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

moved GpuMat's operations implementation to core module

Browse Source
This commit is contained in:
Vladislav Vinogradov
2011-11-14 14:34:36 +00:00
parent 0f53f2993e
commit 2695039a79
34 changed files with 825 additions and 606 deletions
Download Patch File Download Diff File Expand all files Collapse all files

130
modules/core/CMakeLists.txt
View File

@@ -3,4 +3,132 @@ if(ZLIB_FOUND)
else()
include_directories("${CMAKE_CURRENT_SOURCE_DIR}/../../3rdparty/zlib")
endif()
define_opencv_module(core ${ZLIB_LIBRARY})
#define_opencv_module(core ${ZLIB_LIBRARY})
set(name "core")
project(opencv_${name})
include_directories("${CMAKE_CURRENT_SOURCE_DIR}/include"
"${CMAKE_CURRENT_SOURCE_DIR}/src"
"${CMAKE_CURRENT_BINARY_DIR}")
file(GLOB lib_srcs "src/*.cpp")
file(GLOB lib_int_hdrs "src/*.h*")
file(GLOB lib_hdrs "include/opencv2/${name}/*.h*")
file(GLOB lib_hdrs_detail "include/opencv2/${name}/detail/*.h*")
if(COMMAND get_module_external_sources)
get_module_external_sources(${name})
endif()
source_group("Src" FILES ${lib_srcs} ${lib_int_hdrs})
source_group("Include" FILES ${lib_hdrs})
source_group("Include\\detail" FILES ${lib_hdrs_detail})
list(APPEND lib_hdrs ${lib_hdrs_detail})
if (HAVE_CUDA)
file(GLOB lib_cuda "src/cuda/*.cu")
source_group("Cuda" FILES "${lib_cuda}")
include_directories(${CUDA_INCLUDE_DIRS})
include_directories("${OpenCV_SOURCE_DIR}/modules/gpu/src")
include_directories("${OpenCV_SOURCE_DIR}/modules/gpu/src/cuda")
set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -gencode arch=compute_10,code=sm_10
-gencode arch=compute_11,code=sm_11
-gencode arch=compute_12,code=sm_12
-gencode arch=compute_13,code=sm_13
-gencode arch=compute_20,code=sm_20
-gencode arch=compute_20,code=sm_21)
if (UNIX OR APPLE)
set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler;-fPIC;")
#set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}" "-fPIC")
endif()
#set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-keep")
#set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler;/EHsc-;")
if (APPLE)
set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler;-fno-finite-math-only;")
endif()
CUDA_COMPILE(cuda_objs ${lib_cuda})
#CUDA_BUILD_CLEAN_TARGET()
endif()
set(the_target "opencv_${name}")
add_library(${the_target} ${lib_srcs} ${lib_hdrs} ${lib_int_hdrs} ${lib_cuda} ${cuda_objs})
# For dynamic link numbering convenions
if(NOT ANDROID)
# Android SDK build scripts can include only .so files into final .apk
# As result we should not set version properties for Android
set_target_properties(${the_target} PROPERTIES
VERSION ${OPENCV_VERSION}
SOVERSION ${OPENCV_SOVERSION}
)
endif()
set_target_properties(${the_target} PROPERTIES OUTPUT_NAME "${the_target}${OPENCV_DLLVERSION}" )
if(ENABLE_SOLUTION_FOLDERS)
set_target_properties(${the_target} PROPERTIES FOLDER "modules")
endif()
if (BUILD_SHARED_LIBS)
if(MSVC)
set_target_properties(${the_target} PROPERTIES DEFINE_SYMBOL CVAPI_EXPORTS)
else()
add_definitions(-DCVAPI_EXPORTS)
endif()
endif()
# Additional target properties
set_target_properties(${the_target} PROPERTIES
DEBUG_POSTFIX "${OPENCV_DEBUG_POSTFIX}"
ARCHIVE_OUTPUT_DIRECTORY ${LIBRARY_OUTPUT_PATH}
RUNTIME_OUTPUT_DIRECTORY ${EXECUTABLE_OUTPUT_PATH}
INSTALL_NAME_DIR lib
)
# Add the required libraries for linking:
target_link_libraries(${the_target} ${OPENCV_LINKER_LIBS} ${IPP_LIBS} ${ZLIB_LIBRARY})
if (HAVE_CUDA)
target_link_libraries(${the_target} ${CUDA_LIBRARIES})
unset(CUDA_npp_LIBRARY CACHE)
find_cuda_helper_libs(npp)
target_link_libraries(${the_target} ${CUDA_npp_LIBRARY})
endif()
if(MSVC)
if(CMAKE_CROSSCOMPILING)
set_target_properties(${the_target} PROPERTIES
LINK_FLAGS "/NODEFAULTLIB:secchk"
)
endif()
set_target_properties(${the_target} PROPERTIES
LINK_FLAGS "/NODEFAULTLIB:libc /DEBUG"
)
endif()
# Dependencies of this target:
add_dependencies(${the_target} ${ZLIB_LIBRARY})
install(TARGETS ${the_target}
RUNTIME DESTINATION bin COMPONENT main
LIBRARY DESTINATION ${OPENCV_LIB_INSTALL_PATH} COMPONENT main
ARCHIVE DESTINATION ${OPENCV_LIB_INSTALL_PATH} COMPONENT main)
install(FILES ${lib_hdrs}
DESTINATION ${OPENCV_INCLUDE_PREFIX}/opencv2/${name}
COMPONENT main)
add_opencv_precompiled_headers(${the_target})
define_opencv_test(${name})
define_opencv_perf_test(${name})

22
modules/core/include/opencv2/core/gpumat.hpp
View File

@@ -212,27 +212,9 @@ namespace cv { namespace gpu
