/* /*//*/ */
int errors = 0;
- bad_caller = caller;
- bad_caller.objPts = 0;
- errors += run_test_case( CV_StsBadArg, "Zero passed in objPts", bad_caller);
-
- bad_caller = caller;
- bad_caller.imgPts = 0;
- errors += run_test_case( CV_StsBadArg, "Zero passed in imgPts", bad_caller );
-
- bad_caller = caller;
- bad_caller.npoints = 0;
- errors += run_test_case( CV_StsBadArg, "Zero passed in npoints", bad_caller );
-
- bad_caller = caller;
- bad_caller.cameraMatrix = 0;
- errors += run_test_case( CV_StsBadArg, "Zero passed in cameraMatrix", bad_caller );
-
- bad_caller = caller;
- bad_caller.distCoeffs = 0;
- errors += run_test_case( CV_StsBadArg, "Zero passed in distCoeffs", bad_caller );
-
- bad_caller = caller;
- bad_caller.imageSize.width = -1;
- errors += run_test_case( CV_StsOutOfRange, "Bad image width", bad_caller );
-
- bad_caller = caller;
- bad_caller.imageSize.height = -1;
- errors += run_test_case( CV_StsOutOfRange, "Bad image height", bad_caller );
-
- Mat bad_nts_cpp1 = Mat_<float>(M, 1, 1.f);
- Mat bad_nts_cpp2 = Mat_<int>(3, 3, corSize.width * corSize.height);
- CvMat bad_npts_c1 = cvMat(bad_nts_cpp1);
- CvMat bad_npts_c2 = cvMat(bad_nts_cpp2);
-
- bad_caller = caller;
- bad_caller.npoints = &bad_npts_c1;
- errors += run_test_case( CV_StsUnsupportedFormat, "Bad npoints format", bad_caller );
-
- bad_caller = caller;
- bad_caller.npoints = &bad_npts_c2;
- errors += run_test_case( CV_StsUnsupportedFormat, "Bad npoints size", bad_caller );
-
- bad_caller = caller;
- bad_caller.rvecs = (CvMat*)zeros.ptr();
- errors += run_test_case( CV_StsBadArg, "Bad rvecs header", bad_caller );
-
- bad_caller = caller;
- bad_caller.tvecs = (CvMat*)zeros.ptr();
- errors += run_test_case( CV_StsBadArg, "Bad tvecs header", bad_caller );
-
- Mat bad_rvecs_cpp1(M+1, 1, CV_32FC3); CvMat bad_rvecs_c1 = cvMat(bad_rvecs_cpp1);
- Mat bad_tvecs_cpp1(M+1, 1, CV_32FC3); CvMat bad_tvecs_c1 = cvMat(bad_tvecs_cpp1);
-
-
-
- Mat bad_rvecs_cpp2(M, 2, CV_32FC3); CvMat bad_rvecs_c2 = cvMat(bad_rvecs_cpp2);
- Mat bad_tvecs_cpp2(M, 2, CV_32FC3); CvMat bad_tvecs_c2 = cvMat(bad_tvecs_cpp2);
-
- bad_caller = caller;
- bad_caller.rvecs = &bad_rvecs_c1;
- errors += run_test_case( CV_StsBadArg, "Bad tvecs header", bad_caller );
-
- bad_caller = caller;
- bad_caller.rvecs = &bad_rvecs_c2;
- errors += run_test_case( CV_StsBadArg, "Bad tvecs header", bad_caller );
-
- bad_caller = caller;
- bad_caller.tvecs = &bad_tvecs_c1;
- errors += run_test_case( CV_StsBadArg, "Bad tvecs header", bad_caller );
-
- bad_caller = caller;
- bad_caller.tvecs = &bad_tvecs_c2;
- errors += run_test_case( CV_StsBadArg, "Bad tvecs header", bad_caller );
-
- Mat bad_cameraMatrix_cpp1(3, 3, CV_32S); CvMat bad_cameraMatrix_c1 = cvMat(bad_cameraMatrix_cpp1);
- Mat bad_cameraMatrix_cpp2(2, 3, CV_32F); CvMat bad_cameraMatrix_c2 = cvMat(bad_cameraMatrix_cpp2);
- Mat bad_cameraMatrix_cpp3(3, 2, CV_64F); CvMat bad_cameraMatrix_c3 = cvMat(bad_cameraMatrix_cpp3);
