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

minmaxloc

Browse Source
This commit is contained in:
Ilya Lavrenov
2014-06-04 18:22:55 +04:00
parent f30301d171
commit fd5a8b3e97
3 changed files with 399 additions and 153 deletions
Download Patch File Download Diff File Expand all files Collapse all files

175
modules/core/src/stat.cpp
View File

@@ -1311,104 +1311,157 @@ static void ofs2idx(const Mat& a, size_t ofs, int* idx)
#ifdef HAVE_OPENCL
template <typename T>
void getMinMaxRes(const Mat &minv, const Mat &maxv, const Mat &minl, const Mat &maxl, double* minVal,
double* maxVal, int* minLoc, int* maxLoc, const int groupnum, const int cn, const int cols)
void getMinMaxRes(const Mat & db, double* minVal, double* maxVal,
int* minLoc, int* maxLoc,
int groupnum, int cn, int cols)
{
T min = std::numeric_limits<T>::max();
T max = std::numeric_limits<T>::min() > 0 ? -std::numeric_limits<T>::max() : std::numeric_limits<T>::min();
int minloc = INT_MAX, maxloc = INT_MAX;
for (int i = 0; i < groupnum; i++)
uint index_max = std::numeric_limits<uint>::max();
T minval = std::numeric_limits<T>::max();
T maxval = std::numeric_limits<T>::min() > 0 ? -std::numeric_limits<T>::max() : std::numeric_limits<T>::min();
uint minloc = index_max, maxloc = index_max;
int index = 0;
const T * minptr = NULL, * maxptr = NULL;
const uint * minlocptr = NULL, * maxlocptr = NULL;
if (minVal || minLoc)
{
T current_min = minv.at<T>(0,i);
T current_max = maxv.at<T>(0,i);
T oldmin = min, oldmax = max;
min = std::min(min, current_min);
max = std::max(max, current_max);
if (cn == 1)
{
int current_minloc = minl.at<int>(0,i);
int current_maxloc = maxl.at<int>(0,i);
if(current_minloc < 0 || current_maxloc < 0) continue;
minloc = (oldmin == current_min) ? std::min(minloc, current_minloc) : (oldmin < current_min) ? minloc : current_minloc;
maxloc = (oldmax == current_max) ? std::min(maxloc, current_maxloc) : (oldmax > current_max) ? maxloc : current_maxloc;
}
minptr = (const T *)db.data;
index += sizeof(T) * groupnum;
}
if (maxVal || maxLoc)
{
maxptr = (const T *)(db.data + index);
index += sizeof(T) * groupnum;
}
bool zero_mask = (maxloc == INT_MAX) || (minloc == INT_MAX);
if (minVal)
*minVal = zero_mask ? 0 : (double)min;
if (maxVal)
*maxVal = zero_mask ? 0 : (double)max;
if (minLoc)
{
minLoc[0] = zero_mask ? -1 : minloc/cols;
minLoc[1] = zero_mask ? -1 : minloc%cols;
minlocptr = (uint *)(db.data + index);
index += sizeof(uint) * groupnum;
}
if (maxLoc)
maxlocptr = (uint *)(db.data + index);
for (int i = 0; i < groupnum; i++)
{
if (minptr && minptr[i] <= minval)
{
if (minptr[i] == minval)
{
if (minlocptr)
minloc = std::min(minlocptr[i], minloc);
}
else
{
if (minlocptr)
minloc = minlocptr[i];
minval = minptr[i];
}
}
if (maxptr && maxptr[i] >= maxval)
{
if (maxptr[i] == maxval)
{
if (maxlocptr)
maxloc = std::min(maxlocptr[i], maxloc);
}
else
{
if (maxlocptr)
maxloc = maxlocptr[i];
maxval = maxptr[i];
}
}
}
bool zero_mask = (minLoc && minloc == index_max) ||
(maxLoc && maxloc == index_max);
if (minVal)
*minVal = zero_mask ? 0 : (double)minval;
if (maxVal)
*maxVal = zero_mask ? 0 : (double)maxval;
if (minLoc)
{
minLoc[0] = zero_mask ? -1 : minloc / cols;
minLoc[1] = zero_mask ? -1 : minloc % cols;
}
if (maxLoc)
{
maxLoc[0] = zero_mask ? -1 : maxloc/cols;
maxLoc[1] = zero_mask ? -1 : maxloc%cols;
maxLoc[0] = zero_mask ? -1 : maxloc / cols;
maxLoc[1] = zero_mask ? -1 : maxloc % cols;
}
}
typedef void (*getMinMaxResFunc)(const Mat &minv, const Mat &maxv, const Mat &minl, const Mat &maxl, double *minVal,
double *maxVal, int *minLoc, int *maxLoc, const int gropunum, const int cn, const int cols);
typedef void (*getMinMaxResFunc)(const Mat & db, double *minVal, double *maxVal,
int *minLoc, int *maxLoc,
int gropunum, int cn, int cols);
static bool ocl_minMaxIdx( InputArray _src, double* minVal, double* maxVal, int* minLoc, int* maxLoc, InputArray _mask)
{
CV_Assert( (_src.channels() == 1 && (_mask.empty() || _mask.type() == CV_8U)) ||
(_src.channels() >= 1 && _mask.empty() && !minLoc && !maxLoc) );
int type = _src.type(), depth = CV_MAT_DEPTH(type), kercn = 1;
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
