Merge pull request #7047 from tomoaki0705:brushUpFp16Cuda

modules/core/include/opencv2/core/cuda.hpp

@@ -855,7 +855,14 @@ private:
 CV_EXPORTS void printCudaDeviceInfo(int device);
 CV_EXPORTS void printShortCudaDeviceInfo(int device);
 
-CV_EXPORTS void convertFp16Cuda(InputArray _src, OutputArray _dst, Stream& stream = Stream::Null());
+/** @brief Converts an array to half precision floating number.
+
+@param _src input array.
+@param _dst output array.
+@param stream Stream for the asynchronous version.
+@sa convertFp16
+*/
+CV_EXPORTS void convertFp16(InputArray _src, OutputArray _dst, Stream& stream = Stream::Null());
 
 //! @} cudacore_init
 

modules/core/src/cuda/gpu_mat.cu

@@ -594,7 +594,7 @@ void cv::cuda::GpuMat::convertTo(OutputArray _dst, int rtype, double alpha, doub
     funcs[sdepth][ddepth](reshape(1), dst.reshape(1), alpha, beta, stream);
 }
 
-void cv::cuda::convertFp16Cuda(InputArray _src, OutputArray _dst, Stream& stream)
+void cv::cuda::convertFp16(InputArray _src, OutputArray _dst, Stream& stream)
 {
     GpuMat src = _src.getGpuMat();
     int ddepth = 0;

modules/cudev/test/test_cvt.cu

@@ -101,11 +101,11 @@ public:
         GpuMat g_dst;
 
         // Fp32 -> Fp16
-        convertFp16Cuda(g_src, g_dst);
-        convertFp16Cuda(g_dst.clone(), g_dst);
+        cuda::convertFp16(g_src, g_dst);
+        cv::convertFp16(src, dst);
         // Fp16 -> Fp32
-        convertFp16(src, dst);
-        convertFp16(dst, ref);
+        cuda::convertFp16(g_dst.clone(), g_dst);
+        cv::convertFp16(dst, ref);
 
         g_dst.download(dst);
         EXPECT_MAT_NEAR(dst, ref, 0.0);
@@ -127,8 +127,8 @@ public:
         GpuMat g_dst;
 
         // Fp32 -> Fp16
-        convertFp16Cuda(g_src, g_dst);
-        convertFp16(src, ref);
+        cuda::convertFp16(g_src, g_dst);
+        cv::convertFp16(src, ref);
 
         g_dst.download(dst);
         EXPECT_MAT_NEAR(dst, ref, 0.0);