static bool matchTemplate_SQDIFF_NORMED(InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
int type = _image.type(), cn = CV_MAT_CN(type);
ocl::Kernel k("matchTemplate_SQDIFF_NORMED", ocl::imgproc::match_template_oclsrc,
format("-D SQDIFF_NORMED -D T=%s -D cn=%d", ocl::typeToStr(type), cn));
if (k.empty())
return false;
UMat image = _image.getUMat(), templ = _templ.getUMat();
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
UMat result = _result.getUMat();
UMat image_sums, image_sqsums;
integral(image.reshape(1), image_sums, image_sqsums, CV_32F, CV_32F);
UMat templ_sqsum;
if (!sumTemplate(_templ, templ_sqsum))
return false;
k.args(ocl::KernelArg::ReadOnlyNoSize(image_sqsums), ocl::KernelArg::ReadWrite(result),
templ.rows, templ.cols, ocl::KernelArg::PtrReadOnly(templ_sqsum));
size_t globalsize[2] = { result.cols, result.rows };
return k.run(2, globalsize, NULL, false);
}
static bool ocl_accumulate( InputArray _src, InputArray _src2, InputOutputArray _dst, double alpha,
InputArray _mask, int op_type )
{
CV_Assert(op_type == ACCUMULATE || op_type == ACCUMULATE_SQUARE ||
op_type == ACCUMULATE_PRODUCT || op_type == ACCUMULATE_WEIGHTED);
int stype = _src.type(), cn = CV_MAT_CN(stype);
int sdepth = CV_MAT_DEPTH(stype), ddepth = _dst.depth();
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0,
haveMask = !_mask.empty();
if (!doubleSupport && (sdepth == CV_64F || ddepth == CV_64F))
return false;
const char * const opMap[4] = { "ACCUMULATE", "ACCUMULATE_SQUARE", "ACCUMULATE_PRODUCT",
"ACCUMULATE_WEIGHTED" };
ocl::Kernel k("accumulate", ocl::imgproc::accumulate_oclsrc,
format("-D %s%s -D srcT=%s -D cn=%d -D dstT=%s%s",
opMap[op_type], haveMask ? " -D HAVE_MASK" : "",
ocl::typeToStr(sdepth), cn, ocl::typeToStr(ddepth),
doubleSupport ? " -D DOUBLE_SUPPORT" : ""));
if (k.empty())
return false;
UMat src = _src.getUMat(), src2 = _src2.getUMat(), dst = _dst.getUMat(), mask = _mask.getUMat();
ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
src2arg = ocl::KernelArg::ReadOnlyNoSize(src2),
dstarg = ocl::KernelArg::ReadWrite(dst),
maskarg = ocl::KernelArg::ReadOnlyNoSize(mask);
int argidx = k.set(0, srcarg);
if (op_type == ACCUMULATE_PRODUCT)
argidx = k.set(argidx, src2arg);
argidx = k.set(argidx, dstarg);
if (op_type == ACCUMULATE_WEIGHTED)
{
if (ddepth == CV_32F)
argidx = k.set(argidx, (float)alpha);
else
argidx = k.set(argidx, alpha);
}
if (haveMask)
k.set(argidx, maskarg);
size_t globalsize[2] = { src.cols, src.rows };
return k.run(2, globalsize, NULL, false);
}
static bool ocl_dot( InputArray _src1, InputArray _src2, double & res )
{
UMat src1 = _src1.getUMat().reshape(1), src2 = _src2.getUMat().reshape(1);
int type = src1.type(), depth = CV_MAT_DEPTH(type),
kercn = ocl::predictOptimalVectorWidth(src1, src2);
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
if ( !doubleSupport && depth == CV_64F )
return false;
int dbsize = ocl::Device::getDefault().maxComputeUnits();
size_t wgs = ocl::Device::getDefault().maxWorkGroupSize();
int ddepth = std::max(CV_32F, depth);
int wgs2_aligned = 1;
while (wgs2_aligned < (int)wgs)
wgs2_aligned <<= 1;
wgs2_aligned >>= 1;
char cvt[40];
ocl::Kernel k("reduce", ocl::core::reduce_oclsrc,
format("-D srcT=%s -D srcT1=%s -D dstT=%s -D dstTK=%s -D ddepth=%d -D convertToDT=%s -D OP_DOT "
"-D WGS=%d -D WGS2_ALIGNED=%d%s%s%s -D kercn=%d",
ocl::typeToStr(CV_MAKE_TYPE(depth, kercn)), ocl::typeToStr(depth),
ocl::typeToStr(ddepth), ocl::typeToStr(CV_MAKE_TYPE(ddepth, kercn)),
ddepth, ocl::convertTypeStr(depth, ddepth, kercn, cvt),
(int)wgs, wgs2_aligned, doubleSupport ? " -D DOUBLE_SUPPORT" : "",
_src1.isContinuous() ? " -D HAVE_SRC_CONT" : "",
_src2.isContinuous() ? " -D HAVE_SRC2_CONT" : "", kercn));
if (k.empty())
return false;
UMat db(1, dbsize, ddepth);
ocl::KernelArg src1arg = ocl::KernelArg::ReadOnlyNoSize(src1),
src2arg = ocl::KernelArg::ReadOnlyNoSize(src2),
dbarg = ocl::KernelArg::PtrWriteOnly(db);
k.args(src1arg, src1.cols, (int)src1.total(), dbsize, dbarg, src2arg);
size_t globalsize = dbsize * wgs;
if (k.run(1, &globalsize, &wgs, false))
{
res = sum(db.getMat(ACCESS_READ))[0];
return true;
}
return false;
}
static bool ocl_fastNlMeansDenoisingColored( InputArray _src, OutputArray _dst,
float h, float hForColorComponents,
int templateWindowSize, int searchWindowSize)
{
UMat src = _src.getUMat();
_dst.create(src.size(), src.type());
UMat dst = _dst.getUMat();
