From abcf8d9e610e08227de9cada14868e46a651b8d7 Mon Sep 17 00:00:00 2001
From: Ilya Lavrenov <ilya.lavrenov@itseez.com>
Date: Sun, 29 Dec 2013 18:01:01 +0400
Subject: [PATCH] implemented OpenCL version of cv::convertScaleAbs

---
 modules/core/src/convert.cpp          | 39 +++++++++++++++++++++++++++
 modules/core/src/opencl/arithm.cl     |  9 +++----
 modules/core/test/ocl/test_arithm.cpp | 18 +++++++++++++
 3 files changed, 61 insertions(+), 5 deletions(-)

diff --git a/modules/core/src/convert.cpp b/modules/core/src/convert.cpp
index dba8c7b0c..c2014f1be 100644
--- a/modules/core/src/convert.cpp
+++ b/modules/core/src/convert.cpp
@@ -1266,10 +1266,49 @@ static BinaryFunc getConvertScaleFunc(int sdepth, int ddepth)
     return cvtScaleTab[CV_MAT_DEPTH(ddepth)][CV_MAT_DEPTH(sdepth)];
 }
 
+static bool ocl_convertScaleAbs( InputArray _src, OutputArray _dst, double alpha, double beta )
+{
+    int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
+    bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
+
+    if (!doubleSupport && depth == CV_64F)
+        return false;
+
+    char cvt[2][50];
+    int wdepth = std::max(depth, CV_32F);
+    ocl::Kernel k("KF", ocl::core::arithm_oclsrc,
+                  format("-D OP_CONVERT_SCALE_ABS -D UNARY_OP -D dstT=uchar -D srcT1=%s"
+                         " -D workT=%s -D convertToWT1=%s -D convertToDT=%s%s",
+                         ocl::typeToStr(depth), ocl::typeToStr(wdepth),
+                         ocl::convertTypeStr(depth, wdepth, 1, cvt[0]),
+                         ocl::convertTypeStr(wdepth, CV_8U, 1, cvt[1]),
+                         doubleSupport ? " -D DOUBLE_SUPPORT" : ""));
+    if (k.empty())
+        return false;
+
+    _dst.createSameSize(_src, CV_8UC(cn));
+    UMat src = _src.getUMat(), dst = _dst.getUMat();
+
+    ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
+            dstarg = ocl::KernelArg::WriteOnly(dst, cn);
+
+    if (wdepth == CV_32F)
+        k.args(srcarg, dstarg, (float)alpha, (float)beta);
+    else if (wdepth == CV_64F)
+        k.args(srcarg, dstarg, alpha, beta);
+
+    size_t globalsize[2] = { src.cols * cn, src.rows };
+    return k.run(2, globalsize, NULL, false);
+}
+
 }
 
 void cv::convertScaleAbs( InputArray _src, OutputArray _dst, double alpha, double beta )
 {
+    if (ocl::useOpenCL() && _src.dims() <= 2 && _dst.isUMat() &&
+            ocl_convertScaleAbs(_src, _dst, alpha, beta))
+        return;
+
     Mat src = _src.getMat();
     int cn = src.channels();
     double scale[] = {alpha, beta};
diff --git a/modules/core/src/opencl/arithm.cl b/modules/core/src/opencl/arithm.cl
index 1647e8d19..add4b0695 100644
--- a/modules/core/src/opencl/arithm.cl
+++ b/modules/core/src/opencl/arithm.cl
@@ -223,13 +223,12 @@ dstelem = v > (dstT)(0) ? log(v) : log(-v)
 #define convertToWT2
 #define PROCESS_ELEM dstelem = convert_uchar(srcelem1 CMP_OPERATOR srcelem2 ? 255 : 0)
 
-#elif defined OP_CONVERT
-#define PROCESS_ELEM dstelem = convertToDT(srcelem1)
-
-#elif defined OP_CONVERT_SCALE
+#elif defined OP_CONVERT_SCALE_ABS
 #undef EXTRA_PARAMS
 #define EXTRA_PARAMS , workT alpha, workT beta
-#define PROCESS_ELEM dstelem = convertToDT(srcelem1*alpha + beta)
+#define PROCESS_ELEM \
+    workT value = srcelem1 * alpha + beta; \
+    dstelem = convertToDT(value >= 0 ? value : -value)
 
 #elif defined OP_CTP_AD || defined OP_CTP_AR
 #ifdef OP_CTP_AD
diff --git a/modules/core/test/ocl/test_arithm.cpp b/modules/core/test/ocl/test_arithm.cpp
index 3aa47b7d2..df692b818 100644
--- a/modules/core/test/ocl/test_arithm.cpp
+++ b/modules/core/test/ocl/test_arithm.cpp
@@ -1324,6 +1324,23 @@ OCL_TEST_P(InRange, Scalar)
 }
 
 
+//////////////////////////////// ConvertScaleAbs ////////////////////////////////////////////////
+
+typedef ArithmTestBase ConvertScaleAbs;
+
+OCL_TEST_P(ConvertScaleAbs, Mat)
+{
+    for (int j = 0; j < test_loop_times; j++)
+    {
+        generateTestData();
+
+        OCL_OFF(cv::convertScaleAbs(src1_roi, dst1_roi, val[0], val[1]));
+        OCL_ON(cv::convertScaleAbs(usrc1_roi, udst1_roi, val[0], val[1]));
+
+        Near(depth <= CV_32S ? 1 : 1e-6);
+    }
+}
+
 //////////////////////////////////////// Instantiation /////////////////////////////////////////
 
 OCL_INSTANTIATE_TEST_CASE_P(Arithm, Lut, Combine(::testing::Values(CV_8U, CV_8S), OCL_ALL_DEPTHS, OCL_ALL_CHANNELS, Bool(), Bool()));
@@ -1360,6 +1377,7 @@ OCL_INSTANTIATE_TEST_CASE_P(Arithm, Norm, Combine(OCL_ALL_DEPTHS, OCL_ALL_CHANNE
 OCL_INSTANTIATE_TEST_CASE_P(Arithm, Sqrt, Combine(::testing::Values(CV_32F, CV_64F), OCL_ALL_CHANNELS, Bool()));
 OCL_INSTANTIATE_TEST_CASE_P(Arithm, Normalize, Combine(OCL_ALL_DEPTHS, Values(Channels(1)), Bool()));
 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()));
 
 } } // namespace cvtest::ocl