defined (cl_amd_fp64)
-#pragma OPENCL EXTENSION cl_amd_fp64:enable
-#endif
-#endif
-
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////BITWISE_OR////////////////////////////////////////////////////
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-/**************************************bitwise_or without mask**************************************/
-__kernel void arithm_bitwise_or_D0 (__global uchar *src1, int src1_step, int src1_offset,
-                                    __global uchar *src2, int src2_step, int src2_offset,
-                                    __global uchar *dst,  int dst_step,  int dst_offset,
-                                    int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (dst_offset & 3)
-        int src1_index = mad24(y, src1_step, x + src1_offset - dst_align);
-        int src2_index = mad24(y, src2_step, x + src2_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + x & (int)0xfffffffc);
-        int src1_index_fix = src1_index < 0 ? 0 : src1_index;
-        int src2_index_fix = src2_index < 0 ? 0 : src2_index;
-        uchar4 src1_data = vload4(0, src1 + src1_index_fix);
-        uchar4 src2_data = vload4(0, src2 + src2_index_fix);
-        if(src1_index < 0)
-        {
-            uchar4 tmp;
-            tmp.xyzw = (src1_index == -2) ? src1_data.zwxy:src1_data.yzwx;
-            src1_data.xyzw = (src1_index == -1) ? src1_data.wxyz:tmp.xyzw;
-        }
-        if(src2_index < 0)
-        {
-            uchar4 tmp;
-            tmp.xyzw = (src2_index == -2) ? src2_data.zwxy:src2_data.yzwx;
-            src2_data.xyzw = (src2_index == -1) ? src2_data.wxyz:tmp.xyzw;
-        }
-        uchar4 dst_data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data | src2_data;
-
-        dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
-        dst_data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : dst_data.y;
-        dst_data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : dst_data.z;
-        dst_data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : dst_data.w;
-
-        *((__global uchar4 *)(dst + dst_index)) = dst_data;
-    }
-}
-
-
-__kernel void arithm_bitwise_or_D1 (__global char *src1, int src1_step, int src1_offset,
-                                    __global char *src2, int src2_step, int src2_offset,
-                                    __global char *dst,  int dst_step,  int dst_offset,
-                                    int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (dst_offset & 3)
-        int src1_index = mad24(y, src1_step, x + src1_offset - dst_align);
-        int src2_index = mad24(y, src2_step, x + src2_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + x & (int)0xfffffffc);
-
-        char4 src1_data = vload4(0, src1 + src1_index);
-        char4 src2_data = vload4(0, src2 + src2_index);
-
-        char4 dst_data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data | src2_data;
-
-        dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
-        dst_data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : dst_data.y;
-        dst_data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : dst_data.z;
-        dst_data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : dst_data.w;
-
-        *((__global char4 *)(dst + dst_index)) = dst_data;
-    }
-}
-
-
-__kernel void arithm_bitwise_or_D2 (__global ushort *src1, int src1_step, int src1_offset,
-                                    __global ushort *src2, int src2_step, int src2_offset,
-                                    __global ushort *dst,  int dst_step,  int dst_offset,
-                                    int rows, int cols, int dst_step1)
-
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 3)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-        int src2_index = mad24(y, src2_step, (x << 1) + src2_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffff8);
-
-        ushort4 src1_data = vload4(0, (__global ushort *)((__global char *)src1 + src1_index));
-        ushort4 src2_data = vload4(0, (__global ushort *)((__global char *)src2 + src2_index));
-
-        ushort4 dst_data = *((__global ushort4 *)((__global char *)dst + dst_index));
-        ushort4 tmp_data = src1_data | src2_data;
-
-        dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
-        dst_data.y = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : dst_data.y;
-        dst_data.z = ((dst_index + 4 >= dst_start) && (dst_index + 4 < dst_end)) ? tmp_data.z : dst_data.z;
-        dst_data.w = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end)) ? tmp_data.w : dst_data.w;
-
-        *((__global ushort4 *)((__global char *)dst + dst_index)) = dst_data;
-    }
-}
-
-
-
-__kernel void arithm_bitwise_or_D3 (__global short *src1, int src1_step, int src1_offset,
-                                    __global short *src2, int src2_step, int src2_offset,
-                                    __global short *dst,  int dst_step,  int dst_offset,
-                                    int rows, int cols, int dst_step1)
-
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 3)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-        int src2_index = mad24(y, src2_step, (x << 1) + src2_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffff8);
-
-        short4 src1_data = vload4(0, (__global short *)((__global char *)src1 + src1_index));
-        short4 src2_data = vload4(0, (__global short *)((__global char *)src2 + src2_index));
-
-        short4 dst_data = *((__global short4 *)((__global char *)dst + dst_index));
-        short4 tmp_data = src1_data | src2_data;
-
-        dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
-        dst_data.y = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : dst_data.y;
-        dst_data.z = ((dst_index + 4 >= dst_start) && (dst_index + 4 < dst_end)) ? tmp_data.z : dst_data.z;
-        dst_data.w = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end)) ? tmp_data.w : dst_data.w;
-
-        *((__global short4 *)((__global char *)dst + dst_index)) = dst_data;
-    }
-}
-
-
-
-__kernel void arithm_bitwise_or_D4 (__global int *src1, int src1_step, int src1_offset,
-                                    __global int *src2, int src2_step, int src2_offset,
-                                    __global int *dst,  int dst_step,  int dst_offset,
-                                    int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x << 2) + src2_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        int data1 = *((__global int *)((__global char *)src1 + src1_index));
-        int data2 = *((__global int *)((__global char *)src2 + src2_index));
-        int tmp  = data1 | data2;
-
-        *((__global int *)((__global char *)dst + dst_index)) = tmp;
-    }
-}
-
-__kernel void arithm_bitwise_or_D5 (__global char *src1, int src1_step, int src1_offset,
-                                    __global char *src2, int src2_step, int src2_offset,
-                                    __global char *dst,  int dst_step,  int dst_offset,
-                                    int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x << 2) + src2_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        char4 data1 = *((__global char4 *)((__global char *)src1 + src1_index));
-        char4 data2 = *((__global char4 *)((__global char *)src2 + src2_index));
-        char4 tmp = data1 | data2;
-
-        *((__global char4 *)((__global char *)dst + dst_index)) = tmp;
-    }
-}
-
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_bitwise_or_D6 (__global char *src1, int src1_step, int src1_offset,
-                                    __global char *src2, int src2_step, int src2_offset,
-                                    __global char *dst,  int dst_step,  int dst_offset,
-                                    int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x << 3) + src2_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        char8 data1 = *((__global char8 *)((__global char *)src1 + src1_index));
-        char8 data2 = *((__global char8 *)((__global char *)src2 + src2_index));
-
-        *((__global char8 *)((__global char *)dst + dst_index)) = data1 | data2;
-    }
-}
-#endif

modules/ocl/src/opencl/arithm_bitwise_or_scalar.cl

@@ -1,973 +0,0 @@
-/*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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
-// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
-// Third party copyrights are property of their respective owners.
-//
-// @Authors
-//    Jiang Liyuan, jlyuan001.good@163.com
-//
-// 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 oclMaterials 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*/
-#if defined (DOUBLE_SUPPORT)
-#ifdef cl_khr_fp64
-#pragma OPENCL EXTENSION cl_khr_fp64:enable
-#elif defined (cl_amd_fp64)
-#pragma OPENCL EXTENSION cl_amd_fp64:enable
-#endif
-#endif
-
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////BITWISE_OR////////////////////////////////////////////////////
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-/**************************************and with scalar without mask**************************************/
-__kernel void arithm_s_bitwise_or_C1_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (dst_offset & 3)
-        int src1_index = mad24(y, src1_step, x + src1_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + x & (int)0xfffffffc);
-
-        uchar4 src1_data = vload4(0, src1 + src1_index);
-        uchar4 src2_data = (uchar4)(src2.x, src2.x, src2.x, src2.x);
-
-        uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data | src2_data;
-
-        data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
-        data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
-        data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : data.z;
-        data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : data.w;
-
-        *((__global uchar4 *)(dst + dst_index)) = data;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_or_C1_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (dst_offset & 3)
-        int src1_index = mad24(y, src1_step, x + src1_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + x & (int)0xfffffffc);
-
-        char4 src1_data = vload4(0, src1 + src1_index);
-        char4 src2_data = (char4)(src2.x, src2.x, src2.x, src2.x);
-
-        char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data | src2_data;
-
-        data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : data.x;
-        data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : data.y;
-        data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : data.z;
-        data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : data.w;
-
-        *((__global char4 *)(dst + dst_index)) = data;
-    }
-}
-
-__kernel void arithm_s_bitwise_or_C1_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 1)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffffc);
-
-        ushort2 src1_data = vload2(0, (__global ushort *)((__global char *)src1 + src1_index));
-        ushort2 src2_data = (ushort2)(src2.x, src2.x);
-
-        ushort2 data = *((__global ushort2 *)((__global uchar *)dst + dst_index));
-        ushort2 tmp_data = src1_data | src2_data;
-
-        data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x;
-        data.y = (dst_index + 2 <  dst_end  ) ? tmp_data.y : data.y;
-
-        *((__global ushort2 *)((__global uchar *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C1_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 1)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffffc);
-
-        short2 src1_data = vload2(0, (__global short *)((__global char *)src1 + src1_index));
-        short2 src2_data = (short2)(src2.x, src2.x);
-        short2 data = *((__global short2 *)((__global uchar *)dst + dst_index));
-
-        short2 tmp_data = src1_data | src2_data;
-
-        data.x = (dst_index + 0 >= dst_start) ? tmp_data.x : data.x;
-        data.y = (dst_index + 2 <  dst_end  ) ? tmp_data.y : data.y;
-
-        *((__global short2 *)((__global uchar *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C1_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        int src_data1 = *((__global int *)((__global char *)src1 + src1_index));
-        int src_data2 = src2.x;
-
-        int data = src_data1 | src_data2;
-
-        *((__global int *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C1_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        char4 src_data1 = *((__global char4 *)((__global char *)src1 + src1_index));
-        char4 src_data2 = (char4)(src2.s0, src2.s1, src2.s2, src2.s3);
-
-        char4 data = src_data1 | src_data2;
-
-        *((__global char4 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_or_C1_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        short4 src1_data = *((__global short4 *)((__global char *)src1 + src1_index));
-        short4 src2_data = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
-
-        short4 tmp_data = src1_data | src2_data;
-
-        *((__global short4 *)((__global char *)dst + dst_index)) = tmp_data;
-    }
-}
-#endif
-__kernel void arithm_s_bitwise_or_C2_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 1)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffffc);
-
-        uchar4 src1_data = vload4(0, src1 + src1_index);
-        uchar4 src2_data = (uchar4)(src2.x, src2.y, src2.x, src2.y);
-
-        uchar4 data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data | src2_data;
-
-        data.xy = (dst_index + 0 >= dst_start) ? tmp_data.xy : data.xy;
-        data.zw = (dst_index + 2 <  dst_end  ) ? tmp_data.zw : data.zw;
-
-        *((__global uchar4 *)(dst + dst_index)) = data;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_or_C2_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 1)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffffc);
-
-        char4 src1_data = vload4(0, src1 + src1_index);
-        char4 src2_data = (char4)(src2.x, src2.y, src2.x, src2.y);
-
-        char4 data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data | src2_data;
-
-        data.xy = (dst_index + 0 >= dst_start) ? tmp_data.xy : data.xy;
-        data.zw = (dst_index + 2 <  dst_end  ) ? tmp_data.zw : data.zw;
-
-        *((__global char4 *)(dst + dst_index)) = data;
-    }
-}
-
-__kernel void arithm_s_bitwise_or_C2_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        ushort2 src_data1 = *((__global ushort2 *)((__global char *)src1 + src1_index));
-        ushort2 src_data2 = (ushort2)(src2.x, src2.y);
-
-        ushort2 data = src_data1 | src_data2;
-
-        *((__global ushort2 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C2_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        short2 src_data1 = *((__global short2 *)((__global char *)src1 + src1_index));
-        short2 src_data2 = (short2)(src2.x, src2.y);
-
-        short2 data = src_data1 | src_data2;
-
-        *((__global short2 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C2_D4 (__global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        int2 src_data1 = *((__global int2 *)((__global char *)src1 + src1_index));
-        int2 src_data2 = (int2)(src2.x, src2.y);
-
-        int2 data = src_data1 | src_data2;
-        *((__global int2 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C2_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        char8 src_data1 = *((__global char8 *)((__global char *)src1 + src1_index));
-        char8 src_data2 = (char8)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7);
-
-        char8 data = src_data1 | src_data2;
-        *((__global char8 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_or_C2_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 4) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 4) + dst_offset);
-
-        short8 src1_data = *((__global short8 *)((__global char *)src1 + src1_index));
-        short8 src2_data = (short8)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7);
-
-        short8 tmp_data = src1_data & src2_data;
-
-        *((__global short8 *)((__global char *)dst + dst_index)) = tmp_data;
-    }
-}
-#endif
-__kernel void arithm_s_bitwise_or_C3_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        uchar4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        uchar4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        uchar4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        uchar4 src2_data_0 = (uchar4)(src2.x, src2.y, src2.z, src2.x);
-        uchar4 src2_data_1 = (uchar4)(src2.y, src2.z, src2.x, src2.y);
-        uchar4 src2_data_2 = (uchar4)(src2.z, src2.x, src2.y, src2.z);
-
-        uchar4 data_0 = *((__global uchar4 *)(dst + dst_index + 0));
-        uchar4 data_1 = *((__global uchar4 *)(dst + dst_index + 4));
-        uchar4 data_2 = *((__global uchar4 *)(dst + dst_index + 8));
-
-        uchar4 tmp_data_0 =  src1_data_0  |  src2_data_0  ;
-        uchar4 tmp_data_1 =  src1_data_1  |  src2_data_1  ;
-        uchar4 tmp_data_2 =  src1_data_2  |  src2_data_2  ;
-
-        data_0.xyz = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global uchar4 *)(dst + dst_index + 0)) = data_0;
-        *((__global uchar4 *)(dst + dst_index + 4)) = data_1;
-        *((__global uchar4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_or_C3_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 3 ) & 3)
-        int src1_index = mad24(y, src1_step, (x * 3) + src1_offset - (dst_align * 3));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 3) - (dst_align * 3));
-
-        char4 src1_data_0 = vload4(0, src1 + src1_index + 0);
-        char4 src1_data_1 = vload4(0, src1 + src1_index + 4);
-        char4 src1_data_2 = vload4(0, src1 + src1_index + 8);
-
-        char4 src2_data_0 = (char4)(src2.x, src2.y, src2.z, src2.x);
-        char4 src2_data_1 = (char4)(src2.y, src2.z, src2.x, src2.y);
-        char4 src2_data_2 = (char4)(src2.z, src2.x, src2.y, src2.z);
-
-        char4 data_0 = *((__global char4 *)(dst + dst_index + 0));
-        char4 data_1 = *((__global char4 *)(dst + dst_index + 4));
-        char4 data_2 = *((__global char4 *)(dst + dst_index + 8));
-
-        char4 tmp_data_0 =  src1_data_0  |  src2_data_0;
-        char4 tmp_data_1 =  src1_data_1  |  src2_data_1;
-        char4 tmp_data_2 =  src1_data_2  |  src2_data_2;
-
-        data_0.xyz = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xyz : data_0.xyz;
-        data_0.w   = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_0.w : data_0.w;
-
-        data_1.xy  = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end))
-                     ? tmp_data_1.xy : data_1.xy;
-        data_1.zw  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_1.zw : data_1.zw;
-
-        data_2.x   = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                     ? tmp_data_2.x : data_2.x;
-        data_2.yzw = ((dst_index + 9 >= dst_start) && (dst_index + 9 < dst_end))
-                     ? tmp_data_2.yzw : data_2.yzw;
-
-        *((__global char4 *)(dst + dst_index + 0)) = data_0;
-        *((__global char4 *)(dst + dst_index + 4)) = data_1;
-        *((__global char4 *)(dst + dst_index + 8)) = data_2;
-    }
-}
-
-__kernel void arithm_s_bitwise_or_C3_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        ushort2 src1_data_0 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 0));
-        ushort2 src1_data_1 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 4));
-        ushort2 src1_data_2 = vload2(0, (__global ushort *)((__global char *)src1 + src1_index + 8));
-
-        ushort2 src2_data_0 = (ushort2)(src2.x, src2.y);
-        ushort2 src2_data_1 = (ushort2)(src2.z, src2.x);
-        ushort2 src2_data_2 = (ushort2)(src2.y, src2.z);
-
-        ushort2 data_0 = *((__global ushort2 *)((__global char *)dst + dst_index + 0));
-        ushort2 data_1 = *((__global ushort2 *)((__global char *)dst + dst_index + 4));
-        ushort2 data_2 = *((__global ushort2 *)((__global char *)dst + dst_index + 8));
-
-        ushort2 tmp_data_0 = src1_data_0  |  src2_data_0  ;
-        ushort2 tmp_data_1 = src1_data_1  |  src2_data_1  ;
-        ushort2 tmp_data_2 = src1_data_2  |  src2_data_2  ;
-
-        data_0.xy = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global ushort2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global ushort2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_or_C3_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 1;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (((dst_offset % dst_step) / 6 ) & 1)
-        int src1_index = mad24(y, src1_step, (x * 6) + src1_offset - (dst_align * 6));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 6) - (dst_align * 6));
-
-        short2 src1_data_0 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 0));
-        short2 src1_data_1 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 4));
-        short2 src1_data_2 = vload2(0, (__global short *)((__global char *)src1 + src1_index + 8));
-
-        short2 src2_data_0 = (short2)(src2.x, src2.y);
-        short2 src2_data_1 = (short2)(src2.z, src2.x);
-        short2 src2_data_2 = (short2)(src2.y, src2.z);
-
-        short2 data_0 = *((__global short2 *)((__global char *)dst + dst_index + 0));
-        short2 data_1 = *((__global short2 *)((__global char *)dst + dst_index + 4));
-        short2 data_2 = *((__global short2 *)((__global char *)dst + dst_index + 8));
-
-        short2 tmp_data_0 =  src1_data_0  |  src2_data_0  ;
-        short2 tmp_data_1 =  src1_data_1  |  src2_data_1  ;
-        short2 tmp_data_2 =  src1_data_2  |  src2_data_2  ;
-
-        data_0.xy = ((dst_index + 0 >= dst_start)) ? tmp_data_0.xy : data_0.xy;
-
-        data_1.x  = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end))
-                    ? tmp_data_1.x : data_1.x;
-        data_1.y  = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_1.y : data_1.y;
-
-        data_2.xy = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end))
-                    ? tmp_data_2.xy : data_2.xy;
-
-        *((__global short2 *)((__global char *)dst + dst_index + 0))= data_0;
-        *((__global short2 *)((__global char *)dst + dst_index + 4))= data_1;
-        *((__global short2 *)((__global char *)dst + dst_index + 8))= data_2;
-    }
-}
-__kernel void arithm_s_bitwise_or_C3_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        int src1_data_0 = *((__global int *)((__global char *)src1 + src1_index + 0));
-        int src1_data_1 = *((__global int *)((__global char *)src1 + src1_index + 4));
-        int src1_data_2 = *((__global int *)((__global char *)src1 + src1_index + 8));
-
-        int src2_data_0 = src2.x;
-        int src2_data_1 = src2.y;
-        int src2_data_2 = src2.z;
-
-        int data_0 = *((__global int *)((__global char *)dst + dst_index + 0));
-        int data_1 = *((__global int *)((__global char *)dst + dst_index + 4));
-        int data_2 = *((__global int *)((__global char *)dst + dst_index + 8));
-
-        int tmp_data_0 = src1_data_0 | src2_data_0;
-        int tmp_data_1 = src1_data_1 | src2_data_1;
-        int tmp_data_2 = src1_data_2 | src2_data_2;
-
-        *((__global int *)((__global char *)dst + dst_index + 0))= tmp_data_0;
-        *((__global int *)((__global char *)dst + dst_index + 4))= tmp_data_1;
-        *((__global int *)((__global char *)dst + dst_index + 8))= tmp_data_2;
-    }
-}
-__kernel void arithm_s_bitwise_or_C3_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-
-    {
-        int src1_index = mad24(y, src1_step, (x * 12) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 12));
-
-        char4 src1_data_0 = *((__global char4 *)((__global char *)src1 + src1_index + 0));
-        char4 src1_data_1 = *((__global char4 *)((__global char *)src1 + src1_index + 4));
-        char4 src1_data_2 = *((__global char4 *)((__global char *)src1 + src1_index + 8));
-
-        char4 src2_data_0 = (char4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        char4 src2_data_1 = (char4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        char4 src2_data_2 = (char4)(src2.s8, src2.s9, src2.sA, src2.sB);
-
-        char4 tmp_data_0 = src1_data_0 | src2_data_0;
-        char4 tmp_data_1 = src1_data_1 | src2_data_1;
-        char4 tmp_data_2 = src1_data_2 | src2_data_2;
-
-        *((__global char4 *)((__global char *)dst + dst_index + 0))= tmp_data_0;
-        *((__global char4 *)((__global char *)dst + dst_index + 4))= tmp_data_1;
-        *((__global char4 *)((__global char *)dst + dst_index + 8))= tmp_data_2;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_or_C3_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x * 24) + src1_offset);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x * 24));
-
-        short4 src1_data_0 = *((__global short4 *)((__global char *)src1 + src1_index + 0 ));
-        short4 src1_data_1 = *((__global short4 *)((__global char *)src1 + src1_index + 8 ));
-        short4 src1_data_2 = *((__global short4 *)((__global char *)src1 + src1_index + 16));
-
-        short4 src2_data_0 = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        short4 src2_data_1 = (short4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        short4 src2_data_2 = (short4)(src2.s8, src2.s9, src2.sa, src2.sb);
-
-        short4 data_0 = *((__global short4 *)((__global char *)dst + dst_index + 0 ));
-        short4 data_1 = *((__global short4 *)((__global char *)dst + dst_index + 8 ));
-        short4 data_2 = *((__global short4 *)((__global char *)dst + dst_index + 16));
-
-        short4 tmp_data_0 = src1_data_0 | src2_data_0;
-        short4 tmp_data_1 = src1_data_1 | src2_data_1;
-        short4 tmp_data_2 = src1_data_2 | src2_data_2;
-
-        *((__global short4 *)((__global char *)dst + dst_index + 0 ))= tmp_data_0;
-        *((__global short4 *)((__global char *)dst + dst_index + 8 ))= tmp_data_1;
-        *((__global short4 *)((__global char *)dst + dst_index + 16))= tmp_data_2;
-    }
-}
-#endif
-__kernel void arithm_s_bitwise_or_C4_D0 (
-        __global   uchar *src1, int src1_step, int src1_offset,
-        __global   uchar *dst,  int dst_step,  int dst_offset,
-        uchar4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        uchar4 src_data1 = *((__global uchar4 *)(src1 + src1_index));
-
-        uchar4 data = src_data1 | src2;
-
-        *((__global uchar4 *)(dst + dst_index)) = data;
-    }
-}
-
-
-__kernel void arithm_s_bitwise_or_C4_D1 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        char4 src_data1 = *((__global char4 *)(src1 + src1_index));
-
-        char4 data = src_data1 | src2;
-
-        *((__global char4 *)(dst + dst_index)) = data;
-    }
-}
-
-__kernel void arithm_s_bitwise_or_C4_D2 (
-        __global   ushort *src1, int src1_step, int src1_offset,
-        __global   ushort *dst,  int dst_step,  int dst_offset,
-        ushort4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        ushort4 src_data1 = *((__global ushort4 *)((__global char *)src1 + src1_index));
-
-        ushort4 data = src_data1 | src2;
-
-        *((__global ushort4 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C4_D3 (
-        __global   short *src1, int src1_step, int src1_offset,
-        __global   short *dst,  int dst_step,  int dst_offset,
-        short4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        short4 src_data1 = *((__global short4 *)((__global char *)src1 + src1_index));
-
-        short4 data = src_data1 | src2;
-
-        *((__global short4 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C4_D4 (
-        __global   int *src1, int src1_step, int src1_offset,
-        __global   int *dst,  int dst_step,  int dst_offset,
-        int4 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 4) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 4) + dst_offset);
-
-        int4 src_data1 = *((__global int4 *)((__global char *)src1 + src1_index));
-
-        int4 data = src_data1 | src2;
-
-        *((__global int4 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-__kernel void arithm_s_bitwise_or_C4_D5 (
-        __global   char *src1, int src1_step, int src1_offset,
-        __global   char *dst,  int dst_step,  int dst_offset,
-        char16 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 4) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 4) + dst_offset);
-
-        char16 src_data1 = *((__global char16 *)((__global char *)src1 + src1_index));
-        char16 src_data2 = (char16)(src2.s0, src2.s1, src2.s2, src2.s3, src2.s4, src2.s5, src2.s6, src2.s7,
-                                    src2.s8, src2.s9, src2.sa, src2.sb, src2.sc, src2.sd, src2.se, src2.sf);
-
-        char16 data = src_data1 | src_data2;
-
-        *((__global char16 *)((__global char *)dst + dst_index)) = data;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_s_bitwise_or_C4_D6 (
-        __global short *src1, int src1_step, int src1_offset,
-        __global short *dst,  int dst_step,  int dst_offset,
-        short16 src2, int rows, int cols, int dst_step1)
-
-{
-
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 5) + src1_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 5) + dst_offset);
-
-        short4 src1_data_0 = *((__global short4 *)((__global char *)src1 + src1_index + 0));
-        short4 src1_data_1 = *((__global short4 *)((__global char *)src1 + src1_index + 8));
-        short4 src1_data_2 = *((__global short4 *)((__global char *)src1 + src1_index + 16));
-        short4 src1_data_3 = *((__global short4 *)((__global char *)src1 + src1_index + 24));
-
-        short4 src2_data_0 = (short4)(src2.s0, src2.s1, src2.s2, src2.s3);
-        short4 src2_data_1 = (short4)(src2.s4, src2.s5, src2.s6, src2.s7);
-        short4 src2_data_2 = (short4)(src2.s8, src2.s9, src2.sa, src2.sb);
-        short4 src2_data_3 = (short4)(src2.sc, src2.sd, src2.se, src2.sf);
-
-        short4 tmp_data_0 = src1_data_0 | src2_data_0;
-        short4 tmp_data_1 = src1_data_1 | src2_data_1;
-        short4 tmp_data_2 = src1_data_2 | src2_data_2;
-        short4 tmp_data_3 = src1_data_3 | src2_data_3;
-
-        *((__global short4 *)((__global char *)dst + dst_index + 0 ))= tmp_data_0;
-        *((__global short4 *)((__global char *)dst + dst_index + 8 ))= tmp_data_1;
-        *((__global short4 *)((__global char *)dst + dst_index + 16))= tmp_data_2;
-        *((__global short4 *)((__global char *)dst + dst_index + 24))= tmp_data_3;
-
-    }
-}
-#endif

modules/ocl/src/opencl/arithm_bitwise_xor.cl

@@ -1,340 +0,0 @@
-/*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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
-// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
-// Third party copyrights are property of their respective owners.
-//
-// @Authors
-//    Jiang Liyuan, jlyuan001.good@163.com
-//
-// 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 GpuMaterials 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*/
-#if defined (DOUBLE_SUPPORT)
-#ifdef cl_khr_fp64
-#pragma OPENCL EXTENSION cl_khr_fp64:enable
-#elif defined (cl_amd_fp64)
-#pragma OPENCL EXTENSION cl_amd_fp64:enable
-#endif
-#endif
-//////////////////////////////////////////////////////////////////////////////////////////////////////
-////////////////////////////////////////////BITWISE_XOR////////////////////////////////////////////////////
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-/**************************************bitwise_xor without mask**************************************/
-__kernel void arithm_bitwise_xor_D0 (__global uchar *src1, int src1_step, int src1_offset,
-                                     __global uchar *src2, int src2_step, int src2_offset,
-                                     __global uchar *dst,  int dst_step,  int dst_offset,
-                                     int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (dst_offset & 3)
-        int src1_index = mad24(y, src1_step, x + src1_offset - dst_align);
-        int src2_index = mad24(y, src2_step, x + src2_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + x & (int)0xfffffffc);
-
-        int src1_index_fix = src1_index < 0 ? 0 : src1_index;
-        int src2_index_fix = src2_index < 0 ? 0 : src2_index;
-        uchar4 src1_data = vload4(0, src1 + src1_index_fix);
-        uchar4 src2_data = vload4(0, src2 + src2_index_fix);
-
-        if(src1_index < 0)
-        {
-            uchar4 tmp;
-            tmp.xyzw = (src1_index == -2) ? src1_data.zwxy:src1_data.yzwx;
-            src1_data.xyzw = (src1_index == -1) ? src1_data.wxyz:tmp.xyzw;
-        }
-        if(src2_index < 0)
-        {
-            uchar4 tmp;
-            tmp.xyzw = (src2_index == -2) ? src2_data.zwxy:src2_data.yzwx;
-            src2_data.xyzw = (src2_index == -1) ? src2_data.wxyz:tmp.xyzw;
-        }
-        uchar4 dst_data = *((__global uchar4 *)(dst + dst_index));
-        uchar4 tmp_data = src1_data ^ src2_data;
-
-        dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
-        dst_data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : dst_data.y;
-        dst_data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : dst_data.z;
-        dst_data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : dst_data.w;
-
-        *((__global uchar4 *)(dst + dst_index)) = dst_data;
-    }
-}
-
-
-__kernel void arithm_bitwise_xor_D1 (__global char *src1, int src1_step, int src1_offset,
-                                     __global char *src2, int src2_step, int src2_offset,
-                                     __global char *dst,  int dst_step,  int dst_offset,
-                                     int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align (dst_offset & 3)
-        int src1_index = mad24(y, src1_step, x + src1_offset - dst_align);
-        int src2_index = mad24(y, src2_step, x + src2_offset - dst_align);
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + x & (int)0xfffffffc);
-
-        int src1_index_fix = src1_index < 0 ? 0 : src1_index;
-        int src2_index_fix = src2_index < 0 ? 0 : src2_index;
-        char4 src1_data = vload4(0, src1 + src1_index_fix);
-        char4 src2_data = vload4(0, src2 + src2_index_fix);
-
-        if(src1_index < 0)
-        {
-            char4 tmp;
-            tmp.xyzw = (src1_index == -2) ? src1_data.zwxy:src1_data.yzwx;
-            src1_data.xyzw = (src1_index == -1) ? src1_data.wxyz:tmp.xyzw;
-        }
-        if(src2_index < 0)
-        {
-            char4 tmp;
-            tmp.xyzw = (src2_index == -2) ? src2_data.zwxy:src2_data.yzwx;
-            src2_data.xyzw = (src2_index == -1) ? src2_data.wxyz:tmp.xyzw;
-        }
-        char4 dst_data = *((__global char4 *)(dst + dst_index));
-        char4 tmp_data = src1_data ^ src2_data;
-
-        dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
-        dst_data.y = ((dst_index + 1 >= dst_start) && (dst_index + 1 < dst_end)) ? tmp_data.y : dst_data.y;
-        dst_data.z = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.z : dst_data.z;
-        dst_data.w = ((dst_index + 3 >= dst_start) && (dst_index + 3 < dst_end)) ? tmp_data.w : dst_data.w;
-
-        *((__global char4 *)(dst + dst_index)) = dst_data;
-    }
-}
-
-
-__kernel void arithm_bitwise_xor_D2 (__global ushort *src1, int src1_step, int src1_offset,
-                                     __global ushort *src2, int src2_step, int src2_offset,
-                                     __global ushort *dst,  int dst_step,  int dst_offset,
-                                     int rows, int cols, int dst_step1)
-
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 3)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-        int src2_index = mad24(y, src2_step, (x << 1) + src2_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffff8);
-
-        int src1_index_fix = src1_index < 0 ? 0 : src1_index;
-        int src2_index_fix = src2_index < 0 ? 0 : src2_index;
-        ushort4 src1_data = vload4(0, (__global ushort *)((__global char *)src1 + src1_index_fix));
-        ushort4 src2_data = vload4(0, (__global ushort *)((__global char *)src2 + src2_index_fix));
-
-        if(src1_index < 0)
-        {
-            ushort4 tmp;
-            tmp.xyzw = (src1_index == -2) ? src1_data.zwxy:src1_data.yzwx;
-            src1_data.xyzw = (src1_index == -1) ? src1_data.wxyz:tmp.xyzw;
-        }
-        if(src2_index < 0)
-        {
-            ushort4 tmp;
-            tmp.xyzw = (src2_index == -2) ? src2_data.zwxy:src2_data.yzwx;
-            src2_data.xyzw = (src2_index == -1) ? src2_data.wxyz:tmp.xyzw;
-        }
-        ushort4 dst_data = *((__global ushort4 *)((__global char *)dst + dst_index));
-        ushort4 tmp_data = src1_data ^ src2_data;
-
-        dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
-        dst_data.y = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : dst_data.y;
-        dst_data.z = ((dst_index + 4 >= dst_start) && (dst_index + 4 < dst_end)) ? tmp_data.z : dst_data.z;
-        dst_data.w = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end)) ? tmp_data.w : dst_data.w;
-
-        *((__global ushort4 *)((__global char *)dst + dst_index)) = dst_data;
-    }
-}
-
-
-
-__kernel void arithm_bitwise_xor_D3 (__global short *src1, int src1_step, int src1_offset,
-                                     __global short *src2, int src2_step, int src2_offset,
-                                     __global short *dst,  int dst_step,  int dst_offset,
-                                     int rows, int cols, int dst_step1)
-
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        x = x << 2;
-
-#ifdef dst_align
-#undef dst_align
-#endif
-#define dst_align ((dst_offset >> 1) & 3)
-        int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
-        int src2_index = mad24(y, src2_step, (x << 1) + src2_offset - (dst_align << 1));
-
-        int dst_start  = mad24(y, dst_step, dst_offset);
-        int dst_end    = mad24(y, dst_step, dst_offset + dst_step1);
-        int dst_index  = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffff8);
-
-        int src1_index_fix = src1_index < 0 ? 0 : src1_index;
-        int src2_index_fix = src2_index < 0 ? 0 : src2_index;
-        short4 src1_data = vload4(0, (__global short *)((__global char *)src1 + src1_index_fix));
-        short4 src2_data = vload4(0, (__global short *)((__global char *)src2 + src2_index_fix));
-
-        short4 dst_data = *((__global short4 *)((__global char *)dst + dst_index));
-
-        if(src1_index < 0)
-        {
-            short4 tmp;
-            tmp.xyzw = (src1_index == -2) ? src1_data.zwxy:src1_data.yzwx;
-            src1_data.xyzw = (src1_index == -1) ? src1_data.wxyz:tmp.xyzw;
-        }
-        if(src2_index < 0)
-        {
-            short4 tmp;
-            tmp.xyzw = (src2_index == -2) ? src2_data.zwxy:src2_data.yzwx;
-            src2_data.xyzw = (src2_index == -1) ? src2_data.wxyz:tmp.xyzw;
-        }
-
-
-
-        short4 tmp_data = src1_data ^ src2_data;
-
-        dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
-        dst_data.y = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : dst_data.y;
-        dst_data.z = ((dst_index + 4 >= dst_start) && (dst_index + 4 < dst_end)) ? tmp_data.z : dst_data.z;
-        dst_data.w = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end)) ? tmp_data.w : dst_data.w;
-
-        *((__global short4 *)((__global char *)dst + dst_index)) = dst_data;
-    }
-}
-
-
-
-__kernel void arithm_bitwise_xor_D4 (__global int *src1, int src1_step, int src1_offset,
-                                     __global int *src2, int src2_step, int src2_offset,
-                                     __global int *dst,  int dst_step,  int dst_offset,
-                                     int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x << 2) + src2_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        int data1 = *((__global int *)((__global char *)src1 + src1_index));
-        int data2 = *((__global int *)((__global char *)src2 + src2_index));
-        int tmp  = data1 ^ data2;
-
-        *((__global int *)((__global char *)dst + dst_index)) = tmp;
-    }
-}
-
-__kernel void arithm_bitwise_xor_D5 (__global char *src1, int src1_step, int src1_offset,
-                                     __global char *src2, int src2_step, int src2_offset,
-                                     __global char *dst,  int dst_step,  int dst_offset,
-                                     int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x << 2) + src2_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 2) + dst_offset);
-
-        char4 data1 = *((__global char4 *)((__global char *)src1 + src1_index));
-        char4 data2 = *((__global char4 *)((__global char *)src2 + src2_index));
-        char4 tmp = data1 ^ data2;
-
-        *((__global char4 *)((__global char *)dst + dst_index)) = tmp;
-    }
-}
-#if defined (DOUBLE_SUPPORT)
-__kernel void arithm_bitwise_xor_D6 (__global char *src1, int src1_step, int src1_offset,
-                                     __global char *src2, int src2_step, int src2_offset,
-                                     __global char *dst,  int dst_step,  int dst_offset,
-                                     int rows, int cols, int dst_step1)
-{
-    int x = get_global_id(0);
-    int y = get_global_id(1);
-
-    if (x < cols && y < rows)
-    {
-        int src1_index = mad24(y, src1_step, (x << 3) + src1_offset);
-        int src2_index = mad24(y, src2_step, (x << 3) + src2_offset);
-        int dst_index  = mad24(y, dst_step,  (x << 3) + dst_offset);
-
-        char8 data1 = *((__global char8 *)((__global char *)src1 + src1_index));
-        char8 data2 = *((__global char8 *)((__global char *)src2 + src2_index));
-
-        *((__global char8 *)((__global char *)dst + dst_index)) = data1 ^ data2;
-    }
-}
-#endif

modules/ocl/src/opencl/filtering_boxFilter.cl

@@ -79,15 +79,73 @@
 #define ADDR_B(i, b_edge, addr)    ((i) >= (b_edge) ? (i)-(b_edge) : (addr))
 #endif
 