UMat src_lab;
cvtColor(src, src_lab, COLOR_LBGR2Lab);
UMat l(src.size(), CV_8U);
UMat ab(src.size(), CV_8UC2);
std::vector<UMat> l_ab(2), l_ab_denoised(2);
l_ab[0] = l;
l_ab[1] = ab;
l_ab_denoised[0].create(src.size(), CV_8U);
l_ab_denoised[1].create(src.size(), CV_8UC2);
int from_to[] = { 0,0, 1,1, 2,2 };
mixChannels(std::vector<UMat>(1, src_lab), l_ab, from_to, 3);
fastNlMeansDenoising(l_ab[0], l_ab_denoised[0], h, templateWindowSize, searchWindowSize);
fastNlMeansDenoising(l_ab[1], l_ab_denoised[1], hForColorComponents, templateWindowSize, searchWindowSize);
UMat dst_lab(src.size(), CV_8UC3);
mixChannels(l_ab_denoised, std::vector<UMat>(1, dst_lab), from_to, 3);
cvtColor(dst_lab, dst, COLOR_Lab2LBGR, src.channels());
return true;
}
static bool ocl_integral( InputArray _src, OutputArray _sum, int sdepth )
{
if ( _src.type() != CV_8UC1 || _src.step() % vlen != 0 || _src.offset() % vlen != 0 ||
!(sdepth == CV_32S || sdepth == CV_32F) )
return false;
ocl::Kernel k1("integral_sum_cols", ocl::imgproc::integral_sum_oclsrc,
format("-D sdepth=%d", sdepth));
if (k1.empty())
return false;
Size size = _src.size(), t_size = Size(((size.height + vlen - 1) / vlen) * vlen, size.width),
ssize(size.width + 1, size.height + 1);
_sum.create(ssize, sdepth);
UMat src = _src.getUMat(), t_sum(t_size, sdepth), sum = _sum.getUMat();
t_sum = t_sum(Range::all(), Range(0, size.height));
int offset = (int)src.offset / vlen, pre_invalid = (int)src.offset % vlen;
int vcols = (pre_invalid + src.cols + vlen - 1) / vlen;
int sum_offset = (int)sum.offset / vlen;
k1.args(ocl::KernelArg::PtrReadOnly(src), ocl::KernelArg::PtrWriteOnly(t_sum),
offset, pre_invalid, src.rows, src.cols, (int)src.step, (int)t_sum.step);
size_t gt = ((vcols + 1) / 2) * 256, lt = 256;
if (!k1.run(1, >, <, false))
return false;
ocl::Kernel k2("integral_sum_rows", ocl::imgproc::integral_sum_oclsrc,
format("-D sdepth=%d", sdepth));
k2.args(ocl::KernelArg::PtrReadWrite(t_sum), ocl::KernelArg::PtrWriteOnly(sum),
t_sum.rows, t_sum.cols, (int)t_sum.step, (int)sum.step, sum_offset);
size_t gt2 = t_sum.cols * 32, lt2 = 256;
return k2.run(1, >2, <2, false);
}
static bool matchTemplate_CCOEFF(InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
UMat image_sums, temp;
integral(_image, image_sums, CV_32F);
int type = image_sums.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
ocl::Kernel k("matchTemplate_Prepared_CCOEFF", ocl::imgproc::match_template_oclsrc,
format("-D CCOEFF -D T=%s -D T1=%s -D cn=%d", ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat templ = _templ.getUMat();
UMat result = _result.getUMat();
if (cn==1)
{
Scalar templMean = mean(templ);
float templ_sum = (float)templMean[0];
k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadWrite(result), templ.rows, templ.cols, templ_sum);
}
else
{
Vec4f templ_sum = Vec4f::all(0);
templ_sum = (Vec4f)mean(templ);
k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadWrite(result), templ.rows, templ.cols, templ_sum); }
size_t globalsize[2] = { result.cols, result.rows };
return k.run(2, globalsize, NULL, false);
}
static bool ocl_threshold( InputArray _src, OutputArray _dst, double & thresh, double maxval, int thresh_type )
{
int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
kercn = ocl::predictOptimalVectorWidth(_src, _dst), ktype = CV_MAKE_TYPE(depth, kercn);
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
if ( !(thresh_type == THRESH_BINARY || thresh_type == THRESH_BINARY_INV || thresh_type == THRESH_TRUNC ||
thresh_type == THRESH_TOZERO || thresh_type == THRESH_TOZERO_INV) ||
(!doubleSupport && depth == CV_64F))
return false;
const char * const thresholdMap[] = { "THRESH_BINARY", "THRESH_BINARY_INV", "THRESH_TRUNC",
"THRESH_TOZERO", "THRESH_TOZERO_INV" };
ocl::Kernel k("threshold", ocl::imgproc::threshold_oclsrc,
format("-D %s -D T=%s -D T1=%s%s", thresholdMap[thresh_type],
ocl::typeToStr(ktype), ocl::typeToStr(depth),
doubleSupport ? " -D DOUBLE_SUPPORT" : ""));
if (k.empty())
return false;
UMat src = _src.getUMat();
_dst.create(src.size(), type);
UMat dst = _dst.getUMat();
if (depth <= CV_32S)
thresh = cvFloor(thresh);
k.args(ocl::KernelArg::ReadOnlyNoSize(src), ocl::KernelArg::WriteOnly(dst, cn, kercn),
ocl::KernelArg::Constant(Mat(1, 1, depth, Scalar::all(thresh))),
ocl::KernelArg::Constant(Mat(1, 1, depth, Scalar::all(maxval))));
