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

Merge pull request #3531 from jet47:cuda-core-refactoring

Browse Source
This commit is contained in:
Vadim Pisarevsky
2014-12-26 12:12:42 +00:00
parent f85838087a f36546dbd2
commit 0ff67253f7
30 changed files with 487 additions and 441 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
modules/core/include/opencv2/core/base.hpp
View File

@@ -705,7 +705,7 @@ namespace ogl
namespace cuda
{
class CV_EXPORTS GpuMat;
class CV_EXPORTS CudaMem;
class CV_EXPORTS HostMem;
class CV_EXPORTS Stream;
class CV_EXPORTS Event;
}

48
modules/core/include/opencv2/core/cuda.hpp
View File

@@ -67,7 +67,9 @@ namespace cv { namespace cuda {
//! @addtogroup cudacore_struct
//! @{
//////////////////////////////// GpuMat ///////////////////////////////
//===================================================================================
// GpuMat
//===================================================================================
/** @brief Base storage class for GPU memory with reference counting.
@@ -325,13 +327,13 @@ The function does not reallocate memory if the matrix has proper attributes alre
*/
CV_EXPORTS void ensureSizeIsEnough(int rows, int cols, int type, OutputArray arr);
CV_EXPORTS GpuMat allocMatFromBuf(int rows, int cols, int type, GpuMat& mat);
//! BufferPool management (must be called before Stream creation)
CV_EXPORTS void setBufferPoolUsage(bool on);
CV_EXPORTS void setBufferPoolConfig(int deviceId, size_t stackSize, int stackCount);
//////////////////////////////// CudaMem ////////////////////////////////
//===================================================================================
// HostMem
//===================================================================================
/** @brief Class with reference counting wrapping special memory type allocation functions from CUDA.
@@ -348,43 +350,45 @@ Its interface is also Mat-like but with additional memory type parameters.
@note Allocation size of such memory types is usually limited. For more details, see *CUDA 2.2
Pinned Memory APIs* document or *CUDA C Programming Guide*.
*/
class CV_EXPORTS CudaMem
class CV_EXPORTS HostMem
{
public:
enum AllocType { PAGE_LOCKED = 1, SHARED = 2, WRITE_COMBINED = 4 };
explicit CudaMem(AllocType alloc_type = PAGE_LOCKED);
static MatAllocator* getAllocator(AllocType alloc_type = PAGE_LOCKED);
CudaMem(const CudaMem& m);
explicit HostMem(AllocType alloc_type = PAGE_LOCKED);
CudaMem(int rows, int cols, int type, AllocType alloc_type = PAGE_LOCKED);
CudaMem(Size size, int type, AllocType alloc_type = PAGE_LOCKED);
HostMem(const HostMem& m);
HostMem(int rows, int cols, int type, AllocType alloc_type = PAGE_LOCKED);
HostMem(Size size, int type, AllocType alloc_type = PAGE_LOCKED);
//! creates from host memory with coping data
explicit CudaMem(InputArray arr, AllocType alloc_type = PAGE_LOCKED);
explicit HostMem(InputArray arr, AllocType alloc_type = PAGE_LOCKED);
~CudaMem();
~HostMem();
CudaMem& operator =(const CudaMem& m);
HostMem& operator =(const HostMem& m);
//! swaps with other smart pointer
void swap(CudaMem& b);
void swap(HostMem& b);
//! returns deep copy of the matrix, i.e. the data is copied
CudaMem clone() const;
HostMem 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);
//! creates alternative CudaMem header for the same data, with different
//! creates alternative HostMem header for the same data, with different
//! number of channels and/or different number of rows
CudaMem reshape(int cn, int rows = 0) const;
HostMem reshape(int cn, int rows = 0) const;
//! decrements reference counter and released memory if needed.
void release();
//! returns matrix header with disabled reference counting for CudaMem data.
//! returns matrix header with disabled reference counting for HostMem data.
Mat createMatHeader() const;
/** @brief Maps CPU memory to GPU address space and creates the cuda::GpuMat header without reference counting
@@ -433,7 +437,9 @@ CV_EXPORTS void registerPageLocked(Mat& m);
*/
CV_EXPORTS void unregisterPageLocked(Mat& m);
///////////////////////////////// Stream //////////////////////////////////
//===================================================================================
// Stream
//===================================================================================
/** @brief This class encapsulates a queue of asynchronous calls.
@@ -528,7 +534,9 @@ private:
//! @} cudacore_struct
//////////////////////////////// Initialization & Info ////////////////////////
//===================================================================================
// Initialization & Info
//===================================================================================
//! @addtogroup cudacore_init
//! @{
@@ -570,7 +578,9 @@ enum FeatureSet
FEATURE_SET_COMPUTE_20 = 20,
FEATURE_SET_COMPUTE_21 = 21,
FEATURE_SET_COMPUTE_30 = 30,
FEATURE_SET_COMPUTE_32 = 32,
FEATURE_SET_COMPUTE_35 = 35,
FEATURE_SET_COMPUTE_50 = 50,
GLOBAL_ATOMICS = FEATURE_SET_COMPUTE_11,
SHARED_ATOMICS = FEATURE_SET_COMPUTE_12,