+#define THREADS 256
+#define ELEM(i, l_edge, r_edge, elem1, elem2) (i) >= (l_edge) && (i) < (r_edge) ? (elem1) : (elem2)
+
+inline void update_dst_C1_D0(__global uchar *dst, __local uint* temp,
+                             int dst_rows, int dst_cols,
+                             int dst_startX, int dst_x_off,
+                             float alpha)
+{
+    if(get_local_id(0) < anX || get_local_id(0) >= (THREADS-ksX+anX+1))
+    {
+        return;
+    }
+
+    uint4 tmp_sum = 0;
+    int posX = dst_startX - dst_x_off + (get_local_id(0)-anX)*4;
+    int posY = (get_group_id(1) << 1);
+
+    for(int i=-anX; i<=anX; i++)
+    {
+        tmp_sum += vload4(get_local_id(0), temp+i);
+    }
+
+    if(posY < dst_rows && posX < dst_cols)
+    {
+        tmp_sum /= (uint4) alpha;
+        if(posX >= 0 && posX < dst_cols)
+            *(dst) = tmp_sum.x;
+        if(posX+1 >= 0 && posX+1 < dst_cols)
+            *(dst + 1) = tmp_sum.y;
+        if(posX+2 >= 0 && posX+2 < dst_cols)
+            *(dst + 2) = tmp_sum.z;
+        if(posX+3 >= 0 && posX+3 < dst_cols)
+            *(dst + 3) = tmp_sum.w;
+    }
+}
+
+
+inline void update_dst_C4_D0(__global uchar4 *dst, __local uint4* temp,
+                             int dst_rows, int dst_cols,
+                             int dst_startX, int dst_x_off,
+                             float alpha)
+{
+    if(get_local_id(0) >= (THREADS-ksX+1))
+    {
+        return;
+    }
+
+    int posX = dst_startX - dst_x_off + get_local_id(0);
+    int posY = (get_group_id(1) << 1);
+
+    uint4 temp_sum = 0;
+    for(int i=-anX; i<=anX; i++)
+    {
+        temp_sum += temp[get_local_id(0) + anX + i];
+    }
+
+    if(posX >= 0 && posX < dst_cols && posY >= 0 && posY < dst_rows)
+        *dst = convert_uchar4(convert_float4(temp_sum)/alpha);
+}
+
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 /////////////////////////////////////////8uC1////////////////////////////////////////////////////////
 ////////////////////////////////////////////////////////////////////////////////////////////////////
-#define THREADS 256
-#define ELEM(i, l_edge, r_edge, elem1, elem2) (i) >= (l_edge) && (i) < (r_edge) ? (elem1) : (elem2)
 __kernel void boxFilter_C1_D0(__global const uchar * restrict src, __global uchar *dst, float alpha,
-                                     int src_offset, int src_whole_rows, int src_whole_cols, int src_step,
-                                     int dst_offset, int dst_rows, int dst_cols, int dst_step
-                                     )
+                              int src_offset, int src_whole_rows, int src_whole_cols, int src_step,
+                              int dst_offset, int dst_rows, int dst_cols, int dst_step
+                             )
 {
 
     int col = get_local_id(0);
@@ -105,115 +163,84 @@ __kernel void boxFilter_C1_D0(__global const uchar * restrict src, __global ucha
     int dst_startY = (gY << 1) + dst_y_off;
 
     uint4 data[ksY+1];
-    __local uint4 temp[(THREADS<<1)];
+    __local uint4 temp[2][THREADS];
 
 #ifdef BORDER_CONSTANT
 
-        for(int i=0; i < ksY+1; i++)
+    for(int i=0; i < ksY+1; i++)
+    {
+        if(startY+i >=0 && startY+i < src_whole_rows && startX+col*4 >=0 && startX+col*4+3<src_whole_cols)
         {
-            if(startY+i >=0 && startY+i < src_whole_rows && startX+col*4 >=0 && startX+col*4+3<src_whole_cols)
-                data[i] = convert_uint4(vload4(col,(__global uchar*)(src+(startY+i)*src_step + startX)));
-            else
-            {
-                data[i]=0;
-                int con = startY+i >=0 && startY+i < src_whole_rows && startX+col*4 >=0 && startX+col*4<src_whole_cols;
-                if(con)data[i].s0 = *(src+(startY+i)*src_step + startX + col*4);
-                con = startY+i >=0 && startY+i < src_whole_rows && startX+col*4+1 >=0 && startX+col*4+1<src_whole_cols;
-                if(con)data[i].s1 = *(src+(startY+i)*src_step + startX + col*4+1) ;
-                con = startY+i >=0 && startY+i < src_whole_rows && startX+col*4+2 >=0 && startX+col*4+2<src_whole_cols;
-                if(con)data[i].s2 = *(src+(startY+i)*src_step + startX + col*4+2);
-                con = startY+i >=0 && startY+i < src_whole_rows && startX+col*4+3 >=0 && startX+col*4+3<src_whole_cols;
-                if(con)data[i].s3 = *(src+(startY+i)*src_step + startX + col*4+3);
-            }
+            data[i].x = *(src+(startY+i)*src_step + startX + col * 4);
+            data[i].y = *(src+(startY+i)*src_step + startX + col * 4 + 1);
+            data[i].z = *(src+(startY+i)*src_step + startX + col * 4 + 2);
+            data[i].w = *(src+(startY+i)*src_step + startX + col * 4 + 3);
         }
+        else
+        {
+            data[i]=0;
+            int con = startY+i >=0 && startY+i < src_whole_rows && startX+col*4 >=0 && startX+col*4<src_whole_cols;
+            if(con)data[i].s0 = *(src+(startY+i)*src_step + startX + col*4);
+            con = startY+i >=0 && startY+i < src_whole_rows && startX+col*4+1 >=0 && startX+col*4+1<src_whole_cols;
+            if(con)data[i].s1 = *(src+(startY+i)*src_step + startX + col*4+1) ;
+            con = startY+i >=0 && startY+i < src_whole_rows && startX+col*4+2 >=0 && startX+col*4+2<src_whole_cols;
+            if(con)data[i].s2 = *(src+(startY+i)*src_step + startX + col*4+2);
+            con = startY+i >=0 && startY+i < src_whole_rows && startX+col*4+3 >=0 && startX+col*4+3<src_whole_cols;
+            if(con)data[i].s3 = *(src+(startY+i)*src_step + startX + col*4+3);
+        }
+    }
 
 #else
-   int not_all_in_range;
-   for(int i=0; i < ksY+1; i++)
-   {
-      not_all_in_range = (startX+col*4<0) | (startX+col*4+3>src_whole_cols-1)
-                        | (startY+i<0) | (startY+i>src_whole_rows-1);
-      if(not_all_in_range)
-      {
-          int selected_row;
-          int4 selected_col;
-          selected_row = ADDR_H(startY+i, 0, src_whole_rows);
-          selected_row = ADDR_B(startY+i, src_whole_rows, selected_row);
-
-          selected_col.x = ADDR_L(startX+col*4, 0, src_whole_cols);
-          selected_col.x = ADDR_R(startX+col*4, src_whole_cols, selected_col.x);
-
-          selected_col.y = ADDR_L(startX+col*4+1, 0, src_whole_cols);
-          selected_col.y = ADDR_R(startX+col*4+1, src_whole_cols, selected_col.y);
-
-          selected_col.z = ADDR_L(startX+col*4+2, 0, src_whole_cols);
-          selected_col.z = ADDR_R(startX+col*4+2, src_whole_cols, selected_col.z);
-
-          selected_col.w = ADDR_L(startX+col*4+3, 0, src_whole_cols);
-          selected_col.w = ADDR_R(startX+col*4+3, src_whole_cols, selected_col.w);
-
-          data[i].x = *(src + selected_row * src_step + selected_col.x);
-          data[i].y = *(src + selected_row * src_step + selected_col.y);
-          data[i].z = *(src + selected_row * src_step + selected_col.z);
-          data[i].w = *(src + selected_row * src_step + selected_col.w);
-      }
-      else
-      {
-          data[i] =  convert_uint4(vload4(col,(__global uchar*)(src+(startY+i)*src_step + startX)));
-      }
-   }
-#endif
-    uint4 sum0 = 0, sum1 = 0, sum2 = 0;
-    for(int i=1; i < ksY; i++)
+    int not_all_in_range;
+    for(int i=0; i < ksY+1; i++)
     {
-        sum0 += (data[i]);
-    }
-    sum1 = sum0 + (data[0]);
-    sum2 = sum0 + (data[ksY]);
+        not_all_in_range = (startX+col*4<0) | (startX+col*4+3>src_whole_cols-1)
+                           | (startY+i<0) | (startY+i>src_whole_rows-1);
+        if(not_all_in_range)
+        {
+            int selected_row;
+            int4 selected_col;
+            selected_row = ADDR_H(startY+i, 0, src_whole_rows);
+            selected_row = ADDR_B(startY+i, src_whole_rows, selected_row);
 
-    temp[col] = sum1;
-    temp[col+THREADS] = sum2;
-    barrier(CLK_LOCAL_MEM_FENCE);
+            selected_col.x = ADDR_L(startX+col*4, 0, src_whole_cols);
+            selected_col.x = ADDR_R(startX+col*4, src_whole_cols, selected_col.x);
 
-    if(col >= anX && col < (THREADS-ksX+anX+1))
-    {
-        int posX = dst_startX - dst_x_off + (col-anX)*4;
-        int posY = (gY << 1);
-        uint4 tmp_sum1=0, tmp_sum2=0;
-        for(int i=-anX; i<=anX; i++)
-        {
-           tmp_sum1 += vload4(col, (__local uint*)temp+i);
-        }
+            selected_col.y = ADDR_L(startX+col*4+1, 0, src_whole_cols);
+            selected_col.y = ADDR_R(startX+col*4+1, src_whole_cols, selected_col.y);
 
-        for(int i=-anX; i<=anX; i++)
-        {
-           tmp_sum2 += vload4(col, (__local uint*)(temp+THREADS)+i);
-        }
+            selected_col.z = ADDR_L(startX+col*4+2, 0, src_whole_cols);
+            selected_col.z = ADDR_R(startX+col*4+2, src_whole_cols, selected_col.z);
 
-        if(posY < dst_rows && posX < dst_cols)
-        {
-           if(posX >= 0 && posX < dst_cols)
-               *(dst+dst_startY * dst_step + dst_startX + (col-anX)*4) = tmp_sum1.x/alpha;
-           if(posX+1 >= 0 && posX+1 < dst_cols)
-               *(dst+dst_startY * dst_step + dst_startX+1 + (col-anX)*4) = tmp_sum1.y/alpha;
-           if(posX+2 >= 0 && posX+2 < dst_cols)
-               *(dst+dst_startY * dst_step + dst_startX+2 + (col-anX)*4) = tmp_sum1.z/alpha;
-           if(posX+3 >= 0 && posX+3 < dst_cols)
-               *(dst+dst_startY * dst_step + dst_startX+3 + (col-anX)*4) = tmp_sum1.w/alpha;
+            selected_col.w = ADDR_L(startX+col*4+3, 0, src_whole_cols);
+            selected_col.w = ADDR_R(startX+col*4+3, src_whole_cols, selected_col.w);
+
+            data[i].x = *(src + selected_row * src_step + selected_col.x);
+            data[i].y = *(src + selected_row * src_step + selected_col.y);
+            data[i].z = *(src + selected_row * src_step + selected_col.z);
+            data[i].w = *(src + selected_row * src_step + selected_col.w);
         }
-        if(posY+1 < dst_rows && posX < dst_cols)
+        else
         {
-           dst_startY+=1;
-           if(posX >= 0 && posX < dst_cols)
-               *(dst+dst_startY * dst_step + dst_startX + (col-anX)*4) = tmp_sum2.x/alpha;
-           if(posX+1 >= 0 && posX+1 < dst_cols)
-               *(dst+dst_startY * dst_step + dst_startX+1 + (col-anX)*4) = tmp_sum2.y/alpha;
-           if(posX+2 >= 0 && posX+2 < dst_cols)
-               *(dst+dst_startY * dst_step + dst_startX+2 + (col-anX)*4) = tmp_sum2.z/alpha;
-           if(posX+3 >= 0 && posX+3 < dst_cols)
-               *(dst+dst_startY * dst_step + dst_startX+3 + (col-anX)*4) = tmp_sum2.w/alpha;
+            data[i] =  convert_uint4(vload4(col,(__global uchar*)(src+(startY+i)*src_step + startX)));
         }
     }
+#endif
+    uint4 tmp_sum = 0;
+    for(int i=1; i < ksY; i++)
+    {
+        tmp_sum += (data[i]);
+    }
+    
+    int index = dst_startY * dst_step + dst_startX + (col-anX)*4;
+
+    temp[0][col] = tmp_sum + (data[0]);
+    temp[1][col] = tmp_sum + (data[ksY]);
+    barrier(CLK_LOCAL_MEM_FENCE);
+    update_dst_C1_D0(dst+index, (__local uint *)(temp[0]),
+                     dst_rows, dst_cols, dst_startX, dst_x_off, alpha);
+    update_dst_C1_D0(dst+index+dst_step, (__local uint *)(temp[1]),
+                     dst_rows, dst_cols, dst_startX, dst_x_off, alpha);
 
 }
 
@@ -221,9 +248,9 @@ __kernel void boxFilter_C1_D0(__global const uchar * restrict src, __global ucha
 /////////////////////////////////////////8uC4////////////////////////////////////////////////////////
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 __kernel void boxFilter_C4_D0(__global const uchar4 * restrict src, __global uchar4 *dst, float alpha,
-                                     int src_offset, int src_whole_rows, int src_whole_cols, int src_step,
-                                     int dst_offset, int dst_rows, int dst_cols, int dst_step
-                                     )
+                              int src_offset, int src_whole_rows, int src_whole_cols, int src_step,
+                              int dst_offset, int dst_rows, int dst_cols, int dst_step
+                             )
 {
     int col = get_local_id(0);
     const int gX = get_group_id(0);
@@ -238,81 +265,63 @@ __kernel void boxFilter_C4_D0(__global const uchar4 * restrict src, __global uch
     int startY = (gY << 1) - anY + src_y_off;
     int dst_startX = gX * (THREADS-ksX+1) + dst_x_off;
     int dst_startY = (gY << 1) + dst_y_off;
-      //int end_addr = (src_whole_rows-1)*(src_step>>2) + src_whole_cols-4;
 
-      int end_addr = src_whole_cols-4;
     uint4 data[ksY+1];
     __local uint4 temp[2][THREADS];
+
 #ifdef BORDER_CONSTANT
     bool con;
-    uint4 ss;
     for(int i=0; i < ksY+1; i++)
     {
         con = startX+col >= 0 && startX+col < src_whole_cols && startY+i >= 0 && startY+i < src_whole_rows;
-
-            //int cur_addr = clamp((startY+i)*(src_step>>2)+(startX+col),0,end_addr);
-        //ss = convert_uint4(src[cur_addr]);
-
         int cur_col = clamp(startX + col, 0, src_whole_cols);
-        if(con)
-          ss = convert_uint4(src[(startY+i)*(src_step>>2) + cur_col]);
 
-        data[i] = con ? ss : 0;
+        data[i].x = con ? src[(startY+i)*(src_step>>2) + cur_col].x : 0;
+        data[i].y = con ? src[(startY+i)*(src_step>>2) + cur_col].y : 0;
+        data[i].z = con ? src[(startY+i)*(src_step>>2) + cur_col].z : 0;
+        data[i].w = con ? src[(startY+i)*(src_step>>2) + cur_col].w : 0;
     }
 #else
-   for(int i=0; i < ksY+1; i++)
-   {
-          int selected_row;
-          int selected_col;
-          selected_row = ADDR_H(startY+i, 0, src_whole_rows);
-          selected_row = ADDR_B(startY+i, src_whole_rows, selected_row);
+    for(int i=0; i < ksY+1; i++)
+    {
+        int selected_row;
+        int selected_col;
+        selected_row = ADDR_H(startY+i, 0, src_whole_rows);
+        selected_row = ADDR_B(startY+i, src_whole_rows, selected_row);
 
-          selected_col = ADDR_L(startX+col, 0, src_whole_cols);
-          selected_col = ADDR_R(startX+col, src_whole_cols, selected_col);
+        selected_col = ADDR_L(startX+col, 0, src_whole_cols);
+        selected_col = ADDR_R(startX+col, src_whole_cols, selected_col);
 
 
-          data[i] = convert_uint4(src[selected_row * (src_step>>2) + selected_col]);
-   }
+        data[i] = convert_uint4(src[selected_row * (src_step>>2) + selected_col]);
+    }
 
 #endif
-    uint4 sum0 = 0, sum1 = 0, sum2 = 0;
+    uint4 tmp_sum = 0;
     for(int i=1; i < ksY; i++)
     {
-        sum0 += (data[i]);
+        tmp_sum += (data[i]);
     }
-    sum1 = sum0 + (data[0]);
-    sum2 = sum0 + (data[ksY]);
-    temp[0][col] = sum1;
-    temp[1][col] = sum2;
-    barrier(CLK_LOCAL_MEM_FENCE);
-    if(col < (THREADS-(ksX-1)))
-    {
-        col += anX;
-        int posX = dst_startX - dst_x_off + col - anX;
-        int posY = (gY << 1);
 
-        uint4 tmp_sum[2]={(uint4)(0,0,0,0),(uint4)(0,0,0,0)};
-        for(int k=0; k<2; k++)
-            for(int i=-anX; i<=anX; i++)
-            {
-                tmp_sum[k] += temp[k][col+i];
-            }
-        for(int i=0; i<2; i++)
-        {
-            if(posX >= 0 && posX < dst_cols && (posY+i) >= 0 && (posY+i) < dst_rows)
-                dst[(dst_startY+i) * (dst_step>>2)+ dst_startX + col - anX] = convert_uchar4(convert_float4(tmp_sum[i])/alpha);
-        }
+    int index = dst_startY * (dst_step>>2)+ dst_startX + col;
+
+    temp[0][col] = tmp_sum + (data[0]);
+    temp[1][col] = tmp_sum + (data[ksY]);
+    barrier(CLK_LOCAL_MEM_FENCE);
+    update_dst_C4_D0(dst+index, (__local uint4 *)(temp[0]),
+                     dst_rows, dst_cols, dst_startX, dst_x_off, alpha);
+    update_dst_C4_D0(dst+index+(dst_step>>2), (__local uint4 *)(temp[1]),
+                     dst_rows, dst_cols, dst_startX, dst_x_off, alpha);
 
-    }
 }
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 /////////////////////////////////////////32fC1////////////////////////////////////////////////////////
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 __kernel void boxFilter_C1_D5(__global const float *restrict src, __global float *dst, float alpha,
-                                     int src_offset, int src_whole_rows, int src_whole_cols, int src_step,
-                                     int dst_offset, int dst_rows, int dst_cols, int dst_step
-                                     )
+                              int src_offset, int src_whole_rows, int src_whole_cols, int src_step,
+                              int dst_offset, int dst_rows, int dst_cols, int dst_step
+                             )
 {
     int col = get_local_id(0);
     const int gX = get_group_id(0);
@@ -327,7 +336,6 @@ __kernel void boxFilter_C1_D5(__global const float *restrict src, __global float
     int startY = (gY << 1) - anY + src_y_off;
     int dst_startX = gX * (THREADS-ksX+1) + dst_x_off;
     int dst_startY = (gY << 1) + dst_y_off;
-    int end_addr = (src_whole_rows-1)*(src_step>>2) + src_whole_cols-4;
     float data[ksY+1];
     __local float temp[2][THREADS];
 #ifdef BORDER_CONSTANT
@@ -336,28 +344,25 @@ __kernel void boxFilter_C1_D5(__global const float *restrict src, __global float
     for(int i=0; i < ksY+1; i++)
     {
         con = startX+col >= 0 && startX+col < src_whole_cols && startY+i >= 0 && startY+i < src_whole_rows;
-          //int cur_addr = clamp((startY+i)*(src_step>>2)+(startX+col),0,end_addr);
-        //ss = src[cur_addr];
 
         int cur_col = clamp(startX + col, 0, src_whole_cols);
-        //ss = src[(startY+i)*(src_step>>2) + cur_col];
-        ss = (startY+i)<src_whole_rows&&(startY+i)>=0&&cur_col>=0&&cur_col<src_whole_cols?src[(startY+i)*(src_step>>2) + cur_col]:0;
+        ss = (startY+i)<src_whole_rows&&(startY+i)>=0&&cur_col>=0&&cur_col<src_whole_cols?src[(startY+i)*(src_step>>2) + cur_col]:(float)0;
 
         data[i] = con ? ss : 0.f;
     }
 #else
-   for(int i=0; i < ksY+1; i++)
-   {
-          int selected_row;
-          int selected_col;
-          selected_row = ADDR_H(startY+i, 0, src_whole_rows);
-          selected_row = ADDR_B(startY+i, src_whole_rows, selected_row);
+    for(int i=0; i < ksY+1; i++)
+    {
+        int selected_row;
+        int selected_col;
+        selected_row = ADDR_H(startY+i, 0, src_whole_rows);
+        selected_row = ADDR_B(startY+i, src_whole_rows, selected_row);
 
-          selected_col = ADDR_L(startX+col, 0, src_whole_cols);
-          selected_col = ADDR_R(startX+col, src_whole_cols, selected_col);
+        selected_col = ADDR_L(startX+col, 0, src_whole_cols);
+        selected_col = ADDR_R(startX+col, src_whole_cols, selected_col);
 
-          data[i] = src[selected_row * (src_step>>2) + selected_col];
-   }
+        data[i] = src[selected_row * (src_step>>2) + selected_col];
+    }
 
 #endif
     float sum0 = 0.0, sum1 = 0.0, sum2 = 0.0;
@@ -376,7 +381,7 @@ __kernel void boxFilter_C1_D5(__global const float *restrict src, __global float
         int posX = dst_startX - dst_x_off + col - anX;
         int posY = (gY << 1);
 
-        float tmp_sum[2]={0.0, 0.0};
+        float tmp_sum[2]= {0.0, 0.0};
         for(int k=0; k<2; k++)
             for(int i=-anX; i<=anX; i++)
             {
@@ -395,9 +400,9 @@ __kernel void boxFilter_C1_D5(__global const float *restrict src, __global float
 /////////////////////////////////////////32fC4////////////////////////////////////////////////////////
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 __kernel void boxFilter_C4_D5(__global const float4 *restrict src, __global float4 *dst, float alpha,
-                                     int src_offset, int src_whole_rows, int src_whole_cols, int src_step,
-                                     int dst_offset, int dst_rows, int dst_cols, int dst_step
-                                     )
+                              int src_offset, int src_whole_rows, int src_whole_cols, int src_step,
+                              int dst_offset, int dst_rows, int dst_cols, int dst_step
+                             )
 {
     int col = get_local_id(0);
     const int gX = get_group_id(0);
@@ -412,7 +417,6 @@ __kernel void boxFilter_C4_D5(__global const float4 *restrict src, __global floa
     int startY = (gY << 1) - anY + src_y_off;
     int dst_startX = gX * (THREADS-ksX+1) + dst_x_off;
     int dst_startY = (gY << 1) + dst_y_off;
-    int end_addr = (src_whole_rows-1)*(src_step>>4) + src_whole_cols-16;
     float4 data[ksY+1];
     __local float4 temp[2][THREADS];
 #ifdef BORDER_CONSTANT
@@ -421,28 +425,25 @@ __kernel void boxFilter_C4_D5(__global const float4 *restrict src, __global floa
     for(int i=0; i < ksY+1; i++)
     {
         con = startX+col >= 0 && startX+col < src_whole_cols && startY+i >= 0 && startY+i < src_whole_rows;
-            //int cur_addr = clamp((startY+i)*(src_step>>4)+(startX+col),0,end_addr);
-        //ss = src[cur_addr];
 
         int cur_col = clamp(startX + col, 0, src_whole_cols);
-        //ss = src[(startY+i)*(src_step>>4) + cur_col];
-        ss = (startY+i)<src_whole_rows&&(startY+i)>=0&&cur_col>=0&&cur_col<src_whole_cols?src[(startY+i)*(src_step>>4) + cur_col]:0;
+        ss = (startY+i)<src_whole_rows&&(startY+i)>=0&&cur_col>=0&&cur_col<src_whole_cols?src[(startY+i)*(src_step>>4) + cur_col]:(float4)0;
 
         data[i] = con ? ss : (float4)(0.0,0.0,0.0,0.0);
     }
 #else
-   for(int i=0; i < ksY+1; i++)
-   {
-          int selected_row;
-          int selected_col;
-          selected_row = ADDR_H(startY+i, 0, src_whole_rows);
-          selected_row = ADDR_B(startY+i, src_whole_rows, selected_row);
+    for(int i=0; i < ksY+1; i++)
+    {
+        int selected_row;
+        int selected_col;
+        selected_row = ADDR_H(startY+i, 0, src_whole_rows);
+        selected_row = ADDR_B(startY+i, src_whole_rows, selected_row);
 
-          selected_col = ADDR_L(startX+col, 0, src_whole_cols);
-          selected_col = ADDR_R(startX+col, src_whole_cols, selected_col);
+        selected_col = ADDR_L(startX+col, 0, src_whole_cols);
+        selected_col = ADDR_R(startX+col, src_whole_cols, selected_col);
 
-          data[i] = src[selected_row * (src_step>>4) + selected_col];
-   }
+        data[i] = src[selected_row * (src_step>>4) + selected_col];
+    }
 
 #endif
     float4 sum0 = 0.0, sum1 = 0.0, sum2 = 0.0;
@@ -461,7 +462,7 @@ __kernel void boxFilter_C4_D5(__global const float4 *restrict src, __global floa
         int posX = dst_startX - dst_x_off + col - anX;
         int posY = (gY << 1);
 
-        float4 tmp_sum[2]={(float4)(0.0,0.0,0.0,0.0), (float4)(0.0,0.0,0.0,0.0)};
+        float4 tmp_sum[2]= {(float4)(0.0,0.0,0.0,0.0), (float4)(0.0,0.0,0.0,0.0)};
         for(int k=0; k<2; k++)
             for(int i=-anX; i<=anX; i++)
             {

modules/ocl/src/opencl/haarobjectdetect.cl

@@ -112,7 +112,7 @@ typedef struct __attribute__((aligned (64))) GpuHidHaarClassifierCascade
 } GpuHidHaarClassifierCascade;
 
 
-__kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCascade(//constant GpuHidHaarClassifierCascade * cascade,
+__kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCascade(
     global GpuHidHaarStageClassifier * stagecascadeptr,
     global int4 * info,
     global GpuHidHaarTreeNode * nodeptr,
@@ -128,12 +128,7 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
     const int splitnode,
     const int4 p,
     const int4 pq,
-    const float correction
-    //const int width,
-    //const int height,
-    //const int grpnumperline,
-    //const int totalgrp
-)
+    const float correction)
 {
     int grpszx = get_local_size(0);
     int grpszy = get_local_size(1);
@@ -145,13 +140,8 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
     int lcl_sz = mul24(grpszx,grpszy);
     int lcl_id = mad24(lclidy,grpszx,lclidx);
 
-    //assume lcl_sz == 256 or 128 or 64
-    //int lcl_sz_shift = (lcl_sz == 256) ? 8 : 7;
-    //lcl_sz_shift = (lcl_sz == 64) ? 6 : lcl_sz_shift;
     __local int lclshare[1024];
-
-#define OFF 0
-    __local int* lcldata = lclshare + OFF;//for save win data
+    __local int* lcldata = lclshare;//for save win data
     __local int* glboutindex = lcldata + 28*28;//for save global out index
     __local int* lclcount = glboutindex + 1;//for save the numuber of temp pass pixel
     __local int* lcloutindex = lclcount + 1;//for save info of temp pass pixel
@@ -181,7 +171,6 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
         int totalgrp = scaleinfo1.y & 0xffff;
         int imgoff = scaleinfo1.z;
         float factor = as_float(scaleinfo1.w);
-        //int ystep =1;// factor > 2.0 ? 1 : 2;
 
         __global const int * sum = sum1 + imgoff;
         __global const float * sqsum = sqsum1 + imgoff;
@@ -191,8 +180,6 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
             int grpidx = grploop - mul24(grpidy, grpnumperline);
             int x = mad24(grpidx,grpszx,lclidx);
             int y = mad24(grpidy,grpszy,lclidy);
-            //candidate_result.x = convert_int_rtn(x*factor);
-            //candidate_result.y = convert_int_rtn(y*factor);
             int grpoffx = x-lclidx;
             int grpoffy = y-lclidy;
 
@@ -207,18 +194,11 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
                 int glb_x = grpoffx + (lcl_x<<2);
                 int glb_y = grpoffy + lcl_y;
 
-                int glb_off = mad24(glb_y,pixelstep,glb_x);
+                int glb_off = mad24(min(glb_y, height - 1),pixelstep,glb_x);
                 int4 data = *(__global int4*)&sum[glb_off];
                 int lcl_off = mad24(lcl_y, readwidth, lcl_x<<2);
 
-#if OFF
-                lcldata[lcl_off] = data.x;
-                lcldata[lcl_off+1] = data.y;
-                lcldata[lcl_off+2] = data.z;
-                lcldata[lcl_off+3] = data.w;
-#else
                 vstore4(data, 0, &lcldata[lcl_off]);
-#endif
             }
 
             lcloutindex[lcl_id] = 0;
@@ -231,184 +211,170 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
             int lcl_off = mad24(lclidy,readwidth,lclidx);
             int4 cascadeinfo1, cascadeinfo2;
             cascadeinfo1 = p;
-            cascadeinfo2 = pq;// + mad24(y, pixelstep, x);
+            cascadeinfo2 = pq;
 
+            cascadeinfo1.x +=lcl_off;
+            cascadeinfo1.z +=lcl_off;
+            mean = (lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.x)] - lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.z)] -
+                    lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.x)] + lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.z)])
+                    *correction;
 
-            //if((x < width) && (y < height))
-            {
-                cascadeinfo1.x +=lcl_off;
-                cascadeinfo1.z +=lcl_off;
-                mean = (lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.x)] - lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.z)] -
-                        lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.x)] + lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.z)])
-                       *correction;
-
-                int p_offset = mad24(y, pixelstep, x);
-
-                cascadeinfo2.x +=p_offset;
-                cascadeinfo2.z +=p_offset;
-                variance_norm_factor =sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.x)] - sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.z)] -
-                                      sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.x)] + sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.z)];
-
-                variance_norm_factor = variance_norm_factor * correction - mean * mean;
-                variance_norm_factor = variance_norm_factor >=0.f ? sqrt(variance_norm_factor) : 1.f;
-                //if( cascade->is_stump_based )
-                //{
-                for(int stageloop = start_stage; (stageloop < split_stage)  && result; stageloop++ )
-                {
-                    float stage_sum = 0.f;
-                    int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop);
-                    float stagethreshold = as_float(stageinfo.y);
-                    for(int nodeloop = 0; nodeloop < stageinfo.x; nodeloop++ )
-                    {
-                        __global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter);
+            int p_offset = mad24(y, pixelstep, x);
 
-                        int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
-                        int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
-                        int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
-                        float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
-                        float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
-                        float nodethreshold  = w.w * variance_norm_factor;
+            cascadeinfo2.x +=p_offset;
+            cascadeinfo2.z +=p_offset;
+            variance_norm_factor =sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.x)] - sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.z)] -
+                                    sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.x)] + sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.z)];
 
-                        info1.x +=lcl_off;
-                        info1.z +=lcl_off;
-                        info2.x +=lcl_off;
-                        info2.z +=lcl_off;
+            variance_norm_factor = variance_norm_factor * correction - mean * mean;
+            variance_norm_factor = variance_norm_factor >=0.f ? sqrt(variance_norm_factor) : 1.f;
 
-                        float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
-                                          lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
+            for(int stageloop = start_stage; (stageloop < split_stage)  && result; stageloop++ )
+            {
+                float stage_sum = 0.f;
+                int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop);
+                float stagethreshold = as_float(stageinfo.y);
+                for(int nodeloop = 0; nodeloop < stageinfo.x; nodeloop++ )
+                {
+                    __global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter);
 
+                    int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
+                    int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
+                    int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
+                    float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
+                    float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
+                    float nodethreshold  = w.w * variance_norm_factor;
 
-                        classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
-                                     lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
+                    info1.x +=lcl_off;
+                    info1.z +=lcl_off;
+                    info2.x +=lcl_off;
+                    info2.z +=lcl_off;
 
+                    float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
+                                        lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
 
-                        //if((info3.z - info3.x) && (!stageinfo.z))
-                        //{
-                        info3.x +=lcl_off;
-                        info3.z +=lcl_off;
-                        classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
-                                     lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
-                        //}
-                        stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
-                        nodecounter++;
-                    }
+                    classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
+                                    lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
 
-                    result = (stage_sum >= stagethreshold);
-                }
+                    info3.x +=lcl_off;
+                    info3.z +=lcl_off;
+                    classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
+                                    lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
 
-                if(result && (x < width) && (y < height))
-                {
-                    int queueindex = atomic_inc(lclcount);
-                    lcloutindex[queueindex<<1] = (lclidy << 16) | lclidx;
-                    lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
+                    stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
+                    nodecounter++;
                 }
+
+                result = (stage_sum >= stagethreshold);
+            }
+
+            if(result && (x < width) && (y < height))
+            {
+                int queueindex = atomic_inc(lclcount);
+                lcloutindex[queueindex<<1] = (lclidy << 16) | lclidx;
+                lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
+            }
+            barrier(CLK_LOCAL_MEM_FENCE);
+            int queuecount  = lclcount[0];
+            barrier(CLK_LOCAL_MEM_FENCE);
+            nodecounter = splitnode;
+            for(int stageloop = split_stage; stageloop< end_stage && queuecount>0; stageloop++)
+            {
+                lclcount[0]=0;
                 barrier(CLK_LOCAL_MEM_FENCE);
-                int queuecount  = lclcount[0];
-                barrier(CLK_LOCAL_MEM_FENCE);
-                nodecounter = splitnode;
-                for(int stageloop = split_stage; stageloop< end_stage && queuecount>0; stageloop++)
-                {
-                    //barrier(CLK_LOCAL_MEM_FENCE);
-                    //if(lcl_id == 0)
-                    lclcount[0]=0;
-                    barrier(CLK_LOCAL_MEM_FENCE);
 
-                    int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop);
-                    float stagethreshold = as_float(stageinfo.y);
+                int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop);
+                float stagethreshold = as_float(stageinfo.y);
 
-                    int perfscale = queuecount > 4 ? 3 : 2;
-                    int queuecount_loop = (queuecount + (1<<perfscale)-1) >> perfscale;
-                    int lcl_compute_win = lcl_sz >> perfscale;
-                    int lcl_compute_win_id = (lcl_id >>(6-perfscale));
-                    int lcl_loops = (stageinfo.x + lcl_compute_win -1) >> (6-perfscale);
-                    int lcl_compute_id = lcl_id - (lcl_compute_win_id << (6-perfscale));
-                    for(int queueloop=0; queueloop<queuecount_loop/* && lcl_compute_win_id < queuecount*/; queueloop++)
+                int perfscale = queuecount > 4 ? 3 : 2;
+                int queuecount_loop = (queuecount + (1<<perfscale)-1) >> perfscale;
+                int lcl_compute_win = lcl_sz >> perfscale;
+                int lcl_compute_win_id = (lcl_id >>(6-perfscale));
+                int lcl_loops = (stageinfo.x + lcl_compute_win -1) >> (6-perfscale);
+                int lcl_compute_id = lcl_id - (lcl_compute_win_id << (6-perfscale));
+                for(int queueloop=0; queueloop<queuecount_loop; queueloop++)
+                {
+                    float stage_sum = 0.f;
+                    int temp_coord = lcloutindex[lcl_compute_win_id<<1];
+                    float variance_norm_factor = as_float(lcloutindex[(lcl_compute_win_id<<1)+1]);
+                    int queue_pixel = mad24(((temp_coord  & (int)0xffff0000)>>16),readwidth,temp_coord & 0xffff);
+
+                    if(lcl_compute_win_id < queuecount)
                     {
-                        float stage_sum = 0.f;
-                        int temp_coord = lcloutindex[lcl_compute_win_id<<1];
-                        float variance_norm_factor = as_float(lcloutindex[(lcl_compute_win_id<<1)+1]);
-                        int queue_pixel = mad24(((temp_coord  & (int)0xffff0000)>>16),readwidth,temp_coord & 0xffff);
 
-                        //barrier(CLK_LOCAL_MEM_FENCE);
-                        if(lcl_compute_win_id < queuecount)
+                        int tempnodecounter = lcl_compute_id;
+                        float part_sum = 0.f;
+                        for(int lcl_loop=0; lcl_loop<lcl_loops && tempnodecounter<stageinfo.x; lcl_loop++)
                         {
+                            __global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter + tempnodecounter);
+
+                            int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
+                            int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
+                            int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
+                            float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
+                            float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
+                            float nodethreshold  = w.w * variance_norm_factor;
+
+                            info1.x +=queue_pixel;
+                            info1.z +=queue_pixel;
+                            info2.x +=queue_pixel;
+                            info2.z +=queue_pixel;
+
+                            float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
+                                                lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
 
