Skip to content

O
opencv
Commit 16629bff authored by Elena Gvozdeva's avatar Elena Gvozdeva
Browse files
Options

Added parallel version for DCT_ROWS

parent 6880dbfd
Show whitespace changes
Inline Side-by-side
Showing with 102 additions and 72 deletions
modules/core/src/dxt.cpp
View file @ 16629bff
......@@ -2882,125 +2882,153 @@ static void IDCT_64f(const double* src, int src_step, double* dft_src, double* d
namespace cv
{
#if defined HAVE_IPP && IPP_VERSION_MAJOR >= 7
#if defined HAVE_IPP && IPP_VERSION_MAJOR >= 7 && !defined HAVE_IPP_ICV_ONLY
typedef IppStatus (CV_STDCALL * ippiDCTFwdFunc)(const Ipp32f*, int, Ipp32f*, int, const IppiDCTFwdSpec_32f*, Ipp8u*);
typedef IppStatus (CV_STDCALL * ippiDCTInvFunc)(const Ipp32f*, int, Ipp32f*, int, const IppiDCTInvSpec_32f*, Ipp8u*);
typedef IppStatus (CV_STDCALL * ippiDCTFunc)(const Ipp32f*, int, Ipp32f*, int, const void*, Ipp8u*);
typedef IppStatus (CV_STDCALL * ippiDCTInitAlloc)(void**, IppiSize, IppHintAlgorithm);
typedef IppStatus (CV_STDCALL * ippiDCTFree)(void* pDCTSpec);
typedef IppStatus (CV_STDCALL * ippiDCTGetBufSize)(const void*, int*);
static bool ippi_DCT_Fwd(const Mat& src, Mat& dst, bool row)
template <typename Dct>
class DctIPPLoop_Invoker : public ParallelLoopBody
{
if (src.type() != CV_32F)
return false;
public:
DctIPPLoop_Invoker(const Mat& _src, Mat& _dst, const Dct& _ippidct, bool _inv, bool *_ok) :
ParallelLoopBody(), src(_src), dst(_dst), ippidct(_ippidct), inv(_inv), ok(_ok)
{
*ok = true;
}
virtual void operator()(const Range& range) const
{
IppStatus status;
IppiDCTFwdSpec_32f* pDCTSpec;
Ipp8u *pBuffer;
void* pDCTSpec;
AutoBuffer<uchar> buf;
uchar* pBuffer = 0;
int bufSize=0;
ippiDCTFwdFunc ippFunc = (ippiDCTFwdFunc)ippiDCTFwd_32f_C1R;
IppiSize srcRoiSize = {src.cols, 1};
if (ippFunc==0)
return false;
CV_SUPPRESS_DEPRECATED_START
IppiSize srcRoiSize = {src.cols, row ? 1 : src.rows};
ippiDCTInitAlloc ippInitAlloc = inv ? (ippiDCTInitAlloc)ippiDCTInvInitAlloc_32f : (ippiDCTInitAlloc)ippiDCTFwdInitAlloc_32f;
ippiDCTFree ippFree = inv ? (ippiDCTFree)ippiDCTInvFree_32f : (ippiDCTFree)ippiDCTFwdFree_32f;
ippiDCTGetBufSize ippGetBufSize = inv ? (ippiDCTGetBufSize)ippiDCTInvGetBufSize_32f : (ippiDCTGetBufSize)ippiDCTFwdGetBufSize_32f;
CV_SUPPRESS_DEPRECATED_START
status = ippiDCTFwdInitAlloc_32f (&pDCTSpec, srcRoiSize, ippAlgHintNone);
status = ippInitAlloc(&pDCTSpec, srcRoiSize, ippAlgHintNone);
if ( status < 0 )
{
ippiDCTFwdFree_32f(pDCTSpec);
return false;
ippFree(pDCTSpec);
*ok = false;
return;
}
status = ippiDCTFwdGetBufSize_32f (pDCTSpec, &bufSize);
status = ippGetBufSize(pDCTSpec, &bufSize);
