Merge pull request #102 from NikoKJ/ocl4_2.43

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

@@ -378,7 +378,7 @@ namespace cv
         // disabled until fix crash
         // support all types
         CV_EXPORTS Scalar sum(const oclMat &m);
-
+        CV_EXPORTS Scalar absSum(const oclMat &m);
         CV_EXPORTS Scalar sqrSum(const oclMat &m);
 
         //! finds global minimum and maximum array elements and returns their values
@@ -1738,5 +1738,12 @@ namespace cv
 
     }
 }
+#if _MSC_VER >= 1200
+#pragma warning( push)
+#pragma warning( disable: 4267)
+#endif
 #include "opencv2/ocl/matrix_operations.hpp"
+#if _MSC_VER >= 1200
+#pragma warning( pop)
+#endif
 #endif /* __OPENCV_GPU_HPP__ */

modules/ocl/perf/perf_haar.cpp

@@ -137,7 +137,7 @@ TEST_F(Haar, FaceDetect)
     printf( "cpudetection time = %g ms\n", t / (LOOP_TIMES * (double)cvGetTickFrequency() * 1000.) );
 
     cv::ocl::oclMat image;
-    CvSeq *_objects;
+    CvSeq *_objects=NULL;
     t = (double)cvGetTickCount();
     for(int k = 0; k < LOOP_TIMES; k++)
     {

modules/ocl/perf/perf_hog.cpp

@@ -103,10 +103,10 @@ TEST_P(HOG, Performance)
     ASSERT_FALSE(img.empty());
 
     // define HOG related arguments
-    float scale = 1.05;
+    float scale = 1.05f;
     //int nlevels = 13;
-    float gr_threshold = 8;
-    float hit_threshold = 1.4;
+    int gr_threshold = 8;
+    float hit_threshold = 1.4f;
     //bool hit_threshold_auto = true;
 
     int win_width = is48 ? 48 : 64;

modules/ocl/perf/perf_surf.cpp

@@ -54,11 +54,11 @@ using namespace cvtest;
 using namespace testing;
 using namespace std;
 
-extern std::string workdir;
+#define FILTER_IMAGE "../../../samples/gpu/road.png"
 
 TEST(SURF, Performance)
 {
-    cv::Mat img = readImage(workdir+"lena.jpg", cv::IMREAD_GRAYSCALE);
+    cv::Mat img = readImage(FILTER_IMAGE, cv::IMREAD_GRAYSCALE);
     ASSERT_FALSE(img.empty());
 
     ocl::SURF_OCL d_surf;

modules/ocl/perf/precomp.hpp

@@ -55,13 +55,13 @@
 #include "cvconfig.h"
 #include "opencv2/core/core.hpp"
 #include "opencv2/highgui/highgui.hpp"
-#include "opencv2/calib3d/calib3d.hpp"
+//#include "opencv2/calib3d/calib3d.hpp"
 #include "opencv2/imgproc/imgproc.hpp"
 #include "opencv2/video/video.hpp"
 #include "opencv2/ts/ts.hpp"
 #include "opencv2/ts/ts_perf.hpp"
 #include "opencv2/ocl/ocl.hpp"
-#include "opencv2/nonfree/nonfree.hpp"
+//#include "opencv2/nonfree/nonfree.hpp"
 
 #include "utility.hpp"
 #include "interpolation.hpp"

modules/ocl/perf/utility.hpp

@@ -45,7 +45,7 @@
 #ifdef PRINT_KERNEL_RUN_TIME
 #define LOOP_TIMES 1
 #else
-#define LOOP_TIMES 100
+#define LOOP_TIMES 1
 #endif
 #define MWIDTH 1920
 #define MHEIGHT 1080

modules/ocl/src/arithm.cpp

@@ -813,6 +813,22 @@ Scalar cv::ocl::sum(const oclMat &src)
     return func(src, 0);
 }
 