-                            int tempnodecounter = lcl_compute_id;
-                            float part_sum = 0.f;
-                            for(int lcl_loop=0; lcl_loop<lcl_loops && tempnodecounter<stageinfo.x; lcl_loop++)
-                            {
-                                __global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter + tempnodecounter);
-
-                                int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
-                                int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
-                                int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
-                                float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
-                                float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
-                                float nodethreshold  = w.w * variance_norm_factor;
-
-                                info1.x +=queue_pixel;
-                                info1.z +=queue_pixel;
-                                info2.x +=queue_pixel;
-                                info2.z +=queue_pixel;
-
-                                float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
-                                                  lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
-
-
-                                classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
-                                             lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
-                                //if((info3.z - info3.x) && (!stageinfo.z))
-                                //{
-                                info3.x +=queue_pixel;
-                                info3.z +=queue_pixel;
-                                classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
-                                             lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
-                                //}
-                                part_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
-                                tempnodecounter +=lcl_compute_win;
-                            }//end for(int lcl_loop=0;lcl_loop<lcl_loops;lcl_loop++)
-                            partialsum[lcl_id]=part_sum;
+
+                            classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
+                                            lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
+
+                            info3.x +=queue_pixel;
+                            info3.z +=queue_pixel;
+                            classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
+                                            lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
+
+                            part_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
+                            tempnodecounter +=lcl_compute_win;
+                        }//end for(int lcl_loop=0;lcl_loop<lcl_loops;lcl_loop++)
+                        partialsum[lcl_id]=part_sum;
+                    }
+                    barrier(CLK_LOCAL_MEM_FENCE);
+                    if(lcl_compute_win_id < queuecount)
+                    {
+                        for(int i=0; i<lcl_compute_win && (lcl_compute_id==0); i++)
+                        {
+                            stage_sum += partialsum[lcl_id+i];
                         }
-                        barrier(CLK_LOCAL_MEM_FENCE);
-                        if(lcl_compute_win_id < queuecount)
+                        if(stage_sum >= stagethreshold && (lcl_compute_id==0))
                         {
-                            for(int i=0; i<lcl_compute_win && (lcl_compute_id==0); i++)
-                            {
-                                stage_sum += partialsum[lcl_id+i];
-                            }
-                            if(stage_sum >= stagethreshold && (lcl_compute_id==0))
-                            {
-                                int queueindex = atomic_inc(lclcount);
-                                lcloutindex[queueindex<<1] = temp_coord;
-                                lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
-                            }
-                            lcl_compute_win_id +=(1<<perfscale);
+                            int queueindex = atomic_inc(lclcount);
+                            lcloutindex[queueindex<<1] = temp_coord;
+                            lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
                         }
-                        barrier(CLK_LOCAL_MEM_FENCE);
-                    }//end for(int queueloop=0;queueloop<queuecount_loop;queueloop++)
-                    //barrier(CLK_LOCAL_MEM_FENCE);
-                    queuecount = lclcount[0];
+                        lcl_compute_win_id +=(1<<perfscale);
+                    }
                     barrier(CLK_LOCAL_MEM_FENCE);
-                    nodecounter += stageinfo.x;
-                }//end for(int stageloop = splitstage; stageloop< endstage && queuecount>0;stageloop++)
-                //barrier(CLK_LOCAL_MEM_FENCE);
-                if(lcl_id<queuecount)
-                {
-                    int temp = lcloutindex[lcl_id<<1];
-                    int x = mad24(grpidx,grpszx,temp & 0xffff);
-                    int y = mad24(grpidy,grpszy,((temp & (int)0xffff0000) >> 16));
-                    temp = glboutindex[0];
-                    int4 candidate_result;
-                    candidate_result.zw = (int2)convert_int_rtn(factor*20.f);
-                    candidate_result.x = convert_int_rtn(x*factor);
-                    candidate_result.y = convert_int_rtn(y*factor);
-                    atomic_inc(glboutindex);
-                    candidate[outputoff+temp+lcl_id] = candidate_result;
-                }
+                }//end for(int queueloop=0;queueloop<queuecount_loop;queueloop++)
+
+                queuecount = lclcount[0];
                 barrier(CLK_LOCAL_MEM_FENCE);
-            }//end if((x < width) && (y < height))
+                nodecounter += stageinfo.x;
+            }//end for(int stageloop = splitstage; stageloop< endstage && queuecount>0;stageloop++)
+
+            if(lcl_id<queuecount)
+            {
+                int temp = lcloutindex[lcl_id<<1];
+                int x = mad24(grpidx,grpszx,temp & 0xffff);
+                int y = mad24(grpidy,grpszy,((temp & (int)0xffff0000) >> 16));
+                temp = glboutindex[0];
+                int4 candidate_result;
+                candidate_result.zw = (int2)convert_int_rtn(factor*20.f);
+                candidate_result.x = convert_int_rtn(x*factor);
+                candidate_result.y = convert_int_rtn(y*factor);
+                atomic_inc(glboutindex);
+                candidate[outputoff+temp+lcl_id] = candidate_result;
+            }
+            barrier(CLK_LOCAL_MEM_FENCE);
         }//end for(int grploop=grpidx;grploop<totalgrp;grploop+=grpnumx)
-        //outputoff +=mul24(width,height);
     }//end for(int scalei = 0; scalei <loopcount; scalei++)
 }
 

modules/ocl/src/opencl/haarobjectdetect_scaled2.cl

@@ -16,6 +16,7 @@
 //
 // @Authors
 //    Wu Xinglong, wxl370@126.com
+//    Sen Liu, swjtuls1987@126.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -52,11 +53,11 @@ typedef struct  __attribute__((aligned(128)))  GpuHidHaarFeature
 {
     struct __attribute__((aligned(32)))
 {
-        int p0 __attribute__((aligned(4)));
-        int p1 __attribute__((aligned(4)));
-        int p2 __attribute__((aligned(4)));
-        int p3 __attribute__((aligned(4)));
-        float weight __attribute__((aligned(4)));
+    int p0 __attribute__((aligned(4)));
+    int p1 __attribute__((aligned(4)));
+    int p2 __attribute__((aligned(4)));
+    int p3 __attribute__((aligned(4)));
+    float weight __attribute__((aligned(4)));
 }
 rect[CV_HAAR_FEATURE_MAX] __attribute__((aligned(32)));
 }
@@ -113,173 +114,168 @@ __kernel void gpuRunHaarClassifierCascade_scaled2(
     global const int *restrict sum,
     global const float   *restrict sqsum,
     global int4 *candidate,
+    const int rows,
+    const int cols,
     const int step,
     const int loopcount,
     const int start_stage,
     const int split_stage,
     const int end_stage,
     const int startnode,
-    const int splitnode,
     global int4 *p,
-    //const int4 * pq,
     global float *correction,
     const int nodecount)
 {
-        int grpszx = get_local_size(0);
-        int grpszy = get_local_size(1);
-        int grpnumx = get_num_groups(0);
-        int grpidx = get_group_id(0);
-        int lclidx = get_local_id(0);
-        int lclidy = get_local_id(1);
-        int lcl_sz = mul24(grpszx, grpszy);
-        int lcl_id = mad24(lclidy, grpszx, lclidx);
-        __local int lclshare[1024];
-        __local int *glboutindex = lclshare + 0;
-        __local int *lclcount = glboutindex + 1;
-        __local int *lcloutindex = lclcount + 1;
-        __local float *partialsum = (__local float *)(lcloutindex + (lcl_sz << 1));
-        glboutindex[0] = 0;
-        int outputoff = mul24(grpidx, 256);
-        candidate[outputoff + (lcl_id << 2)] = (int4)0;
-        candidate[outputoff + (lcl_id << 2) + 1] = (int4)0;
-        candidate[outputoff + (lcl_id << 2) + 2] = (int4)0;
-        candidate[outputoff + (lcl_id << 2) + 3] = (int4)0;
+    int grpszx = get_local_size(0);
+    int grpszy = get_local_size(1);
+    int grpnumx = get_num_groups(0);
+    int grpidx = get_group_id(0);
+    int lclidx = get_local_id(0);
+    int lclidy = get_local_id(1);
+    int lcl_sz = mul24(grpszx, grpszy);
+    int lcl_id = mad24(lclidy, grpszx, lclidx);
+    __local int glboutindex[1];
+    __local int lclcount[1];
+    __local int lcloutindex[64];
+    glboutindex[0] = 0;
+    int outputoff = mul24(grpidx, 256);
+    candidate[outputoff + (lcl_id << 2)] = (int4)0;
+    candidate[outputoff + (lcl_id << 2) + 1] = (int4)0;
+    candidate[outputoff + (lcl_id << 2) + 2] = (int4)0;
+    candidate[outputoff + (lcl_id << 2) + 3] = (int4)0;
+    int max_idx = rows * cols - 1;
+    for (int scalei = 0; scalei < loopcount; scalei++)
+    {
+        int4 scaleinfo1;
+        scaleinfo1 = info[scalei];
+        int width = (scaleinfo1.x & 0xffff0000) >> 16;
+        int height = scaleinfo1.x & 0xffff;
+        int grpnumperline = (scaleinfo1.y & 0xffff0000) >> 16;
+        int totalgrp = scaleinfo1.y & 0xffff;
+        float factor = as_float(scaleinfo1.w);
+        float correction_t = correction[scalei];
+        int ystep = (int)(max(2.0f, factor) + 0.5f);
 
-        for (int scalei = 0; scalei < loopcount; scalei++)
+        for (int grploop = get_group_id(0); grploop < totalgrp; grploop += grpnumx)
         {
-                int4 scaleinfo1;
-                scaleinfo1 = info[scalei];
-                int width = (scaleinfo1.x & 0xffff0000) >> 16;
-                int height = scaleinfo1.x & 0xffff;
-                int grpnumperline = (scaleinfo1.y & 0xffff0000) >> 16;
-                int totalgrp = scaleinfo1.y & 0xffff;
-                float factor = as_float(scaleinfo1.w);
-                float correction_t = correction[scalei];
-                int ystep = (int)(max(2.0f, factor) + 0.5f);
+            int4 cascadeinfo = p[scalei];
+            int grpidy = grploop / grpnumperline;
+            int grpidx = grploop - mul24(grpidy, grpnumperline);
+            int ix = mad24(grpidx, grpszx, lclidx);
+            int iy = mad24(grpidy, grpszy, lclidy);
+            int x = ix * ystep;
+            int y = iy * ystep;
+            lcloutindex[lcl_id] = 0;
+            lclcount[0] = 0;
+            int nodecounter;
+            float mean, variance_norm_factor;
+            //if((ix < width) && (iy < height))
+            {
+                const int p_offset = mad24(y, step, x);
+                cascadeinfo.x += p_offset;
+                cascadeinfo.z += p_offset;
+                mean = (sum[clamp(mad24(cascadeinfo.y, step, cascadeinfo.x), 0, max_idx)] - sum[clamp(mad24(cascadeinfo.y, step, cascadeinfo.z), 0, max_idx)] -
+                        sum[clamp(mad24(cascadeinfo.w, step, cascadeinfo.x), 0, max_idx)] + sum[clamp(mad24(cascadeinfo.w, step, cascadeinfo.z), 0, max_idx)])
+                       * correction_t;
+                variance_norm_factor = sqsum[clamp(mad24(cascadeinfo.y, step, cascadeinfo.x), 0, max_idx)] - sqsum[clamp(mad24(cascadeinfo.y, step, cascadeinfo.z), 0, max_idx)] -
+                                       sqsum[clamp(mad24(cascadeinfo.w, step, cascadeinfo.x), 0, max_idx)] + sqsum[clamp(mad24(cascadeinfo.w, step, cascadeinfo.z), 0, max_idx)];
+                variance_norm_factor = variance_norm_factor * correction_t - mean * mean;
+                variance_norm_factor = variance_norm_factor >= 0.f ? sqrt(variance_norm_factor) : 1.f;
+                bool result = true;
+                nodecounter = startnode + nodecount * scalei;
 
-                for (int grploop = get_group_id(0); grploop < totalgrp; grploop += grpnumx)
+                for (int stageloop = start_stage; (stageloop < end_stage) && result; stageloop++)
                 {
-                        int4 cascadeinfo = p[scalei];
-                        int grpidy = grploop / grpnumperline;
-                        int grpidx = grploop - mul24(grpidy, grpnumperline);
-                        int ix = mad24(grpidx, grpszx, lclidx);
-                        int iy = mad24(grpidy, grpszy, lclidy);
-                        int x = ix * ystep;
-                        int y = iy * ystep;
-                        lcloutindex[lcl_id] = 0;
-                        lclcount[0] = 0;
-                        int result = 1, nodecounter;
-                        float mean, variance_norm_factor;
-                        //if((ix < width) && (iy < height))
-                        {
-                                const int p_offset = mad24(y, step, x);
-                                cascadeinfo.x += p_offset;
-                                cascadeinfo.z += p_offset;
-                                mean = (sum[mad24(cascadeinfo.y, step, cascadeinfo.x)] - sum[mad24(cascadeinfo.y, step, cascadeinfo.z)] -
-                                        sum[mad24(cascadeinfo.w, step, cascadeinfo.x)] + sum[mad24(cascadeinfo.w, step, cascadeinfo.z)])
-                                       * correction_t;
-                                variance_norm_factor = sqsum[mad24(cascadeinfo.y, step, cascadeinfo.x)] - sqsum[mad24(cascadeinfo.y, step, cascadeinfo.z)] -
-                                                       sqsum[mad24(cascadeinfo.w, step, cascadeinfo.x)] + sqsum[mad24(cascadeinfo.w, step, cascadeinfo.z)];
-                                variance_norm_factor = variance_norm_factor * correction_t - mean * mean;
-                                variance_norm_factor = variance_norm_factor >= 0.f ? sqrt(variance_norm_factor) : 1.f;
-                                result = 1;
-                                nodecounter = startnode + nodecount * scalei;
-
-                                for (int stageloop = start_stage; stageloop < end_stage && result; stageloop++)
-                                {
-                                        float stage_sum = 0.f;
-                                        int4 stageinfo = *(global int4 *)(stagecascadeptr + stageloop);
-                                        float stagethreshold = as_float(stageinfo.y);
-
-                                        for (int nodeloop = 0; nodeloop < stageinfo.x; nodeloop++)
-                                        {
-                                                __global GpuHidHaarTreeNode *currentnodeptr = (nodeptr + nodecounter);
-                                                int4 info1 = *(__global int4 *)(&(currentnodeptr->p[0][0]));
-                                                int4 info2 = *(__global int4 *)(&(currentnodeptr->p[1][0]));
-                                                int4 info3 = *(__global int4 *)(&(currentnodeptr->p[2][0]));
-                                                float4 w = *(__global float4 *)(&(currentnodeptr->weight[0]));
-                                                float2 alpha2 = *(__global float2 *)(&(currentnodeptr->alpha[0]));
-                                                float nodethreshold  = w.w * variance_norm_factor;
-                                                info1.x += p_offset;
-                                                info1.z += p_offset;
-                                                info2.x += p_offset;
-                                                info2.z += p_offset;
-                                                float classsum = (sum[mad24(info1.y, step, info1.x)] - sum[mad24(info1.y, step, info1.z)] -
-                                                                  sum[mad24(info1.w, step, info1.x)] + sum[mad24(info1.w, step, info1.z)]) * w.x;
-                                                classsum += (sum[mad24(info2.y, step, info2.x)] - sum[mad24(info2.y, step, info2.z)] -
-                                                             sum[mad24(info2.w, step, info2.x)] + sum[mad24(info2.w, step, info2.z)]) * w.y;
-                                                info3.x += p_offset;
-                                                info3.z += p_offset;
-                                                classsum += (sum[mad24(info3.y, step, info3.x)] - sum[mad24(info3.y, step, info3.z)] -
-                                                             sum[mad24(info3.w, step, info3.x)] + sum[mad24(info3.w, step, info3.z)]) * w.z;
-                                                stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
-                                                nodecounter++;
-                                        }
-
-                                        result = (stage_sum >= stagethreshold);
-                                }
+                    float stage_sum = 0.f;
+                    int   stagecount = stagecascadeptr[stageloop].count;
+                    for (int nodeloop = 0; nodeloop < stagecount; nodeloop++)
+                    {
+                        __global GpuHidHaarTreeNode *currentnodeptr = (nodeptr + nodecounter);
+                        int4 info1 = *(__global int4 *)(&(currentnodeptr->p[0][0]));
+                        int4 info2 = *(__global int4 *)(&(currentnodeptr->p[1][0]));
+                        int4 info3 = *(__global int4 *)(&(currentnodeptr->p[2][0]));
+                        float4 w = *(__global float4 *)(&(currentnodeptr->weight[0]));
+                        float2 alpha2 = *(__global float2 *)(&(currentnodeptr->alpha[0]));
+                        float nodethreshold  = w.w * variance_norm_factor;
+                        info1.x += p_offset;
+                        info1.z += p_offset;
+                        info2.x += p_offset;
+                        info2.z += p_offset;
+                        float classsum = (sum[clamp(mad24(info1.y, step, info1.x), 0, max_idx)] - sum[clamp(mad24(info1.y, step, info1.z), 0, max_idx)] -
+                                          sum[clamp(mad24(info1.w, step, info1.x), 0, max_idx)] + sum[clamp(mad24(info1.w, step, info1.z), 0, max_idx)]) * w.x;
+                        classsum += (sum[clamp(mad24(info2.y, step, info2.x), 0, max_idx)] - sum[clamp(mad24(info2.y, step, info2.z), 0, max_idx)] -
+                                     sum[clamp(mad24(info2.w, step, info2.x), 0, max_idx)] + sum[clamp(mad24(info2.w, step, info2.z), 0, max_idx)]) * w.y;
+                        info3.x += p_offset;
+                        info3.z += p_offset;
+                        classsum += (sum[clamp(mad24(info3.y, step, info3.x), 0, max_idx)] - sum[clamp(mad24(info3.y, step, info3.z), 0, max_idx)] -
+                                     sum[clamp(mad24(info3.w, step, info3.x), 0, max_idx)] + sum[clamp(mad24(info3.w, step, info3.z), 0, max_idx)]) * w.z;
+                        stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
+                        nodecounter++;
+                    }
+                    result = (bool)(stage_sum >= stagecascadeptr[stageloop].threshold);
+                }
 
-                                if (result && (ix < width) && (iy < height))
-                                {
-                                        int queueindex = atomic_inc(lclcount);
-                                        lcloutindex[queueindex << 1] = (y << 16) | x;
-                                        lcloutindex[(queueindex << 1) + 1] = as_int(variance_norm_factor);
-                                }
+                barrier(CLK_LOCAL_MEM_FENCE);
 
-                                barrier(CLK_LOCAL_MEM_FENCE);
-                                int queuecount = lclcount[0];
-                                nodecounter = splitnode + nodecount * scalei;
+                if (result && (ix < width) && (iy < height))
+                {
+                    int queueindex = atomic_inc(lclcount);
+                    lcloutindex[queueindex] = (y << 16) | x;
+                }
 
-                                if (lcl_id < queuecount)
-                                {
-                                        int temp = lcloutindex[lcl_id << 1];
-                                        int x = temp & 0xffff;
-                                        int y = (temp & (int)0xffff0000) >> 16;
-                                        temp = glboutindex[0];
-                                        int4 candidate_result;
-                                        candidate_result.zw = (int2)convert_int_rtn(factor * 20.f);
-                                        candidate_result.x = x;
-                                        candidate_result.y = y;
-                                        atomic_inc(glboutindex);
-                                        candidate[outputoff + temp + lcl_id] = candidate_result;
-                                }
+                barrier(CLK_LOCAL_MEM_FENCE);
+                int queuecount = lclcount[0];
 
-                                barrier(CLK_LOCAL_MEM_FENCE);
-                        }
+                if (lcl_id < queuecount)
+                {
+                    int temp = lcloutindex[lcl_id];
+                    int x = temp & 0xffff;
+                    int y = (temp & (int)0xffff0000) >> 16;
+                    temp = atomic_inc(glboutindex);
+                    int4 candidate_result;
+                    candidate_result.zw = (int2)convert_int_rtn(factor * 20.f);
+                    candidate_result.x = x;
+                    candidate_result.y = y;
+                    candidate[outputoff + temp + lcl_id] = candidate_result;
                 }
+
+                barrier(CLK_LOCAL_MEM_FENCE);
+            }
         }
+    }
 }
 __kernel void gpuscaleclassifier(global GpuHidHaarTreeNode *orinode, global GpuHidHaarTreeNode *newnode, float scale, float weight_scale, int nodenum)
 {
-        int counter = get_global_id(0);
-        int tr_x[3], tr_y[3], tr_h[3], tr_w[3], i = 0;
-        GpuHidHaarTreeNode t1 = *(orinode + counter);
+    int counter = get_global_id(0);
+    int tr_x[3], tr_y[3], tr_h[3], tr_w[3], i = 0;
+    GpuHidHaarTreeNode t1 = *(orinode + counter);
 #pragma unroll
 
-        for (i = 0; i < 3; i++)
-        {
-                tr_x[i] = (int)(t1.p[i][0] * scale + 0.5f);
-                tr_y[i] = (int)(t1.p[i][1] * scale + 0.5f);
-                tr_w[i] = (int)(t1.p[i][2] * scale + 0.5f);
-                tr_h[i] = (int)(t1.p[i][3] * scale + 0.5f);
-        }
+    for (i = 0; i < 3; i++)
+    {
+        tr_x[i] = (int)(t1.p[i][0] * scale + 0.5f);
+        tr_y[i] = (int)(t1.p[i][1] * scale + 0.5f);
+        tr_w[i] = (int)(t1.p[i][2] * scale + 0.5f);
+        tr_h[i] = (int)(t1.p[i][3] * scale + 0.5f);
+    }
 
-        t1.weight[0] = t1.p[2][0] ? -(t1.weight[1] * tr_h[1] * tr_w[1] + t1.weight[2] * tr_h[2] * tr_w[2]) / (tr_h[0] * tr_w[0]) : -t1.weight[1] * tr_h[1] * tr_w[1] / (tr_h[0] * tr_w[0]);
-        counter += nodenum;
+    t1.weight[0] = t1.p[2][0] ? -(t1.weight[1] * tr_h[1] * tr_w[1] + t1.weight[2] * tr_h[2] * tr_w[2]) / (tr_h[0] * tr_w[0]) : -t1.weight[1] * tr_h[1] * tr_w[1] / (tr_h[0] * tr_w[0]);
+    counter += nodenum;
 #pragma unroll
 
-        for (i = 0; i < 3; i++)
-        {
-                newnode[counter].p[i][0] = tr_x[i];
-                newnode[counter].p[i][1] = tr_y[i];
-                newnode[counter].p[i][2] = tr_x[i] + tr_w[i];
-                newnode[counter].p[i][3] = tr_y[i] + tr_h[i];
-                newnode[counter].weight[i] = t1.weight[i] * weight_scale;
-        }
+    for (i = 0; i < 3; i++)
+    {
+        newnode[counter].p[i][0] = tr_x[i];
+        newnode[counter].p[i][1] = tr_y[i];
+        newnode[counter].p[i][2] = tr_x[i] + tr_w[i];
+        newnode[counter].p[i][3] = tr_y[i] + tr_h[i];
+        newnode[counter].weight[i] = t1.weight[i] * weight_scale;
+    }
 
-        newnode[counter].left = t1.left;
-        newnode[counter].right = t1.right;
-        newnode[counter].threshold = t1.threshold;
-        newnode[counter].alpha[0] = t1.alpha[0];
-        newnode[counter].alpha[1] = t1.alpha[1];
+    newnode[counter].left = t1.left;
+    newnode[counter].right = t1.right;
+    newnode[counter].threshold = t1.threshold;
+    newnode[counter].alpha[0] = t1.alpha[0];
+    newnode[counter].alpha[1] = t1.alpha[1];
 }
+

modules/ocl/src/opencl/imgproc_integral.cl

@@ -60,7 +60,7 @@
 #define GET_CONFLICT_OFFSET(lid) ((lid) >> LOG_NUM_BANKS)
 
 
-kernel void integral_cols(__global uchar4 *src,__global int *sum ,__global float *sqsum,
+kernel void integral_cols_D4(__global uchar4 *src,__global int *sum ,__global float *sqsum,
                           int src_offset,int pre_invalid,int rows,int cols,int src_step,int dst_step)
 {
     unsigned int lid = get_local_id(0);
@@ -159,7 +159,7 @@ kernel void integral_cols(__global uchar4 *src,__global int *sum ,__global float
 }
 
 
-kernel void integral_rows(__global int4 *srcsum,__global float4 * srcsqsum,__global int *sum ,
+kernel void integral_rows_D4(__global int4 *srcsum,__global float4 * srcsqsum,__global int *sum ,
                           __global float *sqsum,int rows,int cols,int src_step,int sum_step,
                           int sqsum_step,int sum_offset,int sqsum_offset)
 {
@@ -275,3 +275,219 @@ kernel void integral_rows(__global int4 *srcsum,__global float4 * srcsqsum,__glo
         barrier(CLK_LOCAL_MEM_FENCE);
     }
 }
+
+kernel void integral_cols_D5(__global uchar4 *src,__global float *sum ,__global float *sqsum,
+                          int src_offset,int pre_invalid,int rows,int cols,int src_step,int dst_step)
+{
+    unsigned int lid = get_local_id(0);
+    unsigned int gid = get_group_id(0);
+    float4 src_t[2], sum_t[2];
+    float4 sqsum_t[2];
+    __local float4 lm_sum[2][LSIZE + LOG_LSIZE];
+    __local float4 lm_sqsum[2][LSIZE + LOG_LSIZE];
+    __local float* sum_p;
+    __local float* sqsum_p;
+    src_step = src_step >> 2;
+    gid = gid << 1;
+    for(int i = 0; i < rows; i =i + LSIZE_1)
+    {
+        src_t[0] = (i + lid < rows ? convert_float4(src[src_offset + (lid+i) * src_step + min(gid, (uint)cols - 1)]) : (float4)0);
+        src_t[1] = (i + lid < rows ? convert_float4(src[src_offset + (lid+i) * src_step + min(gid + 1, (uint)cols - 1)]) : (float4)0);
+
+        sum_t[0] = (i == 0 ? (float4)0 : lm_sum[0][LSIZE_2 + LOG_LSIZE]);
+        sqsum_t[0] = (i == 0 ? (float4)0 : lm_sqsum[0][LSIZE_2 + LOG_LSIZE]);
+        sum_t[1] =  (i == 0 ? (float4)0 : lm_sum[1][LSIZE_2 + LOG_LSIZE]);
+        sqsum_t[1] =  (i == 0 ? (float4)0 : lm_sqsum[1][LSIZE_2 + LOG_LSIZE]);
+        barrier(CLK_LOCAL_MEM_FENCE);
+
+        int bf_loc = lid + GET_CONFLICT_OFFSET(lid);
+        lm_sum[0][bf_loc] = src_t[0];
+        lm_sqsum[0][bf_loc] = convert_float4(src_t[0] * src_t[0]);
+
+        lm_sum[1][bf_loc] = src_t[1];
+        lm_sqsum[1][bf_loc] = convert_float4(src_t[1] * src_t[1]);
+
+        int offset = 1;
+        for(int d = LSIZE >> 1 ;  d > 0; d>>=1)
+        {
+            barrier(CLK_LOCAL_MEM_FENCE);
+            int ai = offset * (((lid & 127)<<1) +1) - 1,bi = ai + offset;
+            ai += GET_CONFLICT_OFFSET(ai);
+            bi += GET_CONFLICT_OFFSET(bi);
+
+            if((lid & 127) < d)
+            {
+                lm_sum[lid >> 7][bi]  +=  lm_sum[lid >> 7][ai];
+                lm_sqsum[lid >> 7][bi]  +=  lm_sqsum[lid >> 7][ai];
+            }
+            offset <<= 1;
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+        if(lid < 2)
+        {
+            lm_sum[lid][LSIZE_2 + LOG_LSIZE] = 0;
+            lm_sqsum[lid][LSIZE_2 + LOG_LSIZE] = 0;
+        }
+        for(int d = 1;  d < LSIZE; d <<= 1)
+        {
+            barrier(CLK_LOCAL_MEM_FENCE);
+            offset >>= 1;
+            int ai = offset * (((lid & 127)<<1) +1) - 1,bi = ai + offset;
+            ai += GET_CONFLICT_OFFSET(ai);
+            bi += GET_CONFLICT_OFFSET(bi);
+
+            if((lid & 127) < d)
+            {
+                lm_sum[lid >> 7][bi] += lm_sum[lid >> 7][ai];
+                lm_sum[lid >> 7][ai] = lm_sum[lid >> 7][bi] - lm_sum[lid >> 7][ai];
+
+                lm_sqsum[lid >> 7][bi] += lm_sqsum[lid >> 7][ai];
+                lm_sqsum[lid >> 7][ai] = lm_sqsum[lid >> 7][bi] - lm_sqsum[lid >> 7][ai];
+            }
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+        int loc_s0 = gid * dst_step + i + lid - 1 - pre_invalid * dst_step / 4, loc_s1 = loc_s0 + dst_step ;
+        if(lid > 0 && (i+lid) <= rows)
+        {
+            lm_sum[0][bf_loc] += sum_t[0];
+            lm_sum[1][bf_loc] += sum_t[1];
+            lm_sqsum[0][bf_loc] += sqsum_t[0];
+            lm_sqsum[1][bf_loc] += sqsum_t[1];
+            sum_p = (__local float*)(&(lm_sum[0][bf_loc]));
+            sqsum_p = (__local float*)(&(lm_sqsum[0][bf_loc]));
+            for(int k = 0; k < 4; k++)
+            {
+                if(gid * 4 + k >= cols + pre_invalid || gid * 4 + k < pre_invalid) continue;
+                sum[loc_s0 + k * dst_step / 4] = sum_p[k];
+                sqsum[loc_s0 + k * dst_step / 4] = sqsum_p[k];
+            }
+            sum_p = (__local float*)(&(lm_sum[1][bf_loc]));
+            sqsum_p = (__local float*)(&(lm_sqsum[1][bf_loc]));
+            for(int k = 0; k < 4; k++)
+            {
+                if(gid * 4 + k + 4 >= cols + pre_invalid) break;
+                sum[loc_s1 + k * dst_step / 4] = sum_p[k];
+                sqsum[loc_s1 + k * dst_step / 4] = sqsum_p[k];
+            }
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+    }
+}
+
+
+kernel void integral_rows_D5(__global float4 *srcsum,__global float4 * srcsqsum,__global float *sum ,
+                          __global float *sqsum,int rows,int cols,int src_step,int sum_step,
+                          int sqsum_step,int sum_offset,int sqsum_offset)
+{
+    unsigned int lid = get_local_id(0);
+    unsigned int gid = get_group_id(0);
+    float4 src_t[2], sum_t[2];
+    float4 sqsrc_t[2],sqsum_t[2];
+    __local float4 lm_sum[2][LSIZE + LOG_LSIZE];
+    __local float4 lm_sqsum[2][LSIZE + LOG_LSIZE];
+    __local float *sum_p;
+    __local float *sqsum_p;
+    src_step = src_step >> 4;
+    for(int i = 0; i < rows; i =i + LSIZE_1)
+    {
+        src_t[0] = i + lid < rows ? srcsum[(lid+i) * src_step + gid * 2] : (float4)0;
+        sqsrc_t[0] = i + lid < rows ? srcsqsum[(lid+i) * src_step + gid * 2] : (float4)0;
+        src_t[1] = i + lid < rows ? srcsum[(lid+i) * src_step + gid * 2 + 1] : (float4)0;
+        sqsrc_t[1] = i + lid < rows ? srcsqsum[(lid+i) * src_step + gid * 2 + 1] : (float4)0;
+
+        sum_t[0] =  (i == 0 ? (float4)0 : lm_sum[0][LSIZE_2 + LOG_LSIZE]);
+        sqsum_t[0] =  (i == 0 ? (float4)0 : lm_sqsum[0][LSIZE_2 + LOG_LSIZE]);
+        sum_t[1] =  (i == 0 ? (float4)0 : lm_sum[1][LSIZE_2 + LOG_LSIZE]);
+        sqsum_t[1] =  (i == 0 ? (float4)0 : lm_sqsum[1][LSIZE_2 + LOG_LSIZE]);
+        barrier(CLK_LOCAL_MEM_FENCE);
+
+        int bf_loc = lid + GET_CONFLICT_OFFSET(lid);
+        lm_sum[0][bf_loc] = src_t[0];
+        lm_sqsum[0][bf_loc] = sqsrc_t[0];
+
+        lm_sum[1][bf_loc] = src_t[1];
+        lm_sqsum[1][bf_loc] = sqsrc_t[1];
+
+        int offset = 1;
+        for(int d = LSIZE >> 1 ;  d > 0; d>>=1)
+        {
+            barrier(CLK_LOCAL_MEM_FENCE);
+            int ai = offset * (((lid & 127)<<1) +1) - 1,bi = ai + offset;
+            ai += GET_CONFLICT_OFFSET(ai);
+            bi += GET_CONFLICT_OFFSET(bi);
+
+            if((lid & 127) < d)
+            {
+                lm_sum[lid >> 7][bi]  +=  lm_sum[lid >> 7][ai];
+                lm_sqsum[lid >> 7][bi]  +=  lm_sqsum[lid >> 7][ai];
+            }
+            offset <<= 1;
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+        if(lid < 2)
+        {
+            lm_sum[lid][LSIZE_2 + LOG_LSIZE] = 0;
+            lm_sqsum[lid][LSIZE_2 + LOG_LSIZE] = 0;
+        }
+        for(int d = 1;  d < LSIZE; d <<= 1)
+        {
+            barrier(CLK_LOCAL_MEM_FENCE);
+            offset >>= 1;
+            int ai = offset * (((lid & 127)<<1) +1) - 1,bi = ai + offset;
+            ai += GET_CONFLICT_OFFSET(ai);
+            bi += GET_CONFLICT_OFFSET(bi);
+
+            if((lid & 127) < d)
+            {
+                lm_sum[lid >> 7][bi] += lm_sum[lid >> 7][ai];
+                lm_sum[lid >> 7][ai] = lm_sum[lid >> 7][bi] - lm_sum[lid >> 7][ai];
+
+                lm_sqsum[lid >> 7][bi] += lm_sqsum[lid >> 7][ai];
+                lm_sqsum[lid >> 7][ai] = lm_sqsum[lid >> 7][bi] - lm_sqsum[lid >> 7][ai];
+            }
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+        if(gid == 0 && (i + lid) <= rows)
+        {
+            sum[sum_offset + i + lid] = 0;
+            sqsum[sqsum_offset + i + lid] = 0;
+        }
+        if(i + lid == 0)
+        {
+            int loc0 = gid * 2 * sum_step;
+            int loc1 = gid * 2 * sqsum_step;
+            for(int k = 1; k <= 8; k++)
+            {
+                if(gid * 8 + k > cols) break;
+                sum[sum_offset + loc0 + k * sum_step / 4] = 0;
+                sqsum[sqsum_offset + loc1 + k * sqsum_step / 4] = 0;
+            }
+        }
+        int loc_s0 = sum_offset + gid * 2 * sum_step + sum_step / 4 + i + lid, loc_s1 = loc_s0 + sum_step ;
+        int loc_sq0 = sqsum_offset + gid * 2 * sqsum_step + sqsum_step / 4 + i + lid, loc_sq1 = loc_sq0 + sqsum_step ;
+        if(lid > 0 && (i+lid) <= rows)
+        {
+            lm_sum[0][bf_loc] += sum_t[0];
+            lm_sum[1][bf_loc] += sum_t[1];
+            lm_sqsum[0][bf_loc] += sqsum_t[0];
+            lm_sqsum[1][bf_loc] += sqsum_t[1];
+            sum_p = (__local float*)(&(lm_sum[0][bf_loc]));
+            sqsum_p = (__local float*)(&(lm_sqsum[0][bf_loc]));
+            for(int k = 0; k < 4; k++)
+            {
+                if(gid * 8 + k >= cols) break;
+                sum[loc_s0 + k * sum_step / 4] = sum_p[k];
+                sqsum[loc_sq0 + k * sqsum_step / 4] = sqsum_p[k];
+            }
+            sum_p = (__local float*)(&(lm_sum[1][bf_loc]));
+            sqsum_p = (__local float*)(&(lm_sqsum[1][bf_loc]));
+            for(int k = 0; k < 4; k++)
+            {
+                if(gid * 8 + 4 + k >= cols) break;
+                sum[loc_s1 + k * sum_step / 4] = sum_p[k];
+                sqsum[loc_sq1 + k * sqsum_step / 4] = sqsum_p[k];
+            }
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+    }
+}

modules/ocl/src/opencl/imgproc_integral_sum.cl

@@ -44,8 +44,13 @@
 //M*/
 
 #if defined (DOUBLE_SUPPORT)
+#ifdef cl_khr_fp64
 #pragma OPENCL EXTENSION cl_khr_fp64:enable
+#elif defined (cl_amd_fp64)
+#pragma OPENCL EXTENSION cl_amd_fp64:enable
 #endif
+#endif
+
 #define LSIZE 256
 #define LSIZE_1 255
 #define LSIZE_2 254
@@ -56,8 +61,8 @@
 #define GET_CONFLICT_OFFSET(lid) ((lid) >> LOG_NUM_BANKS)
 
 
-kernel void integral_sum_cols(__global uchar4 *src,__global int *sum ,
-                          int src_offset,int pre_invalid,int rows,int cols,int src_step,int dst_step)
+kernel void integral_sum_cols_D4(__global uchar4 *src,__global int *sum ,
+                              int src_offset,int pre_invalid,int rows,int cols,int src_step,int dst_step)
 {
     unsigned int lid = get_local_id(0);
     unsigned int gid = get_group_id(0);
@@ -114,7 +119,8 @@ kernel void integral_sum_cols(__global uchar4 *src,__global int *sum ,
             }
         }
         barrier(CLK_LOCAL_MEM_FENCE);
-        if(lid > 0 && (i+lid) <= rows){
+        if(lid > 0 && (i+lid) <= rows)
+        {
             int loc_s0 = gid * dst_step + i + lid - 1 - pre_invalid * dst_step / 4, loc_s1 = loc_s0 + dst_step ;
             lm_sum[0][bf_loc] += sum_t[0];
             lm_sum[1][bf_loc] += sum_t[1];
@@ -136,9 +142,9 @@ kernel void integral_sum_cols(__global uchar4 *src,__global int *sum ,
 }
 
 
-kernel void integral_sum_rows(__global int4 *srcsum,__global int *sum ,
-                          int rows,int cols,int src_step,int sum_step,
-                          int sum_offset)
+kernel void integral_sum_rows_D4(__global int4 *srcsum,__global int *sum ,
+                              int rows,int cols,int src_step,int sum_step,
+                              int sum_offset)
 {
     unsigned int lid = get_local_id(0);
     unsigned int gid = get_group_id(0);
@@ -196,19 +202,20 @@ kernel void integral_sum_rows(__global int4 *srcsum,__global int *sum ,
         barrier(CLK_LOCAL_MEM_FENCE);
         if(gid == 0 && (i + lid) <= rows)
         {
-           sum[sum_offset + i + lid] = 0;
+            sum[sum_offset + i + lid] = 0;
         }
         if(i + lid == 0)
         {
             int loc0 = gid * 2 * sum_step;
-            for(int k = 1;k <= 8;k++)
+            for(int k = 1; k <= 8; k++)
             {
                 if(gid * 8 + k > cols) break;
                 sum[sum_offset + loc0 + k * sum_step / 4] = 0;
             }
         }
 