if ( status < 0 )
{
ippiDCTFwdFree_32f(pDCTSpec);
return false;
ippFree(pDCTSpec);
*ok = false;
return;
}
pBuffer = ippsMalloc_8u( bufSize );
buf.allocate( bufSize );
pBuffer = (uchar*)buf;
if (row)
{
for (int i=0; i<src.rows; i++)
{
status = ippFunc((float*)(src.data+i*src.step), (int)src.step,(float*)(dst.data+i*dst.step), (int)dst.step, pDCTSpec, pBuffer);
if ( status < 0 )
for( int i = range.start; i < range.end; ++i)
if(!ippidct((float*)(src.data+i*src.step), (int)src.step,(float*)(dst.data+i*dst.step), (int)dst.step, pDCTSpec, (Ipp8u*)pBuffer))
{
ippsFree( pBuffer );
ippiDCTFwdFree_32f(pDCTSpec);
return false;
}
}
*ok = false;
}
else
status = ippFunc((float*)src.data, (int)src.step, (float*)dst.data, (int)dst.step, pDCTSpec, pBuffer);
ippsFree( pBuffer );
ippiDCTFwdFree_32f(pDCTSpec);
ippFree( pDCTSpec);
CV_SUPPRESS_DEPRECATED_END
}
return status >= 0;
private:
const Mat& src;
Mat& dst;
const Dct& ippidct;
bool inv;
bool *ok;
const DctIPPLoop_Invoker& operator= (const DctIPPLoop_Invoker&);
};
template <typename Dct>
bool DctIPPLoop(const Mat& src, Mat& dst, const Dct& ippidct, bool inv)
{
bool ok;
parallel_for_(Range(0, src.rows), DctIPPLoop_Invoker<Dct>(src, dst, ippidct, inv, &ok), src.total()/(double)(1<<16) );
return ok;
}
static bool ippi_DCT_Inv(const Mat& src, Mat& dst, bool row)
struct IPPDCTFunctor
{
IPPDCTFunctor(ippiDCTFunc _func) : func(_func){}
bool operator()(const Ipp32f* src, int srcStep, Ipp32f* dst, int dstStep, const void* pDCTSpec, Ipp8u* pBuffer) const
{
return func ? func(src, srcStep, dst, dstStep, pDCTSpec, pBuffer) >= 0 : false;
}
private:
ippiDCTFunc func;
};
static bool ippi_DCT(const Mat& src, Mat& dst, bool inv, bool row)
{
if (src.type() != CV_32F)
return false;
ippiDCTFunc ippFunc = inv ? (ippiDCTFunc)ippiDCTInv_32f_C1R : (ippiDCTFunc)ippiDCTFwd_32f_C1R ;
if (row)
if(DctIPPLoop(src,dst,IPPDCTFunctor(ippFunc),inv))
return true;
else
return false;
else
{
IppStatus status;
IppiDCTInvSpec_32f* pDCTSpec;
Ipp8u *pBuffer;
void* pDCTSpec;
AutoBuffer<uchar> buf;
uchar* pBuffer = 0;
int bufSize=0;
ippiDCTInvFunc ippFunc = (ippiDCTInvFunc)ippiDCTInv_32f_C1R;
IppiSize srcRoiSize = {src.cols, src.rows};
if (ippFunc==0)
return false;
CV_SUPPRESS_DEPRECATED_START
IppiSize srcRoiSize = {src.cols, row ? 1 : src.rows};
ippiDCTInitAlloc ippInitAlloc = inv ? (ippiDCTInitAlloc)ippiDCTInvInitAlloc_32f : (ippiDCTInitAlloc)ippiDCTFwdInitAlloc_32f;
ippiDCTFree ippFree = inv ? (ippiDCTFree)ippiDCTInvFree_32f : (ippiDCTFree)ippiDCTFwdFree_32f;
ippiDCTGetBufSize ippGetBufSize = inv ? (ippiDCTGetBufSize)ippiDCTInvGetBufSize_32f : (ippiDCTGetBufSize)ippiDCTFwdGetBufSize_32f;
