Skip to content

O
opencv
Commit 7d94236c authored by Daniil-Osokin's avatar Daniil-Osokin Committed by Daniil Osokin
Browse files
Options

TBB version of calcHist

parent 09be997e
Hide whitespace changes
Inline Side-by-side
Showing with 579 additions and 1 deletion
modules/imgproc/src/histogram.cpp
View file @ 7d94236c
......@@ -165,11 +165,13 @@ static void histPrepareImages( const Mat* images, int nimages, const int* channe
deltas[dims*2 + 1] = (int)(mask.step/mask.elemSize1());
}
#ifndef HAVE_TBB
if( isContinuous )
{
imsize.width *= imsize.height;
imsize.height = 1;
}
#endif
if( !ranges )
{
......@@ -207,6 +209,538 @@ static void histPrepareImages( const Mat* images, int nimages, const int* channe
////////////////////////////////// C A L C U L A T E H I S T O G R A M ////////////////////////////////////
#ifdef HAVE_TBB
enum {one = 1, two, three}; // array elements number
template<typename T>
class calcHist1D_Invoker
{
public:
calcHist1D_Invoker( const vector<uchar*>& _ptrs, const vector<int>& _deltas,
Mat& hist, const double* _uniranges, int sz, int dims,
Size& imageSize )
: mask_(_ptrs[dims]),
mstep_(_deltas[dims*2 + 1]),
imageWidth_(imageSize.width),
histogramSize_(hist.size()), histogramType_(hist.type()),
globalHistogram_((tbb::atomic<int>*)hist.data)
{
p_[0] = ((T**)&_ptrs[0])[0];
step_[0] = (&_deltas[0])[1];
d_[0] = (&_deltas[0])[0];
a_[0] = (&_uniranges[0])[0];
b_[0] = (&_uniranges[0])[1];
size_[0] = sz;
}
void operator()( const BlockedRange& range ) const
{
T* p0 = p_[0] + range.begin() * (step_[0] + imageWidth_*d_[0]);
uchar* mask = mask_ + range.begin()*mstep_;
for( int row = range.begin(); row < range.end(); row++, p0 += step_[0] )
{
if( !mask_ )
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0] )
{
int idx = cvFloor(*p0*a_[0] + b_[0]);
if( (unsigned)idx < (unsigned)size_[0] )
{
globalHistogram_[idx].fetch_and_add(1);
}
}
}
else
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0] )
{
if( mask[x] )
{
int idx = cvFloor(*p0*a_[0] + b_[0]);
if( (unsigned)idx < (unsigned)size_[0] )
{
globalHistogram_[idx].fetch_and_add(1);
}
}
}
mask += mstep_;
}
}
}
private:
T* p_[one];
uchar* mask_;
int step_[one];
int d_[one];
int mstep_;
double a_[one];
double b_[one];
int size_[one];
int imageWidth_;
Size histogramSize_;
int histogramType_;
tbb::atomic<int>* globalHistogram_;
};
template<typename T>
class calcHist2D_Invoker
{
public:
calcHist2D_Invoker( const vector<uchar*>& _ptrs, const vector<int>& _deltas,
Mat& hist, const double* _uniranges, const int* size,
int dims, Size& imageSize, size_t* hstep )
: mask_(_ptrs[dims]),
mstep_(_deltas[dims*2 + 1]),
imageWidth_(imageSize.width),
histogramSize_(hist.size()), histogramType_(hist.type()),
globalHistogram_(hist.data)
{
p_[0] = ((T**)&_ptrs[0])[0]; p_[1] = ((T**)&_ptrs[0])[1];