-        if(lid > 0 && (i+lid) <= rows){
+        if(lid > 0 && (i+lid) <= rows)
+        {
             int loc_s0 = sum_offset + gid * 2 * sum_step + sum_step / 4 + i + lid, loc_s1 = loc_s0 + sum_step ;
             lm_sum[0][bf_loc] += sum_t[0];
             lm_sum[1][bf_loc] += sum_t[1];
@@ -228,3 +235,178 @@ kernel void integral_sum_rows(__global int4 *srcsum,__global int *sum ,
         barrier(CLK_LOCAL_MEM_FENCE);
     }
 }
+
+kernel void integral_sum_cols_D5(__global uchar4 *src,__global float *sum ,
+                              int src_offset,int pre_invalid,int rows,int cols,int src_step,int dst_step)
+{
+    unsigned int lid = get_local_id(0);
+    unsigned int gid = get_group_id(0);
+    float4 src_t[2], sum_t[2];
+    __local float4 lm_sum[2][LSIZE + LOG_LSIZE];
+    __local float* sum_p;
+    src_step = src_step >> 2;
+    gid = gid << 1;
+    for(int i = 0; i < rows; i =i + LSIZE_1)
+    {
+        src_t[0] = (i + lid < rows ? convert_float4(src[src_offset + (lid+i) * src_step + gid]) : (float4)0);
+        src_t[1] = (i + lid < rows ? convert_float4(src[src_offset + (lid+i) * src_step + gid + 1]) : (float4)0);
+
+        sum_t[0] =  (i == 0 ? (float4)0 : lm_sum[0][LSIZE_2 + LOG_LSIZE]);
+        sum_t[1] =  (i == 0 ? (float4)0 : lm_sum[1][LSIZE_2 + LOG_LSIZE]);
+        barrier(CLK_LOCAL_MEM_FENCE);
+
+        int bf_loc = lid + GET_CONFLICT_OFFSET(lid);
+        lm_sum[0][bf_loc] = src_t[0];
+
+        lm_sum[1][bf_loc] = src_t[1];
+
+        int offset = 1;
+        for(int d = LSIZE >> 1 ;  d > 0; d>>=1)
+        {
+            barrier(CLK_LOCAL_MEM_FENCE);
+            int ai = offset * (((lid & 127)<<1) +1) - 1,bi = ai + offset;
+            ai += GET_CONFLICT_OFFSET(ai);
+            bi += GET_CONFLICT_OFFSET(bi);
+
+            if((lid & 127) < d)
+            {
+                lm_sum[lid >> 7][bi]  +=  lm_sum[lid >> 7][ai];
+            }
+            offset <<= 1;
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+        if(lid < 2)
+        {
+            lm_sum[lid][LSIZE_2 + LOG_LSIZE] = 0;
+        }
+        for(int d = 1;  d < LSIZE; d <<= 1)
+        {
+            barrier(CLK_LOCAL_MEM_FENCE);
+            offset >>= 1;
+            int ai = offset * (((lid & 127)<<1) +1) - 1,bi = ai + offset;
+            ai += GET_CONFLICT_OFFSET(ai);
+            bi += GET_CONFLICT_OFFSET(bi);
+
+            if((lid & 127) < d)
+            {
+                lm_sum[lid >> 7][bi] += lm_sum[lid >> 7][ai];
+                lm_sum[lid >> 7][ai] = lm_sum[lid >> 7][bi] - lm_sum[lid >> 7][ai];
+            }
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+        if(lid > 0 && (i+lid) <= rows)
+        {
+            int loc_s0 = gid * dst_step + i + lid - 1 - pre_invalid * dst_step / 4, loc_s1 = loc_s0 + dst_step ;
+            lm_sum[0][bf_loc] += sum_t[0];
+            lm_sum[1][bf_loc] += sum_t[1];
+            sum_p = (__local float*)(&(lm_sum[0][bf_loc]));
+            for(int k = 0; k < 4; k++)
+            {
+                if(gid * 4 + k >= cols + pre_invalid || gid * 4 + k < pre_invalid) continue;
+                sum[loc_s0 + k * dst_step / 4] = sum_p[k];
+            }
+            sum_p = (__local float*)(&(lm_sum[1][bf_loc]));
+            for(int k = 0; k < 4; k++)
+            {
+                if(gid * 4 + k + 4 >= cols + pre_invalid) break;
+                sum[loc_s1 + k * dst_step / 4] = sum_p[k];
+            }
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+    }
+}
+
+
+kernel void integral_sum_rows_D5(__global float4 *srcsum,__global float *sum ,
+                              int rows,int cols,int src_step,int sum_step,
+                              int sum_offset)
+{
+    unsigned int lid = get_local_id(0);
+    unsigned int gid = get_group_id(0);
+    float4 src_t[2], sum_t[2];
+    __local float4 lm_sum[2][LSIZE + LOG_LSIZE];
+    __local float *sum_p;
+    src_step = src_step >> 4;
+    for(int i = 0; i < rows; i =i + LSIZE_1)
+    {
+        src_t[0] = i + lid < rows ? srcsum[(lid+i) * src_step + gid * 2] : (float4)0;
+        src_t[1] = i + lid < rows ? srcsum[(lid+i) * src_step + gid * 2 + 1] : (float4)0;
+
+        sum_t[0] =  (i == 0 ? (float4)0 : lm_sum[0][LSIZE_2 + LOG_LSIZE]);
+        sum_t[1] =  (i == 0 ? (float4)0 : lm_sum[1][LSIZE_2 + LOG_LSIZE]);
+        barrier(CLK_LOCAL_MEM_FENCE);
+
+        int bf_loc = lid + GET_CONFLICT_OFFSET(lid);
+        lm_sum[0][bf_loc] = src_t[0];
+
+        lm_sum[1][bf_loc] = src_t[1];
+
+        int offset = 1;
+        for(int d = LSIZE >> 1 ;  d > 0; d>>=1)
+        {
+            barrier(CLK_LOCAL_MEM_FENCE);
+            int ai = offset * (((lid & 127)<<1) +1) - 1,bi = ai + offset;
+            ai += GET_CONFLICT_OFFSET(ai);
+            bi += GET_CONFLICT_OFFSET(bi);
+
+            if((lid & 127) < d)
+            {
+                lm_sum[lid >> 7][bi]  +=  lm_sum[lid >> 7][ai];
+            }
+            offset <<= 1;
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+        if(lid < 2)
+        {
+            lm_sum[lid][LSIZE_2 + LOG_LSIZE] = 0;
+        }
+        for(int d = 1;  d < LSIZE; d <<= 1)
+        {
+            barrier(CLK_LOCAL_MEM_FENCE);
+            offset >>= 1;
+            int ai = offset * (((lid & 127)<<1) +1) - 1,bi = ai + offset;
+            ai += GET_CONFLICT_OFFSET(ai);
+            bi += GET_CONFLICT_OFFSET(bi);
+
+            if((lid & 127) < d)
+            {
+                lm_sum[lid >> 7][bi] += lm_sum[lid >> 7][ai];
+                lm_sum[lid >> 7][ai] = lm_sum[lid >> 7][bi] - lm_sum[lid >> 7][ai];
+            }
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+        if(gid == 0 && (i + lid) <= rows)
+        {
+            sum[sum_offset + i + lid] = 0;
+        }
+        if(i + lid == 0)
+        {
+            int loc0 = gid * 2 * sum_step;
+            for(int k = 1; k <= 8; k++)
+            {
+                if(gid * 8 + k > cols) break;
+                sum[sum_offset + loc0 + k * sum_step / 4] = 0;
+            }
+        }
+
+        if(lid > 0 && (i+lid) <= rows)
+        {
+            int loc_s0 = sum_offset + gid * 2 * sum_step + sum_step / 4 + i + lid, loc_s1 = loc_s0 + sum_step ;
+            lm_sum[0][bf_loc] += sum_t[0];
+            lm_sum[1][bf_loc] += sum_t[1];
+            sum_p = (__local float*)(&(lm_sum[0][bf_loc]));
+            for(int k = 0; k < 4; k++)
+            {
+                if(gid * 8 + k >= cols) break;
+                sum[loc_s0 + k * sum_step / 4] = sum_p[k];
+            }
+            sum_p = (__local float*)(&(lm_sum[1][bf_loc]));
+            for(int k = 0; k < 4; k++)
+            {
+                if(gid * 8 + 4 + k >= cols) break;
+                sum[loc_s1 + k * sum_step / 4] = sum_p[k];
+            }
+        }
+        barrier(CLK_LOCAL_MEM_FENCE);
+    }
+}

modules/ocl/src/opencl/moments.cl

@@ -1,3 +1,48 @@
+/*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) 2010-2012, Multicoreware, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// @Authors
+//    Sen Liu, swjtuls1987@126.com
+//
+// 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 oclMaterials 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*/
+
 #if defined (DOUBLE_SUPPORT)
 
 #ifdef cl_khr_fp64
@@ -609,22 +654,33 @@ __kernel void CvMoments_D5( __global float* src_data, int src_rows, int src_cols
     int y = wgidy*TILE_SIZE;  // real Y index of pixel
     int x = wgidx*TILE_SIZE;  // real X index of pixel
     int kcn = (cn==2)?2:4;
-    int rstep = min(src_step/4, TILE_SIZE);
+    src_step /= sizeof(*src_data);
+    int rstep = min(src_step, TILE_SIZE);
     tileSize_height = min(TILE_SIZE, src_rows - y);
     tileSize_width = min(TILE_SIZE, src_cols -x);
-    if(tileSize_width < TILE_SIZE)
-        for(int i = tileSize_width; i < rstep; i++ )
-            *((__global float*)src_data+(y+lidy)*src_step/4+x+i) = 0;
+    int maxIdx = mul24(src_rows, src_cols);
+    int yOff = (y+lidy)*src_step;
+    int index;
+    if(tileSize_width < TILE_SIZE && yOff < src_rows)
+        for(int i = tileSize_width; i < rstep && (yOff+x+i) < maxIdx; i++ )
+            *(src_data+yOff+x+i) = 0;
     if( coi > 0 )
         for(int i=0; i < tileSize_width; i+=VLEN_F)
         {
+#pragma unroll
             for(int j=0; j<4; j++)
-                tmp_coi[j] = *(src_data+(y+lidy)*src_step/4+(x+i+j)*kcn+coi-1);
+            {
+                index = yOff+(x+i+j)*kcn+coi-1;
+                if (index < maxIdx)
+                    tmp_coi[j] = *(src_data+index);
+                else
+                    tmp_coi[j] = 0;
+            }
             tmp[i/VLEN_F] = (float4)(tmp_coi[0],tmp_coi[1],tmp_coi[2],tmp_coi[3]);
         }
     else
-        for(int i=0; i < tileSize_width; i+=VLEN_F)
-            tmp[i/VLEN_F] = (float4)(*(src_data+(y+lidy)*src_step/4+x+i),*(src_data+(y+lidy)*src_step/4+x+i+1),*(src_data+(y+lidy)*src_step/4+x+i+2),*(src_data+(y+lidy)*src_step/4+x+i+3));
+        for(int i=0; i < tileSize_width && (yOff+x+i) < maxIdx; i+=VLEN_F)
+            tmp[i/VLEN_F] = (*(__global float4 *)(src_data+yOff+x+i));
     float4 zero = (float4)(0);
     float4 full = (float4)(255);
     if( binary )
@@ -714,35 +770,59 @@ __kernel void CvMoments_D5( __global float* src_data, int src_rows, int src_cols
         // accumulate moments computed in each tile
         dst_step /= sizeof(F);
 
+        int dst_x_off = mad24(wgidy, dst_cols, wgidx);
+        int dst_off = 0;
+        int max_dst_index = 10 * blocky * get_global_size(1);
+
         // + m00 ( = m00' )
-        *(dst_m + mad24(DST_ROW_00 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[0];
+        dst_off = mad24(DST_ROW_00 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[0];
 
         // + m10 ( = m10' + x*m00' )
-        *(dst_m + mad24(DST_ROW_10 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[1] + xm;
+        dst_off = mad24(DST_ROW_10 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[1] + xm;
 
         // + m01 ( = m01' + y*m00' )
-        *(dst_m + mad24(DST_ROW_01 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[2] + ym;
+        dst_off = mad24(DST_ROW_01 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[2] + ym;
 
         // + m20 ( = m20' + 2*x*m10' + x*x*m00' )
-        *(dst_m + mad24(DST_ROW_20 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[3] + x * (mom[1] * 2 + xm);
+        dst_off = mad24(DST_ROW_20 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[3] + x * (mom[1] * 2 + xm);
 
         // + m11 ( = m11' + x*m01' + y*m10' + x*y*m00' )
-        *(dst_m + mad24(DST_ROW_11 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[4] + x * (mom[2] + ym) + y * mom[1];
+        dst_off = mad24(DST_ROW_11 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[4] + x * (mom[2] + ym) + y * mom[1];
 
         // + m02 ( = m02' + 2*y*m01' + y*y*m00' )
-        *(dst_m + mad24(DST_ROW_02 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[5] + y * (mom[2] * 2 + ym);
+        dst_off = mad24(DST_ROW_02 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[5] + y * (mom[2] * 2 + ym);
 
         // + m30 ( = m30' + 3*x*m20' + 3*x*x*m10' + x*x*x*m00' )
-        *(dst_m + mad24(DST_ROW_30 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[6] + x * (3. * mom[3] + x * (3. * mom[1] + xm));
+        dst_off = mad24(DST_ROW_30 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[6] + x * (3. * mom[3] + x * (3. * mom[1] + xm));
 
         // + m21 ( = m21' + x*(2*m11' + 2*y*m10' + x*m01' + x*y*m00') + y*m20')
-        *(dst_m + mad24(DST_ROW_21 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[7] + x * (2 * (mom[4] + y * mom[1]) + x * (mom[2] + ym)) + y * mom[3];
+        dst_off = mad24(DST_ROW_21 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[7] + x * (2 * (mom[4] + y * mom[1]) + x * (mom[2] + ym)) + y * mom[3];
 
         // + m12 ( = m12' + y*(2*m11' + 2*x*m01' + y*m10' + x*y*m00') + x*m02')
-        *(dst_m + mad24(DST_ROW_12 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[8] + y * (2 * (mom[4] + x * mom[2]) + y * (mom[1] + xm)) + x * mom[5];
+        dst_off = mad24(DST_ROW_12 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[8] + y * (2 * (mom[4] + x * mom[2]) + y * (mom[1] + xm)) + x * mom[5];
 
         // + m03 ( = m03' + 3*y*m02' + 3*y*y*m01' + y*y*y*m00' )
-        *(dst_m + mad24(DST_ROW_03 * blocky, dst_step, mad24(wgidy, dst_cols, wgidx))) = mom[9] + y * (3. * mom[5] + y * (3. * mom[2] + ym));
+        dst_off = mad24(DST_ROW_03 * blocky, dst_step, dst_x_off);
+        if (dst_off < max_dst_index)
+            *(dst_m + dst_off) = mom[9] + y * (3. * mom[5] + y * (3. * mom[2] + ym));
     }
 }
 

modules/ocl/src/opencl/pyr_up.cl

@@ -18,6 +18,7 @@
 //    Zhang Chunpeng	chunpeng@multicorewareinc.com
 //    Dachuan Zhao, dachuan@multicorewareinc.com
 //    Yao Wang, yao@multicorewareinc.com
+//    Peng Xiao, pengxiao@outlook.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -47,7 +48,7 @@
 
 //#pragma OPENCL EXTENSION cl_amd_printf : enable
 
-uchar get_valid_uchar(uchar data)
+uchar get_valid_uchar(float data)
 {
     return (uchar)(data <= 255 ? data : data > 0 ? 255 : 0);
 }
@@ -142,7 +143,7 @@ __kernel void pyrUp_C1_D0(__global uchar* src,__global uchar* dst,
     sum = sum + 0.0625f * s_dstPatch[2 + tidy + 2][tidx];
 
     if ((x < dstCols) && (y < dstRows))
-        dst[x + y * dstStep] = (float)(4.0f * sum);
+        dst[x + y * dstStep] = convert_uchar_sat_rte(4.0f * sum);
 
 }
 
@@ -244,7 +245,7 @@ __kernel void pyrUp_C1_D2(__global ushort* src,__global ushort* dst,
     sum = sum + 0.0625f * s_dstPatch[2 + tidy + 2][get_local_id(0)];
 
     if ((x < dstCols) && (y < dstRows))
-        dst[x + y * dstStep] = (float)(4.0f * sum);
+        dst[x + y * dstStep] = convert_short_sat_rte(4.0f * sum);
 
 }
 
@@ -351,31 +352,6 @@ __kernel void pyrUp_C1_D5(__global float* src,__global float* dst,
 ///////////////////////////////////////////////////////////////////////
 //////////////////////////  CV_8UC4  //////////////////////////////////
 ///////////////////////////////////////////////////////////////////////
-float4 covert_uchar4_to_float4(uchar4 data)
-{
-    float4 f4Data = {0,0,0,0};
-
-    f4Data.x = (float)data.x;
-    f4Data.y = (float)data.y;
-    f4Data.z = (float)data.z;
-    f4Data.w = (float)data.w;
-
-    return f4Data;
-}
-
-
-uchar4 convert_float4_to_uchar4(float4 data)
-{
-    uchar4 u4Data;
-
-    u4Data.x = get_valid_uchar(data.x);
-    u4Data.y = get_valid_uchar(data.y);
-    u4Data.z = get_valid_uchar(data.z);
-    u4Data.w = get_valid_uchar(data.w);
-
-    return u4Data;
-}
-
 __kernel void pyrUp_C4_D0(__global uchar4* src,__global uchar4* dst,
                           int srcRows,int dstRows,int srcCols,int dstCols,
                           int srcOffset,int dstOffset,int srcStep,int dstStep)
@@ -406,7 +382,7 @@ __kernel void pyrUp_C4_D0(__global uchar4* src,__global uchar4* dst,
         srcy = abs(srcy);
         srcy = min(srcRows -1 ,srcy);
 
-        s_srcPatch[tidy][tidx] = covert_uchar4_to_float4(src[srcx + srcy * srcStep]);
+        s_srcPatch[tidy][tidx] = convert_float4(src[srcx + srcy * srcStep]);
     }
 
     barrier(CLK_LOCAL_MEM_FENCE);
@@ -476,38 +452,12 @@ __kernel void pyrUp_C4_D0(__global uchar4* src,__global uchar4* dst,
 
     if ((x < dstCols) && (y < dstRows))
     {
-        dst[x + y * dstStep] = convert_float4_to_uchar4(4.0f * sum);
+        dst[x + y * dstStep] = convert_uchar4_sat_rte(4.0f * sum);
     }
 }
 ///////////////////////////////////////////////////////////////////////
 //////////////////////////  CV_16UC4 //////////////////////////////////
 ///////////////////////////////////////////////////////////////////////
-float4 covert_ushort4_to_float4(ushort4 data)
-{
-    float4 f4Data = {0,0,0,0};
-
-    f4Data.x = (float)data.x;
-    f4Data.y = (float)data.y;
-    f4Data.z = (float)data.z;
-    f4Data.w = (float)data.w;
-
-    return f4Data;
-}
-
-
-ushort4 convert_float4_to_ushort4(float4 data)
-{
-    ushort4 u4Data;
-
-    u4Data.x = (float)data.x;
-    u4Data.y = (float)data.y;
-    u4Data.z = (float)data.z;
-    u4Data.w = (float)data.w;
-
-    return u4Data;
-}
-
-
 __kernel void pyrUp_C4_D2(__global ushort4* src,__global ushort4* dst,
                           int srcRows,int dstRows,int srcCols,int dstCols,
                           int srcOffset,int dstOffset,int srcStep,int dstStep)
@@ -535,7 +485,7 @@ __kernel void pyrUp_C4_D2(__global ushort4* src,__global ushort4* dst,
         srcy = abs(srcy);
         srcy = min(srcRows -1 ,srcy);
 
-        s_srcPatch[get_local_id(1)][get_local_id(0)] = covert_ushort4_to_float4(src[srcx + srcy * srcStep]);
+        s_srcPatch[get_local_id(1)][get_local_id(0)] = convert_float4(src[srcx + srcy * srcStep]);
     }
 
     barrier(CLK_LOCAL_MEM_FENCE);
@@ -570,11 +520,11 @@ __kernel void pyrUp_C4_D2(__global ushort4* src,__global ushort4* dst,
 
         if (eveny)
         {
-            sum = sum + (evenFlag * co3) * s_srcPatch[0][1 + ((tidx - 2) >> 1)];
-            sum = sum + ( oddFlag * co2  ) * s_srcPatch[0][1 + ((tidx - 1) >> 1)];
+            sum = sum + (evenFlag * co3 ) * s_srcPatch[0][1 + ((tidx - 2) >> 1)];
+            sum = sum + (oddFlag * co2  ) * s_srcPatch[0][1 + ((tidx - 1) >> 1)];
             sum = sum + (evenFlag * co1 ) * s_srcPatch[0][1 + ((tidx    ) >> 1)];
-            sum = sum + ( oddFlag * co2  ) * s_srcPatch[0][1 + ((tidx + 1) >> 1)];
-            sum = sum + (evenFlag * co3) * s_srcPatch[0][1 + ((tidx + 2) >> 1)];
+            sum = sum + (oddFlag * co2  ) * s_srcPatch[0][1 + ((tidx + 1) >> 1)];
+            sum = sum + (evenFlag * co3 ) * s_srcPatch[0][1 + ((tidx + 2) >> 1)];
         }
 
         s_dstPatch[get_local_id(1)][get_local_id(0)] = sum;
@@ -610,7 +560,7 @@ __kernel void pyrUp_C4_D2(__global ushort4* src,__global ushort4* dst,
 
     if ((x < dstCols) && (y < dstRows))
     {
-        dst[x + y * dstStep] = convert_float4_to_ushort4(4.0f * sum);
+        dst[x + y * dstStep] = convert_ushort4_sat_rte(4.0f * sum);
     }
 }
 
@@ -681,11 +631,11 @@ __kernel void pyrUp_C4_D5(__global float4* src,__global float4* dst,
 
         if (eveny)
         {
-            sum = sum + (evenFlag * co3) * s_srcPatch[lsizey-16][1 + ((tidx - 2) >> 1)];
-            sum = sum + ( oddFlag * co2  ) * s_srcPatch[lsizey-16][1 + ((tidx - 1) >> 1)];
+            sum = sum + (evenFlag * co3 ) * s_srcPatch[lsizey-16][1 + ((tidx - 2) >> 1)];
+            sum = sum + (oddFlag * co2  ) * s_srcPatch[lsizey-16][1 + ((tidx - 1) >> 1)];
             sum = sum + (evenFlag * co1 ) * s_srcPatch[lsizey-16][1 + ((tidx    ) >> 1)];
-            sum = sum + ( oddFlag * co2  ) * s_srcPatch[lsizey-16][1 + ((tidx + 1) >> 1)];
-            sum = sum + (evenFlag * co3) * s_srcPatch[lsizey-16][1 + ((tidx + 2) >> 1)];
+            sum = sum + ( oddFlag * co2 ) * s_srcPatch[lsizey-16][1 + ((tidx + 1) >> 1)];
+            sum = sum + (evenFlag * co3 ) * s_srcPatch[lsizey-16][1 + ((tidx + 2) >> 1)];
         }
 
         s_dstPatch[tidy][tidx] = sum;

modules/ocl/src/opencl/stereobm.cl

@@ -16,6 +16,8 @@
 //
 // @Authors
 //    Jia Haipeng, jiahaipeng95@gmail.com
+//    Sen Liu, swjtuls1987@126.com
+//    Peng Xiao, pengxiao@outlook.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -50,59 +52,40 @@
 #define STEREO_MIND 0                    // The minimum d range to check
 #define STEREO_DISP_STEP N_DISPARITIES   // the d step, must be <= 1 to avoid aliasing
 
-int SQ(int a)
-{
-    return a * a;
-}
+#ifndef radius
+#define radius 64
+#endif
 
-unsigned int CalcSSD(volatile __local unsigned int *col_ssd_cache,
-                     volatile __local unsigned int *col_ssd, int radius)
+unsigned int CalcSSD(__local unsigned int *col_ssd)
 {
-    unsigned int cache = 0;
-    unsigned int cache2 = 0;
+    unsigned int cache = col_ssd[0];
 
-    for(int i = 1; i <= radius; i++)
+#pragma unroll
+    for(int i = 1; i <= (radius << 1); i++)
         cache += col_ssd[i];
 
-    col_ssd_cache[0] = cache;
-
-    barrier(CLK_LOCAL_MEM_FENCE);
-
-    if (get_local_id(0) < BLOCK_W - radius)
-        cache2 = col_ssd_cache[radius];
-    else
-        for(int i = radius + 1; i < (2 * radius + 1); i++)
-            cache2 += col_ssd[i];
-
-    return col_ssd[0] + cache + cache2;
+    return cache;
 }
 
-uint2 MinSSD(volatile __local unsigned int *col_ssd_cache,
-             volatile __local unsigned int *col_ssd, int radius)
+uint2 MinSSD(__local unsigned int *col_ssd)
 {
     unsigned int ssd[N_DISPARITIES];
-
-    //See above:  #define COL_SSD_SIZE (BLOCK_W + 2 * radius)
-    ssd[0] = CalcSSD(col_ssd_cache, col_ssd + 0 * (BLOCK_W + 2 * radius), radius);
-    barrier(CLK_LOCAL_MEM_FENCE);
-    ssd[1] = CalcSSD(col_ssd_cache, col_ssd + 1 * (BLOCK_W + 2 * radius), radius);
-    barrier(CLK_LOCAL_MEM_FENCE);
-    ssd[2] = CalcSSD(col_ssd_cache, col_ssd + 2 * (BLOCK_W + 2 * radius), radius);
-    barrier(CLK_LOCAL_MEM_FENCE);
-    ssd[3] = CalcSSD(col_ssd_cache, col_ssd + 3 * (BLOCK_W + 2 * radius), radius);
-    barrier(CLK_LOCAL_MEM_FENCE);
-    ssd[4] = CalcSSD(col_ssd_cache, col_ssd + 4 * (BLOCK_W + 2 * radius), radius);
-    barrier(CLK_LOCAL_MEM_FENCE);
-    ssd[5] = CalcSSD(col_ssd_cache, col_ssd + 5 * (BLOCK_W + 2 * radius), radius);
-    barrier(CLK_LOCAL_MEM_FENCE);
-    ssd[6] = CalcSSD(col_ssd_cache, col_ssd + 6 * (BLOCK_W + 2 * radius), radius);
-    barrier(CLK_LOCAL_MEM_FENCE);
-    ssd[7] = CalcSSD(col_ssd_cache, col_ssd + 7 * (BLOCK_W + 2 * radius), radius);
-    barrier(CLK_LOCAL_MEM_FENCE);
+    const int win_size = (radius << 1);
+
+    //See above:  #define COL_SSD_SIZE (BLOCK_W + WIN_SIZE)
+    ssd[0] = CalcSSD(col_ssd + 0 * (BLOCK_W + win_size));
+    ssd[1] = CalcSSD(col_ssd + 1 * (BLOCK_W + win_size));
+    ssd[2] = CalcSSD(col_ssd + 2 * (BLOCK_W + win_size));
+    ssd[3] = CalcSSD(col_ssd + 3 * (BLOCK_W + win_size));
+    ssd[4] = CalcSSD(col_ssd + 4 * (BLOCK_W + win_size));
+    ssd[5] = CalcSSD(col_ssd + 5 * (BLOCK_W + win_size));
+    ssd[6] = CalcSSD(col_ssd + 6 * (BLOCK_W + win_size));
+    ssd[7] = CalcSSD(col_ssd + 7 * (BLOCK_W + win_size));
 
     unsigned int mssd = min(min(min(ssd[0], ssd[1]), min(ssd[4], ssd[5])), min(min(ssd[2], ssd[3]), min(ssd[6], ssd[7])));
 
     int bestIdx = 0;
+
     for (int i = 0; i < N_DISPARITIES; i++)
     {
         if (mssd == ssd[i])
@@ -113,124 +96,66 @@ uint2 MinSSD(volatile __local unsigned int *col_ssd_cache,
 }
 
 void StepDown(int idx1, int idx2, __global unsigned char* imageL,
-              __global unsigned char* imageR, int d, volatile  __local unsigned int *col_ssd, int radius)
+              __global unsigned char* imageR, int d,   __local unsigned int *col_ssd)
 {
-    unsigned char leftPixel1;
-    unsigned char leftPixel2;
-    unsigned char rightPixel1[8];
-    unsigned char rightPixel2[8];
-    unsigned int diff1, diff2;
-
-    leftPixel1 = imageL[idx1];
-    leftPixel2 = imageL[idx2];
-
-    idx1 = idx1 - d;
-    idx2 = idx2 - d;
-
-    rightPixel1[7] = imageR[idx1 - 7];
-    rightPixel1[0] = imageR[idx1 - 0];
-    rightPixel1[1] = imageR[idx1 - 1];
-    rightPixel1[2] = imageR[idx1 - 2];
-    rightPixel1[3] = imageR[idx1 - 3];
-    rightPixel1[4] = imageR[idx1 - 4];
-    rightPixel1[5] = imageR[idx1 - 5];
-    rightPixel1[6] = imageR[idx1 - 6];
-
-    rightPixel2[7] = imageR[idx2 - 7];
-    rightPixel2[0] = imageR[idx2 - 0];
-    rightPixel2[1] = imageR[idx2 - 1];
-    rightPixel2[2] = imageR[idx2 - 2];
-    rightPixel2[3] = imageR[idx2 - 3];
-    rightPixel2[4] = imageR[idx2 - 4];
-    rightPixel2[5] = imageR[idx2 - 5];
-    rightPixel2[6] = imageR[idx2 - 6];
-
-    //See above:  #define COL_SSD_SIZE (BLOCK_W + 2 * radius)
-    diff1 = leftPixel1 - rightPixel1[0];
-    diff2 = leftPixel2 - rightPixel2[0];
-    col_ssd[0 * (BLOCK_W + 2 * radius)] += SQ(diff2) - SQ(diff1);
-
-    diff1 = leftPixel1 - rightPixel1[1];
-    diff2 = leftPixel2 - rightPixel2[1];
-    col_ssd[1 * (BLOCK_W + 2 * radius)] += SQ(diff2) - SQ(diff1);
-
-    diff1 = leftPixel1 - rightPixel1[2];
-    diff2 = leftPixel2 - rightPixel2[2];
-    col_ssd[2 * (BLOCK_W + 2 * radius)] += SQ(diff2) - SQ(diff1);
-
-    diff1 = leftPixel1 - rightPixel1[3];
-    diff2 = leftPixel2 - rightPixel2[3];
-    col_ssd[3 * (BLOCK_W + 2 * radius)] += SQ(diff2) - SQ(diff1);
-
-    diff1 = leftPixel1 - rightPixel1[4];
-    diff2 = leftPixel2 - rightPixel2[4];
-    col_ssd[4 * (BLOCK_W + 2 * radius)] += SQ(diff2) - SQ(diff1);
-
-    diff1 = leftPixel1 - rightPixel1[5];
-    diff2 = leftPixel2 - rightPixel2[5];
-    col_ssd[5 * (BLOCK_W + 2 * radius)] += SQ(diff2) - SQ(diff1);
-
-    diff1 = leftPixel1 - rightPixel1[6];
-    diff2 = leftPixel2 - rightPixel2[6];
-    col_ssd[6 * (BLOCK_W + 2 * radius)] += SQ(diff2) - SQ(diff1);
-
-    diff1 = leftPixel1 - rightPixel1[7];
-    diff2 = leftPixel2 - rightPixel2[7];
-    col_ssd[7 * (BLOCK_W + 2 * radius)] += SQ(diff2) - SQ(diff1);
+    uint8 imgR1 = convert_uint8(vload8(0, imageR + (idx1 - d - 7)));
+    uint8 imgR2 = convert_uint8(vload8(0, imageR + (idx2 - d - 7)));
+    uint8 diff1 = (uint8)(imageL[idx1]) - imgR1;
+    uint8 diff2 = (uint8)(imageL[idx2]) - imgR2;
+    uint8 res = diff2 * diff2 - diff1 * diff1;
+    const int win_size = (radius << 1);
+    col_ssd[0 * (BLOCK_W + win_size)] += res.s7;
+    col_ssd[1 * (BLOCK_W + win_size)] += res.s6;
+    col_ssd[2 * (BLOCK_W + win_size)] += res.s5;
+    col_ssd[3 * (BLOCK_W + win_size)] += res.s4;
+    col_ssd[4 * (BLOCK_W + win_size)] += res.s3;
+    col_ssd[5 * (BLOCK_W + win_size)] += res.s2;
+    col_ssd[6 * (BLOCK_W + win_size)] += res.s1;
+    col_ssd[7 * (BLOCK_W + win_size)] += res.s0;
 }
 
 void InitColSSD(int x_tex, int y_tex, int im_pitch, __global unsigned char* imageL,
                 __global unsigned char* imageR, int d,
-                volatile __local unsigned int *col_ssd, int radius)
+                 __local unsigned int *col_ssd)
 {
-    unsigned char leftPixel1;
-    int idx;
-    unsigned int diffa[] = {0, 0, 0, 0, 0, 0, 0, 0};
-
-    for(int i = 0; i < (2 * radius + 1); i++)
+    uint8 leftPixel1;
+    uint8 diffa = 0;
+    int idx = y_tex * im_pitch + x_tex;
+    const int win_size = (radius << 1);
+    for(int i = 0; i < (win_size + 1); i++)
     {
-        idx = y_tex * im_pitch + x_tex;
-        leftPixel1 = imageL[idx];
-        idx = idx - d;
-
-        diffa[0] += SQ(leftPixel1 - imageR[idx - 0]);
-        diffa[1] += SQ(leftPixel1 - imageR[idx - 1]);
-        diffa[2] += SQ(leftPixel1 - imageR[idx - 2]);
-        diffa[3] += SQ(leftPixel1 - imageR[idx - 3]);
-        diffa[4] += SQ(leftPixel1 - imageR[idx - 4]);
-        diffa[5] += SQ(leftPixel1 - imageR[idx - 5]);
-        diffa[6] += SQ(leftPixel1 - imageR[idx - 6]);
-        diffa[7] += SQ(leftPixel1 - imageR[idx - 7]);
-
-        y_tex += 1;
+        leftPixel1 = (uint8)(imageL[idx]);
+        uint8 imgR = convert_uint8(vload8(0, imageR + (idx - d - 7)));
+        uint8 res = leftPixel1 - imgR;
+        diffa += res * res;
+
+        idx += im_pitch;
     }
-    //See above:  #define COL_SSD_SIZE (BLOCK_W + 2 * radius)
-    col_ssd[0 * (BLOCK_W + 2 * radius)] = diffa[0];
-    col_ssd[1 * (BLOCK_W + 2 * radius)] = diffa[1];
-    col_ssd[2 * (BLOCK_W + 2 * radius)] = diffa[2];
-    col_ssd[3 * (BLOCK_W + 2 * radius)] = diffa[3];
-    col_ssd[4 * (BLOCK_W + 2 * radius)] = diffa[4];
-    col_ssd[5 * (BLOCK_W + 2 * radius)] = diffa[5];
-    col_ssd[6 * (BLOCK_W + 2 * radius)] = diffa[6];
-    col_ssd[7 * (BLOCK_W + 2 * radius)] = diffa[7];
+    //See above:  #define COL_SSD_SIZE (BLOCK_W + WIN_SIZE)
+    col_ssd[0 * (BLOCK_W + win_size)] = diffa.s7;
+    col_ssd[1 * (BLOCK_W + win_size)] = diffa.s6;
+    col_ssd[2 * (BLOCK_W + win_size)] = diffa.s5;
+    col_ssd[3 * (BLOCK_W + win_size)] = diffa.s4;
+    col_ssd[4 * (BLOCK_W + win_size)] = diffa.s3;
+    col_ssd[5 * (BLOCK_W + win_size)] = diffa.s2;
+    col_ssd[6 * (BLOCK_W + win_size)] = diffa.s1;
+    col_ssd[7 * (BLOCK_W + win_size)] = diffa.s0;
 }
 
 __kernel void stereoKernel(__global unsigned char *left, __global unsigned char *right,
                            __global unsigned int *cminSSDImage, int cminSSD_step,
                            __global unsigned char *disp, int disp_step,int cwidth, int cheight,
-                           int img_step, int maxdisp, int radius,
+                           int img_step, int maxdisp,
                            __local unsigned int *col_ssd_cache)
 {
-
-    volatile __local unsigned int *col_ssd = col_ssd_cache + BLOCK_W + get_local_id(0);
-    volatile __local unsigned int *col_ssd_extra = get_local_id(0) < (2 * radius) ? col_ssd + BLOCK_W : 0;
+    __local unsigned int *col_ssd = col_ssd_cache + get_local_id(0);
+    __local unsigned int *col_ssd_extra = get_local_id(0) < (radius << 1) ? col_ssd + BLOCK_W : 0;
 
     int X = get_group_id(0) * BLOCK_W + get_local_id(0) + maxdisp + radius;
-    // int Y = get_group_id(1) * ROWSperTHREAD + radius;
 
 #define Y (get_group_id(1) * ROWSperTHREAD + radius)
 
-    volatile __global unsigned int* minSSDImage = cminSSDImage + X + Y * cminSSD_step;
+    __global unsigned int* minSSDImage = cminSSDImage + X + Y * cminSSD_step;
     __global unsigned char* disparImage = disp + X + Y * disp_step;
 
     int end_row = ROWSperTHREAD < (cheight - Y) ? ROWSperTHREAD:(cheight - Y);
@@ -244,14 +169,14 @@ __kernel void stereoKernel(__global unsigned char *left, __global unsigned char
     {
         y_tex = Y - radius;
 
-        InitColSSD(x_tex, y_tex, img_step, left, right, d, col_ssd, radius);
+        InitColSSD(x_tex, y_tex, img_step, left, right, d, col_ssd);
         if (col_ssd_extra > 0)
             if (x_tex + BLOCK_W < cwidth)
-                InitColSSD(x_tex + BLOCK_W, y_tex, img_step, left, right, d, col_ssd_extra, radius);
+                InitColSSD(x_tex + BLOCK_W, y_tex, img_step, left, right, d, col_ssd_extra);
 
         barrier(CLK_LOCAL_MEM_FENCE); //before MinSSD function
 
-        uint2 minSSD = MinSSD(col_ssd_cache + get_local_id(0), col_ssd, radius);
+        uint2 minSSD = MinSSD(col_ssd);
         if (X < cwidth - radius && Y < cheight - radius)
         {
             if (minSSD.x < minSSDImage[0])
@@ -264,21 +189,18 @@ __kernel void stereoKernel(__global unsigned char *left, __global unsigned char
         for(int row = 1; row < end_row; row++)
         {
             int idx1 = y_tex * img_step + x_tex;
-            int idx2 = min(y_tex + (2 * radius + 1), cheight - 1) * img_step + x_tex;
-
-            barrier(CLK_GLOBAL_MEM_FENCE);
+            int idx2 = min(y_tex + ((radius << 1) + 1), cheight - 1) * img_step + x_tex;
+            
             barrier(CLK_LOCAL_MEM_FENCE);
 
-            StepDown(idx1, idx2, left, right, d, col_ssd, radius);
+            StepDown(idx1, idx2, left, right, d, col_ssd);
             if (col_ssd_extra > 0)
                 if (x_tex + BLOCK_W < cwidth)
-                    StepDown(idx1, idx2, left + BLOCK_W, right + BLOCK_W, d, col_ssd_extra, radius);
-
-            y_tex += 1;
+                    StepDown(idx1, idx2, left + BLOCK_W, right + BLOCK_W, d, col_ssd_extra);
 
             barrier(CLK_LOCAL_MEM_FENCE);
 