CV_EXPORTS void ensureSizeIsEnough(int rows, int cols, int type, GpuMat& m);
CV_EXPORTS void ensureSizeIsEnough(Size size, int type, GpuMat& m);
class CV_EXPORTS GpuFuncTable
{
public:
virtual ~GpuFuncTable() {}
//////////////////////////////// Error handling ////////////////////////
virtual void copy(const Mat& src, GpuMat& dst) const = 0;
virtual void copy(const GpuMat& src, Mat& dst) const = 0;
virtual void copy(const GpuMat& src, GpuMat& dst) const = 0;
virtual void copyWithMask(const GpuMat& src, GpuMat& dst, const GpuMat& mask) const = 0;
virtual void convert(const GpuMat& src, GpuMat& dst) const = 0;
virtual void convert(const GpuMat& src, GpuMat& dst, double alpha, double beta) const = 0;
virtual void setTo(GpuMat& m, Scalar s, const GpuMat& mask) const = 0;
virtual void mallocPitch(void** devPtr, size_t* step, size_t width, size_t height) const = 0;
virtual void free(void* devPtr) const = 0;
};
CV_EXPORTS void setGpuFuncTable(const GpuFuncTable* funcTbl);
CV_EXPORTS void error(const char *error_string, const char *file, const int line, const char *func);
////////////////////////////////////////////////////////////////////////

345
modules/core/src/cuda/matrix_operations.cu Normal file
View File

@@ -0,0 +1,345 @@
/*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 "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/transform.hpp"
#include "opencv2/gpu/device/functional.hpp"
namespace cv { namespace gpu { namespace device
{
template <typename T> struct shift_and_sizeof;
template <> struct shift_and_sizeof<signed char> { enum { shift = 0 }; };
template <> struct shift_and_sizeof<unsigned char> { enum { shift = 0 }; };
template <> struct shift_and_sizeof<short> { enum { shift = 1 }; };
template <> struct shift_and_sizeof<unsigned short> { enum { shift = 1 }; };
template <> struct shift_and_sizeof<int> { enum { shift = 2 }; };
template <> struct shift_and_sizeof<float> { enum { shift = 2 }; };
template <> struct shift_and_sizeof<double> { enum { shift = 3 }; };
///////////////////////////////////////////////////////////////////////////
////////////////////////////////// CopyTo /////////////////////////////////
///////////////////////////////////////////////////////////////////////////
template<typename T>
__global__ void copy_to_with_mask(const T* mat_src, T* mat_dst, const uchar* mask, int cols, int rows, size_t step_mat, size_t step_mask, int channels)
{
size_t x = blockIdx.x * blockDim.x + threadIdx.x;
size_t y = blockIdx.y * blockDim.y + threadIdx.y;
if ((x < cols * channels ) && (y < rows))
if (mask[y * step_mask + x / channels] != 0)
{
size_t idx = y * ( step_mat >> shift_and_sizeof<T>::shift ) + x;
mat_dst[idx] = mat_src[idx];
}
}
template<typename T>
void copy_to_with_mask_run(DevMem2Db mat_src, DevMem2Db mat_dst, DevMem2Db mask, int channels, cudaStream_t stream)
{
dim3 threadsPerBlock(16,16, 1);
dim3 numBlocks ( divUp(mat_src.cols * channels , threadsPerBlock.x) , divUp(mat_src.rows , threadsPerBlock.y), 1);
copy_to_with_mask<T><<<numBlocks,threadsPerBlock, 0, stream>>>
((T*)mat_src.data, (T*)mat_dst.data, (unsigned char*)mask.data, mat_src.cols, mat_src.rows, mat_src.step, mask.step, channels);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall ( cudaDeviceSynchronize() );
}
void copy_to_with_mask(DevMem2Db mat_src, DevMem2Db mat_dst, int depth, DevMem2Db mask, int channels, cudaStream_t stream)
{
typedef void (*CopyToFunc)(DevMem2Db mat_src, DevMem2Db mat_dst, DevMem2Db mask, int channels, cudaStream_t stream);
static CopyToFunc tab[8] =
{
copy_to_with_mask_run<unsigned char>,
copy_to_with_mask_run<signed char>,
copy_to_with_mask_run<unsigned short>,
copy_to_with_mask_run<short>,
copy_to_with_mask_run<int>,
copy_to_with_mask_run<float>,
copy_to_with_mask_run<double>,
0
};
CopyToFunc func = tab[depth];
if (func == 0)
cv::gpu::error("Unsupported copyTo operation", __FILE__, __LINE__);
func(mat_src, mat_dst, mask, channels, stream);
}
///////////////////////////////////////////////////////////////////////////
////////////////////////////////// SetTo //////////////////////////////////
///////////////////////////////////////////////////////////////////////////
__constant__ uchar scalar_8u[4];
__constant__ schar scalar_8s[4];
__constant__ ushort scalar_16u[4];
__constant__ short scalar_16s[4];
__constant__ int scalar_32s[4];
__constant__ float scalar_32f[4];
__constant__ double scalar_64f[4];
template <typename T> __device__ __forceinline__ T readScalar(int i);