67
modules/core/include/opencv2/core/cuda.inl.hpp
View File

@@ -50,7 +50,9 @@
namespace cv { namespace cuda {
//////////////////////////////// GpuMat ///////////////////////////////
//===================================================================================
// GpuMat
//===================================================================================
inline
GpuMat::GpuMat(Allocator* allocator_)
@@ -145,6 +147,7 @@ void GpuMat::swap(GpuMat& b)
std::swap(datastart, b.datastart);
std::swap(dataend, b.dataend);
std::swap(refcount, b.refcount);
std::swap(allocator, b.allocator);
}
inline
@@ -374,16 +377,18 @@ void swap(GpuMat& a, GpuMat& b)
a.swap(b);
}
//////////////////////////////// CudaMem ////////////////////////////////
//===================================================================================
// HostMem
//===================================================================================
inline
CudaMem::CudaMem(AllocType alloc_type_)
HostMem::HostMem(AllocType alloc_type_)
: flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0), alloc_type(alloc_type_)
{
}
inline
CudaMem::CudaMem(const CudaMem& m)
HostMem::HostMem(const HostMem& m)
: flags(m.flags), rows(m.rows), cols(m.cols), step(m.step), data(m.data), refcount(m.refcount), datastart(m.datastart), dataend(m.dataend), alloc_type(m.alloc_type)
{
if( refcount )
@@ -391,7 +396,7 @@ CudaMem::CudaMem(const CudaMem& m)
}
inline
CudaMem::CudaMem(int rows_, int cols_, int type_, AllocType alloc_type_)
HostMem::HostMem(int rows_, int cols_, int type_, AllocType alloc_type_)
: flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0), alloc_type(alloc_type_)
{
if (rows_ > 0 && cols_ > 0)
@@ -399,7 +404,7 @@ CudaMem::CudaMem(int rows_, int cols_, int type_, AllocType alloc_type_)
}
inline
CudaMem::CudaMem(Size size_, int type_, AllocType alloc_type_)
HostMem::HostMem(Size size_, int type_, AllocType alloc_type_)
: flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0), alloc_type(alloc_type_)
{
if (size_.height > 0 && size_.width > 0)
@@ -407,24 +412,24 @@ CudaMem::CudaMem(Size size_, int type_, AllocType alloc_type_)
}
inline
CudaMem::CudaMem(InputArray arr, AllocType alloc_type_)
HostMem::HostMem(InputArray arr, AllocType alloc_type_)
: flags(0), rows(0), cols(0), step(0), data(0), refcount(0), datastart(0), dataend(0), alloc_type(alloc_type_)
{
arr.getMat().copyTo(*this);
}
inline
CudaMem::~CudaMem()
HostMem::~HostMem()
{
release();
}
inline
CudaMem& CudaMem::operator =(const CudaMem& m)
HostMem& HostMem::operator =(const HostMem& m)
{
if (this != &m)
{
CudaMem temp(m);
HostMem temp(m);
swap(temp);
}
@@ -432,7 +437,7 @@ CudaMem& CudaMem::operator =(const CudaMem& m)
}
inline
void CudaMem::swap(CudaMem& b)
void HostMem::swap(HostMem& b)
{
std::swap(flags, b.flags);
std::swap(rows, b.rows);
@@ -446,86 +451,88 @@ void CudaMem::swap(CudaMem& b)
}
inline
CudaMem CudaMem::clone() const
HostMem HostMem::clone() const
{
CudaMem m(size(), type(), alloc_type);
HostMem m(size(), type(), alloc_type);
createMatHeader().copyTo(m);
return m;
}
inline
void CudaMem::create(Size size_, int type_)
void HostMem::create(Size size_, int type_)
{
create(size_.height, size_.width, type_);
}
inline
Mat CudaMem::createMatHeader() const
Mat HostMem::createMatHeader() const
{
return Mat(size(), type(), data, step);
}
inline
bool CudaMem::isContinuous() const
bool HostMem::isContinuous() const
{
return (flags & Mat::CONTINUOUS_FLAG) != 0;
}
inline
size_t CudaMem::elemSize() const
size_t HostMem::elemSize() const
{
return CV_ELEM_SIZE(flags);
}
inline
size_t CudaMem::elemSize1() const
size_t HostMem::elemSize1() const
{
return CV_ELEM_SIZE1(flags);
}
inline
int CudaMem::type() const
int HostMem::type() const
{
return CV_MAT_TYPE(flags);
}
inline
int CudaMem::depth() const
int HostMem::depth() const
{
return CV_MAT_DEPTH(flags);
}
inline
int CudaMem::channels() const
int HostMem::channels() const
{
return CV_MAT_CN(flags);
}
inline
size_t CudaMem::step1() const
size_t HostMem::step1() const
{
return step / elemSize1();
}
inline
Size CudaMem::size() const
Size HostMem::size() const
{
return Size(cols, rows);
}
inline
bool CudaMem::empty() const
bool HostMem::empty() const
{
return data == 0;
}
static inline
void swap(CudaMem& a, CudaMem& b)
void swap(HostMem& a, HostMem& b)
{
a.swap(b);
}
//////////////////////////////// Stream ///////////////////////////////
//===================================================================================
// Stream
//===================================================================================
inline
Stream::Stream(const Ptr<Impl>& impl)
@@ -533,7 +540,9 @@ Stream::Stream(const Ptr<Impl>& impl)
{
}
//////////////////////////////// Initialization & Info ////////////////////////
//===================================================================================
// Initialization & Info
//===================================================================================
inline
bool TargetArchs::has(int major, int minor)
@@ -592,7 +601,9 @@ bool DeviceInfo::supports(FeatureSet feature_set) const
}} // namespace cv { namespace cuda {
//////////////////////////////// Mat ////////////////////////////////
//===================================================================================
// Mat
//===================================================================================
namespace cv {

16
modules/core/include/opencv2/core/mat.hpp
View File

@@ -160,8 +160,8 @@ public:
STD_VECTOR_MAT = 5 << KIND_SHIFT,
EXPR = 6 << KIND_SHIFT,
OPENGL_BUFFER = 7 << KIND_SHIFT,
CUDA_MEM = 8 << KIND_SHIFT,
GPU_MAT = 9 << KIND_SHIFT,
CUDA_HOST_MEM = 8 << KIND_SHIFT,
CUDA_GPU_MAT = 9 << KIND_SHIFT,
UMAT =10 << KIND_SHIFT,
STD_VECTOR_UMAT =11 << KIND_SHIFT
};
@@ -180,7 +180,7 @@ public:
_InputArray(const double& val);
_InputArray(const cuda::GpuMat& d_mat);
_InputArray(const ogl::Buffer& buf);
_InputArray(const cuda::CudaMem& cuda_mem);
_InputArray(const cuda::HostMem& cuda_mem);
template<typename _Tp> _InputArray(const cudev::GpuMat_<_Tp>& m);
_InputArray(const UMat& um);
_InputArray(const std::vector<UMat>& umv);
@@ -277,7 +277,7 @@ public:
_OutputArray(std::vector<Mat>& vec);
_OutputArray(cuda::GpuMat& d_mat);
_OutputArray(ogl::Buffer& buf);
_OutputArray(cuda::CudaMem& cuda_mem);
_OutputArray(cuda::HostMem& cuda_mem);
template<typename _Tp> _OutputArray(cudev::GpuMat_<_Tp>& m);
template<typename _Tp> _OutputArray(std::vector<_Tp>& vec);
template<typename _Tp> _OutputArray(std::vector<std::vector<_Tp> >& vec);
@@ -292,7 +292,7 @@ public:
_OutputArray(const std::vector<Mat>& vec);
_OutputArray(const cuda::GpuMat& d_mat);
_OutputArray(const ogl::Buffer& buf);
_OutputArray(const cuda::CudaMem& cuda_mem);
_OutputArray(const cuda::HostMem& cuda_mem);
template<typename _Tp> _OutputArray(const cudev::GpuMat_<_Tp>& m);
template<typename _Tp> _OutputArray(const std::vector<_Tp>& vec);
template<typename _Tp> _OutputArray(const std::vector<std::vector<_Tp> >& vec);
@@ -310,7 +310,7 @@ public:
virtual UMat& getUMatRef(int i=-1) const;
virtual cuda::GpuMat& getGpuMatRef() const;
virtual ogl::Buffer& getOGlBufferRef() const;
virtual cuda::CudaMem& getCudaMemRef() const;
virtual cuda::HostMem& getHostMemRef() const;
virtual void create(Size sz, int type, int i=-1, bool allowTransposed=false, int fixedDepthMask=0) const;
virtual void create(int rows, int cols, int type, int i=-1, bool allowTransposed=false, int fixedDepthMask=0) const;
virtual void create(int dims, const int* size, int type, int i=-1, bool allowTransposed=false, int fixedDepthMask=0) const;
@@ -333,7 +333,7 @@ public:
_InputOutputArray(std::vector<Mat>& vec);
_InputOutputArray(cuda::GpuMat& d_mat);
_InputOutputArray(ogl::Buffer& buf);
_InputOutputArray(cuda::CudaMem& cuda_mem);
_InputOutputArray(cuda::HostMem& cuda_mem);
template<typename _Tp> _InputOutputArray(cudev::GpuMat_<_Tp>& m);
template<typename _Tp> _InputOutputArray(std::vector<_Tp>& vec);
template<typename _Tp> _InputOutputArray(std::vector<std::vector<_Tp> >& vec);
@@ -348,7 +348,7 @@ public:
_InputOutputArray(const std::vector<Mat>& vec);
_InputOutputArray(const cuda::GpuMat& d_mat);
_InputOutputArray(const ogl::Buffer& buf);
_InputOutputArray(const cuda::CudaMem& cuda_mem);
_InputOutputArray(const cuda::HostMem& cuda_mem);
template<typename _Tp> _InputOutputArray(const cudev::GpuMat_<_Tp>& m);
template<typename _Tp> _InputOutputArray(const std::vector<_Tp>& vec);
template<typename _Tp> _InputOutputArray(const std::vector<std::vector<_Tp> >& vec);