-            uint2 minSSD = MinSSD(col_ssd_cache + get_local_id(0), col_ssd, radius);
+            uint2 minSSD = MinSSD(col_ssd);
             if (X < cwidth - radius && row < cheight - radius - Y)
             {
                 int idx = row * cminSSD_step;
@@ -288,10 +210,11 @@ __kernel void stereoKernel(__global unsigned char *left, __global unsigned char
                     minSSDImage[idx] = minSSD.x;
                 }
             }
+
+            y_tex++;
         } // for row loop
     } // for d loop
 }
-
 //////////////////////////////////////////////////////////////////////////////////////////////////
 //////////////////////////// Sobel Prefiler (signal channel)//////////////////////////////////////
 //////////////////////////////////////////////////////////////////////////////////////////////////

modules/ocl/src/opencl/stereobp.cl

@@ -185,10 +185,10 @@ __kernel void data_step_down(__global T *src, int src_rows,
         for (int d = 0; d < cndisp; ++d)
         {
             float dst_reg;
-            dst_reg  = src[(d * src_rows + (2*y+0)) * src_step + 2*x+0];
-            dst_reg += src[(d * src_rows + (2*y+1)) * src_step + 2*x+0];
-            dst_reg += src[(d * src_rows + (2*y+0)) * src_step + 2*x+1];
-            dst_reg += src[(d * src_rows + (2*y+1)) * src_step + 2*x+1];
+            dst_reg  = src[(d * src_rows + min(2*y+0, src_rows-1)) * src_step + 2*x+0];
+            dst_reg += src[(d * src_rows + min(2*y+1, src_rows-1)) * src_step + 2*x+0];
+            dst_reg += src[(d * src_rows + min(2*y+0, src_rows-1)) * src_step + 2*x+1];
+            dst_reg += src[(d * src_rows + min(2*y+1, src_rows-1)) * src_step + 2*x+1];
 
             dst[(d * dst_rows + y) * dst_step + x] = saturate_cast(dst_reg);
         }

modules/ocl/src/opencl/tvl1flow.cl

@@ -0,0 +1,407 @@
+/*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) 2010-2012, Multicoreware, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// @Authors
+//    Jin Ma jin@multicorewareinc.com
+//
+// 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 oclMaterials 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*/
+
+__kernel void centeredGradientKernel(__global const float* src, int src_col, int src_row, int src_step, 
+__global float* dx, __global float* dy, int dx_step)
+{
+    int x = get_global_id(0);
+    int y = get_global_id(1);
+
+    if((x < src_col)&&(y < src_row))
+    {
+		int src_x1 = (x + 1) < (src_col -1)? (x + 1) : (src_col - 1);
+     	int src_x2 = (x - 1) > 0 ? (x -1) : 0;
+        
+        //if(src[y * src_step + src_x1] == src[y * src_step+ src_x2])
+        //{
+        //    printf("y = %d\n", y);
+        //    printf("src_x1 = %d\n", src_x1);
+        //    printf("src_x2 = %d\n", src_x2);
+        //}
+        dx[y * dx_step+ x] = 0.5f * (src[y * src_step + src_x1] - src[y * src_step+ src_x2]);
+        
+		int src_y1 = (y+1) < (src_row - 1) ? (y + 1) : (src_row - 1);
+        int src_y2 = (y - 1) > 0 ? (y - 1) : 0;
+        dy[y * dx_step+ x] = 0.5f * (src[src_y1 * src_step + x] - src[src_y2 * src_step+ x]);
+    }
+
+}
+
+float bicubicCoeff(float x_)
+{
+
+    float x = fabs(x_);
+    if (x <= 1.0f)
+    {
+        return x * x * (1.5f * x - 2.5f) + 1.0f;
+    }
+    else if (x < 2.0f)
+    {
+        return x * (x * (-0.5f * x + 2.5f) - 4.0f) + 2.0f;
+    }
+    else
+    {
+        return 0.0f;
+    }
+
+}
+
+__kernel void warpBackwardKernel(__global const float* I0, int I0_step, int I0_col, int I0_row,
+	image2d_t tex_I1, image2d_t tex_I1x, image2d_t tex_I1y,  
+    __global const float* u1, int u1_step, 
+    __global const float* u2,
+    __global float* I1w,
+	__global float* I1wx, /*int I1wx_step,*/
+	__global float* I1wy, /*int I1wy_step,*/
+	__global float* grad, /*int grad_step,*/
+	__global float* rho,
+	int I1w_step,
+	int u2_step,
+	int u1_offset_x,
+	int u1_offset_y,
+	int u2_offset_x,
+	int u2_offset_y)
+{
+    const int x = get_global_id(0);
+    const int y = get_global_id(1);
+
+    if(x < I0_col&&y < I0_row)
+    {
+        //const float u1Val = u1(y, x);
+        const float u1Val = u1[(y + u1_offset_y) * u1_step + x + u1_offset_x];
+        //const float u2Val = u2(y, x);
+        const float u2Val = u2[(y + u2_offset_y) * u2_step + x + u2_offset_x];
+
+        const float wx = x + u1Val;
+        const float wy = y + u2Val;
+
+        const int xmin = ceil(wx - 2.0f);
+        const int xmax = floor(wx + 2.0f);
+
+        const int ymin = ceil(wy - 2.0f);
+        const int ymax = floor(wy + 2.0f);
+
+        float sum  = 0.0f;
+        float sumx = 0.0f;
+        float sumy = 0.0f;
+        float wsum = 0.0f;
+        sampler_t sampleri = CLK_NORMALIZED_COORDS_FALSE | CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST;
+
+        for (int cy = ymin; cy <= ymax; ++cy)
+        {
+            for (int cx = xmin; cx <= xmax; ++cx)
+            {
+                const float w = bicubicCoeff(wx - cx) * bicubicCoeff(wy - cy);
+
+                //sum  += w * tex2D(tex_I1 , cx, cy);
+				int2 cood = (int2)(cx, cy);
+                sum += w * read_imagef(tex_I1, sampleri, cood).x;
+                //sumx += w * tex2D(tex_I1x, cx, cy);
+                sumx += w * read_imagef(tex_I1x, sampleri, cood).x;
+                //sumy += w * tex2D(tex_I1y, cx, cy);
+                sumy += w * read_imagef(tex_I1y, sampleri, cood).x;
+
+                wsum += w;
+            }
+        }
+
+        const float coeff = 1.0f / wsum;
+
+        const float I1wVal  = sum  * coeff;
+        const float I1wxVal = sumx * coeff;
+        const float I1wyVal = sumy * coeff;
+
+        I1w[y * I1w_step + x]  = I1wVal;
+        I1wx[y * I1w_step + x] = I1wxVal;
+        I1wy[y * I1w_step + x] = I1wyVal;
+
+        const float Ix2 = I1wxVal * I1wxVal;
+        const float Iy2 = I1wyVal * I1wyVal;
+
+        // store the |Grad(I1)|^2
+        grad[y * I1w_step + x] = Ix2 + Iy2;
+
+        // compute the constant part of the rho function
+        const float I0Val = I0[y * I0_step + x];
+        rho[y * I1w_step + x] = I1wVal - I1wxVal * u1Val - I1wyVal * u2Val - I0Val;
+    }
+
+}
+
+float readImage(__global const float *image,  const int x,  const int y,  const int rows,  const int cols, const int elemCntPerRow)
+{
+    int i0 = clamp(x, 0, cols - 1);
+    int j0 = clamp(y, 0, rows - 1);
+    int i1 = clamp(x + 1, 0, cols - 1);
+    int j1 = clamp(y + 1, 0, rows - 1);
+
+    return image[j0 * elemCntPerRow + i0];
+}
+
+__kernel void warpBackwardKernelNoImage2d(__global const float* I0, int I0_step, int I0_col, int I0_row,
+	__global const float* tex_I1, __global const float* tex_I1x, __global const float* tex_I1y,  
+    __global const float* u1, int u1_step, 
+    __global const float* u2,
+    __global float* I1w,
+	__global float* I1wx, /*int I1wx_step,*/
+	__global float* I1wy, /*int I1wy_step,*/
+	__global float* grad, /*int grad_step,*/
+	__global float* rho,
+	int I1w_step,
+	int u2_step,
+	int I1_step,
+	int I1x_step)
+{
+    const int x = get_global_id(0);
+    const int y = get_global_id(1);
+
+    if(x < I0_col&&y < I0_row)
+    {
+        //const float u1Val = u1(y, x);
+        const float u1Val = u1[y * u1_step + x];
+        //const float u2Val = u2(y, x);
+        const float u2Val = u2[y * u2_step + x];
+
+        const float wx = x + u1Val;
+        const float wy = y + u2Val;
+
+        const int xmin = ceil(wx - 2.0f);
+        const int xmax = floor(wx + 2.0f);
+
+        const int ymin = ceil(wy - 2.0f);
+        const int ymax = floor(wy + 2.0f);
+
+        float sum  = 0.0f;
+        float sumx = 0.0f;
+        float sumy = 0.0f;
+        float wsum = 0.0f;
+
+        for (int cy = ymin; cy <= ymax; ++cy)
+        {
+            for (int cx = xmin; cx <= xmax; ++cx)
+            {
+                const float w = bicubicCoeff(wx - cx) * bicubicCoeff(wy - cy);
+
+				int2 cood = (int2)(cx, cy);
+                sum += w * readImage(tex_I1, cood.x, cood.y, I0_col, I0_row, I1_step);
+                sumx += w * readImage(tex_I1x, cood.x, cood.y, I0_col, I0_row, I1x_step);
+                sumy += w * readImage(tex_I1y, cood.x, cood.y, I0_col, I0_row, I1x_step);
+                wsum += w;
+            }
+        }
+
+        const float coeff = 1.0f / wsum;
+
+        const float I1wVal  = sum  * coeff;
+        const float I1wxVal = sumx * coeff;
+        const float I1wyVal = sumy * coeff;
+
+        I1w[y * I1w_step + x]  = I1wVal;
+        I1wx[y * I1w_step + x] = I1wxVal;
+        I1wy[y * I1w_step + x] = I1wyVal;
+
+        const float Ix2 = I1wxVal * I1wxVal;
+        const float Iy2 = I1wyVal * I1wyVal;
+
+        // store the |Grad(I1)|^2
+        grad[y * I1w_step + x] = Ix2 + Iy2;
+
+        // compute the constant part of the rho function
+        const float I0Val = I0[y * I0_step + x];
+        rho[y * I1w_step + x] = I1wVal - I1wxVal * u1Val - I1wyVal * u2Val - I0Val;
+    }
+
+}
+
+
+__kernel void estimateDualVariablesKernel(__global const float* u1, int u1_col, int u1_row, int u1_step, 
+    __global const float* u2, 
+    __global float* p11, int p11_step, 
+    __global float* p12,
+    __global float* p21,
+    __global float* p22, 
+    const float taut,
+	int u2_step,
+	int u1_offset_x,
+	int u1_offset_y,
+	int u2_offset_x,
+	int u2_offset_y)
+{
+
+    //const int x = blockIdx.x * blockDim.x + threadIdx.x;
+    //const int y = blockIdx.y * blockDim.y + threadIdx.y;
+    const int x = get_global_id(0);
+    const int y = get_global_id(1);
+
+    if(x < u1_col && y < u1_row)
+    {
+		int src_x1 = (x + 1) < (u1_col - 1) ? (x + 1) : (u1_col - 1);
+        const float u1x = u1[(y + u1_offset_y) * u1_step + src_x1 + u1_offset_x] - u1[(y + u1_offset_y) * u1_step + x + u1_offset_x];
+        
+		int src_y1 = (y + 1) < (u1_row - 1) ? (y + 1) : (u1_row - 1);
+        const float u1y = u1[(src_y1 + u1_offset_y) * u1_step + x + u1_offset_x] - u1[(y + u1_offset_y) * u1_step + x + u1_offset_x];
+
+		int src_x2 = (x + 1) < (u1_col - 1) ? (x + 1) : (u1_col - 1);
+        const float u2x = u2[(y + u2_offset_y) * u2_step + src_x2 + u2_offset_x] - u2[(y + u2_offset_y) * u2_step + x + u2_offset_x];
+
+		int src_y2 = (y + 1) <  (u1_row - 1) ? (y + 1) : (u1_row - 1);
+        const float u2y = u2[(src_y2 + u2_offset_y) * u2_step + x + u2_offset_x] - u2[(y + u2_offset_y) * u2_step + x + u2_offset_x];
+
+        const float g1 = hypot(u1x, u1y);
+        const float g2 = hypot(u2x, u2y);
+
+        const float ng1 = 1.0f + taut * g1;
+        const float ng2 = 1.0f + taut * g2;
+
+        p11[y * p11_step + x] = (p11[y * p11_step + x] + taut * u1x) / ng1;
+        p12[y * p11_step + x] = (p12[y * p11_step + x] + taut * u1y) / ng1;
+        p21[y * p11_step + x] = (p21[y * p11_step + x] + taut * u2x) / ng2;
+        p22[y * p11_step + x] = (p22[y * p11_step + x] + taut * u2y) / ng2;
+    }
+
+}
+
+float divergence(__global const float* v1, __global const float* v2, int y, int x, int v1_step, int v2_step)
+{
+
+    if (x > 0 && y > 0)
+    {
+        const float v1x = v1[y * v1_step + x] - v1[y * v1_step + x - 1];
+        const float v2y = v2[y * v2_step + x] - v2[(y - 1) * v2_step + x];
+        return v1x + v2y;
+    }
+    else
+    {
+        if (y > 0)
+            return v1[y * v1_step + 0] + v2[y * v2_step + 0] - v2[(y - 1) * v2_step + 0];
+        else
+        {
+            if (x > 0)
+                return v1[0 * v1_step + x] - v1[0 * v1_step + x - 1] + v2[0 * v2_step + x];
+            else
+                return v1[0 * v1_step + 0] + v2[0 * v2_step + 0];
+        }
+    }
+
+}
+
+__kernel void estimateUKernel(__global const float* I1wx, int I1wx_col, int I1wx_row, int I1wx_step,
+    __global const float* I1wy, /*int I1wy_step,*/
+    __global const float* grad, /*int grad_step,*/ 
+    __global const float* rho_c, /*int rho_c_step,*/
+    __global const float* p11, /*int p11_step,*/
+    __global const float* p12, /*int p12_step,*/
+    __global const float* p21, /*int p21_step,*/
+    __global const float* p22, /*int p22_step,*/
+    __global float* u1, int u1_step, 
+    __global float* u2, 
+    __global float* error, const float l_t, const float theta, int u2_step,
+	int u1_offset_x,
+	int u1_offset_y,
+	int u2_offset_x,
+	int u2_offset_y)
+{
+
+    //const int x = blockIdx.x * blockDim.x + threadIdx.x;
+    //const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+    int x = get_global_id(0);
+    int y = get_global_id(1);
+
+
+    if(x < I1wx_col && y < I1wx_row)
+    {
+        const float I1wxVal = I1wx[y * I1wx_step + x];
+        const float I1wyVal = I1wy[y * I1wx_step + x];
+        const float gradVal = grad[y * I1wx_step + x];
+        const float u1OldVal = u1[(y + u1_offset_y) * u1_step + x + u1_offset_x];
+        const float u2OldVal = u2[(y + u2_offset_y) * u2_step + x + u2_offset_x];
+
+        const float rho = rho_c[y * I1wx_step + x] + (I1wxVal * u1OldVal + I1wyVal * u2OldVal);
+
+        // estimate the values of the variable (v1, v2) (thresholding operator TH)
+
+        float d1 = 0.0f;
+        float d2 = 0.0f;
+
+        if (rho < -l_t * gradVal)
+        {
+            d1 = l_t * I1wxVal;
+            d2 = l_t * I1wyVal;
+        }
+        else if (rho > l_t * gradVal)
+        {
+            d1 = -l_t * I1wxVal;
+            d2 = -l_t * I1wyVal;
+        }
+        else if (gradVal > 1.192092896e-07f)
+        {
+            const float fi = -rho / gradVal;
+            d1 = fi * I1wxVal;
+            d2 = fi * I1wyVal;
+        }
+
+        const float v1 = u1OldVal + d1;
+        const float v2 = u2OldVal + d2;
+
+        // compute the divergence of the dual variable (p1, p2)
+
+        const float div_p1 = divergence(p11, p12, y, x, I1wx_step, I1wx_step);
+        const float div_p2 = divergence(p21, p22, y, x, I1wx_step, I1wx_step);
+
+        // estimate the values of the optical flow (u1, u2)
+
+        const float u1NewVal = v1 + theta * div_p1;
+        const float u2NewVal = v2 + theta * div_p2;
+
+        u1[(y + u1_offset_y) * u1_step + x + u1_offset_x] = u1NewVal;
+        u2[(y + u2_offset_y) * u2_step + x + u2_offset_x] = u2NewVal;
+
+        const float n1 = (u1OldVal - u1NewVal) * (u1OldVal - u1NewVal);
+        const float n2 = (u2OldVal - u2NewVal) * (u2OldVal - u2NewVal);
+        error[y * I1wx_step + x] = n1 + n2;
+    }
+
+}

modules/ocl/src/stereo_csbp.cpp

@@ -0,0 +1,756 @@
+/*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) 2010-2012, Multicoreware, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// @Authors
+//    Jia Haipeng, jiahaipeng95@gmail.com
+//    Jin Ma, jin@multicorewareinc.com
+// 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 oclMaterials 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"
+
+using namespace cv;
+using namespace cv::ocl;
+using namespace std;
+
+#if !defined (HAVE_OPENCL)
+
+namespace cv
+{
+    namespace ocl
+    {
+
+        void cv::ocl::StereoConstantSpaceBP::estimateRecommendedParams(int, int, int &, int &, int &, int &)
+        {
+            throw_nogpu();
+        }
+        cv::ocl::StereoConstantSpaceBP::StereoConstantSpaceBP(int, int, int, int, int)
+        {
+            throw_nogpu();
+        }
+        cv::ocl::StereoConstantSpaceBP::StereoConstantSpaceBP(int, int, int, int, float, float,
+            float, float, int, int)
+        {
+            throw_nogpu();
+        }
+
+        void cv::ocl::StereoConstantSpaceBP::operator()(const oclMat &, const oclMat &, oclMat &)
+        {
+            throw_nogpu();
+        }
+    }
+}
+
+#else /* !defined (HAVE_OPENCL) */
+
+namespace cv
+{
+    namespace ocl
+    {
+
+        ///////////////////////////OpenCL kernel Strings///////////////////////////
+        extern const char *stereocsbp;
+    }
+
+}
+namespace cv
+{
+    namespace ocl
+    {
+        namespace stereoCSBP
+        {
+            //////////////////////////////////////////////////////////////////////////
+            //////////////////////////////common////////////////////////////////////
+            ////////////////////////////////////////////////////////////////////////
+            static inline int divUp(int total, int grain)
+            {
+                return (total + grain - 1) / grain;
+            }
+            static String get_kernel_name(String kernel_name, int data_type)
+            {
+                return kernel_name + (data_type == CV_16S ? "0" : "1");
+            }
+            using cv::ocl::StereoConstantSpaceBP;
+            //////////////////////////////////////////////////////////////////////////////////
+            /////////////////////////////////init_data_cost//////////////////////////////////
+            //////////////////////////////////////////////////////////////////////////////////
+            static void init_data_cost_caller(const oclMat &left, const oclMat &right, oclMat &temp,
+                StereoConstantSpaceBP &rthis,
+                int msg_step, int h, int w, int level)
+            {
+                Context  *clCxt = left.clCxt;
+                int data_type = rthis.msg_type;
+                int channels = left.oclchannels();
+
+                String kernelName = get_kernel_name("init_data_cost_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+
+                //size_t blockSize = 256;
+                size_t localThreads[]  = {32, 8 ,1};
+                size_t globalThreads[] = {divUp(w, localThreads[0]) *localThreads[0],
+                    divUp(h, localThreads[1]) *localThreads[1],
+                    1
+                };
+
+                int cdisp_step1 = msg_step * h;
+                openCLVerifyKernel(clCxt, kernel,  localThreads);
+                openCLSafeCall(clSetKernelArg(kernel, 0, sizeof(cl_mem),  (void *)&temp.data));
+                openCLSafeCall(clSetKernelArg(kernel, 1, sizeof(cl_mem),  (void *)&left.data));
+                openCLSafeCall(clSetKernelArg(kernel, 2, sizeof(cl_mem),  (void *)&right.data));
+                openCLSafeCall(clSetKernelArg(kernel, 3, sizeof(cl_int),  (void *)&h));
+                openCLSafeCall(clSetKernelArg(kernel, 4, sizeof(cl_int),  (void *)&w));
+                openCLSafeCall(clSetKernelArg(kernel, 5, sizeof(cl_int),  (void *)&level));
+                openCLSafeCall(clSetKernelArg(kernel, 6, sizeof(cl_int),  (void *)&channels));
+                openCLSafeCall(clSetKernelArg(kernel, 7, sizeof(cl_int),  (void *)&msg_step));
+                openCLSafeCall(clSetKernelArg(kernel, 8, sizeof(cl_float), (void *)&rthis.data_weight));
+                openCLSafeCall(clSetKernelArg(kernel, 9, sizeof(cl_float), (void *)&rthis.max_data_term));
+                openCLSafeCall(clSetKernelArg(kernel, 10, sizeof(cl_int), (void *)&cdisp_step1));
+                openCLSafeCall(clSetKernelArg(kernel, 11, sizeof(cl_int), (void *)&rthis.min_disp_th));
+                openCLSafeCall(clSetKernelArg(kernel, 12, sizeof(cl_int), (void *)&left.step));
+                openCLSafeCall(clSetKernelArg(kernel, 13, sizeof(cl_int), (void *)&rthis.ndisp));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 2, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+
+            static void init_data_cost_reduce_caller(const oclMat &left, const oclMat &right, oclMat &temp,
+                StereoConstantSpaceBP &rthis,
+                int msg_step, int h, int w, int level)
+            {
+
+                Context  *clCxt = left.clCxt;
+                int data_type = rthis.msg_type;
+                int channels = left.oclchannels();
+                int win_size = (int)std::pow(2.f, level);
+
+                String kernelName = get_kernel_name("init_data_cost_reduce_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+
+                const int threadsNum = 256;
+                //size_t blockSize = threadsNum;
+                size_t localThreads[3]  = {win_size, 1, threadsNum / win_size};
+                size_t globalThreads[3] = {w *localThreads[0],
+                    h * divUp(rthis.ndisp, localThreads[2]) *localThreads[1], 1 * localThreads[2]
+                };
+
+                int local_mem_size = threadsNum * sizeof(float);
+                int cdisp_step1 = msg_step * h;
+
+                openCLVerifyKernel(clCxt, kernel, localThreads);
+
+                openCLSafeCall(clSetKernelArg(kernel, 0,  sizeof(cl_mem),  (void *)&temp.data));
+                openCLSafeCall(clSetKernelArg(kernel, 1,  sizeof(cl_mem),  (void *)&left.data));
+                openCLSafeCall(clSetKernelArg(kernel, 2,  sizeof(cl_mem),  (void *)&right.data));
+                openCLSafeCall(clSetKernelArg(kernel, 3,  local_mem_size,  (void *)NULL));
+                openCLSafeCall(clSetKernelArg(kernel, 4,  sizeof(cl_int),  (void *)&level));
+                openCLSafeCall(clSetKernelArg(kernel, 5,  sizeof(cl_int),  (void *)&left.rows));
+                openCLSafeCall(clSetKernelArg(kernel, 6,  sizeof(cl_int),  (void *)&left.cols));
+                openCLSafeCall(clSetKernelArg(kernel, 7,  sizeof(cl_int),  (void *)&h));
+                openCLSafeCall(clSetKernelArg(kernel, 8,  sizeof(cl_int),  (void *)&win_size));
+                openCLSafeCall(clSetKernelArg(kernel, 9,  sizeof(cl_int),  (void *)&channels));
+                openCLSafeCall(clSetKernelArg(kernel, 10, sizeof(cl_int),  (void *)&rthis.ndisp));
+                openCLSafeCall(clSetKernelArg(kernel, 11, sizeof(cl_int),  (void *)&left.step));
+                openCLSafeCall(clSetKernelArg(kernel, 12, sizeof(cl_float), (void *)&rthis.data_weight));
+                openCLSafeCall(clSetKernelArg(kernel, 13, sizeof(cl_float), (void *)&rthis.max_data_term));
+                openCLSafeCall(clSetKernelArg(kernel, 14, sizeof(cl_int),  (void *)&rthis.min_disp_th));
+                openCLSafeCall(clSetKernelArg(kernel, 15, sizeof(cl_int),  (void *)&cdisp_step1));
+                openCLSafeCall(clSetKernelArg(kernel, 16, sizeof(cl_int),  (void *)&msg_step));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 3, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+
+            static void get_first_initial_local_caller(uchar *data_cost_selected, uchar *disp_selected_pyr,
+                oclMat &temp, StereoConstantSpaceBP &rthis,
+                int h, int w, int nr_plane, int msg_step)
+            {
+                Context  *clCxt = temp.clCxt;
+                int data_type = rthis.msg_type;
+
+                String kernelName = get_kernel_name("get_first_k_initial_local_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+
+                //size_t blockSize = 256;
+                size_t localThreads[]  = {32, 8 ,1};
+                size_t globalThreads[] = {divUp(w, localThreads[0]) *localThreads[0],
+                    divUp(h, localThreads[1]) *localThreads[1],
+                    1
+                };
+
+                int disp_step = msg_step * h;
+                openCLVerifyKernel(clCxt, kernel, localThreads);
+                openCLSafeCall(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&data_cost_selected));
+                openCLSafeCall(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&disp_selected_pyr));
+                openCLSafeCall(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *)&temp.data));
+                openCLSafeCall(clSetKernelArg(kernel, 3, sizeof(cl_int), (void *)&h));
+                openCLSafeCall(clSetKernelArg(kernel, 4, sizeof(cl_int), (void *)&w));
+                openCLSafeCall(clSetKernelArg(kernel, 5, sizeof(cl_int), (void *)&nr_plane));
+                openCLSafeCall(clSetKernelArg(kernel, 6, sizeof(cl_int), (void *)&msg_step));
+                openCLSafeCall(clSetKernelArg(kernel, 7, sizeof(cl_int), (void *)&disp_step));
+                openCLSafeCall(clSetKernelArg(kernel, 8, sizeof(cl_int), (void *)&rthis.ndisp));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 2, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+            static void get_first_initial_global_caller(uchar *data_cost_selected, uchar *disp_selected_pyr,
+                oclMat &temp, StereoConstantSpaceBP &rthis,
+                int h, int w, int nr_plane, int msg_step)
+            {
+                Context  *clCxt = temp.clCxt;
+                int data_type = rthis.msg_type;
+
+                String kernelName = get_kernel_name("get_first_k_initial_global_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+
+                //size_t blockSize = 256;
+                size_t localThreads[]  = {32, 8, 1};
+                size_t globalThreads[] = {divUp(w, localThreads[0]) *localThreads[0],
+                    divUp(h, localThreads[1]) *localThreads[1],
+                    1
+                };
+
+                int disp_step = msg_step * h;
+                openCLVerifyKernel(clCxt, kernel, localThreads);
+                openCLSafeCall(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&data_cost_selected));
+                openCLSafeCall(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&disp_selected_pyr));
+                openCLSafeCall(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *)&temp.data));
+                openCLSafeCall(clSetKernelArg(kernel, 3, sizeof(cl_int), (void *)&h));
+                openCLSafeCall(clSetKernelArg(kernel, 4, sizeof(cl_int), (void *)&w));
+                openCLSafeCall(clSetKernelArg(kernel, 5, sizeof(cl_int), (void *)&nr_plane));
+                openCLSafeCall(clSetKernelArg(kernel, 6, sizeof(cl_int), (void *)&msg_step));
+                openCLSafeCall(clSetKernelArg(kernel, 7, sizeof(cl_int), (void *)&disp_step));
+                openCLSafeCall(clSetKernelArg(kernel, 8, sizeof(cl_int), (void *)&rthis.ndisp));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 2, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+
+            static void init_data_cost(const oclMat &left, const oclMat &right, oclMat &temp, StereoConstantSpaceBP &rthis,
+                uchar *disp_selected_pyr, uchar *data_cost_selected,
+                size_t msg_step, int h, int w, int level, int nr_plane)
+            {
+
+                if(level <= 1)
+                    init_data_cost_caller(left, right, temp, rthis, msg_step, h, w, level);
+                else
+                    init_data_cost_reduce_caller(left, right, temp, rthis, msg_step, h, w, level);
+
+                if(rthis.use_local_init_data_cost == true)
+                {
+                    get_first_initial_local_caller(data_cost_selected, disp_selected_pyr, temp, rthis, h, w, nr_plane, msg_step);
+                }
+                else
+                {
+                    get_first_initial_global_caller(data_cost_selected, disp_selected_pyr, temp, rthis, h, w,
+                        nr_plane, msg_step);
+                }
+            }
+
+            ///////////////////////////////////////////////////////////////////////////////////////////////////
+            ///////////////////////////////////compute_data_cost//////////////////////////////////////////////
+            ////////////////////////////////////////////////////////////////////////////////////////////////
+            static void compute_data_cost_caller(uchar *disp_selected_pyr, uchar *data_cost,
+                StereoConstantSpaceBP &rthis, int msg_step1,
+                int msg_step2, const oclMat &left, const oclMat &right, int h,
+                int w, int h2, int level, int nr_plane)
+            {
+                Context  *clCxt = left.clCxt;
+                int channels = left.oclchannels();
+                int data_type = rthis.msg_type;
+
+                String kernelName = get_kernel_name("compute_data_cost_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+
+                //size_t blockSize = 256;
+                size_t localThreads[]  = {32, 8, 1};
+                size_t globalThreads[] = {divUp(w, localThreads[0]) *localThreads[0],
+                    divUp(h, localThreads[1]) *localThreads[1],
+                    1
+                };
+
+                int disp_step1 = msg_step1 * h;
+                int disp_step2 = msg_step2 * h2;
+                openCLVerifyKernel(clCxt, kernel, localThreads);
+                openCLSafeCall(clSetKernelArg(kernel, 0,  sizeof(cl_mem),  (void *)&disp_selected_pyr));
+                openCLSafeCall(clSetKernelArg(kernel, 1,  sizeof(cl_mem),  (void *)&data_cost));
+                openCLSafeCall(clSetKernelArg(kernel, 2,  sizeof(cl_mem),  (void *)&left.data));
+                openCLSafeCall(clSetKernelArg(kernel, 3,  sizeof(cl_mem),  (void *)&right.data));
+                openCLSafeCall(clSetKernelArg(kernel, 4,  sizeof(cl_int),  (void *)&h));
+                openCLSafeCall(clSetKernelArg(kernel, 5,  sizeof(cl_int),  (void *)&w));
+                openCLSafeCall(clSetKernelArg(kernel, 6,  sizeof(cl_int),  (void *)&level));
+                openCLSafeCall(clSetKernelArg(kernel, 7,  sizeof(cl_int),  (void *)&nr_plane));
+                openCLSafeCall(clSetKernelArg(kernel, 8,  sizeof(cl_int),  (void *)&channels));
+                openCLSafeCall(clSetKernelArg(kernel, 9,  sizeof(cl_int),  (void *)&msg_step1));
+                openCLSafeCall(clSetKernelArg(kernel, 10, sizeof(cl_int),  (void *)&msg_step2));
+                openCLSafeCall(clSetKernelArg(kernel, 11, sizeof(cl_int),  (void *)&disp_step1));
+                openCLSafeCall(clSetKernelArg(kernel, 12, sizeof(cl_int),  (void *)&disp_step2));
+                openCLSafeCall(clSetKernelArg(kernel, 13, sizeof(cl_float), (void *)&rthis.data_weight));
+                openCLSafeCall(clSetKernelArg(kernel, 14, sizeof(cl_float), (void *)&rthis.max_data_term));
+                openCLSafeCall(clSetKernelArg(kernel, 15, sizeof(cl_int),  (void *)&left.step));
+                openCLSafeCall(clSetKernelArg(kernel, 16, sizeof(cl_int),  (void *)&rthis.min_disp_th));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 2, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+            static void compute_data_cost_reduce_caller(uchar *disp_selected_pyr, uchar *data_cost,
+                StereoConstantSpaceBP &rthis, int msg_step1,
+                int msg_step2, const oclMat &left, const oclMat &right, int h,
+                int w, int h2, int level, int nr_plane)
+            {
+                Context  *clCxt = left.clCxt;
+                int data_type = rthis.msg_type;
+                int channels = left.oclchannels();
+                int win_size = (int)std::pow(2.f, level);
+
+                String kernelName = get_kernel_name("compute_data_cost_reduce_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+
+                const size_t threadsNum = 256;
+                //size_t blockSize = threadsNum;
+                size_t localThreads[3]  = {win_size, 1, threadsNum / win_size};
+                size_t globalThreads[3] = {w *localThreads[0],
+                    h * divUp(nr_plane, localThreads[2]) *localThreads[1], 1 * localThreads[2]
+                };
+
+                int disp_step1 = msg_step1 * h;
+                int disp_step2 = msg_step2 * h2;
+                size_t local_mem_size = threadsNum * sizeof(float);
+                openCLVerifyKernel(clCxt, kernel, localThreads);
+                openCLSafeCall(clSetKernelArg(kernel, 0,  sizeof(cl_mem),  (void *)&disp_selected_pyr));
+                openCLSafeCall(clSetKernelArg(kernel, 1,  sizeof(cl_mem),  (void *)&data_cost));
+                openCLSafeCall(clSetKernelArg(kernel, 2,  sizeof(cl_mem),  (void *)&left.data));
+                openCLSafeCall(clSetKernelArg(kernel, 3,  sizeof(cl_mem),  (void *)&right.data));
+                openCLSafeCall(clSetKernelArg(kernel, 4, local_mem_size,   (void *)NULL));
+                openCLSafeCall(clSetKernelArg(kernel, 5,  sizeof(cl_int),  (void *)&level));
+                openCLSafeCall(clSetKernelArg(kernel, 6,  sizeof(cl_int),  (void *)&left.rows));
+                openCLSafeCall(clSetKernelArg(kernel, 7,  sizeof(cl_int),  (void *)&left.cols));
+                openCLSafeCall(clSetKernelArg(kernel, 8,  sizeof(cl_int),  (void *)&h));
+                openCLSafeCall(clSetKernelArg(kernel, 9,  sizeof(cl_int),  (void *)&nr_plane));
+                openCLSafeCall(clSetKernelArg(kernel, 10, sizeof(cl_int),  (void *)&channels));
+                openCLSafeCall(clSetKernelArg(kernel, 11, sizeof(cl_int),  (void *)&win_size));
+                openCLSafeCall(clSetKernelArg(kernel, 12, sizeof(cl_int),  (void *)&msg_step1));
+                openCLSafeCall(clSetKernelArg(kernel, 13, sizeof(cl_int),  (void *)&msg_step2));
+                openCLSafeCall(clSetKernelArg(kernel, 14, sizeof(cl_int),  (void *)&disp_step1));
+                openCLSafeCall(clSetKernelArg(kernel, 15, sizeof(cl_int),  (void *)&disp_step2));
+                openCLSafeCall(clSetKernelArg(kernel, 16, sizeof(cl_float), (void *)&rthis.data_weight));
+                openCLSafeCall(clSetKernelArg(kernel, 17, sizeof(cl_float), (void *)&rthis.max_data_term));
+                openCLSafeCall(clSetKernelArg(kernel, 18, sizeof(cl_int),  (void *)&left.step));
+                openCLSafeCall(clSetKernelArg(kernel, 19, sizeof(cl_int),  (void *)&rthis.min_disp_th));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 3, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+            static void compute_data_cost(uchar *disp_selected_pyr, uchar *data_cost, StereoConstantSpaceBP &rthis,
+                int msg_step1, int msg_step2, const oclMat &left, const oclMat &right, int h, int w,
+                int h2, int level, int nr_plane)
+            {
+                if(level <= 1)
+                    compute_data_cost_caller(disp_selected_pyr, data_cost, rthis, msg_step1, msg_step2,
+                    left, right, h, w, h2, level, nr_plane);
+                else
+                    compute_data_cost_reduce_caller(disp_selected_pyr, data_cost, rthis,  msg_step1, msg_step2,
+                    left, right, h, w, h2, level, nr_plane);
+            }
+            ////////////////////////////////////////////////////////////////////////////////////////////////
+            //////////////////////////////////////init message//////////////////////////////////////////////
+            ////////////////////////////////////////////////////////////////////////////////////////////////
+            static void init_message(uchar *u_new, uchar *d_new, uchar *l_new, uchar *r_new,
+                uchar *u_cur, uchar *d_cur, uchar *l_cur, uchar *r_cur,
+                uchar *disp_selected_pyr_new, uchar *disp_selected_pyr_cur,
+                uchar *data_cost_selected, uchar *data_cost, oclMat &temp, StereoConstantSpaceBP rthis,
+                size_t msg_step1, size_t msg_step2, int h, int w, int nr_plane,
+                int h2, int w2, int nr_plane2)
+            {
+                Context  *clCxt = temp.clCxt;
+                int data_type = rthis.msg_type;
+
+                String kernelName = get_kernel_name("init_message_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+
+                //size_t blockSize = 256;
+                size_t localThreads[]  = {32, 8, 1};
+                size_t globalThreads[] = {divUp(w, localThreads[0]) *localThreads[0],
+                    divUp(h, localThreads[1]) *localThreads[1],
+                    1
+                };
+
+                int disp_step1 = msg_step1 * h;
+                int disp_step2 = msg_step2 * h2;
+                openCLVerifyKernel(clCxt, kernel, localThreads);
+                openCLSafeCall(clSetKernelArg(kernel, 0,  sizeof(cl_mem), (void *)&u_new));
+                openCLSafeCall(clSetKernelArg(kernel, 1,  sizeof(cl_mem), (void *)&d_new));
+                openCLSafeCall(clSetKernelArg(kernel, 2,  sizeof(cl_mem), (void *)&l_new));
+                openCLSafeCall(clSetKernelArg(kernel, 3,  sizeof(cl_mem), (void *)&r_new));
+                openCLSafeCall(clSetKernelArg(kernel, 4,  sizeof(cl_mem), (void *)&u_cur));
+                openCLSafeCall(clSetKernelArg(kernel, 5,  sizeof(cl_mem), (void *)&d_cur));
+                openCLSafeCall(clSetKernelArg(kernel, 6,  sizeof(cl_mem), (void *)&l_cur));
+                openCLSafeCall(clSetKernelArg(kernel, 7,  sizeof(cl_mem), (void *)&r_cur));
+                openCLSafeCall(clSetKernelArg(kernel, 8,  sizeof(cl_mem), (void *)&temp.data));
+                openCLSafeCall(clSetKernelArg(kernel, 9,  sizeof(cl_mem), (void *)&disp_selected_pyr_new));
+                openCLSafeCall(clSetKernelArg(kernel, 10, sizeof(cl_mem), (void *)&disp_selected_pyr_cur));
+                openCLSafeCall(clSetKernelArg(kernel, 11, sizeof(cl_mem), (void *)&data_cost_selected));
+                openCLSafeCall(clSetKernelArg(kernel, 12, sizeof(cl_mem), (void *)&data_cost));
+                openCLSafeCall(clSetKernelArg(kernel, 13, sizeof(cl_int), (void *)&h));
+                openCLSafeCall(clSetKernelArg(kernel, 14, sizeof(cl_int), (void *)&w));
+                openCLSafeCall(clSetKernelArg(kernel, 15, sizeof(cl_int), (void *)&nr_plane));
+                openCLSafeCall(clSetKernelArg(kernel, 16, sizeof(cl_int), (void *)&h2));
+                openCLSafeCall(clSetKernelArg(kernel, 17, sizeof(cl_int), (void *)&w2));
+                openCLSafeCall(clSetKernelArg(kernel, 18, sizeof(cl_int), (void *)&nr_plane2));
+                openCLSafeCall(clSetKernelArg(kernel, 19, sizeof(cl_int), (void *)&disp_step1));
+                openCLSafeCall(clSetKernelArg(kernel, 20, sizeof(cl_int), (void *)&disp_step2));
+                openCLSafeCall(clSetKernelArg(kernel, 21, sizeof(cl_int), (void *)&msg_step1));
+                openCLSafeCall(clSetKernelArg(kernel, 22, sizeof(cl_int), (void *)&msg_step2));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 2, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+            ////////////////////////////////////////////////////////////////////////////////////////////////
+            ///////////////////////////calc_all_iterations////////////////////////////////////////////////
+            //////////////////////////////////////////////////////////////////////////////////////////////
+            static void calc_all_iterations_caller(uchar *u, uchar *d, uchar *l, uchar *r, uchar *data_cost_selected,
+                uchar *disp_selected_pyr, oclMat &temp, StereoConstantSpaceBP rthis,
+                int msg_step, int h, int w, int nr_plane, int i)
+            {
+                Context  *clCxt = temp.clCxt;
+                int data_type = rthis.msg_type;
+
+                String kernelName = get_kernel_name("compute_message_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+                size_t localThreads[]  = {32, 8, 1};
+                size_t globalThreads[] = {divUp(w, (localThreads[0]) << 1) *localThreads[0],
+                    divUp(h, localThreads[1]) *localThreads[1],
+                    1
+                };
+
+                int disp_step = msg_step * h;
+                openCLVerifyKernel(clCxt, kernel, localThreads);
+                openCLSafeCall(clSetKernelArg(kernel, 0,  sizeof(cl_mem),  (void *)&u));
+                openCLSafeCall(clSetKernelArg(kernel, 1,  sizeof(cl_mem),  (void *)&d));
+                openCLSafeCall(clSetKernelArg(kernel, 2,  sizeof(cl_mem),  (void *)&l));
+                openCLSafeCall(clSetKernelArg(kernel, 3,  sizeof(cl_mem),  (void *)&r));
+                openCLSafeCall(clSetKernelArg(kernel, 4,  sizeof(cl_mem),  (void *)&data_cost_selected));
+                openCLSafeCall(clSetKernelArg(kernel, 5,  sizeof(cl_mem),  (void *)&disp_selected_pyr));
+                openCLSafeCall(clSetKernelArg(kernel, 6,  sizeof(cl_mem),  (void *)&temp.data));
+                openCLSafeCall(clSetKernelArg(kernel, 7,  sizeof(cl_int),  (void *)&h));
+                openCLSafeCall(clSetKernelArg(kernel, 8,  sizeof(cl_int),  (void *)&w));
+                openCLSafeCall(clSetKernelArg(kernel, 9,  sizeof(cl_int),  (void *)&nr_plane));
+                openCLSafeCall(clSetKernelArg(kernel, 10, sizeof(cl_int),  (void *)&i));
+                openCLSafeCall(clSetKernelArg(kernel, 11, sizeof(cl_float), (void *)&rthis.max_disc_term));
+                openCLSafeCall(clSetKernelArg(kernel, 12, sizeof(cl_int),  (void *)&disp_step));
+                openCLSafeCall(clSetKernelArg(kernel, 13, sizeof(cl_int),  (void *)&msg_step));
+                openCLSafeCall(clSetKernelArg(kernel, 14, sizeof(cl_float), (void *)&rthis.disc_single_jump));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 2, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+            static void calc_all_iterations(uchar *u, uchar *d, uchar *l, uchar *r, uchar *data_cost_selected,
+                uchar *disp_selected_pyr, oclMat &temp, StereoConstantSpaceBP rthis,
+                int msg_step, int h, int w, int nr_plane)
+            {
+                for(int t = 0; t < rthis.iters; t++)
+                    calc_all_iterations_caller(u, d, l, r, data_cost_selected, disp_selected_pyr, temp, rthis,
+                    msg_step, h, w, nr_plane, t & 1);
+            }
+
+            ///////////////////////////////////////////////////////////////////////////////////////////////
+            //////////////////////////compute_disp////////////////////////////////////////////////////////
+            /////////////////////////////////////////////////////////////////////////////////////////////
+            static void compute_disp(uchar *u, uchar *d, uchar *l, uchar *r, uchar *data_cost_selected,
+                uchar *disp_selected_pyr, StereoConstantSpaceBP &rthis, size_t msg_step,
+                oclMat &disp, int nr_plane)
+            {
+                Context  *clCxt = disp.clCxt;
+                int data_type = rthis.msg_type;
+
+                String kernelName = get_kernel_name("compute_disp_", data_type);
+
+                cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereocsbp, kernelName);
+
+                //size_t blockSize = 256;
+                size_t localThreads[]  = {32, 8, 1};
+                size_t globalThreads[] = {divUp(disp.cols, localThreads[0]) *localThreads[0],
+                    divUp(disp.rows, localThreads[1]) *localThreads[1],
+                    1
+                };
+
+                int step_size = disp.step / disp.elemSize();
+                int disp_step = disp.rows * msg_step;
+                openCLVerifyKernel(clCxt, kernel, localThreads);
+                openCLSafeCall(clSetKernelArg(kernel, 0,  sizeof(cl_mem), (void *)&u));
+                openCLSafeCall(clSetKernelArg(kernel, 1,  sizeof(cl_mem), (void *)&d));
+                openCLSafeCall(clSetKernelArg(kernel, 2,  sizeof(cl_mem), (void *)&l));
+                openCLSafeCall(clSetKernelArg(kernel, 3,  sizeof(cl_mem), (void *)&r));
+                openCLSafeCall(clSetKernelArg(kernel, 4,  sizeof(cl_mem), (void *)&data_cost_selected));
+                openCLSafeCall(clSetKernelArg(kernel, 5,  sizeof(cl_mem), (void *)&disp_selected_pyr));
+                openCLSafeCall(clSetKernelArg(kernel, 6,  sizeof(cl_mem), (void *)&disp.data));
+                openCLSafeCall(clSetKernelArg(kernel, 7,  sizeof(cl_int), (void *)&step_size));
+                openCLSafeCall(clSetKernelArg(kernel, 8,  sizeof(cl_int), (void *)&disp.cols));
+                openCLSafeCall(clSetKernelArg(kernel, 9,  sizeof(cl_int), (void *)&disp.rows));
+                openCLSafeCall(clSetKernelArg(kernel, 10, sizeof(cl_int), (void *)&nr_plane));
+                openCLSafeCall(clSetKernelArg(kernel, 11, sizeof(cl_int), (void *)&msg_step));
+                openCLSafeCall(clSetKernelArg(kernel, 12, sizeof(cl_int), (void *)&disp_step));
+                openCLSafeCall(clEnqueueNDRangeKernel(*(cl_command_queue*)getoclCommandQueue(), kernel, 2, NULL,
+                    globalThreads, localThreads, 0, NULL, NULL));
+
+                clFinish(*(cl_command_queue*)getoclCommandQueue());
+                openCLSafeCall(clReleaseKernel(kernel));
+            }
+        }
+    }
+}
+namespace
+{
+    const float DEFAULT_MAX_DATA_TERM = 30.0f;
+    const float DEFAULT_DATA_WEIGHT = 1.0f;
+    const float DEFAULT_MAX_DISC_TERM = 160.0f;
+    const float DEFAULT_DISC_SINGLE_JUMP = 10.0f;
+}
+
+void cv::ocl::StereoConstantSpaceBP::estimateRecommendedParams(int width, int height, int &ndisp, int &iters, int &levels, int &nr_plane)
+{
+    ndisp = (int) ((float) width / 3.14f);
+    if ((ndisp & 1) != 0)
+        ndisp++;
+
+    int mm = ::max(width, height);
+    iters = mm / 100 + ((mm > 1200) ? - 4 : 4);
+
+    levels = (int)::log(static_cast<double>(mm)) * 2 / 3;
+    if (levels == 0) levels++;
+
+    nr_plane = (int) ((float) ndisp / std::pow(2.0, levels + 1));
+}
+
+cv::ocl::StereoConstantSpaceBP::StereoConstantSpaceBP(int ndisp_, int iters_, int levels_, int nr_plane_,
+    int msg_type_)
+
+    : ndisp(ndisp_), iters(iters_), levels(levels_), nr_plane(nr_plane_),
+    max_data_term(DEFAULT_MAX_DATA_TERM), data_weight(DEFAULT_DATA_WEIGHT),
+    max_disc_term(DEFAULT_MAX_DISC_TERM), disc_single_jump(DEFAULT_DISC_SINGLE_JUMP), min_disp_th(0),
+    msg_type(msg_type_), use_local_init_data_cost(true)
+{
+    CV_Assert(msg_type_ == CV_32F || msg_type_ == CV_16S);
+}
+
+
+cv::ocl::StereoConstantSpaceBP::StereoConstantSpaceBP(int ndisp_, int iters_, int levels_, int nr_plane_,
+    float max_data_term_, float data_weight_, float max_disc_term_, float disc_single_jump_,
+    int min_disp_th_, int msg_type_)
+    : ndisp(ndisp_), iters(iters_), levels(levels_), nr_plane(nr_plane_),
+    max_data_term(max_data_term_), data_weight(data_weight_),
+    max_disc_term(max_disc_term_), disc_single_jump(disc_single_jump_), min_disp_th(min_disp_th_),
+    msg_type(msg_type_), use_local_init_data_cost(true)
+{
+    CV_Assert(msg_type_ == CV_32F || msg_type_ == CV_16S);
+}
+
+template<class T>
+static void csbp_operator(StereoConstantSpaceBP &rthis, oclMat u[2], oclMat d[2], oclMat l[2], oclMat r[2],
+    oclMat disp_selected_pyr[2], oclMat &data_cost, oclMat &data_cost_selected,
+    oclMat &temp, oclMat &out, const oclMat &left, const oclMat &right, oclMat &disp)
+{
+    CV_DbgAssert(0 < rthis.ndisp && 0 < rthis.iters && 0 < rthis.levels && 0 < rthis.nr_plane
+        && left.rows == right.rows && left.cols == right.cols && left.type() == right.type());
+
+    CV_Assert(rthis.levels <= 8 && (left.type() == CV_8UC1 || left.type() == CV_8UC3));
+
+    const Scalar zero = Scalar::all(0);
+
+    ////////////////////////////////////Init///////////////////////////////////////////////////
+    int rows = left.rows;
+    int cols = left.cols;
+
+    rthis.levels = min(rthis.levels, int(log((double)rthis.ndisp) / log(2.0)));
+    int levels = rthis.levels;
+
+    AutoBuffer<int> buf(levels * 4);
+
+    int *cols_pyr = buf;
+    int *rows_pyr = cols_pyr + levels;
+    int *nr_plane_pyr = rows_pyr + levels;
+    int *step_pyr = nr_plane_pyr + levels;
+
+    cols_pyr[0] = cols;
+    rows_pyr[0] = rows;
+    nr_plane_pyr[0] = rthis.nr_plane;
+
+    const int n = 64;
+    step_pyr[0] = alignSize(cols * sizeof(T), n) / sizeof(T);
+    for (int i = 1; i < levels; i++)
+    {
+        cols_pyr[i] = cols_pyr[i - 1]  / 2;
+        rows_pyr[i] = rows_pyr[i - 1]/ 2;
+
+        nr_plane_pyr[i] = nr_plane_pyr[i - 1] * 2;
+
+        step_pyr[i] = alignSize(cols_pyr[i] * sizeof(T), n) / sizeof(T);
+    }
+
+    Size msg_size(step_pyr[0], rows * nr_plane_pyr[0]);
+    Size data_cost_size(step_pyr[0], rows * nr_plane_pyr[0] * 2);
+
+    u[0].create(msg_size, DataType<T>::type);
+    d[0].create(msg_size, DataType<T>::type);
+    l[0].create(msg_size, DataType<T>::type);
+    r[0].create(msg_size, DataType<T>::type);
+
+    u[1].create(msg_size, DataType<T>::type);
+    d[1].create(msg_size, DataType<T>::type);
+    l[1].create(msg_size, DataType<T>::type);
+    r[1].create(msg_size, DataType<T>::type);
+
+    disp_selected_pyr[0].create(msg_size, DataType<T>::type);
+    disp_selected_pyr[1].create(msg_size, DataType<T>::type);
+
+    data_cost.create(data_cost_size, DataType<T>::type);
+    data_cost_selected.create(msg_size, DataType<T>::type);
+
+    Size temp_size = data_cost_size;
+    if (data_cost_size.width * data_cost_size.height < step_pyr[0] * rows_pyr[levels - 1] * rthis.ndisp)
+        temp_size = Size(step_pyr[0], rows_pyr[levels - 1] * rthis.ndisp);
+
+    temp.create(temp_size, DataType<T>::type);
+    temp = zero;
+
+    ///////////////////////////////// Compute////////////////////////////////////////////////
+
+    //csbp::load_constants(rthis.ndisp, rthis.max_data_term, rthis.data_weight,
+    //   rthis.max_disc_term, rthis.disc_single_jump, rthis.min_disp_th, left, right, temp);
+
+    l[0] = zero;
+    d[0] = zero;
+    r[0] = zero;
+    u[0] = zero;
+    disp_selected_pyr[0] = zero;
+
+    l[1] = zero;
+    d[1] = zero;
+    r[1] = zero;
+    u[1] = zero;
+    disp_selected_pyr[1] = zero;
+
+    data_cost = zero;
+
+    data_cost_selected = zero;
+
+    int cur_idx = 0;
+
+    for (int i = levels - 1; i >= 0; i--)
+    {
+        if (i == levels - 1)
+        {
+            cv::ocl::stereoCSBP::init_data_cost(left, right, temp, rthis, disp_selected_pyr[cur_idx].data,
+                data_cost_selected.data, step_pyr[0], rows_pyr[i], cols_pyr[i],
+                i, nr_plane_pyr[i]);
+        }
+        else
+        {
+            cv::ocl::stereoCSBP::compute_data_cost(
+                disp_selected_pyr[cur_idx].data, data_cost.data, rthis, step_pyr[0],
+                step_pyr[0], left, right, rows_pyr[i], cols_pyr[i], rows_pyr[i + 1], i,
+                nr_plane_pyr[i + 1]);
+
+            int new_idx = (cur_idx + 1) & 1;
+
+            cv::ocl::stereoCSBP::init_message(u[new_idx].data, d[new_idx].data, l[new_idx].data, r[new_idx].data,
+                u[cur_idx].data, d[cur_idx].data, l[cur_idx].data, r[cur_idx].data,
+                disp_selected_pyr[new_idx].data, disp_selected_pyr[cur_idx].data,
+                data_cost_selected.data, data_cost.data, temp, rthis, step_pyr[0],
+                step_pyr[0], rows_pyr[i], cols_pyr[i], nr_plane_pyr[i], rows_pyr[i + 1],
+                cols_pyr[i + 1], nr_plane_pyr[i + 1]);
+            cur_idx = new_idx;
+        }
+        cv::ocl::stereoCSBP::calc_all_iterations(u[cur_idx].data, d[cur_idx].data, l[cur_idx].data, r[cur_idx].data,
+            data_cost_selected.data, disp_selected_pyr[cur_idx].data, temp,
+            rthis, step_pyr[0], rows_pyr[i], cols_pyr[i], nr_plane_pyr[i]);
+    }
+
+    if (disp.empty())
+        disp.create(rows, cols, CV_16S);
+
+    out = ((disp.type() == CV_16S) ? disp : (out.create(rows, cols, CV_16S), out));
+    out = zero;
+
+    stereoCSBP::compute_disp(u[cur_idx].data, d[cur_idx].data, l[cur_idx].data, r[cur_idx].data,
+        data_cost_selected.data, disp_selected_pyr[cur_idx].data, rthis, step_pyr[0],
+        out, nr_plane_pyr[0]);
+    if (disp.type() != CV_16S)
+        out.convertTo(disp, disp.type());
+}
+
+
+typedef void (*csbp_operator_t)(StereoConstantSpaceBP &rthis, oclMat u[2], oclMat d[2], oclMat l[2], oclMat r[2],
+    oclMat disp_selected_pyr[2], oclMat &data_cost, oclMat &data_cost_selected,
+    oclMat &temp, oclMat &out, const oclMat &left, const oclMat &right, oclMat &disp);
+
+const static csbp_operator_t operators[] = {0, 0, 0, csbp_operator<short>, 0, csbp_operator<float>, 0, 0};
+
+void cv::ocl::StereoConstantSpaceBP::operator()(const oclMat &left, const oclMat &right, oclMat &disp)
+{
+
+    CV_Assert(msg_type == CV_32F || msg_type == CV_16S);
+    operators[msg_type](*this, u, d, l, r, disp_selected_pyr, data_cost, data_cost_selected, temp, out,
+        left, right, disp);
+}
+
+#endif /* !defined (HAVE_OPENCL) */

