mirror of
https://github.com/opencv/opencv.git
synced 2026-07-30 07:43:03 +04:00
gpuarithm module for arithmetics operations on matrices
This commit is contained in:
Vladislav Vinogradov
@@ -1,565 +0,0 @@
|
||||
/*M///////////////////////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
|
||||
//
|
||||
// By downloading, copying, installing or using the software you agree to this license.
|
||||
// If you do not agree to this license, do not download, install,
|
||||
// copy or use the software.
|
||||
//
|
||||
//
|
||||
// License Agreement
|
||||
// For Open Source Computer Vision Library
|
||||
//
|
||||
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
|
||||
// Third party copyrights are property of their respective owners.
|
||||
//
|
||||
// Redistribution and use in source and binary forms, with or without modification,
|
||||
// are permitted provided that the following conditions are met:
|
||||
//
|
||||
// * Redistribution's of source code must retain the above copyright notice,
|
||||
// this list of conditions and the following disclaimer.
|
||||
//
|
||||
// * Redistribution's in binary form must reproduce the above copyright notice,
|
||||
// this list of conditions and the following disclaimer in the documentation
|
||||
// and/or other materials provided with the distribution.
|
||||
//
|
||||
// * The name of the copyright holders may not be used to endorse or promote products
|
||||
// derived from this software without specific prior written permission.
|
||||
//
|
||||
// This software is provided by the copyright holders and contributors "as is" and
|
||||
// any express or implied warranties, including, but not limited to, the implied
|
||||
// warranties of merchantability and fitness for a particular purpose are disclaimed.
|
||||
// In no event shall the Intel Corporation or contributors be liable for any direct,
|
||||
// indirect, incidental, special, exemplary, or consequential damages
|
||||
// (including, but not limited to, procurement of substitute goods or services;
|
||||
// loss of use, data, or profits; or business interruption) however caused
|
||||
// and on any theory of liability, whether in contract, strict liability,
|
||||
// or tort (including negligence or otherwise) arising in any way out of
|
||||
// the use of this software, even if advised of the possibility of such damage.
|
||||
//
|
||||
//M*/
|
||||
|
||||
#include "precomp.hpp"
|
||||
|
||||
using namespace cv;
|
||||
using namespace cv::gpu;
|
||||
|
||||
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
|
||||
|
||||
void cv::gpu::gemm(const GpuMat&, const GpuMat&, double, const GpuMat&, double, GpuMat&, int, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::transpose(const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::flip(const GpuMat&, GpuMat&, int, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::LUT(const GpuMat&, const Mat&, GpuMat&, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::magnitude(const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::magnitudeSqr(const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::magnitude(const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::magnitudeSqr(const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::phase(const GpuMat&, const GpuMat&, GpuMat&, bool, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::cartToPolar(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::polarToCart(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool, Stream&) { throw_no_cuda(); }
|
||||
void cv::gpu::normalize(const GpuMat&, GpuMat&, double, double, int, int, const GpuMat&) { throw_no_cuda(); }
|
||||
void cv::gpu::normalize(const GpuMat&, GpuMat&, double, double, int, int, const GpuMat&, GpuMat&, GpuMat&) { throw_no_cuda(); }
|
||||
|
||||
#else /* !defined (HAVE_CUDA) */
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// gemm
|
||||
|
||||
void cv::gpu::gemm(const GpuMat& src1, const GpuMat& src2, double alpha, const GpuMat& src3, double beta, GpuMat& dst, int flags, Stream& stream)
|
||||
{
|
||||
#ifndef HAVE_CUBLAS
|
||||
(void)src1;
|
||||
(void)src2;
|
||||
(void)alpha;
|
||||
(void)src3;
|
||||
(void)beta;
|
||||
(void)dst;
|
||||
(void)flags;
|
||||
(void)stream;
|
||||
CV_Error(cv::Error::StsNotImplemented, "The library was build without CUBLAS");
|
||||
#else
|
||||
// CUBLAS works with column-major matrices
|
||||
|
||||
CV_Assert(src1.type() == CV_32FC1 || src1.type() == CV_32FC2 || src1.type() == CV_64FC1 || src1.type() == CV_64FC2);
|
||||
CV_Assert(src2.type() == src1.type() && (src3.empty() || src3.type() == src1.type()));
|
||||
|
||||
if (src1.depth() == CV_64F)
|
||||
{
|
||||
if (!deviceSupports(NATIVE_DOUBLE))
|
||||
CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
|
||||
}
|
||||
|
||||
bool tr1 = (flags & GEMM_1_T) != 0;
|
||||
bool tr2 = (flags & GEMM_2_T) != 0;
|
||||
bool tr3 = (flags & GEMM_3_T) != 0;
|
||||
|
||||
if (src1.type() == CV_64FC2)
|
||||
{
|
||||
if (tr1 || tr2 || tr3)
|
||||
CV_Error(cv::Error::StsNotImplemented, "transpose operation doesn't implemented for CV_64FC2 type");
|
||||
}
|
||||
|
||||
Size src1Size = tr1 ? Size(src1.rows, src1.cols) : src1.size();
|
||||
Size src2Size = tr2 ? Size(src2.rows, src2.cols) : src2.size();
|
||||
Size src3Size = tr3 ? Size(src3.rows, src3.cols) : src3.size();
|
||||
Size dstSize(src2Size.width, src1Size.height);
|
||||
|
||||
CV_Assert(src1Size.width == src2Size.height);
|
||||
CV_Assert(src3.empty() || src3Size == dstSize);
|
||||
|
||||
dst.create(dstSize, src1.type());
|
||||
|
||||
if (beta != 0)
|
||||
{
|
||||
if (src3.empty())
|
||||
{
|
||||
if (stream)
|
||||
stream.enqueueMemSet(dst, Scalar::all(0));
|
||||
else
|
||||
dst.setTo(Scalar::all(0));
|
||||
}
|
||||
else
|
||||
{
|
||||
if (tr3)
|
||||
{
|
||||
transpose(src3, dst, stream);
|
||||
}
|
||||
else
|
||||
{
|
||||
if (stream)
|
||||
stream.enqueueCopy(src3, dst);
|
||||
else
|
||||
src3.copyTo(dst);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
cublasHandle_t handle;
|
||||
cublasSafeCall( cublasCreate_v2(&handle) );
|
||||
|
||||
cublasSafeCall( cublasSetStream_v2(handle, StreamAccessor::getStream(stream)) );
|
||||
|
||||
cublasSafeCall( cublasSetPointerMode_v2(handle, CUBLAS_POINTER_MODE_HOST) );
|
||||
|
||||
const float alphaf = static_cast<float>(alpha);
|
||||
const float betaf = static_cast<float>(beta);
|
||||
|
||||
const cuComplex alphacf = make_cuComplex(alphaf, 0);
|
||||
const cuComplex betacf = make_cuComplex(betaf, 0);
|
||||
|
||||
const cuDoubleComplex alphac = make_cuDoubleComplex(alpha, 0);
|
||||
const cuDoubleComplex betac = make_cuDoubleComplex(beta, 0);
|
||||
|
||||
cublasOperation_t transa = tr2 ? CUBLAS_OP_T : CUBLAS_OP_N;
|
||||
cublasOperation_t transb = tr1 ? CUBLAS_OP_T : CUBLAS_OP_N;
|
||||
|
||||
switch (src1.type())
|
||||
{
|
||||
case CV_32FC1:
|
||||
cublasSafeCall( cublasSgemm_v2(handle, transa, transb, tr2 ? src2.rows : src2.cols, tr1 ? src1.cols : src1.rows, tr2 ? src2.cols : src2.rows,
|
||||
&alphaf,
|
||||
src2.ptr<float>(), static_cast<int>(src2.step / sizeof(float)),
|
||||
src1.ptr<float>(), static_cast<int>(src1.step / sizeof(float)),
|
||||
&betaf,
|
||||
dst.ptr<float>(), static_cast<int>(dst.step / sizeof(float))) );
|
||||
break;
|
||||
|
||||
case CV_64FC1:
|
||||
cublasSafeCall( cublasDgemm_v2(handle, transa, transb, tr2 ? src2.rows : src2.cols, tr1 ? src1.cols : src1.rows, tr2 ? src2.cols : src2.rows,
|
||||
&alpha,
|
||||
src2.ptr<double>(), static_cast<int>(src2.step / sizeof(double)),
|
||||
src1.ptr<double>(), static_cast<int>(src1.step / sizeof(double)),
|
||||
&beta,
|
||||
dst.ptr<double>(), static_cast<int>(dst.step / sizeof(double))) );
|
||||
break;
|
||||
|
||||
case CV_32FC2:
|
||||
cublasSafeCall( cublasCgemm_v2(handle, transa, transb, tr2 ? src2.rows : src2.cols, tr1 ? src1.cols : src1.rows, tr2 ? src2.cols : src2.rows,
|
||||
&alphacf,
|
||||
src2.ptr<cuComplex>(), static_cast<int>(src2.step / sizeof(cuComplex)),
|
||||
src1.ptr<cuComplex>(), static_cast<int>(src1.step / sizeof(cuComplex)),
|
||||
&betacf,
|
||||
dst.ptr<cuComplex>(), static_cast<int>(dst.step / sizeof(cuComplex))) );
|
||||
break;
|
||||
|
||||
case CV_64FC2:
|
||||
cublasSafeCall( cublasZgemm_v2(handle, transa, transb, tr2 ? src2.rows : src2.cols, tr1 ? src1.cols : src1.rows, tr2 ? src2.cols : src2.rows,
|
||||
&alphac,
|
||||
src2.ptr<cuDoubleComplex>(), static_cast<int>(src2.step / sizeof(cuDoubleComplex)),
|
||||
src1.ptr<cuDoubleComplex>(), static_cast<int>(src1.step / sizeof(cuDoubleComplex)),
|
||||
&betac,
|
||||
dst.ptr<cuDoubleComplex>(), static_cast<int>(dst.step / sizeof(cuDoubleComplex))) );
|
||||
break;
|
||||
}
|
||||
|
||||
cublasSafeCall( cublasDestroy_v2(handle) );
|
||||
#endif
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// transpose
|
||||
|
||||
void cv::gpu::transpose(const GpuMat& src, GpuMat& dst, Stream& s)
|
||||
{
|
||||
CV_Assert(src.elemSize() == 1 || src.elemSize() == 4 || src.elemSize() == 8);
|
||||
|
||||
dst.create( src.cols, src.rows, src.type() );
|
||||
|
||||
cudaStream_t stream = StreamAccessor::getStream(s);
|
||||
|
||||
if (src.elemSize() == 1)
|
||||
{
|
||||
NppStreamHandler h(stream);
|
||||
|
||||
NppiSize sz;
|
||||
sz.width = src.cols;
|
||||
sz.height = src.rows;
|
||||
|
||||
nppSafeCall( nppiTranspose_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
|
||||
dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz) );
|
||||
}
|
||||
else if (src.elemSize() == 4)
|
||||
{
|
||||
NppStStreamHandler h(stream);
|
||||
|
||||
NcvSize32u sz;
|
||||
sz.width = src.cols;
|
||||
sz.height = src.rows;
|
||||
|
||||
ncvSafeCall( nppiStTranspose_32u_C1R(const_cast<Ncv32u*>(src.ptr<Ncv32u>()), static_cast<int>(src.step),
|
||||
dst.ptr<Ncv32u>(), static_cast<int>(dst.step), sz) );
|
||||
}
|
||||
else // if (src.elemSize() == 8)
|
||||
{
|
||||
if (!deviceSupports(NATIVE_DOUBLE))
|
||||
CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
|
||||
|
||||
NppStStreamHandler h(stream);
|
||||
|
||||
NcvSize32u sz;
|
||||
sz.width = src.cols;
|
||||
sz.height = src.rows;
|
||||
|
||||
ncvSafeCall( nppiStTranspose_64u_C1R(const_cast<Ncv64u*>(src.ptr<Ncv64u>()), static_cast<int>(src.step),
|
||||
dst.ptr<Ncv64u>(), static_cast<int>(dst.step), sz) );
|
||||
}
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// flip
|
||||
|
||||
namespace
|
||||
{
|
||||
template<int DEPTH> struct NppTypeTraits;
|
||||
template<> struct NppTypeTraits<CV_8U> { typedef Npp8u npp_t; };
