void cv::gpu::min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
{
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
min_caller<uchar>, min_caller<schar>, min_caller<ushort>, min_caller<short>, min_caller<int>,
min_caller<float>, min_caller<double>
};
funcs[src1.depth()](src1, src2, dst, StreamAccessor::getStream(stream));
}
void cv::gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point* minLoc, Point* maxLoc,
const GpuMat& mask, GpuMat& valBuf, GpuMat& locBuf)
{
typedef void (*func_t)(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
#ifdef OPENCV_TINY_GPU_MODULE
static const func_t funcs[] =
{
::minMaxLoc::run<uchar>,
0/*::minMaxLoc::run<schar>*/,
0/*::minMaxLoc::run<ushort>*/,
0/*::minMaxLoc::run<short>*/,
::minMaxLoc::run<int>,
::minMaxLoc::run<float>,
0/*::minMaxLoc::run<double>*/,
};
#else
static const func_t funcs[] =
{
::minMaxLoc::run<uchar>,
::minMaxLoc::run<schar>,
::minMaxLoc::run<ushort>,
::minMaxLoc::run<short>,
::minMaxLoc::run<int>,
::minMaxLoc::run<float>,
::minMaxLoc::run<double>,
};
#endif
CV_Assert( src.channels() == 1 );
CV_Assert( mask.empty() || (mask.size() == src.size() && mask.type() == CV_8U) );
if (src.depth() == CV_64F)
{
if (!deviceSupports(NATIVE_DOUBLE))
CV_Error(CV_StsUnsupportedFormat, "The device doesn't support double");
}
Size valbuf_size, locbuf_size;
::minMaxLoc::getBufSize(src.cols, src.rows, src.elemSize(), valbuf_size.width, valbuf_size.height, locbuf_size.width, locbuf_size.height);
ensureSizeIsEnough(valbuf_size, CV_8U, valBuf);
ensureSizeIsEnough(locbuf_size, CV_8U, locBuf);
const func_t func = funcs[src.depth()];
if (!func)
CV_Error(CV_StsUnsupportedFormat, "Unsupported combination of source and destination types");
double temp1, temp2;
Point temp3, temp4;
func(src, mask, minVal ? minVal : &temp1, maxVal ? maxVal : &temp2, minLoc ? &minLoc->x : &temp3.x, maxLoc ? &maxLoc->x : &temp4.x, valBuf, locBuf);
}
void cv::cuda::BFMatcher_CUDA::matchSingle(const GpuMat& query, const GpuMat& train,
GpuMat& trainIdx, GpuMat& distance,
const GpuMat& mask, Stream& stream)
{
if (query.empty() || train.empty())
return;
using namespace cv::cuda::device::bf_match;
typedef void (*caller_t)(const PtrStepSzb& query, const PtrStepSzb& train, const PtrStepSzb& mask,
const PtrStepSzi& trainIdx, const PtrStepSzf& distance,
cudaStream_t stream);
static const caller_t callersL1[] =
{
matchL1_gpu<unsigned char>, 0/*matchL1_gpu<signed char>*/,
matchL1_gpu<unsigned short>, matchL1_gpu<short>,
matchL1_gpu<int>, matchL1_gpu<float>
};
static const caller_t callersL2[] =
{
0/*matchL2_gpu<unsigned char>*/, 0/*matchL2_gpu<signed char>*/,
0/*matchL2_gpu<unsigned short>*/, 0/*matchL2_gpu<short>*/,
0/*matchL2_gpu<int>*/, matchL2_gpu<float>
};
static const caller_t callersHamming[] =
{
matchHamming_gpu<unsigned char>, 0/*matchHamming_gpu<signed char>*/,
matchHamming_gpu<unsigned short>, 0/*matchHamming_gpu<short>*/,
matchHamming_gpu<int>, 0/*matchHamming_gpu<float>*/
};
CV_Assert(query.channels() == 1 && query.depth() < CV_64F);
CV_Assert(train.cols == query.cols && train.type() == query.type());
CV_Assert(norm == NORM_L1 || norm == NORM_L2 || norm == NORM_HAMMING);
const caller_t* callers = norm == NORM_L1 ? callersL1 : norm == NORM_L2 ? callersL2 : callersHamming;
const int nQuery = query.rows;
ensureSizeIsEnough(1, nQuery, CV_32S, trainIdx);
ensureSizeIsEnough(1, nQuery, CV_32F, distance);
caller_t func = callers[query.depth()];
CV_Assert(func != 0);
func(query, train, mask, trainIdx, distance, StreamAccessor::getStream(stream));
}
Scalar cv::gpu::sqrSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
{
typedef void (*func_t)(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
#ifdef OPENCV_TINY_GPU_MODULE
static const func_t funcs[7][5] =
{
{0, ::sum::runSqr<uchar , 1>, 0, 0, 0},
{0, 0, 0, 0, 0},
{0, 0, 0, 0, 0},
{0, 0, 0, 0, 0},
{0, 0, 0, 0, 0},
{0, ::sum::runSqr<float , 1>, 0, 0, 0},
{0, 0, 0, 0, 0},
};
#else
static const func_t funcs[7][5] =
{
{0, ::sum::runSqr<uchar , 1>, ::sum::runSqr<uchar , 2>, ::sum::runSqr<uchar , 3>, ::sum::runSqr<uchar , 4>},