size_t globalsize[2] = { dst.cols * cn / kercn, dst.rows };
return k.run(2, globalsize, NULL, false);
}
void detect( InputArray _image, std::vector<KeyPoint>& keypoints, InputArray _mask )
{
CV_INSTRUMENT_REGION()
std::vector<Point2f> corners;
if (_image.isUMat())
{
UMat ugrayImage;
if( _image.type() != CV_8U )
cvtColor( _image, ugrayImage, COLOR_BGR2GRAY );
else
ugrayImage = _image.getUMat();
goodFeaturesToTrack( ugrayImage, corners, nfeatures, qualityLevel, minDistance, _mask,
blockSize, useHarrisDetector, k );
}
else
{
Mat image = _image.getMat(), grayImage = image;
if( image.type() != CV_8U )
cvtColor( image, grayImage, COLOR_BGR2GRAY );
goodFeaturesToTrack( grayImage, corners, nfeatures, qualityLevel, minDistance, _mask,
blockSize, useHarrisDetector, k );
}
keypoints.resize(corners.size());
std::vector<Point2f>::const_iterator corner_it = corners.begin();
std::vector<KeyPoint>::iterator keypoint_it = keypoints.begin();
for( ; corner_it != corners.end(); ++corner_it, ++keypoint_it )
*keypoint_it = KeyPoint( *corner_it, (float)blockSize );
}
static bool sumTemplate(InputArray _src, UMat & result)
{
int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
int wdepth = CV_32F, wtype = CV_MAKE_TYPE(wdepth, cn);
size_t wgs = ocl::Device::getDefault().maxWorkGroupSize();
int wgs2_aligned = 1;
while (wgs2_aligned < (int)wgs)
wgs2_aligned <<= 1;
wgs2_aligned >>= 1;
char cvt[40];
ocl::Kernel k("calcSum", ocl::imgproc::match_template_oclsrc,
format("-D CALC_SUM -D T=%s -D T1=%s -D WT=%s -D cn=%d -D convertToWT=%s -D WGS=%d -D WGS2_ALIGNED=%d",
ocl::typeToStr(type), ocl::typeToStr(depth), ocl::typeToStr(wtype), cn,
ocl::convertTypeStr(depth, wdepth, cn, cvt),
(int)wgs, wgs2_aligned));
if (k.empty())
return false;
UMat src = _src.getUMat();
result.create(1, 1, CV_32FC1);
ocl::KernelArg srcarg = ocl::KernelArg::ReadOnlyNoSize(src),
resarg = ocl::KernelArg::PtrWriteOnly(result);
k.args(srcarg, src.cols, (int)src.total(), resarg);
size_t globalsize = wgs;
return k.run(1, &globalsize, &wgs, false);
}
static bool convolve_32F(InputArray _image, InputArray _templ, OutputArray _result)
{
_result.create(_image.rows() - _templ.rows() + 1, _image.cols() - _templ.cols() + 1, CV_32F);
if (_image.channels() == 1)
return(convolve_dft(_image, _templ, _result));
else
{
UMat image = _image.getUMat();
UMat templ = _templ.getUMat();
UMat result_(image.rows-templ.rows+1,(image.cols-templ.cols+1)*image.channels(), CV_32F);
bool ok = convolve_dft(image.reshape(1), templ.reshape(1), result_);
if (ok==false)
return false;
UMat result = _result.getUMat();
return (extractFirstChannel_32F(result_, _result, _image.channels()));
}
}
static bool ocl_integral( InputArray _src, OutputArray _sum, OutputArray _sqsum, int sdepth, int sqdepth )
{
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
if ( _src.type() != CV_8UC1 || _src.step() % vlen != 0 || _src.offset() % vlen != 0 ||
(!doubleSupport && (sdepth == CV_64F || sqdepth == CV_64F)) )
return false;
char cvt[40];
String opts = format("-D sdepth=%d -D sqdepth=%d -D TYPE=%s -D TYPE4=%s4 -D convert_TYPE4=%s%s",
sdepth, sqdepth, ocl::typeToStr(sqdepth), ocl::typeToStr(sqdepth),
ocl::convertTypeStr(sdepth, sqdepth, 4, cvt),
doubleSupport ? " -D DOUBLE_SUPPORT" : "");
ocl::Kernel k1("integral_cols", ocl::imgproc::integral_sqrsum_oclsrc, opts);
if (k1.empty())
return false;
Size size = _src.size(), dsize = Size(size.width + 1, size.height + 1),
t_size = Size(((size.height + vlen - 1) / vlen) * vlen, size.width);
UMat src = _src.getUMat(), t_sum(t_size, sdepth), t_sqsum(t_size, sqdepth);
t_sum = t_sum(Range::all(), Range(0, size.height));
t_sqsum = t_sqsum(Range::all(), Range(0, size.height));
_sum.create(dsize, sdepth);
_sqsum.create(dsize, sqdepth);
UMat sum = _sum.getUMat(), sqsum = _sqsum.getUMat();
int offset = src.offset / vlen;
int pre_invalid = src.offset % vlen;
int vcols = (pre_invalid + src.cols + vlen - 1) / vlen;
int sum_offset = sum.offset / sum.elemSize();
int sqsum_offset = sqsum.offset / sqsum.elemSize();
CV_Assert(sqsum.offset % sqsum.elemSize() == 0);
k1.args(ocl::KernelArg::PtrReadOnly(src), ocl::KernelArg::PtrWriteOnly(t_sum),
ocl::KernelArg::PtrWriteOnly(t_sqsum), offset, pre_invalid, src.rows,
src.cols, (int)src.step, (int)t_sum.step, (int)t_sqsum.step);
size_t gt = ((vcols + 1) / 2) * 256, lt = 256;
if (!k1.run(1, >, <, false))
return false;
ocl::Kernel k2("integral_rows", ocl::imgproc::integral_sqrsum_oclsrc, opts);
