generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Implementation detector and selector for IPP and OpenCL;

Browse Source 
IPP can be switched on and off on runtime;

Optional implementation collector was added (switched off by default in CMake). Gathers data of implementation used in functions and report this info through performance TS;

TS modifications for implementations control;
This commit is contained in:
Pavel Vlasov
2014-10-03 15:17:28 +04:00
parent 83ef276697
commit 45958eaabc
49 changed files with 4279 additions and 2799 deletions
Download Patch File Download Diff File Expand all files Collapse all files

366
modules/imgproc/src/accum.cpp
View File

@@ -856,54 +856,60 @@ void cv::accumulate( InputArray _src, InputOutputArray _dst, InputArray _mask )
Mat src = _src.getMat(), dst = _dst.getMat(), mask = _mask.getMat();
#if defined HAVE_IPP
if (src.dims <= 2 || (src.isContinuous() && dst.isContinuous() && (mask.empty() || mask.isContinuous())))
CV_IPP_CHECK()
{
typedef IppStatus (CV_STDCALL * ippiAdd)(const void * pSrc, int srcStep, Ipp32f * pSrcDst, int srcdstStep, IppiSize roiSize);
typedef IppStatus (CV_STDCALL * ippiAddMask)(const void * pSrc, int srcStep, const Ipp8u * pMask, int maskStep, Ipp32f * pSrcDst,
int srcDstStep, IppiSize roiSize);
ippiAdd ippFunc = 0;
ippiAddMask ippFuncMask = 0;
if (mask.empty())
if (src.dims <= 2 || (src.isContinuous() && dst.isContinuous() && (mask.empty() || mask.isContinuous())))
{
CV_SUPPRESS_DEPRECATED_START
ippFunc = sdepth == CV_8U && ddepth == CV_32F ? (ippiAdd)ippiAdd_8u32f_C1IR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAdd)ippiAdd_16u32f_C1IR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAdd)ippiAdd_32f_C1IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else if (scn == 1)
{
ippFuncMask = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddMask)ippiAdd_8u32f_C1IMR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddMask)ippiAdd_16u32f_C1IMR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddMask)ippiAdd_32f_C1IMR : 0;
}
if (ippFunc || ippFuncMask)
{
IppStatus status = ippStsNoErr;
Size size = src.size();
int srcstep = (int)src.step, dststep = (int)dst.step, maskstep = (int)mask.step;
if (src.isContinuous() && dst.isContinuous() && mask.isContinuous())
{
srcstep = static_cast<int>(src.total() * src.elemSize());
dststep = static_cast<int>(dst.total() * dst.elemSize());
maskstep = static_cast<int>(mask.total() * mask.elemSize());
size.width = static_cast<int>(src.total());
size.height = 1;
}
size.width *= scn;
typedef IppStatus (CV_STDCALL * ippiAdd)(const void * pSrc, int srcStep, Ipp32f * pSrcDst, int srcdstStep, IppiSize roiSize);
typedef IppStatus (CV_STDCALL * ippiAddMask)(const void * pSrc, int srcStep, const Ipp8u * pMask, int maskStep, Ipp32f * pSrcDst,
int srcDstStep, IppiSize roiSize);
ippiAdd ippFunc = 0;
ippiAddMask ippFuncMask = 0;
if (mask.empty())
status = ippFunc(src.ptr(), srcstep, dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
else
status = ippFuncMask(src.ptr(), srcstep, mask.ptr<Ipp8u>(), maskstep,
dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
{
CV_SUPPRESS_DEPRECATED_START
ippFunc = sdepth == CV_8U && ddepth == CV_32F ? (ippiAdd)ippiAdd_8u32f_C1IR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAdd)ippiAdd_16u32f_C1IR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAdd)ippiAdd_32f_C1IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else if (scn == 1)
{
ippFuncMask = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddMask)ippiAdd_8u32f_C1IMR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddMask)ippiAdd_16u32f_C1IMR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddMask)ippiAdd_32f_C1IMR : 0;
}
if (status >= 0)
return;
setIppErrorStatus();
if (ippFunc || ippFuncMask)
{
IppStatus status = ippStsNoErr;
Size size = src.size();
int srcstep = (int)src.step, dststep = (int)dst.step, maskstep = (int)mask.step;
if (src.isContinuous() && dst.isContinuous() && mask.isContinuous())
{
srcstep = static_cast<int>(src.total() * src.elemSize());
dststep = static_cast<int>(dst.total() * dst.elemSize());
maskstep = static_cast<int>(mask.total() * mask.elemSize());
size.width = static_cast<int>(src.total());