template <> __device__ __forceinline__ uchar readScalar<uchar>(int i) {return scalar_8u[i];}
template <> __device__ __forceinline__ schar readScalar<schar>(int i) {return scalar_8s[i];}
template <> __device__ __forceinline__ ushort readScalar<ushort>(int i) {return scalar_16u[i];}
template <> __device__ __forceinline__ short readScalar<short>(int i) {return scalar_16s[i];}
template <> __device__ __forceinline__ int readScalar<int>(int i) {return scalar_32s[i];}
template <> __device__ __forceinline__ float readScalar<float>(int i) {return scalar_32f[i];}
template <> __device__ __forceinline__ double readScalar<double>(int i) {return scalar_64f[i];}
void writeScalar(const uchar* vals)
{
cudaSafeCall( cudaMemcpyToSymbol(scalar_8u, vals, sizeof(uchar) * 4) );
}
void writeScalar(const schar* vals)
{
cudaSafeCall( cudaMemcpyToSymbol(scalar_8s, vals, sizeof(schar) * 4) );
}
void writeScalar(const ushort* vals)
{
cudaSafeCall( cudaMemcpyToSymbol(scalar_16u, vals, sizeof(ushort) * 4) );
}
void writeScalar(const short* vals)
{
cudaSafeCall( cudaMemcpyToSymbol(scalar_16s, vals, sizeof(short) * 4) );
}
void writeScalar(const int* vals)
{
cudaSafeCall( cudaMemcpyToSymbol(scalar_32s, vals, sizeof(int) * 4) );
}
void writeScalar(const float* vals)
{
cudaSafeCall( cudaMemcpyToSymbol(scalar_32f, vals, sizeof(float) * 4) );
}
void writeScalar(const double* vals)
{
cudaSafeCall( cudaMemcpyToSymbol(scalar_64f, vals, sizeof(double) * 4) );
}
template<typename T>
__global__ void set_to_without_mask(T* mat, int cols, int rows, size_t step, int channels)
{
size_t x = blockIdx.x * blockDim.x + threadIdx.x;
size_t y = blockIdx.y * blockDim.y + threadIdx.y;
if ((x < cols * channels ) && (y < rows))
{
size_t idx = y * ( step >> shift_and_sizeof<T>::shift ) + x;
mat[idx] = readScalar<T>(x % channels);
}
}
template<typename T>
__global__ void set_to_with_mask(T* mat, const uchar* mask, int cols, int rows, size_t step, int channels, size_t step_mask)
{
size_t x = blockIdx.x * blockDim.x + threadIdx.x;
size_t y = blockIdx.y * blockDim.y + threadIdx.y;
if ((x < cols * channels ) && (y < rows))
if (mask[y * step_mask + x / channels] != 0)
{
size_t idx = y * ( step >> shift_and_sizeof<T>::shift ) + x;
mat[idx] = readScalar<T>(x % channels);
}
}
template <typename T>
void set_to_gpu(DevMem2Db mat, const T* scalar, DevMem2Db mask, int channels, cudaStream_t stream)
{
writeScalar(scalar);
dim3 threadsPerBlock(32, 8, 1);
dim3 numBlocks (mat.cols * channels / threadsPerBlock.x + 1, mat.rows / threadsPerBlock.y + 1, 1);
set_to_with_mask<T><<<numBlocks, threadsPerBlock, 0, stream>>>((T*)mat.data, (uchar*)mask.data, mat.cols, mat.rows, mat.step, channels, mask.step);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall ( cudaDeviceSynchronize() );
}
template void set_to_gpu<uchar >(DevMem2Db mat, const uchar* scalar, DevMem2Db mask, int channels, cudaStream_t stream);
template void set_to_gpu<schar >(DevMem2Db mat, const schar* scalar, DevMem2Db mask, int channels, cudaStream_t stream);
template void set_to_gpu<ushort>(DevMem2Db mat, const ushort* scalar, DevMem2Db mask, int channels, cudaStream_t stream);
template void set_to_gpu<short >(DevMem2Db mat, const short* scalar, DevMem2Db mask, int channels, cudaStream_t stream);
template void set_to_gpu<int >(DevMem2Db mat, const int* scalar, DevMem2Db mask, int channels, cudaStream_t stream);
template void set_to_gpu<float >(DevMem2Db mat, const float* scalar, DevMem2Db mask, int channels, cudaStream_t stream);
template void set_to_gpu<double>(DevMem2Db mat, const double* scalar, DevMem2Db mask, int channels, cudaStream_t stream);
template <typename T>
void set_to_gpu(DevMem2Db mat, const T* scalar, int channels, cudaStream_t stream)
{
writeScalar(scalar);
dim3 threadsPerBlock(32, 8, 1);
dim3 numBlocks (mat.cols * channels / threadsPerBlock.x + 1, mat.rows / threadsPerBlock.y + 1, 1);
set_to_without_mask<T><<<numBlocks, threadsPerBlock, 0, stream>>>((T*)mat.data, mat.cols, mat.rows, mat.step, channels);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall ( cudaDeviceSynchronize() );
}
template void set_to_gpu<uchar >(DevMem2Db mat, const uchar* scalar, int channels, cudaStream_t stream);
template void set_to_gpu<schar >(DevMem2Db mat, const schar* scalar, int channels, cudaStream_t stream);