|
||||
template<> struct NppTypeTraits<CV_8S> { typedef Npp8s npp_t; };
|
||||
template<> struct NppTypeTraits<CV_16U> { typedef Npp16u npp_t; };
|
||||
template<> struct NppTypeTraits<CV_16S> { typedef Npp16s npp_t; };
|
||||
template<> struct NppTypeTraits<CV_32S> { typedef Npp32s npp_t; };
|
||||
template<> struct NppTypeTraits<CV_32F> { typedef Npp32f npp_t; };
|
||||
template<> struct NppTypeTraits<CV_64F> { typedef Npp64f npp_t; };
|
||||
|
||||
template <int DEPTH> struct NppMirrorFunc
|
||||
{
|
||||
typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
|
||||
|
||||
typedef NppStatus (*func_t)(const npp_t* pSrc, int nSrcStep, npp_t* pDst, int nDstStep, NppiSize oROI, NppiAxis flip);
|
||||
};
|
||||
|
||||
template <int DEPTH, typename NppMirrorFunc<DEPTH>::func_t func> struct NppMirror
|
||||
{
|
||||
typedef typename NppMirrorFunc<DEPTH>::npp_t npp_t;
|
||||
|
||||
static void call(const GpuMat& src, GpuMat& dst, int flipCode, cudaStream_t stream)
|
||||
{
|
||||
NppStreamHandler h(stream);
|
||||
|
||||
NppiSize sz;
|
||||
sz.width = src.cols;
|
||||
sz.height = src.rows;
|
||||
|
||||
nppSafeCall( func(src.ptr<npp_t>(), static_cast<int>(src.step),
|
||||
dst.ptr<npp_t>(), static_cast<int>(dst.step), sz,
|
||||
(flipCode == 0 ? NPP_HORIZONTAL_AXIS : (flipCode > 0 ? NPP_VERTICAL_AXIS : NPP_BOTH_AXIS))) );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
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));
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// LUT
|
||||
|
||||
void cv::gpu::LUT(const GpuMat& src, const Mat& lut, GpuMat& dst, Stream& s)
|
||||
{
|
||||
const int cn = src.channels();
|
||||
|
||||
CV_Assert( src.type() == CV_8UC1 || src.type() == CV_8UC3 );
|
||||
CV_Assert( lut.depth() == CV_8U );
|
||||
CV_Assert( lut.channels() == 1 || lut.channels() == cn );
|
||||
CV_Assert( lut.rows * lut.cols == 256 && lut.isContinuous() );
|
||||
|
||||
dst.create(src.size(), CV_MAKE_TYPE(lut.depth(), cn));
|
||||
|
||||
NppiSize sz;
|
||||
sz.height = src.rows;
|
||||
sz.width = src.cols;
|
||||
|
||||
Mat nppLut;
|
||||
lut.convertTo(nppLut, CV_32S);
|
||||
|
||||
int nValues3[] = {256, 256, 256};
|
||||
|
||||
Npp32s pLevels[256];
|
||||
for (int i = 0; i < 256; ++i)
|
||||
pLevels[i] = i;
|
||||
|
||||
const Npp32s* pLevels3[3];
|
||||
|
||||
#if (CUDA_VERSION <= 4020)
|
||||
pLevels3[0] = pLevels3[1] = pLevels3[2] = pLevels;
|
||||
#else
|
||||
GpuMat d_pLevels;
|
||||
d_pLevels.upload(Mat(1, 256, CV_32S, pLevels));
|
||||
pLevels3[0] = pLevels3[1] = pLevels3[2] = d_pLevels.ptr<Npp32s>();
|
||||
#endif
|
||||
|
||||
cudaStream_t stream = StreamAccessor::getStream(s);
|
||||
NppStreamHandler h(stream);
|
||||
|
||||
if (src.type() == CV_8UC1)
|
||||
{
|
||||
#if (CUDA_VERSION <= 4020)
|
||||
nppSafeCall( nppiLUT_Linear_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
|
||||
dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, nppLut.ptr<Npp32s>(), pLevels, 256) );
|
||||
#else
|
||||
GpuMat d_nppLut(Mat(1, 256, CV_32S, nppLut.data));
|
||||
nppSafeCall( nppiLUT_Linear_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
|
||||
dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, d_nppLut.ptr<Npp32s>(), d_pLevels.ptr<Npp32s>(), 256) );
|
||||
#endif
|
||||
}
|
||||
else
|
||||
{
|
||||
const Npp32s* pValues3[3];
|
||||
|
||||
Mat nppLut3[3];
|
||||
if (nppLut.channels() == 1)
|
||||
{
|
||||
#if (CUDA_VERSION <= 4020)
|
||||
pValues3[0] = pValues3[1] = pValues3[2] = nppLut.ptr<Npp32s>();
|
||||
#else
|
||||
GpuMat d_nppLut(Mat(1, 256, CV_32S, nppLut.data));
|
||||
pValues3[0] = pValues3[1] = pValues3[2] = d_nppLut.ptr<Npp32s>();
|
||||
#endif
|
||||
}
|
||||
else
|
||||
{
|
||||
cv::split(nppLut, nppLut3);
|
||||
|
||||
#if (CUDA_VERSION <= 4020)
|
||||
pValues3[0] = nppLut3[0].ptr<Npp32s>();
|
||||
pValues3[1] = nppLut3[1].ptr<Npp32s>();
|
||||
pValues3[2] = nppLut3[2].ptr<Npp32s>();
|
||||
#else
|
||||
GpuMat d_nppLut0(Mat(1, 256, CV_32S, nppLut3[0].data));
|
||||
GpuMat d_nppLut1(Mat(1, 256, CV_32S, nppLut3[1].data));
|
||||
GpuMat d_nppLut2(Mat(1, 256, CV_32S, nppLut3[2].data));
|
||||
|
||||
pValues3[0] = d_nppLut0.ptr<Npp32s>();
|
||||
pValues3[1] = d_nppLut1.ptr<Npp32s>();
|
||||
pValues3[2] = d_nppLut2.ptr<Npp32s>();
|
||||
#endif
|
||||
}
|
||||
|
||||
nppSafeCall( nppiLUT_Linear_8u_C3R(src.ptr<Npp8u>(), static_cast<int>(src.step),
|
||||
dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, pValues3, pLevels3, nValues3) );
|
||||
}
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// NPP magnitide
|
||||
|
||||
namespace
|
||||
{
|
||||
typedef NppStatus (*nppMagnitude_t)(const Npp32fc* pSrc, int nSrcStep, Npp32f* pDst, int nDstStep, NppiSize oSizeROI);
|
||||
|
||||
inline void npp_magnitude(const GpuMat& src, GpuMat& dst, nppMagnitude_t func, cudaStream_t stream)
|
||||
{
|
||||
CV_Assert(src.type() == CV_32FC2);
|
||||
|
||||
dst.create(src.size(), CV_32FC1);
|
||||
|
||||
NppiSize sz;
|
||||
sz.width = src.cols;
|
||||
sz.height = src.rows;
|
||||
|
||||
NppStreamHandler h(stream);
|
||||
|
||||
nppSafeCall( func(src.ptr<Npp32fc>(), static_cast<int>(src.step), dst.ptr<Npp32f>(), static_cast<int>(dst.step), sz) );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
}
|
||||
}
|
||||
|
||||
void cv::gpu::magnitude(const GpuMat& src, GpuMat& dst, Stream& stream)
|
||||
{
|
||||
npp_magnitude(src, dst, nppiMagnitude_32fc32f_C1R, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
void cv::gpu::magnitudeSqr(const GpuMat& src, GpuMat& dst, Stream& stream)
|
||||
{
|
||||
npp_magnitude(src, dst, nppiMagnitudeSqr_32fc32f_C1R, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// Polar <-> Cart
|
||||
|
||||
namespace cv { namespace gpu { namespace cudev
|
||||
{
|
||||
namespace mathfunc
|
||||
{
|
||||
void cartToPolar_gpu(PtrStepSzf x, PtrStepSzf y, PtrStepSzf mag, bool magSqr, PtrStepSzf angle, bool angleInDegrees, cudaStream_t stream);
|
||||
void polarToCart_gpu(PtrStepSzf mag, PtrStepSzf angle, PtrStepSzf x, PtrStepSzf y, bool angleInDegrees, cudaStream_t stream);
|
||||
}
|
||||
}}}
|
||||
|
||||
namespace
|
||||
{
|
||||
inline void cartToPolar_caller(const GpuMat& x, const GpuMat& y, GpuMat* mag, bool magSqr, GpuMat* angle, bool angleInDegrees, cudaStream_t stream)
|
||||
{
|
||||
using namespace ::cv::gpu::cudev::mathfunc;
|
||||
|
||||
CV_Assert(x.size() == y.size() && x.type() == y.type());
|
||||
CV_Assert(x.depth() == CV_32F);
|
||||
|
||||
if (mag)
|
||||
mag->create(x.size(), x.type());
|
||||
if (angle)
|
||||
angle->create(x.size(), x.type());
|
||||
|
||||
GpuMat x1cn = x.reshape(1);
|
||||
GpuMat y1cn = y.reshape(1);
|
||||
GpuMat mag1cn = mag ? mag->reshape(1) : GpuMat();
|
||||
GpuMat angle1cn = angle ? angle->reshape(1) : GpuMat();
|
||||
|
||||
cartToPolar_gpu(x1cn, y1cn, mag1cn, magSqr, angle1cn, angleInDegrees, stream);
|
||||
}
|
||||
|
||||
inline void polarToCart_caller(const GpuMat& mag, const GpuMat& angle, GpuMat& x, GpuMat& y, bool angleInDegrees, cudaStream_t stream)
|
||||
{
|
||||
using namespace ::cv::gpu::cudev::mathfunc;
|
||||
|
||||
CV_Assert((mag.empty() || mag.size() == angle.size()) && mag.type() == angle.type());
|
||||
CV_Assert(mag.depth() == CV_32F);
|
||||
|
||||
x.create(mag.size(), mag.type());
|
||||
y.create(mag.size(), mag.type());
|
||||
|
||||
GpuMat mag1cn = mag.reshape(1);
|
||||
GpuMat angle1cn = angle.reshape(1);
|
||||
GpuMat x1cn = x.reshape(1);
|
||||
GpuMat y1cn = y.reshape(1);
|
||||
|
||||
polarToCart_gpu(mag1cn, angle1cn, x1cn, y1cn, angleInDegrees, stream);
|
||||
}
|
||||
}
|
||||
|
||||
void cv::gpu::magnitude(const GpuMat& x, const GpuMat& y, GpuMat& dst, Stream& stream)
|
||||
{
|
||||
cartToPolar_caller(x, y, &dst, false, 0, false, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
void cv::gpu::magnitudeSqr(const GpuMat& x, const GpuMat& y, GpuMat& dst, Stream& stream)
|
||||
{
|
||||
cartToPolar_caller(x, y, &dst, true, 0, false, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
void cv::gpu::phase(const GpuMat& x, const GpuMat& y, GpuMat& angle, bool angleInDegrees, Stream& stream)
|
||||
{
|
||||
cartToPolar_caller(x, y, 0, false, &angle, angleInDegrees, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
void cv::gpu::cartToPolar(const GpuMat& x, const GpuMat& y, GpuMat& mag, GpuMat& angle, bool angleInDegrees, Stream& stream)
|
||||
{
|
||||
cartToPolar_caller(x, y, &mag, false, &angle, angleInDegrees, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
void cv::gpu::polarToCart(const GpuMat& magnitude, const GpuMat& angle, GpuMat& x, GpuMat& y, bool angleInDegrees, Stream& stream)
|
||||
{
|
||||
polarToCart_caller(magnitude, angle, x, y, angleInDegrees, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// normalize
|
||||
|
||||
void cv::gpu::normalize(const GpuMat& src, GpuMat& dst, double a, double b, int norm_type, int dtype, const GpuMat& mask)
|
||||
{
|
||||
GpuMat norm_buf;
|
||||
GpuMat cvt_buf;
|
||||
normalize(src, dst, a, b, norm_type, dtype, mask, norm_buf, cvt_buf);
|
||||
}
|
||||
|
||||
void cv::gpu::normalize(const GpuMat& src, GpuMat& dst, double a, double b, int norm_type, int dtype, const GpuMat& mask, GpuMat& norm_buf, GpuMat& cvt_buf)
|
||||
{
|
||||
double scale = 1, shift = 0;
|
||||
if (norm_type == NORM_MINMAX)
|
||||
{
|
||||
double smin = 0, smax = 0;
|
||||
double dmin = std::min(a, b), dmax = std::max(a, b);
|
||||
minMax(src, &smin, &smax, mask, norm_buf);
|
||||
scale = (dmax - dmin) * (smax - smin > std::numeric_limits<double>::epsilon() ? 1.0 / (smax - smin) : 0.0);
|
||||
shift = dmin - smin * scale;
|
||||
}
|
||||
else if (norm_type == NORM_L2 || norm_type == NORM_L1 || norm_type == NORM_INF)
|
||||
{
|
||||
scale = norm(src, norm_type, mask, norm_buf);
|
||||
scale = scale > std::numeric_limits<double>::epsilon() ? a / scale : 0.0;
|
||||
shift = 0;
|
||||
}
|
||||
else
|
||||
{
|
||||
CV_Error(cv::Error::StsBadArg, "Unknown/unsupported norm type");
|
||||
}
|
||||
|
||||
if (mask.empty())
|
||||
{
|
||||
src.convertTo(dst, dtype, scale, shift);
|
||||
}
|
||||
else
|
||||
{
|
||||
src.convertTo(cvt_buf, dtype, scale, shift);
|
||||
cvt_buf.copyTo(dst, mask);
|
||||
}
|
||||
}
|
||||
|
||||
#endif /* !defined (HAVE_CUDA) */
|
||||
File diff suppressed because it is too large
Load Diff
@@ -1,217 +0,0 @@
|
||||
/*M///////////////////////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
|
||||
//
|
||||
// By downloading, copying, installing or using the software you agree to this license.