CV_SUPPRESS_DEPRECATED_START
status = ippiDCTInvInitAlloc_32f (&pDCTSpec, srcRoiSize, ippAlgHintNone);
status = ippInitAlloc(&pDCTSpec, srcRoiSize, ippAlgHintNone);
if ( status < 0 )
{
ippiDCTInvFree_32f(pDCTSpec);
ippFree(pDCTSpec);
return false;
}
status = ippiDCTInvGetBufSize_32f (pDCTSpec, &bufSize);
status = ippGetBufSize(pDCTSpec, &bufSize);
if ( status < 0 )
{
ippiDCTInvFree_32f(pDCTSpec);
ippFree(pDCTSpec);
return false;
}
pBuffer = ippsMalloc_8u( bufSize );
buf.allocate( bufSize );
pBuffer = (uchar*)buf;
if (row)
{
for (int i=0; i<src.rows; i++)
{
status = ippFunc((float*)(src.data+i*src.step), (int)src.step,(float*)(dst.data+i*dst.step), (int)dst.step, pDCTSpec, pBuffer);
if ( status < 0 )
{
ippsFree( pBuffer );
ippiDCTInvFree_32f(pDCTSpec);
return false;
}
}
}
else
status = ippFunc((float*)src.data, (int)src.step, (float*)dst.data, (int)dst.step, pDCTSpec, pBuffer);
status = ippFunc((float*)src.data, (int)src.step, (float*)dst.data, (int)dst.step, pDCTSpec, (Ipp8u*)pBuffer);
ippFree( pBuffer );
ippiDCTInvFree_32f(pDCTSpec);
ippFree(pDCTSpec);
CV_SUPPRESS_DEPRECATED_END
return status >= 0;
}
}
#endif
......@@ -3019,7 +3047,7 @@ void cv::dct( InputArray _src0, OutputArray _dst, int flags )
bool inv = (flags & DCT_INVERSE) != 0;
Mat src0 = _src0.getMat(), src = src0;
int type = src.type(), depth = src.depth();
void /* *spec_dft = 0, */ *spec = 0;
void *spec = 0;
double scale = 1.;
int prev_len = 0, nf = 0, stage, end_stage;
......@@ -3036,12 +3064,14 @@ void cv::dct( InputArray _src0, OutputArray _dst, int flags )
_dst.create( src.rows, src.cols, type );
Mat dst = _dst.getMat();
#if defined HAVE_IPP && IPP_VERSION_MAJOR >= 7
#if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7) && !defined HAVE_IPP_ICV_ONLY
bool row = (flags & DCT_ROWS) != 0;
if (inv && ippi_DCT_Inv(src,dst,row))
return;
if(ippi_DCT_Fwd(src,dst,row))
if(!row || src.rows>(int)(1<<8))
{
if(ippi_DCT(src,dst,inv, row))
return;
setIppErrorStatus();
}
#endif
DCTFunc dct_func = dct_tbl[(int)inv + (depth == CV_64F)*2];
......
modules/imgproc/src/distransform.cpp
View file @ 16629bff
......@@ -577,7 +577,7 @@ trueDistTrans( const Mat& src, Mat& dst )
for( ; i <= m*3; i++ )
sat_tab[i] = i - shift;
cv::parallel_for_(cv::Range(0, n), cv::DTColumnInvoker(&src, &dst, sat_tab, sqr_tab));
cv::parallel_for_(cv::Range(0, n), cv::DTColumnInvoker(&src, &dst, sat_tab, sqr_tab), src.total()/(double)(1<<16));
// stage 2: compute modified distance transform for each row
float* inv_tab = sqr_tab + n;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment