Skip to content

O
opencv
Commit 07825bad authored by Anatoly Baksheev's avatar Anatoly Baksheev
Browse files
Options

compilation with no cuda re factored

parent 20e2dc84
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 550 additions and 313 deletions
modules/gpu/CMakeLists.txt
View file @ 07825bad
set(name "gpu")
set(DEPS "opencv_core")
set(name "gpu")
set(DEPS "opencv_core")
set(the_target "opencv_${name}") set(the_target "opencv_${name}")
...@@ -9,25 +10,25 @@ project(${the_target}) ...@@ -9,25 +10,25 @@ project(${the_target})
add_definitions(-DCVAPI_EXPORTS) add_definitions(-DCVAPI_EXPORTS)
include_directories("${CMAKE_CURRENT_SOURCE_DIR}/include" include_directories("${CMAKE_CURRENT_SOURCE_DIR}/include"
"${CMAKE_CURRENT_SOURCE_DIR}/cuda" "${CMAKE_CURRENT_SOURCE_DIR}/src/cuda"
"${CMAKE_CURRENT_SOURCE_DIR}/src" "${CMAKE_CURRENT_SOURCE_DIR}/src"
"${CMAKE_CURRENT_BINARY_DIR}") "${CMAKE_CURRENT_BINARY_DIR}")
foreach(d ${DEPS}) foreach(d ${DEPS})
if(${d} MATCHES "opencv_") if(${d} MATCHES "opencv_")
string(REPLACE "opencv_" "${CMAKE_CURRENT_SOURCE_DIR}/../" d_dir ${d}) string(REPLACE "opencv_" "${CMAKE_CURRENT_SOURCE_DIR}/../" d_dir ${d})
include_directories("${d_dir}/include") include_directories("${d_dir}/include")
endif() endif()
endforeach() endforeach()
file(GLOB lib_srcs "src/*.cpp") file(GLOB lib_srcs "src/*.cpp")
file(GLOB lib_int_hdrs "src/*.h*") file(GLOB lib_int_hdrs "src/*.h*")
file(GLOB lib_cuda "cuda/*.cu") file(GLOB lib_cuda "src/cuda/*.cu")
file(GLOB lib_cuda_hdrs "cuda/*.h*") file(GLOB lib_cuda_hdrs "src/cuda/*.h*")
source_group("Src" FILES ${lib_srcs} ${lib_int_hdrs}) source_group("Src" FILES ${lib_srcs} ${lib_int_hdrs})
source_group("Cuda" FILES ${lib_cuda} ${lib_cuda_hdrs}) source_group("Cuda" FILES ${lib_cuda} ${lib_cuda_hdrs})
file(GLOB lib_hdrs "include/opencv2/${name}/*.h*") file(GLOB lib_hdrs "include/opencv2/${name}/*.h*")
source_group("Include" FILES ${lib_hdrs}) source_group("Include" FILES ${lib_hdrs})
if (HAVE_CUDA) if (HAVE_CUDA)
...@@ -35,13 +36,13 @@ if (HAVE_CUDA) ...@@ -35,13 +36,13 @@ if (HAVE_CUDA)
link_directories(${CUDA_LIBRARIES}) link_directories(${CUDA_LIBRARIES})
if (UNIX OR APPLE) if (UNIX OR APPLE)
set (CUDA_NVCC_FLAGS "-Xcompiler;-fPIC") set (CUDA_NVCC_FLAGS "-Xcompiler;-fPIC;")
endif() endif()
CUDA_COMPILE(cuda_objs ${lib_cuda}) CUDA_COMPILE(cuda_objs ${lib_cuda})
#CUDA_BUILD_CLEAN_TARGET() #CUDA_BUILD_CLEAN_TARGET()
endif() endif()
add_library(${the_target} ${lib_srcs} ${lib_hdrs} ${lib_int_hdrs} ${lib_cuda} ${lib_cuda_hdrs} ${cuda_objs}) add_library(${the_target} ${lib_srcs} ${lib_hdrs} ${lib_int_hdrs} ${lib_cuda} ${lib_cuda_hdrs} ${cuda_objs})
...@@ -50,7 +51,7 @@ if(PCHSupport_FOUND) ...@@ -50,7 +51,7 @@ if(PCHSupport_FOUND)
if(${CMAKE_GENERATOR} MATCHES "Visual*" OR ${CMAKE_GENERATOR} MATCHES "Xcode*") if(${CMAKE_GENERATOR} MATCHES "Visual*" OR ${CMAKE_GENERATOR} MATCHES "Xcode*")
if(${CMAKE_GENERATOR} MATCHES "Visual*") if(${CMAKE_GENERATOR} MATCHES "Visual*")
set(${the_target}_pch "src/precomp.cpp") set(${the_target}_pch "src/precomp.cpp")
endif() endif()
add_native_precompiled_header(${the_target} ${pch_header}) add_native_precompiled_header(${the_target} ${pch_header})
elseif(CMAKE_COMPILER_IS_GNUCXX AND ${CMAKE_GENERATOR} MATCHES ".*Makefiles") elseif(CMAKE_COMPILER_IS_GNUCXX AND ${CMAKE_GENERATOR} MATCHES ".*Makefiles")
add_precompiled_header(${the_target} ${pch_header}) add_precompiled_header(${the_target} ${pch_header})
......