if (k2.empty())
return false;
k2.args(ocl::KernelArg::PtrReadOnly(t_sum), ocl::KernelArg::PtrReadOnly(t_sqsum),
ocl::KernelArg::PtrWriteOnly(sum), ocl::KernelArg::PtrWriteOnly(sqsum),
t_sum.rows, t_sum.cols, (int)t_sum.step, (int)t_sqsum.step,
(int)sum.step, (int)sqsum.step, sum_offset, sqsum_offset);
size_t gt2 = t_sum.cols * 32, lt2 = 256;
return k2.run(1, >2, <2, false);
}
static bool ocl_sepFilter3x3_8UC1(InputArray _src, OutputArray _dst, int ddepth,
InputArray _kernelX, InputArray _kernelY, double delta, int borderType)
{
const ocl::Device & dev = ocl::Device::getDefault();
int type = _src.type(), sdepth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
if ( !(dev.isIntel() && (type == CV_8UC1) && (ddepth == CV_8U) &&
(_src.offset() == 0) && (_src.step() % 4 == 0) &&
(_src.cols() % 16 == 0) && (_src.rows() % 2 == 0)) )
return false;
Mat kernelX = _kernelX.getMat().reshape(1, 1);
if (kernelX.cols % 2 != 1)
return false;
Mat kernelY = _kernelY.getMat().reshape(1, 1);
if (kernelY.cols % 2 != 1)
return false;
if (ddepth < 0)
ddepth = sdepth;
Size size = _src.size();
size_t globalsize[2] = { 0, 0 };
size_t localsize[2] = { 0, 0 };
globalsize[0] = size.width / 16;
globalsize[1] = size.height / 2;
const char * const borderMap[] = { "BORDER_CONSTANT", "BORDER_REPLICATE", "BORDER_REFLECT", 0, "BORDER_REFLECT_101" };
char build_opts[1024];
sprintf(build_opts, "-D %s %s%s", borderMap[borderType],
ocl::kernelToStr(kernelX, CV_32F, "KERNEL_MATRIX_X").c_str(),
ocl::kernelToStr(kernelY, CV_32F, "KERNEL_MATRIX_Y").c_str());
ocl::Kernel kernel("sepFilter3x3_8UC1_cols16_rows2", cv::ocl::imgproc::sepFilter3x3_oclsrc, build_opts);
if (kernel.empty())
return false;
UMat src = _src.getUMat();
_dst.create(size, CV_MAKETYPE(ddepth, cn));
if (!(_dst.offset() == 0 && _dst.step() % 4 == 0))
return false;
UMat dst = _dst.getUMat();
int idxArg = kernel.set(0, ocl::KernelArg::PtrReadOnly(src));
idxArg = kernel.set(idxArg, (int)src.step);
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(dst));
idxArg = kernel.set(idxArg, (int)dst.step);
idxArg = kernel.set(idxArg, (int)dst.rows);
idxArg = kernel.set(idxArg, (int)dst.cols);
idxArg = kernel.set(idxArg, static_cast<float>(delta));
return kernel.run(2, globalsize, (localsize[0] == 0) ? NULL : localsize, false);
}
static bool matchTemplate_CCORR(InputArray _image, InputArray _templ, OutputArray _result)
{
if (useNaive(_templ.size()))
return( matchTemplateNaive_CCORR(_image, _templ, _result));
else
{
if(_image.depth() == CV_8U)
{
UMat imagef, templf;
UMat image = _image.getUMat();
UMat templ = _templ.getUMat();
image.convertTo(imagef, CV_32F);
templ.convertTo(templf, CV_32F);
return(convolve_32F(imagef, templf, _result));
}
else
{
return(convolve_32F(_image, _templ, _result));
}
}
}
static bool ocl_blendLinear( InputArray _src1, InputArray _src2, InputArray _weights1, InputArray _weights2, OutputArray _dst )
{
int type = _src1.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
char cvt[30];
ocl::Kernel k("blendLinear", ocl::imgproc::blend_linear_oclsrc,
format("-D T=%s -D cn=%d -D convertToT=%s", ocl::typeToStr(depth),
cn, ocl::convertTypeStr(CV_32F, depth, 1, cvt)));
if (k.empty())
return false;
UMat src1 = _src1.getUMat(), src2 = _src2.getUMat(), weights1 = _weights1.getUMat(),
weights2 = _weights2.getUMat(), dst = _dst.getUMat();
k.args(ocl::KernelArg::ReadOnlyNoSize(src1), ocl::KernelArg::ReadOnlyNoSize(src2),
ocl::KernelArg::ReadOnlyNoSize(weights1), ocl::KernelArg::ReadOnlyNoSize(weights2),
ocl::KernelArg::WriteOnly(dst));
size_t globalsize[2] = { (size_t)dst.cols, (size_t)dst.rows };
return k.run(2, globalsize, NULL, false);
}
static bool ocl_repeat(InputArray _src, int ny, int nx, OutputArray _dst)
{
UMat src = _src.getUMat(), dst = _dst.getUMat();
for (int y = 0; y < ny; ++y)
for (int x = 0; x < nx; ++x)
{
Rect roi(x * src.cols, y * src.rows, src.cols, src.rows);
UMat hdr(dst, roi);
src.copyTo(hdr);
}
return true;
}
static bool matchTemplateNaive_SQDIFF(InputArray _image, InputArray _templ, OutputArray _result)
{
int type = _image.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
int wdepth = CV_32F, wtype = CV_MAKE_TYPE(wdepth, cn);
char cvt[40];
ocl::Kernel k("matchTemplate_Naive_SQDIFF", ocl::imgproc::match_template_oclsrc,
format("-D SQDIFF -D T=%s -D T1=%s -D WT=%s -D convertToWT=%s -D cn=%d", ocl::typeToStr(type), ocl::typeToStr(depth),