+Scalar cv::ocl::absSum(const oclMat &src)
+{
+    if(src.clCxt->impl->double_support == 0 && src.depth() == CV_64F)
+    {
+        CV_Error(CV_GpuNotSupported, "select device don't support double");
+    }
+    static sumFunc functab[2] =
+    {
+        arithmetic_sum<float>,
+        arithmetic_sum<double>
+    };
+
+    sumFunc func;
+    func = functab[src.clCxt->impl->double_support];
+    return func(src, 1);
+}
 
 Scalar cv::ocl::sqrSum(const oclMat &src)
 {
@@ -945,13 +961,17 @@ template <typename T> void arithmetic_minMax(const oclMat &src, double *minVal,
     T *p = new T[groupnum * vlen * 2];
     memset(p, 0, dbsize);
     openCLReadBuffer(clCxt, dstBuffer, (void *)p, dbsize);
-    for(int i = 0; i < vlen * (int)groupnum; i++)
-    {
-        *minVal = *minVal < p[i] ? *minVal : p[i];
+    if(minVal != NULL){
+        for(int i = 0; i < vlen * (int)groupnum; i++)
+        {
+            *minVal = *minVal < p[i] ? *minVal : p[i];
+        }
     }
-    for(int i = vlen * (int)groupnum; i < 2 * vlen * (int)groupnum; i++)
-    {
-        *maxVal = *maxVal > p[i] ? *maxVal : p[i];
+    if(maxVal != NULL){
+        for(int i = vlen * (int)groupnum; i < 2 * vlen * (int)groupnum; i++)
+        {
+            *maxVal = *maxVal > p[i] ? *maxVal : p[i];
+        }
     }
     delete[] p;
     openCLFree(dstBuffer);

modules/ocl/src/brute_force_matcher.cpp

@@ -1548,8 +1548,7 @@ void cv::ocl::BruteForceMatcher_OCL_base::knnMatch(const oclMat &query, vector<
         temp.reserve(2 * k);
 
         matches.resize(query.rows);
-        for(size_t queryIdx = 0; queryIdx < matches.size(); queryIdx++ )
-            matches[queryIdx].reserve(k);
+        for_each(matches.begin(), matches.end(), bind2nd(mem_fun_ref(&vector<DMatch>::reserve), k));
 
         for (size_t imgIdx = 0, size = trainDescCollection.size(); imgIdx < size; ++imgIdx)
         {
@@ -1573,15 +1572,8 @@ void cv::ocl::BruteForceMatcher_OCL_base::knnMatch(const oclMat &query, vector<
 
         if (compactResult)
         {
-            size_t i, j = 0;
-            for( i = 0; i < matches.size(); i++ )
-                if( !matches[i].empty() )
-                {
-                    if( i > j )
-                        matches[j] = matches[i];
-                    j++;
-                }
-            matches.resize(j);
+            vector< vector<DMatch> >::iterator new_end = remove_if(matches.begin(), matches.end(), mem_fun_ref(&vector<DMatch>::empty));
+            matches.erase(new_end, matches.end());
         }
     }
 }

modules/ocl/src/build_warps.cpp

@@ -76,7 +76,7 @@ namespace cv
 //////////////////////////////////////////////////////////////////////////////
 // buildWarpPlaneMaps
 
-void cv::ocl::buildWarpPlaneMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, const Mat &T,
+void cv::ocl::buildWarpPlaneMaps(Size /*src_size*/, Rect dst_roi, const Mat &K, const Mat &R, const Mat &T,
                                  float scale, oclMat &map_x, oclMat &map_y)
 {
     CV_Assert(K.size() == Size(3, 3) && K.type() == CV_32F);
@@ -121,7 +121,7 @@ void cv::ocl::buildWarpPlaneMaps(Size src_size, Rect dst_roi, const Mat &K, cons
 //////////////////////////////////////////////////////////////////////////////
 // buildWarpCylyndricalMaps
 
-void cv::ocl::buildWarpCylindricalMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, float scale,
+void cv::ocl::buildWarpCylindricalMaps(Size /*src_size*/, Rect dst_roi, const Mat &K, const Mat &R, float scale,
                                        oclMat &map_x, oclMat &map_y)
 {
     CV_Assert(K.size() == Size(3, 3) && K.type() == CV_32F);
@@ -160,7 +160,7 @@ void cv::ocl::buildWarpCylindricalMaps(Size src_size, Rect dst_roi, const Mat &K
 