modules/ocl/src/stereobm.cpp

@@ -72,28 +72,21 @@ namespace stereoBM
 ////////////////////////////////////////////////////////////////////////
 static void prefilter_xsobel(const oclMat &input, oclMat &output, int prefilterCap)
 {
-    Context *clCxt = input.clCxt;
-
     String kernelName = "prefilter_xsobel";
-    cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereobm, kernelName);
 
     size_t blockSize = 1;
     size_t globalThreads[3] = { input.cols, input.rows, 1 };
     size_t localThreads[3]  = { blockSize, blockSize, 1 };
 
-    openCLVerifyKernel(clCxt, kernel,  localThreads);
-    openCLSafeCall(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&input.data));
-    openCLSafeCall(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&output.data));
-    openCLSafeCall(clSetKernelArg(kernel, 2, sizeof(cl_int), (void *)&input.rows));
-    openCLSafeCall(clSetKernelArg(kernel, 3, sizeof(cl_int), (void *)&input.cols));
-    openCLSafeCall(clSetKernelArg(kernel, 4, sizeof(cl_int), (void *)&prefilterCap));
-
-    openCLSafeCall(clEnqueueNDRangeKernel((cl_command_queue)clCxt->oclCommandQueue(), kernel, 3, NULL,
-                                          globalThreads, localThreads, 0, NULL, NULL));
-
-    clFinish((cl_command_queue)clCxt->oclCommandQueue());
-    openCLSafeCall(clReleaseKernel(kernel));
+    std::vector< std::pair<size_t, const void *> > args;
+    args.push_back(std::make_pair(sizeof(cl_mem), (void *)&input.data));
+    args.push_back(std::make_pair(sizeof(cl_mem), (void *)&output.data));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&input.rows));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&input.cols));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&prefilterCap));
 
+    openCLExecuteKernel(Context::getContext(), &stereobm, kernelName,
+        globalThreads, localThreads, args, -1, -1);
 }
 //////////////////////////////////////////////////////////////////////////
 //////////////////////////////common////////////////////////////////////
@@ -113,16 +106,13 @@ static void stereo_bm(const oclMat &left, const oclMat &right,  oclMat &disp,
 {
     int winsz2 = winSize >> 1;
 
-    Context *clCxt = left.clCxt;
-
     String kernelName = "stereoKernel";
-    cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereobm, kernelName);
 
     disp.setTo(Scalar_<unsigned char>::all(0));
     minSSD_buf.setTo(Scalar_<unsigned int>::all(0xFFFFFFFF));
 
     size_t minssd_step = minSSD_buf.step / minSSD_buf.elemSize();
-    size_t local_mem_size = (BLOCK_W + N_DISPARITIES * (BLOCK_W + 2 * winsz2)) *
+    size_t local_mem_size = (N_DISPARITIES * (BLOCK_W + 2 * winsz2)) *
                             sizeof(cl_uint);
     //size_t blockSize = 1;
     size_t localThreads[]  = { BLOCK_W, 1,1};
@@ -131,26 +121,23 @@ static void stereo_bm(const oclMat &left, const oclMat &right,  oclMat &disp,
                                1
                              };
 
-    openCLVerifyKernel(clCxt, kernel, localThreads);
-    openCLSafeCall(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&left.data));
-    openCLSafeCall(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&right.data));
-    openCLSafeCall(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *)&minSSD_buf.data));
-    openCLSafeCall(clSetKernelArg(kernel, 3, sizeof(cl_int), (void *)&minssd_step));
-    openCLSafeCall(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *)&disp.data));
-    openCLSafeCall(clSetKernelArg(kernel, 5, sizeof(cl_int), (void *)&disp.step));
-    openCLSafeCall(clSetKernelArg(kernel, 6, sizeof(cl_int), (void *)&left.cols));
-    openCLSafeCall(clSetKernelArg(kernel, 7, sizeof(cl_int), (void *)&left.rows));
-    openCLSafeCall(clSetKernelArg(kernel, 8, sizeof(cl_int), (void *)&left.step));
-    openCLSafeCall(clSetKernelArg(kernel, 9, sizeof(cl_int), (void *)&maxdisp));
-    openCLSafeCall(clSetKernelArg(kernel, 10, sizeof(cl_int), (void *)&winsz2));
-    openCLSafeCall(clSetKernelArg(kernel, 11, local_mem_size, (void *)NULL));
-
-    openCLSafeCall(clEnqueueNDRangeKernel((cl_command_queue)clCxt->oclCommandQueue(), kernel, 2, NULL,
-                                          globalThreads, localThreads, 0, NULL, NULL));
-
-
-    clFinish((cl_command_queue)clCxt->oclCommandQueue());
-    openCLSafeCall(clReleaseKernel(kernel));
+    std::vector< std::pair<size_t, const void *> > args;
+    args.push_back(std::make_pair(sizeof(cl_mem), (void *)&left.data));
+    args.push_back(std::make_pair(sizeof(cl_mem), (void *)&right.data));
+    args.push_back(std::make_pair(sizeof(cl_mem), (void *)&minSSD_buf.data));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&minssd_step));
+    args.push_back(std::make_pair(sizeof(cl_mem), (void *)&disp.data));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&disp.step));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&left.cols));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&left.rows));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&left.step));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&maxdisp));
+    args.push_back(std::make_pair(local_mem_size, (void *)NULL));
+
+    char opt [128];
+    sprintf(opt, "-D radius=%d", winsz2);
+    openCLExecuteKernel(Context::getContext(), &stereobm, kernelName,
+        globalThreads, localThreads, args, -1, -1, opt);
 }
 ////////////////////////////////////////////////////////////////////////////
 ///////////////////////////////postfilter_textureness///////////////////////
@@ -158,10 +145,7 @@ static void stereo_bm(const oclMat &left, const oclMat &right,  oclMat &disp,
 static void postfilter_textureness(oclMat &left, int winSize,
                             float avergeTexThreshold, oclMat &disparity)
 {
-    Context *clCxt = left.clCxt;
-
     String kernelName = "textureness_kernel";
-    cl_kernel kernel = openCLGetKernelFromSource(clCxt, &stereobm, kernelName);
 
     size_t blockSize = 1;
     size_t localThreads[]  = { BLOCK_W, blockSize ,1};
@@ -172,22 +156,19 @@ static void postfilter_textureness(oclMat &left, int winSize,
 
     size_t local_mem_size = (localThreads[0] + localThreads[0] + (winSize / 2) * 2) * sizeof(float);
 
-    openCLVerifyKernel(clCxt, kernel,  localThreads);
-    openCLSafeCall(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&disparity.data));
-    openCLSafeCall(clSetKernelArg(kernel, 1, sizeof(cl_int), (void *)&disparity.rows));
-    openCLSafeCall(clSetKernelArg(kernel, 2, sizeof(cl_int), (void *)&disparity.cols));
-    openCLSafeCall(clSetKernelArg(kernel, 3, sizeof(cl_int), (void *)&disparity.step));
-    openCLSafeCall(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *)&left.data));
-    openCLSafeCall(clSetKernelArg(kernel, 5, sizeof(cl_int), (void *)&left.rows));
-    openCLSafeCall(clSetKernelArg(kernel, 6, sizeof(cl_int), (void *)&left.cols));
-    openCLSafeCall(clSetKernelArg(kernel, 7, sizeof(cl_int), (void *)&winSize));
-    openCLSafeCall(clSetKernelArg(kernel, 8, sizeof(cl_float), (void *)&avergeTexThreshold));
-    openCLSafeCall(clSetKernelArg(kernel, 9, local_mem_size, NULL));
-    openCLSafeCall(clEnqueueNDRangeKernel((cl_command_queue)clCxt->oclCommandQueue(), kernel, 2, NULL,
-                                          globalThreads, localThreads, 0, NULL, NULL));
-
-    clFinish((cl_command_queue)clCxt->oclCommandQueue());
-    openCLSafeCall(clReleaseKernel(kernel));
+    std::vector< std::pair<size_t, const void *> > args;
+    args.push_back(std::make_pair(sizeof(cl_mem), (void *)&disparity.data));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&disparity.rows));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&disparity.cols));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&disparity.step));
+    args.push_back(std::make_pair(sizeof(cl_mem), (void *)&left.data));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&left.rows));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&left.cols));
+    args.push_back(std::make_pair(sizeof(cl_int), (void *)&winSize));
+    args.push_back(std::make_pair(sizeof(cl_float), (void *)&avergeTexThreshold));
+    args.push_back(std::make_pair(local_mem_size, (void*)NULL));
+    openCLExecuteKernel(Context::getContext(), &stereobm, kernelName,
+        globalThreads, localThreads, args, -1, -1);
 }
 //////////////////////////////////////////////////////////////////////////////
 /////////////////////////////////////operator/////////////////////////////////

modules/ocl/src/tvl1flow.cpp

@@ -0,0 +1,475 @@
+/*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) 2010-2012, Multicoreware, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// @Authors
+//		Jin Ma, jin@multicorewareinc.com
+// 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 oclMaterials 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"
+using namespace std;
+using namespace cv;
+using namespace cv::ocl;
+
+namespace cv
+{
+    namespace ocl
+    {
+        ///////////////////////////OpenCL kernel strings///////////////////////////
+        extern const char* tvl1flow;
+    }
+}
+
+cv::ocl::OpticalFlowDual_TVL1_OCL::OpticalFlowDual_TVL1_OCL()
+{
+    tau            = 0.25;
+    lambda         = 0.15;
+    theta          = 0.3;
+    nscales        = 5;
+    warps          = 5;
+    epsilon        = 0.01;
+    iterations     = 300;
+    useInitialFlow = false;
+}
+
+void cv::ocl::OpticalFlowDual_TVL1_OCL::operator()(const oclMat& I0, const oclMat& I1, oclMat& flowx, oclMat& flowy)
+{
+    CV_Assert( I0.type() == CV_8UC1 || I0.type() == CV_32FC1 );
+    CV_Assert( I0.size() == I1.size() );
+    CV_Assert( I0.type() == I1.type() );
+    CV_Assert( !useInitialFlow || (flowx.size() == I0.size() && flowx.type() == CV_32FC1 && flowy.size() == flowx.size() && flowy.type() == flowx.type()) );
+    CV_Assert( nscales > 0 );
+
+    // allocate memory for the pyramid structure
+    I0s.resize(nscales);
+    I1s.resize(nscales);
+    u1s.resize(nscales);
+    u2s.resize(nscales);
+    //I0s_step == I1s_step
+    I0.convertTo(I0s[0], CV_32F, I0.depth() == CV_8U ? 1.0 : 255.0);
+    I1.convertTo(I1s[0], CV_32F, I1.depth() == CV_8U ? 1.0 : 255.0);
+
+
+    if (!useInitialFlow)
+    {
+        flowx.create(I0.size(), CV_32FC1);
+        flowy.create(I0.size(), CV_32FC1);
+    }
+    //u1s_step != u2s_step
+    u1s[0] = flowx;
+    u2s[0] = flowy;
+
+    I1x_buf.create(I0.size(), CV_32FC1);
+    I1y_buf.create(I0.size(), CV_32FC1);
+
+    I1w_buf.create(I0.size(), CV_32FC1);
+    I1wx_buf.create(I0.size(), CV_32FC1);
+    I1wy_buf.create(I0.size(), CV_32FC1);
+
+    grad_buf.create(I0.size(), CV_32FC1);
+    rho_c_buf.create(I0.size(), CV_32FC1);
+
+    p11_buf.create(I0.size(), CV_32FC1);
+    p12_buf.create(I0.size(), CV_32FC1);
+    p21_buf.create(I0.size(), CV_32FC1);
+    p22_buf.create(I0.size(), CV_32FC1);
+
+    diff_buf.create(I0.size(), CV_32FC1);
+
+    // create the scales
+    for (int s = 1; s < nscales; ++s)
+    {
+        ocl::pyrDown(I0s[s - 1], I0s[s]);
+        ocl::pyrDown(I1s[s - 1], I1s[s]);
+
+        if (I0s[s].cols < 16 || I0s[s].rows < 16)
+        {
+            nscales = s;
+            break;
+        }
+
+        if (useInitialFlow)
+        {
+            ocl::pyrDown(u1s[s - 1], u1s[s]);
+            ocl::pyrDown(u2s[s - 1], u2s[s]);
+
+            //ocl::multiply(u1s[s], Scalar::all(0.5), u1s[s]);
+            multiply(0.5, u1s[s], u1s[s]);
+            //ocl::multiply(u2s[s], Scalar::all(0.5), u2s[s]);
+            multiply(0.5, u1s[s], u2s[s]);
+        }
+    }
+
+    // pyramidal structure for computing the optical flow
+    for (int s = nscales - 1; s >= 0; --s)
+    {
+        // compute the optical flow at the current scale
+        procOneScale(I0s[s], I1s[s], u1s[s], u2s[s]);
+
+        // if this was the last scale, finish now
+        if (s == 0)
+            break;
+
+        // otherwise, upsample the optical flow
+
+        // zoom the optical flow for the next finer scale
+        ocl::resize(u1s[s], u1s[s - 1], I0s[s - 1].size());
+        ocl::resize(u2s[s], u2s[s - 1], I0s[s - 1].size());
+
+        // scale the optical flow with the appropriate zoom factor
+        multiply(2, u1s[s - 1], u1s[s - 1]);
+        multiply(2, u2s[s - 1], u2s[s - 1]);
+
+    }
+
+}
+
+namespace ocl_tvl1flow
+{
+    void centeredGradient(const oclMat &src, oclMat &dx, oclMat &dy);
+
+    void warpBackward(const oclMat &I0, const oclMat &I1, oclMat &I1x, oclMat &I1y,
+        oclMat &u1, oclMat &u2, oclMat &I1w, oclMat &I1wx, oclMat &I1wy,
+        oclMat &grad, oclMat &rho);
+
+    void estimateU(oclMat &I1wx, oclMat &I1wy, oclMat &grad,
+        oclMat &rho_c, oclMat &p11, oclMat &p12,
+        oclMat &p21, oclMat &p22, oclMat &u1,
+        oclMat &u2, oclMat &error, float l_t, float theta);
+
+    void estimateDualVariables(oclMat &u1, oclMat &u2,
+        oclMat &p11, oclMat &p12, oclMat &p21, oclMat &p22, float taut);
+}
+
+void cv::ocl::OpticalFlowDual_TVL1_OCL::procOneScale(const oclMat &I0, const oclMat &I1, oclMat &u1, oclMat &u2)
+{
+    using namespace ocl_tvl1flow;
+
+    const double scaledEpsilon = epsilon * epsilon * I0.size().area();
+
+    CV_DbgAssert( I1.size() == I0.size() );
+    CV_DbgAssert( I1.type() == I0.type() );
+    CV_DbgAssert( u1.empty() || u1.size() == I0.size() );
+    CV_DbgAssert( u2.size() == u1.size() );
+
+    if (u1.empty())
+    {
+        u1.create(I0.size(), CV_32FC1);
+        u1.setTo(Scalar::all(0));
+
+        u2.create(I0.size(), CV_32FC1);
+        u2.setTo(Scalar::all(0));
+    }
+
+    oclMat I1x = I1x_buf(Rect(0, 0, I0.cols, I0.rows));
+    oclMat I1y = I1y_buf(Rect(0, 0, I0.cols, I0.rows));
+
+    centeredGradient(I1, I1x, I1y);
+
+    oclMat I1w = I1w_buf(Rect(0, 0, I0.cols, I0.rows));
+    oclMat I1wx = I1wx_buf(Rect(0, 0, I0.cols, I0.rows));
+    oclMat I1wy = I1wy_buf(Rect(0, 0, I0.cols, I0.rows));
+
+    oclMat grad = grad_buf(Rect(0, 0, I0.cols, I0.rows));
+    oclMat rho_c = rho_c_buf(Rect(0, 0, I0.cols, I0.rows));
+
+    oclMat p11 = p11_buf(Rect(0, 0, I0.cols, I0.rows));
+    oclMat p12 = p12_buf(Rect(0, 0, I0.cols, I0.rows));
+    oclMat p21 = p21_buf(Rect(0, 0, I0.cols, I0.rows));
+    oclMat p22 = p22_buf(Rect(0, 0, I0.cols, I0.rows));
+    p11.setTo(Scalar::all(0));
+    p12.setTo(Scalar::all(0));
+    p21.setTo(Scalar::all(0));
+    p22.setTo(Scalar::all(0));
+
+    oclMat diff = diff_buf(Rect(0, 0, I0.cols, I0.rows));
+
+    const float l_t = static_cast<float>(lambda * theta);
+    const float taut = static_cast<float>(tau / theta);
+
+    for (int warpings = 0; warpings < warps; ++warpings)
+    {
+        warpBackward(I0, I1, I1x, I1y, u1, u2, I1w, I1wx, I1wy, grad, rho_c);
+
+        double error = numeric_limits<double>::max();
+        for (int n = 0; error > scaledEpsilon && n < iterations; ++n)
+        {
+            estimateU(I1wx, I1wy, grad, rho_c, p11, p12, p21, p22,
+                u1, u2, diff, l_t, static_cast<float>(theta));
+
+            error = ocl::sum(diff)[0];
+
+            estimateDualVariables(u1, u2, p11, p12, p21, p22, taut);
+
+        }
+    }
+
+}
+
+void cv::ocl::OpticalFlowDual_TVL1_OCL::collectGarbage()
+{
+    I0s.clear();
+    I1s.clear();
+    u1s.clear();
+    u2s.clear();
+
+    I1x_buf.release();
+    I1y_buf.release();
+
+    I1w_buf.release();
+    I1wx_buf.release();
+    I1wy_buf.release();
+
+    grad_buf.release();
+    rho_c_buf.release();
+
+    p11_buf.release();
+    p12_buf.release();
+    p21_buf.release();
+    p22_buf.release();
+
+    diff_buf.release();
+    norm_buf.release();
+}
+
+void ocl_tvl1flow::centeredGradient(const oclMat &src, oclMat &dx, oclMat &dy)
+{
+    Context  *clCxt = src.clCxt;
+    size_t localThreads[3] = {32, 8, 1};
+    size_t globalThreads[3] = {src.cols, src.rows, 1};
+
+    int srcElementSize = src.elemSize();
+    int src_step = src.step/srcElementSize;
+
+    int dElememntSize = dx.elemSize();
+    int dx_step = dx.step/dElememntSize;
+
+    String kernelName = "centeredGradientKernel";
+    vector< pair<size_t, const void *> > args;
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&src.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&src.cols));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&src.rows));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&src_step));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&dx.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&dy.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&dx_step));
+    openCLExecuteKernel(clCxt, &tvl1flow, kernelName, globalThreads, localThreads, args, -1, -1);
+
+}
+
+void ocl_tvl1flow::estimateDualVariables(oclMat &u1, oclMat &u2, oclMat &p11, oclMat &p12, oclMat &p21, oclMat &p22, float taut)
+{
+    Context *clCxt = u1.clCxt;
+
+    size_t localThread[] = {32, 8, 1};
+    size_t globalThread[] =
+    {
+        u1.cols,
+        u1.rows,
+        1
+    };
+
+    int u1_element_size = u1.elemSize();
+    int u1_step = u1.step/u1_element_size;
+
+    int u2_element_size = u2.elemSize();
+    int u2_step = u2.step/u2_element_size;
+
+    int p11_element_size = p11.elemSize();
+    int p11_step = p11.step/p11_element_size;
+
+    int u1_offset_y = u1.offset/u1.step;
+    int u1_offset_x = u1.offset%u1.step;
+    u1_offset_x = u1_offset_x/u1.elemSize();
+
+    int u2_offset_y = u2.offset/u2.step;
+    int u2_offset_x = u2.offset%u2.step;
+    u2_offset_x = u2_offset_x/u2.elemSize();
+
+    String kernelName = "estimateDualVariablesKernel";
+    vector< pair<size_t, const void *> > args;
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&u1.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1.cols));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1.rows));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1_step));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&u2.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&p11.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&p11_step));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&p12.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&p21.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&p22.data));
+    args.push_back( make_pair( sizeof(cl_float), (void*)&taut));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2_step));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1_offset_x));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1_offset_y));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2_offset_x));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2_offset_y));
+
+    openCLExecuteKernel(clCxt, &tvl1flow, kernelName, globalThread, localThread, args, -1, -1);
+}
+
+void ocl_tvl1flow::estimateU(oclMat &I1wx, oclMat &I1wy, oclMat &grad,
+    oclMat &rho_c, oclMat &p11, oclMat &p12,
+    oclMat &p21, oclMat &p22, oclMat &u1,
+    oclMat &u2, oclMat &error, float l_t, float theta)
+{
+    Context* clCxt = I1wx.clCxt;
+
+    size_t localThread[] = {32, 8, 1};
+    size_t globalThread[] =
+    {
+        I1wx.cols,
+        I1wx.rows,
+        1
+    };
+
+    int I1wx_element_size = I1wx.elemSize();
+    int I1wx_step = I1wx.step/I1wx_element_size;
+
+    int u1_element_size = u1.elemSize();
+    int u1_step = u1.step/u1_element_size;
+
+    int u2_element_size = u2.elemSize();
+    int u2_step = u2.step/u2_element_size;
+
+    int u1_offset_y = u1.offset/u1.step;
+    int u1_offset_x = u1.offset%u1.step;
+    u1_offset_x = u1_offset_x/u1.elemSize();
+
+    int u2_offset_y = u2.offset/u2.step;
+    int u2_offset_x = u2.offset%u2.step;
+    u2_offset_x = u2_offset_x/u2.elemSize();
+
+    String kernelName = "estimateUKernel";
+    vector< pair<size_t, const void *> > args;
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I1wx.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&I1wx.cols));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&I1wx.rows));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&I1wx_step));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I1wy.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&grad.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&rho_c.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&p11.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&p12.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&p21.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&p22.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&u1.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1_step));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&u2.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&error.data));
+    args.push_back( make_pair( sizeof(cl_float), (void*)&l_t));
+    args.push_back( make_pair( sizeof(cl_float), (void*)&theta));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2_step));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1_offset_x));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1_offset_y));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2_offset_x));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2_offset_y));
+
+    openCLExecuteKernel(clCxt, &tvl1flow, kernelName, globalThread, localThread, args, -1, -1);
+}
+
+void ocl_tvl1flow::warpBackward(const oclMat &I0, const oclMat &I1, oclMat &I1x, oclMat &I1y, oclMat &u1, oclMat &u2, oclMat &I1w, oclMat &I1wx, oclMat &I1wy, oclMat &grad, oclMat &rho)
+{
+    Context* clCxt = I0.clCxt;
+    const bool isImgSupported = support_image2d(clCxt);
+
+    CV_Assert(isImgSupported);
+
+    int u1ElementSize = u1.elemSize();
+    int u1Step = u1.step/u1ElementSize;
+
+    int u2ElementSize = u2.elemSize();
+    int u2Step = u2.step/u2ElementSize;
+
+    int I0ElementSize = I0.elemSize();
+    int I0Step = I0.step/I0ElementSize;
+
+    int I1w_element_size = I1w.elemSize();
+    int I1w_step = I1w.step/I1w_element_size;
+
+    int u1_offset_y = u1.offset/u1.step;
+    int u1_offset_x = u1.offset%u1.step;
+    u1_offset_x = u1_offset_x/u1.elemSize();
+
+    int u2_offset_y = u2.offset/u2.step;
+    int u2_offset_x = u2.offset%u2.step;
+    u2_offset_x = u2_offset_x/u2.elemSize();
+
+    size_t localThread[] = {32, 8, 1};
+    size_t globalThread[] =
+    {
+        I0.cols,
+        I0.rows,
+        1
+    };
+
+    cl_mem I1_tex;
+    cl_mem I1x_tex;
+    cl_mem I1y_tex;
+    I1_tex = bindTexture(I1);
+    I1x_tex = bindTexture(I1x);
+    I1y_tex = bindTexture(I1y);
+
+    String kernelName = "warpBackwardKernel";
+    vector< pair<size_t, const void *> > args;
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I0.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&I0Step));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&I0.cols));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&I0.rows));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I1_tex));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I1x_tex));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I1y_tex));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&u1.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1Step));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&u2.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I1w.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I1wx.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&I1wy.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&grad.data));
+    args.push_back( make_pair( sizeof(cl_mem), (void*)&rho.data));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&I1w_step));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2Step));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1_offset_x));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u1_offset_y));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2_offset_x));
+    args.push_back( make_pair( sizeof(cl_int), (void*)&u2_offset_y));
+
+    openCLExecuteKernel(clCxt, &tvl1flow, kernelName, globalThread, localThread, args, -1, -1);
+}