step_[0] = (&_deltas[0])[1]; step_[1] = (&_deltas[0])[3];
d_[0] = (&_deltas[0])[0]; d_[1] = (&_deltas[0])[2];
a_[0] = (&_uniranges[0])[0]; a_[1] = (&_uniranges[0])[2];
b_[0] = (&_uniranges[0])[1]; b_[1] = (&_uniranges[0])[3];
size_[0] = size[0]; size_[1] = size[1];
hstep_[0] = hstep[0];
}
void operator()(const BlockedRange& range) const
{
T* p0 = p_[0] + range.begin()*(step_[0] + imageWidth_*d_[0]);
T* p1 = p_[1] + range.begin()*(step_[1] + imageWidth_*d_[1]);
uchar* mask = mask_ + range.begin()*mstep_;
for( int row = range.begin(); row < range.end(); row++, p0 += step_[0], p1 += step_[1] )
{
if( !mask_ )
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0], p1 += d_[1] )
{
int idx0 = cvFloor(*p0*a_[0] + b_[0]);
int idx1 = cvFloor(*p1*a_[1] + b_[1]);
if( (unsigned)idx0 < (unsigned)size_[0] && (unsigned)idx1 < (unsigned)size_[1] )
( (tbb::atomic<int>*)(globalHistogram_ + hstep_[0]*idx0) )[idx1].fetch_and_add(1);
}
}
else
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0], p1 += d_[1] )
{
if( mask[x] )
{
int idx0 = cvFloor(*p0*a_[0] + b_[0]);
int idx1 = cvFloor(*p1*a_[1] + b_[1]);
if( (unsigned)idx0 < (unsigned)size_[0] && (unsigned)idx1 < (unsigned)size_[1] )
((tbb::atomic<int>*)(globalHistogram_ + hstep_[0]*idx0))[idx1].fetch_and_add(1);
}
}
mask += mstep_;
}
}
}
private:
T* p_[two];
uchar* mask_;
int step_[two];
int d_[two];
int mstep_;
double a_[two];
double b_[two];
int size_[two];
const int imageWidth_;
size_t hstep_[one];
Size histogramSize_;
int histogramType_;
uchar* globalHistogram_;
};
template<typename T>
class calcHist3D_Invoker
{
public:
calcHist3D_Invoker( const vector<uchar*>& _ptrs, const vector<int>& _deltas,
Size imsize, Mat& hist, const double* uniranges, int _dims,
size_t* hstep, int* size )
: mask_(_ptrs[_dims]),
mstep_(_deltas[_dims*2 + 1]),
imageWidth_(imsize.width),
globalHistogram_(hist.data)
{
p_[0] = ((T**)&_ptrs[0])[0]; p_[1] = ((T**)&_ptrs[0])[1]; p_[2] = ((T**)&_ptrs[0])[2];
step_[0] = (&_deltas[0])[1]; step_[1] = (&_deltas[0])[3]; step_[2] = (&_deltas[0])[5];
d_[0] = (&_deltas[0])[0]; d_[1] = (&_deltas[0])[2]; d_[2] = (&_deltas[0])[4];
a_[0] = uniranges[0]; a_[1] = uniranges[2]; a_[2] = uniranges[4];
b_[0] = uniranges[1]; b_[1] = uniranges[3]; b_[2] = uniranges[5];
size_[0] = size[0]; size_[1] = size[1]; size_[2] = size[2];
hstep_[0] = hstep[0]; hstep_[1] = hstep[1];
}
void operator()( const BlockedRange& range ) const
{
T* p0 = p_[0] + range.begin()*(imageWidth_*d_[0] + step_[0]);
T* p1 = p_[1] + range.begin()*(imageWidth_*d_[1] + step_[1]);
T* p2 = p_[2] + range.begin()*(imageWidth_*d_[2] + step_[2]);
uchar* mask = mask_ + range.begin()*mstep_;
for( int i = range.begin(); i < range.end(); i++, p0 += step_[0], p1 += step_[1], p2 += step_[2] )
