OSSForks/opencv
1
0
Fork 0
mirror of https://github.com/zebrajr/opencv.git synced 2025-12-06 12:19:50 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Dropped inefficient (disabled) IPP integration for LUT.

Browse Source
This commit is contained in:
Alexander Smorkalov 2025-04-07 14:11:13 +03:00
parent 1b3db545a3
commit 91e078be93
2 changed files with 0 additions and 183 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
modules/core/include/opencv2/core/private.hpp
View File

@ -213,7 +213,6 @@ T* allocSingletonNew() { return new(allocSingletonNewBuffer(sizeof(T))) T(); }
// Temporary disabled named IPP region. Performance // Temporary disabled named IPP region. Performance
#define IPP_DISABLE_PERF_COPYMAKE 1 // performance variations #define IPP_DISABLE_PERF_COPYMAKE 1 // performance variations
#define IPP_DISABLE_PERF_LUT 1 // there are no performance benefits (PR #2653)
#define IPP_DISABLE_PERF_TRUE_DIST_MT 1 // cv::distanceTransform OpenCV MT performance is better #define IPP_DISABLE_PERF_TRUE_DIST_MT 1 // cv::distanceTransform OpenCV MT performance is better
#define IPP_DISABLE_PERF_CANNY_MT 1 // cv::Canny OpenCV MT performance is better #define IPP_DISABLE_PERF_CANNY_MT 1 // cv::Canny OpenCV MT performance is better

182
modules/core/src/lut.cpp
View File

@ -104,184 +104,6 @@ static bool ocl_LUT(InputArray _src, InputArray _lut, OutputArray _dst)
#endif #endif
#if defined(HAVE_IPP)
#if !IPP_DISABLE_PERF_LUT // there are no performance benefits (PR #2653)
namespace ipp {
class IppLUTParallelBody_LUTC1 : public ParallelLoopBody
{
public:
bool* ok;
const Mat& src_;
const Mat& lut_;
Mat& dst_;
int width;
size_t elemSize1;
IppLUTParallelBody_LUTC1(const Mat& src, const Mat& lut, Mat& dst, bool* _ok)
: ok(_ok), src_(src), lut_(lut), dst_(dst)
{
width = dst.cols * dst.channels();
elemSize1 = CV_ELEM_SIZE1(dst.depth());
CV_DbgAssert(elemSize1 == 1 || elemSize1 == 4);
*ok = true;
}
void operator()( const cv::Range& range ) const
{
if (!*ok)
return;
const int row0 = range.start;
const int row1 = range.end;
Mat src = src_.rowRange(row0, row1);
Mat dst = dst_.rowRange(row0, row1);
IppiSize sz = { width, dst.rows };
if (elemSize1 == 1)
{
if (CV_INSTRUMENT_FUN_IPP(ippiLUTPalette_8u_C1R, (const Ipp8u*)src.data, (int)src.step[0], dst.data, (int)dst.step[0], sz, lut_.data, 8) >= 0)
return;
}
else if (elemSize1 == 4)
{
if (CV_INSTRUMENT_FUN_IPP(ippiLUTPalette_8u32u_C1R, (const Ipp8u*)src.data, (int)src.step[0], (Ipp32u*)dst.data, (int)dst.step[0], sz, (Ipp32u*)lut_.data, 8) >= 0)
return;
}
*ok = false;
}
private:
IppLUTParallelBody_LUTC1(const IppLUTParallelBody_LUTC1&);
IppLUTParallelBody_LUTC1& operator=(const IppLUTParallelBody_LUTC1&);
};
class IppLUTParallelBody_LUTCN : public ParallelLoopBody
{
public:
bool *ok;
const Mat& src_;
const Mat& lut_;
Mat& dst_;
int lutcn;
uchar* lutBuffer;
uchar* lutTable[4];
IppLUTParallelBody_LUTCN(const Mat& src, const Mat& lut, Mat& dst, bool* _ok)
: ok(_ok), src_(src), lut_(lut), dst_(dst), lutBuffer(NULL)
{
lutcn = lut.channels();
IppiSize sz256 = {256, 1};
size_t elemSize1 = dst.elemSize1();
CV_DbgAssert(elemSize1 == 1);
lutBuffer = (uchar*)CV_IPP_MALLOC(256 * (int)elemSize1 * 4);
lutTable[0] = lutBuffer + 0;
lutTable[1] = lutBuffer + 1 * 256 * elemSize1;
lutTable[2] = lutBuffer + 2 * 256 * elemSize1;
lutTable[3] = lutBuffer + 3 * 256 * elemSize1;
CV_DbgAssert(lutcn == 3 || lutcn == 4);
if (lutcn == 3)
{
IppStatus status = CV_INSTRUMENT_FUN_IPP(ippiCopy_8u_C3P3R, lut.ptr(), (int)lut.step[0], lutTable, (int)lut.step[0], sz256);
if (status < 0)
return;
}
else if (lutcn == 4)
{
IppStatus status = CV_INSTRUMENT_FUN_IPP(ippiCopy_8u_C4P4R, lut.ptr(), (int)lut.step[0], lutTable, (int)lut.step[0], sz256);
if (status < 0)
return;
}
*ok = true;
}
~IppLUTParallelBody_LUTCN()
{
if (lutBuffer != NULL)
ippFree(lutBuffer);
lutBuffer = NULL;
lutTable[0] = NULL;
}
void operator()( const cv::Range& range ) const
{
if (!*ok)
return;
const int row0 = range.start;
const int row1 = range.end;
Mat src = src_.rowRange(row0, row1);
Mat dst = dst_.rowRange(row0, row1);
if (lutcn == 3)
{
if (CV_INSTRUMENT_FUN_IPP(ippiLUTPalette_8u_C3R, src.ptr(), (int)src.step[0], dst.ptr(), (int)dst.step[0], ippiSize(dst.size()), lutTable, 8) >= 0)
return;
}
else if (lutcn == 4)
{
if (CV_INSTRUMENT_FUN_IPP(ippiLUTPalette_8u_C4R, src.ptr(), (int)src.step[0], dst.ptr(), (int)dst.step[0], ippiSize(dst.size()), lutTable, 8) >= 0)
return;
}
*ok = false;
}
private:
IppLUTParallelBody_LUTCN(const IppLUTParallelBody_LUTCN&);
IppLUTParallelBody_LUTCN& operator=(const IppLUTParallelBody_LUTCN&);
};
} // namespace ipp
static bool ipp_lut(Mat &src, Mat &lut, Mat &dst)
{
CV_INSTRUMENT_REGION_IPP();
int lutcn = lut.channels();
if(src.dims > 2)
return false;
bool ok = false;
Ptr<ParallelLoopBody> body;
size_t elemSize1 = CV_ELEM_SIZE1(dst.depth());
if (lutcn == 1)
{
ParallelLoopBody* p = new ipp::IppLUTParallelBody_LUTC1(src, lut, dst, &ok);
body.reset(p);
}
else if ((lutcn == 3 || lutcn == 4) && elemSize1 == 1)
{
ParallelLoopBody* p = new ipp::IppLUTParallelBody_LUTCN(src, lut, dst, &ok);
body.reset(p);
}
if (body != NULL && ok)
{
Range all(0, dst.rows);
if (dst.total()>>18)
parallel_for_(all, *body, (double)std::max((size_t)1, dst.total()>>16));
else
(*body)(all);
if (ok)
return true;
}
return false;
}
#endif
#endif // IPP
class LUTParallelBody : public ParallelLoopBody class LUTParallelBody : public ParallelLoopBody
{ {
public: public:
@ -348,10 +170,6 @@ void cv::LUT( InputArray _src, InputArray _lut, OutputArray _dst )
CALL_HAL(LUT, cv_hal_lut, src.data, src.step, src.type(), lut.data, CALL_HAL(LUT, cv_hal_lut, src.data, src.step, src.type(), lut.data,
lut.elemSize1(), lutcn, dst.data, dst.step, src.cols, src.rows); lut.elemSize1(), lutcn, dst.data, dst.step, src.cols, src.rows);
#if !IPP_DISABLE_PERF_LUT
CV_IPP_RUN(_src.dims() <= 2, ipp_lut(src, lut, dst));
#endif
if (_src.dims() <= 2) if (_src.dims() <= 2)
{ {
bool ok = false; bool ok = false;