template void set_to_gpu<ushort>(DevMem2Db mat, const ushort* scalar, int channels, cudaStream_t stream);
template void set_to_gpu<short >(DevMem2Db mat, const short* scalar, int channels, cudaStream_t stream);
template void set_to_gpu<int >(DevMem2Db mat, const int* scalar, int channels, cudaStream_t stream);
template void set_to_gpu<float >(DevMem2Db mat, const float* scalar, int channels, cudaStream_t stream);
template void set_to_gpu<double>(DevMem2Db mat, const double* scalar, int channels, cudaStream_t stream);
///////////////////////////////////////////////////////////////////////////
//////////////////////////////// ConvertTo ////////////////////////////////
///////////////////////////////////////////////////////////////////////////
template <typename T, typename D> struct Convertor : unary_function<T, D>
{
Convertor(double alpha_, double beta_) : alpha(alpha_), beta(beta_) {}
__device__ __forceinline__ D operator()(const T& src) const
{
return saturate_cast<D>(alpha * src + beta);
}
const double alpha, beta;
};
namespace detail
{
template <size_t src_size, size_t dst_size, typename F> struct ConvertTraitsDispatcher : DefaultTransformFunctorTraits<F>
{
};
template <typename F> struct ConvertTraitsDispatcher<1, 1, F> : DefaultTransformFunctorTraits<F>
{
enum { smart_shift = 8 };
};
template <typename F> struct ConvertTraitsDispatcher<1, 2, F> : DefaultTransformFunctorTraits<F>
{
enum { smart_shift = 4 };
};
template <typename F> struct ConvertTraitsDispatcher<1, 4, F> : DefaultTransformFunctorTraits<F>
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
template <typename F> struct ConvertTraitsDispatcher<2, 2, F> : DefaultTransformFunctorTraits<F>
{
enum { smart_shift = 4 };
};
template <typename F> struct ConvertTraitsDispatcher<2, 4, F> : DefaultTransformFunctorTraits<F>
{
enum { smart_shift = 2 };
};
template <typename F> struct ConvertTraitsDispatcher<4, 2, F> : DefaultTransformFunctorTraits<F>
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
template <typename F> struct ConvertTraitsDispatcher<4, 4, F> : DefaultTransformFunctorTraits<F>
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 2 };
};
template <typename F> struct ConvertTraits : ConvertTraitsDispatcher<sizeof(typename F::argument_type), sizeof(typename F::result_type), F>
{
};
}
template <typename T, typename D> struct TransformFunctorTraits< Convertor<T, D> > : detail::ConvertTraits< Convertor<T, D> >
{
};
template<typename T, typename D>
void cvt_(DevMem2Db src, DevMem2Db dst, double alpha, double beta, cudaStream_t stream)
{
cudaSafeCall( cudaSetDoubleForDevice(&alpha) );
cudaSafeCall( cudaSetDoubleForDevice(&beta) );
Convertor<T, D> op(alpha, beta);
::cv::gpu::device::transform((DevMem2D_<T>)src, (DevMem2D_<D>)dst, op, stream);
}
void convert_gpu(DevMem2Db src, int sdepth, DevMem2Db dst, int ddepth, double alpha, double beta, cudaStream_t stream)
{
typedef void (*caller_t)(DevMem2Db src, DevMem2Db dst, double alpha, double beta, cudaStream_t stream);
static const caller_t tab[8][8] =
{
{cvt_<uchar, uchar>, cvt_<uchar, schar>, cvt_<uchar, ushort>, cvt_<uchar, short>,
cvt_<uchar, int>, cvt_<uchar, float>, cvt_<uchar, double>, 0},
{cvt_<schar, uchar>, cvt_<schar, schar>, cvt_<schar, ushort>, cvt_<schar, short>,
cvt_<schar, int>, cvt_<schar, float>, cvt_<schar, double>, 0},
{cvt_<ushort, uchar>, cvt_<ushort, schar>, cvt_<ushort, ushort>, cvt_<ushort, short>,
cvt_<ushort, int>, cvt_<ushort, float>, cvt_<ushort, double>, 0},
{cvt_<short, uchar>, cvt_<short, schar>, cvt_<short, ushort>, cvt_<short, short>,
cvt_<short, int>, cvt_<short, float>, cvt_<short, double>, 0},
{cvt_<int, uchar>, cvt_<int, schar>, cvt_<int, ushort>,
cvt_<int, short>, cvt_<int, int>, cvt_<int, float>, cvt_<int, double>, 0},
{cvt_<float, uchar>, cvt_<float, schar>, cvt_<float, ushort>,
cvt_<float, short>, cvt_<float, int>, cvt_<float, float>, cvt_<float, double>, 0},
{cvt_<double, uchar>, cvt_<double, schar>, cvt_<double, ushort>,
cvt_<double, short>, cvt_<double, int>, cvt_<double, float>, cvt_<double, double>, 0},
{0,0,0,0,0,0,0,0}
};
caller_t func = tab[sdepth][ddepth];
if (!func)
cv::gpu::error("Unsupported convert operation", __FILE__, __LINE__);
func(src, dst, alpha, beta, stream);
}
}}} // namespace cv { namespace gpu { namespace device

499
modules/core/src/gpumat.cpp
View File

@@ -43,6 +43,14 @@
#include "precomp.hpp"
#include "opencv2/core/gpumat.hpp"