ocl::typeToStr(wtype), ocl::convertTypeStr(depth, wdepth, cn, cvt), cn));
if (k.empty())
return false;
UMat image = _image.getUMat(), templ = _templ.getUMat();
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
UMat result = _result.getUMat();
k.args(ocl::KernelArg::ReadOnlyNoSize(image), ocl::KernelArg::ReadOnly(templ),
ocl::KernelArg::WriteOnly(result));
size_t globalsize[2] = { result.cols, result.rows };
return k.run(2, globalsize, NULL, false);
}
static bool matchTemplateNaive_CCORR(InputArray _image, InputArray _templ, OutputArray _result)
{
int type = _image.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
int wdepth = CV_32F, wtype = CV_MAKE_TYPE(wdepth, cn);
ocl::Device dev = ocl::Device::getDefault();
int pxPerWIx = (cn==1 && dev.isIntel() && (dev.type() & ocl::Device::TYPE_GPU)) ? 4 : 1;
int rated_cn = cn;
int wtype1 = wtype;
if (pxPerWIx!=1)
{
rated_cn = pxPerWIx;
type = CV_MAKE_TYPE(depth, rated_cn);
wtype1 = CV_MAKE_TYPE(wdepth, rated_cn);
}
char cvt[40];
char cvt1[40];
const char* convertToWT1 = ocl::convertTypeStr(depth, wdepth, cn, cvt);
const char* convertToWT = ocl::convertTypeStr(depth, wdepth, rated_cn, cvt1);
ocl::Kernel k("matchTemplate_Naive_CCORR", ocl::imgproc::match_template_oclsrc,
format("-D CCORR -D T=%s -D T1=%s -D WT=%s -D WT1=%s -D convertToWT=%s -D convertToWT1=%s -D cn=%d -D PIX_PER_WI_X=%d", ocl::typeToStr(type), ocl::typeToStr(depth), ocl::typeToStr(wtype1), ocl::typeToStr(wtype),
convertToWT, convertToWT1, cn, pxPerWIx));
if (k.empty())
return false;
UMat image = _image.getUMat(), templ = _templ.getUMat();
_result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32FC1);
UMat result = _result.getUMat();
k.args(ocl::KernelArg::ReadOnlyNoSize(image), ocl::KernelArg::ReadOnly(templ),
ocl::KernelArg::WriteOnly(result));
size_t globalsize[2] = { (result.cols+pxPerWIx-1)/pxPerWIx, result.rows};
return k.run(2, globalsize, NULL, false);
}
static bool ocl_MultiBandBlender_feed(InputArray _src, InputArray _weight,
InputOutputArray _dst, InputOutputArray _dst_weight)
{
String buildOptions = "-D DEFINE_feed";
ocl::buildOptionsAddMatrixDescription(buildOptions, "src", _src);
ocl::buildOptionsAddMatrixDescription(buildOptions, "weight", _weight);
ocl::buildOptionsAddMatrixDescription(buildOptions, "dst", _dst);
ocl::buildOptionsAddMatrixDescription(buildOptions, "dstWeight", _dst_weight);
ocl::Kernel k("feed", ocl::stitching::multibandblend_oclsrc, buildOptions);
if (k.empty())
return false;
UMat src = _src.getUMat();
k.args(ocl::KernelArg::ReadOnly(src),
ocl::KernelArg::ReadOnly(_weight.getUMat()),
ocl::KernelArg::ReadWrite(_dst.getUMat()),
ocl::KernelArg::ReadWrite(_dst_weight.getUMat())
);
size_t globalsize[2] = {(size_t)src.cols, (size_t)src.rows };
return k.run(2, globalsize, NULL, false);
}
static bool matchTemplate_CCOEFF(InputArray _image, InputArray _templ, OutputArray _result)
{
matchTemplate(_image, _templ, _result, CV_TM_CCORR);
UMat image_sums, temp;
integral(_image, temp);
if (temp.depth() == CV_64F)
temp.convertTo(image_sums, CV_32F);
else
image_sums = temp;
int type = image_sums.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
ocl::Kernel k("matchTemplate_Prepared_CCOEFF", ocl::imgproc::match_template_oclsrc,
format("-D CCOEFF -D T=%s -D elem_type=%s -D cn=%d", ocl::typeToStr(type), ocl::typeToStr(depth), cn));
if (k.empty())
return false;
UMat templ = _templ.getUMat();
Size size = _image.size(), tsize = templ.size();
_result.create(size.height - templ.rows + 1, size.width - templ.cols + 1, CV_32F);
UMat result = _result.getUMat();
if (cn == 1)
{
float templ_sum = static_cast<float>(sum(_templ)[0]) / tsize.area();
k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadWrite(result),
templ.rows, templ.cols, templ_sum);
}
else
{
Vec4f templ_sum = Vec4f::all(0);
templ_sum = sum(templ) / tsize.area();
if (cn == 2)
k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadWrite(result), templ.rows, templ.cols,
templ_sum[0], templ_sum[1]);
else if (cn==3)
k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadWrite(result), templ.rows, templ.cols,
templ_sum[0], templ_sum[1], templ_sum[2]);
else
k.args(ocl::KernelArg::ReadOnlyNoSize(image_sums), ocl::KernelArg::ReadWrite(result), templ.rows, templ.cols,
templ_sum[0], templ_sum[1], templ_sum[2], templ_sum[3]);
}
size_t globalsize[2] = { result.cols, result.rows };