30
modules/core/include/opencv2/core/mat.inl.hpp
View File

@@ -100,13 +100,13 @@ inline _InputArray::_InputArray(const MatExpr& expr)
{ init(FIXED_TYPE + FIXED_SIZE + EXPR + ACCESS_READ, &expr); }
inline _InputArray::_InputArray(const cuda::GpuMat& d_mat)
{ init(GPU_MAT + ACCESS_READ, &d_mat); }
{ init(CUDA_GPU_MAT + ACCESS_READ, &d_mat); }
inline _InputArray::_InputArray(const ogl::Buffer& buf)
{ init(OPENGL_BUFFER + ACCESS_READ, &buf); }
inline _InputArray::_InputArray(const cuda::CudaMem& cuda_mem)
{ init(CUDA_MEM + ACCESS_READ, &cuda_mem); }
inline _InputArray::_InputArray(const cuda::HostMem& cuda_mem)
{ init(CUDA_HOST_MEM + ACCESS_READ, &cuda_mem); }
inline _InputArray::~_InputArray() {}
@@ -174,13 +174,13 @@ _OutputArray::_OutputArray(const _Tp* vec, int n)
{ init(FIXED_TYPE + FIXED_SIZE + MATX + DataType<_Tp>::type + ACCESS_WRITE, vec, Size(n, 1)); }
inline _OutputArray::_OutputArray(cuda::GpuMat& d_mat)
{ init(GPU_MAT + ACCESS_WRITE, &d_mat); }
{ init(CUDA_GPU_MAT + ACCESS_WRITE, &d_mat); }
inline _OutputArray::_OutputArray(ogl::Buffer& buf)
{ init(OPENGL_BUFFER + ACCESS_WRITE, &buf); }
inline _OutputArray::_OutputArray(cuda::CudaMem& cuda_mem)
{ init(CUDA_MEM + ACCESS_WRITE, &cuda_mem); }
inline _OutputArray::_OutputArray(cuda::HostMem& cuda_mem)
{ init(CUDA_HOST_MEM + ACCESS_WRITE, &cuda_mem); }
inline _OutputArray::_OutputArray(const Mat& m)
{ init(FIXED_TYPE + FIXED_SIZE + MAT + ACCESS_WRITE, &m); }
@@ -195,13 +195,13 @@ inline _OutputArray::_OutputArray(const std::vector<UMat>& vec)
{ init(FIXED_SIZE + STD_VECTOR_UMAT + ACCESS_WRITE, &vec); }
inline _OutputArray::_OutputArray(const cuda::GpuMat& d_mat)
{ init(FIXED_TYPE + FIXED_SIZE + GPU_MAT + ACCESS_WRITE, &d_mat); }
{ init(FIXED_TYPE + FIXED_SIZE + CUDA_GPU_MAT + ACCESS_WRITE, &d_mat); }
inline _OutputArray::_OutputArray(const ogl::Buffer& buf)
{ init(FIXED_TYPE + FIXED_SIZE + OPENGL_BUFFER + ACCESS_WRITE, &buf); }
inline _OutputArray::_OutputArray(const cuda::CudaMem& cuda_mem)