#include <iostream>
#include <sstream>
#ifdef HAVE_CUDA
#include <cuda_runtime.h>
#include <npp.h>
#endif
using namespace std;
using namespace cv;
using namespace cv::gpu;
@@ -283,6 +291,31 @@ cv::Mat::Mat(const GpuMat& m) : flags(0), dims(0), rows(0), cols(0), data(0), re
m.download(*this);
}
namespace
{
class CV_EXPORTS GpuFuncTable
{
public:
virtual ~GpuFuncTable() {}
virtual void copy(const Mat& src, GpuMat& dst) const = 0;
virtual void copy(const GpuMat& src, Mat& dst) const = 0;
virtual void copy(const GpuMat& src, GpuMat& dst) const = 0;
virtual void copyWithMask(const GpuMat& src, GpuMat& dst, const GpuMat& mask) const = 0;
virtual void convert(const GpuMat& src, GpuMat& dst) const = 0;
virtual void convert(const GpuMat& src, GpuMat& dst, double alpha, double beta) const = 0;
virtual void setTo(GpuMat& m, Scalar s, const GpuMat& mask) const = 0;
virtual void mallocPitch(void** devPtr, size_t* step, size_t width, size_t height) const = 0;
virtual void free(void* devPtr) const = 0;
};
}
#ifndef HAVE_CUDA
namespace
{
void throw_nogpu()
@@ -308,20 +341,460 @@ namespace
void free(void*) const {}
};
const GpuFuncTable* g_funcTbl = 0;
const GpuFuncTable* gpuFuncTable()
{
static EmptyFuncTable empty;
return g_funcTbl ? g_funcTbl : &empty;
return &empty;
}
}
void cv::gpu::setGpuFuncTable(const GpuFuncTable* funcTbl)
#else // HAVE_CUDA
namespace cv { namespace gpu { namespace device
{
g_funcTbl = funcTbl;
void copy_to_with_mask(DevMem2Db src, DevMem2Db dst, int depth, DevMem2Db mask, int channels, cudaStream_t stream);
template <typename T>
void set_to_gpu(DevMem2Db mat, const T* scalar, int channels, cudaStream_t stream);
template <typename T>
void set_to_gpu(DevMem2Db mat, const T* scalar, DevMem2Db mask, int channels, cudaStream_t stream);
void convert_gpu(DevMem2Db src, int sdepth, DevMem2Db dst, int ddepth, double alpha, double beta, cudaStream_t stream);
}}}
namespace
{
#if defined(__GNUC__)
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__, __func__)
#define nppSafeCall(expr) ___nppSafeCall(expr, __FILE__, __LINE__, __func__)
#else /* defined(__CUDACC__) || defined(__MSVC__) */
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__)
#define nppSafeCall(expr) ___nppSafeCall(expr, __FILE__, __LINE__)
#endif
inline void ___cudaSafeCall(cudaError_t err, const char *file, const int line, const char *func = "")
{
if (cudaSuccess != err)
cv::gpu::error(cudaGetErrorString(err), file, line, func);
}
inline void ___nppSafeCall(int err, const char *file, const int line, const char *func = "")
{
if (err < 0)
{
std::ostringstream msg;
msg << "NPP API Call Error: " << err;
cv::gpu::error(msg.str().c_str(), file, line, func);
}
}
}
namespace
{
template <typename T> void kernelSetCaller(GpuMat& src, Scalar s, cudaStream_t stream)
{
Scalar_<T> sf = s;
::cv::gpu::device::set_to_gpu(src, sf.val, src.channels(), stream);
}
template <typename T> void kernelSetCaller(GpuMat& src, Scalar s, const GpuMat& mask, cudaStream_t stream)
{
Scalar_<T> sf = s;
::cv::gpu::device::set_to_gpu(src, sf.val, mask, src.channels(), stream);
}
}
namespace cv { namespace gpu
{
CV_EXPORTS void copyWithMask(const GpuMat& src, GpuMat& dst, const GpuMat& mask, cudaStream_t stream = 0)
{
::cv::gpu::device::copy_to_with_mask(src, dst, src.depth(), mask, src.channels(), stream);
}
CV_EXPORTS void convertTo(const GpuMat& src, GpuMat& dst)
{
::cv::gpu::device::convert_gpu(src.reshape(1), src.depth(), dst.reshape(1), dst.depth(), 1.0, 0.0, 0);
}
CV_EXPORTS void convertTo(const GpuMat& src, GpuMat& dst, double alpha, double beta, cudaStream_t stream = 0)
{
::cv::gpu::device::convert_gpu(src.reshape(1), src.depth(), dst.reshape(1), dst.depth(), alpha, beta, stream);
}
CV_EXPORTS void setTo(GpuMat& src, Scalar s, cudaStream_t stream)
{
typedef void (*caller_t)(GpuMat& src, Scalar s, cudaStream_t stream);
static const caller_t callers[] =
{
kernelSetCaller<uchar>, kernelSetCaller<schar>, kernelSetCaller<ushort>, kernelSetCaller<short>, kernelSetCaller<int>,
kernelSetCaller<float>, kernelSetCaller<double>
};
callers[src.depth()](src, s, stream);
}
CV_EXPORTS void setTo(GpuMat& src, Scalar s, const GpuMat& mask, cudaStream_t stream)
{
typedef void (*caller_t)(GpuMat& src, Scalar s, const GpuMat& mask, cudaStream_t stream);
static const caller_t callers[] =
{
kernelSetCaller<uchar>, kernelSetCaller<schar>, kernelSetCaller<ushort>, kernelSetCaller<short>, kernelSetCaller<int>,