-
-
-
- bad_caller = caller;
- bad_caller.cameraMatrix = &bad_cameraMatrix_c1;
- errors += run_test_case( CV_StsBadArg, "Bad camearaMatrix header", bad_caller );
+ caller.initArgs();
+ caller.objPts_arg = noArray();
+ errors += run_test_case( CV_StsBadArg, "None passed in objPts", caller);
- bad_caller = caller;
- bad_caller.cameraMatrix = &bad_cameraMatrix_c2;
- errors += run_test_case( CV_StsBadArg, "Bad camearaMatrix header", bad_caller );
+ caller.initArgs();
+ caller.imgPts_arg = noArray();
+ errors += run_test_case( CV_StsBadArg, "None passed in imgPts", caller );
- bad_caller = caller;
- bad_caller.cameraMatrix = &bad_cameraMatrix_c3;
- errors += run_test_case( CV_StsBadArg, "Bad camearaMatrix header", bad_caller );
+ caller.initArgs();
+ caller.cameraMatrix_arg = noArray();
+ errors += run_test_case( CV_StsBadArg, "Zero passed in cameraMatrix", caller );
- Mat bad_distCoeffs_cpp1(1, 5, CV_32S); CvMat bad_distCoeffs_c1 = cvMat(bad_distCoeffs_cpp1);
- Mat bad_distCoeffs_cpp2(2, 2, CV_64F); CvMat bad_distCoeffs_c2 = cvMat(bad_distCoeffs_cpp2);
- Mat bad_distCoeffs_cpp3(1, 6, CV_64F); CvMat bad_distCoeffs_c3 = cvMat(bad_distCoeffs_cpp3);
+ caller.initArgs();
+ caller.distCoeffs_arg = noArray();
+ errors += run_test_case( CV_StsBadArg, "Zero passed in distCoeffs", caller );
+ caller.initArgs();
+ caller.imageSize.width = -1;
+ errors += run_test_case( CV_StsOutOfRange, "Bad image width", caller );
+ caller.initArgs();
+ caller.imageSize.height = -1;
+ errors += run_test_case( CV_StsOutOfRange, "Bad image height", caller );
- bad_caller = caller;
- bad_caller.distCoeffs = &bad_distCoeffs_c1;
- errors += run_test_case( CV_StsBadArg, "Bad distCoeffs header", bad_caller );
+ caller.initArgs();
+ caller.imgPts[0].clear();
- errors += run_test_case( CV_StsUnsupportedFormat, "Bad imgpts[0]", caller );
++ errors += run_test_case( CV_StsBadSize, "Bad imgpts[0]", caller );
+ caller.imgPts[0] = caller.imgPts[1];
- bad_caller = caller;
- bad_caller.distCoeffs = &bad_distCoeffs_c2;
- errors += run_test_case( CV_StsBadArg, "Bad distCoeffs header", bad_caller );
+ caller.initArgs();
+ caller.objPts[1].clear();
- errors += run_test_case( CV_StsUnsupportedFormat, "Bad objpts[1]", caller );
++ errors += run_test_case( CV_StsBadSize, "Bad objpts[1]", caller );
+ caller.objPts[1] = caller.objPts[0];
+ caller.initArgs();
+ Mat badCM = Mat::zeros(4, 4, CV_64F);
+ caller.cameraMatrix_arg = badCM;
+ caller.flags = CALIB_USE_INTRINSIC_GUESS;
+ errors += run_test_case( CV_StsBadArg, "Bad camearaMatrix header", caller );
- bad_caller = caller;
- bad_caller.distCoeffs = &bad_distCoeffs_c3;
- errors += run_test_case( CV_StsBadArg, "Bad distCoeffs header", bad_caller );
-
- double CM[] = {0, 0, 0, /**/0, 0, 0, /**/0, 0, 0};
- Mat bad_cameraMatrix_cpp4(3, 3, CV_64F, CM); CvMat bad_cameraMatrix_c4 = cvMat(bad_cameraMatrix_cpp4);
-
- bad_caller = caller;
- bad_caller.flags |= CV_CALIB_USE_INTRINSIC_GUESS;
- bad_caller.cameraMatrix = &bad_cameraMatrix_c4;
- CM[0] = 0; //bad fx
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[0] = 500; CM[4] = 0; //bad fy
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[0] = 500; CM[4] = 500; CM[2] = -1; //bad cx
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[0] = 500; CM[4] = 500; CM[2] = imgSize.width*2; //bad cx
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[0] = 500; CM[4] = 500; CM[2] = imgSize.width/2; CM[5] = -1; //bad cy
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[0] = 500; CM[4] = 500; CM[2] = imgSize.width/2; CM[5] = imgSize.height*2; //bad cy
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[0] = 500; CM[4] = 500; CM[2] = imgSize.width/2; CM[5] = imgSize.height/2;
- CM[1] = 0.1; //Non-zero skew
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[1] = 0;
- CM[3] = 0.1; /* mad matrix shape */
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[3] = 0; CM[6] = 0.1; /* mad matrix shape */
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[3] = 0; CM[6] = 0; CM[7] = 0.1; /* mad matrix shape */
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
-
- CM[3] = 0; CM[6] = 0; CM[7] = 0; CM[8] = 1.1; /* mad matrix shape */
- errors += run_test_case( CV_StsOutOfRange, "Bad camearaMatrix data", bad_caller );
- CM[8] = 1.0;
-
- /////////////////////////////////////////////////////////////////////////////////////
- bad_caller = caller;
- Mat bad_objPts_cpp5 = objPts_cpp.clone(); CvMat bad_objPts_c5 = cvMat(bad_objPts_cpp5);
- bad_caller.objPts = &bad_objPts_c5;
-
- cv::RNG& rng = theRNG();
- for(int i = 0; i < bad_objPts_cpp5.rows; ++i)
- bad_objPts_cpp5.at<Point3f>(0, i).z += ((float)rng - 0.5f);
-
- errors += run_test_case( CV_StsBadArg, "Bad objPts data", bad_caller );
+ caller.initArgs();
+ Mat badDC = Mat::zeros(10, 10, CV_64F);
+ caller.distCoeffs_arg = badDC;
+ caller.flags = CALIB_USE_INTRINSIC_GUESS;
+ errors += run_test_case( CV_StsBadArg, "Bad camearaMatrix header", caller );
if (errors)
ts->set_failed_test_info(cvtest::TS::FAIL_MISMATCH);
--- /dev/null
-static BinaryFunc get_cvt32f16f()
-{
- CV_INSTRUMENT_REGION();
- CV_CPU_DISPATCH(get_cvt32f16f, (),
- CV_CPU_DISPATCH_MODES_ALL);
-}
-static BinaryFunc get_cvt16f32f()
-{
- CV_INSTRUMENT_REGION();
- CV_CPU_DISPATCH(get_cvt16f32f, (),
- CV_CPU_DISPATCH_MODES_ALL);
-}
-
+ // This file is part of OpenCV project.
+ // It is subject to the license terms in the LICENSE file found in the top-level directory
+ // of this distribution and at http://opencv.org/license.html
+
+ #include "precomp.hpp"
+ #include "opencl_kernels_core.hpp"
+
+ #include "convert.simd.hpp"
+ #include "convert.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
+
+ namespace cv {
+
++namespace hal {
++void cvt16f32f(const float16_t* src, float* dst, int len)
++{
++ CV_INSTRUMENT_REGION();
++ CV_CPU_DISPATCH(cvt16f32f, (src, dst, len),
++ CV_CPU_DISPATCH_MODES_ALL);
++}
++void cvt32f16f(const float* src, float16_t* dst, int len)
++{
++ CV_INSTRUMENT_REGION();
++ CV_CPU_DISPATCH(cvt32f16f, (src, dst, len),
++ CV_CPU_DISPATCH_MODES_ALL);
++}
++void addRNGBias32f(float* arr, const float* scaleBiasPairs, int len)
++{
++ CV_INSTRUMENT_REGION();
++ CV_CPU_DISPATCH(addRNGBias32f, (arr, scaleBiasPairs, len),
++ CV_CPU_DISPATCH_MODES_ALL);
++}
++void addRNGBias64f(double* arr, const double* scaleBiasPairs, int len)
++{
++ CV_INSTRUMENT_REGION();
++ CV_CPU_DISPATCH(addRNGBias64f, (arr, scaleBiasPairs, len),
++ CV_CPU_DISPATCH_MODES_ALL);
++}
++
++} // namespace
++
++
+ /* [TODO] Recover IPP calls
+ #if defined(HAVE_IPP)
+ #define DEF_CVT_FUNC_F(suffix, stype, dtype, ippFavor) \
+ static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
+ dtype* dst, size_t dstep, Size size, double*) \
+ { \
+ CV_IPP_RUN(src && dst, CV_INSTRUMENT_FUN_IPP(ippiConvert_##ippFavor, src, (int)sstep, dst, (int)dstep, ippiSize(size.width, size.height)) >= 0) \
+ cvt_(src, sstep, dst, dstep, size); \
+ }
+
+ #define DEF_CVT_FUNC_F2(suffix, stype, dtype, ippFavor) \
+ static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
+ dtype* dst, size_t dstep, Size size, double*) \
+ { \
+ CV_IPP_RUN(src && dst, CV_INSTRUMENT_FUN_IPP(ippiConvert_##ippFavor, src, (int)sstep, dst, (int)dstep, ippiSize(size.width, size.height), ippRndFinancial, 0) >= 0) \
+ cvt_(src, sstep, dst, dstep, size); \
+ }
+ #else
+ #define DEF_CVT_FUNC_F(suffix, stype, dtype, ippFavor) \
+ static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
+ dtype* dst, size_t dstep, Size size, double*) \
+ { \
+ cvt_(src, sstep, dst, dstep, size); \
+ }
+ #define DEF_CVT_FUNC_F2 DEF_CVT_FUNC_F
+ #endif
+
+ #define DEF_CVT_FUNC(suffix, stype, dtype) \
+ static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
+ dtype* dst, size_t dstep, Size size, double*) \
+ { \
+ cvt_(src, sstep, dst, dstep, size); \
+ }
+
+ #define DEF_CPY_FUNC(suffix, stype) \
+ static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
+ stype* dst, size_t dstep, Size size, double*) \
+ { \
+ cpy_(src, sstep, dst, dstep, size); \
+ }
+
+ DEF_CPY_FUNC(8u, uchar)
+ DEF_CVT_FUNC_F(8s8u, schar, uchar, 8s8u_C1Rs)
+ DEF_CVT_FUNC_F(16u8u, ushort, uchar, 16u8u_C1R)
+ DEF_CVT_FUNC_F(16s8u, short, uchar, 16s8u_C1R)
+ DEF_CVT_FUNC_F(32s8u, int, uchar, 32s8u_C1R)
+ DEF_CVT_FUNC_F2(32f8u, float, uchar, 32f8u_C1RSfs)
+ DEF_CVT_FUNC(64f8u, double, uchar)
+
+ DEF_CVT_FUNC_F2(8u8s, uchar, schar, 8u8s_C1RSfs)
+ DEF_CVT_FUNC_F2(16u8s, ushort, schar, 16u8s_C1RSfs)
+ DEF_CVT_FUNC_F2(16s8s, short, schar, 16s8s_C1RSfs)
+ DEF_CVT_FUNC_F(32s8s, int, schar, 32s8s_C1R)
+ DEF_CVT_FUNC_F2(32f8s, float, schar, 32f8s_C1RSfs)
+ DEF_CVT_FUNC(64f8s, double, schar)
+
+ DEF_CVT_FUNC_F(8u16u, uchar, ushort, 8u16u_C1R)
+ DEF_CVT_FUNC_F(8s16u, schar, ushort, 8s16u_C1Rs)
+ DEF_CPY_FUNC(16u, ushort)
+ DEF_CVT_FUNC_F(16s16u, short, ushort, 16s16u_C1Rs)
+ DEF_CVT_FUNC_F2(32s16u, int, ushort, 32s16u_C1RSfs)
+ DEF_CVT_FUNC_F2(32f16u, float, ushort, 32f16u_C1RSfs)
+ DEF_CVT_FUNC(64f16u, double, ushort)
+
+ DEF_CVT_FUNC_F(8u16s, uchar, short, 8u16s_C1R)
+ DEF_CVT_FUNC_F(8s16s, schar, short, 8s16s_C1R)
+ DEF_CVT_FUNC_F2(16u16s, ushort, short, 16u16s_C1RSfs)
+ DEF_CVT_FUNC_F2(32s16s, int, short, 32s16s_C1RSfs)
+ DEF_CVT_FUNC(32f16s, float, short)
+ DEF_CVT_FUNC(64f16s, double, short)
+
+ DEF_CVT_FUNC_F(8u32s, uchar, int, 8u32s_C1R)
+ DEF_CVT_FUNC_F(8s32s, schar, int, 8s32s_C1R)
+ DEF_CVT_FUNC_F(16u32s, ushort, int, 16u32s_C1R)
+ DEF_CVT_FUNC_F(16s32s, short, int, 16s32s_C1R)
+ DEF_CPY_FUNC(32s, int)
+ DEF_CVT_FUNC_F2(32f32s, float, int, 32f32s_C1RSfs)
+ DEF_CVT_FUNC(64f32s, double, int)
+
+ DEF_CVT_FUNC_F(8u32f, uchar, float, 8u32f_C1R)
+ DEF_CVT_FUNC_F(8s32f, schar, float, 8s32f_C1R)
+ DEF_CVT_FUNC_F(16u32f, ushort, float, 16u32f_C1R)
+ DEF_CVT_FUNC_F(16s32f, short, float, 16s32f_C1R)
+ DEF_CVT_FUNC_F(32s32f, int, float, 32s32f_C1R)
+ DEF_CVT_FUNC(64f32f, double, float)
+
+ DEF_CVT_FUNC(8u64f, uchar, double)
+ DEF_CVT_FUNC(8s64f, schar, double)
+ DEF_CVT_FUNC(16u64f, ushort, double)
+ DEF_CVT_FUNC(16s64f, short, double)
+ DEF_CVT_FUNC(32s64f, int, double)
+ DEF_CVT_FUNC(32f64f, float, double)
+ DEF_CPY_FUNC(64s, int64)
+ */
+
+ BinaryFunc getConvertFunc(int sdepth, int ddepth)
+ {
+ CV_INSTRUMENT_REGION();
+ CV_CPU_DISPATCH(getConvertFunc, (sdepth, ddepth),
+ CV_CPU_DISPATCH_MODES_ALL);
+ }
+
- CV_Assert(ddepth == CV_16S /*|| ddepth == CV_16F*/);
+ #ifdef HAVE_OPENCL
+ static bool ocl_convertFp16( InputArray _src, OutputArray _dst, int sdepth, int ddepth )
+ {
+ int type = _src.type(), cn = CV_MAT_CN(type);
+
+ _dst.createSameSize( _src, CV_MAKETYPE(ddepth, cn) );
+ int kercn = 1;
+ int rowsPerWI = 1;
+ String build_opt = format("-D HALF_SUPPORT -D srcT=%s -D dstT=%s -D rowsPerWI=%d%s",
+ sdepth == CV_32F ? "float" : "half",
+ sdepth == CV_32F ? "half" : "float",
+ rowsPerWI,
+ sdepth == CV_32F ? " -D FLOAT_TO_HALF " : "");
+ ocl::Kernel k("convertFp16", ocl::core::halfconvert_oclsrc, build_opt);
+ if (k.empty())
+ return false;
+
+ UMat src = _src.getUMat();
+ UMat dst = _dst.getUMat();
+
+ ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
+ dstarg = ocl::KernelArg::WriteOnly(dst, cn, kercn);
+
+ k.args(srcarg, dstarg);
+
+ size_t globalsize[2] = { (size_t)src.cols * cn / kercn, ((size_t)src.rows + rowsPerWI - 1) / rowsPerWI };
+ return k.run(2, globalsize, NULL, false);
+ }
+ #endif
+
+ void Mat::convertTo(OutputArray _dst, int _type, double alpha, double beta) const
+ {
+ CV_INSTRUMENT_REGION();
+
+ if( empty() )
+ {
+ _dst.release();
+ return;
+ }
+
+ bool noScale = fabs(alpha-1) < DBL_EPSILON && fabs(beta) < DBL_EPSILON;
+
+ if( _type < 0 )
+ _type = _dst.fixedType() ? _dst.type() : type();
+ else
+ _type = CV_MAKETYPE(CV_MAT_DEPTH(_type), channels());
+
+ int sdepth = depth(), ddepth = CV_MAT_DEPTH(_type);
+ if( sdepth == ddepth && noScale )
+ {
+ copyTo(_dst);
+ return;
+ }
+
+ Mat src = *this;
+ if( dims <= 2 )
+ _dst.create( size(), _type );
+ else
+ _dst.create( dims, size, _type );
+ Mat dst = _dst.getMat();
+
+ BinaryFunc func = noScale ? getConvertFunc(sdepth, ddepth) : getConvertScaleFunc(sdepth, ddepth);
+ double scale[] = {alpha, beta};
+ int cn = channels();
+ CV_Assert( func != 0 );
+
+ if( dims <= 2 )
+ {
+ Size sz = getContinuousSize2D(src, dst, cn);
+ func( src.data, src.step, 0, 0, dst.data, dst.step, sz, scale );
+ }
+ else
+ {
+ const Mat* arrays[] = {&src, &dst, 0};
+ uchar* ptrs[2] = {};
+ NAryMatIterator it(arrays, ptrs);
+ Size sz((int)(it.size*cn), 1);
+
+ for( size_t i = 0; i < it.nplanes; i++, ++it )
+ func(ptrs[0], 1, 0, 0, ptrs[1], 1, sz, scale);
+ }
+ }
+
+ //==================================================================================================
+
+ void convertFp16(InputArray _src, OutputArray _dst)
+ {
+ CV_INSTRUMENT_REGION();
+
+ int sdepth = _src.depth(), ddepth = 0;
+ BinaryFunc func = 0;
+
+ switch( sdepth )
+ {
+ case CV_32F:
+ if(_dst.fixedType())
+ {
+ ddepth = _dst.depth();
- func = (BinaryFunc)get_cvt32f16f();
++ CV_Assert(ddepth == CV_16S || ddepth == CV_16F);
+ CV_Assert(_dst.channels() == _src.channels());
+ }
+ else
+ ddepth = CV_16S;
- //case CV_16F:
++ func = (BinaryFunc)getConvertFunc(CV_32F, CV_16F);
+ break;
+ case CV_16S:
- func = (BinaryFunc)get_cvt16f32f();
++ case CV_16F:
+ ddepth = CV_32F;
++ func = (BinaryFunc)getConvertFunc(CV_16F, CV_32F);
+ break;
+ default:
+ CV_Error(Error::StsUnsupportedFormat, "Unsupported input depth");
+ return;
+ }
+
+ CV_OCL_RUN(_src.dims() <= 2 && _dst.isUMat(),
+ ocl_convertFp16(_src, _dst, sdepth, ddepth))
+
+ Mat src = _src.getMat();
+
+ int type = CV_MAKETYPE(ddepth, src.channels());
+ _dst.create( src.dims, src.size, type );
+ Mat dst = _dst.getMat();
+ int cn = src.channels();
+
+ CV_Assert( func != 0 );
+
+ if( src.dims <= 2 )
+ {
+ Size sz = getContinuousSize2D(src, dst, cn);
+ func( src.data, src.step, 0, 0, dst.data, dst.step, sz, 0);
+ }
+ else
+ {
+ const Mat* arrays[] = {&src, &dst, 0};
+ uchar* ptrs[2] = {};
+ NAryMatIterator it(arrays, ptrs);
+ Size sz((int)(it.size*cn), 1);
+
+ for( size_t i = 0; i < it.nplanes; i++, ++it )
+ func(ptrs[0], 0, 0, 0, ptrs[1], 0, sz, 0);
+ }
+ }
+
+ } // namespace cv
#include "convert.hpp"
namespace cv {
-BinaryFunc get_cvt32f16f();
-BinaryFunc get_cvt16f32f();
++namespace hal {
++CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
++
++void cvt16f32f(const float16_t* src, float* dst, int len);
++void cvt32f16f(const float* src, float16_t* dst, int len);
++void addRNGBias32f(float* arr, const float* scaleBiasPairs, int len);
++void addRNGBias64f(double* arr, const double* scaleBiasPairs, int len);
++
++CV_CPU_OPTIMIZATION_NAMESPACE_END
++} // namespace cv::hal
++
+ CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
+
+ BinaryFunc getConvertFunc(int sdepth, int ddepth);
++
++CV_CPU_OPTIMIZATION_NAMESPACE_END
+
+ #ifndef CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
-/*namespace hal {
+namespace hal {
++CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
++
++BinaryFunc getConvertFunc(int sdepth, int ddepth);
void cvt16f32f( const float16_t* src, float* dst, int len )
{
++ CV_INSTRUMENT_REGION();
int j = 0;
#if CV_SIMD
const int VECSZ = v_float32::nlanes;
void cvt32f16f( const float* src, float16_t* dst, int len )
{
++ CV_INSTRUMENT_REGION();
int j = 0;
#if CV_SIMD
const int VECSZ = v_float32::nlanes;
dst[j] = float16_t(src[j]);
}
-/*void addRNGBias32f( float* arr, const float* scaleBiasPairs, int len )
+void addRNGBias32f( float* arr, const float* scaleBiasPairs, int len )
{
++ CV_INSTRUMENT_REGION();
// the loop is simple enough, so we let the compiler to vectorize it
for( int i = 0; i < len; i++ )
- arr[i] = scaleBiasPairs[i*2 + 1];
+ arr[i] += scaleBiasPairs[i*2 + 1];
}
void addRNGBias64f( double* arr, const double* scaleBiasPairs, int len )
{
++ CV_INSTRUMENT_REGION();
// the loop is simple enough, so we let the compiler to vectorize it
for( int i = 0; i < len; i++ )
- arr[i] = scaleBiasPairs[i*2 + 1];
+ arr[i] += scaleBiasPairs[i*2 + 1];
}
- }
-}*/
++CV_CPU_OPTIMIZATION_NAMESPACE_END
++} // namespace cv::hal
++
++// cv::
++CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
- template<typename _Ts, typename _Td, typename _Twvec> inline void
+ template<typename _Ts, typename _Td, typename _Twvec> static inline void
cvt_( const _Ts* src, size_t sstep, _Td* dst, size_t dstep, Size size )
{
sstep /= sizeof(src[0]);
return cvtTab[CV_MAT_DEPTH(ddepth)][CV_MAT_DEPTH(sdepth)];
}
- #ifdef HAVE_OPENCL
- static bool ocl_convertFp16( InputArray _src, OutputArray _dst, int sdepth, int ddepth )
-BinaryFunc get_cvt32f16f()
--{
- int type = _src.type(), cn = CV_MAT_CN(type);
-
- _dst.createSameSize( _src, CV_MAKETYPE(ddepth, cn) );
- int kercn = 1;
- int rowsPerWI = 1;
- String build_opt = format("-D HALF_SUPPORT -D srcT=%s -D dstT=%s -D rowsPerWI=%d%s",
- sdepth == CV_32F ? "float" : "half",
- sdepth == CV_32F ? "half" : "float",
- rowsPerWI,
- sdepth == CV_32F ? " -D FLOAT_TO_HALF " : "");
- ocl::Kernel k("convertFp16", ocl::core::halfconvert_oclsrc, build_opt);
- if (k.empty())
- return false;
-
- UMat src = _src.getUMat();
- UMat dst = _dst.getUMat();
- return (BinaryFunc)cvt32f16f;
-}
--
- ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
- dstarg = ocl::KernelArg::WriteOnly(dst, cn, kercn);
-BinaryFunc get_cvt16f32f()
-{
- return (BinaryFunc)cvt16f32f;
-}
--
- k.args(srcarg, dstarg);
-#endif
--
- size_t globalsize[2] = { (size_t)src.cols * cn / kercn, ((size_t)src.rows + rowsPerWI - 1) / rowsPerWI };
- return k.run(2, globalsize, NULL, false);
- }
+ CV_CPU_OPTIMIZATION_NAMESPACE_END
+#endif
-
- } // cv::
-
- void cv::Mat::convertTo(OutputArray _dst, int _type, double alpha, double beta) const
- {
- CV_INSTRUMENT_REGION();
-
- if( empty() )
- {
- _dst.release();
- return;
- }
-
- bool noScale = fabs(alpha-1) < DBL_EPSILON && fabs(beta) < DBL_EPSILON;
-
- if( _type < 0 )
- _type = _dst.fixedType() ? _dst.type() : type();
- else
- _type = CV_MAKETYPE(CV_MAT_DEPTH(_type), channels());
-
- int sdepth = depth(), ddepth = CV_MAT_DEPTH(_type);
- if( sdepth == ddepth && noScale )
- {
- copyTo(_dst);
- return;
- }
-
- Mat src = *this;
- if( dims <= 2 )
- _dst.create( size(), _type );
- else
- _dst.create( dims, size, _type );
- Mat dst = _dst.getMat();
-
- BinaryFunc func = noScale ? getConvertFunc(sdepth, ddepth) : getConvertScaleFunc(sdepth, ddepth);
- double scale[] = {alpha, beta};
- int cn = channels();
- CV_Assert( func != 0 );
-
- if( dims <= 2 )
- {
- Size sz = getContinuousSize2D(src, dst, cn);
- func( src.data, src.step, 0, 0, dst.data, dst.step, sz, scale );
- }
- else
- {
- const Mat* arrays[] = {&src, &dst, 0};
- uchar* ptrs[2] = {};
- NAryMatIterator it(arrays, ptrs);
- Size sz((int)(it.size*cn), 1);
-
- for( size_t i = 0; i < it.nplanes; i++, ++it )
- func(ptrs[0], 1, 0, 0, ptrs[1], 1, sz, scale);
- }
- }
-
- //==================================================================================================
-
- void cv::convertFp16( InputArray _src, OutputArray _dst )
- {
- CV_INSTRUMENT_REGION();
-
- int sdepth = _src.depth(), ddepth = 0;
- BinaryFunc func = 0;
-
- switch( sdepth )
- {
- case CV_32F:
- if(_dst.fixedType())
- {
- ddepth = _dst.depth();
- CV_Assert(ddepth == CV_16S || ddepth == CV_16F);
- CV_Assert(_dst.channels() == _src.channels());
- }
- else
- ddepth = CV_16S;
- func = (BinaryFunc)cvt32f16f;
- break;
- case CV_16S:
- case CV_16F:
- ddepth = CV_32F;
- func = (BinaryFunc)cvt16f32f;
- break;
- default:
- CV_Error(Error::StsUnsupportedFormat, "Unsupported input depth");
- return;
- }
-
- CV_OCL_RUN(_src.dims() <= 2 && _dst.isUMat(),
- ocl_convertFp16(_src, _dst, sdepth, ddepth))
-
- Mat src = _src.getMat();
-
- int type = CV_MAKETYPE(ddepth, src.channels());
- _dst.create( src.dims, src.size, type );
- Mat dst = _dst.getMat();
- int cn = src.channels();
-
- CV_Assert( func != 0 );
-
- if( src.dims <= 2 )
- {
- Size sz = getContinuousSize2D(src, dst, cn);
- func( src.data, src.step, 0, 0, dst.data, dst.step, sz, 0);
- }
- else
- {
- const Mat* arrays[] = {&src, &dst, 0};
- uchar* ptrs[2] = {};
- NAryMatIterator it(arrays, ptrs);
- Size sz((int)(it.size*cn), 1);
-
- for( size_t i = 0; i < it.nplanes; i++, ++it )
- func(ptrs[0], 0, 0, 0, ptrs[1], 0, sz, 0);
- }
- }
+ } // namespace
DEF_CVT_SCALE_FUNC(32s16f, cvt1_32f, int, float16_t, float)
DEF_CVT_SCALE_FUNC(32f16f, cvt1_32f, float, float16_t, float)
DEF_CVT_SCALE_FUNC(64f16f, cvt_64f, double, float16_t, double)
-DEF_CVT_SCALE_FUNC(16f, cvt1_32f, float16_t, float16_t, float)*/
+DEF_CVT_SCALE_FUNC(16f, cvt1_32f, float16_t, float16_t, float)
- static BinaryFunc getCvtScaleAbsFunc(int depth)
+ BinaryFunc getCvtScaleAbsFunc(int depth)
{
static BinaryFunc cvtScaleAbsTab[] =
{
projects / platform / upstream / opencv.git / commitdiff