{0, ::sum::runSqr<schar , 1>, ::sum::runSqr<schar , 2>, ::sum::runSqr<schar , 3>, ::sum::runSqr<schar , 4>},
{0, ::sum::runSqr<ushort, 1>, ::sum::runSqr<ushort, 2>, ::sum::runSqr<ushort, 3>, ::sum::runSqr<ushort, 4>},
{0, ::sum::runSqr<short , 1>, ::sum::runSqr<short , 2>, ::sum::runSqr<short , 3>, ::sum::runSqr<short , 4>},
{0, ::sum::runSqr<int , 1>, ::sum::runSqr<int , 2>, ::sum::runSqr<int , 3>, ::sum::runSqr<int , 4>},
{0, ::sum::runSqr<float , 1>, ::sum::runSqr<float , 2>, ::sum::runSqr<float , 3>, ::sum::runSqr<float , 4>},
{0, ::sum::runSqr<double, 1>, ::sum::runSqr<double, 2>, ::sum::runSqr<double, 3>, ::sum::runSqr<double, 4>}
};
#endif
CV_Assert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
if (src.depth() == CV_64F)
{
if (!deviceSupports(NATIVE_DOUBLE))
CV_Error(CV_StsUnsupportedFormat, "The device doesn't support double");
}
Size buf_size;
::sum::getBufSize(src.cols, src.rows, src.channels(), buf_size.width, buf_size.height);
ensureSizeIsEnough(buf_size, CV_8U, buf);
buf.setTo(Scalar::all(0));
const func_t func = funcs[src.depth()][src.channels()];
if (!func)
CV_Error(CV_StsUnsupportedFormat, "Unsupported combination of source and destination types");
double result[4];
func(src, buf.data, result, mask);
return Scalar(result[0], result[1], result[2], result[3]);
}
void cv::gpu::Stream::enqueueMemSet(GpuMat& src, Scalar val, const GpuMat& mask)
{
CV_Assert((src.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
CV_Assert(mask.type() == CV_8UC1);
typedef void (*set_caller_t)(GpuMat& src, const Scalar& s, const GpuMat& mask, cudaStream_t stream);
static const set_caller_t set_callers[] =
{
kernelSetMask<uchar>, kernelSetMask<schar>, kernelSetMask<ushort>, kernelSetMask<short>,
kernelSetMask<int>, kernelSetMask<float>, kernelSetMask<double>
};
set_callers[src.depth()](src, val, mask, impl->stream);
}
void cv::gpu::BruteForceMatcher_GPU_base::matchCollection(const GpuMat& query, const GpuMat& trainCollection,
GpuMat& trainIdx, GpuMat& imgIdx, GpuMat& distance,
const GpuMat& masks, Stream& stream)
{
if (query.empty() || trainCollection.empty())
return;
using namespace ::cv::gpu::device::bf_match;
typedef void (*caller_t)(const DevMem2Db& query, const DevMem2Db& trains, const DevMem2D_<PtrStepb>& masks,
const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
int cc, cudaStream_t stream);
static const caller_t callers[3][6] =
{
{
matchL1_gpu<unsigned char>, 0/*matchL1_gpu<signed char>*/,
matchL1_gpu<unsigned short>, matchL1_gpu<short>,
matchL1_gpu<int>, matchL1_gpu<float>
},
{
0/*matchL2_gpu<unsigned char>*/, 0/*matchL2_gpu<signed char>*/,
0/*matchL2_gpu<unsigned short>*/, 0/*matchL2_gpu<short>*/,
0/*matchL2_gpu<int>*/, matchL2_gpu<float>
},
{
matchHamming_gpu<unsigned char>, 0/*matchHamming_gpu<signed char>*/,
matchHamming_gpu<unsigned short>, 0/*matchHamming_gpu<short>*/,
matchHamming_gpu<int>, 0/*matchHamming_gpu<float>*/
}
};
CV_Assert(query.channels() == 1 && query.depth() < CV_64F);
const int nQuery = query.rows;
ensureSizeIsEnough(1, nQuery, CV_32S, trainIdx);
ensureSizeIsEnough(1, nQuery, CV_32S, imgIdx);
ensureSizeIsEnough(1, nQuery, CV_32F, distance);
caller_t func = callers[distType][query.depth()];
CV_Assert(func != 0);
DeviceInfo info;
int cc = info.majorVersion() * 10 + info.minorVersion();
func(query, trainCollection, masks, trainIdx, imgIdx, distance, cc, StreamAccessor::getStream(stream));
}
void cv::gpu::connectivityMask(const GpuMat& image, GpuMat& mask, const cv::Scalar& lo, const cv::Scalar& hi, Stream& s)
{
CV_Assert(!image.empty());
int ch = image.channels();
CV_Assert(ch <= 4);
int depth = image.depth();
typedef void (*func_t)(const DevMem2D& image, DevMem2D edges, const float4& lo, const float4& hi, cudaStream_t stream);
static const func_t suppotLookup[8][4] =
{ // 1, 2, 3, 4
{ device::ccl::computeEdges<uchar>, 0, device::ccl::computeEdges<uchar3>, device::ccl::computeEdges<uchar4> },// CV_8U
{ 0, 0, 0, 0 },// CV_16U
{ device::ccl::computeEdges<ushort>, 0, device::ccl::computeEdges<ushort3>, device::ccl::computeEdges<ushort4> },// CV_8S