kernelSetCaller<float>, kernelSetCaller<double>
};
callers[src.depth()](src, s, mask, stream);
}
CV_EXPORTS void setTo(GpuMat& src, Scalar s)
{
setTo(src, s, 0);
}
CV_EXPORTS void setTo(GpuMat& src, Scalar s, const GpuMat& mask)
{
setTo(src, s, mask, 0);
}
}}
namespace
{
//////////////////////////////////////////////////////////////////////////
// Convert
template<int n> struct NPPTypeTraits;
template<> struct NPPTypeTraits<CV_8U> { typedef Npp8u npp_type; };
template<> struct NPPTypeTraits<CV_16U> { typedef Npp16u npp_type; };
template<> struct NPPTypeTraits<CV_16S> { typedef Npp16s npp_type; };
template<> struct NPPTypeTraits<CV_32S> { typedef Npp32s npp_type; };
template<> struct NPPTypeTraits<CV_32F> { typedef Npp32f npp_type; };
template<int SDEPTH, int DDEPTH> struct NppConvertFunc
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef typename NPPTypeTraits<DDEPTH>::npp_type dst_t;
typedef NppStatus (*func_ptr)(const src_t* pSrc, int nSrcStep, dst_t* pDst, int nDstStep, NppiSize oSizeROI);
};
template<int DDEPTH> struct NppConvertFunc<CV_32F, DDEPTH>
{
typedef typename NPPTypeTraits<DDEPTH>::npp_type dst_t;
typedef NppStatus (*func_ptr)(const Npp32f* pSrc, int nSrcStep, dst_t* pDst, int nDstStep, NppiSize oSizeROI, NppRoundMode eRoundMode);
};
template<int SDEPTH, int DDEPTH, typename NppConvertFunc<SDEPTH, DDEPTH>::func_ptr func> struct NppCvt
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef typename NPPTypeTraits<DDEPTH>::npp_type dst_t;
static void cvt(const GpuMat& src, GpuMat& dst)
{
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
nppSafeCall( func(src.ptr<src_t>(), static_cast<int>(src.step), dst.ptr<dst_t>(), static_cast<int>(dst.step), sz) );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template<int DDEPTH, typename NppConvertFunc<CV_32F, DDEPTH>::func_ptr func> struct NppCvt<CV_32F, DDEPTH, func>
{
typedef typename NPPTypeTraits<DDEPTH>::npp_type dst_t;
static void cvt(const GpuMat& src, GpuMat& dst)
{
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
nppSafeCall( func(src.ptr<Npp32f>(), static_cast<int>(src.step), dst.ptr<dst_t>(), static_cast<int>(dst.step), sz, NPP_RND_NEAR) );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
//////////////////////////////////////////////////////////////////////////
// Set
template<int SDEPTH, int SCN> struct NppSetFunc
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef NppStatus (*func_ptr)(const src_t values[], src_t* pSrc, int nSrcStep, NppiSize oSizeROI);
};
template<int SDEPTH> struct NppSetFunc<SDEPTH, 1>
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef NppStatus (*func_ptr)(src_t val, src_t* pSrc, int nSrcStep, NppiSize oSizeROI);
};
template<int SDEPTH, int SCN, typename NppSetFunc<SDEPTH, SCN>::func_ptr func> struct NppSet
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
static void set(GpuMat& src, Scalar s)
{
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
Scalar_<src_t> nppS = s;
nppSafeCall( func(nppS.val, src.ptr<src_t>(), static_cast<int>(src.step), sz) );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template<int SDEPTH, typename NppSetFunc<SDEPTH, 1>::func_ptr func> struct NppSet<SDEPTH, 1, func>
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
static void set(GpuMat& src, Scalar s)
{
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
Scalar_<src_t> nppS = s;
nppSafeCall( func(nppS[0], src.ptr<src_t>(), static_cast<int>(src.step), sz) );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template<int SDEPTH, int SCN> struct NppSetMaskFunc
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef NppStatus (*func_ptr)(const src_t values[], src_t* pSrc, int nSrcStep, NppiSize oSizeROI, const Npp8u* pMask, int nMaskStep);
};
template<int SDEPTH> struct NppSetMaskFunc<SDEPTH, 1>
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef NppStatus (*func_ptr)(src_t val, src_t* pSrc, int nSrcStep, NppiSize oSizeROI, const Npp8u* pMask, int nMaskStep);
};
template<int SDEPTH, int SCN, typename NppSetMaskFunc<SDEPTH, SCN>::func_ptr func> struct NppSetMask
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
static void set(GpuMat& src, Scalar s, const GpuMat& mask)
{
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
Scalar_<src_t> nppS = s;
