potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 15:23:05 +04:00
Code Packages Releases Wiki Activity

Changes for fp16 conversion

Browse Source
This commit is contained in:
Om Navin Rope
2026-05-20 14:35:52 +05:30
parent cc8f683b6a
commit 1e731a2e2a
3 changed files with 2 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/cuda.hpp
-9
View File
@@ -1327,15 +1327,6 @@ private:
CV_EXPORTS_W void printCudaDeviceInfo(int device);
CV_EXPORTS_W void printShortCudaDeviceInfo(int device);
/** @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 cv::Mat::convertTo
*/
CV_EXPORTS void convertFp16(InputArray _src, OutputArray _dst, Stream& stream = Stream::Null());
//! @} cudacore_init
}} // namespace cv { namespace cuda {
modules/core/src/cuda/gpu_mat.cu
+2 -36
View File
@@ -565,9 +565,9 @@ void cv::cuda::GpuMat::convertTo(OutputArray _dst, int rtype, Stream& stream) co
{convertToNoScale<ushort, uchar>, convertToNoScale<ushort, schar>, 0, convertToNoScale<ushort, short>, convertToNoScale<ushort, int>, convertToNoScale<ushort, float>, convertToNoScale<ushort, double>, 0, 0, 0, convertToNoScale<ushort, uint64_t>, convertToNoScale<ushort, int64_t>, convertToNoScale<ushort, uint32_t>},
{convertToNoScale<short, uchar>, convertToNoScale<short, schar>, convertToNoScale<short, ushort>, 0, convertToNoScale<short, int>, convertToNoScale<short, float>, convertToNoScale<short, double>, 0, 0, 0, convertToNoScale<short, uint64_t>, convertToNoScale<short, int64_t>, convertToNoScale<short, uint32_t>},
{convertToNoScale<int, uchar>, convertToNoScale<int, schar>, convertToNoScale<int, ushort>, convertToNoScale<int, short>, 0, convertToNoScale<int, float>, convertToNoScale<int, double>, 0, 0, 0, convertToNoScale<int, uint64_t>, convertToNoScale<int, int64_t>, convertToNoScale<int, uint32_t>},
{convertToNoScale<float, uchar>, convertToNoScale<float, schar>, convertToNoScale<float, ushort>, convertToNoScale<float, short>, convertToNoScale<float, int>, 0, convertToNoScale<float, double>, 0, 0, 0, convertToNoScale<float, uint64_t>, convertToNoScale<float, int64_t>, convertToNoScale<float, uint32_t>},
{convertToNoScale<float, uchar>, convertToNoScale<float, schar>, convertToNoScale<float, ushort>, convertToNoScale<float, short>, convertToNoScale<float, int>, 0, convertToNoScale<float, double>, convertScaleHalf<float, short>, 0, 0, convertToNoScale<float, uint64_t>, convertToNoScale<float, int64_t>, convertToNoScale<float, uint32_t>},
{convertToNoScale<double, uchar>, convertToNoScale<double, schar>, convertToNoScale<double, ushort>, convertToNoScale<double, short>, convertToNoScale<double, int>, convertToNoScale<double, float>, 0, 0, 0, 0, convertToNoScale<double, uint64_t>, convertToNoScale<double, int64_t>, convertToNoScale<double, uint32_t>},
{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
{0, 0, 0, 0, 0, convertScaleHalf<short, float>, 0, 0, 0, 0, 0, 0, 0},
{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
{convertToNoScale<uint64_t, uchar>, convertToNoScale<uint64_t, schar>, convertToNoScale<uint64_t, ushort>, convertToNoScale<uint64_t, short>, convertToNoScale<uint64_t, int>, convertToNoScale<uint64_t, float>, convertToNoScale<uint64_t, double>, 0, 0, 0, 0, convertToNoScale<uint64_t, int64_t>, convertToNoScale<uint64_t, uint32_t>},
@@ -619,38 +619,4 @@ void cv::cuda::GpuMat::convertTo(OutputArray _dst, int rtype, double alpha, doub
func(src.reshape(1), dst.reshape(1), alpha, beta, stream);
}
void cv::cuda::convertFp16(InputArray _src, OutputArray _dst, Stream& stream)
{
GpuMat src = _src.getGpuMat();
int ddepth = 0;
switch(src.depth())
{
case CV_32F:
ddepth = CV_16F;
break;
case CV_16F:
ddepth = CV_32F;
break;
default:
CV_Error(Error::StsUnsupportedFormat, "Unsupported input depth");
return;
}
int type = CV_MAKE_TYPE(CV_MAT_DEPTH(ddepth), src.channels());
_dst.create(src.size(), type);
GpuMat dst = _dst.getGpuMat();
typedef void (*func_t)(const GpuMat& src, const GpuMat& dst, Stream& stream);
static const func_t funcs[] =
{
0, 0, 0,
0, 0, convertScaleHalf<short, float>,
0, convertScaleHalf<float, short>, 0,
};
func_t func = funcs[ddepth];
CV_Assert(func);
func(src.reshape(1), dst.reshape(1), stream);
}
#endif
modules/core/test/test_io.cpp
-1
View File
@@ -1977,7 +1977,6 @@ static void test_20279(FileStorage& fs)
m32fc1.at<float>((int)i) = v * 0.5f;
}
Mat m16fc1;
// produces CV_16S output: convertFp16(m32fc1, m16fc1);
m32fc1.convertTo(m16fc1, CV_16FC1);
EXPECT_EQ(CV_16FC1, m16fc1.type()) << typeToString(m16fc1.type());
//std::cout << m16fc1 << std::endl;