{ 0, 0, 0, 0 },// CV_16S
{ device::ccl::computeEdges<int>, 0, 0, 0 },// CV_32S
{ device::ccl::computeEdges<float>, 0, 0, 0 },// CV_32F
{ 0, 0, 0, 0 },// CV_64F
{ 0, 0, 0, 0 } // CV_USRTYPE1
};
func_t f = suppotLookup[depth][ch - 1];
CV_Assert(f);
if (image.size() != mask.size() || mask.type() != CV_8UC1)
mask.create(image.size(), CV_8UC1);
cudaStream_t stream = StreamAccessor::getStream(s);
float4 culo = scalarToCudaType(lo), cuhi = scalarToCudaType(hi);
f(image, mask, culo, cuhi, stream);
}
Scalar cv::gpu::sqrSum(const GpuMat& src, GpuMat& buf)
{
using namespace mathfunc;
typedef void (*Caller)(const DevMem2D, PtrStep, double*, int);
static Caller multipass_callers[7] = {
sqrSumMultipassCaller<unsigned char>, sqrSumMultipassCaller<char>,
sqrSumMultipassCaller<unsigned short>, sqrSumMultipassCaller<short>,
sqrSumMultipassCaller<int>, sqrSumMultipassCaller<float>, 0 };
static Caller singlepass_callers[7] = {
sqrSumCaller<unsigned char>, sqrSumCaller<char>,
sqrSumCaller<unsigned short>, sqrSumCaller<short>,
sqrSumCaller<int>, sqrSumCaller<float>, 0 };
Caller* callers = multipass_callers;
if (TargetArchs::builtWith(GLOBAL_ATOMICS) && DeviceInfo().supports(GLOBAL_ATOMICS))
callers = singlepass_callers;
Size buf_size;
sums::getBufSizeRequired(src.cols, src.rows, src.channels(),
buf_size.width, buf_size.height);
ensureSizeIsEnough(buf_size, CV_8U, buf);
Caller caller = callers[src.depth()];
if (!caller) CV_Error(CV_StsBadArg, "sqrSum: unsupported type");
double result[4];
caller(src, buf, result, src.channels());
return Scalar(result[0], result[1], result[2], result[3]);
}
void cv::gpu::pow(const GpuMat& src, double power, GpuMat& dst, Stream& stream)
{
CV_Assert( src.depth() != CV_64F );
dst.create(src.size(), src.type());
typedef void (*caller_t)(const DevMem2D& src, float power, DevMem2D dst, cudaStream_t stream);
static const caller_t callers[] =
{
mathfunc::pow_caller<unsigned char>, mathfunc::pow_caller<signed char>,
mathfunc::pow_caller<unsigned short>, mathfunc::pow_caller<short>,
mathfunc::pow_caller<int>, mathfunc::pow_caller<float>
};
callers[src.depth()](src.reshape(1), (float)power, dst.reshape(1), StreamAccessor::getStream(stream));
}
void cv::cuda::fastNlMeansDenoising(InputArray _src, OutputArray _dst, float h, int search_window, int block_window, Stream& stream)
{
const GpuMat src = _src.getGpuMat();
CV_Assert(src.depth() == CV_8U && src.channels() < 4);
int border_size = search_window/2 + block_window/2;
Size esize = src.size() + Size(border_size, border_size) * 2;
BufferPool pool(stream);
GpuMat extended_src = pool.getBuffer(esize, src.type());
cv::cuda::copyMakeBorder(src, extended_src, border_size, border_size, border_size, border_size, cv::BORDER_DEFAULT, Scalar(), stream);
GpuMat src_hdr = extended_src(Rect(Point2i(border_size, border_size), src.size()));
int bcols, brows;
device::imgproc::nln_fast_get_buffer_size(src_hdr, search_window, block_window, bcols, brows);
GpuMat buffer = pool.getBuffer(brows, bcols, CV_32S);
using namespace cv::cuda::device::imgproc;
typedef void (*nlm_fast_t)(const PtrStepSzb&, PtrStepSzb, PtrStepi, int, int, float, cudaStream_t);
static const nlm_fast_t funcs[] = { nlm_fast_gpu<uchar>, nlm_fast_gpu<uchar2>, nlm_fast_gpu<uchar3>, 0};
_dst.create(src.size(), src.type());
GpuMat dst = _dst.getGpuMat();
funcs[src.channels()-1](src_hdr, dst, buffer, search_window, block_window, h, StreamAccessor::getStream(stream));
}
void cv::gpu::transpose(const GpuMat& src, GpuMat& dst)
{
using namespace cv::gpu::mathfunc;
typedef void (*func_t)(const DevMem2D& src, const DevMem2D& dst);
static const func_t funcs[] =
{
transpose_gpu<uchar4>, transpose_gpu<char4>, transpose_gpu<ushort2>, transpose_gpu<short2>,
transpose_gpu<int>, transpose_gpu<float>
};
CV_Assert(src.type() == CV_8UC1 || src.type() == CV_8UC4 || src.type() == CV_8SC4
|| src.type() == CV_16UC2 || src.type() == CV_16SC2 || src.type() == CV_32SC1 || src.type() == CV_32FC1);
dst.create( src.cols, src.rows, src.type() );
if (src.type() == CV_8UC1)
{
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