{ init(FIXED_TYPE + FIXED_SIZE + CUDA_MEM + ACCESS_WRITE, &cuda_mem); }
inline _OutputArray::_OutputArray(const cuda::HostMem& cuda_mem)
{ init(FIXED_TYPE + FIXED_SIZE + CUDA_HOST_MEM + ACCESS_WRITE, &cuda_mem); }
///////////////////////////////////////////////////////////////////////////////////////////
@@ -261,13 +261,13 @@ _InputOutputArray::_InputOutputArray(const _Tp* vec, int n)
{ init(FIXED_TYPE + FIXED_SIZE + MATX + DataType<_Tp>::type + ACCESS_RW, vec, Size(n, 1)); }
inline _InputOutputArray::_InputOutputArray(cuda::GpuMat& d_mat)
{ init(GPU_MAT + ACCESS_RW, &d_mat); }
{ init(CUDA_GPU_MAT + ACCESS_RW, &d_mat); }
inline _InputOutputArray::_InputOutputArray(ogl::Buffer& buf)
{ init(OPENGL_BUFFER + ACCESS_RW, &buf); }
inline _InputOutputArray::_InputOutputArray(cuda::CudaMem& cuda_mem)
{ init(CUDA_MEM + ACCESS_RW, &cuda_mem); }
inline _InputOutputArray::_InputOutputArray(cuda::HostMem& cuda_mem)
{ init(CUDA_HOST_MEM + ACCESS_RW, &cuda_mem); }
inline _InputOutputArray::_InputOutputArray(const Mat& m)
{ init(FIXED_TYPE + FIXED_SIZE + MAT + ACCESS_RW, &m); }
@@ -282,13 +282,13 @@ inline _InputOutputArray::_InputOutputArray(const std::vector<UMat>& vec)
{ init(FIXED_SIZE + STD_VECTOR_UMAT + ACCESS_RW, &vec); }
inline _InputOutputArray::_InputOutputArray(const cuda::GpuMat& d_mat)
{ init(FIXED_TYPE + FIXED_SIZE + GPU_MAT + ACCESS_RW, &d_mat); }
{ init(FIXED_TYPE + FIXED_SIZE + CUDA_GPU_MAT + ACCESS_RW, &d_mat); }
inline _InputOutputArray::_InputOutputArray(const ogl::Buffer& buf)
{ init(FIXED_TYPE + FIXED_SIZE + OPENGL_BUFFER + ACCESS_RW, &buf); }
inline _InputOutputArray::_InputOutputArray(const cuda::CudaMem& cuda_mem)
{ init(FIXED_TYPE + FIXED_SIZE + CUDA_MEM + ACCESS_RW, &cuda_mem); }
inline _InputOutputArray::_InputOutputArray(const cuda::HostMem& cuda_mem)
{ init(FIXED_TYPE + FIXED_SIZE + CUDA_HOST_MEM + ACCESS_RW, &cuda_mem); }
//////////////////////////////////////////// Mat //////////////////////////////////////////