{
if( !mask_ )
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0], p1 += d_[1], p2 += d_[2] )
{
int idx0 = cvFloor(*p0*a_[0] + b_[0]);
int idx1 = cvFloor(*p1*a_[1] + b_[1]);
int idx2 = cvFloor(*p2*a_[2] + b_[2]);
if( (unsigned)idx0 < (unsigned)size_[0] &&
(unsigned)idx1 < (unsigned)size_[1] &&
(unsigned)idx2 < (unsigned)size_[2] )
{
( (tbb::atomic<int>*)(globalHistogram_ + hstep_[0]*idx0 + hstep_[1]*idx1) )[idx2].fetch_and_add(1);
}
}
}
else
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0], p1 += d_[1], p2 += d_[2] )
{
if( mask[x] )
{
int idx0 = cvFloor(*p0*a_[0] + b_[0]);
int idx1 = cvFloor(*p1*a_[1] + b_[1]);
int idx2 = cvFloor(*p2*a_[2] + b_[2]);
if( (unsigned)idx0 < (unsigned)size_[0] &&
(unsigned)idx1 < (unsigned)size_[1] &&
(unsigned)idx2 < (unsigned)size_[2] )
{
( (tbb::atomic<int>*)(globalHistogram_ + hstep_[0]*idx0 + hstep_[1]*idx1) )[idx2].fetch_and_add(1);
}
}
}
mask += mstep_;
}
}
}
static bool isFit( const Mat& histogram, const Size imageSize )
{
return ( imageSize.width * imageSize.height >= 320*240
&& histogram.total() >= 8*8*8 );
}
private:
T* p_[three];
uchar* mask_;
int step_[three];
int d_[three];
const int mstep_;
double a_[three];
double b_[three];
int size_[three];
int imageWidth_;
size_t hstep_[two];
uchar* globalHistogram_;
};
class CalcHist1D_8uInvoker
{
public:
CalcHist1D_8uInvoker( const vector<uchar*>& ptrs, const vector<int>& deltas,
Size imsize, Mat& hist, int dims, const vector<size_t>& tab,
tbb::mutex* lock )
: mask_(ptrs[dims]),
mstep_(deltas[dims*2 + 1]),
imageWidth_(imsize.width),
imageSize_(imsize),
histSize_(hist.size()), histType_(hist.type()),
tab_((size_t*)&tab[0]),
histogramWriteLock_(lock),
globalHistogram_(hist.data)
{
p_[0] = (&ptrs[0])[0];
step_[0] = (&deltas[0])[1];
d_[0] = (&deltas[0])[0];
}
void operator()( const BlockedRange& range ) const
{
int localHistogram[256] = { 0, };
uchar* mask = mask_;
uchar* p0 = p_[0];
int x;
tbb::mutex::scoped_lock lock;
if( !mask_ )
{
int n = (imageWidth_ - 4) / 4 + 1;
int tail = imageWidth_ - n*4;
int xN = 4*n;
p0 += (xN*d_[0] + tail*d_[0] + step_[0]) * range.begin();
}
else
{
p0 += (imageWidth_*d_[0] + step_[0]) * range.begin();
mask += mstep_*range.begin();
}
for( int i = range.begin(); i < range.end(); i++, p0 += step_[0] )
{
if( !mask_ )
{
if( d_[0] == 1 )
{
for( x = 0; x <= imageWidth_ - 4; x += 4 )
{
int t0 = p0[x], t1 = p0[x+1];
localHistogram[t0]++; localHistogram[t1]++;
t0 = p0[x+2]; t1 = p0[x+3];
localHistogram[t0]++; localHistogram[t1]++;
}
p0 += x;
}
else
{
for( x = 0; x <= imageWidth_ - 4; x += 4 )
{
int t0 = p0[0], t1 = p0[d_[0]];
localHistogram[t0]++; localHistogram[t1]++;
p0 += d_[0]*2;
t0 = p0[0]; t1 = p0[d_[0]];
localHistogram[t0]++; localHistogram[t1]++;
p0 += d_[0]*2;
}
}