nppSafeCall( nppiTranspose_8u_C1R(src.ptr<Npp8u>(), src.step, dst.ptr<Npp8u>(), dst.step, sz) );
}
else
{
funcs[src.depth()](src, dst);
}
}
int cv::gpu::CascadeClassifier_GPU::detectMultiScale( const GpuMat& image, GpuMat& objectsBuf, double scaleFactor, int minNeighbors, Size minSize)
{
CV_Assert( scaleFactor > 1 && image.depth() == CV_8U);
CV_Assert( !this->empty());
const int defaultObjSearchNum = 100;
if (objectsBuf.empty())
{
objectsBuf.create(1, defaultObjSearchNum, DataType<Rect>::type);
}
NcvSize32u ncvMinSize = impl->getClassifierSize();
if (ncvMinSize.width < (unsigned)minSize.width && ncvMinSize.height < (unsigned)minSize.height)
{
ncvMinSize.width = minSize.width;
ncvMinSize.height = minSize.height;
}
unsigned int numDetections;
NCVStatus ncvStat = impl->process(image, objectsBuf, (float)scaleFactor, minNeighbors, findLargestObject, visualizeInPlace, ncvMinSize, numDetections);
if (ncvStat != NCV_SUCCESS)
{
CV_Error(CV_GpuApiCallError, "Error in face detectioln");
}
return numDetections;
}
void cv::gpu::matchTemplate(
const GpuMat& image, const GpuMat& templ, GpuMat& result, int method,
MatchTemplateBuf &buf, Stream& stream)
{
CV_Assert(image.type() == templ.type());
CV_Assert(image.cols >= templ.cols && image.rows >= templ.rows);
typedef void (*Caller)(const GpuMat&, const GpuMat&, GpuMat&, MatchTemplateBuf&, Stream& stream);
static const Caller callers8U[] = { ::matchTemplate_SQDIFF_8U, ::matchTemplate_SQDIFF_NORMED_8U,
::matchTemplate_CCORR_8U, ::matchTemplate_CCORR_NORMED_8U,
::matchTemplate_CCOFF_8U, ::matchTemplate_CCOFF_NORMED_8U };
static const Caller callers32F[] = { ::matchTemplate_SQDIFF_32F, 0,
::matchTemplate_CCORR_32F, 0, 0, 0 };
const Caller* callers = 0;
switch (image.depth())
{
case CV_8U: callers = callers8U; break;
case CV_32F: callers = callers32F; break;
default: CV_Error(CV_StsBadArg, "matchTemplate: unsupported data type");
}
Caller caller = callers[method];
CV_Assert(caller);
caller(image, templ, result, buf, stream);
}
void cv::gpu::blendLinear(const GpuMat& img1, const GpuMat& img2, const GpuMat& weights1, const GpuMat& weights2,
GpuMat& result, Stream& stream)
{
CV_Assert(img1.size() == img2.size());
CV_Assert(img1.type() == img2.type());
CV_Assert(weights1.size() == img1.size());
CV_Assert(weights2.size() == img2.size());
CV_Assert(weights1.type() == CV_32F);
CV_Assert(weights2.type() == CV_32F);
const Size size = img1.size();
const int depth = img1.depth();
const int cn = img1.channels();
result.create(size, CV_MAKE_TYPE(depth, cn));
switch (depth)
{
case CV_8U:
if (cn != 4)
blendLinearCaller<uchar>(size.height, size.width, cn, img1, img2, weights1, weights2, result, StreamAccessor::getStream(stream));
else
blendLinearCaller8UC4(size.height, size.width, img1, img2, weights1, weights2, result, StreamAccessor::getStream(stream));
break;
case CV_32F:
blendLinearCaller<float>(size.height, size.width, cn, img1, img2, weights1, weights2, result, StreamAccessor::getStream(stream));
break;
default:
CV_Error(CV_StsUnsupportedFormat, "bad image depth in linear blending function");
}
}
void cv::gpu::ImagePyramid::build(const GpuMat& img, int numLayers, Stream& stream)
{
using namespace cv::gpu::device::pyramid;
typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
static const func_t funcs[6][4] =
{
{kernelDownsampleX2_gpu<uchar1> , 0 /*kernelDownsampleX2_gpu<uchar2>*/ , kernelDownsampleX2_gpu<uchar3> , kernelDownsampleX2_gpu<uchar4> },
{0 /*kernelDownsampleX2_gpu<char1>*/ , 0 /*kernelDownsampleX2_gpu<char2>*/ , 0 /*kernelDownsampleX2_gpu<char3>*/ , 0 /*kernelDownsampleX2_gpu<char4>*/ },
{kernelDownsampleX2_gpu<ushort1> , 0 /*kernelDownsampleX2_gpu<ushort2>*/, kernelDownsampleX2_gpu<ushort3> , kernelDownsampleX2_gpu<ushort4> },
{0 /*kernelDownsampleX2_gpu<short1>*/ , 0 /*kernelDownsampleX2_gpu<short2>*/ , 0 /*kernelDownsampleX2_gpu<short3>*/, 0 /*kernelDownsampleX2_gpu<short4>*/},
{0 /*kernelDownsampleX2_gpu<int1>*/ , 0 /*kernelDownsampleX2_gpu<int2>*/ , 0 /*kernelDownsampleX2_gpu<int3>*/ , 0 /*kernelDownsampleX2_gpu<int4>*/ },
{kernelDownsampleX2_gpu<float1> , 0 /*kernelDownsampleX2_gpu<float2>*/ , kernelDownsampleX2_gpu<float3> , kernelDownsampleX2_gpu<float4> }
};
CV_Assert(img.depth() <= CV_32F && img.channels() <= 4);
const func_t func = funcs[img.depth()][img.channels() - 1];
CV_Assert(func != 0);
layer0_ = img;
Size szLastLayer = img.size();
nLayers_ = 1;
if (numLayers <= 0)
numLayers = 255; //it will cut-off when any of the dimensions goes 1
pyramid_.resize(numLayers);
for (int i = 0; i < numLayers - 1; ++i)
{
Size szCurLayer(szLastLayer.width / 2, szLastLayer.height / 2);
if (szCurLayer.width == 0 || szCurLayer.height == 0)
break;
ensureSizeIsEnough(szCurLayer, img.type(), pyramid_[i]);
nLayers_++;
const GpuMat& prevLayer = i == 0 ? layer0_ : pyramid_[i - 1];
func(prevLayer, pyramid_[i], StreamAccessor::getStream(stream));
szLastLayer = szCurLayer;
}
}
void cv::gpu::ImagePyramid::getLayer(GpuMat& outImg, Size outRoi, Stream& stream) const
{
using namespace cv::gpu::device::pyramid;
typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
static const func_t funcs[6][4] =
{
{kernelInterpolateFrom1_gpu<uchar1> , 0 /*kernelInterpolateFrom1_gpu<uchar2>*/ , kernelInterpolateFrom1_gpu<uchar3> , kernelInterpolateFrom1_gpu<uchar4> },
{0 /*kernelInterpolateFrom1_gpu<char1>*/ , 0 /*kernelInterpolateFrom1_gpu<char2>*/ , 0 /*kernelInterpolateFrom1_gpu<char3>*/ , 0 /*kernelInterpolateFrom1_gpu<char4>*/ },
{kernelInterpolateFrom1_gpu<ushort1> , 0 /*kernelInterpolateFrom1_gpu<ushort2>*/, kernelInterpolateFrom1_gpu<ushort3> , kernelInterpolateFrom1_gpu<ushort4> },
{0 /*kernelInterpolateFrom1_gpu<short1>*/, 0 /*kernelInterpolateFrom1_gpu<short2>*/ , 0 /*kernelInterpolateFrom1_gpu<short3>*/, 0 /*kernelInterpolateFrom1_gpu<short4>*/},
{0 /*kernelInterpolateFrom1_gpu<int1>*/ , 0 /*kernelInterpolateFrom1_gpu<int2>*/ , 0 /*kernelInterpolateFrom1_gpu<int3>*/ , 0 /*kernelInterpolateFrom1_gpu<int4>*/ },
{kernelInterpolateFrom1_gpu<float1> , 0 /*kernelInterpolateFrom1_gpu<float2>*/ , kernelInterpolateFrom1_gpu<float3> , kernelInterpolateFrom1_gpu<float4> }
};
CV_Assert(outRoi.width <= layer0_.cols && outRoi.height <= layer0_.rows && outRoi.width > 0 && outRoi.height > 0);
ensureSizeIsEnough(outRoi, layer0_.type(), outImg);
const func_t func = funcs[outImg.depth()][outImg.channels() - 1];
CV_Assert(func != 0);
if (outRoi.width == layer0_.cols && outRoi.height == layer0_.rows)
{
if (stream)
stream.enqueueCopy(layer0_, outImg);
else
layer0_.copyTo(outImg);
}
float lastScale = 1.0f;
float curScale;
GpuMat lastLayer = layer0_;
GpuMat curLayer;
for (int i = 0; i < nLayers_ - 1; ++i)
{
curScale = lastScale * 0.5f;
curLayer = pyramid_[i];
if (outRoi.width == curLayer.cols && outRoi.height == curLayer.rows)
{
if (stream)
stream.enqueueCopy(curLayer, outImg);
else
curLayer.copyTo(outImg);
}
if (outRoi.width >= curLayer.cols && outRoi.height >= curLayer.rows)
break;
lastScale = curScale;
lastLayer = curLayer;
}
func(lastLayer, outImg, StreamAccessor::getStream(stream));
}
int cv::gpu::countNonZero(const GpuMat& src, GpuMat& buf)
{
typedef int (*func_t)(const PtrStepSzb src, PtrStep<unsigned int> buf);
#ifdef OPENCV_TINY_GPU_MODULE
static const func_t funcs[] =
{
::countNonZero::run<uchar>,
0/*::countNonZero::run<schar>*/,
0/*::countNonZero::run<ushort>*/,
0/*::countNonZero::run<short>*/,
0/*::countNonZero::run<int>*/,
::countNonZero::run<float>,
0/*::countNonZero::run<double>*/,
};
#else
static const func_t funcs[] =
{
::countNonZero::run<uchar>,
::countNonZero::run<schar>,
::countNonZero::run<ushort>,
::countNonZero::run<short>,
::countNonZero::run<int>,
::countNonZero::run<float>,
::countNonZero::run<double>,
};
#endif
CV_Assert(src.channels() == 1);
if (src.depth() == CV_64F)
{
if (!deviceSupports(NATIVE_DOUBLE))
CV_Error(CV_StsUnsupportedFormat, "The device doesn't support double");
}
Size buf_size;
::countNonZero::getBufSize(src.cols, src.rows, buf_size.width, buf_size.height);
ensureSizeIsEnough(buf_size, CV_8U, buf);
const func_t func = funcs[src.depth()];
if (!func)
CV_Error(CV_StsUnsupportedFormat, "Unsupported combination of source and destination types");
return func(src, buf);
}
void cv::gpu::BruteForceMatcher_GPU_base::matchSingle(const GpuMat& query, const GpuMat& train,
GpuMat& trainIdx, GpuMat& distance,
const GpuMat& mask, Stream& stream)
{
if (query.empty() || train.empty())
return;
using namespace cv::gpu::device::bf_match;
typedef void (*caller_t)(const PtrStepSzb& query, const PtrStepSzb& train, const PtrStepSzb& mask,
const PtrStepSzi& trainIdx, const PtrStepSzf& distance,
cudaStream_t stream);
static const caller_t callers[3][6] =
{
{
matchL1_gpu<unsigned char>, 0/*matchL1_gpu<signed char>*/,
matchL1_gpu<unsigned short>, matchL1_gpu<short>,
matchL1_gpu<int>, matchL1_gpu<float>
},
{
0/*matchL2_gpu<unsigned char>*/, 0/*matchL2_gpu<signed char>*/,
0/*matchL2_gpu<unsigned short>*/, 0/*matchL2_gpu<short>*/,
0/*matchL2_gpu<int>*/, matchL2_gpu<float>
},
{
matchHamming_gpu<unsigned char>, 0/*matchHamming_gpu<signed char>*/,
matchHamming_gpu<unsigned short>, 0/*matchHamming_gpu<short>*/,
matchHamming_gpu<int>, 0/*matchHamming_gpu<float>*/
}
};
CV_Assert(query.channels() == 1 && query.depth() < CV_64F);
CV_Assert(train.cols == query.cols && train.type() == query.type());
const int nQuery = query.rows;
ensureSizeIsEnough(1, nQuery, CV_32S, trainIdx);
ensureSizeIsEnough(1, nQuery, CV_32F, distance);
caller_t func = callers[distType][query.depth()];
CV_Assert(func != 0);
func(query, train, mask, trainIdx, distance, StreamAccessor::getStream(stream));
}
void cv::gpu::Stream::enqueueMemSet(GpuMat& src, Scalar val, const GpuMat& mask)
{
CV_Assert((src.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
CV_Assert(mask.type() == CV_8UC1);
setTo(src, val, mask, Impl::getStream(impl));
}
void cv::gpu::sepFilter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& kernelX, const Mat& kernelY, Point anchor)
{
if( ddepth < 0 )
ddepth = src.depth();
dst.create(src.size(), CV_MAKETYPE(ddepth, src.channels()));
Ptr<FilterEngine_GPU> f = createSeparableLinearFilter_GPU(src.type(), dst.type(), kernelX, kernelY, anchor);
f->apply(src, dst);
}
void cv::gpu::filter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& kernel, Point anchor, Stream& stream)
{
if( ddepth < 0 )
ddepth = src.depth();
dst.create(src.size(), CV_MAKETYPE(ddepth, src.channels()));
Ptr<FilterEngine_GPU> f = createLinearFilter_GPU(src.type(), dst.type(), kernel, anchor);
f->apply(src, dst, Rect(0, 0, -1, -1), stream);
}
void cv::gpu::remap(const GpuMat& src, GpuMat& dst, const GpuMat& xmap, const GpuMat& ymap, int interpolation, int borderMode, Scalar borderValue, Stream& stream)
{
using namespace cv::gpu::device::imgproc;
typedef void (*func_t)(PtrStepSzb src, PtrStepSzb srcWhole, int xoff, int yoff, PtrStepSzf xmap, PtrStepSzf ymap, PtrStepSzb dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
static const func_t funcs[6][4] =
{
{remap_gpu<uchar> , 0 /*remap_gpu<uchar2>*/ , remap_gpu<uchar3> , remap_gpu<uchar4> },
{0 /*remap_gpu<schar>*/, 0 /*remap_gpu<char2>*/ , 0 /*remap_gpu<char3>*/, 0 /*remap_gpu<char4>*/},
{remap_gpu<ushort> , 0 /*remap_gpu<ushort2>*/, remap_gpu<ushort3> , remap_gpu<ushort4> },
{remap_gpu<short> , 0 /*remap_gpu<short2>*/ , remap_gpu<short3> , remap_gpu<short4> },
{0 /*remap_gpu<int>*/ , 0 /*remap_gpu<int2>*/ , 0 /*remap_gpu<int3>*/ , 0 /*remap_gpu<int4>*/ },
{remap_gpu<float> , 0 /*remap_gpu<float2>*/ , remap_gpu<float3> , remap_gpu<float4> }
};
CV_Assert(src.depth() <= CV_32F && src.channels() <= 4);
CV_Assert(xmap.type() == CV_32F && ymap.type() == CV_32F && xmap.size() == ymap.size());
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
CV_Assert(borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP);
const func_t func = funcs[src.depth()][src.channels() - 1];
CV_Assert(func != 0);
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderMode, gpuBorderType));
dst.create(xmap.size(), src.type());
Scalar_<float> borderValueFloat;
borderValueFloat = borderValue;
DeviceInfo info;
int cc = info.majorVersion() * 10 + info.minorVersion();
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
func(src, PtrStepSzb(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, xmap, ymap,