const ocl::Device & dev = ocl::Device::getDefault();
bool doubleSupport = dev.doubleFPConfig() > 0, haveMask = !_mask.empty();
int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
kercn = haveMask ? 1 : std::min(4, ocl::predictOptimalVectorWidth(_src));
if (depth == CV_64F && !doubleSupport)
return false;
int groupnum = ocl::Device::getDefault().maxComputeUnits();
size_t wgs = ocl::Device::getDefault().maxWorkGroupSize();
int groupnum = dev.maxComputeUnits();
size_t wgs = dev.maxWorkGroupSize();
int wgs2_aligned = 1;
while (wgs2_aligned < (int)wgs)
wgs2_aligned <<= 1;
wgs2_aligned >>= 1;
String opts = format("-D DEPTH_%d -D srcT=%s -D OP_MIN_MAX_LOC%s -D WGS=%d"
" -D WGS2_ALIGNED=%d%s%s%s -D kercn=%d",
depth, ocl::typeToStr(depth), _mask.empty() ? "" : "_MASK", (int)wgs,
wgs2_aligned, doubleSupport ? " -D DOUBLE_SUPPORT" : "",
_src.isContinuous() ? " -D HAVE_SRC_CONT" : "",
_mask.isContinuous() ? " -D HAVE_MASK_CONT" : "", kercn);
bool needMinVal = minVal || minLoc, needMinLoc = minLoc != NULL,
needMaxVal = maxVal || maxLoc, needMaxLoc = maxLoc != NULL;
ocl::Kernel k("reduce", ocl::core::reduce_oclsrc, opts);
// in case of mask we must know whether mask is filled with zeros or not
// so let's calculate min or max location, if it's undefined, so mask is zeros
if (!(needMaxLoc || needMinLoc) && haveMask)
if (needMinVal)
needMinLoc = true;
else
needMaxVal = true;
String opts = format("-D DEPTH_%d -D srcT1=%s%s -D WGS=%d -D srcT=%s"
" -D WGS2_ALIGNED=%d%s%s%s -D kercn=%d%s%s%s%s",
depth, ocl::typeToStr(depth), haveMask ? " -D HAVE_MASK" : "", (int)wgs,
ocl::typeToStr(CV_MAKE_TYPE(depth, kercn)), wgs2_aligned,
doubleSupport ? " -D DOUBLE_SUPPORT" : "",
_src.isContinuous() ? " -D HAVE_SRC_CONT" : "",
_mask.isContinuous() ? " -D HAVE_MASK_CONT" : "", kercn,
needMinVal ? " -D NEED_MINVAL" : "", needMaxVal ? " -D NEED_MAXVAL" : "",
needMinLoc ? " -D NEED_MINLOC" : "", needMaxLoc ? " -D NEED_MAXLOC" : "");
ocl::Kernel k("minmaxloc", ocl::core::minmaxloc_oclsrc, opts);
if (k.empty())
return false;
UMat src = _src.getUMat(), minval(1, groupnum, src.type()),
maxval(1, groupnum, src.type()), minloc( 1, groupnum, CV_32SC1),
maxloc( 1, groupnum, CV_32SC1), mask;
if (!_mask.empty())
mask = _mask.getUMat();
int esz = CV_ELEM_SIZE(depth), esz32s = CV_ELEM_SIZE1(CV_32S),
dbsize = groupnum * ((needMinVal ? esz : 0) + (needMaxVal ? esz : 0) +
(needMinLoc ? esz32s : 0) + (needMaxLoc ? esz32s : 0));
UMat src = _src.getUMat(), db(1, dbsize, CV_8UC1), mask = _mask.getUMat();
if (src.channels() > 1)
if (cn > 1)
src = src.reshape(1);
if (mask.empty())
if (!haveMask)
k.args(ocl::KernelArg::ReadOnlyNoSize(src), src.cols, (int)src.total(),
groupnum, ocl::KernelArg::PtrWriteOnly(minval), ocl::KernelArg::PtrWriteOnly(maxval),
ocl::KernelArg::PtrWriteOnly(minloc), ocl::KernelArg::PtrWriteOnly(maxloc));
groupnum, ocl::KernelArg::PtrWriteOnly(db));
else
k.args(ocl::KernelArg::ReadOnlyNoSize(src), src.cols, (int)src.total(), groupnum,
ocl::KernelArg::PtrWriteOnly(minval), ocl::KernelArg::PtrWriteOnly(maxval),
ocl::KernelArg::PtrWriteOnly(minloc), ocl::KernelArg::PtrWriteOnly(maxloc), ocl::KernelArg::ReadOnlyNoSize(mask));
k.args(ocl::KernelArg::ReadOnlyNoSize(src), src.cols, (int)src.total(),
groupnum, ocl::KernelArg::PtrWriteOnly(db), ocl::KernelArg::ReadOnlyNoSize(mask));
size_t globalsize = groupnum * wgs;
if (!k.run(1, &globalsize, &wgs, false))
return false;
Mat minv = minval.getMat(ACCESS_READ), maxv = maxval.getMat(ACCESS_READ),
minl = minloc.getMat(ACCESS_READ), maxl = maxloc.getMat(ACCESS_READ);
static getMinMaxResFunc functab[7] =
static const getMinMaxResFunc functab[7] =
{
getMinMaxRes<uchar>,
getMinMaxRes<char>,
@@ -1419,10 +1472,12 @@ static bool ocl_minMaxIdx( InputArray _src, double* minVal, double* maxVal, int*
getMinMaxRes<double>
};
getMinMaxResFunc func;
getMinMaxResFunc func = functab[depth];
func = functab[depth];
func(minv, maxv, minl, maxl, minVal, maxVal, minLoc, maxLoc, groupnum, src.channels(), src.cols);
int locTemp[2];
func(db.getMat(ACCESS_READ), minVal, maxVal,
needMinLoc ? minLoc ? minLoc : locTemp : minLoc,
needMaxLoc ? maxLoc ? maxLoc : locTemp : maxLoc, groupnum, cn, src.cols);
return true;
}