return k.run(2, globalsize, NULL, false);
}
static bool ocl_pyrUp( InputArray _src, OutputArray _dst, const Size& _dsz, int borderType)
{
int type = _src.type(), depth = CV_MAT_DEPTH(type), channels = CV_MAT_CN(type);
if (channels > 4 || borderType != BORDER_DEFAULT)
return false;
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
if (depth == CV_64F && !doubleSupport)
return false;
Size ssize = _src.size();
if ((_dsz.area() != 0) && (_dsz != Size(ssize.width * 2, ssize.height * 2)))
return false;
UMat src = _src.getUMat();
Size dsize = Size(ssize.width * 2, ssize.height * 2);
_dst.create( dsize, src.type() );
UMat dst = _dst.getUMat();
int float_depth = depth == CV_64F ? CV_64F : CV_32F;
const int local_size = 16;
char cvt[2][50];
String buildOptions = format(
"-D T=%s -D FT=%s -D convertToT=%s -D convertToFT=%s%s "
"-D T1=%s -D cn=%d -D LOCAL_SIZE=%d",
ocl::typeToStr(type), ocl::typeToStr(CV_MAKETYPE(float_depth, channels)),
ocl::convertTypeStr(float_depth, depth, channels, cvt[0]),
ocl::convertTypeStr(depth, float_depth, channels, cvt[1]),
doubleSupport ? " -D DOUBLE_SUPPORT" : "",
ocl::typeToStr(depth), channels, local_size
);
size_t globalThreads[2] = { dst.cols, dst.rows };
size_t localThreads[2] = { local_size, local_size };
ocl::Kernel k;
if (ocl::Device::getDefault().isIntel() && channels == 1)
{
k.create("pyrUp_unrolled", ocl::imgproc::pyr_up_oclsrc, buildOptions);
globalThreads[0] = dst.cols/2; globalThreads[1] = dst.rows/2;
}
else
k.create("pyrUp", ocl::imgproc::pyr_up_oclsrc, buildOptions);
if (k.empty())
return false;
k.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnly(dst));
return k.run(2, globalThreads, localThreads, false);
}
static bool ocl_updateMotionHistory( InputArray _silhouette, InputOutputArray _mhi,
float timestamp, float delbound )
{
ocl::Kernel k("updateMotionHistory", ocl::video::updatemotionhistory_oclsrc);
if (k.empty())
return false;
UMat silh = _silhouette.getUMat(), mhi = _mhi.getUMat();
k.args(ocl::KernelArg::ReadOnlyNoSize(silh), ocl::KernelArg::ReadWrite(mhi),
timestamp, delbound);
size_t globalsize[2] = { silh.cols, silh.rows };
return k.run(2, globalsize, NULL, false);
}
static bool ocl_pyrDown( InputArray _src, OutputArray _dst, const Size& _dsz, int borderType)
{
int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
if (cn > 4 || (depth == CV_64F && !doubleSupport))
return false;
Size ssize = _src.size();
Size dsize = _dsz.area() == 0 ? Size((ssize.width + 1) / 2, (ssize.height + 1) / 2) : _dsz;
if (dsize.height < 2 || dsize.width < 2)
return false;
CV_Assert( ssize.width > 0 && ssize.height > 0 &&
std::abs(dsize.width*2 - ssize.width) <= 2 &&
std::abs(dsize.height*2 - ssize.height) <= 2 );
UMat src = _src.getUMat();
_dst.create( dsize, src.type() );
UMat dst = _dst.getUMat();
int float_depth = depth == CV_64F ? CV_64F : CV_32F;
const int local_size = 256;
int kercn = 1;
if (depth == CV_8U && float_depth == CV_32F && cn == 1 && ocl::Device::getDefault().isIntel())
kercn = 4;
const char * const borderMap[] = { "BORDER_CONSTANT", "BORDER_REPLICATE", "BORDER_REFLECT", "BORDER_WRAP",
"BORDER_REFLECT_101" };
char cvt[2][50];
String buildOptions = format(
"-D T=%s -D FT=%s -D convertToT=%s -D convertToFT=%s%s "
"-D T1=%s -D cn=%d -D kercn=%d -D fdepth=%d -D %s -D LOCAL_SIZE=%d",
ocl::typeToStr(type), ocl::typeToStr(CV_MAKETYPE(float_depth, cn)),
ocl::convertTypeStr(float_depth, depth, cn, cvt[0]),
ocl::convertTypeStr(depth, float_depth, cn, cvt[1]),
doubleSupport ? " -D DOUBLE_SUPPORT" : "", ocl::typeToStr(depth),
cn, kercn, float_depth, borderMap[borderType], local_size
);
ocl::Kernel k("pyrDown", ocl::imgproc::pyr_down_oclsrc, buildOptions);
if (k.empty())
return false;
k.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnly(dst));
size_t localThreads[2] = { local_size/kercn, 1 };
size_t globalThreads[2] = { (src.cols + (kercn-1))/kercn, (dst.rows + 1) / 2 };
return k.run(2, globalThreads, localThreads, false);
}
UMat cv::superres::arrGetUMat(InputArray arr, UMat& buf)
{
switch (arr.kind())
{
case _InputArray::GPU_MAT:
arr.getGpuMat().download(buf);
return buf;
case _InputArray::OPENGL_BUFFER:
arr.getOGlBuffer().copyTo(buf);
return buf;
default:
return arr.getUMat();
}
}
void UMat::copyTo(OutputArray _dst, InputArray _mask) const
{
if( _mask.empty() )
{
copyTo(_dst);
return;
}
#ifdef HAVE_OPENCL