size.height = 1;
}
size.width *= scn;
if (mask.empty())
status = ippFunc(src.ptr(), srcstep, dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
else
status = ippFuncMask(src.ptr(), srcstep, mask.ptr<Ipp8u>(), maskstep,
dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
}
#endif
@@ -935,52 +941,58 @@ void cv::accumulateSquare( InputArray _src, InputOutputArray _dst, InputArray _m
Mat src = _src.getMat(), dst = _dst.getMat(), mask = _mask.getMat();
#if defined(HAVE_IPP)
if (src.dims <= 2 || (src.isContinuous() && dst.isContinuous() && (mask.empty() || mask.isContinuous())))
CV_IPP_CHECK()
{
typedef IppStatus (CV_STDCALL * ippiAddSquare)(const void * pSrc, int srcStep, Ipp32f * pSrcDst, int srcdstStep, IppiSize roiSize);
typedef IppStatus (CV_STDCALL * ippiAddSquareMask)(const void * pSrc, int srcStep, const Ipp8u * pMask, int maskStep, Ipp32f * pSrcDst,
int srcDstStep, IppiSize roiSize);
ippiAddSquare ippFunc = 0;
ippiAddSquareMask ippFuncMask = 0;
if (mask.empty())
if (src.dims <= 2 || (src.isContinuous() && dst.isContinuous() && (mask.empty() || mask.isContinuous())))
{
ippFunc = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddSquare)ippiAddSquare_8u32f_C1IR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddSquare)ippiAddSquare_16u32f_C1IR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddSquare)ippiAddSquare_32f_C1IR : 0;
}
else if (scn == 1)
{
ippFuncMask = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddSquareMask)ippiAddSquare_8u32f_C1IMR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddSquareMask)ippiAddSquare_16u32f_C1IMR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddSquareMask)ippiAddSquare_32f_C1IMR : 0;
}
if (ippFunc || ippFuncMask)
{
IppStatus status = ippStsNoErr;
Size size = src.size();
int srcstep = (int)src.step, dststep = (int)dst.step, maskstep = (int)mask.step;
if (src.isContinuous() && dst.isContinuous() && mask.isContinuous())
{
srcstep = static_cast<int>(src.total() * src.elemSize());
dststep = static_cast<int>(dst.total() * dst.elemSize());
maskstep = static_cast<int>(mask.total() * mask.elemSize());
size.width = static_cast<int>(src.total());
size.height = 1;
}
size.width *= scn;
typedef IppStatus (CV_STDCALL * ippiAddSquare)(const void * pSrc, int srcStep, Ipp32f * pSrcDst, int srcdstStep, IppiSize roiSize);
typedef IppStatus (CV_STDCALL * ippiAddSquareMask)(const void * pSrc, int srcStep, const Ipp8u * pMask, int maskStep, Ipp32f * pSrcDst,
int srcDstStep, IppiSize roiSize);
ippiAddSquare ippFunc = 0;
ippiAddSquareMask ippFuncMask = 0;
if (mask.empty())
status = ippFunc(src.ptr(), srcstep, dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
else
status = ippFuncMask(src.ptr(), srcstep, mask.ptr<Ipp8u>(), maskstep,
dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
{
ippFunc = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddSquare)ippiAddSquare_8u32f_C1IR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddSquare)ippiAddSquare_16u32f_C1IR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddSquare)ippiAddSquare_32f_C1IR : 0;
}
else if (scn == 1)
{
ippFuncMask = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddSquareMask)ippiAddSquare_8u32f_C1IMR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddSquareMask)ippiAddSquare_16u32f_C1IMR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddSquareMask)ippiAddSquare_32f_C1IMR : 0;
}
if (status >= 0)
return;
setIppErrorStatus();
if (ippFunc || ippFuncMask)
{
IppStatus status = ippStsNoErr;
Size size = src.size();
int srcstep = (int)src.step, dststep = (int)dst.step, maskstep = (int)mask.step;
if (src.isContinuous() && dst.isContinuous() && mask.isContinuous())
{
srcstep = static_cast<int>(src.total() * src.elemSize());
dststep = static_cast<int>(dst.total() * dst.elemSize());
maskstep = static_cast<int>(mask.total() * mask.elemSize());
size.width = static_cast<int>(src.total());
size.height = 1;
}
size.width *= scn;
if (mask.empty())
status = ippFunc(src.ptr(), srcstep, dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
else
status = ippFuncMask(src.ptr(), srcstep, mask.ptr<Ipp8u>(), maskstep,
dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
}
#endif
@@ -1014,55 +1026,61 @@ void cv::accumulateProduct( InputArray _src1, InputArray _src2,
