potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 07:13:02 +04:00
Code Packages Releases Wiki Activity

core: rework code locality

Browse Source 
- to reduce binaries size of FFmpeg Windows wrapper
- MinGW linker doesn't support -ffunction-sections (used for FFmpeg Windows wrapper)
- move code to improve locality with its used dependencies
- move UMat::dot() to matmul.dispatch.cpp (Mat::dot() is already there)
- move UMat::inv() to lapack.cpp
- move UMat::mul() to arithm.cpp
- move UMat:eye() to matrix_operations.cpp (near setIdentity() implementation)
- move normalize(): convert_scale.cpp => norm.cpp
- move convertAndUnrollScalar(): arithm.cpp => copy.cpp
- move scalarToRawData(): array.cpp => copy.cpp
- move transpose(): matrix_operations.cpp => matrix_transform.cpp
- move flip(), rotate(): copy.cpp => matrix_transform.cpp (rotate90 uses flip and transpose)
- add 'OPENCV_CORE_EXCLUDE_C_API' CMake variable to exclude compilation of C-API functions from the core module
- matrix_wrap.cpp: add compile-time checks for CUDA/OpenGL calls
- the steps above allow to reduce FFmpeg wrapper size for ~1.5Mb (initial size of OpenCV part is about 3Mb)
This commit is contained in:
Alexander Alekhin
2021-02-23 00:22:06 +00:00
parent 7bcb51eded
commit 65eb946756
19 changed files with 1206 additions and 1108 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/src/convert_scale.dispatch.cpp
-140
View File
@@ -9,7 +9,6 @@
#include "convert_scale.simd.hpp"
#include "convert_scale.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
namespace cv
{
@@ -117,143 +116,4 @@ void convertScaleAbs(InputArray _src, OutputArray _dst, double alpha, double bet
}
}
//==================================================================================================
#ifdef HAVE_OPENCL
static bool ocl_normalize( InputArray _src, InputOutputArray _dst, InputArray _mask, int dtype,
double scale, double delta )
{
UMat src = _src.getUMat();
if( _mask.empty() )
src.convertTo( _dst, dtype, scale, delta );
else if (src.channels() <= 4)
{
const ocl::Device & dev = ocl::Device::getDefault();
int stype = _src.type(), sdepth = CV_MAT_DEPTH(stype), cn = CV_MAT_CN(stype),
ddepth = CV_MAT_DEPTH(dtype), wdepth = std::max(CV_32F, std::max(sdepth, ddepth)),
rowsPerWI = dev.isIntel() ? 4 : 1;
float fscale = static_cast<float>(scale), fdelta = static_cast<float>(delta);
bool haveScale = std::fabs(scale - 1) > DBL_EPSILON,
haveZeroScale = !(std::fabs(scale) > DBL_EPSILON),
haveDelta = std::fabs(delta) > DBL_EPSILON,
doubleSupport = dev.doubleFPConfig() > 0;
if (!haveScale && !haveDelta && stype == dtype)
{
_src.copyTo(_dst, _mask);
return true;
}
if (haveZeroScale)
{
_dst.setTo(Scalar(delta), _mask);
return true;
}
if ((sdepth == CV_64F || ddepth == CV_64F) && !doubleSupport)
return false;
char cvt[2][40];
String opts = format("-D srcT=%s -D dstT=%s -D convertToWT=%s -D cn=%d -D rowsPerWI=%d"
" -D convertToDT=%s -D workT=%s%s%s%s -D srcT1=%s -D dstT1=%s",
ocl::typeToStr(stype), ocl::typeToStr(dtype),
ocl::convertTypeStr(sdepth, wdepth, cn, cvt[0]), cn,
rowsPerWI, ocl::convertTypeStr(wdepth, ddepth, cn, cvt[1]),
ocl::typeToStr(CV_MAKE_TYPE(wdepth, cn)),
doubleSupport ? " -D DOUBLE_SUPPORT" : "",
haveScale ? " -D HAVE_SCALE" : "",
haveDelta ? " -D HAVE_DELTA" : "",
ocl::typeToStr(sdepth), ocl::typeToStr(ddepth));
ocl::Kernel k("normalizek", ocl::core::normalize_oclsrc, opts);
if (k.empty())
return false;
UMat mask = _mask.getUMat(), dst = _dst.getUMat();
ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
maskarg = ocl::KernelArg::ReadOnlyNoSize(mask),
dstarg = ocl::KernelArg::ReadWrite(dst);
if (haveScale)
{
if (haveDelta)
k.args(srcarg, maskarg, dstarg, fscale, fdelta);
else
k.args(srcarg, maskarg, dstarg, fscale);
}
else
{
if (haveDelta)
k.args(srcarg, maskarg, dstarg, fdelta);
else
k.args(srcarg, maskarg, dstarg);
}
size_t globalsize[2] = { (size_t)src.cols, ((size_t)src.rows + rowsPerWI - 1) / rowsPerWI };
return k.run(2, globalsize, NULL, false);
}
else
{
UMat temp;
src.convertTo( temp, dtype, scale, delta );
temp.copyTo( _dst, _mask );
}
return true;
}
#endif
void normalize(InputArray _src, InputOutputArray _dst, double a, double b,
int norm_type, int rtype, InputArray _mask)
{
CV_INSTRUMENT_REGION();
double scale = 1, shift = 0;
int type = _src.type(), depth = CV_MAT_DEPTH(type);
if( rtype < 0 )
rtype = _dst.fixedType() ? _dst.depth() : depth;
if( norm_type == CV_MINMAX )
{
double smin = 0, smax = 0;
double dmin = MIN( a, b ), dmax = MAX( a, b );
minMaxIdx( _src, &smin, &smax, 0, 0, _mask );
scale = (dmax - dmin)*(smax - smin > DBL_EPSILON ? 1./(smax - smin) : 0);
if( rtype == CV_32F )
{
scale = (float)scale;
shift = (float)dmin - (float)(smin*scale);
}
else
shift = dmin - smin*scale;
}
else if( norm_type == CV_L2 || norm_type == CV_L1 || norm_type == CV_C )
{
scale = norm( _src, norm_type, _mask );
scale = scale > DBL_EPSILON ? a/scale : 0.;
shift = 0;
}
else
CV_Error( CV_StsBadArg, "Unknown/unsupported norm type" );
CV_OCL_RUN(_dst.isUMat(),
ocl_normalize(_src, _dst, _mask, rtype, scale, shift))
Mat src = _src.getMat();
if( _mask.empty() )
src.convertTo( _dst, rtype, scale, shift );
else
{
Mat temp;
src.convertTo( temp, rtype, scale, shift );
temp.copyTo( _dst, _mask );
}
}
} // namespace