modules/gpu/cuda/mat_operators.cu deleted 100644 → 0
View file @ 20e2dc84
/*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 <stddef.h>
#include "cuda_shared.hpp"
#include "cuda_runtime.h"
__constant__ float scalar_d[4];
namespace mat_operators
{
template <typename T, int channels, int count = channels>
struct unroll
{
__device__ static void unroll_set(T * mat, size_t i)
{
mat[i] = static_cast<T>(scalar_d[i % channels]);
unroll<T, channels, count - 1>::unroll_set(mat, i+1);
}
__device__ static void unroll_set_with_mask(T * mat, float mask, size_t i)
{
mat[i] = mask * static_cast<T>(scalar_d[i % channels]);
unroll<T, channels, count - 1>::unroll_set_with_mask(mat, mask, i+1);
}
};
template <typename T, int channels>
struct unroll<T,channels,0>
{
__device__ static void unroll_set(T * , size_t){}
__device__ static void unroll_set_with_mask(T * , float, size_t){}
};
template <typename T, int channels>
__global__ void kernel_set_to_without_mask(T * mat)
{
size_t i = (blockIdx.x * blockDim.x + threadIdx.x) * sizeof(T);
unroll<T, channels>::unroll_set(mat, i);
}
template <typename T, int channels>
__global__ void kernel_set_to_with_mask(T * mat, const float * mask)
{
size_t i = (blockIdx.x * blockDim.x + threadIdx.x) * sizeof(T);
unroll<T, channels>::unroll_set_with_mask(mat, i, mask[i]);
}
}
extern "C" void cv::gpu::impl::set_to_with_mask(const DevMem2D& mat, const double * scalar, const DevMem2D& mask, int depth, int channels)
{
scalar_d[0] = scalar[0];
scalar_d[1] = scalar[1];
scalar_d[2] = scalar[2];
scalar_d[3] = scalar[3];
int numBlocks = mat.rows * mat.step / 256;
dim3 threadsPerBlock(256);
if (channels == 1)
{
if (depth == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 1><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 1><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 1><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr);
}
if (channels == 2)
{
if (depth == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 2><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 2><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 2><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr);
}
if (channels == 3)
{
if (depth == 1) ::mat_operators::kernel_set_to_with_mask<unsigned char, 3><<<numBlocks,threadsPerBlock>>>(mat.ptr, (float *)mask.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_with_mask<unsigned short, 3><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr, (float *)mask.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_with_mask<unsigned int, 3><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr, (float *)mask.ptr);
}
}
extern "C" void cv::gpu::impl::set_to_without_mask(const DevMem2D& mat, const double * scalar, int depth, int channels)
{
scalar_d[0] = scalar[0];
scalar_d[1] = scalar[1];
scalar_d[2] = scalar[2];
scalar_d[3] = scalar[3];
int numBlocks = mat.rows * mat.step / 256;
dim3 threadsPerBlock(256);
if (channels == 1)
{
if (depth == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 1><<<numBlocks,threadsPerBlock>>>(mat.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 1><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 1><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr);
}
if (channels == 2)
{
if (depth == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 2><<<numBlocks,threadsPerBlock>>>(mat.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 2><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 2><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr);
}
if (channels == 3)
{
if (depth == 1) ::mat_operators::kernel_set_to_without_mask<unsigned char, 3><<<numBlocks,threadsPerBlock>>>(mat.ptr);
if (depth == 2) ::mat_operators::kernel_set_to_without_mask<unsigned short, 3><<<numBlocks,threadsPerBlock>>>((unsigned short *)mat.ptr);
if (depth == 4) ::mat_operators::kernel_set_to_without_mask<unsigned int, 3><<<numBlocks,threadsPerBlock>>>((unsigned int *)mat.ptr);
}
}
modules/gpu/include/opencv2/gpu/devmem2d.hpp
View file @ 07825bad
...@@ -48,12 +48,13 @@ namespace cv ...@@ -48,12 +48,13 @@ namespace cv
namespace gpu namespace gpu
{ {
// Simple lightweight structure that encapsulates image ptr on device, its pitch and its sizes. // Simple lightweight structure that encapsulates image ptr on device, its pitch and its sizes.
// It is intended to pass to nvcc-compiled code. // It is intended to pass to nvcc-compiled code. GpuMat depends on headers that nvcc can't compile
template<typename T = unsigned char> template<typename T = unsigned char>
struct DevMem2D_ struct DevMem2D_
{ {
enum { elem_size = sizeof(T) }; typedef T elem_t;
enum { elem_size = sizeof(elem_t) };
int cols; int cols;
int rows; int rows;
......
modules/gpu/include/opencv2/gpu/gpu.hpp
View file @ 07825bad
...@@ -52,15 +52,20 @@ namespace cv ...@@ -52,15 +52,20 @@ namespace cv
{ {
//////////////////////////////// Initialization //////////////////////// //////////////////////////////// Initialization ////////////////////////
//! This is the only function that do not throw exceptions if the library is compiled without Cuda.
CV_EXPORTS int getCudaEnabledDeviceCount(); CV_EXPORTS int getCudaEnabledDeviceCount();
//! Functions below throw cv::Expception if the library is compiled without Cuda.
CV_EXPORTS string getDeviceName(int device); CV_EXPORTS string getDeviceName(int device);
CV_EXPORTS void setDevice(int device); CV_EXPORTS void setDevice(int device);
CV_EXPORTS int getDevice();
CV_EXPORTS void getComputeCapability(int device, int* major, int* minor); CV_EXPORTS void getComputeCapability(int device, int* major, int* minor);
CV_EXPORTS int getNumberOfSMs(int device); CV_EXPORTS int getNumberOfSMs(int device);
//////////////////////////////// GpuMat //////////////////////////////// //////////////////////////////// GpuMat ////////////////////////////////
//! Smart pointer for GPU memory with reference counting. Its interface is mostly similar with cv::Mat.
class CV_EXPORTS GpuMat class CV_EXPORTS GpuMat
{ {
public: public:
...@@ -85,7 +90,7 @@ namespace cv ...@@ -85,7 +90,7 @@ namespace cv
GpuMat(const GpuMat& m, const Rect& roi); GpuMat(const GpuMat& m, const Rect& roi);
//! builds GpuMat from Mat. Perfom blocking upload to device. //! builds GpuMat from Mat. Perfom blocking upload to device.
GpuMat (const Mat& m); explicit GpuMat (const Mat& m);
//! destructor - calls release() //! destructor - calls release()
~GpuMat(); ~GpuMat();
...@@ -211,44 +216,109 @@ namespace cv ...@@ -211,44 +216,109 @@ namespace cv
uchar* dataend; uchar* dataend;
}; };
//////////////////////////////// MatPL ////////////////////////////////
// MatPL is limited cv::Mat with page locked memory allocation.
// Page locked memory is only needed for async and faster coping to GPU.
// It is convertable to cv::Mat header without reference counting
// so you can use it with other opencv functions.
class CV_EXPORTS MatPL
{
public:
//Not supported. Now behaviour is like ALLOC_DEFAULT.
//enum { ALLOC_DEFAULT = 0, ALLOC_PORTABLE = 1, ALLOC_WRITE_COMBINED = 4 }
MatPL();
MatPL(const MatPL& m);
MatPL(int _rows, int _cols, int _type);
MatPL(Size _size, int _type);
//! creates from cv::Mat with coping data
explicit MatPL(const Mat& m);
~MatPL();
MatPL& operator = (const MatPL& m);
//! returns deep copy of the matrix, i.e. the data is copied
MatPL clone() const;
//! allocates new matrix data unless the matrix already has specified size and type.
void create(int _rows, int _cols, int _type);
void create(Size _size, int _type);
//! decrements reference counter and released memory if needed.
void release();
//! returns matrix header with disabled reference counting for MatPL data.
