Make Integral sum support cv_32f, sqsum support cv_64f.

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

@@ -861,10 +861,10 @@ namespace cv
         CV_EXPORTS void warpPerspective(const oclMat &src, oclMat &dst, const Mat &M, Size dsize, int flags = INTER_LINEAR);
 
         //! computes the integral image and integral for the squared image
-        // sum will have CV_32S type, sqsum - CV32F type
+        // sum will support CV_32S, CV_32F, sqsum - support CV32F, CV_64F
         // supports only CV_8UC1 source type
-        CV_EXPORTS void integral(const oclMat &src, oclMat &sum, oclMat &sqsum);
-        CV_EXPORTS void integral(const oclMat &src, oclMat &sum);
+        CV_EXPORTS void integral(const oclMat &src, oclMat &sum, oclMat &sqsum, int sdepth=-1 );
+        CV_EXPORTS void integral(const oclMat &src, oclMat &sum, int sdepth=-1 );
         CV_EXPORTS void cornerHarris(const oclMat &src, oclMat &dst, int blockSize, int ksize, double k, int bordertype = cv::BORDER_DEFAULT);
         CV_EXPORTS void cornerHarris_dxdy(const oclMat &src, oclMat &dst, oclMat &Dx, oclMat &Dy,
             int blockSize, int ksize, double k, int bordertype = cv::BORDER_DEFAULT);

modules/ocl/src/imgproc.cpp

@@ -783,7 +783,7 @@ namespace cv
         ////////////////////////////////////////////////////////////////////////
         // integral
 
-        void integral(const oclMat &src, oclMat &sum, oclMat &sqsum)
+        void integral(const oclMat &src, oclMat &sum, oclMat &sqsum, int sdepth)
         {
             CV_Assert(src.type() == CV_8UC1);
             if (!src.clCxt->supportsFeature(ocl::FEATURE_CL_DOUBLE) && src.depth() == CV_64F)
@@ -792,6 +792,12 @@ namespace cv
                 return;
             }
 
+            int depth = src.depth();
+            if( sdepth <= 0 )
+                sdepth = CV_32S;
+            sdepth = CV_MAT_DEPTH(sdepth);
+            int type = CV_MAKE_TYPE(sdepth, 1);
+
             int vlen = 4;
             int offset = src.offset / vlen;
             int pre_invalid = src.offset % vlen;
@@ -799,17 +805,26 @@ namespace cv
 
             oclMat t_sum , t_sqsum;
             int w = src.cols + 1, h = src.rows + 1;
-            int depth = src.depth() == CV_8U ? CV_32S : CV_64F;
-            int type = CV_MAKE_TYPE(depth, 1);
+
+            char build_option[250];
+            if(Context::getContext()->supportsFeature(ocl::FEATURE_CL_DOUBLE))
+            {
+                t_sqsum.create(src.cols, src.rows, CV_64FC1);
+                sqsum.create(h, w, CV_64FC1);
+                sprintf(build_option, "-D TYPE=double -D TYPE4=double4 -D convert_TYPE4=convert_double4");
+            }
+            else
+            {
+                t_sqsum.create(src.cols, src.rows, CV_32FC1);
+                sqsum.create(h, w, CV_32FC1);
+                sprintf(build_option, "-D TYPE=float -D TYPE4=float4 -D convert_TYPE4=convert_float4");
+            }
 
             t_sum.create(src.cols, src.rows, type);
             sum.create(h, w, type);
 
-            t_sqsum.create(src.cols, src.rows, CV_32FC1);
-            sqsum.create(h, w, CV_32FC1);
-
-            int sum_offset = sum.offset / vlen;
-            int sqsum_offset = sqsum.offset / vlen;
+            int sum_offset = sum.offset / sum.elemSize();
+            int sqsum_offset = sqsum.offset / sqsum.elemSize();
 
             vector<pair<size_t , const void *> > args;
             args.push_back( make_pair( sizeof(cl_mem) , (void *)&src.data ));
@@ -821,8 +836,9 @@ namespace cv
             args.push_back( make_pair( sizeof(cl_int) , (void *)&src.cols ));
             args.push_back( make_pair( sizeof(cl_int) , (void *)&src.step ));
             args.push_back( make_pair( sizeof(cl_int) , (void *)&t_sum.step));
+            args.push_back( make_pair( sizeof(cl_int) , (void *)&t_sqsum.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, depth);
+            openCLExecuteKernel(src.clCxt, &imgproc_integral, "integral_cols", gt, lt, args, -1, sdepth, build_option);
 
             args.clear();
             args.push_back( make_pair( sizeof(cl_mem) , (void *)&t_sum.data ));
@@ -832,15 +848,16 @@ namespace cv
             args.push_back( make_pair( sizeof(cl_int) , (void *)&t_sum.rows ));
             args.push_back( make_pair( sizeof(cl_int) , (void *)&t_sum.cols ));
             args.push_back( make_pair( sizeof(cl_int) , (void *)&t_sum.step ));
+            args.push_back( make_pair( sizeof(cl_int) , (void *)&t_sqsum.step));
             args.push_back( make_pair( sizeof(cl_int) , (void *)&sum.step));
             args.push_back( make_pair( sizeof(cl_int) , (void *)&sqsum.step));
             args.push_back( make_pair( sizeof(cl_int) , (void *)&sum_offset));
             args.push_back( 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, depth);
+            openCLExecuteKernel(src.clCxt, &imgproc_integral, "integral_rows", gt2, lt2, args, -1, sdepth, build_option);
         }
 
-        void integral(const oclMat &src, oclMat &sum)
+        void integral(const oclMat &src, oclMat &sum, int sdepth)
         {
             CV_Assert(src.type() == CV_8UC1);
             int vlen = 4;
@@ -848,10 +865,13 @@ namespace cv
             int pre_invalid = src.offset % vlen;
             int vcols = (pre_invalid + src.cols + vlen - 1) / vlen;
 
+            if( sdepth <= 0 )
+                sdepth = CV_32S;
+            sdepth = CV_MAT_DEPTH(sdepth);
+            int type = CV_MAKE_TYPE(sdepth, 1);
+
             oclMat t_sum;
             int w = src.cols + 1, h = src.rows + 1;
-            int depth = src.depth() == CV_8U ? CV_32S : CV_32F;
-            int type = CV_MAKE_TYPE(depth, 1);
 
             t_sum.create(src.cols, src.rows, type);
             sum.create(h, w, type);
@@ -867,7 +887,7 @@ namespace cv
             args.push_back( make_pair( sizeof(cl_int) , (void *)&src.step ));
             args.push_back( 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, depth);
+            openCLExecuteKernel(src.clCxt, &imgproc_integral_sum, "integral_sum_cols", gt, lt, args, -1, sdepth);
 
             args.clear();
             args.push_back( make_pair( sizeof(cl_mem) , (void *)&t_sum.data ));
@@ -878,7 +898,7 @@ namespace cv
             args.push_back( make_pair( sizeof(cl_int) , (void *)&sum.step));
             args.push_back( 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, depth);
+            openCLExecuteKernel(src.clCxt, &imgproc_integral_sum, "integral_sum_rows", gt2, lt2, args, -1, sdepth);
         }
 
         /////////////////////// corner //////////////////////////////

modules/ocl/test/test_imgproc.cpp

@@ -275,23 +275,33 @@ OCL_TEST_P(CornerHarris, Mat)
 
 //////////////////////////////////integral/////////////////////////////////////////////////
 
-typedef ImgprocTestBase Integral;
+struct Integral :
+        public ImgprocTestBase
+{
+    int sdepth;
 
+    virtual void SetUp()
+    {
+        type = GET_PARAM(0);
+        blockSize = GET_PARAM(1);
+        sdepth = GET_PARAM(2);
+        useRoi = GET_PARAM(3);
+    }
+};
 OCL_TEST_P(Integral, Mat1)
 {
     for (int j = 0; j < LOOP_TIMES; j++)
     {
         random_roi();
 
-        ocl::integral(gsrc_roi, gdst_roi);
-        integral(src_roi, dst_roi);
+        ocl::integral(gsrc_roi, gdst_roi, sdepth);
+        integral(src_roi, dst_roi, sdepth);
 
         Near();
     }
 }
 
-// TODO wrong output type
-OCL_TEST_P(Integral, DISABLED_Mat2)
+OCL_TEST_P(Integral, Mat2)
 {
     Mat dst1;
     ocl::oclMat gdst1;
@@ -300,10 +310,12 @@ OCL_TEST_P(Integral, DISABLED_Mat2)
     {
         random_roi();
 
-        integral(src_roi, dst1, dst_roi);
-        ocl::integral(gsrc_roi, gdst1, gdst_roi);
+        integral(src_roi, dst_roi, dst1, sdepth);
+        ocl::integral(gsrc_roi, gdst_roi, gdst1, sdepth);
 
         Near();
+        if(gdst1.clCxt->supportsFeature(ocl::FEATURE_CL_DOUBLE))
+            EXPECT_MAT_NEAR(dst1, Mat(gdst1), 0.);
     }
 }
 
@@ -543,7 +555,7 @@ INSTANTIATE_TEST_CASE_P(Imgproc, CornerHarris, Combine(
 INSTANTIATE_TEST_CASE_P(Imgproc, Integral, Combine(
                             Values((MatType)CV_8UC1), // TODO does not work with CV_32F, CV_64F
                             Values(0), // not used
-                            Values(0), // not used
+                            Values((MatType)CV_32SC1, (MatType)CV_32FC1),
                             Bool()));
 
 INSTANTIATE_TEST_CASE_P(Imgproc, Threshold, Combine(