for( ; x < imageWidth_; x++, p0 += d_[0] )
{
localHistogram[*p0]++;
}
}
else
{
for( x = 0; x < imageWidth_; x++, p0 += d_[0] )
{
if( mask[x] )
{
localHistogram[*p0]++;
}
}
mask += mstep_;
}
}
lock.acquire(*histogramWriteLock_);
for(int i = 0; i < 256; i++ )
{
size_t hidx = tab_[i];
if( hidx < OUT_OF_RANGE )
{
*(int*)((globalHistogram_ + hidx)) += localHistogram[i];
}
}
lock.release();
}
static bool isFit( const Mat& histogram, const Size imageSize )
{
return ( histogram.total() >= 8
&& imageSize.width * imageSize.height >= 160*120 );
}
private:
uchar* p_[one];
uchar* mask_;
int mstep_;
int step_[one];
int d_[one];
int imageWidth_;
Size imageSize_;
Size histSize_;
int histType_;
size_t* tab_;
tbb::mutex* histogramWriteLock_;
uchar* globalHistogram_;
};
class CalcHist2D_8uInvoker
{
public:
CalcHist2D_8uInvoker( const vector<uchar*>& _ptrs, const vector<int>& _deltas,
Size imsize, Mat& hist, int dims, const vector<size_t>& _tab,
tbb::mutex* lock )
: mask_(_ptrs[dims]),
mstep_(_deltas[dims*2 + 1]),
imageWidth_(imsize.width),
histSize_(hist.size()), histType_(hist.type()),
tab_((size_t*)&_tab[0]),
histogramWriteLock_(lock),
globalHistogram_(hist.data)
{
p_[0] = (uchar*)(&_ptrs[0])[0]; p_[1] = (uchar*)(&_ptrs[0])[1];
step_[0] = (&_deltas[0])[1]; step_[1] = (&_deltas[0])[3];
d_[0] = (&_deltas[0])[0]; d_[1] = (&_deltas[0])[2];
}
void operator()( const BlockedRange& range ) const
{
uchar* p0 = p_[0] + range.begin()*(step_[0] + imageWidth_*d_[0]);
uchar* p1 = p_[1] + range.begin()*(step_[1] + imageWidth_*d_[1]);
uchar* mask = mask_ + range.begin()*mstep_;
Mat localHist = Mat::zeros(histSize_, histType_);
uchar* localHistData = localHist.data;
tbb::mutex::scoped_lock lock;
for(int i = range.begin(); i < range.end(); i++, p0 += step_[0], p1 += step_[1])
{
if( !mask_ )
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0], p1 += d_[1] )
{
size_t idx = tab_[*p0] + tab_[*p1 + 256];
if( idx < OUT_OF_RANGE )
{
++*(int*)(localHistData + idx);
}
}
}
else
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0], p1 += d_[1] )
{
size_t idx;
if( mask[x] && (idx = tab_[*p0] + tab_[*p1 + 256]) < OUT_OF_RANGE )
{
++*(int*)(localHistData + idx);
}
}
mask += mstep_;
}
}
lock.acquire(*histogramWriteLock_);
for(int i = 0; i < histSize_.width*histSize_.height; i++)
{
((int*)globalHistogram_)[i] += ((int*)localHistData)[i];
}
lock.release();
}
static bool isFit( const Mat& histogram, const Size imageSize )
{
return ( (histogram.total() > 4*4 && histogram.total() <= 116*116
&& imageSize.width * imageSize.height >= 320*240)
|| (histogram.total() > 116*116 && imageSize.width * imageSize.height >= 1280*720) );
}
private:
uchar* p_[two];
uchar* mask_;
int step_[two];
int d_[two];
int mstep_;
int imageWidth_;
Size histSize_;
int histType_;
size_t* tab_;