|
||||
// If you do not agree to this license, do not download, install,
|
||||
// copy or use the software.
|
||||
//
|
||||
//
|
||||
// License Agreement
|
||||
// For Open Source Computer Vision Library
|
||||
//
|
||||
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
|
||||
// Third party copyrights are property of their respective owners.
|
||||
//
|
||||
// Redistribution and use in source and binary forms, with or without modification,
|
||||
// are permitted provided that the following conditions are met:
|
||||
//
|
||||
// * Redistribution's of source code must retain the above copyright notice,
|
||||
// this list of conditions and the following disclaimer.
|
||||
//
|
||||
// * Redistribution's in binary form must reproduce the above copyright notice,
|
||||
// this list of conditions and the following disclaimer in the documentation
|
||||
// and/or other materials provided with the distribution.
|
||||
//
|
||||
// * The name of the copyright holders may not be used to endorse or promote products
|
||||
// derived from this software without specific prior written permission.
|
||||
//
|
||||
// This software is provided by the copyright holders and contributors "as is" and
|
||||
// any express or implied warranties, including, but not limited to, the implied
|
||||
// warranties of merchantability and fitness for a particular purpose are disclaimed.
|
||||
// In no event shall the Intel Corporation or contributors be liable for any direct,
|
||||
// indirect, incidental, special, exemplary, or consequential damages
|
||||
// (including, but not limited to, procurement of substitute goods or services;
|
||||
// loss of use, data, or profits; or business interruption) however caused
|
||||
// and on any theory of liability, whether in contract, strict liability,
|
||||
// or tort (including negligence or otherwise) arising in any way out of
|
||||
// the use of this software, even if advised of the possibility of such damage.
|
||||
//
|
||||
//M*/
|
||||
|
||||
#if !defined CUDA_DISABLER
|
||||
|
||||
#include "opencv2/core/cuda/common.hpp"
|
||||
|
||||
namespace cv { namespace gpu { namespace cudev
|
||||
{
|
||||
namespace mathfunc
|
||||
{
|
||||
//////////////////////////////////////////////////////////////////////////////////////
|
||||
// Cart <-> Polar
|
||||
|
||||
struct Nothing
|
||||
{
|
||||
static __device__ __forceinline__ void calc(int, int, float, float, float*, size_t, float)
|
||||
{
|
||||
}
|
||||
};
|
||||
struct Magnitude
|
||||
{
|
||||
static __device__ __forceinline__ void calc(int x, int y, float x_data, float y_data, float* dst, size_t dst_step, float)
|
||||
{
|
||||
dst[y * dst_step + x] = ::sqrtf(x_data * x_data + y_data * y_data);
|
||||
}
|
||||
};
|
||||
struct MagnitudeSqr
|
||||
{
|
||||
static __device__ __forceinline__ void calc(int x, int y, float x_data, float y_data, float* dst, size_t dst_step, float)
|
||||
{
|
||||
dst[y * dst_step + x] = x_data * x_data + y_data * y_data;
|
||||
}
|
||||
};
|
||||
struct Atan2
|
||||
{
|
||||
static __device__ __forceinline__ void calc(int x, int y, float x_data, float y_data, float* dst, size_t dst_step, float scale)
|
||||
{
|
||||
float angle = ::atan2f(y_data, x_data);
|
||||
angle += (angle < 0) * 2.0f * CV_PI_F;
|
||||
dst[y * dst_step + x] = scale * angle;
|
||||
}
|
||||
};
|
||||
template <typename Mag, typename Angle>
|
||||
__global__ void cartToPolar(const float* xptr, size_t x_step, const float* yptr, size_t y_step,
|
||||
float* mag, size_t mag_step, float* angle, size_t angle_step, float scale, int width, int height)
|
||||
{
|
||||
const int x = blockDim.x * blockIdx.x + threadIdx.x;
|
||||
const int y = blockDim.y * blockIdx.y + threadIdx.y;
|
||||
|
||||
if (x < width && y < height)
|
||||
{
|
||||
float x_data = xptr[y * x_step + x];
|
||||
float y_data = yptr[y * y_step + x];
|
||||
|
||||
Mag::calc(x, y, x_data, y_data, mag, mag_step, scale);
|
||||
Angle::calc(x, y, x_data, y_data, angle, angle_step, scale);
|
||||
}
|
||||
}
|
||||
|
||||
struct NonEmptyMag
|
||||
{
|
||||
static __device__ __forceinline__ float get(const float* mag, size_t mag_step, int x, int y)
|
||||
{
|
||||
return mag[y * mag_step + x];
|
||||
}
|
||||
};
|
||||
struct EmptyMag
|
||||
{
|
||||
static __device__ __forceinline__ float get(const float*, size_t, int, int)
|
||||
{
|
||||
return 1.0f;
|
||||
}
|
||||
};
|
||||
template <typename Mag>
|
||||
__global__ void polarToCart(const float* mag, size_t mag_step, const float* angle, size_t angle_step, float scale,
|
||||
float* xptr, size_t x_step, float* yptr, size_t y_step, int width, int height)
|
||||
{
|
||||
const int x = blockDim.x * blockIdx.x + threadIdx.x;
|
||||
const int y = blockDim.y * blockIdx.y + threadIdx.y;
|
||||
|
||||
if (x < width && y < height)
|
||||
{
|
||||
float mag_data = Mag::get(mag, mag_step, x, y);
|
||||
float angle_data = angle[y * angle_step + x];
|
||||
float sin_a, cos_a;
|
||||
|
||||
::sincosf(scale * angle_data, &sin_a, &cos_a);
|
||||
|
||||
xptr[y * x_step + x] = mag_data * cos_a;
|
||||
yptr[y * y_step + x] = mag_data * sin_a;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename Mag, typename Angle>
|
||||
void cartToPolar_caller(PtrStepSzf x, PtrStepSzf y, PtrStepSzf mag, PtrStepSzf angle, bool angleInDegrees, cudaStream_t stream)
|
||||
{
|
||||
dim3 threads(32, 8, 1);
|
||||
dim3 grid(1, 1, 1);
|
||||
|
||||
grid.x = divUp(x.cols, threads.x);
|
||||
grid.y = divUp(x.rows, threads.y);
|
||||
|
||||
const float scale = angleInDegrees ? (180.0f / CV_PI_F) : 1.f;
|
||||
|
||||
cartToPolar<Mag, Angle><<<grid, threads, 0, stream>>>(
|
||||
x.data, x.step/x.elemSize(), y.data, y.step/y.elemSize(),
|
||||
mag.data, mag.step/mag.elemSize(), angle.data, angle.step/angle.elemSize(), scale, x.cols, x.rows);
|
||||
cudaSafeCall( cudaGetLastError() );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
}
|
||||
|
||||
void cartToPolar_gpu(PtrStepSzf x, PtrStepSzf y, PtrStepSzf mag, bool magSqr, PtrStepSzf angle, bool angleInDegrees, cudaStream_t stream)
|
||||
{
|
||||
typedef void (*caller_t)(PtrStepSzf x, PtrStepSzf y, PtrStepSzf mag, PtrStepSzf angle, bool angleInDegrees, cudaStream_t stream);
|
||||
static const caller_t callers[2][2][2] =
|
||||
{
|
||||
{
|
||||
{
|
||||
cartToPolar_caller<Magnitude, Atan2>,
|
||||
cartToPolar_caller<Magnitude, Nothing>
|
||||
},
|
||||
{
|
||||
cartToPolar_caller<MagnitudeSqr, Atan2>,
|
||||
cartToPolar_caller<MagnitudeSqr, Nothing>,
|
||||
}
|
||||
},
|
||||
{
|
||||
{
|
||||
cartToPolar_caller<Nothing, Atan2>,
|
||||
cartToPolar_caller<Nothing, Nothing>
|
||||
},
|
||||
{
|
||||
cartToPolar_caller<Nothing, Atan2>,
|
||||
cartToPolar_caller<Nothing, Nothing>,
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
callers[mag.data == 0][magSqr][angle.data == 0](x, y, mag, angle, angleInDegrees, stream);
|
||||
}
|
||||
|
||||
template <typename Mag>
|
||||
void polarToCart_caller(PtrStepSzf mag, PtrStepSzf angle, PtrStepSzf x, PtrStepSzf y, bool angleInDegrees, cudaStream_t stream)
|
||||
{
|
||||
dim3 threads(32, 8, 1);
|
||||
dim3 grid(1, 1, 1);
|
||||
|
||||
grid.x = divUp(mag.cols, threads.x);
|
||||
grid.y = divUp(mag.rows, threads.y);
|
||||
|
||||
const float scale = angleInDegrees ? (CV_PI_F / 180.0f) : 1.0f;
|
||||
|
||||
polarToCart<Mag><<<grid, threads, 0, stream>>>(mag.data, mag.step/mag.elemSize(),
|
||||
angle.data, angle.step/angle.elemSize(), scale, x.data, x.step/x.elemSize(), y.data, y.step/y.elemSize(), mag.cols, mag.rows);
|
||||
cudaSafeCall( cudaGetLastError() );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
}
|
||||
|
||||
void polarToCart_gpu(PtrStepSzf mag, PtrStepSzf angle, PtrStepSzf x, PtrStepSzf y, bool angleInDegrees, cudaStream_t stream)
|
||||
{
|
||||
typedef void (*caller_t)(PtrStepSzf mag, PtrStepSzf angle, PtrStepSzf x, PtrStepSzf y, bool angleInDegrees, cudaStream_t stream);
|
||||
static const caller_t callers[2] =
|
||||
{
|
||||
polarToCart_caller<NonEmptyMag>,
|
||||
polarToCart_caller<EmptyMag>
|
||||
};
|
||||
|
||||
callers[mag.data == 0](mag, angle, x, y, angleInDegrees, stream);
|
||||
}
|
||||
} // namespace mathfunc
|
||||
}}} // namespace cv { namespace gpu { namespace cudev
|
||||
|
||||
#endif /* CUDA_DISABLER */
|
||||
File diff suppressed because it is too large
Load Diff
@@ -45,24 +45,20 @@
|
||||
|
||||
#include <cuda_runtime_api.h>
|
||||
#include <cufft.h>
|
||||
#include <cublas.h>
|
||||
#include "NCV.hpp"
|
||||
|
||||
#if defined(__GNUC__)
|
||||
#define ncvSafeCall(expr) ___ncvSafeCall(expr, __FILE__, __LINE__, __func__)
|
||||
#define cufftSafeCall(expr) ___cufftSafeCall(expr, __FILE__, __LINE__, __func__)
|
||||
#define cublasSafeCall(expr) ___cublasSafeCall(expr, __FILE__, __LINE__, __func__)
|
||||
#else /* defined(__CUDACC__) || defined(__MSVC__) */
|
||||
#define ncvSafeCall(expr) ___ncvSafeCall(expr, __FILE__, __LINE__)
|
||||
#define cufftSafeCall(expr) ___cufftSafeCall(expr, __FILE__, __LINE__)
|
||||
#define cublasSafeCall(expr) ___cublasSafeCall(expr, __FILE__, __LINE__)
|
||||
#endif
|
||||
|
||||
namespace cv { namespace gpu
|
||||
{
|
||||
void ncvError(int err, const char *file, const int line, const char *func = "");
|
||||
void cufftError(int err, const char *file, const int line, const char *func = "");
|
||||
void cublasError(int err, const char *file, const int line, const char *func = "");
|
||||
}}
|
||||
|
||||
static inline void ___ncvSafeCall(int err, const char *file, const int line, const char *func = "")
|
||||
@@ -77,10 +73,4 @@ static inline void ___cufftSafeCall(cufftResult_t err, const char *file, const i
|
||||
cv::gpu::cufftError(err, file, line, func);
|
||||
}
|
||||
|
||||
static inline void ___cublasSafeCall(cublasStatus_t err, const char *file, const int line, const char *func = "")
|
||||
{
|
||||
if (CUBLAS_STATUS_SUCCESS != err)
|
||||
cv::gpu::cublasError(err, file, line, func);
|
||||
}
|
||||
|
||||
#endif /* __OPENCV_CUDA_SAFE_CALL_HPP__ */
|
||||
|
||||
@@ -1,511 +0,0 @@
|
||||
/*M///////////////////////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
|
||||
//
|
||||
// By downloading, copying, installing or using the software you agree to this license.