 //////////////////////////////////////////////////////////////////////////////
 // buildWarpSphericalMaps
-void cv::ocl::buildWarpSphericalMaps(Size src_size, Rect dst_roi, const Mat &K, const Mat &R, float scale,
+void cv::ocl::buildWarpSphericalMaps(Size /*src_size*/, Rect dst_roi, const Mat &K, const Mat &R, float scale,
                                      oclMat &map_x, oclMat &map_y)
 {
     CV_Assert(K.size() == Size(3, 3) && K.type() == CV_32F);

modules/ocl/src/color.cpp

@@ -96,7 +96,7 @@ namespace
         size_t gt[3] = {src.cols, src.rows, 1}, lt[3] = {16, 16, 1};
         openCLExecuteKernel(src.clCxt, &cvt_color, "RGB2Gray", gt, lt, args, -1, -1, build_options);
     }
-    void cvtColor_caller(const oclMat &src, oclMat &dst, int code, int dcn)
+    void cvtColor_caller(const oclMat &src, oclMat &dst, int code, int /*dcn*/)
     {
         Size sz = src.size();
         int scn = src.oclchannels(), depth = src.depth(), bidx;

modules/ocl/src/columnsum.cpp

@@ -88,7 +88,7 @@ void cv::ocl::columnSum(const oclMat &src, oclMat &dst)
     args.push_back( make_pair( sizeof(cl_int), (void *)&dst.step));
 
     size_t globalThreads[3] = {dst.cols, 1, 1};
-    size_t localThreads[3]  = {16, 16, 1};
+    size_t localThreads[3]  = {256, 1, 1};
 
     openCLExecuteKernel(clCxt, &imgproc_columnsum, kernelName, globalThreads, localThreads, args, src.channels(), src.depth());
 

modules/ocl/src/filtering.cpp

@@ -1235,7 +1235,7 @@ void linearRowFilter_gpu(const oclMat &src, const oclMat &dst, oclMat mat_kernel
     openCLExecuteKernel(clCxt, &filter_sep_row, kernelName, globalThreads, localThreads, args, channels, src.depth(), compile_option);
 }
 
-Ptr<BaseRowFilter_GPU> cv::ocl::getLinearRowFilter_GPU(int srcType, int bufType, const Mat &rowKernel, int anchor, int bordertype)
+Ptr<BaseRowFilter_GPU> cv::ocl::getLinearRowFilter_GPU(int srcType, int /*bufType*/, const Mat &rowKernel, int anchor, int bordertype)
 {
     static const gpuFilter1D_t gpuFilter1D_callers[6] =
     {
@@ -1391,7 +1391,7 @@ void linearColumnFilter_gpu(const oclMat &src, const oclMat &dst, oclMat mat_ker
     openCLExecuteKernel(clCxt, &filter_sep_col, kernelName, globalThreads, localThreads, args, -1, -1, compile_option);
 }
 
-Ptr<BaseColumnFilter_GPU> cv::ocl::getLinearColumnFilter_GPU(int bufType, int dstType, const Mat &columnKernel, int anchor, int bordertype, double delta)
+Ptr<BaseColumnFilter_GPU> cv::ocl::getLinearColumnFilter_GPU(int /*bufType*/, int dstType, const Mat &columnKernel, int anchor, int bordertype, double /*delta*/)
 {
     static const gpuFilter1D_t gpuFilter1D_callers[6] =
     {

modules/ocl/src/haar.cpp

@@ -162,7 +162,7 @@ typedef _ALIGNED_ON(128) struct  GpuHidHaarFeature
         _ALIGNED_ON(4)  int    p3 ;
         _ALIGNED_ON(4)  float weight  ;
     }
-    _ALIGNED_ON(32) rect[CV_HAAR_FEATURE_MAX] ;
+    /*_ALIGNED_ON(32)*/ rect[CV_HAAR_FEATURE_MAX] ;
 }
 GpuHidHaarFeature;
 