nppSafeCall( func(nppS.val, src.ptr<src_t>(), static_cast<int>(src.step), sz, mask.ptr<Npp8u>(), static_cast<int>(mask.step)) );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template<int SDEPTH, typename NppSetMaskFunc<SDEPTH, 1>::func_ptr func> struct NppSetMask<SDEPTH, 1, func>
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
static void set(GpuMat& src, Scalar s, const GpuMat& mask)
{
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
Scalar_<src_t> nppS = s;
nppSafeCall( func(nppS[0], src.ptr<src_t>(), static_cast<int>(src.step), sz, mask.ptr<Npp8u>(), static_cast<int>(mask.step)) );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
class CudaFuncTable : public GpuFuncTable
{
public:
void copy(const Mat& src, GpuMat& dst) const
{
cudaSafeCall( cudaMemcpy2D(dst.data, dst.step, src.data, src.step, src.cols * src.elemSize(), src.rows, cudaMemcpyHostToDevice) );
}
void copy(const GpuMat& src, Mat& dst) const
{
cudaSafeCall( cudaMemcpy2D(dst.data, dst.step, src.data, src.step, src.cols * src.elemSize(), src.rows, cudaMemcpyDeviceToHost) );
}
void copy(const GpuMat& src, GpuMat& dst) const
{
cudaSafeCall( cudaMemcpy2D(dst.data, dst.step, src.data, src.step, src.cols * src.elemSize(), src.rows, cudaMemcpyDeviceToDevice) );
}
void copyWithMask(const GpuMat& src, GpuMat& dst, const GpuMat& mask) const
{
::cv::gpu::copyWithMask(src, dst, mask);
}
void convert(const GpuMat& src, GpuMat& dst) const
{
typedef void (*caller_t)(const GpuMat& src, GpuMat& dst);
static const caller_t callers[7][7][7] =
{
{
/* 8U -> 8U */ {0, 0, 0, 0},
/* 8U -> 8S */ {::cv::gpu::convertTo, ::cv::gpu::convertTo, ::cv::gpu::convertTo, ::cv::gpu::convertTo},
/* 8U -> 16U */ {NppCvt<CV_8U, CV_16U, nppiConvert_8u16u_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,NppCvt<CV_8U, CV_16U, nppiConvert_8u16u_C4R>::cvt},
/* 8U -> 16S */ {NppCvt<CV_8U, CV_16S, nppiConvert_8u16s_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,NppCvt<CV_8U, CV_16S, nppiConvert_8u16s_C4R>::cvt},
/* 8U -> 32S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 8U -> 32F */ {NppCvt<CV_8U, CV_32F, nppiConvert_8u32f_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 8U -> 64F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo}
},
{
/* 8S -> 8U */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 8S -> 8S */ {0,0,0,0},
/* 8S -> 16U */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 8S -> 16S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 8S -> 32S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 8S -> 32F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 8S -> 64F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo}
},
{
/* 16U -> 8U */ {NppCvt<CV_16U, CV_8U, nppiConvert_16u8u_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,NppCvt<CV_16U, CV_8U, nppiConvert_16u8u_C4R>::cvt},
/* 16U -> 8S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 16U -> 16U */ {0,0,0,0},
/* 16U -> 16S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 16U -> 32S */ {NppCvt<CV_16U, CV_32S, nppiConvert_16u32s_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 16U -> 32F */ {NppCvt<CV_16U, CV_32F, nppiConvert_16u32f_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 16U -> 64F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo}
},
{
/* 16S -> 8U */ {NppCvt<CV_16S, CV_8U, nppiConvert_16s8u_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,NppCvt<CV_16S, CV_8U, nppiConvert_16s8u_C4R>::cvt},
/* 16S -> 8S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 16S -> 16U */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 16S -> 16S */ {0,0,0,0},
/* 16S -> 32S */ {NppCvt<CV_16S, CV_32S, nppiConvert_16s32s_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 16S -> 32F */ {NppCvt<CV_16S, CV_32F, nppiConvert_16s32f_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 16S -> 64F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo}
},
{
/* 32S -> 8U */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32S -> 8S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32S -> 16U */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32S -> 16S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32S -> 32S */ {0,0,0,0},
/* 32S -> 32F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32S -> 64F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo}
},
{