196
modules/core/perf/cuda/perf_gpumat.cpp Normal file
View File

@@ -0,0 +1,196 @@
/*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 "../perf_precomp.hpp"
#ifdef HAVE_CUDA
#include "opencv2/core/cuda.hpp"
#include "opencv2/ts/cuda_perf.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// SetTo
PERF_TEST_P(Sz_Depth_Cn, CUDA_GpuMat_SetTo,
Combine(CUDA_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_32F, CV_64F),
CUDA_CHANNELS_1_3_4))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
const int type = CV_MAKE_TYPE(depth, channels);
const cv::Scalar val(1, 2, 3, 4);
if (PERF_RUN_CUDA())
{
cv::cuda::GpuMat dst(size, type);
TEST_CYCLE() dst.setTo(val);
}
else
{
cv::Mat dst(size, type);
TEST_CYCLE() dst.setTo(val);
}
SANITY_CHECK_NOTHING();
}
//////////////////////////////////////////////////////////////////////
// SetToMasked
PERF_TEST_P(Sz_Depth_Cn, CUDA_GpuMat_SetToMasked,
Combine(CUDA_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_32F, CV_64F),
CUDA_CHANNELS_1_3_4))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
const int type = CV_MAKE_TYPE(depth, channels);
cv::Mat src(size, type);
cv::Mat mask(size, CV_8UC1);
declare.in(src, mask, WARMUP_RNG);
const cv::Scalar val(1, 2, 3, 4);
if (PERF_RUN_CUDA())
{
cv::cuda::GpuMat dst(src);
const cv::cuda::GpuMat d_mask(mask);
TEST_CYCLE() dst.setTo(val, d_mask);
}
else
{
cv::Mat dst = src;
TEST_CYCLE() dst.setTo(val, mask);
}
SANITY_CHECK_NOTHING();
}
//////////////////////////////////////////////////////////////////////
// CopyToMasked
PERF_TEST_P(Sz_Depth_Cn, CUDA_GpuMat_CopyToMasked,
Combine(CUDA_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_32F, CV_64F),
CUDA_CHANNELS_1_3_4))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
const int type = CV_MAKE_TYPE(depth, channels);
cv::Mat src(size, type);
cv::Mat mask(size, CV_8UC1);
declare.in(src, mask, WARMUP_RNG);
if (PERF_RUN_CUDA())
{
const cv::cuda::GpuMat d_src(src);
const cv::cuda::GpuMat d_mask(mask);
cv::cuda::GpuMat dst(d_src.size(), d_src.type(), cv::Scalar::all(0));
TEST_CYCLE() d_src.copyTo(dst, d_mask);
}
else
{
cv::Mat dst(src.size(), src.type(), cv::Scalar::all(0));
TEST_CYCLE() src.copyTo(dst, mask);
}
SANITY_CHECK_NOTHING();
}
//////////////////////////////////////////////////////////////////////
// ConvertTo
DEF_PARAM_TEST(Sz_2Depth, cv::Size, MatDepth, MatDepth);
PERF_TEST_P(Sz_2Depth, CUDA_GpuMat_ConvertTo,
Combine(CUDA_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_32F, CV_64F),
Values(CV_8U, CV_16U, CV_32F, CV_64F)))
{
const cv::Size size = GET_PARAM(0);
const int depth1 = GET_PARAM(1);
const int depth2 = GET_PARAM(2);
cv::Mat src(size, depth1);
declare.in(src, WARMUP_RNG);
const double a = 0.5;
const double b = 1.0;
if (PERF_RUN_CUDA())
{
const cv::cuda::GpuMat d_src(src);
cv::cuda::GpuMat dst;
TEST_CYCLE() d_src.convertTo(dst, depth2, a, b);
}
else
{
cv::Mat dst;
TEST_CYCLE() src.convertTo(dst, depth2, a, b);
}
SANITY_CHECK_NOTHING();
}
#endif