|
||||
// If you do not agree to this license, do not download, install,
|
||||
// copy or use the software.
|
||||
//
|
||||
//
|
||||
// License Agreement
|
||||
// For Open Source Computer Vision Library
|
||||
//
|
||||
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
|
||||
// Third party copyrights are property of their respective owners.
|
||||
//
|
||||
// Redistribution and use in source and binary forms, with or without modification,
|
||||
// are permitted provided that the following conditions are met:
|
||||
//
|
||||
// * Redistribution's of source code must retain the above copyright notice,
|
||||
// this list of conditions and the following disclaimer.
|
||||
//
|
||||
// * Redistribution's in binary form must reproduce the above copyright notice,
|
||||
// this list of conditions and the following disclaimer in the documentation
|
||||
// and/or other materials provided with the distribution.
|
||||
//
|
||||
// * The name of the copyright holders may not be used to endorse or promote products
|
||||
// derived from this software without specific prior written permission.
|
||||
//
|
||||
// This software is provided by the copyright holders and contributors "as is" and
|
||||
// any express or implied warranties, including, but not limited to, the implied
|
||||
// warranties of merchantability and fitness for a particular purpose are disclaimed.
|
||||
// In no event shall the Intel Corporation or contributors be liable for any direct,
|
||||
// indirect, incidental, special, exemplary, or consequential damages
|
||||
// (including, but not limited to, procurement of substitute goods or services;
|
||||
// loss of use, data, or profits; or business interruption) however caused
|
||||
// and on any theory of liability, whether in contract, strict liability,
|
||||
// or tort (including negligence or otherwise) arising in any way out of
|
||||
// the use of this software, even if advised of the possibility of such damage.
|
||||
//
|
||||
//M*/
|
||||
|
||||
#if !defined CUDA_DISABLER
|
||||
|
||||
#include "opencv2/core/cuda/common.hpp"
|
||||
|
||||
namespace cv { namespace gpu { namespace cudev
|
||||
{
|
||||
namespace split_merge
|
||||
{
|
||||
template <typename T, size_t elem_size = sizeof(T)>
|
||||
struct TypeTraits
|
||||
{
|
||||
typedef T type;
|
||||
typedef T type2;
|
||||
typedef T type3;
|
||||
typedef T type4;
|
||||
};
|
||||
|
||||
template <typename T>
|
||||
struct TypeTraits<T, 1>
|
||||
{
|
||||
typedef char type;
|
||||
typedef char2 type2;
|
||||
typedef char3 type3;
|
||||
typedef char4 type4;
|
||||
};
|
||||
|
||||
template <typename T>
|
||||
struct TypeTraits<T, 2>
|
||||
{
|
||||
typedef short type;
|
||||
typedef short2 type2;
|
||||
typedef short3 type3;
|
||||
typedef short4 type4;
|
||||
};
|
||||
|
||||
template <typename T>
|
||||
struct TypeTraits<T, 4>
|
||||
{
|
||||
typedef int type;
|
||||
typedef int2 type2;
|
||||
typedef int3 type3;
|
||||
typedef int4 type4;
|
||||
};
|
||||
|
||||
template <typename T>
|
||||
struct TypeTraits<T, 8>
|
||||
{
|
||||
typedef double type;
|
||||
typedef double2 type2;
|
||||
//typedef double3 type3;
|
||||
//typedef double4 type3;
|
||||
};
|
||||
|
||||
typedef void (*MergeFunction)(const PtrStepSzb* src, PtrStepSzb& dst, const cudaStream_t& stream);
|
||||
typedef void (*SplitFunction)(const PtrStepSzb& src, PtrStepSzb* dst, const cudaStream_t& stream);
|
||||
|
||||
//------------------------------------------------------------
|
||||
// Merge
|
||||
|
||||
template <typename T>
|
||||
__global__ void mergeC2_(const uchar* src0, size_t src0_step,
|
||||
const uchar* src1, size_t src1_step,
|
||||
int rows, int cols, uchar* dst, size_t dst_step)
|
||||
{
|
||||
typedef typename TypeTraits<T>::type2 dst_type;
|
||||
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const T* src0_y = (const T*)(src0 + y * src0_step);
|
||||
const T* src1_y = (const T*)(src1 + y * src1_step);
|
||||
dst_type* dst_y = (dst_type*)(dst + y * dst_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
dst_type dst_elem;
|
||||
dst_elem.x = src0_y[x];
|
||||
dst_elem.y = src1_y[x];
|
||||
dst_y[x] = dst_elem;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
__global__ void mergeC3_(const uchar* src0, size_t src0_step,
|
||||
const uchar* src1, size_t src1_step,
|
||||
const uchar* src2, size_t src2_step,
|
||||
int rows, int cols, uchar* dst, size_t dst_step)
|
||||
{
|
||||
typedef typename TypeTraits<T>::type3 dst_type;
|
||||
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const T* src0_y = (const T*)(src0 + y * src0_step);
|
||||
const T* src1_y = (const T*)(src1 + y * src1_step);
|
||||
const T* src2_y = (const T*)(src2 + y * src2_step);
|
||||
dst_type* dst_y = (dst_type*)(dst + y * dst_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
dst_type dst_elem;
|
||||
dst_elem.x = src0_y[x];
|
||||
dst_elem.y = src1_y[x];
|
||||
dst_elem.z = src2_y[x];
|
||||
dst_y[x] = dst_elem;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <>
|
||||
__global__ void mergeC3_<double>(const uchar* src0, size_t src0_step,
|
||||
const uchar* src1, size_t src1_step,
|
||||
const uchar* src2, size_t src2_step,
|
||||
int rows, int cols, uchar* dst, size_t dst_step)
|
||||
{
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const double* src0_y = (const double*)(src0 + y * src0_step);
|
||||
const double* src1_y = (const double*)(src1 + y * src1_step);
|
||||
const double* src2_y = (const double*)(src2 + y * src2_step);
|
||||
double* dst_y = (double*)(dst + y * dst_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
dst_y[3 * x] = src0_y[x];
|
||||
dst_y[3 * x + 1] = src1_y[x];
|
||||
dst_y[3 * x + 2] = src2_y[x];
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
__global__ void mergeC4_(const uchar* src0, size_t src0_step,
|
||||
const uchar* src1, size_t src1_step,
|
||||
const uchar* src2, size_t src2_step,
|
||||
const uchar* src3, size_t src3_step,
|
||||
int rows, int cols, uchar* dst, size_t dst_step)
|
||||
{
|
||||
typedef typename TypeTraits<T>::type4 dst_type;
|
||||
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const T* src0_y = (const T*)(src0 + y * src0_step);
|
||||
const T* src1_y = (const T*)(src1 + y * src1_step);
|
||||
const T* src2_y = (const T*)(src2 + y * src2_step);
|
||||
const T* src3_y = (const T*)(src3 + y * src3_step);
|
||||
dst_type* dst_y = (dst_type*)(dst + y * dst_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
dst_type dst_elem;
|
||||
dst_elem.x = src0_y[x];
|
||||
dst_elem.y = src1_y[x];
|
||||
dst_elem.z = src2_y[x];
|
||||
dst_elem.w = src3_y[x];
|
||||
dst_y[x] = dst_elem;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <>
|
||||
__global__ void mergeC4_<double>(const uchar* src0, size_t src0_step,
|
||||
const uchar* src1, size_t src1_step,
|
||||
const uchar* src2, size_t src2_step,
|
||||
const uchar* src3, size_t src3_step,
|
||||
int rows, int cols, uchar* dst, size_t dst_step)
|
||||
{
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const double* src0_y = (const double*)(src0 + y * src0_step);
|
||||
const double* src1_y = (const double*)(src1 + y * src1_step);
|
||||
const double* src2_y = (const double*)(src2 + y * src2_step);
|
||||
const double* src3_y = (const double*)(src3 + y * src3_step);
|
||||
double2* dst_y = (double2*)(dst + y * dst_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
dst_y[2 * x] = make_double2(src0_y[x], src1_y[x]);
|
||||
dst_y[2 * x + 1] = make_double2(src2_y[x], src3_y[x]);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
static void mergeC2_(const PtrStepSzb* src, PtrStepSzb& dst, const cudaStream_t& stream)
|
||||
{
|
||||
dim3 block(32, 8);
|
||||
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
|
||||
mergeC2_<T><<<grid, block, 0, stream>>>(
|
||||
src[0].data, src[0].step,
|
||||
src[1].data, src[1].step,
|
||||
dst.rows, dst.cols, dst.data, dst.step);
|
||||
cudaSafeCall( cudaGetLastError() );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall(cudaDeviceSynchronize());
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
static void mergeC3_(const PtrStepSzb* src, PtrStepSzb& dst, const cudaStream_t& stream)
|
||||
{
|
||||
dim3 block(32, 8);
|
||||
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
|
||||
mergeC3_<T><<<grid, block, 0, stream>>>(
|
||||
src[0].data, src[0].step,
|
||||
src[1].data, src[1].step,
|
||||
src[2].data, src[2].step,
|
||||
dst.rows, dst.cols, dst.data, dst.step);
|
||||
cudaSafeCall( cudaGetLastError() );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall(cudaDeviceSynchronize());
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
static void mergeC4_(const PtrStepSzb* src, PtrStepSzb& dst, const cudaStream_t& stream)
|
||||
{
|
||||
dim3 block(32, 8);
|
||||
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
|
||||
mergeC4_<T><<<grid, block, 0, stream>>>(
|
||||
src[0].data, src[0].step,
|
||||
src[1].data, src[1].step,
|
||||
src[2].data, src[2].step,
|
||||
src[3].data, src[3].step,
|
||||
dst.rows, dst.cols, dst.data, dst.step);
|
||||
cudaSafeCall( cudaGetLastError() );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall(cudaDeviceSynchronize());
|
||||
}
|
||||
|
||||
|
||||
void merge_caller(const PtrStepSzb* src, PtrStepSzb& dst,
|
||||
int total_channels, size_t elem_size,
|
||||
const cudaStream_t& stream)
|
||||
{
|
||||
static MergeFunction merge_func_tbl[] =
|
||||
{
|
||||
mergeC2_<char>, mergeC2_<short>, mergeC2_<int>, 0, mergeC2_<double>,
|
||||
mergeC3_<char>, mergeC3_<short>, mergeC3_<int>, 0, mergeC3_<double>,
|
||||
mergeC4_<char>, mergeC4_<short>, mergeC4_<int>, 0, mergeC4_<double>,
|
||||
};
|
||||
|
||||
size_t merge_func_id = (total_channels - 2) * 5 + (elem_size >> 1);
|
||||
MergeFunction merge_func = merge_func_tbl[merge_func_id];
|
||||
|
||||
if (merge_func == 0)
|
||||
CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported channel count or data type");
|
||||
|
||||
merge_func(src, dst, stream);
|
||||
}
|
||||
|
||||
|
||||
|
||||
//------------------------------------------------------------
|
||||
// Split
|
||||
|
||||
|
||||
template <typename T>
|
||||
__global__ void splitC2_(const uchar* src, size_t src_step,
|
||||
int rows, int cols,
|
||||
uchar* dst0, size_t dst0_step,
|
||||
uchar* dst1, size_t dst1_step)
|
||||
{
|
||||
typedef typename TypeTraits<T>::type2 src_type;
|
||||
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const src_type* src_y = (const src_type*)(src + y * src_step);
|
||||
T* dst0_y = (T*)(dst0 + y * dst0_step);
|
||||
T* dst1_y = (T*)(dst1 + y * dst1_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
src_type src_elem = src_y[x];
|
||||
dst0_y[x] = src_elem.x;
|
||||
dst1_y[x] = src_elem.y;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
__global__ void splitC3_(const uchar* src, size_t src_step,
|