/* 32F -> 8U */ {NppCvt<CV_32F, CV_8U, nppiConvert_32f8u_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32F -> 8S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32F -> 16U */ {NppCvt<CV_32F, CV_16U, nppiConvert_32f16u_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32F -> 16S */ {NppCvt<CV_32F, CV_16S, nppiConvert_32f16s_C1R>::cvt,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32F -> 32S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 32F -> 32F */ {0,0,0,0},
/* 32F -> 64F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo}
},
{
/* 64F -> 8U */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 64F -> 8S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 64F -> 16U */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 64F -> 16S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 64F -> 32S */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 64F -> 32F */ {::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo,::cv::gpu::convertTo},
/* 64F -> 64F */ {0,0,0,0}
}
};
caller_t func = callers[src.depth()][dst.depth()][src.channels() - 1];
CV_DbgAssert(func != 0);
func(src, dst);
}
void convert(const GpuMat& src, GpuMat& dst, double alpha, double beta) const
{
::cv::gpu::convertTo(src, dst, alpha, beta);
}
void setTo(GpuMat& m, Scalar s, const GpuMat& mask) const
{
NppiSize sz;
sz.width = m.cols;
sz.height = m.rows;
if (mask.empty())
{
if (s[0] == 0.0 && s[1] == 0.0 && s[2] == 0.0 && s[3] == 0.0)
{
cudaSafeCall( cudaMemset2D(m.data, m.step, 0, m.cols * m.elemSize(), m.rows) );
return;
}
if (m.depth() == CV_8U)
{
int cn = m.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( cudaMemset2D(m.data, m.step, val, m.cols * m.elemSize(), m.rows) );
return;
}
}
typedef void (*caller_t)(GpuMat& src, Scalar s);
static const caller_t callers[7][4] =
{
{NppSet<CV_8U, 1, nppiSet_8u_C1R>::set, ::cv::gpu::setTo, ::cv::gpu::setTo, NppSet<CV_8U, 4, nppiSet_8u_C4R>::set},
{::cv::gpu::setTo, ::cv::gpu::setTo, ::cv::gpu::setTo, ::cv::gpu::setTo},
{NppSet<CV_16U, 1, nppiSet_16u_C1R>::set, NppSet<CV_16U, 2, nppiSet_16u_C2R>::set, ::cv::gpu::setTo, NppSet<CV_16U, 4, nppiSet_16u_C4R>::set},
{NppSet<CV_16S, 1, nppiSet_16s_C1R>::set, NppSet<CV_16S, 2, nppiSet_16s_C2R>::set, ::cv::gpu::setTo, NppSet<CV_16S, 4, nppiSet_16s_C4R>::set},
{NppSet<CV_32S, 1, nppiSet_32s_C1R>::set, ::cv::gpu::setTo, ::cv::gpu::setTo, NppSet<CV_32S, 4, nppiSet_32s_C4R>::set},
{NppSet<CV_32F, 1, nppiSet_32f_C1R>::set, ::cv::gpu::setTo, ::cv::gpu::setTo, NppSet<CV_32F, 4, nppiSet_32f_C4R>::set},
{::cv::gpu::setTo, ::cv::gpu::setTo, ::cv::gpu::setTo, ::cv::gpu::setTo}
};
callers[m.depth()][m.channels() - 1](m, s);
}
else
{
typedef void (*caller_t)(GpuMat& src, Scalar s, const GpuMat& mask);
static const caller_t callers[7][4] =
{
{NppSetMask<CV_8U, 1, nppiSet_8u_C1MR>::set, ::cv::gpu::setTo, ::cv::gpu::setTo, NppSetMask<CV_8U, 4, nppiSet_8u_C4MR>::set},
{::cv::gpu::setTo, ::cv::gpu::setTo, ::cv::gpu::setTo, ::cv::gpu::setTo},
{NppSetMask<CV_16U, 1, nppiSet_16u_C1MR>::set, ::cv::gpu::setTo, ::cv::gpu::setTo, NppSetMask<CV_16U, 4, nppiSet_16u_C4MR>::set},
{NppSetMask<CV_16S, 1, nppiSet_16s_C1MR>::set, ::cv::gpu::setTo, ::cv::gpu::setTo, NppSetMask<CV_16S, 4, nppiSet_16s_C4MR>::set},
{NppSetMask<CV_32S, 1, nppiSet_32s_C1MR>::set, ::cv::gpu::setTo, ::cv::gpu::setTo, NppSetMask<CV_32S, 4, nppiSet_32s_C4MR>::set},
{NppSetMask<CV_32F, 1, nppiSet_32f_C1MR>::set, ::cv::gpu::setTo, ::cv::gpu::setTo, NppSetMask<CV_32F, 4, nppiSet_32f_C4MR>::set},
{::cv::gpu::setTo, ::cv::gpu::setTo, ::cv::gpu::setTo, ::cv::gpu::setTo}
};
callers[m.depth()][m.channels() - 1](m, s, mask);
}
}
void mallocPitch(void** devPtr, size_t* step, size_t width, size_t height) const
{
cudaSafeCall( cudaMallocPitch(devPtr, step, width, height) );
}
void free(void* devPtr) const
{
cudaFree(devPtr);
}
};
const GpuFuncTable* gpuFuncTable()
{
static CudaFuncTable funcTable;
return &funcTable;
}
}
#endif // HAVE_CUDA
void cv::gpu::GpuMat::upload(const Mat& m)
{
CV_DbgAssert(!m.empty());
@@ -458,3 +931,19 @@ void cv::gpu::GpuMat::release()
step = rows = cols = 0;
refcount = 0;
}
void cv::gpu::error(const char *error_string, const char *file, const int line, const char *func)
{
int code = CV_GpuApiCallError;
if (uncaught_exception())
{
const char* errorStr = cvErrorStr(code);
const char* function = func ? func : "unknown function";
cerr << "OpenCV Error: " << errorStr << "(" << error_string << ") in " << function << ", file " << file << ", line " << line;
cerr.flush();
}
else
cv::error( cv::Exception(code, error_string, func, file, line) );
}