20
modules/core/src/cuda_gpu_mat.cpp
View File

@@ -275,12 +275,12 @@ void cv::cuda::createContinuous(int rows, int cols, int type, OutputArray arr)
::createContinuousImpl(rows, cols, type, arr.getMatRef());
break;
case _InputArray::GPU_MAT:
case _InputArray::CUDA_GPU_MAT:
::createContinuousImpl(rows, cols, type, arr.getGpuMatRef());
break;
case _InputArray::CUDA_MEM:
::createContinuousImpl(rows, cols, type, arr.getCudaMemRef());
case _InputArray::CUDA_HOST_MEM:
::createContinuousImpl(rows, cols, type, arr.getHostMemRef());
break;
default:
@@ -329,12 +329,12 @@ void cv::cuda::ensureSizeIsEnough(int rows, int cols, int type, OutputArray arr)
::ensureSizeIsEnoughImpl(rows, cols, type, arr.getMatRef());
break;
case _InputArray::GPU_MAT:
case _InputArray::CUDA_GPU_MAT:
::ensureSizeIsEnoughImpl(rows, cols, type, arr.getGpuMatRef());
break;
case _InputArray::CUDA_MEM:
::ensureSizeIsEnoughImpl(rows, cols, type, arr.getCudaMemRef());
case _InputArray::CUDA_HOST_MEM:
::ensureSizeIsEnoughImpl(rows, cols, type, arr.getHostMemRef());
break;
default:
@@ -342,14 +342,6 @@ void cv::cuda::ensureSizeIsEnough(int rows, int cols, int type, OutputArray arr)
}
}
GpuMat cv::cuda::allocMatFromBuf(int rows, int cols, int type, GpuMat& mat)
{
if (!mat.empty() && mat.type() == type && mat.rows >= rows && mat.cols >= cols)
return mat(Rect(0, 0, cols, rows));
return mat = GpuMat(rows, cols, type);
}
#ifndef HAVE_CUDA
GpuMat::Allocator* cv::cuda::GpuMat::defaultAllocator()