Mat createMatHeader() const;
operator Mat() const;
// Please see cv::Mat for descriptions
bool isContinuous() const;
size_t elemSize() const;
size_t elemSize1() const;
int type() const;
int depth() const;
int channels() const;
size_t step1() const;
Size size() const;
bool empty() const;
// Please see cv::Mat for descriptions
int flags;
int rows, cols;
size_t step;
uchar* data;
int* refcount;
uchar* datastart;
uchar* dataend;
};
//////////////////////////////// CudaStream //////////////////////////////// //////////////////////////////// CudaStream ////////////////////////////////
// Encapculates Cuda Stream. Provides interface for async coping.
// Passed to each function that supports async kernel execution.
// Reference counting is enabled
class CudaStream class CV_EXPORTS CudaStream
{ {
public: public:
static CudaStream empty();
CudaStream(); CudaStream();
~CudaStream(); ~CudaStream();
CudaStream(const CudaStream&);
CudaStream& operator=(const CudaStream&);
bool queryIfComplete(); bool queryIfComplete();
void waitForCompletion(); void waitForCompletion();
//calls cudaMemcpyAsync //! downloads asynchronously.
// Warning! cv::Mat must point to page locked memory (i.e. to MatPL data or to its subMat)
void enqueueDownload(const GpuMat& src, MatPL& dst);
void enqueueDownload(const GpuMat& src, Mat& dst); void enqueueDownload(const GpuMat& src, Mat& dst);
void enqueueUpload(const Mat& src, GpuMat& dst);
void enqueueCopy(const GpuMat& src, GpuMat& dst);
// calls cudaMemset2D asynchronous for single channel. Invoke kernel for some multichannel. //! uploads asynchronously.
void enqueueMemSet(const GpuMat& src, Scalar val); // Warning! cv::Mat must point to page locked memory (i.e. to MatPL data or to its ROI)
void enqueueUpload(const MatPL& src, GpuMat& dst);
void enqueueUpload(const Mat& src, GpuMat& dst);
// invoke kernel asynchronous because of mask void enqueueCopy(const GpuMat& src, GpuMat& dst);
void enqueueMemSet(const GpuMat& src, Scalar val);
void enqueueMemSet(const GpuMat& src, Scalar val, const GpuMat& mask); void enqueueMemSet(const GpuMat& src, Scalar val, const GpuMat& mask);
// converts matrix type, ex from float to uchar depending on type // converts matrix type, ex from float to uchar depending on type
void enqueueConvert(const GpuMat& src, GpuMat& dst, int type); void enqueueConvert(const GpuMat& src, GpuMat& dst, int type, double a = 1, double b = 0);
struct Impl;
const Impl& getImpl() const;
private: private:
void create();
Impl *impl; void release();
struct Impl;
Impl *impl;
friend struct StreamAccessor;
CudaStream(const CudaStream&);
CudaStream& operator=(const CudaStream&);
}; };
//////////////////////////////// StereoBM_GPU //////////////////////////////// //////////////////////////////// StereoBM_GPU ////////////////////////////////
...@@ -265,17 +335,22 @@ namespace cv ...@@ -265,17 +335,22 @@ namespace cv
StereoBM_GPU(int preset, int ndisparities=0); StereoBM_GPU(int preset, int ndisparities=0);
//! the stereo correspondence operator. Finds the disparity for the specified rectified stereo pair //! the stereo correspondence operator. Finds the disparity for the specified rectified stereo pair
//! Output disparity has CV_8U type. //! Output disparity has CV_8U type.
void operator() ( const GpuMat& left, const GpuMat& right, GpuMat& disparity) const; void operator() ( const GpuMat& left, const GpuMat& right, GpuMat& disparity);
//! Acync version
void operator() ( const GpuMat& left, const GpuMat& right, GpuMat& disparity, const CudaStream& stream);
//! Some heuristics that tries to estmate
// if current GPU will be faster then CPU in this algorithm.
// It queries current active device.
static bool checkIfGpuCallReasonable();
private: private:
mutable GpuMat minSSD; GpuMat minSSD;
int preset; int preset;
int ndisp; int ndisp;
}; };
} }
} }
#include "opencv2/gpu/matrix_operations.hpp"
#include "opencv2/gpu/gpumat.hpp"
#endif /* __OPENCV_GPU_HPP__ */ #endif /* __OPENCV_GPU_HPP__ */
\ No newline at end of file
modules/gpu/include/opencv2/gpu/matpl.hpp deleted 100644 → 0
View file @ 20e2dc84
This diff is collapsed.
modules/gpu/include/opencv2/gpu/gpumat.hpp → modules/gpu/include/opencv2/gpu/matrix_operations.hpp
View file @ 07825bad
...@@ -43,27 +43,25 @@ ...@@ -43,27 +43,25 @@
#ifndef __OPENCV_GPU_MATRIX_OPERATIONS_HPP__ #ifndef __OPENCV_GPU_MATRIX_OPERATIONS_HPP__
#define __OPENCV_GPU_MATRIX_OPERATIONS_HPP__ #define __OPENCV_GPU_MATRIX_OPERATIONS_HPP__
namespace cv namespace cv
{ {
namespace gpu namespace gpu
{ {
////////////////////////////////////////////////////////////////////////
//////////////////////////////// GpuMat //////////////////////////////// //////////////////////////////// GpuMat ////////////////////////////////
////////////////////////////////////////////////////////////////////////
inline GpuMat::GpuMat() inline GpuMat::GpuMat() : flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0) {}
: flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0) {}
inline GpuMat::GpuMat(int _rows, int _cols, int _type) inline GpuMat::GpuMat(int _rows, int _cols, int _type) : flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0)
: flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0)
{ {
if( _rows > 0 && _cols > 0 ) if( _rows > 0 && _cols > 0 )
create( _rows, _cols, _type ); create( _rows, _cols, _type );
} }
inline GpuMat::GpuMat(Size _size, int _type) inline GpuMat::GpuMat(Size _size, int _type) : flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0)
: flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0)
{ {
if( _size.height > 0 && _size.width > 0 ) if( _size.height > 0 && _size.width > 0 )
create( _size.height, _size.width, _type ); create( _size.height, _size.width, _type );
...@@ -249,12 +247,9 @@ inline void GpuMat::assignTo( GpuMat& m, int type ) const ...@@ -249,12 +247,9 @@ inline void GpuMat::assignTo( GpuMat& m, int type ) const
//CPP GpuMat& GpuMat::operator = (const Scalar& s); //CPP GpuMat& GpuMat::operator = (const Scalar& s);
//CPP GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask=GpuMat()); //CPP GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask=GpuMat());
//CPP GpuMat GpuMat::reshape(int _cn, int _rows=0) const; //CPP GpuMat GpuMat::reshape(int _cn, int _rows=0) const;
//CPP void GpuMat::create(int _rows, int _cols, int _type);
inline void GpuMat::create(Size _size, int _type) { create(_size.height, _size.width, _type); } inline void GpuMat::create(Size _size, int _type) { create(_size.height, _size.width, _type); }
//CPP void GpuMat::create(int _rows, int _cols, int _type);
//CPP void GpuMat::release(); //CPP void GpuMat::release();
inline void GpuMat::swap(GpuMat& b) inline void GpuMat::swap(GpuMat& b)
...@@ -343,6 +338,87 @@ template<typename _Tp> inline const _Tp* GpuMat::ptr(int y) const ...@@ -343,6 +338,87 @@ template<typename _Tp> inline const _Tp* GpuMat::ptr(int y) const
static inline void swap( GpuMat& a, GpuMat& b ) { a.swap(b); } static inline void swap( GpuMat& a, GpuMat& b ) { a.swap(b); }
///////////////////////////////////////////////////////////////////////
//////////////////////////////// MatPL ////////////////////////////////
///////////////////////////////////////////////////////////////////////
MatPL::MatPL() : flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0) {}
MatPL::MatPL(int _rows, int _cols, int _type) : flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0)
{
if( _rows > 0 && _cols > 0 )
create( _rows, _cols, _type );
}
MatPL::MatPL(Size _size, int _type) : flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0)
{
if( _size.height > 0 && _size.width > 0 )
create( _size.height, _size.width, _type );
}
MatPL::MatPL(const MatPL& m) : flags(m.flags), rows(m.rows), cols(m.cols), step(m.step), data(m.data), refcount(m.refcount), datastart(0), dataend(0)
{
if( refcount )
CV_XADD(refcount, 1);
}
MatPL::MatPL(const Mat& m) : flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0)
{
if( m.rows > 0 && m.cols > 0 )
create( m.size(), m.type() );
Mat tmp = createMatHeader();
m.copyTo(tmp);
}
MatPL::~MatPL()
{
release();
}
MatPL& MatPL::operator = (const MatPL& m)
{
if( this != &m )
{
if( m.refcount )
CV_XADD(m.refcount, 1);
release();
flags = m.flags;
rows = m.rows; cols = m.cols;
step = m.step; data = m.data;
datastart = m.datastart;
dataend = m.dataend;
refcount = m.refcount;
}
return *this;
}
MatPL MatPL::clone() const
{
MatPL m(size(), type());
Mat to = m;
Mat from = *this;
from.copyTo(to);
return m;
}
inline void MatPL::create(Size _size, int _type) { create(_size.height, _size.width, _type); }
//CCP void MatPL::create(int _rows, int _cols, int _type);
//CPP void MatPL::release();
inline Mat MatPL::createMatHeader() const { return Mat(size(), type(), data); }
inline MatPL::operator Mat() const { return createMatHeader(); }
inline bool MatPL::isContinuous() const { return (flags & Mat::CONTINUOUS_FLAG) != 0; }
inline size_t MatPL::elemSize() const { return CV_ELEM_SIZE(flags); }
inline size_t MatPL::elemSize1() const { return CV_ELEM_SIZE1(flags); }
inline int MatPL::type() const { return CV_MAT_TYPE(flags); }
inline int MatPL::depth() const { return CV_MAT_DEPTH(flags); }
inline int MatPL::channels() const { return CV_MAT_CN(flags); }
inline size_t MatPL::step1() const { return step/elemSize1(); }
inline Size MatPL::size() const { return Size(cols, rows); }
inline bool MatPL::empty() const { return data == 0; }
} /* end of namespace gpu */ } /* end of namespace gpu */
} /* end of namespace cv */ } /* end of namespace cv */
......
modules/gpu/include/opencv2/gpu/stream_accessor.hpp 0 → 100644
View file @ 07825bad
/*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 GpuMaterials 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*/
#ifndef __OPENCV_GPU_STREAM_ACCESSOR_HPP__
#define __OPENCV_GPU_STREAM_ACCESSOR_HPP__
#include "opencv2/gpu/gpu.hpp"
#include "cuda_runtime_api.h"
namespace cv
{
namespace gpu
{
// This is only header file that depends on Cuda. All other headers are independent.
// So if you use OpenCV binaries you do noot need to install Cuda Toolkit.
// But of you wanna use GPU by yourself, may get cuda stream instance using the class below.
// In this case you have to install Cuda Toolkit.
struct StreamAccessor
{
CV_EXPORTS static cudaStream_t getStream(const CudaStream& stream);
};
}
}
#endif /* __OPENCV_GPU_STREAM_ACCESSOR_HPP__ */
\ No newline at end of file
modules/gpu/cuda/cuda_shared.hpp → modules/gpu/src/cuda/cuda_shared.hpp
View file @ 07825bad
...@@ -44,31 +44,37 @@ ...@@ -44,31 +44,37 @@
#define __OPENCV_CUDA_SHARED_HPP__ #define __OPENCV_CUDA_SHARED_HPP__
#include "opencv2/gpu/devmem2d.hpp" #include "opencv2/gpu/devmem2d.hpp"
#include "cuda_runtime_api.h"
namespace cv namespace cv
{ {
namespace gpu namespace gpu
{ {
typedef unsigned char uchar; typedef unsigned char uchar;
typedef unsigned short ushort; typedef unsigned short ushort;
