Tegra optimization for image filtering

modules/imgproc/perf/perf_filter2d.cpp

+#include "perf_precomp.hpp"
+
+using namespace std;
+using namespace cv;
+using namespace perf;
+using namespace testing;
+using std::tr1::make_tuple;
+using std::tr1::get;
+
+
+CV_ENUM(BorderMode, BORDER_CONSTANT, BORDER_REPLICATE, BORDER_REFLECT_101);
+
+typedef TestBaseWithParam< tr1::tuple<Size, int, BorderMode> > TestFilter2d;
+
+
+PERF_TEST_P( TestFilter2d, Filter2d,
+             Combine(
+                Values( Size(320, 240), szVGA, sz720p, sz1080p ),
+                Values( 3, 5 ),
+                ValuesIn( BorderMode::all() )
+             )
+)
+{
+    Size sz;
+    int borderMode, kSize;
+    sz         = get<0>(GetParam());
+    kSize      = get<1>(GetParam());
+    borderMode = get<2>(GetParam());
+
+    Mat src(sz, CV_8UC4);
+    Mat dst(sz, CV_8UC4);
+
+    Mat kernel(kSize, kSize, CV_32FC1);
+    randu(kernel, -3, 10);
+    float s = fabs( sum(kernel)[0] );
+    if(s > 1e-3) kernel /= s;
+
+    declare.in(src, WARMUP_RNG).out(dst).time(20);
+
+    TEST_CYCLE() filter2D(src, dst, CV_8UC4, kernel, Point(1, 1), 0., borderMode);
+
+    SANITY_CHECK(dst);
+}
+
+

modules/imgproc/src/filter.cpp

@@ -3010,6 +3010,11 @@ void cv::filter2D( InputArray _src, OutputArray _dst, int ddepth,
     Mat dst = _dst.getMat();
     anchor = normalizeAnchor(anchor, kernel.size());
 
+#ifdef HAVE_TEGRA_OPTIMIZATION
+    if( tegra::filter2D(src, dst, kernel, anchor, delta, borderType) )
+        return;
+#endif
+
     if( kernel.cols*kernel.rows >= dft_filter_size )
     {
         Mat temp;

modules/imgproc/src/smooth.cpp

@@ -311,6 +311,12 @@ void cv::boxFilter( InputArray _src, OutputArray _dst, int ddepth,
         if( src.cols == 1 )
             ksize.width = 1;
     }
+
+#ifdef HAVE_TEGRA_OPTIMIZATION
+    if ( tegra::boxFilter(src, dst, ksize, anchor, normalize, borderType) )
+        return;
+#endif
+
     Ptr<FilterEngine> f = createBoxFilter( src.type(), dst.type(),
                         ksize, anchor, normalize, borderType );
     f->apply( src, dst );