From 3e1bec52486bab3002e39fd912727b1a85d0a30a Mon Sep 17 00:00:00 2001
From: Ilya Lavrenov <ilya.lavrenov@itseez.com>
Date: Mon, 30 Dec 2013 01:21:04 +0400
Subject: [PATCH] added OpenCL version of cv::patchNaNs

---
 modules/core/src/mathfuncs.cpp        | 23 +++++++++-
 modules/core/src/opencl/arithm.cl     |  7 ++++
 modules/core/test/ocl/test_arithm.cpp | 60 +++++++++++++++++++++++++++
 3 files changed, 88 insertions(+), 2 deletions(-)

diff --git a/modules/core/src/mathfuncs.cpp b/modules/core/src/mathfuncs.cpp
index 0b596071a9..90e0d74a49 100644
--- a/modules/core/src/mathfuncs.cpp
+++ b/modules/core/src/mathfuncs.cpp
@@ -2364,12 +2364,31 @@ bool checkRange(InputArray _src, bool quiet, Point* pt, double minVal, double ma
     return badPt.x < 0;
 }
 
+static bool ocl_patchNaNs( InputOutputArray _a, float value )
+{
+    ocl::Kernel k("KF", ocl::core::arithm_oclsrc,
+                     format("-D UNARY_OP -D OP_PATCH_NANS -D dstT=int"));
+    if (k.empty())
+        return false;
+
+    UMat a = _a.getUMat();
+    int cn = a.channels();
+
+    k.args(ocl::KernelArg::ReadOnlyNoSize(a),
+           ocl::KernelArg::WriteOnly(a), (float)value);
+
+    size_t globalsize[2] = { a.cols * cn, a.rows };
+    return k.run(2, globalsize, NULL, false);
+}
 
 void patchNaNs( InputOutputArray _a, double _val )
 {
-    Mat a = _a.getMat();
-    CV_Assert( a.depth() == CV_32F );
+    CV_Assert( _a.depth() == CV_32F );
 
+    if (ocl::useOpenCL() && _a.isUMat() && _a.dims() <= 2 && ocl_patchNaNs(_a, (float)_val))
+        return;
+
+    Mat a = _a.getMat();
     const Mat* arrays[] = {&a, 0};
     int* ptrs[1];
     NAryMatIterator it(arrays, (uchar**)ptrs);
diff --git a/modules/core/src/opencl/arithm.cl b/modules/core/src/opencl/arithm.cl
index 605fe4785b..c8fd99eeff 100644
--- a/modules/core/src/opencl/arithm.cl
+++ b/modules/core/src/opencl/arithm.cl
@@ -271,6 +271,13 @@ dstelem = v > (dstT)(0) ? log(v) : log(-v)
     dstelem = cos(alpha) * x; \
     dstelem2 = sin(alpha) * x
 
+#elif defined OP_PATCH_NANS
+#undef EXTRA_PARAMS
+#define EXTRA_PARAMS , int val
+#define PROCESS_ELEM \
+    if (( srcelem1 & 0x7fffffff) > 0x7f800000 ) \
+        dstelem = val
+
 #else
 #error "unknown op type"
 #endif
diff --git a/modules/core/test/ocl/test_arithm.cpp b/modules/core/test/ocl/test_arithm.cpp
index f2b9875143..03d8422182 100644
--- a/modules/core/test/ocl/test_arithm.cpp
+++ b/modules/core/test/ocl/test_arithm.cpp
@@ -42,6 +42,8 @@
 #include "test_precomp.hpp"
 #include "opencv2/ts/ocl_test.hpp"
 
+#include <cmath>
+
 #ifdef HAVE_OPENCL
 
 namespace cvtest {
@@ -1357,6 +1359,63 @@ OCL_TEST_P(ScaleAdd, Mat)
     }
 }
 
+//////////////////////////////// PatchNans ////////////////////////////////////////////////
+
+PARAM_TEST_CASE(PatchNaNs, Channels, bool)
+{
+    int cn;
+    bool use_roi;
+    double value;
+
+    TEST_DECLARE_INPUT_PARAMETER(src)
+
+    virtual void SetUp()
+    {
+        cn = GET_PARAM(0);
+        use_roi = GET_PARAM(1);
+    }
+
+    virtual void generateTestData()
+    {
+        const int type = CV_MAKE_TYPE(CV_32F, cn);
+
+        Size roiSize = randomSize(1, 10);
+        Border srcBorder = randomBorder(0, use_roi ? MAX_VALUE : 0);
+        randomSubMat(src, src_roi, roiSize, srcBorder, type, -40, 40);
+
+        // generating NaNs
+        roiSize.width *= cn;
+        for (int y = 0; y < roiSize.height; ++y)
+        {
+            float * const ptr = src_roi.ptr<float>(y);
+            for (int x = 0; x < roiSize.width; ++x)
+                ptr[x] = randomInt(-1, 1) == 0 ? std::numeric_limits<float>::quiet_NaN() : ptr[x];
+        }
+
+        value = randomDouble(-100, 100);
+
+        UMAT_UPLOAD_INPUT_PARAMETER(src)
+    }
+
+    void Near()
+    {
+        OCL_EXPECT_MATS_NEAR(src, 0)
+    }
+};
+
+OCL_TEST_P(PatchNaNs, Mat)
+{
+    for (int j = 0; j < test_loop_times; j++)
+    {
+        generateTestData();
+
+        OCL_OFF(cv::patchNaNs(src_roi, value));
+        OCL_ON(cv::patchNaNs(usrc_roi, value));
+
+        Near();
+    }
+}
+
 //////////////////////////////////////// Instantiation /////////////////////////////////////////
 
 OCL_INSTANTIATE_TEST_CASE_P(Arithm, Lut, Combine(::testing::Values(CV_8U, CV_8S), OCL_ALL_DEPTHS, OCL_ALL_CHANNELS, Bool(), Bool()));
@@ -1395,6 +1454,7 @@ OCL_INSTANTIATE_TEST_CASE_P(Arithm, Normalize, Combine(OCL_ALL_DEPTHS, Values(Ch
 OCL_INSTANTIATE_TEST_CASE_P(Arithm, InRange, Combine(OCL_ALL_DEPTHS, OCL_ALL_CHANNELS, Bool(), Bool()));
 OCL_INSTANTIATE_TEST_CASE_P(Arithm, ConvertScaleAbs, Combine(OCL_ALL_DEPTHS, OCL_ALL_CHANNELS, Bool()));
 OCL_INSTANTIATE_TEST_CASE_P(Arithm, ScaleAdd, Combine(OCL_ALL_DEPTHS, OCL_ALL_CHANNELS, Bool()));
+OCL_INSTANTIATE_TEST_CASE_P(Arithm, PatchNaNs, Combine(OCL_ALL_CHANNELS, Bool()));
 
 } } // namespace cvtest::ocl