||||
int rows, int cols,
|
||||
uchar* dst0, size_t dst0_step,
|
||||
uchar* dst1, size_t dst1_step,
|
||||
uchar* dst2, size_t dst2_step)
|
||||
{
|
||||
typedef typename TypeTraits<T>::type3 src_type;
|
||||
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const src_type* src_y = (const src_type*)(src + y * src_step);
|
||||
T* dst0_y = (T*)(dst0 + y * dst0_step);
|
||||
T* dst1_y = (T*)(dst1 + y * dst1_step);
|
||||
T* dst2_y = (T*)(dst2 + y * dst2_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
src_type src_elem = src_y[x];
|
||||
dst0_y[x] = src_elem.x;
|
||||
dst1_y[x] = src_elem.y;
|
||||
dst2_y[x] = src_elem.z;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <>
|
||||
__global__ void splitC3_<double>(
|
||||
const uchar* src, size_t src_step, int rows, int cols,
|
||||
uchar* dst0, size_t dst0_step,
|
||||
uchar* dst1, size_t dst1_step,
|
||||
uchar* dst2, size_t dst2_step)
|
||||
{
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const double* src_y = (const double*)(src + y * src_step);
|
||||
double* dst0_y = (double*)(dst0 + y * dst0_step);
|
||||
double* dst1_y = (double*)(dst1 + y * dst1_step);
|
||||
double* dst2_y = (double*)(dst2 + y * dst2_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
dst0_y[x] = src_y[3 * x];
|
||||
dst1_y[x] = src_y[3 * x + 1];
|
||||
dst2_y[x] = src_y[3 * x + 2];
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
__global__ void splitC4_(const uchar* src, size_t src_step, int rows, int cols,
|
||||
uchar* dst0, size_t dst0_step,
|
||||
uchar* dst1, size_t dst1_step,
|
||||
uchar* dst2, size_t dst2_step,
|
||||
uchar* dst3, size_t dst3_step)
|
||||
{
|
||||
typedef typename TypeTraits<T>::type4 src_type;
|
||||
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const src_type* src_y = (const src_type*)(src + y * src_step);
|
||||
T* dst0_y = (T*)(dst0 + y * dst0_step);
|
||||
T* dst1_y = (T*)(dst1 + y * dst1_step);
|
||||
T* dst2_y = (T*)(dst2 + y * dst2_step);
|
||||
T* dst3_y = (T*)(dst3 + y * dst3_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
src_type src_elem = src_y[x];
|
||||
dst0_y[x] = src_elem.x;
|
||||
dst1_y[x] = src_elem.y;
|
||||
dst2_y[x] = src_elem.z;
|
||||
dst3_y[x] = src_elem.w;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
template <>
|
||||
__global__ void splitC4_<double>(
|
||||
const uchar* src, size_t src_step, int rows, int cols,
|
||||
uchar* dst0, size_t dst0_step,
|
||||
uchar* dst1, size_t dst1_step,
|
||||
uchar* dst2, size_t dst2_step,
|
||||
uchar* dst3, size_t dst3_step)
|
||||
{
|
||||
const int x = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
const int y = blockIdx.y * blockDim.y + threadIdx.y;
|
||||
|
||||
const double2* src_y = (const double2*)(src + y * src_step);
|
||||
double* dst0_y = (double*)(dst0 + y * dst0_step);
|
||||
double* dst1_y = (double*)(dst1 + y * dst1_step);
|
||||
double* dst2_y = (double*)(dst2 + y * dst2_step);
|
||||
double* dst3_y = (double*)(dst3 + y * dst3_step);
|
||||
|
||||
if (x < cols && y < rows)
|
||||
{
|
||||
double2 src_elem1 = src_y[2 * x];
|
||||
double2 src_elem2 = src_y[2 * x + 1];
|
||||
dst0_y[x] = src_elem1.x;
|
||||
dst1_y[x] = src_elem1.y;
|
||||
dst2_y[x] = src_elem2.x;
|
||||
dst3_y[x] = src_elem2.y;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
static void splitC2_(const PtrStepSzb& src, PtrStepSzb* dst, const cudaStream_t& stream)
|
||||
{
|
||||
dim3 block(32, 8);
|
||||
dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
|
||||
splitC2_<T><<<grid, block, 0, stream>>>(
|
||||
src.data, src.step, src.rows, src.cols,
|
||||
dst[0].data, dst[0].step,
|
||||
dst[1].data, dst[1].step);
|
||||
cudaSafeCall( cudaGetLastError() );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall(cudaDeviceSynchronize());
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
static void splitC3_(const PtrStepSzb& src, PtrStepSzb* dst, const cudaStream_t& stream)
|
||||
{
|
||||
dim3 block(32, 8);
|
||||
dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
|
||||
splitC3_<T><<<grid, block, 0, stream>>>(
|
||||
src.data, src.step, src.rows, src.cols,
|
||||
dst[0].data, dst[0].step,
|
||||
dst[1].data, dst[1].step,
|
||||
dst[2].data, dst[2].step);
|
||||
cudaSafeCall( cudaGetLastError() );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall(cudaDeviceSynchronize());
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
static void splitC4_(const PtrStepSzb& src, PtrStepSzb* dst, const cudaStream_t& stream)
|
||||
{
|
||||
dim3 block(32, 8);
|
||||
dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
|
||||
splitC4_<T><<<grid, block, 0, stream>>>(
|
||||
src.data, src.step, src.rows, src.cols,
|
||||
dst[0].data, dst[0].step,
|
||||
dst[1].data, dst[1].step,
|
||||
dst[2].data, dst[2].step,
|
||||
dst[3].data, dst[3].step);
|
||||
cudaSafeCall( cudaGetLastError() );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall(cudaDeviceSynchronize());
|
||||
}
|
||||
|
||||
|
||||
void split_caller(const PtrStepSzb& src, PtrStepSzb* dst, int num_channels, size_t elem_size1, const cudaStream_t& stream)
|
||||
{
|
||||
static SplitFunction split_func_tbl[] =
|
||||
{
|
||||
splitC2_<char>, splitC2_<short>, splitC2_<int>, 0, splitC2_<double>,
|
||||
splitC3_<char>, splitC3_<short>, splitC3_<int>, 0, splitC3_<double>,
|
||||
splitC4_<char>, splitC4_<short>, splitC4_<int>, 0, splitC4_<double>,
|
||||
};
|
||||
|
||||
size_t split_func_id = (num_channels - 2) * 5 + (elem_size1 >> 1);
|
||||
SplitFunction split_func = split_func_tbl[split_func_id];
|
||||
|
||||
if (split_func == 0)
|
||||
CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported channel count or data type");
|
||||
|
||||
split_func(src, dst, stream);
|
||||
}
|
||||
} // namespace split_merge
|
||||
}}} // namespace cv { namespace gpu { namespace cudev
|
||||
|
||||
|
||||
#endif /* CUDA_DISABLER */
|
||||
File diff suppressed because it is too large
Load Diff
@@ -142,23 +142,6 @@ namespace
|
||||
};
|
||||
|
||||
const int cufft_error_num = sizeof(cufft_errors) / sizeof(cufft_errors[0]);
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////
|
||||
// CUBLAS errors
|
||||
|
||||
const ErrorEntry cublas_errors[] =
|
||||
{
|
||||
error_entry( CUBLAS_STATUS_SUCCESS ),
|
||||
error_entry( CUBLAS_STATUS_NOT_INITIALIZED ),
|
||||
error_entry( CUBLAS_STATUS_ALLOC_FAILED ),
|
||||
error_entry( CUBLAS_STATUS_INVALID_VALUE ),
|
||||
error_entry( CUBLAS_STATUS_ARCH_MISMATCH ),
|
||||
error_entry( CUBLAS_STATUS_MAPPING_ERROR ),
|
||||
error_entry( CUBLAS_STATUS_EXECUTION_FAILED ),
|
||||
error_entry( CUBLAS_STATUS_INTERNAL_ERROR )
|
||||
};
|
||||
|
||||
const int cublas_error_num = sizeof(cublas_errors) / sizeof(cublas_errors[0]);
|
||||
}
|
||||
|
||||
namespace cv
|
||||
@@ -176,12 +159,6 @@ namespace cv
|
||||
String msg = getErrorString(code, cufft_errors, cufft_error_num);
|
||||
cv::error(cv::Error::GpuApiCallError, msg, func, file, line);
|
||||
}
|
||||
|
||||
void cublasError(int code, const char* file, const int line, const char* func)
|
||||
{
|
||||
String msg = getErrorString(code, cublas_errors, cublas_error_num);
|
||||
cv::error(cv::Error::GpuApiCallError, msg, func, file, line);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -92,6 +92,7 @@ void cv::gpu::Canny(const GpuMat&, const GpuMat&, CannyBuf&, GpuMat&, double, do
|
||||
void cv::gpu::CannyBuf::create(const Size&, int) { throw_no_cuda(); }
|
||||
void cv::gpu::CannyBuf::release() { throw_no_cuda(); }
|
||||
cv::Ptr<cv::gpu::CLAHE> cv::gpu::createCLAHE(double, cv::Size) { throw_no_cuda(); return cv::Ptr<cv::gpu::CLAHE>(); }
|
||||
void cv::gpu::alphaComp(const GpuMat&, const GpuMat&, GpuMat&, int, Stream&) { throw_no_cuda(); }
|
||||
|
||||
#else /* !defined (HAVE_CUDA) */
|
||||
|
||||
@@ -1672,4 +1673,77 @@ cv::Ptr<cv::gpu::CLAHE> cv::gpu::createCLAHE(double clipLimit, cv::Size tileGrid
|
||||
return new CLAHE_Impl(clipLimit, tileGridSize.width, tileGridSize.height);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// alphaComp
|
||||
|
||||
namespace
|
||||
{
|
||||
template <int DEPTH> struct NppAlphaCompFunc
|
||||
{
|
||||
typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
|
||||
|
||||
typedef NppStatus (*func_t)(const npp_t* pSrc1, int nSrc1Step, const npp_t* pSrc2, int nSrc2Step, npp_t* pDst, int nDstStep, NppiSize oSizeROI, NppiAlphaOp eAlphaOp);
|
||||
};
|
||||
|
||||
template <int DEPTH, typename NppAlphaCompFunc<DEPTH>::func_t func> struct NppAlphaComp
|
||||
{
|
||||
typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
|
||||
|
||||
static void call(const GpuMat& img1, const GpuMat& img2, GpuMat& dst, NppiAlphaOp eAlphaOp, cudaStream_t stream)
|
||||
{
|
||||
NppStreamHandler h(stream);
|
||||
|
||||
NppiSize oSizeROI;
|
||||
oSizeROI.width = img1.cols;
|
||||
oSizeROI.height = img2.rows;
|
||||
|
||||
nppSafeCall( func(img1.ptr<npp_t>(), static_cast<int>(img1.step), img2.ptr<npp_t>(), static_cast<int>(img2.step),
|
||||
dst.ptr<npp_t>(), static_cast<int>(dst.step), oSizeROI, eAlphaOp) );
|
||||
|
||||
if (stream == 0)
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
void cv::gpu::alphaComp(const GpuMat& img1, const GpuMat& img2, GpuMat& dst, int alpha_op, Stream& stream)
|
||||
{
|
||||
static const NppiAlphaOp npp_alpha_ops[] = {
|
||||
NPPI_OP_ALPHA_OVER,
|
||||
NPPI_OP_ALPHA_IN,
|
||||
NPPI_OP_ALPHA_OUT,
|
||||
NPPI_OP_ALPHA_ATOP,
|
||||
NPPI_OP_ALPHA_XOR,
|
||||
NPPI_OP_ALPHA_PLUS,
|
||||
NPPI_OP_ALPHA_OVER_PREMUL,
|
||||
NPPI_OP_ALPHA_IN_PREMUL,
|
||||
NPPI_OP_ALPHA_OUT_PREMUL,
|
||||
NPPI_OP_ALPHA_ATOP_PREMUL,
|
||||
NPPI_OP_ALPHA_XOR_PREMUL,
|
||||
NPPI_OP_ALPHA_PLUS_PREMUL,
|
||||
NPPI_OP_ALPHA_PREMUL
|
||||
};
|
||||
|
||||
typedef void (*func_t)(const GpuMat& img1, const GpuMat& img2, GpuMat& dst, NppiAlphaOp eAlphaOp, cudaStream_t stream);
|
||||
|
||||
static const func_t funcs[] =
|
||||
{
|
||||
NppAlphaComp<CV_8U, nppiAlphaComp_8u_AC4R>::call,
|
||||
0,
|
||||
NppAlphaComp<CV_16U, nppiAlphaComp_16u_AC4R>::call,
|
||||
0,
|
||||
NppAlphaComp<CV_32S, nppiAlphaComp_32s_AC4R>::call,
|
||||
NppAlphaComp<CV_32F, nppiAlphaComp_32f_AC4R>::call
|
||||
};
|
||||
|
||||
CV_Assert( img1.type() == CV_8UC4 || img1.type() == CV_16UC4 || img1.type() == CV_32SC4 || img1.type() == CV_32FC4 );
|
||||
CV_Assert( img1.size() == img2.size() && img1.type() == img2.type() );
|
||||
|
||||
dst.create(img1.size(), img1.type());
|
||||
|
||||
const func_t func = funcs[img1.depth()];
|
||||
|
||||
func(img1, img2, dst, npp_alpha_ops[alpha_op], StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
#endif /* !defined (HAVE_CUDA) */
|
||||
|
||||
@@ -1,700 +0,0 @@
|
||||
/*M///////////////////////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
|
||||
//
|
||||
// By downloading, copying, installing or using the software you agree to this license.