typedef unsigned int uint; typedef unsigned int uint;
extern "C" void error( const char *error_string, const char *file, const int line, const char *func = ""); extern "C" void error( const char *error_string, const char *file, const int line, const char *func = "");
namespace impl namespace impl
{ {
static inline int divUp(int a, int b) { return (a % b == 0) ? a/b : a/b + 1; } static inline int divUp(int a, int b) { return (a % b == 0) ? a/b : a/b + 1; }
extern "C" void stereoBM_GPU(const DevMem2D& left, const DevMem2D& right, DevMem2D& disp, int maxdisp, DevMem2D_<uint>& minSSD_buf); extern "C" void stereoBM_GPU(const DevMem2D& left, const DevMem2D& right, DevMem2D& disp, int maxdisp, DevMem2D_<uint>& minSSD_buf);
extern "C" void set_to_without_mask (const DevMem2D& mat, const double * scalar, int depth, int channels);
extern "C" void set_to_with_mask (const DevMem2D& mat, const double * scalar, const DevMem2D& mask, int depth, int channels);
} }
} }
} }
#ifdef __CUDACC__ #if defined(__GNUC__)
#define cudaSafeCall(expr) { cudaError_t err = expr; if( cudaSuccess != err) cv::gpu::error(cudaGetErrorString(err), __FILE__, __LINE__); } #define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__, __func__);
#else /* defined(__CUDACC__) || defined(__MSVC__) */
#define cudaSafeCall(expr) ___cudaSafeCall(expr, __FILE__, __LINE__)
#endif #endif
static 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);
}
#endif /* __OPENCV_CUDA_SHARED_HPP__ */ #endif /* __OPENCV_CUDA_SHARED_HPP__ */
modules/gpu/src/cudastream.cpp
View file @ 07825bad
...@@ -41,57 +41,119 @@ ...@@ -41,57 +41,119 @@
//M*/ //M*/
#include "precomp.hpp" #include "precomp.hpp"
//#include "opencv2/gpu/stream_access.hpp"
using namespace cv; using namespace cv;
using namespace cv::gpu; using namespace cv::gpu;
cv::gpu::CudaStream::CudaStream() //: impl( (Impl*)fastMalloc(sizeof(Impl)) ) #if !defined (HAVE_CUDA)
void cv::gpu::CudaStream::create() { throw_nogpu(); }
void cv::gpu::CudaStream::release() { throw_nogpu(); }
cv::gpu::CudaStream::CudaStream() : impl(0) { throw_nogpu(); }
cv::gpu::CudaStream::~CudaStream() { throw_nogpu(); }
cv::gpu::CudaStream::CudaStream(const CudaStream& stream) { throw_nogpu(); }
CudaStream& cv::gpu::CudaStream::operator=(const CudaStream& stream) { throw_nogpu(); return *this; }
bool cv::gpu::CudaStream::queryIfComplete() { throw_nogpu(); return true; }
void cv::gpu::CudaStream::waitForCompletion() { throw_nogpu(); }
void cv::gpu::CudaStream::enqueueDownload(const GpuMat& src, Mat& dst) { throw_nogpu(); }
void cv::gpu::CudaStream::enqueueDownload(const GpuMat& src, MatPL& dst) { throw_nogpu(); }
void cv::gpu::CudaStream::enqueueUpload(const MatPL& src, GpuMat& dst) { throw_nogpu(); }
void cv::gpu::CudaStream::enqueueUpload(const Mat& src, GpuMat& dst) { throw_nogpu(); }
void cv::gpu::CudaStream::enqueueCopy(const GpuMat& src, GpuMat& dst) { throw_nogpu(); }
void cv::gpu::CudaStream::enqueueMemSet(const GpuMat& src, Scalar val) { throw_nogpu(); }
void cv::gpu::CudaStream::enqueueMemSet(const GpuMat& src, Scalar val, const GpuMat& mask) { throw_nogpu(); }
void cv::gpu::CudaStream::enqueueConvert(const GpuMat& src, GpuMat& dst, int type, double a, double b) { throw_nogpu(); }
#else /* !defined (HAVE_CUDA) */
#include "opencv2/gpu/stream_accessor.hpp"
struct CudaStream::Impl
{ {
//cudaSafeCall( cudaStreamCreate( &impl->stream) ); cudaStream_t stream;
} int ref_counter;
cv::gpu::CudaStream::~CudaStream() };
namespace
{ {
if (impl) template<class S, class D> void devcopy(const S& src, D& dst, cudaStream_t s, cudaMemcpyKind k)
{ {
cudaSafeCall( cudaStreamDestroy( *(cudaStream_t*)impl ) ); dst.create(src.size(), src.type());
cv::fastFree( impl ); size_t bwidth = src.cols * src.elemSize();
} cudaSafeCall( cudaMemcpy2DAsync(dst.data, dst.step, src.data, src.step, bwidth, src.rows, k, s) );
};
} }
bool cv::gpu::CudaStream::queryIfComplete() CV_EXPORTS cudaStream_t cv::gpu::StreamAccessor::getStream(const CudaStream& stream) { return stream.impl->stream; };
void cv::gpu::CudaStream::create()
{ {
//cudaError_t err = cudaStreamQuery( *(cudaStream_t*)impl ); if (impl)
release();
//if (err == cudaSuccess) cudaStream_t stream;
// return true; cudaSafeCall( cudaStreamCreate( &stream ) );
//if (err == cudaErrorNotReady) impl = (CudaStream::Impl*)fastMalloc(sizeof(CudaStream::Impl));
// return false;
////cudaErrorInvalidResourceHandle impl->stream = stream;
//cudaSafeCall( err ); impl->ref_counter = 1;
return true;
} }
void cv::gpu::CudaStream::waitForCompletion()
void cv::gpu::CudaStream::release()
{ {
cudaSafeCall( cudaStreamSynchronize( *(cudaStream_t*)impl ) ); if( impl && CV_XADD(&impl->ref_counter, -1) == 1 )
{
cudaSafeCall( cudaStreamDestroy( impl->stream ) );
cv::fastFree( impl );
}
} }
void cv::gpu::CudaStream::enqueueDownload(const GpuMat& src, Mat& dst) cv::gpu::CudaStream::CudaStream() : impl(0) { create(); }
cv::gpu::CudaStream::~CudaStream() { release(); }
cv::gpu::CudaStream::CudaStream(const CudaStream& stream) : impl(stream.impl)
{ {
// cudaMemcpy2DAsync(dst.data, dst.step, src.data, src.step, src.cols * src.elemSize(), src.rows, cudaMemcpyDeviceToHost, if( impl )
CV_XADD(&impl->ref_counter, 1);
} }
void cv::gpu::CudaStream::enqueueUpload(const Mat& src, GpuMat& dst) CudaStream& cv::gpu::CudaStream::operator=(const CudaStream& stream)
{ {
CV_Assert(!"Not implemented"); if( this != &stream )
{
if( stream.impl )