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(stream), cc);
}
void cv::cuda::remap(InputArray _src, OutputArray _dst, InputArray _xmap, InputArray _ymap, int interpolation, int borderMode, Scalar borderValue, Stream& stream)
{
using namespace cv::cuda::device::imgproc;
typedef void (*func_t)(PtrStepSzb src, PtrStepSzb srcWhole, int xoff, int yoff, PtrStepSzf xmap, PtrStepSzf ymap, PtrStepSzb dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, bool cc20);
static const func_t funcs[6][4] =
{
{remap_gpu<uchar> , 0 /*remap_gpu<uchar2>*/ , remap_gpu<uchar3> , remap_gpu<uchar4> },
{0 /*remap_gpu<schar>*/, 0 /*remap_gpu<char2>*/ , 0 /*remap_gpu<char3>*/, 0 /*remap_gpu<char4>*/},
{remap_gpu<ushort> , 0 /*remap_gpu<ushort2>*/, remap_gpu<ushort3> , remap_gpu<ushort4> },
{remap_gpu<short> , 0 /*remap_gpu<short2>*/ , remap_gpu<short3> , remap_gpu<short4> },
{0 /*remap_gpu<int>*/ , 0 /*remap_gpu<int2>*/ , 0 /*remap_gpu<int3>*/ , 0 /*remap_gpu<int4>*/ },
{remap_gpu<float> , 0 /*remap_gpu<float2>*/ , remap_gpu<float3> , remap_gpu<float4> }
};
GpuMat src = _src.getGpuMat();
GpuMat xmap = _xmap.getGpuMat();
GpuMat ymap = _ymap.getGpuMat();
CV_Assert( src.depth() <= CV_32F && src.channels() <= 4 );
CV_Assert( xmap.type() == CV_32F && ymap.type() == CV_32F && xmap.size() == ymap.size() );
CV_Assert( interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC );
CV_Assert( borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP );
const func_t func = funcs[src.depth()][src.channels() - 1];
if (!func)
CV_Error(Error::StsUnsupportedFormat, "Unsupported input type");
_dst.create(xmap.size(), src.type());
GpuMat dst = _dst.getGpuMat();
Scalar_<float> borderValueFloat;
borderValueFloat = borderValue;
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
func(src, PtrStepSzb(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, xmap, ymap,
dst, interpolation, borderMode, borderValueFloat.val, StreamAccessor::getStream(stream), deviceSupports(FEATURE_SET_COMPUTE_20));
}
void cv::gpu::flip(const GpuMat& src, GpuMat& dst, int flipCode, Stream& stream)
{
typedef void (*func_t)(const GpuMat& src, GpuMat& dst, int flipCode, cudaStream_t stream);
static const func_t funcs[6][4] =
{
{NppMirror<CV_8U, nppiMirror_8u_C1R>::call, 0, NppMirror<CV_8U, nppiMirror_8u_C3R>::call, NppMirror<CV_8U, nppiMirror_8u_C4R>::call},
{0,0,0,0},
{NppMirror<CV_16U, nppiMirror_16u_C1R>::call, 0, NppMirror<CV_16U, nppiMirror_16u_C3R>::call, NppMirror<CV_16U, nppiMirror_16u_C4R>::call},
{0,0,0,0},
{NppMirror<CV_32S, nppiMirror_32s_C1R>::call, 0, NppMirror<CV_32S, nppiMirror_32s_C3R>::call, NppMirror<CV_32S, nppiMirror_32s_C4R>::call},
{NppMirror<CV_32F, nppiMirror_32f_C1R>::call, 0, NppMirror<CV_32F, nppiMirror_32f_C3R>::call, NppMirror<CV_32F, nppiMirror_32f_C4R>::call}
};
CV_Assert(src.depth() == CV_8U || src.depth() == CV_16U || src.depth() == CV_32S || src.depth() == CV_32F);
CV_Assert(src.channels() == 1 || src.channels() == 3 || src.channels() == 4);
dst.create(src.size(), src.type());
funcs[src.depth()][src.channels() - 1](src, dst, flipCode, StreamAccessor::getStream(stream));
}
void cv::gpu::boxFilter(const GpuMat& src, GpuMat& dst, int ddepth, Size ksize, Point anchor)
{
int sdepth = src.depth(), cn = src.channels();
if( ddepth < 0 )
ddepth = sdepth;
dst.create(src.size(), CV_MAKETYPE(ddepth, cn));
Ptr<FilterEngine_GPU> f = createBoxFilter_GPU(src.type(), dst.type(), ksize, anchor);
f->apply(src, dst);
}
void cv::gpu::sepFilter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& kernelX, const Mat& kernelY, Point anchor, int rowBorderType, int columnBorderType,
Stream& stream)
{
if( ddepth < 0 )
ddepth = src.depth();
dst.create(src.size(), CV_MAKETYPE(ddepth, src.channels()));
Ptr<FilterEngine_GPU> f = createSeparableLinearFilter_GPU(src.type(), dst.type(), kernelX, kernelY, anchor, rowBorderType, columnBorderType);
f->apply(src, dst, Rect(0, 0, src.cols, src.rows), stream);
}
int cv::gpu::countNonZero(const GpuMat& src, GpuMat& buf)
{
using namespace ::cv::gpu::device::matrix_reductions::countnonzero;