Mat src1 = _src1.getMat(), src2 = _src2.getMat(), dst = _dst.getMat(), mask = _mask.getMat();
#if defined(HAVE_IPP)
if (src1.dims <= 2 || (src1.isContinuous() && src2.isContinuous() && dst.isContinuous()))
CV_IPP_CHECK()
{
typedef IppStatus (CV_STDCALL * ippiAddProduct)(const void * pSrc1, int src1Step, const void * pSrc2,
int src2Step, Ipp32f * pSrcDst, int srcDstStep, IppiSize roiSize);
typedef IppStatus (CV_STDCALL * ippiAddProductMask)(const void * pSrc1, int src1Step, const void * pSrc2, int src2Step,
const Ipp8u * pMask, int maskStep, Ipp32f * pSrcDst, int srcDstStep, IppiSize roiSize);
ippiAddProduct ippFunc = 0;
ippiAddProductMask ippFuncMask = 0;
if (mask.empty())
if (src1.dims <= 2 || (src1.isContinuous() && src2.isContinuous() && dst.isContinuous()))
{
ippFunc = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddProduct)ippiAddProduct_8u32f_C1IR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddProduct)ippiAddProduct_16u32f_C1IR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddProduct)ippiAddProduct_32f_C1IR : 0;
}
else if (scn == 1)
{
ippFuncMask = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddProductMask)ippiAddProduct_8u32f_C1IMR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddProductMask)ippiAddProduct_16u32f_C1IMR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddProductMask)ippiAddProduct_32f_C1IMR : 0;
}
if (ippFunc || ippFuncMask)
{
IppStatus status = ippStsNoErr;
Size size = src1.size();
int src1step = (int)src1.step, src2step = (int)src2.step, dststep = (int)dst.step, maskstep = (int)mask.step;
if (src1.isContinuous() && src2.isContinuous() && dst.isContinuous() && mask.isContinuous())
{
src1step = static_cast<int>(src1.total() * src1.elemSize());
src2step = static_cast<int>(src2.total() * src2.elemSize());
dststep = static_cast<int>(dst.total() * dst.elemSize());
maskstep = static_cast<int>(mask.total() * mask.elemSize());
size.width = static_cast<int>(src1.total());
size.height = 1;
}
size.width *= scn;
typedef IppStatus (CV_STDCALL * ippiAddProduct)(const void * pSrc1, int src1Step, const void * pSrc2,
int src2Step, Ipp32f * pSrcDst, int srcDstStep, IppiSize roiSize);
typedef IppStatus (CV_STDCALL * ippiAddProductMask)(const void * pSrc1, int src1Step, const void * pSrc2, int src2Step,
const Ipp8u * pMask, int maskStep, Ipp32f * pSrcDst, int srcDstStep, IppiSize roiSize);
ippiAddProduct ippFunc = 0;
ippiAddProductMask ippFuncMask = 0;
if (mask.empty())
status = ippFunc(src1.ptr(), src1step, src2.ptr(), src2step, dst.ptr<Ipp32f>(),
dststep, ippiSize(size.width, size.height));
else
status = ippFuncMask(src1.ptr(), src1step, src2.ptr(), src2step, mask.ptr<Ipp8u>(), maskstep,
dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
{
ippFunc = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddProduct)ippiAddProduct_8u32f_C1IR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddProduct)ippiAddProduct_16u32f_C1IR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddProduct)ippiAddProduct_32f_C1IR : 0;
}
else if (scn == 1)
{
ippFuncMask = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddProductMask)ippiAddProduct_8u32f_C1IMR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddProductMask)ippiAddProduct_16u32f_C1IMR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddProductMask)ippiAddProduct_32f_C1IMR : 0;
}
if (status >= 0)
return;
setIppErrorStatus();
if (ippFunc || ippFuncMask)
{
IppStatus status = ippStsNoErr;
Size size = src1.size();
int src1step = (int)src1.step, src2step = (int)src2.step, dststep = (int)dst.step, maskstep = (int)mask.step;
if (src1.isContinuous() && src2.isContinuous() && dst.isContinuous() && mask.isContinuous())
{
src1step = static_cast<int>(src1.total() * src1.elemSize());
src2step = static_cast<int>(src2.total() * src2.elemSize());
dststep = static_cast<int>(dst.total() * dst.elemSize());
maskstep = static_cast<int>(mask.total() * mask.elemSize());
size.width = static_cast<int>(src1.total());
size.height = 1;
}
size.width *= scn;
if (mask.empty())
status = ippFunc(src1.ptr(), src1step, src2.ptr(), src2step, dst.ptr<Ipp32f>(),
dststep, ippiSize(size.width, size.height));
else