CV_XADD(&stream.impl->ref_counter, 1);
release();
impl = stream.impl;
}
return *this;
} }
void cv::gpu::CudaStream::enqueueCopy(const GpuMat& src, GpuMat& dst)
bool cv::gpu::CudaStream::queryIfComplete()
{ {
CV_Assert(!"Not implemented"); cudaError_t err = cudaStreamQuery( impl->stream );
if (err == cudaErrorNotReady || err == cudaSuccess)
return err == cudaSuccess;
cudaSafeCall(err);
} }
void cv::gpu::CudaStream::waitForCompletion() { cudaSafeCall( cudaStreamSynchronize( impl->stream ) ); }
void cv::gpu::CudaStream::enqueueDownload(const GpuMat& src, Mat& dst)
{
// if not -> allocation will be done, but after that dst will not point to page locked memory
CV_Assert(src.cols == dst.cols && src.rows == dst.rows && src.type() == dst.type() )
devcopy(src, dst, impl->stream, cudaMemcpyDeviceToHost);
}
void cv::gpu::CudaStream::enqueueDownload(const GpuMat& src, MatPL& dst) { devcopy(src, dst, impl->stream, cudaMemcpyDeviceToHost); }
void cv::gpu::CudaStream::enqueueUpload(const MatPL& src, GpuMat& dst){ devcopy(src, dst, impl->stream, cudaMemcpyHostToDevice); }
void cv::gpu::CudaStream::enqueueUpload(const Mat& src, GpuMat& dst) { devcopy(src, dst, impl->stream, cudaMemcpyHostToDevice); }
void cv::gpu::CudaStream::enqueueCopy(const GpuMat& src, GpuMat& dst) { devcopy(src, dst, impl->stream, cudaMemcpyDeviceToDevice); }
void cv::gpu::CudaStream::enqueueMemSet(const GpuMat& src, Scalar val) void cv::gpu::CudaStream::enqueueMemSet(const GpuMat& src, Scalar val)
{ {
CV_Assert(!"Not implemented"); CV_Assert(!"Not implemented");
...@@ -102,11 +164,10 @@ void cv::gpu::CudaStream::enqueueMemSet(const GpuMat& src, Scalar val, const Gpu ...@@ -102,11 +164,10 @@ void cv::gpu::CudaStream::enqueueMemSet(const GpuMat& src, Scalar val, const Gpu
CV_Assert(!"Not implemented"); CV_Assert(!"Not implemented");
} }
void cv::gpu::CudaStream::enqueueConvert(const GpuMat& src, GpuMat& dst, int type) void cv::gpu::CudaStream::enqueueConvert(const GpuMat& src, GpuMat& dst, int type, double a, double b)
{ {
CV_Assert(!"Not implemented"); CV_Assert(!"Not implemented");
} }
//struct cudaStream_t& cv::gpu::CudaStream::getStream() { return stream; }
#endif /* !defined (HAVE_CUDA) */
\ No newline at end of file
modules/gpu/src/initialization.cpp
View file @ 07825bad
...@@ -45,15 +45,18 @@ ...@@ -45,15 +45,18 @@
using namespace cv; using namespace cv;
using namespace cv::gpu; using namespace cv::gpu;
#ifndef HAVE_CUDA
#if !defined (HAVE_CUDA)
CV_EXPORTS int cv::gpu::getCudaEnabledDeviceCount() { return 0; } CV_EXPORTS int cv::gpu::getCudaEnabledDeviceCount() { return 0; }
CV_EXPORTS string cv::gpu::getDeviceName(int /*device*/) { cudaSafeCall(0); return 0; } CV_EXPORTS string cv::gpu::getDeviceName(int /*device*/) { throw_nogpu(); return 0; }
CV_EXPORTS void cv::gpu::setDevice(int /*device*/) { cudaSafeCall(0); } CV_EXPORTS void cv::gpu::setDevice(int /*device*/) { throw_nogpu(); }
CV_EXPORTS void cv::gpu::getComputeCapability(int /*device*/, int* /*major*/, int* /*minor*/) { cudaSafeCall(0); } CV_EXPORTS int cv::gpu::getDevice() { throw_nogpu(); return 0; }
CV_EXPORTS int cv::gpu::getNumberOfSMs(int /*device*/) { cudaSafeCall(0); return 0; } CV_EXPORTS void cv::gpu::getComputeCapability(int /*device*/, int* /*major*/, int* /*minor*/) { throw_nogpu(); }
CV_EXPORTS int cv::gpu::getNumberOfSMs(int /*device*/) { throw_nogpu(); return 0; }
#else #else /* !defined (HAVE_CUDA) */
CV_EXPORTS int cv::gpu::getCudaEnabledDeviceCount() CV_EXPORTS int cv::gpu::getCudaEnabledDeviceCount()
{ {
...@@ -73,6 +76,12 @@ CV_EXPORTS void cv::gpu::setDevice(int device) ...@@ -73,6 +76,12 @@ CV_EXPORTS void cv::gpu::setDevice(int device)
{ {
cudaSafeCall( cudaSetDevice( device ) ); cudaSafeCall( cudaSetDevice( device ) );
} }
CV_EXPORTS int cv::gpu::getDevice()
{
int device;
cudaSafeCall( cudaGetDevice( &device ) );
return device;
}
CV_EXPORTS void cv::gpu::getComputeCapability(int device, int* major, int* minor) CV_EXPORTS void cv::gpu::getComputeCapability(int device, int* major, int* minor)
{ {
...@@ -90,4 +99,5 @@ CV_EXPORTS int cv::gpu::getNumberOfSMs(int device) ...@@ -90,4 +99,5 @@ CV_EXPORTS int cv::gpu::getNumberOfSMs(int device)
return prop.multiProcessorCount; return prop.multiProcessorCount;
} }
#endif #endif
\ No newline at end of file
modules/gpu/src/gpumat.cpp → modules/gpu/src/matrix_operations.cpp
View file @ 07825bad
...@@ -45,23 +45,53 @@ ...@@ -45,23 +45,53 @@
using namespace cv; using namespace cv;
using namespace cv::gpu; using namespace cv::gpu;
////////////////////////////////////////////////////////////////////////
//////////////////////////////// GpuMat //////////////////////////////// //////////////////////////////// GpuMat ////////////////////////////////
////////////////////////////////////////////////////////////////////////
void GpuMat::upload(const Mat& m)
#if !defined (HAVE_CUDA)
namespace cv
{
namespace gpu
{
void GpuMat::upload(const Mat& /*m*/) { throw_nogpu(); }
void GpuMat::download(cv::Mat& /*m*/) const { throw_nogpu(); }
void GpuMat::copyTo( GpuMat& /*m*/ ) const { throw_nogpu(); }
void GpuMat::copyTo( GpuMat& /*m*/, const GpuMat&/* mask */) const { throw_nogpu(); }
void GpuMat::convertTo( GpuMat& /*m*/, int /*rtype*/, double /*alpha*/, double /*beta*/ ) const { throw_nogpu(); }
GpuMat& GpuMat::operator = (const Scalar& /*s*/) { throw_nogpu(); return *this; }
GpuMat& GpuMat::setTo(const Scalar& /*s*/, const GpuMat& /*mask*/) { throw_nogpu(); return *this; }
GpuMat GpuMat::reshape(int /*new_cn*/, int /*new_rows*/) const { throw_nogpu(); return GpuMat(); }
void GpuMat::create(int /*_rows*/, int /*_cols*/, int /*_type*/) { throw_nogpu(); }
void GpuMat::release() { throw_nogpu(); }
void MatPL::create(int /*_rows*/, int /*_cols*/, int /*_type*/) { throw_nogpu(); }
void MatPL::release() { throw_nogpu(); }
}
}
#else /* !defined (HAVE_CUDA) */
void cv::gpu::GpuMat::upload(const Mat& m)
{ {
CV_DbgAssert(!m.empty()); CV_DbgAssert(!m.empty());
create(m.size(), m.type()); create(m.size(), m.type());
cudaSafeCall( cudaMemcpy2D(data, step, m.data, m.step, cols * elemSize(), rows, cudaMemcpyHostToDevice) ); cudaSafeCall( cudaMemcpy2D(data, step, m.data, m.step, cols * elemSize(), rows, cudaMemcpyHostToDevice) );
} }
void GpuMat::download(cv::Mat& m) const void cv::gpu::GpuMat::download(cv::Mat& m) const
{ {
CV_DbgAssert(!this->empty()); CV_DbgAssert(!this->empty());
m.create(size(), type()); m.create(size(), type());
cudaSafeCall( cudaMemcpy2D(m.data, m.step, data, step, cols * elemSize(), rows, cudaMemcpyDeviceToHost) ); cudaSafeCall( cudaMemcpy2D(m.data, m.step, data, step, cols * elemSize(), rows, cudaMemcpyDeviceToHost) );
} }
void GpuMat::copyTo( GpuMat& m ) const void cv::gpu::GpuMat::copyTo( GpuMat& m ) const
{ {
CV_DbgAssert(!this->empty()); CV_DbgAssert(!this->empty());
m.create(size(), type()); m.create(size(), type());
...@@ -69,45 +99,30 @@ void GpuMat::copyTo( GpuMat& m ) const ...@@ -69,45 +99,30 @@ void GpuMat::copyTo( GpuMat& m ) const
cudaSafeCall( cudaThreadSynchronize() ); cudaSafeCall( cudaThreadSynchronize() );
} }
void GpuMat::copyTo( GpuMat& /*m*/, const GpuMat&/* mask */) const void cv::gpu::GpuMat::copyTo( GpuMat& /*m*/, const GpuMat&/* mask */) const
{ {
CV_Assert(!"Not implemented"); CV_Assert(!"Not implemented");
} }
void GpuMat::convertTo( GpuMat& /*m*/, int /*rtype*/, double /*alpha*/, double /*beta*/ ) const void cv::gpu::GpuMat::convertTo( GpuMat& /*m*/, int /*rtype*/, double /*alpha*/, double /*beta*/ ) const
{ {
CV_Assert(!"Not implemented"); CV_Assert(!"Not implemented");
} }
GpuMat& GpuMat::operator = (const Scalar& s) GpuMat& cv::gpu::GpuMat::operator = (const Scalar& /*s*/)
{ {
cv::gpu::impl::set_to_without_mask(*this, s.val, this->depth(), this->channels()); CV_Assert(!"Not implemented");
return *this; return *this;
} }
GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask) GpuMat& cv::gpu::GpuMat::setTo(const Scalar& /*s*/, const GpuMat& /*mask*/)
{ {
CV_Assert(mask.type() == CV_32F); CV_Assert(!"Not implemented");
CV_DbgAssert(!this->empty());
this->channels();
this->depth();
if (mask.empty())
{
cv::gpu::impl::set_to_without_mask(*this, s.val, this->depth(), this->channels());
}
else
{
cv::gpu::impl::set_to_with_mask(*this, s.val, mask, this->depth(), this->channels());
}
return *this; return *this;
} }
GpuMat GpuMat::reshape(int new_cn, int new_rows) const GpuMat cv::gpu::GpuMat::reshape(int new_cn, int new_rows) const
{ {
GpuMat hdr = *this; GpuMat hdr = *this;
...@@ -148,7 +163,7 @@ GpuMat GpuMat::reshape(int new_cn, int new_rows) const ...@@ -148,7 +163,7 @@ GpuMat GpuMat::reshape(int new_cn, int new_rows) const
return hdr; return hdr;
} }
void GpuMat::create(int _rows, int _cols, int _type) void cv::gpu::GpuMat::create(int _rows, int _cols, int _type)
{ {
_type &= TYPE_MASK; _type &= TYPE_MASK;
if( rows == _rows && cols == _cols && type() == _type && data ) if( rows == _rows && cols == _cols && type() == _type && data )
...@@ -162,7 +177,7 @@ void GpuMat::create(int _rows, int _cols, int _type) ...@@ -162,7 +177,7 @@ void GpuMat::create(int _rows, int _cols, int _type)
rows = _rows; rows = _rows;
cols = _cols; cols = _cols;
size_t esz = elemSize(); size_t esz = elemSize();
void *dev_ptr; void *dev_ptr;
cudaSafeCall( cudaMallocPitch(&dev_ptr, &step, esz * cols, rows) ); cudaSafeCall( cudaMallocPitch(&dev_ptr, &step, esz * cols, rows) );
...@@ -174,19 +189,19 @@ void GpuMat::create(int _rows, int _cols, int _type) ...@@ -174,19 +189,19 @@ void GpuMat::create(int _rows, int _cols, int _type)
size_t nettosize = (size_t)_nettosize; size_t nettosize = (size_t)_nettosize;
datastart = data = (uchar*)dev_ptr; datastart = data = (uchar*)dev_ptr;
dataend = data + nettosize; dataend = data + nettosize;
refcount = (int*)fastMalloc(sizeof(*refcount)); refcount = (int*)fastMalloc(sizeof(*refcount));
*refcount = 1; *refcount = 1;
} }
} }
void GpuMat::release() void cv::gpu::GpuMat::release()
{ {
if( refcount && CV_XADD(refcount, -1) == 1 ) if( refcount && CV_XADD(refcount, -1) == 1 )
{ {
fastFree(refcount); fastFree(refcount);
cudaSafeCall( cudaFree(datastart) ); cudaSafeCall( cudaFree(datastart) );
} }
data = datastart = dataend = 0; data = datastart = dataend = 0;
step = rows = cols = 0; step = rows = cols = 0;
...@@ -194,7 +209,52 @@ void GpuMat::release() ...@@ -194,7 +209,52 @@ void GpuMat::release()
} }
///////////////////////////////////////////////////////////////////////
//////////////////////////////// MatPL ////////////////////////////////
///////////////////////////////////////////////////////////////////////
void cv::gpu::MatPL::create(int _rows, int _cols, int _type)
{
_type &= TYPE_MASK;
if( rows == _rows && cols == _cols && type() == _type && data )
return;
if( data )
release();
CV_DbgAssert( _rows >= 0 && _cols >= 0 );
if( _rows > 0 && _cols > 0 )
{
flags = Mat::MAGIC_VAL + Mat::CONTINUOUS_FLAG + _type;
rows = _rows;