modules/ocl/test/interpolation.hpp

@@ -1,120 +0,0 @@
-/*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.
-//
-//
-//                        Intel License Agreement
-//                For Open Source Computer Vision Library
-//
-// Copyright (C) 2000, Intel Corporation, 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 Intel Corporation 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*/
-
-#ifndef __OPENCV_TEST_INTERPOLATION_HPP__
-#define __OPENCV_TEST_INTERPOLATION_HPP__
-
-template <typename T> T readVal(const cv::Mat &src, int y, int x, int c, int border_type, cv::Scalar borderVal = cv::Scalar())
-{
-    if (border_type == cv::BORDER_CONSTANT)
-        return (y >= 0 && y < src.rows && x >= 0 && x < src.cols) ? src.at<T>(y, x * src.channels() + c) : cv::saturate_cast<T>(borderVal.val[c]);
-
-    return src.at<T>(cv::borderInterpolate(y, src.rows, border_type), cv::borderInterpolate(x, src.cols, border_type) * src.channels() + c);
-}
-
-template <typename T> struct NearestInterpolator
-{
-    static T getValue(const cv::Mat &src, float y, float x, int c, int border_type, cv::Scalar borderVal = cv::Scalar())
-    {
-        return readVal<T>(src, cvFloor(y), cvFloor(x), c, border_type, borderVal);
-    }
-};
-
-template <typename T> struct LinearInterpolator
-{
-    static T getValue(const cv::Mat &src, float y, float x, int c, int border_type, cv::Scalar borderVal = cv::Scalar())
-    {
-        x -= 0.5f;
-        y -= 0.5f;
-
-        int x1 = cvFloor(x);
-        int y1 = cvFloor(y);
-        int x2 = x1 + 1;
-        int y2 = y1 + 1;
-
-        float res = 0;
-
-        res += readVal<T>(src, y1, x1, c, border_type, borderVal) * ((x2 - x) * (y2 - y));
-        res += readVal<T>(src, y1, x2, c, border_type, borderVal) * ((x - x1) * (y2 - y));
-        res += readVal<T>(src, y2, x1, c, border_type, borderVal) * ((x2 - x) * (y - y1));
-        res += readVal<T>(src, y2, x2, c, border_type, borderVal) * ((x - x1) * (y - y1));
-
-        return cv::saturate_cast<T>(res);
-    }
-};
-
-template <typename T> struct CubicInterpolator
-{
-    static float getValue(float p[4], float x)
-    {
-        return p[1] + 0.5 * x * (p[2] - p[0] + x * (2.0 * p[0] - 5.0 * p[1] + 4.0 * p[2] - p[3] + x * (3.0 * (p[1] - p[2]) + p[3] - p[0])));
-    }
-
-    static float getValue(float p[4][4], float x, float y)
-    {
-        float arr[4];
-
-        arr[0] = getValue(p[0], x);
-        arr[1] = getValue(p[1], x);
-        arr[2] = getValue(p[2], x);
-        arr[3] = getValue(p[3], x);
-
-        return getValue(arr, y);
-    }
-
-    static T getValue(const cv::Mat &src, float y, float x, int c, int border_type, cv::Scalar borderVal = cv::Scalar())
-    {
-        int ix = cvRound(x);
-        int iy = cvRound(y);
-
-        float vals[4][4] =
-        {
-            {readVal<T>(src, iy - 2, ix - 2, c, border_type, borderVal), readVal<T>(src, iy - 2, ix - 1, c, border_type, borderVal), readVal<T>(src, iy - 2, ix, c, border_type, borderVal), readVal<T>(src, iy - 2, ix + 1, c, border_type, borderVal)},
-            {readVal<T>(src, iy - 1, ix - 2, c, border_type, borderVal), readVal<T>(src, iy - 1, ix - 1, c, border_type, borderVal), readVal<T>(src, iy - 1, ix, c, border_type, borderVal), readVal<T>(src, iy - 1, ix + 1, c, border_type, borderVal)},
-            {readVal<T>(src, iy    , ix - 2, c, border_type, borderVal), readVal<T>(src, iy    , ix - 1, c, border_type, borderVal), readVal<T>(src, iy    , ix, c, border_type, borderVal), readVal<T>(src, iy    , ix + 1, c, border_type, borderVal)},
-            {readVal<T>(src, iy + 1, ix - 2, c, border_type, borderVal), readVal<T>(src, iy + 1, ix - 1, c, border_type, borderVal), readVal<T>(src, iy + 1, ix, c, border_type, borderVal), readVal<T>(src, iy + 1, ix + 1, c, border_type, borderVal)},
-        };
-
-        return cv::saturate_cast<T>(getValue(vals, (x - ix + 2.0) / 4.0, (y - iy + 2.0) / 4.0));
-    }
-};
-
-#endif // __OPENCV_TEST_INTERPOLATION_HPP__

modules/ocl/test/precomp.hpp

@@ -68,9 +68,7 @@
 #include "opencv2/ocl.hpp"
 
 #include "utility.hpp"
-#include "interpolation.hpp"
 
 #include "opencv2/core/private.hpp"
 
 #endif
-

modules/ocl/test/test_arithm.cpp

@@ -12,6 +12,7 @@
 //
 // Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
 // Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
 // Third party copyrights are property of their respective owners.
 //
 // @Authors
@@ -21,6 +22,7 @@
 //    Jiang Liyuan,jlyuan001.good@163.com
 //    Rock Li, Rock.Li@amd.com
 //    Zailong Wu, bullet@yeah.net
+//    Yao Wang, bitwangyaoyao@gmail.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -87,14 +89,13 @@ PARAM_TEST_CASE(ArithmTestBase, MatType, bool)
     int maskx;
     int masky;
 
-
     //src mat with roi
     cv::Mat mat1_roi;
     cv::Mat mat2_roi;
     cv::Mat mask_roi;
     cv::Mat dst_roi;
     cv::Mat dst1_roi; //bak
-    //std::vector<cv::ocl::Info> oclinfo;
+
     //ocl dst mat for testing
     cv::ocl::oclMat gdst_whole;
     cv::ocl::oclMat gdst1_whole; //bak
@@ -125,10 +126,6 @@ PARAM_TEST_CASE(ArithmTestBase, MatType, bool)
 
         val = cv::Scalar(rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0));
 
-        //int devnums = getDevice(oclinfo, OPENCV_DEFAULT_OPENCL_DEVICE);
-        //CV_Assert(devnums > 0);
-        ////if you want to use undefault device, set it here
-        ////setDevice(oclinfo[0]);
     }
 
     void random_roi()
@@ -175,14 +172,22 @@ PARAM_TEST_CASE(ArithmTestBase, MatType, bool)
         gmask = mask_roi; //end
     }
 
+    void Near(double threshold = 0.)
+    {
+        EXPECT_MAT_NEAR(dst, Mat(gdst_whole), threshold);
+    }
+
+    void Near1(double threshold = 0.)
+    {     
+        EXPECT_MAT_NEAR(dst1, Mat(gdst1_whole), threshold);
+    }
+
 };
 ////////////////////////////////lut/////////////////////////////////////////////////
-
 struct Lut : ArithmTestBase {};
 #define VARNAME(A) string(#A);
 
 
-
 TEST_P(Lut, Mat)
 {
 
@@ -203,20 +208,12 @@ TEST_P(Lut, Mat)
 
         cv::LUT(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::LUT(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download (cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0, s);
+        Near(0);
     }
 }
 
 
-
-
 ////////////////////////////////exp/////////////////////////////////////////////////
-
 struct Exp : ArithmTestBase {};
 
 TEST_P(Exp, Mat)
@@ -227,20 +224,12 @@ TEST_P(Exp, Mat)
 
         cv::exp(mat1_roi, dst_roi);
         cv::ocl::exp(gmat1, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 2, s);
-
+        Near(2);
     }
 }
 
 
 ////////////////////////////////log/////////////////////////////////////////////////
-
 struct Log : ArithmTestBase {};
 
 TEST_P(Log, Mat)
@@ -249,24 +238,14 @@ TEST_P(Log, Mat)
     {
         random_roi();
 
-
         cv::log(mat1_roi, dst_roi);
         cv::ocl::log(gmat1, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1, s);
-
+        Near(1);
     }
 }
 
 
-
-
 ////////////////////////////////add/////////////////////////////////////////////////
-
 struct Add : ArithmTestBase {};
 
 TEST_P(Add, Mat)
@@ -277,12 +256,7 @@ TEST_P(Add, Mat)
 
         cv::add(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::add(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, s);
+        Near(0);
     }
 }
 
@@ -294,14 +268,10 @@ TEST_P(Add, Mat_Mask)
 
         cv::add(mat1_roi, mat2_roi, dst_roi, mask_roi);
         cv::ocl::add(gmat1, gmat2, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, s);
+        Near(0);
     }
 }
+
 TEST_P(Add, Scalar)
 {
     for(int j = 0; j < LOOP_TIMES; j++)
@@ -310,12 +280,7 @@ TEST_P(Add, Scalar)
 
         cv::add(mat1_roi, val, dst_roi);
         cv::ocl::add(gmat1, val, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, s);
+        Near(1e-5);
     }
 }
 
@@ -327,12 +292,7 @@ TEST_P(Add, Scalar_Mask)
 
         cv::add(mat1_roi, val, dst_roi, mask_roi);
         cv::ocl::add(gmat1, val, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, s);
+        Near(1e-5);
     }
 }
 
@@ -349,12 +309,7 @@ TEST_P(Sub, Mat)
 
         cv::subtract(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::subtract(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, s);
+        Near(0);
     }
 }
 
@@ -366,14 +321,10 @@ TEST_P(Sub, Mat_Mask)
 
         cv::subtract(mat1_roi, mat2_roi, dst_roi, mask_roi);
         cv::ocl::subtract(gmat1, gmat2, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, s);
+        Near(0);
     }
 }
+
 TEST_P(Sub, Scalar)
 {
     for(int j = 0; j < LOOP_TIMES; j++)
@@ -382,12 +333,7 @@ TEST_P(Sub, Scalar)
 
         cv::subtract(mat1_roi, val, dst_roi);
         cv::ocl::subtract(gmat1, val, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, s);
+        Near(1e-5);
     }
 }
 
@@ -399,12 +345,7 @@ TEST_P(Sub, Scalar_Mask)
 
         cv::subtract(mat1_roi, val, dst_roi, mask_roi);
         cv::ocl::subtract(gmat1, val, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, s);
+        Near(1e-5);
     }
 }
 
@@ -421,12 +362,7 @@ TEST_P(Mul, Mat)
 
         cv::multiply(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::multiply(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char s[1024];
-        sprintf(s, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, s);
+        Near(0);
     }
 }
 
@@ -441,12 +377,7 @@ TEST_P(Mul, Mat_Scalar)
 
         cv::multiply(mat1_roi, mat2_roi, dst_roi, s);
         cv::ocl::multiply(gmat1, gmat2, gdst, s);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.001, sss);
+        Near(.001);
     }
 }
 
@@ -462,13 +393,7 @@ TEST_P(Div, Mat)
 
         cv::divide(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::divide(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1, sss);
+        Near(1);
     }
 }
 
@@ -483,13 +408,7 @@ TEST_P(Div, Mat_Scalar)
 
         cv::divide(mat1_roi, mat2_roi, dst_roi, s);
         cv::ocl::divide(gmat1, gmat2, gdst, s);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.001, sss);
+        Near(.001);
     }
 }
 
@@ -504,13 +423,7 @@ TEST_P(Absdiff, Mat)
 
         cv::absdiff(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::absdiff(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0, sss);
+        Near(0);
     }
 }
 
@@ -522,13 +435,7 @@ TEST_P(Absdiff, Mat_Scalar)
 
         cv::absdiff(mat1_roi, val, dst_roi);
         cv::ocl::absdiff(gmat1, val, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
@@ -544,16 +451,8 @@ TEST_P(CartToPolar, angleInDegree)
 
         cv::cartToPolar(mat1_roi, mat2_roi, dst_roi, dst1_roi, 1);
         cv::ocl::cartToPolar(gmat1, gmat2, gdst, gdst1, 1);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-
-        cv::Mat cpu_dst1;
-        gdst1_whole.download(cpu_dst1);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.5, sss);
-        EXPECT_MAT_NEAR(dst1, cpu_dst1, 0.5, sss);
+        Near(.5);
+        Near1(.5);
     }
 }
 
@@ -565,22 +464,12 @@ TEST_P(CartToPolar, angleInRadians)
 
         cv::cartToPolar(mat1_roi, mat2_roi, dst_roi, dst1_roi, 0);
         cv::ocl::cartToPolar(gmat1, gmat2, gdst, gdst1, 0);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-
-        cv::Mat cpu_dst1;
-        gdst1_whole.download(cpu_dst1);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.5, sss);
-        EXPECT_MAT_NEAR(dst1, cpu_dst1, 0.5, sss);
+        Near(.5);
+        Near1(.5);
     }
 }
 
 
-
-
 struct PolarToCart : ArithmTestBase {};
 
 TEST_P(PolarToCart, angleInDegree)
@@ -591,17 +480,8 @@ TEST_P(PolarToCart, angleInDegree)
 
         cv::polarToCart(mat1_roi, mat2_roi, dst_roi, dst1_roi, 1);
         cv::ocl::polarToCart(gmat1, gmat2, gdst, gdst1, 1);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-
-        cv::Mat cpu_dst1;
-        gdst1_whole.download(cpu_dst1);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.5, sss);
-        EXPECT_MAT_NEAR(dst1, cpu_dst1, 0.5, sss);
+        Near(.5);
+        Near1(.5);
     }
 }
 
@@ -613,17 +493,8 @@ TEST_P(PolarToCart, angleInRadians)
 
         cv::polarToCart(mat1_roi, mat2_roi, dst_roi, dst1_roi, 0);
         cv::ocl::polarToCart(gmat1, gmat2, gdst, gdst1, 0);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-
-        cv::Mat cpu_dst1;
-        gdst1_whole.download(cpu_dst1);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.5, sss);
-        EXPECT_MAT_NEAR(dst1, cpu_dst1, 0.5, sss);
+        Near(.5);
+        Near1(.5);
     }
 }
 
@@ -640,19 +511,11 @@ TEST_P(Magnitude, Mat)
 
         cv::magnitude(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::magnitude(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
 
-
-
 struct Transpose : ArithmTestBase {};
 
 TEST_P(Transpose, Mat)
@@ -663,20 +526,11 @@ TEST_P(Transpose, Mat)
 
         cv::transpose(mat1_roi, dst_roi);
         cv::ocl::transpose(gmat1, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
 
-
-
-
 struct Flip : ArithmTestBase {};
 
 TEST_P(Flip, X)
@@ -687,13 +541,7 @@ TEST_P(Flip, X)
 
         cv::flip(mat1_roi, dst_roi, 0);
         cv::ocl::flip(gmat1, gdst, 0);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
@@ -705,13 +553,7 @@ TEST_P(Flip, Y)
 
         cv::flip(mat1_roi, dst_roi, 1);
         cv::ocl::flip(gmat1, gdst, 1);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
@@ -723,18 +565,11 @@ TEST_P(Flip, BOTH)
 
         cv::flip(mat1_roi, dst_roi, -1);
         cv::ocl::flip(gmat1, gdst, -1);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
 
-
 struct MinMax : ArithmTestBase {};
 
 TEST_P(MinMax, MAT)
@@ -765,12 +600,8 @@ TEST_P(MinMax, MAT)
         double minVal_, maxVal_;
         cv::ocl::minMax(gmat1, &minVal_, &maxVal_);
 
-        //check results
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_DOUBLE_EQ(minVal_, minVal) << sss;
-        EXPECT_DOUBLE_EQ(maxVal_, maxVal) << sss;
+        EXPECT_DOUBLE_EQ(minVal_, minVal);
+        EXPECT_DOUBLE_EQ(maxVal_, maxVal);
     }
 }
 
@@ -803,12 +634,8 @@ TEST_P(MinMax, MASK)
         double minVal_, maxVal_;
         cv::ocl::minMax(gmat1, &minVal_, &maxVal_, gmask);
 
-        //check results
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_DOUBLE_EQ(minVal_, minVal) << sss;
-        EXPECT_DOUBLE_EQ(maxVal_, maxVal) << sss;
+        EXPECT_DOUBLE_EQ(minVal_, minVal);
+        EXPECT_DOUBLE_EQ(maxVal_, maxVal);
     }
 }
 
@@ -919,17 +746,13 @@ TEST_P(MinMaxLoc, MAT)
             error1 = ::abs(mat1_roi.at<double>(maxLoc_) - mat1_roi.at<double>(maxLoc));
         }
 
-        //check results
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_DOUBLE_EQ(minVal_, minVal) << sss;
-        EXPECT_DOUBLE_EQ(maxVal_, maxVal) << sss;
-        EXPECT_DOUBLE_EQ(minlocVal_, minlocVal) << sss;
-        EXPECT_DOUBLE_EQ(maxlocVal_, maxlocVal) << sss;
+        EXPECT_DOUBLE_EQ(minVal_, minVal);
+        EXPECT_DOUBLE_EQ(maxVal_, maxVal);
+        EXPECT_DOUBLE_EQ(minlocVal_, minlocVal);
+        EXPECT_DOUBLE_EQ(maxlocVal_, maxlocVal);
 
-        EXPECT_DOUBLE_EQ(error0, 0.0) << sss;
-        EXPECT_DOUBLE_EQ(error1, 0.0) << sss;
+        EXPECT_DOUBLE_EQ(error0, 0.0);
+        EXPECT_DOUBLE_EQ(error1, 0.0);
     }
 }
 
@@ -1040,17 +863,13 @@ TEST_P(MinMaxLoc, MASK)
             error1 = ::abs(mat1_roi.at<double>(maxLoc_) - mat1_roi.at<double>(maxLoc));
         }
 
-        //check results
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_DOUBLE_EQ(minVal_, minVal) << sss;
-        EXPECT_DOUBLE_EQ(maxVal_, maxVal) << sss;
-        EXPECT_DOUBLE_EQ(minlocVal_, minlocVal) << sss;
-        EXPECT_DOUBLE_EQ(maxlocVal_, maxlocVal) << sss;
+        EXPECT_DOUBLE_EQ(minVal_, minVal);
+        EXPECT_DOUBLE_EQ(maxVal_, maxVal);
+        EXPECT_DOUBLE_EQ(minlocVal_, minlocVal);
+        EXPECT_DOUBLE_EQ(maxlocVal_, maxlocVal);
 
-        EXPECT_DOUBLE_EQ(error0, 0.0) << sss;
-        EXPECT_DOUBLE_EQ(error1, 0.0) << sss;
+        EXPECT_DOUBLE_EQ(error0, 0.0);
+        EXPECT_DOUBLE_EQ(error1, 0.0);
     }
 }
 
@@ -1064,14 +883,12 @@ TEST_P(Sum, MAT)
         random_roi();
         Scalar cpures = cv::sum(mat1_roi);
         Scalar gpures = cv::ocl::sum(gmat1);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
 
         //check results
-        EXPECT_NEAR(cpures[0], gpures[0], 0.1) << sss;
-        EXPECT_NEAR(cpures[1], gpures[1], 0.1) << sss;
-        EXPECT_NEAR(cpures[2], gpures[2], 0.1) << sss;
-        EXPECT_NEAR(cpures[3], gpures[3], 0.1) << sss;
+        EXPECT_NEAR(cpures[0], gpures[0], 0.1);
+        EXPECT_NEAR(cpures[1], gpures[1], 0.1);
+        EXPECT_NEAR(cpures[2], gpures[2], 0.1);
+        EXPECT_NEAR(cpures[3], gpures[3], 0.1);
     }
 }
 
@@ -1086,11 +903,7 @@ TEST_P(CountNonZero, MAT)
         int cpures = cv::countNonZero(mat1_roi);
         int gpures = cv::ocl::countNonZero(gmat1);
 
-        //check results
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_DOUBLE_EQ((double)cpures, (double)gpures) << sss;
+        EXPECT_DOUBLE_EQ((double)cpures, (double)gpures);
     }
 }
 
@@ -1112,13 +925,7 @@ TEST_P(Phase, Mat)
             random_roi();
             cv::phase(mat1_roi, mat2_roi, dst_roi, angelInDegrees ? true : false);
             cv::ocl::phase(gmat1, gmat2, gdst, angelInDegrees ? true : false);
-
-            cv::Mat cpu_dst;
-            gdst_whole.download(cpu_dst);
-
-            char sss[1024];
-            sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-            EXPECT_MAT_NEAR(dst, cpu_dst, 1e-2, sss);
+            Near(1e-2);
         }
     }
 }
@@ -1135,13 +942,7 @@ TEST_P(Bitwise_and, Mat)
 
         cv::bitwise_and(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::bitwise_and(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, sss);
+        Near(0);
     }
 }
 
@@ -1153,15 +954,10 @@ TEST_P(Bitwise_and, Mat_Mask)
 
         cv::bitwise_and(mat1_roi, mat2_roi, dst_roi, mask_roi);
         cv::ocl::bitwise_and(gmat1, gmat2, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, sss);
+        Near(0);
     }
 }
+
 TEST_P(Bitwise_and, Scalar)
 {
     for(int j = 0; j < LOOP_TIMES; j++)
@@ -1170,14 +966,7 @@ TEST_P(Bitwise_and, Scalar)
 
         cv::bitwise_and(mat1_roi, val, dst_roi);
         cv::ocl::bitwise_and(gmat1, val, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
-
+        Near(1e-5);
     }
 }
 
@@ -1189,14 +978,7 @@ TEST_P(Bitwise_and, Scalar_Mask)
 
         cv::bitwise_and(mat1_roi, val, dst_roi, mask_roi);
         cv::ocl::bitwise_and(gmat1, val, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char *sss = new char[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
-        delete[] sss;
+        Near(1e-5);
     }
 }
 
@@ -1214,13 +996,7 @@ TEST_P(Bitwise_or, Mat)
 
         cv::bitwise_or(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::bitwise_or(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, sss);
+        Near(0);
     }
 }
 
@@ -1232,15 +1008,10 @@ TEST_P(Bitwise_or, Mat_Mask)
 
         cv::bitwise_or(mat1_roi, mat2_roi, dst_roi, mask_roi);
         cv::ocl::bitwise_or(gmat1, gmat2, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, sss);
+        Near(0);
     }
 }
+
 TEST_P(Bitwise_or, Scalar)
 {
     for(int j = 0; j < LOOP_TIMES; j++)
@@ -1249,13 +1020,7 @@ TEST_P(Bitwise_or, Scalar)
 
         cv::bitwise_or(mat1_roi, val, dst_roi);
         cv::ocl::bitwise_or(gmat1, val, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
@@ -1267,13 +1032,7 @@ TEST_P(Bitwise_or, Scalar_Mask)
 
         cv::bitwise_or(mat1_roi, val, dst_roi, mask_roi);
         cv::ocl::bitwise_or(gmat1, val, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
@@ -1291,13 +1050,7 @@ TEST_P(Bitwise_xor, Mat)
 
         cv::bitwise_xor(mat1_roi, mat2_roi, dst_roi);
         cv::ocl::bitwise_xor(gmat1, gmat2, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, sss);
+        Near(0);
     }
 }
 
@@ -1309,15 +1062,10 @@ TEST_P(Bitwise_xor, Mat_Mask)
 
         cv::bitwise_xor(mat1_roi, mat2_roi, dst_roi, mask_roi);
         cv::ocl::bitwise_xor(gmat1, gmat2, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, sss);
+        Near(0);
     }
 }
+
 TEST_P(Bitwise_xor, Scalar)
 {
     for(int j = 0; j < LOOP_TIMES; j++)
@@ -1326,13 +1074,7 @@ TEST_P(Bitwise_xor, Scalar)
 
         cv::bitwise_xor(mat1_roi, val, dst_roi);
         cv::ocl::bitwise_xor(gmat1, val, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
@@ -1344,13 +1086,7 @@ TEST_P(Bitwise_xor, Scalar_Mask)
 
         cv::bitwise_xor(mat1_roi, val, dst_roi, mask_roi);
         cv::ocl::bitwise_xor(gmat1, val, gdst, gmask);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
@@ -1367,13 +1103,7 @@ TEST_P(Bitwise_not, Mat)
 
         cv::bitwise_not(mat1_roi, dst_roi);
         cv::ocl::bitwise_not(gmat1, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, sss);
+        Near(0);
     }
 }
 
@@ -1390,7 +1120,7 @@ TEST_P(Compare, Mat)
     }
 
     int cmp_codes[] = {CMP_EQ, CMP_GT, CMP_GE, CMP_LT, CMP_LE, CMP_NE};
-    const char *cmp_str[] = {"CMP_EQ", "CMP_GT", "CMP_GE", "CMP_LT", "CMP_LE", "CMP_NE"};
+    //const char *cmp_str[] = {"CMP_EQ", "CMP_GT", "CMP_GE", "CMP_LT", "CMP_LE", "CMP_NE"};
     int cmp_num = sizeof(cmp_codes) / sizeof(int);
 
     for (int i = 0; i < cmp_num; ++i)
@@ -1402,13 +1132,7 @@ TEST_P(Compare, Mat)
 
             cv::compare(mat1_roi, mat2_roi, dst_roi, cmp_codes[i]);
             cv::ocl::compare(gmat1, gmat2, gdst, cmp_codes[i]);
-
-            cv::Mat cpu_dst;
-            gdst_whole.download(cpu_dst);
-            char sss[1024];
-            sprintf(sss, "cmptype=%s, roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", cmp_str[i], roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-            EXPECT_MAT_NEAR(dst, cpu_dst, 0.0, sss);
+            Near(0);
         }
     }
 
@@ -1430,14 +1154,7 @@ TEST_P(Pow, Mat)
         double p = 4.5;
         cv::pow(mat1_roi, p, dst_roi);
         cv::ocl::pow(gmat1, p, gdst);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1, sss);
+        Near(1);
     }
 }
 
@@ -1448,36 +1165,18 @@ TEST_P(MagnitudeSqr, Mat)
 {
     for(int j = 0; j < LOOP_TIMES; j++)
     {
-        // random_roi();
-        // int64 start, end;
-        // start = cv::getTickCount();
+        random_roi();
         for(int i = 0; i < mat1.rows; ++i)
             for(int j = 0; j < mat1.cols; ++j)
             {
                 float val1 = mat1.at<float>(i, j);
                 float val2 = mat2.at<float>(i, j);
-
                 ((float *)(dst.data))[i * dst.step / 4 + j] = val1 * val1 + val2 * val2;
-
-                //        float val1 =((float *)( mat1.data))[(i*mat1.step/8 +j)*2];
-                //
-                //     float val2 =((float *)( mat1.data))[(i*mat1.step/8 +j)*2+ 1 ];
-
-                //  ((float *)(dst.data))[i*dst.step/4 +j]= val1 * val1 +val2 * val2;
             }
-        // end = cv::getTickCount();
 
-
-
-        cv::ocl::oclMat clmat1(mat1), clmat2(mat2), cldst;
-        cv::ocl::magnitudeSqr(clmat1, clmat2, cldst);
-
-        cv::Mat cpu_dst;
-        cldst.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1, sss);
+        cv::ocl::oclMat clmat1(mat1), clmat2(mat2);
+        cv::ocl::magnitudeSqr(clmat1, clmat2, gdst);
+        Near(1);
     }
 }
 
@@ -1498,21 +1197,13 @@ TEST_P(AddWeighted, Mat)
 
         cv::ocl::addWeighted(gmat1, alpha, gmat2, beta, gama, gdst);
 
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d,maskx=%d,masky=%d,src2x=%d,src2y=%d", roicols, roirows, src1x, src1y, dstx, dsty, maskx, masky, src2x, src2y);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
 }
 
 
 
 
-
 //********test****************
 
 INSTANTIATE_TEST_CASE_P(Arithm, Lut, Combine(

modules/ocl/test/test_blend.cpp

@@ -1,3 +1,47 @@
+/*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) 2010-2012, Multicoreware, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// @Authors
+//    Nathan, liujun@multicorewareinc.com
+//
+// 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 oclMaterials 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"
 #include <iomanip>
 
@@ -33,20 +77,14 @@ void blendLinearGold(const cv::Mat &img1, const cv::Mat &img2, const cv::Mat &we
 
 PARAM_TEST_CASE(Blend, cv::Size, MatType/*, UseRoi*/)
 {
-    //std::vector<cv::ocl::Info> oclinfo;
     cv::Size size;
     int type;
     bool useRoi;
 
     virtual void SetUp()
     {
-        //devInfo = GET_PARAM(0);
         size = GET_PARAM(0);
         type = GET_PARAM(1);
-        /*useRoi = GET_PARAM(3);*/
-
-        //int devnums = getDevice(oclinfo, OPENCV_DEFAULT_OPENCL_DEVICE);
-        //CV_Assert(devnums > 0);
     }
 };
 
@@ -59,12 +97,9 @@ TEST_P(Blend, Accuracy)
     cv::Mat weights1 = randomMat(size, CV_32F, 0, 1);
     cv::Mat weights2 = randomMat(size, CV_32F, 0, 1);
 
-    cv::ocl::oclMat gimg1(size, type), gimg2(size, type), gweights1(size, CV_32F), gweights2(size, CV_32F);
-    cv::ocl::oclMat dst(size, type);
-    gimg1.upload(img1);
-    gimg2.upload(img2);
-    gweights1.upload(weights1);
-    gweights2.upload(weights2);
+    cv::ocl::oclMat gimg1(img1), gimg2(img2), gweights1(weights1), gweights2(weights2);
+    cv::ocl::oclMat dst;
+
     cv::ocl::blendLinear(gimg1, gimg2, gweights1, gweights2, dst);
     cv::Mat result;
     cv::Mat result_gold;
@@ -74,10 +109,10 @@ TEST_P(Blend, Accuracy)
     else
         blendLinearGold<float>(img1, img2, weights1, weights2, result_gold);
 
-    EXPECT_MAT_NEAR(result_gold, result, CV_MAT_DEPTH(type) == CV_8U ? 1.f : 1e-5f, 0);
+    EXPECT_MAT_NEAR(result_gold, result, CV_MAT_DEPTH(type) == CV_8U ? 1.f : 1e-5f);
 }
 
-INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Blend, Combine(
+INSTANTIATE_TEST_CASE_P(OCL_ImgProc, Blend, Combine(
                             DIFFERENT_SIZES,
                             testing::Values(MatType(CV_8UC1), MatType(CV_8UC3), MatType(CV_8UC4), MatType(CV_32FC1), MatType(CV_32FC4))
                         ));

modules/ocl/test/test_brute_force_matcher.cpp

@@ -7,12 +7,16 @@
 //  copy or use the software.
 //
 //
-//                        Intel License Agreement
+//                           License Agreement
 //                For Open Source Computer Vision Library
 //
-// Copyright (C) 2010-2012, Multicoreware inc., all rights reserved.
+// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
 // Third party copyrights are property of their respective owners.
 //
+// @Authors
+//    Nathan, liujun@multicorewareinc.com
+//
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
 //
@@ -21,12 +25,12 @@
 //
 //   * 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.
+//     and/or other oclMaterials provided with the distribution.
 //
-//   * The name of Intel Corporation may not be used to endorse or promote products
+//   * 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
+// 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,

modules/ocl/test/test_calib3d.cpp

@@ -129,9 +129,69 @@ TEST_P(StereoMatchBP, Regression)
     bp(d_left, d_right, d_disp);
     d_disp.download(disp);
     disp.convertTo(disp, disp_gold.depth());
-    EXPECT_MAT_NEAR(disp_gold, disp, 0.0, "");
+    EXPECT_MAT_NEAR(disp_gold, disp, 0.0);
 }
 INSTANTIATE_TEST_CASE_P(OCL_Calib3D, StereoMatchBP, testing::Combine(testing::Values(64),
     testing::Values(8),testing::Values(2),testing::Values(25.0f),
     testing::Values(0.1f),testing::Values(15.0f),testing::Values(1.0f)));
+
+//////////////////////////////////////////////////////////////////////////
+//  ConstSpaceBeliefPropagation
+PARAM_TEST_CASE(StereoMatchConstSpaceBP, int, int, int, int, float, float, float, float, int, int)
+{
+    int ndisp_;
+    int iters_;
+    int levels_;
+    int nr_plane_;
+    float max_data_term_;
+    float data_weight_;
+    float max_disc_term_;
+    float disc_single_jump_;
+    int min_disp_th_;
+    int msg_type_;
+
+    virtual void SetUp()
+    {
+        ndisp_          = GET_PARAM(0);
+        iters_          = GET_PARAM(1);
+        levels_         = GET_PARAM(2);
+        nr_plane_ = GET_PARAM(3);
+        max_data_term_  = GET_PARAM(4);
+        data_weight_    = GET_PARAM(5);
+        max_disc_term_     = GET_PARAM(6);
+        disc_single_jump_  = GET_PARAM(7);
+        min_disp_th_ = GET_PARAM(8);
+        msg_type_  = GET_PARAM(9);
+    }
+};
+TEST_P(StereoMatchConstSpaceBP, Regression)
+{
+    Mat left_image  = readImage("csstereobp/aloe-L.png");
+    Mat right_image = readImage("csstereobp/aloe-R.png");
+    Mat disp_gold   = readImage("csstereobp/aloe-disp.png", IMREAD_GRAYSCALE);
+
+    ocl::oclMat d_left, d_right;
+    ocl::oclMat d_disp;
+
+    Mat  disp;
+    ASSERT_FALSE(left_image.empty());
+    ASSERT_FALSE(right_image.empty());
+    ASSERT_FALSE(disp_gold.empty());
+
+    d_left.upload(left_image);
+    d_right.upload(right_image);
+
+    ocl::StereoConstantSpaceBP bp(ndisp_, iters_, levels_, nr_plane_, max_data_term_, data_weight_,
+        max_disc_term_, disc_single_jump_, 0, CV_32F);
+    bp(d_left, d_right, d_disp);
+    d_disp.download(disp);
+    disp.convertTo(disp, disp_gold.depth());
+
+    EXPECT_MAT_SIMILAR(disp_gold, disp, 1e-4);
+    //EXPECT_MAT_NEAR(disp_gold, disp, 1.0, "");
+}
+INSTANTIATE_TEST_CASE_P(OCL_Calib3D, StereoMatchConstSpaceBP, testing::Combine(testing::Values(128),
+    testing::Values(16),testing::Values(4), testing::Values(4), testing::Values(30.0f),
+    testing::Values(1.0f),testing::Values(160.0f),
+    testing::Values(10.0f), testing::Values(0), testing::Values(CV_32F)));
 #endif // HAVE_OPENCL

modules/ocl/test/test_color.cpp

@@ -103,7 +103,7 @@ PARAM_TEST_CASE(CvtColor, cv::Size, MatDepth)
     cv::cvtColor(src, dst_gold, CVTCODE(name));\
     cv::Mat dst_mat;\
     dst.download(dst_mat);\
-    EXPECT_MAT_NEAR(dst_gold, dst_mat, 1e-5, "");\
+    EXPECT_MAT_NEAR(dst_gold, dst_mat, 1e-5);\
 }
 
 //add new ones here using macro
@@ -144,7 +144,7 @@ TEST_P(CvtColor_Gray2RGB, Accuracy)
     cv::cvtColor(src, dst_gold, code);
     cv::Mat dst_mat;
     dst.download(dst_mat);
-    EXPECT_MAT_NEAR(dst_gold, dst_mat, 1e-5, "");
+    EXPECT_MAT_NEAR(dst_gold, dst_mat, 1e-5);
 }
 
 
@@ -174,7 +174,7 @@ TEST_P(CvtColor_YUV420, Accuracy)
     cv::Mat dst_mat;
     dst.download(dst_mat);
     MAT_DIFF(dst_mat, dst_gold);
-    EXPECT_MAT_NEAR(dst_gold, dst_mat, 1e-5, "");
+    EXPECT_MAT_NEAR(dst_gold, dst_mat, 1e-5);
 }
 
 INSTANTIATE_TEST_CASE_P(OCL_ImgProc, CvtColor, testing::Combine(

modules/ocl/test/test_columnsum.cpp

@@ -47,27 +47,16 @@
 #include "precomp.hpp"
 #include <iomanip>
 
-///////////////////////////////////////////////////////////////////////////////
-/// ColumnSum
-
 #ifdef HAVE_OPENCL
 
-////////////////////////////////////////////////////////////////////////
-// ColumnSum
-
-PARAM_TEST_CASE(ColumnSum, cv::Size, bool )
+PARAM_TEST_CASE(ColumnSum, cv::Size)
 {
     cv::Size size;
     cv::Mat src;
-    bool useRoi;
-    //std::vector<cv::ocl::Info> oclinfo;
 
     virtual void SetUp()
     {
         size = GET_PARAM(0);
-        useRoi = GET_PARAM(1);
-        //int devnums = getDevice(oclinfo, OPENCV_DEFAULT_OPENCL_DEVICE);
-        //CV_Assert(devnums > 0);
     }
 };
 
@@ -99,8 +88,7 @@ TEST_P(ColumnSum, Accuracy)
     }
 }
 
-INSTANTIATE_TEST_CASE_P(GPU_ImgProc, ColumnSum, testing::Combine(
-                            DIFFERENT_SIZES, testing::Values(Inverse(false), Inverse(true))));
+INSTANTIATE_TEST_CASE_P(OCL_ImgProc, ColumnSum, DIFFERENT_SIZES);
 