|
||||
// If you do not agree to this license, do not download, install,
|
||||
// copy or use the software.
|
||||
//
|
||||
//
|
||||
// License Agreement
|
||||
// For Open Source Computer Vision Library
|
||||
//
|
||||
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
|
||||
// Third party copyrights are property of their respective owners.
|
||||
//
|
||||
// Redistribution and use in source and binary forms, with or without modification,
|
||||
// are permitted provided that the following conditions are met:
|
||||
//
|
||||
// * Redistribution's of source code must retain the above copyright notice,
|
||||
// this list of conditions and the following disclaimer.
|
||||
//
|
||||
// * Redistribution's in binary form must reproduce the above copyright notice,
|
||||
// this list of conditions and the following disclaimer in the documentation
|
||||
// and/or other materials provided with the distribution.
|
||||
//
|
||||
// * The name of the copyright holders may not be used to endorse or promote products
|
||||
// derived from this software without specific prior written permission.
|
||||
//
|
||||
// This software is provided by the copyright holders and contributors "as is" and
|
||||
// any express or implied warranties, including, but not limited to, the implied
|
||||
// warranties of merchantability and fitness for a particular purpose are disclaimed.
|
||||
// In no event shall the Intel Corporation or contributors be liable for any direct,
|
||||
// indirect, incidental, special, exemplary, or consequential damages
|
||||
// (including, but not limited to, procurement of substitute goods or services;
|
||||
// loss of use, data, or profits; or business interruption) however caused
|
||||
// and on any theory of liability, whether in contract, strict liability,
|
||||
// or tort (including negligence or otherwise) arising in any way out of
|
||||
// the use of this software, even if advised of the possibility of such damage.
|
||||
//
|
||||
//M*/
|
||||
|
||||
#include "precomp.hpp"
|
||||
|
||||
using namespace cv;
|
||||
using namespace cv::gpu;
|
||||
|
||||
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
|
||||
|
||||
void cv::gpu::meanStdDev(const GpuMat&, Scalar&, Scalar&) { throw_no_cuda(); }
|
||||
void cv::gpu::meanStdDev(const GpuMat&, Scalar&, Scalar&, GpuMat&) { throw_no_cuda(); }
|
||||
double cv::gpu::norm(const GpuMat&, int) { throw_no_cuda(); return 0.0; }
|
||||
double cv::gpu::norm(const GpuMat&, int, GpuMat&) { throw_no_cuda(); return 0.0; }
|
||||
double cv::gpu::norm(const GpuMat&, int, const GpuMat&, GpuMat&) { throw_no_cuda(); return 0.0; }
|
||||
double cv::gpu::norm(const GpuMat&, const GpuMat&, int) { throw_no_cuda(); return 0.0; }
|
||||
Scalar cv::gpu::sum(const GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
Scalar cv::gpu::sum(const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
Scalar cv::gpu::sum(const GpuMat&, const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
Scalar cv::gpu::absSum(const GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
Scalar cv::gpu::absSum(const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
Scalar cv::gpu::absSum(const GpuMat&, const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
Scalar cv::gpu::sqrSum(const GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
Scalar cv::gpu::sqrSum(const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
Scalar cv::gpu::sqrSum(const GpuMat&, const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
|
||||
void cv::gpu::minMax(const GpuMat&, double*, double*, const GpuMat&) { throw_no_cuda(); }
|
||||
void cv::gpu::minMax(const GpuMat&, double*, double*, const GpuMat&, GpuMat&) { throw_no_cuda(); }
|
||||
void cv::gpu::minMaxLoc(const GpuMat&, double*, double*, Point*, Point*, const GpuMat&) { throw_no_cuda(); }
|
||||
void cv::gpu::minMaxLoc(const GpuMat&, double*, double*, Point*, Point*, const GpuMat&, GpuMat&, GpuMat&) { throw_no_cuda(); }
|
||||
int cv::gpu::countNonZero(const GpuMat&) { throw_no_cuda(); return 0; }
|
||||
int cv::gpu::countNonZero(const GpuMat&, GpuMat&) { throw_no_cuda(); return 0; }
|
||||
void cv::gpu::reduce(const GpuMat&, GpuMat&, int, int, int, Stream&) { throw_no_cuda(); }
|
||||
|
||||
#else
|
||||
#include "opencv2/core/utility.hpp"
|
||||
|
||||
namespace
|
||||
{
|
||||
class DeviceBuffer
|
||||
{
|
||||
public:
|
||||
explicit DeviceBuffer(int count_ = 1) : count(count_)
|
||||
{
|
||||
cudaSafeCall( cudaMalloc(&pdev, count * sizeof(double)) );
|
||||
}
|
||||
~DeviceBuffer()
|
||||
{
|
||||
cudaSafeCall( cudaFree(pdev) );
|
||||
}
|
||||
|
||||
operator double*() {return pdev;}
|
||||
|
||||
void download(double* hptr)
|
||||
{
|
||||
double hbuf;
|
||||
cudaSafeCall( cudaMemcpy(&hbuf, pdev, sizeof(double), cudaMemcpyDeviceToHost) );
|
||||
*hptr = hbuf;
|
||||
}
|
||||
void download(double** hptrs)
|
||||
{
|
||||
AutoBuffer<double, 2 * sizeof(double)> hbuf(count);
|
||||
cudaSafeCall( cudaMemcpy((void*)hbuf, pdev, count * sizeof(double), cudaMemcpyDeviceToHost) );
|
||||
for (int i = 0; i < count; ++i)
|
||||
*hptrs[i] = hbuf[i];
|
||||
}
|
||||
|
||||
private:
|
||||
double* pdev;
|
||||
int count;
|
||||
};
|
||||
}
|
||||
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// meanStdDev
|
||||
|
||||
void cv::gpu::meanStdDev(const GpuMat& src, Scalar& mean, Scalar& stddev)
|
||||
{
|
||||
GpuMat buf;
|
||||
meanStdDev(src, mean, stddev, buf);
|
||||
}
|
||||
|
||||
void cv::gpu::meanStdDev(const GpuMat& src, Scalar& mean, Scalar& stddev, GpuMat& buf)
|
||||
{
|
||||
CV_Assert(src.type() == CV_8UC1);
|
||||
|
||||
if (!deviceSupports(FEATURE_SET_COMPUTE_13))
|
||||
CV_Error(cv::Error::StsNotImplemented, "Not sufficient compute capebility");
|
||||
|
||||
NppiSize sz;
|
||||
sz.width = src.cols;
|
||||
sz.height = src.rows;
|
||||
|
||||
DeviceBuffer dbuf(2);
|
||||
|
||||
int bufSize;
|
||||
#if (CUDA_VERSION <= 4020)
|
||||
nppSafeCall( nppiMeanStdDev8uC1RGetBufferHostSize(sz, &bufSize) );
|
||||
#else
|
||||
nppSafeCall( nppiMeanStdDevGetBufferHostSize_8u_C1R(sz, &bufSize) );
|
||||
#endif
|
||||
|
||||
ensureSizeIsEnough(1, bufSize, CV_8UC1, buf);
|
||||
|
||||
nppSafeCall( nppiMean_StdDev_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step), sz, buf.ptr<Npp8u>(), dbuf, (double*)dbuf + 1) );
|
||||
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
|
||||
double* ptrs[2] = {mean.val, stddev.val};
|
||||
dbuf.download(ptrs);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// norm
|
||||
|
||||
double cv::gpu::norm(const GpuMat& src, int normType)
|
||||
{
|
||||
GpuMat buf;
|
||||
return norm(src, normType, GpuMat(), buf);
|
||||
}
|
||||
|
||||
double cv::gpu::norm(const GpuMat& src, int normType, GpuMat& buf)
|
||||
{
|
||||
return norm(src, normType, GpuMat(), buf);
|
||||
}
|
||||
|
||||
double cv::gpu::norm(const GpuMat& src, int normType, const GpuMat& mask, GpuMat& buf)
|
||||
{
|
||||
CV_Assert(normType == NORM_INF || normType == NORM_L1 || normType == NORM_L2);
|
||||
CV_Assert(mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size() && src.channels() == 1));
|
||||
|
||||
GpuMat src_single_channel = src.reshape(1);
|
||||
|
||||
if (normType == NORM_L1)
|
||||
return absSum(src_single_channel, mask, buf)[0];
|
||||
|
||||
if (normType == NORM_L2)
|
||||
return std::sqrt(sqrSum(src_single_channel, mask, buf)[0]);
|
||||
|
||||
// NORM_INF
|
||||
double min_val, max_val;
|
||||
minMax(src_single_channel, &min_val, &max_val, mask, buf);
|
||||
return std::max(std::abs(min_val), std::abs(max_val));
|
||||
}
|
||||
|
||||
double cv::gpu::norm(const GpuMat& src1, const GpuMat& src2, int normType)
|
||||
{
|
||||
CV_Assert(src1.type() == CV_8UC1);
|
||||
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
|
||||
CV_Assert(normType == NORM_INF || normType == NORM_L1 || normType == NORM_L2);
|
||||
|
||||
typedef NppStatus (*npp_norm_diff_func_t)(const Npp8u* pSrc1, int nSrcStep1, const Npp8u* pSrc2, int nSrcStep2,
|
||||
NppiSize oSizeROI, Npp64f* pRetVal);
|
||||
|
||||
static const npp_norm_diff_func_t npp_norm_diff_func[] = {nppiNormDiff_Inf_8u_C1R, nppiNormDiff_L1_8u_C1R, nppiNormDiff_L2_8u_C1R};
|
||||
|
||||
NppiSize sz;
|
||||
sz.width = src1.cols;
|
||||
sz.height = src1.rows;
|
||||
|
||||
int funcIdx = normType >> 1;
|
||||
|
||||
double retVal;
|
||||
|
||||
DeviceBuffer dbuf;
|
||||
|
||||
nppSafeCall( npp_norm_diff_func[funcIdx](src1.ptr<Npp8u>(), static_cast<int>(src1.step), src2.ptr<Npp8u>(), static_cast<int>(src2.step), sz, dbuf) );
|
||||
|
||||
cudaSafeCall( cudaDeviceSynchronize() );
|
||||
|
||||
dbuf.download(&retVal);
|
||||
|
||||
return retVal;
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// Sum
|
||||
|
||||
namespace sum
|
||||
{
|
||||
void getBufSize(int cols, int rows, int cn, int& bufcols, int& bufrows);
|
||||
|
||||
template <typename T, int cn>
|
||||
void run(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
|
||||
|
||||
template <typename T, int cn>
|
||||
void runAbs(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
|
||||
|
||||
template <typename T, int cn>
|
||||
void runSqr(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::sum(const GpuMat& src)
|
||||
{
|
||||
GpuMat buf;
|
||||
return sum(src, GpuMat(), buf);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::sum(const GpuMat& src, GpuMat& buf)
|
||||
{
|
||||
return sum(src, GpuMat(), buf);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::sum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
|
||||
{
|
||||
typedef void (*func_t)(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
|
||||
static const func_t funcs[7][5] =
|
||||
{
|
||||
{0, ::sum::run<uchar , 1>, ::sum::run<uchar , 2>, ::sum::run<uchar , 3>, ::sum::run<uchar , 4>},
|
||||
{0, ::sum::run<schar , 1>, ::sum::run<schar , 2>, ::sum::run<schar , 3>, ::sum::run<schar , 4>},
|
||||
{0, ::sum::run<ushort, 1>, ::sum::run<ushort, 2>, ::sum::run<ushort, 3>, ::sum::run<ushort, 4>},
|
||||
{0, ::sum::run<short , 1>, ::sum::run<short , 2>, ::sum::run<short , 3>, ::sum::run<short , 4>},
|
||||
{0, ::sum::run<int , 1>, ::sum::run<int , 2>, ::sum::run<int , 3>, ::sum::run<int , 4>},
|
||||
{0, ::sum::run<float , 1>, ::sum::run<float , 2>, ::sum::run<float , 3>, ::sum::run<float , 4>},
|
||||
{0, ::sum::run<double, 1>, ::sum::run<double, 2>, ::sum::run<double, 3>, ::sum::run<double, 4>}
|
||||
};
|
||||
|
||||
CV_Assert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
|
||||
|
||||
if (src.depth() == CV_64F)
|
||||
{
|
||||
if (!deviceSupports(NATIVE_DOUBLE))
|
||||
CV_Error(cv::Error::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()];
|
||||
|
||||
double result[4];
|
||||
func(src, buf.data, result, mask);
|
||||
|
||||
return Scalar(result[0], result[1], result[2], result[3]);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::absSum(const GpuMat& src)
|
||||
{
|
||||
GpuMat buf;
|
||||
return absSum(src, GpuMat(), buf);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::absSum(const GpuMat& src, GpuMat& buf)
|
||||
{
|
||||
return absSum(src, GpuMat(), buf);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::absSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
|
||||
{
|
||||
typedef void (*func_t)(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
|
||||
static const func_t funcs[7][5] =
|
||||
{
|
||||
{0, ::sum::runAbs<uchar , 1>, ::sum::runAbs<uchar , 2>, ::sum::runAbs<uchar , 3>, ::sum::runAbs<uchar , 4>},
|
||||
{0, ::sum::runAbs<schar , 1>, ::sum::runAbs<schar , 2>, ::sum::runAbs<schar , 3>, ::sum::runAbs<schar , 4>},
|
||||
{0, ::sum::runAbs<ushort, 1>, ::sum::runAbs<ushort, 2>, ::sum::runAbs<ushort, 3>, ::sum::runAbs<ushort, 4>},
|
||||
{0, ::sum::runAbs<short , 1>, ::sum::runAbs<short , 2>, ::sum::runAbs<short , 3>, ::sum::runAbs<short , 4>},
|
||||
{0, ::sum::runAbs<int , 1>, ::sum::runAbs<int , 2>, ::sum::runAbs<int , 3>, ::sum::runAbs<int , 4>},
|
||||
{0, ::sum::runAbs<float , 1>, ::sum::runAbs<float , 2>, ::sum::runAbs<float , 3>, ::sum::runAbs<float , 4>},
|
||||
{0, ::sum::runAbs<double, 1>, ::sum::runAbs<double, 2>, ::sum::runAbs<double, 3>, ::sum::runAbs<double, 4>}
|
||||
};
|
||||
|
||||
CV_Assert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
|
||||
|
||||
if (src.depth() == CV_64F)
|
||||
{
|
||||
if (!deviceSupports(NATIVE_DOUBLE))
|
||||
CV_Error(cv::Error::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()];
|
||||
|
||||
double result[4];
|
||||
func(src, buf.data, result, mask);
|
||||
|
||||
return Scalar(result[0], result[1], result[2], result[3]);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::sqrSum(const GpuMat& src)
|
||||
{
|
||||
GpuMat buf;
|
||||
return sqrSum(src, GpuMat(), buf);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::sqrSum(const GpuMat& src, GpuMat& buf)
|
||||
{
|
||||
return sqrSum(src, GpuMat(), buf);
|
||||
}
|
||||
|
||||
Scalar cv::gpu::sqrSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
|
||||
{
|
||||
typedef void (*func_t)(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
|
||||
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>}
|
||||
};
|
||||
|
||||
CV_Assert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
|
||||
|
||||
if (src.depth() == CV_64F)
|
||||
{
|
||||
if (!deviceSupports(NATIVE_DOUBLE))
|
||||
CV_Error(cv::Error::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()];
|
||||
|
||||
double result[4];
|
||||
func(src, buf.data, result, mask);
|
||||
|
||||
return Scalar(result[0], result[1], result[2], result[3]);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// minMax
|
||||
|
||||
namespace minMax
|
||||
{
|
||||
void getBufSize(int cols, int rows, int& bufcols, int& bufrows);
|
||||
|
||||
template <typename T>
|
||||
void run(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
|
||||
}
|
||||
|
||||
void cv::gpu::minMax(const GpuMat& src, double* minVal, double* maxVal, const GpuMat& mask)
|
||||
{
|
||||
GpuMat buf;
|
||||
minMax(src, minVal, maxVal, mask, buf);
|
||||
}
|
||||
|
||||
void cv::gpu::minMax(const GpuMat& src, double* minVal, double* maxVal, const GpuMat& mask, GpuMat& buf)
|
||||
{
|
||||
typedef void (*func_t)(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
|
||||
static const func_t funcs[] =
|
||||
{
|
||||
::minMax::run<uchar>,
|
||||
::minMax::run<schar>,
|
||||
::minMax::run<ushort>,
|
||||
::minMax::run<short>,
|
||||
::minMax::run<int>,
|
||||
::minMax::run<float>,
|
||||
::minMax::run<double>
|
||||
};
|
||||
|
||||
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::Error::StsUnsupportedFormat, "The device doesn't support double");
|
||||
}
|
||||
|
||||
Size buf_size;
|
||||
::minMax::getBufSize(src.cols, src.rows, buf_size.width, buf_size.height);
|
||||
ensureSizeIsEnough(buf_size, CV_8U, buf);
|
||||
|
||||
const func_t func = funcs[src.depth()];
|
||||
|
||||
double temp1, temp2;
|
||||
func(src, mask, minVal ? minVal : &temp1, maxVal ? maxVal : &temp2, buf);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////
|
||||
// minMaxLoc
|
||||
|
||||
namespace minMaxLoc
|
||||
{
|
||||
void getBufSize(int cols, int rows, size_t elem_size, int& b1cols, int& b1rows, int& b2cols, int& b2rows);
|
||||
|
||||
template <typename T>
|
||||
void run(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
|
||||
}
|
||||
|
||||
void cv::gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point* minLoc, Point* maxLoc, const GpuMat& mask)
|
||||
{
|
||||
GpuMat valBuf, locBuf;
|
||||
minMaxLoc(src, minVal, maxVal, minLoc, maxLoc, mask, valBuf, locBuf);
|
||||
}
|
||||
|
||||
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);
|
||||
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>
|
||||
};
|
||||
|
||||
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::Error::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()];
|
||||
|
||||
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);
|
||||
}
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////////
|
||||
// countNonZero
|
||||
|
||||
namespace countNonZero
|
||||
{
|
||||
void getBufSize(int cols, int rows, int& bufcols, int& bufrows);
|
||||
|
||||
template <typename T>
|
||||
int run(const PtrStepSzb src, PtrStep<unsigned int> buf);
|
||||
}
|
||||
|
||||
int cv::gpu::countNonZero(const GpuMat& src)
|
||||
{
|
||||
GpuMat buf;
|
||||
return countNonZero(src, buf);
|
||||
}
|
||||
|
||||
int cv::gpu::countNonZero(const GpuMat& src, GpuMat& buf)
|
||||
{
|
||||
typedef int (*func_t)(const PtrStepSzb src, PtrStep<unsigned int> buf);
|
||||
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>
|
||||
};
|
||||
|
||||
CV_Assert(src.channels() == 1);
|
||||
|
||||
if (src.depth() == CV_64F)
|
||||
{
|
||||
if (!deviceSupports(NATIVE_DOUBLE))
|
||||
CV_Error(cv::Error::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()];
|
||||
|
||||
return func(src, buf);
|
||||
}
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////////
|
||||
// reduce
|
||||
|
||||
namespace reduce
|
||||
{
|
||||
template <typename T, typename S, typename D>
|
||||
void rows(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
|
||||
|
||||
template <typename T, typename S, typename D>
|
||||
void cols(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
|
||||
}
|
||||
|
||||
void cv::gpu::reduce(const GpuMat& src, GpuMat& dst, int dim, int reduceOp, int dtype, Stream& stream)
|
||||
{
|
||||
CV_Assert( src.channels() <= 4 );
|
||||
CV_Assert( dim == 0 || dim == 1 );
|
||||
CV_Assert( reduceOp == REDUCE_SUM || reduceOp == REDUCE_AVG || reduceOp == REDUCE_MAX || reduceOp == REDUCE_MIN );
|
||||
|
||||
if (dtype < 0)
|
||||
dtype = src.depth();
|
||||
|
||||
dst.create(1, dim == 0 ? src.cols : src.rows, CV_MAKE_TYPE(CV_MAT_DEPTH(dtype), src.channels()));
|
||||
|
||||
if (dim == 0)
|
||||
{
|
||||
typedef void (*func_t)(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
|
||||
static const func_t funcs[7][7] =
|
||||
{
|
||||
{
|
||||
::reduce::rows<unsigned char, int, unsigned char>,
|
||||
0/*::reduce::rows<unsigned char, int, signed char>*/,
|
||||
0/*::reduce::rows<unsigned char, int, unsigned short>*/,
|
||||
0/*::reduce::rows<unsigned char, int, short>*/,
|
||||
::reduce::rows<unsigned char, int, int>,
|
||||
::reduce::rows<unsigned char, float, float>,
|
||||
::reduce::rows<unsigned char, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::rows<signed char, int, unsigned char>*/,
|
||||
0/*::reduce::rows<signed char, int, signed char>*/,
|
||||
0/*::reduce::rows<signed char, int, unsigned short>*/,
|
||||
0/*::reduce::rows<signed char, int, short>*/,
|
||||
0/*::reduce::rows<signed char, int, int>*/,
|
||||
0/*::reduce::rows<signed char, float, float>*/,
|
||||
0/*::reduce::rows<signed char, double, double>*/
|
||||
},
|
||||
{
|
||||
0/*::reduce::rows<unsigned short, int, unsigned char>*/,
|
||||
0/*::reduce::rows<unsigned short, int, signed char>*/,
|
||||
::reduce::rows<unsigned short, int, unsigned short>,
|
||||
0/*::reduce::rows<unsigned short, int, short>*/,
|
||||
::reduce::rows<unsigned short, int, int>,
|
||||
::reduce::rows<unsigned short, float, float>,
|
||||
::reduce::rows<unsigned short, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::rows<short, int, unsigned char>*/,
|
||||
0/*::reduce::rows<short, int, signed char>*/,
|
||||
0/*::reduce::rows<short, int, unsigned short>*/,
|
||||
::reduce::rows<short, int, short>,
|
||||
::reduce::rows<short, int, int>,
|
||||
::reduce::rows<short, float, float>,
|
||||
::reduce::rows<short, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::rows<int, int, unsigned char>*/,
|
||||
0/*::reduce::rows<int, int, signed char>*/,
|
||||
0/*::reduce::rows<int, int, unsigned short>*/,
|
||||
0/*::reduce::rows<int, int, short>*/,
|
||||
::reduce::rows<int, int, int>,
|
||||
::reduce::rows<int, float, float>,
|
||||
::reduce::rows<int, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::rows<float, float, unsigned char>*/,
|
||||
0/*::reduce::rows<float, float, signed char>*/,
|
||||
0/*::reduce::rows<float, float, unsigned short>*/,
|
||||
0/*::reduce::rows<float, float, short>*/,
|
||||
0/*::reduce::rows<float, float, int>*/,
|
||||
::reduce::rows<float, float, float>,
|
||||
::reduce::rows<float, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::rows<double, double, unsigned char>*/,
|
||||
0/*::reduce::rows<double, double, signed char>*/,
|
||||
0/*::reduce::rows<double, double, unsigned short>*/,
|
||||
0/*::reduce::rows<double, double, short>*/,
|
||||
0/*::reduce::rows<double, double, int>*/,
|
||||
0/*::reduce::rows<double, double, float>*/,
|
||||
::reduce::rows<double, double, double>
|
||||
}
|
||||
};
|
||||
|
||||
const func_t func = funcs[src.depth()][dst.depth()];
|
||||
|
||||
if (!func)
|
||||
CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of input and output array formats");
|
||||
|
||||
func(src.reshape(1), dst.data, reduceOp, StreamAccessor::getStream(stream));
|
||||
}
|
||||
else
|
||||
{
|
||||
typedef void (*func_t)(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
|
||||
static const func_t funcs[7][7] =
|
||||
{
|
||||
{
|
||||
::reduce::cols<unsigned char, int, unsigned char>,
|
||||
0/*::reduce::cols<unsigned char, int, signed char>*/,
|
||||
0/*::reduce::cols<unsigned char, int, unsigned short>*/,
|
||||
0/*::reduce::cols<unsigned char, int, short>*/,
|
||||
::reduce::cols<unsigned char, int, int>,
|
||||
::reduce::cols<unsigned char, float, float>,
|
||||
::reduce::cols<unsigned char, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::cols<signed char, int, unsigned char>*/,
|
||||
0/*::reduce::cols<signed char, int, signed char>*/,
|
||||
0/*::reduce::cols<signed char, int, unsigned short>*/,
|
||||
0/*::reduce::cols<signed char, int, short>*/,
|
||||
0/*::reduce::cols<signed char, int, int>*/,
|
||||
0/*::reduce::cols<signed char, float, float>*/,
|
||||
0/*::reduce::cols<signed char, double, double>*/
|
||||
},
|
||||
{
|
||||
0/*::reduce::cols<unsigned short, int, unsigned char>*/,
|
||||
0/*::reduce::cols<unsigned short, int, signed char>*/,
|
||||
::reduce::cols<unsigned short, int, unsigned short>,
|
||||
0/*::reduce::cols<unsigned short, int, short>*/,
|
||||
::reduce::cols<unsigned short, int, int>,
|
||||
::reduce::cols<unsigned short, float, float>,
|
||||
::reduce::cols<unsigned short, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::cols<short, int, unsigned char>*/,
|
||||
0/*::reduce::cols<short, int, signed char>*/,
|
||||
0/*::reduce::cols<short, int, unsigned short>*/,
|
||||
::reduce::cols<short, int, short>,
|
||||
::reduce::cols<short, int, int>,
|
||||
::reduce::cols<short, float, float>,
|
||||
::reduce::cols<short, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::cols<int, int, unsigned char>*/,
|
||||
0/*::reduce::cols<int, int, signed char>*/,
|
||||
0/*::reduce::cols<int, int, unsigned short>*/,
|
||||
0/*::reduce::cols<int, int, short>*/,
|
||||
::reduce::cols<int, int, int>,
|
||||
::reduce::cols<int, float, float>,
|
||||
::reduce::cols<int, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::cols<float, float, unsigned char>*/,
|
||||
0/*::reduce::cols<float, float, signed char>*/,
|
||||
0/*::reduce::cols<float, float, unsigned short>*/,
|
||||
0/*::reduce::cols<float, float, short>*/,
|
||||
0/*::reduce::cols<float, float, int>*/,
|
||||
::reduce::cols<float, float, float>,
|
||||
::reduce::cols<float, double, double>
|
||||
},
|
||||
{
|
||||
0/*::reduce::cols<double, double, unsigned char>*/,
|
||||
0/*::reduce::cols<double, double, signed char>*/,
|
||||
0/*::reduce::cols<double, double, unsigned short>*/,
|
||||
0/*::reduce::cols<double, double, short>*/,
|
||||
0/*::reduce::cols<double, double, int>*/,
|
||||
0/*::reduce::cols<double, double, float>*/,
|
||||
::reduce::cols<double, double, double>
|
||||
}
|
||||
};
|
||||
|
||||
const func_t func = funcs[src.depth()][dst.depth()];
|
||||
|
||||
if (!func)
|
||||
CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of input and output array formats");
|
||||
|
||||
func(src, dst.data, src.channels(), reduceOp, StreamAccessor::getStream(stream));
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
@@ -76,10 +76,6 @@
|
||||
#include <cufft.h>
|
||||
#endif
|
||||
|
||||
#ifdef HAVE_CUBLAS
|
||||
#include <cublas.h>
|
||||
#endif
|
||||
|
||||
#include "internal_shared.hpp"
|
||||
#include "opencv2/core/stream_accessor.hpp"
|
||||
|
||||
|
||||
@@ -1,171 +0,0 @@
|
||||
/*M///////////////////////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
|
||||
//
|
||||
// By downloading, copying, installing or using the software you agree to this license.
|
||||
// If you do not agree to this license, do not download, install,
|
||||
// copy or use the software.
|
||||
//
|
||||
//
|
||||
// License Agreement
|
||||
// For Open Source Computer Vision Library
|
||||
//
|
||||
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
|
||||
// Third party copyrights are property of their respective owners.
|
||||
//
|
||||
// Redistribution and use in source and binary forms, with or without modification,
|
||||
// are permitted provided that the following conditions are met:
|
||||
//
|
||||
// * Redistribution's of source code must retain the above copyright notice,
|
||||
// this list of conditions and the following disclaimer.
|
||||
//
|
||||
// * Redistribution's in binary form must reproduce the above copyright notice,
|
||||
// this list of conditions and the following disclaimer in the documentation
|
||||
// and/or other materials provided with the distribution.
|
||||
//
|
||||
// * The name of the copyright holders may not be used to endorse or promote products
|
||||
// derived from this software without specific prior written permission.
|
||||
//
|
||||
// This software is provided by the copyright holders and contributors "as is" and
|
||||
// any express or implied warranties, including, but not limited to, the implied
|
||||
// warranties of merchantability and fitness for a particular purpose are disclaimed.
|
||||
// In no event shall the Intel Corporation or contributors be liable for any direct,
|
||||
// indirect, incidental, special, exemplary, or consequential damages
|
||||
// (including, but not limited to, procurement of substitute goods or services;
|
||||
// loss of use, data, or profits; or business interruption) however caused
|
||||
// and on any theory of liability, whether in contract, strict liability,
|
||||
// or tort (including negligence or otherwise) arising in any way out of
|
||||
// the use of this software, even if advised of the possibility of such damage.
|
||||
//
|
||||
//M*/
|
||||
|
||||
#include "precomp.hpp"
|
||||
|
||||
using namespace cv;
|
||||
using namespace cv::gpu;
|
||||
|
||||
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
|
||||
|
||||
void cv::gpu::merge(const GpuMat* /*src*/, size_t /*count*/, GpuMat& /*dst*/, Stream& /*stream*/) { throw_no_cuda(); }
|
||||
void cv::gpu::merge(const std::vector<GpuMat>& /*src*/, GpuMat& /*dst*/, Stream& /*stream*/) { throw_no_cuda(); }
|
||||
void cv::gpu::split(const GpuMat& /*src*/, GpuMat* /*dst*/, Stream& /*stream*/) { throw_no_cuda(); }
|
||||
void cv::gpu::split(const GpuMat& /*src*/, std::vector<GpuMat>& /*dst*/, Stream& /*stream*/) { throw_no_cuda(); }
|
||||
|
||||
#else /* !defined (HAVE_CUDA) */
|
||||
|
||||
namespace cv { namespace gpu { namespace cudev
|
||||
{
|
||||
namespace split_merge
|
||||
{
|
||||
void merge_caller(const PtrStepSzb* src, PtrStepSzb& dst, int total_channels, size_t elem_size, const cudaStream_t& stream);
|
||||
void split_caller(const PtrStepSzb& src, PtrStepSzb* dst, int num_channels, size_t elem_size1, const cudaStream_t& stream);
|
||||
}
|
||||
}}}
|
||||
|
||||
namespace
|
||||
{
|
||||
void merge(const GpuMat* src, size_t n, GpuMat& dst, const cudaStream_t& stream)
|
||||
{
|
||||
using namespace ::cv::gpu::cudev::split_merge;
|
||||
|
||||
CV_Assert(src);
|
||||
CV_Assert(n > 0);
|
||||
|
||||
int depth = src[0].depth();
|
||||
Size size = src[0].size();
|
||||
|
||||
if (depth == CV_64F)
|
||||
{
|
||||
if (!deviceSupports(NATIVE_DOUBLE))
|
||||
CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
|
||||
}
|
||||
|
||||
bool single_channel_only = true;
|
||||
int total_channels = 0;
|
||||
|
||||
for (size_t i = 0; i < n; ++i)
|
||||
{
|
||||
CV_Assert(src[i].size() == size);
|
||||
CV_Assert(src[i].depth() == depth);
|
||||
single_channel_only = single_channel_only && src[i].channels() == 1;
|
||||
total_channels += src[i].channels();
|
||||
}
|
||||
|
||||
CV_Assert(single_channel_only);
|
||||
CV_Assert(total_channels <= 4);
|
||||
|
||||
if (total_channels == 1)
|
||||
src[0].copyTo(dst);
|
||||
else
|
||||
{
|
||||
dst.create(size, CV_MAKETYPE(depth, total_channels));
|
||||
|
||||
PtrStepSzb src_as_devmem[4];
|
||||
for(size_t i = 0; i < n; ++i)
|
||||
src_as_devmem[i] = src[i];
|
||||
|
||||
PtrStepSzb dst_as_devmem(dst);
|
||||
merge_caller(src_as_devmem, dst_as_devmem, total_channels, CV_ELEM_SIZE(depth), stream);
|
||||
}
|
||||
}
|
||||
|
||||
void split(const GpuMat& src, GpuMat* dst, const cudaStream_t& stream)
|
||||
{
|
||||
using namespace ::cv::gpu::cudev::split_merge;
|
||||
|
||||
CV_Assert(dst);
|
||||
|
||||
int depth = src.depth();
|
||||
int num_channels = src.channels();
|
||||
|
||||
if (depth == CV_64F)
|
||||
{
|
||||
if (!deviceSupports(NATIVE_DOUBLE))
|
||||
CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
|
||||
}
|
||||
|
||||
if (num_channels == 1)
|
||||
{
|
||||
src.copyTo(dst[0]);
|
||||
return;
|
||||
}
|
||||
|
||||
for (int i = 0; i < num_channels; ++i)
|
||||
dst[i].create(src.size(), depth);
|
||||
|
||||
CV_Assert(num_channels <= 4);
|
||||
|
||||
PtrStepSzb dst_as_devmem[4];
|
||||
for (int i = 0; i < num_channels; ++i)
|
||||
dst_as_devmem[i] = dst[i];
|
||||
|
||||
PtrStepSzb src_as_devmem(src);
|
||||
split_caller(src_as_devmem, dst_as_devmem, num_channels, src.elemSize1(), stream);
|
||||
}
|
||||
}
|
||||
|
||||
void cv::gpu::merge(const GpuMat* src, size_t n, GpuMat& dst, Stream& stream)
|
||||
{
|
||||
::merge(src, n, dst, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
|
||||
void cv::gpu::merge(const std::vector<GpuMat>& src, GpuMat& dst, Stream& stream)
|
||||
{
|
||||
::merge(&src[0], src.size(), dst, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
void cv::gpu::split(const GpuMat& src, GpuMat* dst, Stream& stream)
|
||||
{
|
||||
::split(src, dst, StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
void cv::gpu::split(const GpuMat& src, std::vector<GpuMat>& dst, Stream& stream)
|
||||
{
|
||||
dst.resize(src.channels());
|
||||
if(src.channels() > 0)
|
||||
::split(src, &dst[0], StreamAccessor::getStream(stream));
|
||||
}
|
||||
|
||||
#endif /* !defined (HAVE_CUDA) */
|
||||
Reference in New Issue
Block a user