tbb::mutex* histogramWriteLock_;
uchar* globalHistogram_;
};
class CalcHist3D_8uInvoker
{
public:
CalcHist3D_8uInvoker( const vector<uchar*>& _ptrs, const vector<int>& _deltas,
Size imsize, Mat& hist, int dims, const vector<size_t>& tab )
: mask_(_ptrs[dims]),
mstep_(_deltas[dims*2 + 1]),
histogramSize_(hist.size.p), histogramType_(hist.type()),
imageWidth_(imsize.width),
tab_((size_t*)&tab[0]),
globalHistogram_(hist.data)
{
p_[0] = (uchar*)(&_ptrs[0])[0]; p_[1] = (uchar*)(&_ptrs[0])[1]; p_[2] = (uchar*)(&_ptrs[0])[2];
step_[0] = (&_deltas[0])[1]; step_[1] = (&_deltas[0])[3]; step_[2] = (&_deltas[0])[5];
d_[0] = (&_deltas[0])[0]; d_[1] = (&_deltas[0])[2]; d_[2] = (&_deltas[0])[4];
}
void operator()( const BlockedRange& range ) const
{
uchar* p0 = p_[0] + range.begin()*(step_[0] + imageWidth_*d_[0]);
uchar* p1 = p_[1] + range.begin()*(step_[1] + imageWidth_*d_[1]);
uchar* p2 = p_[2] + range.begin()*(step_[2] + imageWidth_*d_[2]);
uchar* mask = mask_ + range.begin()*mstep_;
for(int i = range.begin(); i < range.end(); i++, p0 += step_[0], p1 += step_[1], p2 += step_[2] )
{
if( !mask_ )
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0], p1 += d_[1], p2 += d_[2] )
{
size_t idx = tab_[*p0] + tab_[*p1 + 256] + tab_[*p2 + 512];
if( idx < OUT_OF_RANGE )
{
( *(tbb::atomic<int>*)(globalHistogram_ + idx) ).fetch_and_add(1);
}
}
}
else
{
for( int x = 0; x < imageWidth_; x++, p0 += d_[0], p1 += d_[1], p2 += d_[2] )
{
size_t idx;
if( mask[x] && (idx = tab_[*p0] + tab_[*p1 + 256] + tab_[*p2 + 512]) < OUT_OF_RANGE )
{
(*(tbb::atomic<int>*)(globalHistogram_ + idx)).fetch_and_add(1);
}
}
mask += mstep_;
}
}
}
static bool isFit( const Mat& histogram, const Size imageSize )
{
return ( histogram.total() >= 128*128*128
&& imageSize.width * imageSize.width >= 320*240 );
}
private:
uchar* p_[three];
uchar* mask_;
int mstep_;
int step_[three];
int d_[three];
int* histogramSize_;
int histogramType_;
int imageWidth_;
size_t* tab_;
uchar* globalHistogram_;
};
static void
callCalcHist2D_8u( vector<uchar*>& _ptrs, const vector<int>& _deltas,
Size imsize, Mat& hist, int dims, vector<size_t>& _tab )
{
int grainSize = imsize.height / tbb::task_scheduler_init::default_num_threads();
tbb::mutex histogramWriteLock;
CalcHist2D_8uInvoker body(_ptrs, _deltas, imsize, hist, dims, _tab, &histogramWriteLock);
parallel_for(BlockedRange(0, imsize.height, grainSize), body);
}
static void
callCalcHist3D_8u( vector<uchar*>& _ptrs, const vector<int>& _deltas,
Size imsize, Mat& hist, int dims, vector<size_t>& _tab )
{
CalcHist3D_8uInvoker body(_ptrs, _deltas, imsize, hist, dims, _tab);
parallel_for(BlockedRange(0, imsize.height), body);
}
#endif
template<typename T> static void
calcHist_( vector<uchar*>& _ptrs, const vector<int>& _deltas,
......@@ -234,6 +768,11 @@ calcHist_( vector<uchar*>& _ptrs, const vector<int>& _deltas,
if( dims == 1 )
{
#ifdef HAVE_TBB
calcHist1D_Invoker<T> body(_ptrs, _deltas, hist, _uniranges, size[0], dims, imsize);
parallel_for(BlockedRange(0, imsize.height), body);
return;
#endif
double a = uniranges[0], b = uniranges[1];
int sz = size[0], d0 = deltas[0], step0 = deltas[1];
const T* p0 = (const T*)ptrs[0];
......@@ -259,6 +798,11 @@ calcHist_( vector<uchar*>& _ptrs, const vector<int>& _deltas,
}
else if( dims == 2 )
{
#ifdef HAVE_TBB
calcHist2D_Invoker<T> body(_ptrs, _deltas, hist, _uniranges, size, dims, imsize, hstep);
parallel_for(BlockedRange(0, imsize.height), body);
return;
#endif
double a0 = uniranges[0], b0 = uniranges[1], a1 = uniranges[2], b1 = uniranges[3];
int sz0 = size[0], sz1 = size[1];
int d0 = deltas[0], step0 = deltas[1],
......@@ -290,6 +834,14 @@ calcHist_( vector<uchar*>& _ptrs, const vector<int>& _deltas,
}
else if( dims == 3 )
{
#ifdef HAVE_TBB
if( calcHist3D_Invoker<T>::isFit(hist, imsize) )
{
calcHist3D_Invoker<T> body(_ptrs, _deltas, imsize, hist, uniranges, dims, hstep, size);
parallel_for(BlockedRange(0, imsize.height), body);
return;
}
#endif
double a0 = uniranges[0], b0 = uniranges[1],
a1 = uniranges[2], b1 = uniranges[3],
a2 = uniranges[4], b2 = uniranges[5];
......@@ -441,8 +993,20 @@ calcHist_8u( vector<uchar*>& _ptrs, const vector<int>& _deltas,
if( dims == 1 )
{
#ifdef HAVE_TBB
if( CalcHist1D_8uInvoker::isFit(hist, imsize) )
{
int treadsNumber = tbb::task_scheduler_init::default_num_threads();
int grainSize = imsize.height/treadsNumber;
tbb::mutex histogramWriteLock;
CalcHist1D_8uInvoker body(_ptrs, _deltas, imsize, hist, dims, _tab, &histogramWriteLock);
parallel_for(BlockedRange(0, imsize.height, grainSize), body);
return;
}
#endif
int d0 = deltas[0], step0 = deltas[1];
int matH[256] = {0};
int matH[256] = { 0, };
const uchar* p0 = (const uchar*)ptrs[0];
for( ; imsize.height--; p0 += step0, mask += mstep )
......@@ -489,6 +1053,13 @@ calcHist_8u( vector<uchar*>& _ptrs, const vector<int>& _deltas,
}
else if( dims == 2 )
{
#ifdef HAVE_TBB
if( CalcHist2D_8uInvoker::isFit(hist, imsize) )
{
callCalcHist2D_8u(_ptrs, _deltas, imsize, hist, dims, _tab);
return;
}
#endif
int d0 = deltas[0], step0 = deltas[1],
d1 = deltas[2], step1 = deltas[3];
const uchar* p0 = (const uchar*)ptrs[0];
......@@ -514,6 +1085,13 @@ calcHist_8u( vector<uchar*>& _ptrs, const vector<int>& _deltas,
}
else if( dims == 3 )
{
#ifdef HAVE_TBB
if( CalcHist3D_8uInvoker::isFit(hist, imsize) )
{
callCalcHist3D_8u(_ptrs, _deltas, imsize, hist, dims, _tab);
return;
}
#endif
int d0 = deltas[0], step0 = deltas[1],
d1 = deltas[2], step1 = deltas[3],
d2 = deltas[4], step2 = deltas[5];
......
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