 
 #endif

modules/ocl/test/test_fft.cpp

@@ -68,7 +68,7 @@ TEST_P(Dft, C2C)
 
     cv::dft(a, b_gold, dft_flags);
     cv::ocl::dft(cv::ocl::oclMat(a), d_b, a.size(), dft_flags);
-    EXPECT_MAT_NEAR(b_gold, cv::Mat(d_b), a.size().area() * 1e-4, "");
+    EXPECT_MAT_NEAR(b_gold, cv::Mat(d_b), a.size().area() * 1e-4);
 }
 
 TEST_P(Dft, R2C)
@@ -81,11 +81,11 @@ TEST_P(Dft, R2C)
     cv::dft(a, b_gold, cv::DFT_COMPLEX_OUTPUT | dft_flags);
 
     b_gold_roi = b_gold(cv::Rect(0, 0, d_b.cols, d_b.rows));
-    EXPECT_MAT_NEAR(b_gold_roi, cv::Mat(d_b), a.size().area() * 1e-4, "");
+    EXPECT_MAT_NEAR(b_gold_roi, cv::Mat(d_b), a.size().area() * 1e-4);
 
     cv::Mat c_gold;
     cv::dft(b_gold, c_gold, cv::DFT_INVERSE | cv::DFT_REAL_OUTPUT | cv::DFT_SCALE);
-    EXPECT_MAT_NEAR(b_gold_roi, cv::Mat(d_b), a.size().area() * 1e-4, "");
+    EXPECT_MAT_NEAR(b_gold_roi, cv::Mat(d_b), a.size().area() * 1e-4);
 }
 
 TEST_P(Dft, R2CthenC2R)
@@ -95,7 +95,7 @@ TEST_P(Dft, R2CthenC2R)
     cv::ocl::oclMat d_b, d_c;
     cv::ocl::dft(cv::ocl::oclMat(a), d_b, a.size(), 0);
     cv::ocl::dft(d_b, d_c, a.size(), cv::DFT_SCALE | cv::DFT_INVERSE | cv::DFT_REAL_OUTPUT);
-    EXPECT_MAT_NEAR(a, d_c, a.size().area() * 1e-4, "");
+    EXPECT_MAT_NEAR(a, d_c, a.size().area() * 1e-4);
 }
 
 

modules/ocl/test/test_filters.cpp

@@ -12,6 +12,7 @@
 //
 // Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
 // Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
 // Third party copyrights are property of their respective owners.
 //
 // @Authors
@@ -19,6 +20,7 @@
 //    Jia Haipeng, jiahaipeng95@gmail.com
 //    Zero Lin, Zero.Lin@amd.com
 //    Zhang Ying, zhangying913@gmail.com
+//    Yao Wang, bitwangyaoyao@gmail.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -55,191 +57,93 @@ using namespace testing;
 using namespace std;
 
 
-PARAM_TEST_CASE(FilterTestBase, MatType, bool)
+PARAM_TEST_CASE(FilterTestBase, 
+                MatType, 
+                cv::Size, // kernel size
+                cv::Size, // dx,dy
+                int       // border type, or iteration
+                )
 {
-    int type;
-    cv::Scalar val;
-
     //src mat
     cv::Mat mat1;
-    cv::Mat mat2;
-    cv::Mat mask;
     cv::Mat dst;
-    cv::Mat dst1; //bak, for two outputs
 
     // set up roi
     int roicols;
     int roirows;
     int src1x;
     int src1y;
-    int src2x;
-    int src2y;
     int dstx;
     int dsty;
-    int maskx;
-    int masky;
 
     //src mat with roi
     cv::Mat mat1_roi;
-    cv::Mat mat2_roi;
-    cv::Mat mask_roi;
     cv::Mat dst_roi;
-    cv::Mat dst1_roi; //bak
-    //std::vector<cv::ocl::Info> oclinfo;
+
     //ocl dst mat for testing
     cv::ocl::oclMat gdst_whole;
-    cv::ocl::oclMat gdst1_whole; //bak
 
     //ocl mat with roi
     cv::ocl::oclMat gmat1;
-    cv::ocl::oclMat gmat2;
     cv::ocl::oclMat gdst;
-    cv::ocl::oclMat gdst1;   //bak
-    cv::ocl::oclMat gmask;
-
-    virtual void SetUp()
-    {
-        type = GET_PARAM(0);
-
-        cv::RNG &rng = TS::ptr()->get_rng();
-        cv::Size size(MWIDTH, MHEIGHT);
-
-        mat1 = randomMat(rng, size, type, 5, 16, false);
-        mat2 = randomMat(rng, size, type, 5, 16, false);
-        dst  = randomMat(rng, size, type, 5, 16, false);
-        dst1  = randomMat(rng, size, type, 5, 16, false);
-        mask = randomMat(rng, size, CV_8UC1, 0, 2,  false);
-
-        cv::threshold(mask, mask, 0.5, 255., CV_8UC1);
-
-        val = cv::Scalar(rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0));
-    }
 
     void random_roi()
     {
 #ifdef RANDOMROI
         //randomize ROI
         cv::RNG &rng = TS::ptr()->get_rng();
-        roicols = rng.uniform(1, mat1.cols);
-        roirows = rng.uniform(1, mat1.rows);
+        roicols = rng.uniform(2, mat1.cols);
+        roirows = rng.uniform(2, mat1.rows);
         src1x   = rng.uniform(0, mat1.cols - roicols);
         src1y   = rng.uniform(0, mat1.rows - roirows);
-        src2x   = rng.uniform(0, mat2.cols - roicols);
-        src2y   = rng.uniform(0, mat2.rows - roirows);
         dstx    = rng.uniform(0, dst.cols  - roicols);
         dsty    = rng.uniform(0, dst.rows  - roirows);
-        maskx   = rng.uniform(0, mask.cols - roicols);
-        masky   = rng.uniform(0, mask.rows - roirows);
 #else
         roicols = mat1.cols;
         roirows = mat1.rows;
         src1x = 0;
         src1y = 0;
-        src2x = 0;
-        src2y = 0;
         dstx = 0;
         dsty = 0;
-        maskx = 0;
-        masky = 0;
 #endif
+
         mat1_roi = mat1(Rect(src1x, src1y, roicols, roirows));
-        mat2_roi = mat2(Rect(src2x, src2y, roicols, roirows));
-        mask_roi = mask(Rect(maskx, masky, roicols, roirows));
         dst_roi  = dst(Rect(dstx, dsty, roicols, roirows));
-        dst1_roi = dst1(Rect(dstx, dsty, roicols, roirows));
 
         gdst_whole = dst;
         gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
 
-        gdst1_whole = dst1;
-        gdst1 = gdst1_whole(Rect(dstx, dsty, roicols, roirows));
-
         gmat1 = mat1_roi;
-        gmat2 = mat2_roi;
-        gmask = mask_roi;
     }
 
+    void Init(int mat_type)
+    {
+        cv::Size size(MWIDTH, MHEIGHT);
+        mat1 = randomMat(size, mat_type, 5, 16);
+        dst  = randomMat(size, mat_type, 5, 16);
+    }
+
+    void Near(double threshold)
+    {
+        EXPECT_MAT_NEAR(dst, Mat(gdst_whole), threshold);
+    }
 };
 
 /////////////////////////////////////////////////////////////////////////////////////////////////
 // blur
-
-PARAM_TEST_CASE(Blur, MatType, cv::Size, int)
+struct Blur : FilterTestBase
 {
     int type;
     cv::Size ksize;
     int bordertype;
 
-    //src mat
-    cv::Mat mat1;
-    cv::Mat dst;
-
-    // set up roi
-    int roicols;
-    int roirows;
-    int src1x;
-    int src1y;
-    int dstx;
-    int dsty;
-
-    //src mat with roi
-    cv::Mat mat1_roi;
-    cv::Mat dst_roi;
-    //std::vector<cv::ocl::Info> oclinfo;
-    //ocl dst mat for testing
-    cv::ocl::oclMat gdst_whole;
-
-    //ocl mat with roi
-    cv::ocl::oclMat gmat1;
-    cv::ocl::oclMat gdst;
-
     virtual void SetUp()
     {
         type = GET_PARAM(0);
         ksize = GET_PARAM(1);
-        bordertype = GET_PARAM(2);
-
-        cv::RNG &rng = TS::ptr()->get_rng();
-        cv::Size size(MWIDTH, MHEIGHT);
-
-        mat1 = randomMat(rng, size, type, 5, 16, false);
-        dst  = randomMat(rng, size, type, 5, 16, false);
-        //int devnums = getDevice(oclinfo, OPENCV_DEFAULT_OPENCL_DEVICE);
-        //CV_Assert(devnums > 0);
-        ////if you want to use undefault device, set it here
-        ////setDevice(oclinfo[0]);
-    }
-
-    void random_roi()
-    {
-#ifdef RANDOMROI
-        //randomize ROI
-        cv::RNG &rng = TS::ptr()->get_rng();
-        roicols = rng.uniform(2, mat1.cols);
-        roirows = rng.uniform(2, mat1.rows);
-        src1x   = rng.uniform(0, mat1.cols - roicols);
-        src1y   = rng.uniform(0, mat1.rows - roirows);
-        dstx    = rng.uniform(0, dst.cols  - roicols);
-        dsty    = rng.uniform(0, dst.rows  - roirows);
-#else
-        roicols = mat1.cols;
-        roirows = mat1.rows;
-        src1x = 0;
-        src1y = 0;
-        dstx = 0;
-        dsty = 0;
-#endif
-
-        mat1_roi = mat1(Rect(src1x, src1y, roicols, roirows));
-        dst_roi  = dst(Rect(dstx, dsty, roicols, roirows));
-
-        gdst_whole = dst;
-        gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
-
-
-        gmat1 = mat1_roi;
+        bordertype = GET_PARAM(3);
+        Init(type);
     }
-
 };
 
 TEST_P(Blur, Mat)
@@ -247,116 +151,36 @@ TEST_P(Blur, Mat)
     for(int j = 0; j < LOOP_TIMES; j++)
     {
         random_roi();
-
         cv::blur(mat1_roi, dst_roi, ksize, Point(-1, -1), bordertype);
         cv::ocl::blur(gmat1, gdst, ksize, Point(-1, -1), bordertype);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d", roicols, roirows, src1x, src1y, dstx, dsty);
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1.0, sss);
+        Near(1.0);
     }
-
 }
 
 
-
 /////////////////////////////////////////////////////////////////////////////////////////////////
 //Laplacian
-
-PARAM_TEST_CASE(LaplacianTestBase, MatType, int)
+struct Laplacian : FilterTestBase
 {
     int type;
-    int ksize;
-
-    //src mat
-    cv::Mat mat;
-    cv::Mat dst;
-
-    // set up roi
-    int roicols;
-    int roirows;
-    int srcx;
-    int srcy;
-    int dstx;
-    int dsty;
-
-    //src mat with roi
-    cv::Mat mat_roi;
-    cv::Mat dst_roi;
-    //std::vector<cv::ocl::Info> oclinfo;
-    //ocl dst mat for testing
-    cv::ocl::oclMat gdst_whole;
-
-    //ocl mat with roi
-    cv::ocl::oclMat gmat;
-    cv::ocl::oclMat gdst;
+    cv::Size ksize;
 
     virtual void SetUp()
     {
         type = GET_PARAM(0);
         ksize = GET_PARAM(1);
-
-        cv::RNG &rng = TS::ptr()->get_rng();
-        cv::Size size(MWIDTH, MHEIGHT);
-
-        mat  = randomMat(rng, size, type, 5, 16, false);
-        dst  = randomMat(rng, size, type, 5, 16, false);
-
-        //int devnums = getDevice(oclinfo, OPENCV_DEFAULT_OPENCL_DEVICE);
-        //CV_Assert(devnums > 0);
-        ////if you want to use undefault device, set it here
-        ////setDevice(oclinfo[0]);
-    }
-
-    void random_roi()
-    {
-#ifdef RANDOMROI
-        //randomize ROI
-        cv::RNG &rng = TS::ptr()->get_rng();
-        roicols = rng.uniform(2, mat.cols);
-        roirows = rng.uniform(2, mat.rows);
-        srcx   = rng.uniform(0, mat.cols - roicols);
-        srcy   = rng.uniform(0, mat.rows - roirows);
-        dstx    = rng.uniform(0, dst.cols  - roicols);
-        dsty    = rng.uniform(0, dst.rows  - roirows);
-#else
-        roicols = mat.cols;
-        roirows = mat.rows;
-        srcx = 0;
-        srcy = 0;
-        dstx = 0;
-        dsty = 0;
-#endif
-
-        mat_roi  = mat(Rect(srcx, srcy, roicols, roirows));
-        dst_roi  = dst(Rect(dstx, dsty, roicols, roirows));
-
-        gdst_whole = dst;
-        gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
-
-        gmat = mat_roi;
+        Init(type);
     }
 };
 
-struct Laplacian : LaplacianTestBase {};
-
 TEST_P(Laplacian, Accuracy)
 {
     for(int j = 0; j < LOOP_TIMES; j++)
     {
         random_roi();
-
-        cv::Laplacian(mat_roi, dst_roi, -1, ksize, 1);
-        cv::ocl::Laplacian(gmat, gdst, -1, ksize, 1);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d", roicols, roirows, srcx, srcy, dstx, dsty);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        cv::Laplacian(mat1_roi, dst_roi, -1, ksize.width, 1);
+        cv::ocl::Laplacian(gmat1, gdst, -1, ksize.width, 1);
+        Near(1e-5);
     }
 }
 
@@ -364,8 +188,7 @@ TEST_P(Laplacian, Accuracy)
 
 /////////////////////////////////////////////////////////////////////////////////////////////////
 // erode & dilate
-
-PARAM_TEST_CASE(ErodeDilateBase, MatType, int)
+struct ErodeDilate : FilterTestBase
 {
     int type;
     int iterations;
@@ -373,210 +196,54 @@ PARAM_TEST_CASE(ErodeDilateBase, MatType, int)
     //erode or dilate kernel
     cv::Mat kernel;
 
-    //src mat
-    cv::Mat mat1;
-    cv::Mat dst;
-
-    // set up roi
-    int roicols;
-    int roirows;
-    int src1x;
-    int src1y;
-    int dstx;
-    int dsty;
-
-    //src mat with roi
-    cv::Mat mat1_roi;
-    cv::Mat dst_roi;
-    //std::vector<cv::ocl::Info> oclinfo;
-    //ocl dst mat for testing
-    cv::ocl::oclMat gdst_whole;
-
-    //ocl mat with roi
-    cv::ocl::oclMat gmat1;
-    cv::ocl::oclMat gdst;
-
     virtual void SetUp()
     {
         type = GET_PARAM(0);
-        iterations = GET_PARAM(1);
-
-        cv::RNG &rng = TS::ptr()->get_rng();
-        cv::Size size(MWIDTH, MHEIGHT);
-
-        mat1 = randomMat(rng, size, type, 5, 16, false);
-        dst  = randomMat(rng, size, type, 5, 16, false);
+        iterations = GET_PARAM(3);
+        Init(type);
         //		rng.fill(kernel, cv::RNG::UNIFORM, cv::Scalar::all(0), cv::Scalar::all(3));
-        kernel = randomMat(rng, Size(3, 3), CV_8UC1, 0, 3, false);
-
-    }
-
-    void random_roi()
-    {
-#ifdef RANDOMROI
-        //randomize ROI
-        cv::RNG &rng = TS::ptr()->get_rng();
-        roicols = rng.uniform(2, mat1.cols);
-        roirows = rng.uniform(2, mat1.rows);
-        src1x   = rng.uniform(0, mat1.cols - roicols);
-        src1y   = rng.uniform(0, mat1.rows - roirows);
-        dstx    = rng.uniform(0, dst.cols  - roicols);
-        dsty    = rng.uniform(0, dst.rows  - roirows);
-#else
-        roicols = mat1.cols;
-        roirows = mat1.rows;
-        src1x = 0;
-        src1y = 0;
-        dstx = 0;
-        dsty = 0;
-#endif
-
-        mat1_roi = mat1(Rect(src1x, src1y, roicols, roirows));
-        dst_roi  = dst(Rect(dstx, dsty, roicols, roirows));
-
-        gdst_whole = dst;
-        gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
-
-
-        gmat1 = mat1_roi;
+        kernel = randomMat(Size(3, 3), CV_8UC1, 0, 3);
     }
 
 };
 
-// erode
-
-struct Erode : ErodeDilateBase {};
-
-TEST_P(Erode, Mat)
+TEST_P(ErodeDilate, Mat)
 {
     for(int j = 0; j < LOOP_TIMES; j++)
     {
         random_roi();
-
         cv::erode(mat1_roi, dst_roi, kernel, Point(-1, -1), iterations);
         cv::ocl::erode(gmat1, gdst, kernel, Point(-1, -1), iterations);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d", roicols, roirows, src1x, src1y, dstx, dsty);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        Near(1e-5);
     }
-
-}
-
-
-
-
-
-// dilate
-
-struct Dilate : ErodeDilateBase {};
-
-TEST_P(Dilate, Mat)
-{
     for(int j = 0; j < LOOP_TIMES; j++)
     {
         random_roi();
-        cv::erode(mat1_roi, dst_roi, kernel, Point(-1, -1), iterations);
-        cv::ocl::erode(gmat1, gdst, kernel, Point(-1, -1), iterations);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d", roicols, roirows, src1x, src1y, dstx, dsty);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1e-5, sss);
+        cv::dilate(mat1_roi, dst_roi, kernel, Point(-1, -1), iterations);
+        cv::ocl::dilate(gmat1, gdst, kernel, Point(-1, -1), iterations);
+        Near(1e-5);
     }
-
 }
 
 
-
-
 /////////////////////////////////////////////////////////////////////////////////////////////////
 // Sobel
-
-PARAM_TEST_CASE(Sobel, MatType, int, int, int, int)
+struct Sobel : FilterTestBase
 {
     int type;
     int dx, dy, ksize, bordertype;
 
-    //src mat
-    cv::Mat mat1;
-    cv::Mat dst;
-
-    // set up roi
-    int roicols;
-    int roirows;
-    int src1x;
-    int src1y;
-    int dstx;
-    int dsty;
-
-    //src mat with roi
-    cv::Mat mat1_roi;
-    cv::Mat dst_roi;
-    //std::vector<cv::ocl::Info> oclinfo;
-    //ocl dst mat for testing
-    cv::ocl::oclMat gdst_whole;
-
-    //ocl mat with roi
-    cv::ocl::oclMat gmat1;
-    cv::ocl::oclMat gdst;
-
     virtual void SetUp()
     {
         type = GET_PARAM(0);
-        dx = GET_PARAM(1);
-        dy = GET_PARAM(2);
-        ksize = GET_PARAM(3);
-        bordertype = GET_PARAM(4);
-
-
-        cv::RNG &rng = TS::ptr()->get_rng();
-        cv::Size size(MWIDTH, MHEIGHT);
-
-        mat1 = randomMat(rng, size, type, 5, 16, false);
-        dst  = randomMat(rng, size, type, 5, 16, false);
-
-        //int devnums = getDevice(oclinfo, OPENCV_DEFAULT_OPENCL_DEVICE);
-        //CV_Assert(devnums > 0);
-        ////if you want to use undefault device, set it here
-        ////setDevice(oclinfo[0]);
-    }
-
-    void random_roi()
-    {
-#ifdef RANDOMROI
-        //randomize ROI
-        cv::RNG &rng = TS::ptr()->get_rng();
-        roicols = rng.uniform(2, mat1.cols);
-        roirows = rng.uniform(2, mat1.rows);
-        src1x   = rng.uniform(0, mat1.cols - roicols);
-        src1y   = rng.uniform(0, mat1.rows - roirows);
-        dstx    = rng.uniform(0, dst.cols  - roicols);
-        dsty    = rng.uniform(0, dst.rows  - roirows);
-#else
-        roicols = mat1.cols;
-        roirows = mat1.rows;
-        src1x = 0;
-        src1y = 0;
-        dstx = 0;
-        dsty = 0;
-#endif
-
-        mat1_roi = mat1(Rect(src1x, src1y, roicols, roirows));
-        dst_roi  = dst(Rect(dstx, dsty, roicols, roirows));
-
-        gdst_whole = dst;
-        gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
-
-
-        gmat1 = mat1_roi;
+        Size s = GET_PARAM(1);
+        ksize = s.width;
+        s = GET_PARAM(2);
+        dx = s.width;
+        dy = s.height;
+        bordertype = GET_PARAM(3);
+        Init(type);
     }
-
 };
 
 TEST_P(Sobel, Mat)
@@ -584,103 +251,29 @@ TEST_P(Sobel, Mat)
     for(int j = 0; j < LOOP_TIMES; j++)
     {
         random_roi();
-
         cv::Sobel(mat1_roi, dst_roi, -1, dx, dy, ksize, /*scale*/0.00001,/*delta*/0, bordertype);
         cv::ocl::Sobel(gmat1, gdst, -1, dx, dy, ksize,/*scale*/0.00001,/*delta*/0, bordertype);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d", roicols, roirows, src1x, src1y, dstx, dsty);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1, sss);
+        Near(1);
     }
-
 }
 
 
 /////////////////////////////////////////////////////////////////////////////////////////////////
 // Scharr
-
-PARAM_TEST_CASE(Scharr, MatType, int, int, int)
+struct Scharr : FilterTestBase
 {
     int type;
     int dx, dy, bordertype;
 
-    //src mat
-    cv::Mat mat1;
-    cv::Mat dst;
-
-    // set up roi
-    int roicols;
-    int roirows;
-    int src1x;
-    int src1y;
-    int dstx;
-    int dsty;
-
-    //src mat with roi
-    cv::Mat mat1_roi;
-    cv::Mat dst_roi;
-    //std::vector<cv::ocl::Info> oclinfo;
-    //ocl dst mat for testing
-    cv::ocl::oclMat gdst_whole;
-
-    //ocl mat with roi
-    cv::ocl::oclMat gmat1;
-    cv::ocl::oclMat gdst;
-
     virtual void SetUp()
     {
         type = GET_PARAM(0);
-        dx = GET_PARAM(1);
-        dy = GET_PARAM(2);
+        Size s = GET_PARAM(2);
+        dx = s.width;
+        dy = s.height;
         bordertype = GET_PARAM(3);
-        dx = 1;
-        dy = 0;
-
-        cv::RNG &rng = TS::ptr()->get_rng();
-        cv::Size size(MWIDTH, MHEIGHT);
-
-        mat1 = randomMat(rng, size, type, 5, 16, false);
-        dst  = randomMat(rng, size, type, 5, 16, false);
-
-        //int devnums = getDevice(oclinfo, OPENCV_DEFAULT_OPENCL_DEVICE);
-        //CV_Assert(devnums > 0);
-        ////if you want to use undefault device, set it here
-        ////setDevice(oclinfo[0]);
+        Init(type);
     }
-
-    void random_roi()
-    {
-#ifdef RANDOMROI
-        //randomize ROI
-        cv::RNG &rng = TS::ptr()->get_rng();
-        roicols = rng.uniform(2, mat1.cols);
-        roirows = rng.uniform(2, mat1.rows);
-        src1x   = rng.uniform(0, mat1.cols - roicols);
-        src1y   = rng.uniform(0, mat1.rows - roirows);
-        dstx    = rng.uniform(0, dst.cols  - roicols);
-        dsty    = rng.uniform(0, dst.rows  - roirows);
-#else
-        roicols = mat1.cols;
-        roirows = mat1.rows;
-        src1x = 0;
-        src1y = 0;
-        dstx = 0;
-        dsty = 0;
-#endif
-
-        mat1_roi = mat1(Rect(src1x, src1y, roicols, roirows));
-        dst_roi  = dst(Rect(dstx, dsty, roicols, roirows));
-
-        gdst_whole = dst;
-        gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
-
-
-        gmat1 = mat1_roi;
-    }
-
 };
 
 TEST_P(Scharr, Mat)
@@ -688,16 +281,9 @@ TEST_P(Scharr, Mat)
     for(int j = 0; j < LOOP_TIMES; j++)
     {
         random_roi();
-
         cv::Scharr(mat1_roi, dst_roi, -1, dx, dy, /*scale*/1,/*delta*/0, bordertype);
         cv::ocl::Scharr(gmat1, gdst, -1, dx, dy,/*scale*/1,/*delta*/0, bordertype);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d", roicols, roirows, src1x, src1y, dstx, dsty);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1, sss);
+        Near(1);
     }
 
 }
@@ -705,89 +291,23 @@ TEST_P(Scharr, Mat)
 
 /////////////////////////////////////////////////////////////////////////////////////////////////
 // GaussianBlur
-
-PARAM_TEST_CASE(GaussianBlur, MatType, cv::Size, int)
+struct GaussianBlur : FilterTestBase
 {
     int type;
     cv::Size ksize;
     int bordertype;
-
     double sigma1, sigma2;
 
-    //src mat
-    cv::Mat mat1;
-    cv::Mat dst;
-
-    // set up roi
-    int roicols;
-    int roirows;
-    int src1x;
-    int src1y;
-    int dstx;
-    int dsty;
-
-    //src mat with roi
-    cv::Mat mat1_roi;
-    cv::Mat dst_roi;
-    //std::vector<cv::ocl::Info> oclinfo;
-    //ocl dst mat for testing
-    cv::ocl::oclMat gdst_whole;
-
-    //ocl mat with roi
-    cv::ocl::oclMat gmat1;
-    cv::ocl::oclMat gdst;
-
     virtual void SetUp()
     {
         type = GET_PARAM(0);
         ksize = GET_PARAM(1);
-        bordertype = GET_PARAM(2);
-
+        bordertype = GET_PARAM(3);
+        Init(type);
         cv::RNG &rng = TS::ptr()->get_rng();
-        cv::Size size(MWIDTH, MHEIGHT);
-
         sigma1 = rng.uniform(0.1, 1.0);
         sigma2 = rng.uniform(0.1, 1.0);
-
-        mat1 = randomMat(rng, size, type, 5, 16, false);
-        dst  = randomMat(rng, size, type, 5, 16, false);
-
-        //int devnums = getDevice(oclinfo, OPENCV_DEFAULT_OPENCL_DEVICE);
-        //CV_Assert(devnums > 0);
-        ////if you want to use undefault device, set it here
-        ////setDevice(oclinfo[0]);
     }
-
-    void random_roi()
-    {
-#ifdef RANDOMROI
-        //randomize ROI
-        cv::RNG &rng = TS::ptr()->get_rng();
-        roicols = rng.uniform(2, mat1.cols);
-        roirows = rng.uniform(2, mat1.rows);
-        src1x   = rng.uniform(0, mat1.cols - roicols);
-        src1y   = rng.uniform(0, mat1.rows - roirows);
-        dstx    = rng.uniform(0, dst.cols  - roicols);
-        dsty    = rng.uniform(0, dst.rows  - roirows);
-#else
-        roicols = mat1.cols;
-        roirows = mat1.rows;
-        src1x = 0;
-        src1y = 0;
-        dstx = 0;
-        dsty = 0;
-#endif
-
-        mat1_roi = mat1(Rect(src1x, src1y, roicols, roirows));
-        dst_roi  = dst(Rect(dstx, dsty, roicols, roirows));
-
-        gdst_whole = dst;
-        gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
-
-
-        gmat1 = mat1_roi;
-    }
-
 };
 
 TEST_P(GaussianBlur, Mat)
@@ -795,53 +315,53 @@ TEST_P(GaussianBlur, Mat)
     for(int j = 0; j < LOOP_TIMES; j++)
     {
         random_roi();
-
         cv::GaussianBlur(mat1_roi, dst_roi, ksize, sigma1, sigma2, bordertype);
         cv::ocl::GaussianBlur(gmat1, gdst, ksize, sigma1, sigma2, bordertype);
-
-        cv::Mat cpu_dst;
-        gdst_whole.download(cpu_dst);
-        char sss[1024];
-        sprintf(sss, "roicols=%d,roirows=%d,src1x=%d,src1y=%d,dstx=%d,dsty=%d", roicols, roirows, src1x, src1y, dstx, dsty);
-
-        EXPECT_MAT_NEAR(dst, cpu_dst, 1.0, sss);
+        Near(1);
     }
 
 }
 
 
 
-INSTANTIATE_TEST_CASE_P(Filter, Blur, Combine(Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC4),
+INSTANTIATE_TEST_CASE_P(Filter, Blur, Combine(
+                        Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC4),
                         Values(cv::Size(3, 3), cv::Size(5, 5), cv::Size(7, 7)),
+                        Values(Size(0, 0)), //not use
                         Values((MatType)cv::BORDER_CONSTANT, (MatType)cv::BORDER_REPLICATE, (MatType)cv::BORDER_REFLECT, (MatType)cv::BORDER_REFLECT_101)));
 
 
 INSTANTIATE_TEST_CASE_P(Filters, Laplacian, Combine(
-                            Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC3, CV_32FC4),
-                            Values(1, 3)));
-
-INSTANTIATE_TEST_CASE_P(Filter, Erode, Combine(Values(CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4), Values(1)));
-
-//INSTANTIATE_TEST_CASE_P(Filter, Erode, Combine(Values(CV_8UC1, CV_8UC1), Values(false)));
-
-INSTANTIATE_TEST_CASE_P(Filter, Dilate, Combine(Values(CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4), Values(1)));
+                        Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC3, CV_32FC4),
+                        Values(Size(3, 3)),
+                        Values(Size(0, 0)), //not use
+                        Values(0)));        //not use
 
-//INSTANTIATE_TEST_CASE_P(Filter, Dilate, Combine(Values(CV_8UC1, CV_8UC1), Values(false)));
+INSTANTIATE_TEST_CASE_P(Filter, ErodeDilate, Combine(
+                        Values(CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4), 
+                        Values(Size(0, 0)), //not use
+                        Values(Size(0, 0)), //not use
+                        Values(1)));
 
 
-INSTANTIATE_TEST_CASE_P(Filter, Sobel, Combine(Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC3, CV_32FC4),
-                        Values(1, 2), Values(0, 1), Values(3, 5), Values((MatType)cv::BORDER_CONSTANT,
-                                (MatType)cv::BORDER_REPLICATE)));
+INSTANTIATE_TEST_CASE_P(Filter, Sobel, Combine(
+                        Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC3, CV_32FC4),
+                        Values(Size(3, 3), Size(5, 5)),
+                        Values(Size(1, 0), Size(1, 1), Size(2, 0), Size(2, 1)),
+                        Values((MatType)cv::BORDER_CONSTANT, (MatType)cv::BORDER_REPLICATE)));
 
 
 INSTANTIATE_TEST_CASE_P(Filter, Scharr, Combine(
-                            Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC4), Values(0, 1), Values(0, 1),
-                            Values((MatType)cv::BORDER_CONSTANT, (MatType)cv::BORDER_REPLICATE)));
+                        Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC4),
+                        Values(Size(0, 0)), //not use
+                        Values(Size(0, 1), Size(1, 0)), 
+                        Values((MatType)cv::BORDER_CONSTANT, (MatType)cv::BORDER_REPLICATE)));
 
 INSTANTIATE_TEST_CASE_P(Filter, GaussianBlur, Combine(
-                            Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC4),
-                            Values(cv::Size(3, 3), cv::Size(5, 5)),
-                            Values((MatType)cv::BORDER_CONSTANT, (MatType)cv::BORDER_REPLICATE)));
+                        Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC4),
+                        Values(Size(3, 3), Size(5, 5)),
+                        Values(Size(0, 0)), //not use
+                        Values((MatType)cv::BORDER_CONSTANT, (MatType)cv::BORDER_REPLICATE)));
 
 
 

modules/ocl/test/test_gemm.cpp

@@ -53,13 +53,12 @@ PARAM_TEST_CASE(Gemm, int, cv::Size, int)
     int      type;
     cv::Size mat_size;
     int		 flags;
-    //vector<cv::ocl::Info> info;
+
     virtual void SetUp()
     {
         type     = GET_PARAM(0);
         mat_size = GET_PARAM(1);
         flags    = GET_PARAM(2);
-        //cv::ocl::getDevice(info);
     }
 };
 

modules/ocl/test/test_haar.cpp

@@ -12,10 +12,12 @@
 //
 // Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
 // Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
 // Third party copyrights are property of their respective owners.
 //
 // @Authors
 //    Jia Haipeng, jiahaipeng95@gmail.com
+//    Sen Liu, swjutls1987@126.com
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
@@ -61,40 +63,31 @@ struct getRect
     }
 };
 
-PARAM_TEST_CASE(HaarTestBase, int, int)
+PARAM_TEST_CASE(Haar, double, int)
 {
-    //std::vector<cv::ocl::Info> oclinfo;
     cv::ocl::OclCascadeClassifier cascade, nestedCascade;
+    cv::ocl::OclCascadeClassifierBuf cascadebuf;
     cv::CascadeClassifier cpucascade, cpunestedCascade;
-    //    Mat img;
 
     double scale;
-    int index;
+    int flags;
 
     virtual void SetUp()
     {
-        scale = 1.0;
-        index = 0;
+        scale = GET_PARAM(0);
+        flags = GET_PARAM(1);
         string cascadeName = workdir + "../../data/haarcascades/haarcascade_frontalface_alt.xml";
 
-        if( (!cascade.load( cascadeName )) || (!cpucascade.load(cascadeName)))
+        if( (!cascade.load( cascadeName )) || (!cpucascade.load(cascadeName)) || (!cascadebuf.load( cascadeName )))
         {
             cout << "ERROR: Could not load classifier cascade" << endl;
             return;
         }
-        //int devnums = getDevice(oclinfo);
-        //CV_Assert(devnums>0);
-        ////if you want to use undefault device, set it here
-        ////setDevice(oclinfo[0]);
-        //cv::ocl::setBinpath("E:\\");
     }
 };
 
 ////////////////////////////////faceDetect/////////////////////////////////////////////////
-
-struct Haar : HaarTestBase {};
-
-TEST_F(Haar, FaceDetect)
+TEST_P(Haar, FaceDetect)
 {
     string imgName = workdir + "lena.jpg";
     Mat img = imread( imgName, 1 );
@@ -105,59 +98,65 @@ TEST_F(Haar, FaceDetect)
         return ;
     }
 
-    //int i = 0;
-    //double t = 0;
     vector<Rect> faces, oclfaces;
 
-    // const static Scalar colors[] =  { CV_RGB(0, 0, 255),
-    //                                   CV_RGB(0, 128, 255),
-    //                                   CV_RGB(0, 255, 255),
-    //                                   CV_RGB(0, 255, 0),
-    //                                   CV_RGB(255, 128, 0),
-    //                                   CV_RGB(255, 255, 0),
-    //                                   CV_RGB(255, 0, 0),
-    //                                   CV_RGB(255, 0, 255)
-    //                                 } ;
-
     Mat gray, smallImg(cvRound (img.rows / scale), cvRound(img.cols / scale), CV_8UC1 );
     MemStorage storage(cvCreateMemStorage(0));
     cvtColor( img, gray, COLOR_BGR2GRAY );
     resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR );
     equalizeHist( smallImg, smallImg );
 
-
     cv::ocl::oclMat image;
     CvSeq *_objects;
     image.upload(smallImg);
     _objects = cascade.oclHaarDetectObjects( image, storage, 1.1,
-               3, 0
-               | CV_HAAR_SCALE_IMAGE
-               , Size(30, 30), Size(0, 0) );
+                   3, flags, Size(30, 30), Size(0, 0) );
     vector<CvAvgComp> vecAvgComp;
     Seq<CvAvgComp>(_objects).copyTo(vecAvgComp);
     oclfaces.resize(vecAvgComp.size());
     std::transform(vecAvgComp.begin(), vecAvgComp.end(), oclfaces.begin(), getRect());
 
-    cpucascade.detectMultiScale( smallImg, faces,  1.1,
-                                 3, 0
-                                 | CV_HAAR_SCALE_IMAGE
-                                 , Size(30, 30), Size(0, 0) );
+    cpucascade.detectMultiScale( smallImg, faces,  1.1, 3,
+                                 flags,
+                                 Size(30, 30), Size(0, 0) );
     EXPECT_EQ(faces.size(), oclfaces.size());
-    /*	for( vector<Rect>::const_iterator r = faces.begin(); r != faces.end(); r++, i++ )
+}
+
+TEST_P(Haar, FaceDetectUseBuf)
+{
+    string imgName = workdir + "lena.jpg";
+    Mat img = imread( imgName, 1 );
+
+    if(img.empty())
     {
-    Mat smallImgROI;
-    Point center;
-    Scalar color = colors[i%8];
-    int radius;
-    center.x = cvRound((r->x + r->width*0.5)*scale);
-    center.y = cvRound((r->y + r->height*0.5)*scale);
-    radius = cvRound((r->width + r->height)*0.25*scale);
-    circle( img, center, radius, color, 3, 8, 0 );
-    } */
-    //namedWindow("result");
-    //imshow("result",img);
-    //waitKey(0);
-    //destroyAllWindows();
+        std::cout << "Couldn't read " << imgName << std::endl;
+        return ;
+    }
 
+    vector<Rect> faces, oclfaces;
+
+    Mat gray, smallImg(cvRound (img.rows / scale), cvRound(img.cols / scale), CV_8UC1 );
+    MemStorage storage(cvCreateMemStorage(0));
+    cvtColor( img, gray, CV_BGR2GRAY );
+    resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR );
+    equalizeHist( smallImg, smallImg );
+
+    cv::ocl::oclMat image;
+    image.upload(smallImg);
+
+    cascadebuf.detectMultiScale( image, oclfaces,  1.1, 3,
+                                 flags,
+                                 Size(30, 30), Size(0, 0) );
+    cascadebuf.release();
+
+    cpucascade.detectMultiScale( smallImg, faces,  1.1, 3,
+                                 flags,
+                                 Size(30, 30), Size(0, 0) );
+    EXPECT_EQ(faces.size(), oclfaces.size());
 }
+
+INSTANTIATE_TEST_CASE_P(FaceDetect, Haar,
+    Combine(Values(1.0),
+            Values(CV_HAAR_SCALE_IMAGE, 0)));
+
 #endif // HAVE_OPENCL

modules/ocl/test/test_hog.cpp

@@ -240,12 +240,11 @@ TEST_P(HOG, Detect)
         }
     }
 
-    char s[100] = {0};
-    EXPECT_MAT_NEAR(cv::Mat(d_comp), cv::Mat(comp), 3, s);
+    EXPECT_MAT_NEAR(cv::Mat(d_comp), cv::Mat(comp), 3);
 }
 
 
-INSTANTIATE_TEST_CASE_P(GPU_ImgProc, HOG, testing::Combine(
+INSTANTIATE_TEST_CASE_P(OCL_ObjDetect, HOG, testing::Combine(
                             testing::Values(cv::Size(64, 128), cv::Size(48, 96)),
                             testing::Values(MatType(CV_8UC1), MatType(CV_8UC4))));