@@ -867,8 +867,8 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
     std::vector<cv::Rect> rectList;
     std::vector<int> rweights;
     double factor;
-    int datasize;
-    int totalclassifier;
+    int datasize=0;
+    int totalclassifier=0;
 
     void *out;
     GpuHidHaarClassifierCascade *gcascade;

modules/ocl/src/initialization.cpp

@@ -644,7 +644,7 @@ namespace cv
             size_t kernelWorkGroupSize;
             openCLSafeCall(clGetKernelWorkGroupInfo(kernel, clCxt->impl->devices[0],
                                                     CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &kernelWorkGroupSize, 0));
-            CV_DbgAssert( (localThreads[0] <= clCxt->impl->maxWorkItemSizes[0]) &&
+            CV_Assert( (localThreads[0] <= clCxt->impl->maxWorkItemSizes[0]) &&
                           (localThreads[1] <= clCxt->impl->maxWorkItemSizes[1]) &&
                           (localThreads[2] <= clCxt->impl->maxWorkItemSizes[2]) &&
                           ((localThreads[0] * localThreads[1] * localThreads[2]) <= kernelWorkGroupSize) &&

modules/ocl/src/interpolate_frames.cpp

@@ -218,7 +218,7 @@ void interpolate::vectorWarp(const oclMat &src, const oclMat &u, const oclMat &v
     normalizeKernel(buffer, src.rows, b_offset, d_offset);
 }
 
-void interpolate::blendFrames(const oclMat &frame0, const oclMat &frame1, const oclMat &buffer, float pos, oclMat &newFrame, cl_mem &tex_src0, cl_mem &tex_src1)
+void interpolate::blendFrames(const oclMat &frame0, const oclMat &/*frame1*/, const oclMat &buffer, float pos, oclMat &newFrame, cl_mem &tex_src0, cl_mem &tex_src1)
 {
     int step = buffer.step / sizeof(float);
 

modules/ocl/src/mssegmentation.cpp

@@ -150,7 +150,7 @@ namespace
     // Implementation
     //
 
-    DjSets DjSets::operator = (const DjSets &obj)
+    DjSets DjSets::operator = (const DjSets &/*obj*/)
     {
         //cout << "Invalid DjSets constructor\n";
         CV_Error(-1, "Invalid DjSets constructor\n");

modules/ocl/src/precomp.hpp

@@ -47,7 +47,7 @@
 #define __OPENCV_PRECOMP_H__
 
 #if _MSC_VER >= 1200
-#pragma warning( disable: 4244 4251 4710 4711 4514 4996 )
+#pragma warning( disable: 4267 4324 4244 4251 4710 4711 4514 4996 )
 #endif
 
 #ifdef HAVE_CVCONFIG_H

modules/ocl/src/pyrlk.cpp

@@ -581,26 +581,26 @@ void pyrDown_cus(const oclMat &src, oclMat &dst)
 //
 //void callF(const oclMat& src, oclMat& dst, MultiplyScalar op, int mask)
 //{
-//	Mat srcTemp;
-//	Mat dstTemp;
-//	src.download(srcTemp);
-//	dst.download(dstTemp);
+//  Mat srcTemp;
+//  Mat dstTemp;
+//  src.download(srcTemp);
+//  dst.download(dstTemp);
 //
-//	int i;
-//	int j;
-//	int k;
-//	for(i = 0; i < srcTemp.rows; i++)
-//	{
-//		for(j = 0; j < srcTemp.cols; j++)
-//		{
-//			for(k = 0; k < srcTemp.channels(); k++)
-//			{
-//				((float*)dstTemp.data)[srcTemp.channels() * (i * srcTemp.rows + j) + k] = (float)op(((float*)srcTemp.data)[srcTemp.channels() * (i * srcTemp.rows + j) + k]);
-//			}
-//		}
-//	}
+//  int i;
+//  int j;
+//  int k;
+//  for(i = 0; i < srcTemp.rows; i++)
+//  {
+//      for(j = 0; j < srcTemp.cols; j++)
+//      {
+//          for(k = 0; k < srcTemp.channels(); k++)
+//          {
+//              ((float*)dstTemp.data)[srcTemp.channels() * (i * srcTemp.rows + j) + k] = (float)op(((float*)srcTemp.data)[srcTemp.channels() * (i * srcTemp.rows + j) + k]);
+//          }
+//      }
+//  }
 //
-//	dst = dstTemp;
+//  dst = dstTemp;
 //}
 //
 //static inline bool isAligned(const unsigned char* ptr, size_t size)
@@ -622,54 +622,54 @@ void pyrDown_cus(const oclMat &src, oclMat &dst)
 //        return;
 //    }
 //
-//	Mat srcTemp;
-//	Mat dstTemp;
-//	src.download(srcTemp);
-//	dst.download(dstTemp);
+//  Mat srcTemp;
+//  Mat dstTemp;
+//  src.download(srcTemp);
+//  dst.download(dstTemp);
 //
-//	int x_shifted;
+//  int x_shifted;
 //
-//	int i;
-//	int j;
-//	for(i = 0; i < srcTemp.rows; i++)
-//	{
-//		const double* srcRow = (const double*)srcTemp.data + i * srcTemp.rows;
+//  int i;
+//  int j;
+//  for(i = 0; i < srcTemp.rows; i++)
+//  {
+//      const double* srcRow = (const double*)srcTemp.data + i * srcTemp.rows;
 //        double* dstRow = (double*)dstTemp.data + i * dstTemp.rows;;
 //
-//		for(j = 0; j < srcTemp.cols; j++)
-//		{
-//			x_shifted = j * 4;
+//      for(j = 0; j < srcTemp.cols; j++)
+//      {
+//          x_shifted = j * 4;
 //
-//			if(x_shifted + 4 - 1 < srcTemp.cols)
-//			{
-//				dstRow[x_shifted    ] = op(srcRow[x_shifted    ]);
-//				dstRow[x_shifted + 1] = op(srcRow[x_shifted + 1]);
-//				dstRow[x_shifted + 2] = op(srcRow[x_shifted + 2]);
-//				dstRow[x_shifted + 3] = op(srcRow[x_shifted + 3]);
-//			}
-//			else
-//			{
-//				for (int real_x = x_shifted; real_x < srcTemp.cols; ++real_x)
-//				{
-//					((float*)dstTemp.data)[i * srcTemp.rows + real_x] = op(((float*)srcTemp.data)[i * srcTemp.rows + real_x]);
-//				}
-//			}
-//		}
-//	}
+//          if(x_shifted + 4 - 1 < srcTemp.cols)
+//          {
+//              dstRow[x_shifted    ] = op(srcRow[x_shifted    ]);
+//              dstRow[x_shifted + 1] = op(srcRow[x_shifted + 1]);
+//              dstRow[x_shifted + 2] = op(srcRow[x_shifted + 2]);
+//              dstRow[x_shifted + 3] = op(srcRow[x_shifted + 3]);
+//          }
+//          else
+//          {
+//              for (int real_x = x_shifted; real_x < srcTemp.cols; ++real_x)
+//              {
+//                  ((float*)dstTemp.data)[i * srcTemp.rows + real_x] = op(((float*)srcTemp.data)[i * srcTemp.rows + real_x]);
+//              }
+//          }
+//      }
+//  }
 //}
 //
 //void multiply(const oclMat& src1, double val, oclMat& dst, double scale = 1.0f);
 //void multiply(const oclMat& src1, double val, oclMat& dst, double scale)
 //{
 //    MultiplyScalar op(val, scale);
-//	//if(src1.channels() == 1 && dst.channels() == 1)
-//	//{
-//	//    callT(src1, dst, op, 0);
-//	//}
-//	//else
-//	//{
-//	    callF(src1, dst, op, 0);
-//	//}
+//  //if(src1.channels() == 1 && dst.channels() == 1)
+//  //{
+//  //    callT(src1, dst, op, 0);
+//  //}
+//  //else
+//  //{
+//      callF(src1, dst, op, 0);
+//  //}
 //}
 
 cl_mem bindTexture(const oclMat &mat, int depth, int channels)
@@ -792,6 +792,12 @@ void lkSparse_run(oclMat &I, oclMat &J,
 
 void cv::ocl::PyrLKOpticalFlow::sparse(const oclMat &prevImg, const oclMat &nextImg, const oclMat &prevPts, oclMat &nextPts, oclMat &status, oclMat *err)
 {
+    if (prevImg.clCxt->impl->devName.find("Intel(R) HD Graphics") != string::npos)
+    {
+        cout << " Intel HD GPU device unsupported " << endl;
+        return;
+    }
+
     if (prevPts.empty())
     {
         nextPts.release();

modules/ocl/src/surf.cpp

@@ -197,11 +197,11 @@ public:
             throw std::exception();
             //!FIXME
             // temp fix for missing min overload
-            oclMat temp(mask.size(), mask.type());
-            temp.setTo(Scalar::all(1.0));
-            //cv::ocl::min(mask, temp, surf_.mask1);           ///////// disable this
-            integral(surf_.mask1, surf_.maskSum);
-            bindImgTex(surf_.maskSum, maskSumTex);
+            //oclMat temp(mask.size(), mask.type());
+            //temp.setTo(Scalar::all(1.0));
+            ////cv::ocl::min(mask, temp, surf_.mask1);           ///////// disable this
+            //integral(surf_.mask1, surf_.maskSum);
+            //bindImgTex(surf_.maskSum, maskSumTex);
         }
     }
 
@@ -415,6 +415,11 @@ void cv::ocl::SURF_OCL::operator()(const oclMat &img, const oclMat &mask, oclMat
 {
     if (!img.empty())
     {
+        if (img.clCxt->impl->devName.find("Intel(R) HD Graphics") != string::npos)
+        {
+            cout << " Intel HD GPU device unsupported " << endl;
+            return;
+        }
         SURF_OCL_Invoker surf(*this, img, mask);
 
         surf.detectKeypoints(keypoints);
@@ -426,6 +431,11 @@ void cv::ocl::SURF_OCL::operator()(const oclMat &img, const oclMat &mask, oclMat
 {
     if (!img.empty())
     {
+        if (img.clCxt->impl->devName.find("Intel(R) HD Graphics") != string::npos)
+        {
+            cout << " Intel HD GPU device unsupported " << endl;
+            return;
+        }
         SURF_OCL_Invoker surf(*this, img, mask);
 
         if (!useProvidedKeypoints)

modules/ocl/test/precomp.hpp

@@ -55,13 +55,13 @@
 #include "cvconfig.h"
 #include "opencv2/core/core.hpp"
 #include "opencv2/highgui/highgui.hpp"
-#include "opencv2/calib3d/calib3d.hpp"
+//#include "opencv2/calib3d/calib3d.hpp"
 #include "opencv2/imgproc/imgproc.hpp"
 #include "opencv2/video/video.hpp"
 #include "opencv2/ts/ts.hpp"
 #include "opencv2/ts/ts_perf.hpp"
 #include "opencv2/ocl/ocl.hpp"
-#include "opencv2/nonfree/nonfree.hpp"
+//#include "opencv2/nonfree/nonfree.hpp"
 
 #include "utility.hpp"
 #include "interpolation.hpp"

modules/ocl/test/test_hog.cpp

@@ -199,8 +199,8 @@ TEST_P(HOG, Detect)
 
     int threshold = 10;
     int val = 32;
-    d_comp[0] = d_found.size();
-    comp[0] = found.size();
+    d_comp[0] = (int)d_found.size();
+    comp[0] = (int)found.size();
     if (winSize == cv::Size(48, 96))
     {
         for(int i = 0; i < (int)d_found.size(); i++)

modules/ocl/test/test_imgproc.cpp

@@ -213,8 +213,8 @@ COOR do_meanShift(int x0, int y0, uchar *sptr, uchar *dptr, int sstep, cv::Size
     dptr[3] = (uchar)c3;
 
     COOR coor;
-    coor.x = x0;
-    coor.y = y0;
+    coor.x = (short)x0;
+    coor.y = (short)y0;
     return coor;
 }