int cn = channels(), mtype = _mask.type(), mdepth = CV_MAT_DEPTH(mtype), mcn = CV_MAT_CN(mtype);
CV_Assert( mdepth == CV_8U && (mcn == 1 || mcn == cn) );
if (ocl::useOpenCL() && _dst.isUMat() && dims <= 2)
{
UMatData * prevu = _dst.getUMat().u;
_dst.create( dims, size, type() );
UMat dst = _dst.getUMat();
bool haveDstUninit = false;
if( prevu != dst.u ) // do not leave dst uninitialized
haveDstUninit = true;
String opts = format("-D COPY_TO_MASK -D T1=%s -D scn=%d -D mcn=%d%s",
ocl::memopTypeToStr(depth()), cn, mcn,
haveDstUninit ? " -D HAVE_DST_UNINIT" : "");
ocl::Kernel k("copyToMask", ocl::core::copyset_oclsrc, opts);
if (!k.empty())
{
k.args(ocl::KernelArg::ReadOnlyNoSize(*this),
ocl::KernelArg::ReadOnlyNoSize(_mask.getUMat()),
haveDstUninit ? ocl::KernelArg::WriteOnly(dst) :
ocl::KernelArg::ReadWrite(dst));
size_t globalsize[2] = { cols, rows };
if (k.run(2, globalsize, NULL, false))
{
CV_IMPL_ADD(CV_IMPL_OCL);
return;
}
}
}
#endif
Mat src = getMat(ACCESS_READ);
src.copyTo(_dst, _mask);
}
static bool ocl_norm( InputArray _src, int normType, InputArray _mask, double & result )
{
const ocl::Device & d = ocl::Device::getDefault();
#ifdef __ANDROID__
if (d.isNVidia())
return false;
#endif
const int cn = _src.channels();
if (cn > 4)
return false;
int type = _src.type(), depth = CV_MAT_DEPTH(type);
bool doubleSupport = d.doubleFPConfig() > 0,
haveMask = _mask.kind() != _InputArray::NONE;
if ( !(normType == NORM_INF || normType == NORM_L1 || normType == NORM_L2 || normType == NORM_L2SQR) ||
(!doubleSupport && depth == CV_64F))
return false;
UMat src = _src.getUMat();
if (normType == NORM_INF)
{
if (!ocl_minMaxIdx(_src, NULL, &result, NULL, NULL, _mask,
std::max(depth, CV_32S), depth != CV_8U && depth != CV_16U))
return false;
}
else if (normType == NORM_L1 || normType == NORM_L2 || normType == NORM_L2SQR)
{
Scalar sc;
bool unstype = depth == CV_8U || depth == CV_16U;
if ( !ocl_sum(haveMask ? src : src.reshape(1), sc, normType == NORM_L2 || normType == NORM_L2SQR ?
OCL_OP_SUM_SQR : (unstype ? OCL_OP_SUM : OCL_OP_SUM_ABS), _mask) )
return false;
double s = 0.0;
for (int i = 0; i < (haveMask ? cn : 1); ++i)
s += sc[i];
result = normType == NORM_L1 || normType == NORM_L2SQR ? s : std::sqrt(s);
}
return true;
}
static bool ocl_diffSign(InputArray _src1, OutputArray _src2, OutputArray _dst)
{
ocl::Kernel k("diffSign", ocl::superres::superres_btvl1_oclsrc);
if (k.empty())
return false;
UMat src1 = _src1.getUMat(), src2 = _src2.getUMat();
_dst.create(src1.size(), src1.type());
UMat dst = _dst.getUMat();
int cn = src1.channels();
k.args(ocl::KernelArg::ReadOnlyNoSize(src1),
ocl::KernelArg::ReadOnlyNoSize(src2),
ocl::KernelArg::WriteOnly(dst, cn));
size_t globalsize[2] = { (size_t)src1.cols * cn, (size_t)src1.rows };
return k.run(2, globalsize, NULL, false);
}
static bool ocl_threshold( InputArray _src, OutputArray _dst, double & thresh, double maxval, int thresh_type )
{
int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
kercn = ocl::predictOptimalVectorWidth(_src, _dst), ktype = CV_MAKE_TYPE(depth, kercn);
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
if ( !(thresh_type == THRESH_BINARY || thresh_type == THRESH_BINARY_INV || thresh_type == THRESH_TRUNC ||
thresh_type == THRESH_TOZERO || thresh_type == THRESH_TOZERO_INV) ||
(!doubleSupport && depth == CV_64F))
return false;
const char * const thresholdMap[] = { "THRESH_BINARY", "THRESH_BINARY_INV", "THRESH_TRUNC",
"THRESH_TOZERO", "THRESH_TOZERO_INV" };
ocl::Device dev = ocl::Device::getDefault();
int stride_size = dev.isIntel() && (dev.type() & ocl::Device::TYPE_GPU) ? 4 : 1;
ocl::Kernel k("threshold", ocl::imgproc::threshold_oclsrc,
format("-D %s -D T=%s -D T1=%s -D STRIDE_SIZE=%d%s", thresholdMap[thresh_type],
ocl::typeToStr(ktype), ocl::typeToStr(depth), stride_size,
doubleSupport ? " -D DOUBLE_SUPPORT" : ""));
if (k.empty())
return false;
UMat src = _src.getUMat();
_dst.create(src.size(), type);
UMat dst = _dst.getUMat();
if (depth <= CV_32S)
thresh = cvFloor(thresh);
const double min_vals[] = { 0, CHAR_MIN, 0, SHRT_MIN, INT_MIN, -FLT_MAX, -DBL_MAX, 0 };
double min_val = min_vals[depth];
k.args(ocl::KernelArg::ReadOnlyNoSize(src), ocl::KernelArg::WriteOnly(dst, cn, kercn),
ocl::KernelArg::Constant(Mat(1, 1, depth, Scalar::all(thresh))),
ocl::KernelArg::Constant(Mat(1, 1, depth, Scalar::all(maxval))),
ocl::KernelArg::Constant(Mat(1, 1, depth, Scalar::all(min_val))));