cols = _cols;
step = elemSize()*cols;
int64 _nettosize = (int64)step*rows;
size_t nettosize = (size_t)_nettosize;
if( _nettosize != (int64)nettosize )
CV_Error(CV_StsNoMem, "Too big buffer is allocated");
size_t datasize = alignSize(nettosize, (int)sizeof(*refcount));
//datastart = data = (uchar*)fastMalloc(datasize + sizeof(*refcount));
void *ptr;
cudaSafeCall( cudaHostAlloc( &ptr, datasize, cudaHostAllocDefault) );
datastart = data = (uchar*)ptr;
dataend = data + nettosize;
refcount = (int*)cv::fastMalloc(sizeof(*refcount));
*refcount = 1;
}
}
void cv::gpu::MatPL::release()
{
if( refcount && CV_XADD(refcount, -1) == 1 )
{
cudaSafeCall( cudaFreeHost(datastart ) );
fastFree(refcount);
}
data = datastart = dataend = 0;
step = rows = cols = 0;
refcount = 0;
}
#endif /* !defined (HAVE_CUDA) */
\ No newline at end of file
modules/gpu/src/precomp.cpp
View file @ 07825bad
...@@ -44,7 +44,13 @@ ...@@ -44,7 +44,13 @@
/* End of file. */ /* End of file. */
extern "C" void cv::gpu::error( const char *error_string, const char *file, const int line, const char *func) namespace cv
{ {
cv::error( cv::Exception(CV_GpuApiCallError, error_string, func, file, line) ); namespace gpu
} {
extern "C" void error(const char *error_string, const char *file, const int line, const char *func)
{
cv::error( cv::Exception(CV_GpuApiCallError, error_string, func, file, line) );
}
}
}
modules/gpu/src/precomp.hpp
View file @ 07825bad
...@@ -53,30 +53,17 @@ ...@@ -53,30 +53,17 @@
#include <iostream> #include <iostream>
#include "opencv2/gpu/gpu.hpp" #include "opencv2/gpu/gpu.hpp"
#include "cuda_shared.hpp"
#ifndef HAVE_CUDA
#define cudaSafeCall(expr) CV_Error(CV_GpuNotFound, "The library is compilled with no GPU support") #if defined(HAVE_CUDA)
#define cudaCallerSafeCall(expr) CV_Error(CV_GpuNotFound, "The library is compilled with no GPU support")
#else /* HAVE_CUDA */ #include "cuda_shared.hpp"
#include "cuda_runtime_api.h"
#if _MSC_VER >= 1200 #else /* defined(HAVE_CUDA) */
#pragma warning (disable : 4100 4211 4201 4408)
#endif
#include "cuda_runtime_api.h"
#ifdef __GNUC__
#define cudaSafeCall(expr) { cudaError_t err = expr; if(cudaSuccess != err) cv::gpu::error(cudaGetErrorString(err), __FILE__, __LINE__, __func__); }
#else
#define cudaSafeCall(expr) { cudaError_t err = expr; if(cudaSuccess != err) cv::gpu::error(cudaGetErrorString(err), __FILE__, __LINE__); }
#endif
#define cudaCallerSafeCall(expr) expr; static inline void throw_nogpu() { CV_Error(CV_GpuNotFound, "The library is compilled with no GPU support"); }
#endif /* defined(HAVE_CUDA) */
#endif /* __OPENCV_PRECOMP_H__ */
#endif /* HAVE_CUDA */
#endif
modules/gpu/src/stereobm_gpu.cpp
View file @ 07825bad
...@@ -44,15 +44,45 @@ ...@@ -44,15 +44,45 @@
using namespace cv; using namespace cv;
using namespace cv::gpu; using namespace cv::gpu;
#if !defined (HAVE_CUDA)
cv::gpu::StereoBM_GPU::StereoBM_GPU() { throw_nogpu(); }
cv::gpu::StereoBM_GPU::StereoBM_GPU(int preset_, int ndisparities_) { throw_nogpu(); }
bool cv::gpu::StereoBM_GPU::checkIfGpuCallReasonable() { throw_nogpu(); return false; }
void cv::gpu::StereoBM_GPU::operator() ( const GpuMat& left, const GpuMat& right, GpuMat& disparity) { throw_nogpu(); }
void cv::gpu::StereoBM_GPU::operator() ( const GpuMat& left, const GpuMat& right, GpuMat& disparity, const CudaStream& stream) { throw_nogpu(); }
#else /* !defined (HAVE_CUDA) */
StereoBM_GPU::StereoBM_GPU() : preset(BASIC_PRESET), ndisp(64) {} cv::gpu::StereoBM_GPU::StereoBM_GPU() : preset(BASIC_PRESET), ndisp(64) {}
StereoBM_GPU::StereoBM_GPU(int preset_, int ndisparities_) : preset(preset_), ndisp(ndisparities_) cv::gpu::StereoBM_GPU::StereoBM_GPU(int preset_, int ndisparities_) : preset(preset_), ndisp(ndisparities_)
{ {
const int max_supported_ndisp = 1 << (sizeof(unsigned char) * 8); const int max_supported_ndisp = 1 << (sizeof(unsigned char) * 8);
CV_Assert(ndisp <= max_supported_ndisp); CV_Assert(ndisp <= max_supported_ndisp);
CV_Assert(ndisp % 8 == 0);
}
bool cv::gpu::StereoBM_GPU::checkIfGpuCallReasonable()
{
if (0 == getCudaEnabledDeviceCount())
return false;
int device = getDevice();
int minor, major;
getComputeCapability(device, &major, &minor);
int numSM = getNumberOfSMs(device);
if (major > 1 || numSM > 16)
return true;
return false;
} }
void StereoBM_GPU::operator() ( const GpuMat& left, const GpuMat& right, GpuMat& disparity) const void cv::gpu::StereoBM_GPU::operator() ( const GpuMat& left, const GpuMat& right, GpuMat& disparity)
{ {
CV_DbgAssert(left.rows == right.rows && left.cols == right.cols); CV_DbgAssert(left.rows == right.rows && left.cols == right.cols);
CV_DbgAssert(left.type() == CV_8UC1); CV_DbgAssert(left.type() == CV_8UC1);
...@@ -67,6 +97,13 @@ void StereoBM_GPU::operator() ( const GpuMat& left, const GpuMat& right, GpuMat& ...@@ -67,6 +97,13 @@ void StereoBM_GPU::operator() ( const GpuMat& left, const GpuMat& right, GpuMat&
} }
DevMem2D disp = disparity; DevMem2D disp = disparity;
DevMem2D_<uint> mssd = minSSD; DevMem2D_<unsigned int> mssd = minSSD;
cudaCallerSafeCall( impl::stereoBM_GPU(left, right, disp, ndisp, mssd) ); impl::stereoBM_GPU(left, right, disp, ndisp, mssd);
} }
void cv::gpu::StereoBM_GPU::operator() ( const GpuMat& left, const GpuMat& right, GpuMat& disparity, const CudaStream& stream)
{
CV_Assert(!"Not implemented");
}
#endif /* !defined (HAVE_CUDA) */
\ No newline at end of file
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment