Merge pull request #2437 from ilya-lavrenov:tapi_morph

modules/imgproc/src/morph.cpp

@@ -1286,21 +1286,30 @@ static bool IPPMorphOp(int op, InputArray _src, OutputArray _dst,
 
 #ifdef HAVE_OPENCL
 
-static bool ocl_morphology_op(InputArray _src, OutputArray _dst, InputArray _kernel, Size &ksize, const Point anchor, int iterations, int op)
+static bool ocl_morphology_op(InputArray _src, OutputArray _dst, Mat kernel,
+                              const Size & ksize, const Point & anchor, int iterations, int op)
 {
+    CV_Assert(op == MORPH_ERODE || op == MORPH_DILATE);
+
     bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
 
     if (_src.depth() == CV_64F && !doubleSupport)
         return false;
 
     UMat kernel8U;
-    _kernel.getUMat().convertTo(kernel8U, CV_8U);
-    UMat kernel = kernel8U.reshape(1, 1);
+    kernel.convertTo(kernel8U, CV_8U);
+    kernel8U = kernel8U.reshape(1, 1);
 
     bool rectKernel = true;
-    for(int i = 0; i < kernel.rows * kernel.cols; ++i)
-        if(kernel.getMat(ACCESS_READ).at<uchar>(i) != 1)
-            rectKernel = false;
+    {
+        Mat m = kernel.reshape(1, 1);
+        for (int i = 0; i < m.size().area(); ++i)
+            if (m.at<uchar>(i) != 1)
+            {
+                rectKernel = false;
+                break;
+            }
+    }
 
     UMat src = _src.getUMat();
 
@@ -1311,19 +1320,19 @@ static bool ocl_morphology_op(InputArray _src, OutputArray _dst, InputArray _ker
 #endif
     size_t globalThreads[3] = {(src.cols + localThreads[0] - 1) / localThreads[0] *localThreads[0], (src.rows + localThreads[1] - 1) / localThreads[1] *localThreads[1], 1};
 
-    if(localThreads[0]*localThreads[1] * 2 < (localThreads[0] + ksize.width - 1) * (localThreads[1] + ksize.height - 1))
+    if (localThreads[0]*localThreads[1] * 2 < (localThreads[0] + ksize.width - 1) * (localThreads[1] + ksize.height - 1))
         return false;
 
-    char compile_option[128];
-    static const char* op2str[] = {"ERODE", "DILATE"};
-    sprintf(compile_option, "-D RADIUSX=%d -D RADIUSY=%d -D LSIZE0=%d -D LSIZE1=%d -D %s %s %s -D GENTYPE=%s -D DEPTH_%d",
-        anchor.x, anchor.y, (int)localThreads[0], (int)localThreads[1], op2str[op], doubleSupport?"-D DOUBLE_SUPPORT" :"", rectKernel?"-D RECTKERNEL":"",
-        ocl::typeToStr(_src.type()), _src.depth() );
+    static const char * const op2str[] = { "ERODE", "DILATE" };
+    String buildOptions = format("-D RADIUSX=%d -D RADIUSY=%d -D LSIZE0=%d -D LSIZE1=%d -D %s%s%s -D GENTYPE=%s -D DEPTH_%d",
+                                 anchor.x, anchor.y, (int)localThreads[0], (int)localThreads[1], op2str[op],
+                                 doubleSupport ? " -D DOUBLE_SUPPORT" : "", rectKernel ? " -D RECTKERNEL" : "",
+                                 ocl::typeToStr(_src.type()), _src.depth() );
 
     std::vector<ocl::Kernel> kernels;
-    for(int i = 0; i<iterations; i++)
+    for (int i = 0; i<iterations; i++)
     {
-        ocl::Kernel k( "morph", ocl::imgproc::morph_oclsrc, compile_option);
+        ocl::Kernel k("morph", ocl::imgproc::morph_oclsrc, buildOptions);
         if (k.empty())
             return false;
         kernels.push_back(k);
@@ -1346,7 +1355,7 @@ static bool ocl_morphology_op(InputArray _src, OutputArray _dst, InputArray _ker
         idxArg = kernels[0].set(idxArg, ofs.y);
         idxArg = kernels[0].set(idxArg, src.cols);
         idxArg = kernels[0].set(idxArg, src.rows);
-        idxArg = kernels[0].set(idxArg, ocl::KernelArg::PtrReadOnly(kernel));
+        idxArg = kernels[0].set(idxArg, ocl::KernelArg::PtrReadOnly(kernel8U));
         idxArg = kernels[0].set(idxArg, wholecols);
         idxArg = kernels[0].set(idxArg, wholerows);
 
@@ -1387,7 +1396,7 @@ static bool ocl_morphology_op(InputArray _src, OutputArray _dst, InputArray _ker
         idxArg = kernels[i].set(idxArg, ofs.y);
         idxArg = kernels[i].set(idxArg, source.cols);
         idxArg = kernels[i].set(idxArg, source.rows);
-        idxArg = kernels[i].set(idxArg, ocl::KernelArg::PtrReadOnly(kernel));
+        idxArg = kernels[i].set(idxArg, ocl::KernelArg::PtrReadOnly(kernel8U));
         idxArg = kernels[i].set(idxArg, wholecols);
         idxArg = kernels[i].set(idxArg, wholerows);
 
@@ -1413,8 +1422,6 @@ static void morphOp( int op, InputArray _src, OutputArray _dst,
     Size ksize = kernel.data ? kernel.size() : Size(3,3);
     anchor = normalizeAnchor(anchor, ksize);
 
-    CV_Assert( anchor.inside(Rect(0, 0, ksize.width, ksize.height)) );
-
 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
     if( IPPMorphOp(op, _src, _dst, kernel, anchor, iterations, borderType, borderValue) )
         return;
@@ -1443,7 +1450,7 @@ static void morphOp( int op, InputArray _src, OutputArray _dst,
     }
 
     CV_OCL_RUN(_dst.isUMat() && _src.size() == _dst.size() && src_type == dst_type &&
-               _src.dims() <= 2 && (src_cn == 1 || src_cn == 4) && anchor.x == -1 && anchor.y == -1 &&
+               _src.dims() <= 2 && (src_cn == 1 || src_cn == 4) &&
                (src_depth == CV_8U || src_depth == CV_32F || src_depth == CV_64F ) &&
                borderType == cv::BORDER_CONSTANT && borderValue == morphologyDefaultBorderValue() &&
                (op == MORPH_ERODE || op == MORPH_DILATE),