size_t globalsize[2] = { static_cast<size_t>(dst.cols * cn / kercn), static_cast<size_t>(dst.rows) };
globalsize[1] = (globalsize[1] + stride_size - 1) / stride_size;
return k.run(2, globalsize, NULL, false);
}
static bool ocl_integral( InputArray _src, OutputArray _sum, OutputArray _sqsum, int sdepth, int sqdepth )
{
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
if ( _src.type() != CV_8UC1 || (!doubleSupport && (sdepth == CV_64F || sqdepth == CV_64F)) )
return false;
static const int tileSize = 16;
String build_opt = format("-D SUM_SQUARE -D sumT=%s -D sumSQT=%s -D LOCAL_SUM_SIZE=%d%s",
ocl::typeToStr(sdepth), ocl::typeToStr(sqdepth),
tileSize,
doubleSupport ? " -D DOUBLE_SUPPORT" : "");
ocl::Kernel kcols("integral_sum_cols", ocl::imgproc::integral_sum_oclsrc, build_opt);
if (kcols.empty())
return false;
UMat src = _src.getUMat();
Size src_size = src.size();
Size bufsize(((src_size.height + tileSize - 1) / tileSize) * tileSize, ((src_size.width + tileSize - 1) / tileSize) * tileSize);
UMat buf(bufsize, sdepth);
UMat buf_sq(bufsize, sqdepth);
kcols.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnlyNoSize(buf), ocl::KernelArg::WriteOnlyNoSize(buf_sq));
size_t gt = src.cols, lt = tileSize;
if (!kcols.run(1, >, <, false))
return false;
ocl::Kernel krows("integral_sum_rows", ocl::imgproc::integral_sum_oclsrc, build_opt);
if (krows.empty())
return false;
Size sumsize(src_size.width + 1, src_size.height + 1);
_sum.create(sumsize, sdepth);
UMat sum = _sum.getUMat();
_sqsum.create(sumsize, sqdepth);
UMat sum_sq = _sqsum.getUMat();
krows.args(ocl::KernelArg::ReadOnlyNoSize(buf), ocl::KernelArg::ReadOnlyNoSize(buf_sq), ocl::KernelArg::WriteOnly(sum), ocl::KernelArg::WriteOnlyNoSize(sum_sq));
gt = src.rows;
return krows.run(1, >, <, false);
}
static bool ocl_pyrDown( InputArray _src, OutputArray _dst, const Size& _dsz, int borderType)
{
int type = _src.type(), depth = CV_MAT_DEPTH(type), channels = CV_MAT_CN(type);
if (channels > 4 || borderType != BORDER_DEFAULT)
return false;
bool doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
if ((depth == CV_64F) && !(doubleSupport))
return false;
Size ssize = _src.size();
Size dsize = _dsz.area() == 0 ? Size((ssize.width + 1) / 2, (ssize.height + 1) / 2) : _dsz;
CV_Assert( ssize.width > 0 && ssize.height > 0 &&
std::abs(dsize.width*2 - ssize.width) <= 2 &&
std::abs(dsize.height*2 - ssize.height) <= 2 );
UMat src = _src.getUMat();
_dst.create( dsize, src.type() );
UMat dst = _dst.getUMat();
int float_depth = depth == CV_64F ? CV_64F : CV_32F;
char cvt[2][50];
String buildOptions = format(
"-D T=%s -D FT=%s -D convertToT=%s -D convertToFT=%s%s "
"-D T1=%s -D cn=%d",
ocl::typeToStr(type), ocl::typeToStr(CV_MAKETYPE(float_depth, channels)),
ocl::convertTypeStr(float_depth, depth, channels, cvt[0]),
ocl::convertTypeStr(depth, float_depth, channels, cvt[1]),
doubleSupport ? " -D DOUBLE_SUPPORT" : "",
ocl::typeToStr(depth), channels
);
ocl::Kernel k("pyrDown", ocl::imgproc::pyr_down_oclsrc, buildOptions);
if (k.empty())
return false;
k.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnly(dst));
size_t localThreads[2] = { 256, 1 };
size_t globalThreads[2] = { src.cols, dst.rows };
return k.run(2, globalThreads, localThreads, false);
}
static bool ocl_flip(InputArray _src, OutputArray _dst, int flipCode )
{
int type = _src.type(), cn = CV_MAT_CN(type);
if (cn > 4 || cn == 3)
return false;
const char * kernelName;
int flipType;
if (flipCode == 0)
kernelName = "arithm_flip_rows", flipType = FLIP_ROWS;
else if (flipCode > 0)
kernelName = "arithm_flip_cols", flipType = FLIP_COLS;
else
kernelName = "arithm_flip_rows_cols", flipType = FLIP_BOTH;
Size size = _src.size();
int cols = size.width, rows = size.height;
if ((cols == 1 && flipType == FLIP_COLS) ||
(rows == 1 && flipType == FLIP_ROWS) ||
(rows == 1 && cols == 1 && flipType == FLIP_BOTH))
{
_src.copyTo(_dst);
return true;
}
ocl::Kernel k(kernelName, ocl::core::flip_oclsrc,
format( "-D type=%s", ocl::memopTypeToStr(type)));
if (k.empty())
return false;
_dst.create(size, type);
UMat src = _src.getUMat(), dst = _dst.getUMat();
cols = flipType == FLIP_COLS ? ((cols+1)/2) : cols;
rows = flipType & FLIP_ROWS ? ((rows+1)/2) : rows;
size_t globalsize[2] = { cols, rows };
return k.args(ocl::KernelArg::ReadOnlyNoSize(src), ocl::KernelArg::WriteOnly(dst), rows, cols).run(2, globalsize, NULL, false);
}