typedef int (*Caller)(const PtrStepSzb src, PtrStepb buf);
static Caller multipass_callers[7] =
{
countNonZeroMultipassCaller<unsigned char>, countNonZeroMultipassCaller<char>,
countNonZeroMultipassCaller<unsigned short>, countNonZeroMultipassCaller<short>,
countNonZeroMultipassCaller<int>, countNonZeroMultipassCaller<float>, 0
};
static Caller singlepass_callers[7] =
{
countNonZeroCaller<unsigned char>, countNonZeroCaller<char>,
countNonZeroCaller<unsigned short>, countNonZeroCaller<short>,
countNonZeroCaller<int>, countNonZeroCaller<float>, countNonZeroCaller<double>
};
CV_Assert(src.depth() <= CV_64F);
CV_Assert(src.channels() == 1);
if (src.depth() == CV_64F)
{
if (!TargetArchs::builtWith(NATIVE_DOUBLE) || !DeviceInfo().supports(NATIVE_DOUBLE))
CV_Error(CV_StsUnsupportedFormat, "The device doesn't support double");
}
Size buf_size;
getBufSizeRequired(src.cols, src.rows, buf_size.width, buf_size.height);
ensureSizeIsEnough(buf_size, CV_8U, buf);
Caller* callers = multipass_callers;
if (TargetArchs::builtWith(GLOBAL_ATOMICS) && DeviceInfo().supports(GLOBAL_ATOMICS))
callers = singlepass_callers;
Caller caller = callers[src.type()];
CV_Assert(caller != 0);
return caller(src, buf);
}
void cv::cuda::lshift(InputArray _src, Scalar_<int> val, OutputArray _dst, Stream& stream)
{
typedef void (*func_t)(const GpuMat& src, Scalar_<Npp32u> sc, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[5][4] =
{
{NppShift<CV_8U , 1, nppiLShiftC_8u_C1R>::call , 0, NppShift<CV_8U , 3, nppiLShiftC_8u_C3R>::call , NppShift<CV_8U , 4, nppiLShiftC_8u_C4R>::call },
{0 , 0, 0 , 0 },
{NppShift<CV_16U, 1, nppiLShiftC_16u_C1R>::call, 0, NppShift<CV_16U, 3, nppiLShiftC_16u_C3R>::call, NppShift<CV_16U, 4, nppiLShiftC_16u_C4R>::call},
{0 , 0, 0 , 0 },
{NppShift<CV_32S, 1, nppiLShiftC_32s_C1R>::call, 0, NppShift<CV_32S, 3, nppiLShiftC_32s_C3R>::call, NppShift<CV_32S, 4, nppiLShiftC_32s_C4R>::call},
};
GpuMat src = _src.getGpuMat();
CV_Assert( src.depth() == CV_8U || src.depth() == CV_16U || src.depth() == CV_32S );
CV_Assert( src.channels() == 1 || src.channels() == 3 || src.channels() == 4 );
_dst.create(src.size(), src.type());
GpuMat dst = _dst.getGpuMat();
funcs[src.depth()][src.channels() - 1](src, val, dst, StreamAccessor::getStream(stream));
}
void cv::gpu::histEven(InputArray _src, OutputArray hist, InputOutputArray buf, int histSize, int lowerLevel, int upperLevel, Stream& stream)
{
typedef void (*hist_t)(const GpuMat& src, OutputArray hist, InputOutputArray buf, int levels, int lowerLevel, int upperLevel, cudaStream_t stream);
static const hist_t hist_callers[] =
{
NppHistogramEvenC1<CV_8U , nppiHistogramEven_8u_C1R , nppiHistogramEvenGetBufferSize_8u_C1R >::hist,
0,
NppHistogramEvenC1<CV_16U, nppiHistogramEven_16u_C1R, nppiHistogramEvenGetBufferSize_16u_C1R>::hist,
NppHistogramEvenC1<CV_16S, nppiHistogramEven_16s_C1R, nppiHistogramEvenGetBufferSize_16s_C1R>::hist
};
GpuMat src = _src.getGpuMat();
if (src.depth() == CV_8U && deviceSupports(FEATURE_SET_COMPUTE_30))
{
histEven8u(src, hist.getGpuMatRef(), histSize, lowerLevel, upperLevel, StreamAccessor::getStream(stream));
return;
}
CV_Assert( src.type() == CV_8UC1 || src.type() == CV_16UC1 || src.type() == CV_16SC1 );
hist_callers[src.depth()](src, hist, buf, histSize, lowerLevel, upperLevel, StreamAccessor::getStream(stream));
}
void cv::gpu::Stream::enqueueMemSet(GpuMat& src, Scalar s)
{
CV_Assert((src.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
if (s[0] == 0.0 && s[1] == 0.0 && s[2] == 0.0 && s[3] == 0.0)
{
cudaSafeCall( cudaMemset2DAsync(src.data, src.step, 0, src.cols * src.elemSize(), src.rows, Impl::getStream(impl)) );
return;
}
if (src.depth() == CV_8U)
{
int cn = src.channels();
if (cn == 1 || (cn == 2 && s[0] == s[1]) || (cn == 3 && s[0] == s[1] && s[0] == s[2]) || (cn == 4 && s[0] == s[1] && s[0] == s[2] && s[0] == s[3]))
{
int val = saturate_cast<uchar>(s[0]);
cudaSafeCall( cudaMemset2DAsync(src.data, src.step, val, src.cols * src.elemSize(), src.rows, Impl::getStream(impl)) );
return;
}
}
setTo(src, s, Impl::getStream(impl));
}