status = ippFuncMask(src1.ptr(), src1step, src2.ptr(), src2step, mask.ptr<Ipp8u>(), maskstep,
dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
}
#endif
@@ -1095,54 +1113,60 @@ void cv::accumulateWeighted( InputArray _src, InputOutputArray _dst,
Mat src = _src.getMat(), dst = _dst.getMat(), mask = _mask.getMat();
#if defined(HAVE_IPP)
if (src.dims <= 2 || (src.isContinuous() && dst.isContinuous() && mask.isContinuous()))
CV_IPP_CHECK()
{
typedef IppStatus (CV_STDCALL * ippiAddWeighted)(const void * pSrc, int srcStep, Ipp32f * pSrcDst, int srcdstStep,
IppiSize roiSize, Ipp32f alpha);
typedef IppStatus (CV_STDCALL * ippiAddWeightedMask)(const void * pSrc, int srcStep, const Ipp8u * pMask,
int maskStep, Ipp32f * pSrcDst,
int srcDstStep, IppiSize roiSize, Ipp32f alpha);
ippiAddWeighted ippFunc = 0;
ippiAddWeightedMask ippFuncMask = 0;
if (mask.empty())
if (src.dims <= 2 || (src.isContinuous() && dst.isContinuous() && mask.isContinuous()))
{
ippFunc = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddWeighted)ippiAddWeighted_8u32f_C1IR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddWeighted)ippiAddWeighted_16u32f_C1IR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddWeighted)ippiAddWeighted_32f_C1IR : 0;
}
else if (scn == 1)
{
ippFuncMask = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddWeightedMask)ippiAddWeighted_8u32f_C1IMR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddWeightedMask)ippiAddWeighted_16u32f_C1IMR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddWeightedMask)ippiAddWeighted_32f_C1IMR : 0;
}
if (ippFunc || ippFuncMask)
{
IppStatus status = ippStsNoErr;
Size size = src.size();
int srcstep = (int)src.step, dststep = (int)dst.step, maskstep = (int)mask.step;
if (src.isContinuous() && dst.isContinuous() && mask.isContinuous())
{
srcstep = static_cast<int>(src.total() * src.elemSize());
dststep = static_cast<int>(dst.total() * dst.elemSize());
maskstep = static_cast<int>(mask.total() * mask.elemSize());
size.width = static_cast<int>((int)src.total());
size.height = 1;
}
size.width *= scn;
typedef IppStatus (CV_STDCALL * ippiAddWeighted)(const void * pSrc, int srcStep, Ipp32f * pSrcDst, int srcdstStep,
IppiSize roiSize, Ipp32f alpha);
typedef IppStatus (CV_STDCALL * ippiAddWeightedMask)(const void * pSrc, int srcStep, const Ipp8u * pMask,
int maskStep, Ipp32f * pSrcDst,
int srcDstStep, IppiSize roiSize, Ipp32f alpha);
ippiAddWeighted ippFunc = 0;
ippiAddWeightedMask ippFuncMask = 0;
if (mask.empty())
status = ippFunc(src.ptr(), srcstep, dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height), (Ipp32f)alpha);
else
status = ippFuncMask(src.ptr(), srcstep, mask.ptr<Ipp8u>(), maskstep,
dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height), (Ipp32f)alpha);
{
ippFunc = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddWeighted)ippiAddWeighted_8u32f_C1IR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddWeighted)ippiAddWeighted_16u32f_C1IR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddWeighted)ippiAddWeighted_32f_C1IR : 0;
}
else if (scn == 1)
{
ippFuncMask = sdepth == CV_8U && ddepth == CV_32F ? (ippiAddWeightedMask)ippiAddWeighted_8u32f_C1IMR :
sdepth == CV_16U && ddepth == CV_32F ? (ippiAddWeightedMask)ippiAddWeighted_16u32f_C1IMR :
sdepth == CV_32F && ddepth == CV_32F ? (ippiAddWeightedMask)ippiAddWeighted_32f_C1IMR : 0;
}
if (status >= 0)
return;
setIppErrorStatus();
if (ippFunc || ippFuncMask)
{
IppStatus status = ippStsNoErr;
Size size = src.size();
int srcstep = (int)src.step, dststep = (int)dst.step, maskstep = (int)mask.step;
if (src.isContinuous() && dst.isContinuous() && mask.isContinuous())
{
srcstep = static_cast<int>(src.total() * src.elemSize());
dststep = static_cast<int>(dst.total() * dst.elemSize());
maskstep = static_cast<int>(mask.total() * mask.elemSize());
size.width = static_cast<int>((int)src.total());
size.height = 1;
}
size.width *= scn;
if (mask.empty())
status = ippFunc(src.ptr(), srcstep, dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height), (Ipp32f)alpha);
else
status = ippFuncMask(src.ptr(), srcstep, mask.ptr<Ipp8u>(), maskstep,
dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height), (Ipp32f)alpha);
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
}
#endif