124
modules/core/src/cuda_host_mem.cpp
View File

@@ -42,10 +42,124 @@
//M*/
#include "precomp.hpp"
#include <map>
using namespace cv;
using namespace cv::cuda;
#ifdef HAVE_CUDA
namespace {
class HostMemAllocator : public MatAllocator
{
public:
explicit HostMemAllocator(unsigned int flags) : flags_(flags)
{
}
UMatData* allocate(int dims, const int* sizes, int type,
void* data0, size_t* step,
int /*flags*/, UMatUsageFlags /*usageFlags*/) const
{
size_t total = CV_ELEM_SIZE(type);
for (int i = dims-1; i >= 0; i--)
{
if (step)
{
if (data0 && step[i] != CV_AUTOSTEP)
{
CV_Assert(total <= step[i]);
total = step[i];
}
else
{
step[i] = total;
}
}
total *= sizes[i];
}
UMatData* u = new UMatData(this);
u->size = total;
if (data0)
{
u->data = u->origdata = static_cast<uchar*>(data0);
u->flags |= UMatData::USER_ALLOCATED;
}
else
{
void* ptr = 0;
cudaSafeCall( cudaHostAlloc(&ptr, total, flags_) );
u->data = u->origdata = static_cast<uchar*>(ptr);
}
return u;
}
bool allocate(UMatData* u, int /*accessFlags*/, UMatUsageFlags /*usageFlags*/) const
{
return (u != NULL);
}
void deallocate(UMatData* u) const
{
CV_Assert(u->urefcount >= 0);
CV_Assert(u->refcount >= 0);
if (u && u->refcount == 0)
{
if ( !(u->flags & UMatData::USER_ALLOCATED) )
{
cudaFreeHost(u->origdata);
u->origdata = 0;
}
delete u;
}
}
private:
unsigned int flags_;
};
} // namespace
#endif
MatAllocator* cv::cuda::HostMem::getAllocator(AllocType alloc_type)
{
#ifndef HAVE_CUDA
(void) alloc_type;
throw_no_cuda();
return NULL;
#else
static std::map<unsigned int, Ptr<MatAllocator> > allocators;
unsigned int flag = cudaHostAllocDefault;
switch (alloc_type)
{
case PAGE_LOCKED: flag = cudaHostAllocDefault; break;
case SHARED: flag = cudaHostAllocMapped; break;
case WRITE_COMBINED: flag = cudaHostAllocWriteCombined; break;
default: CV_Error(cv::Error::StsBadFlag, "Invalid alloc type");
}
Ptr<MatAllocator>& a = allocators[flag];
if (a.empty())
{
a = makePtr<HostMemAllocator>(flag);
}
return a.get();
#endif
}
#ifdef HAVE_CUDA
namespace
{
@@ -59,7 +173,7 @@ namespace
}
#endif
void cv::cuda::CudaMem::create(int rows_, int cols_, int type_)
void cv::cuda::HostMem::create(int rows_, int cols_, int type_)
{
#ifndef HAVE_CUDA
(void) rows_;
@@ -123,9 +237,9 @@ void cv::cuda::CudaMem::create(int rows_, int cols_, int type_)
#endif
}
CudaMem cv::cuda::CudaMem::reshape(int new_cn, int new_rows) const
HostMem cv::cuda::HostMem::reshape(int new_cn, int new_rows) const
{
CudaMem hdr = *this;
HostMem hdr = *this;
int cn = channels();
if (new_cn == 0)
@@ -166,7 +280,7 @@ CudaMem cv::cuda::CudaMem::reshape(int new_cn, int new_rows) const
return hdr;
}
void cv::cuda::CudaMem::release()
void cv::cuda::HostMem::release()
{
#ifdef HAVE_CUDA
if (refcount && CV_XADD(refcount, -1) == 1)
@@ -181,7 +295,7 @@ void cv::cuda::CudaMem::release()
#endif
}
GpuMat cv::cuda::CudaMem::createGpuMatHeader() const
GpuMat cv::cuda::HostMem::createGpuMatHeader() const
{
#ifndef HAVE_CUDA
throw_no_cuda();

94
modules/core/src/matrix.cpp
View File

@@ -1187,18 +1187,18 @@ Mat _InputArray::getMat(int i) const
return Mat();
}
if( k == GPU_MAT )
if( k == CUDA_GPU_MAT )
{
CV_Assert( i < 0 );
CV_Error(cv::Error::StsNotImplemented, "You should explicitly call download method for cuda::GpuMat object");
return Mat();
}
if( k == CUDA_MEM )
if( k == CUDA_HOST_MEM )
{
CV_Assert( i < 0 );
const cuda::CudaMem* cuda_mem = (const cuda::CudaMem*)obj;
const cuda::HostMem* cuda_mem = (const cuda::HostMem*)obj;
return cuda_mem->createMatHeader();
}
@@ -1391,15 +1391,15 @@ cuda::GpuMat _InputArray::getGpuMat() const
{
int k = kind();
if (k == GPU_MAT)
if (k == CUDA_GPU_MAT)
{
const cuda::GpuMat* d_mat = (const cuda::GpuMat*)obj;
return *d_mat;
}
if (k == CUDA_MEM)
if (k == CUDA_HOST_MEM)
{
const cuda::CudaMem* cuda_mem = (const cuda::CudaMem*)obj;
const cuda::HostMem* cuda_mem = (const cuda::HostMem*)obj;
return cuda_mem->createGpuMatHeader();
}
@@ -1412,7 +1412,7 @@ cuda::GpuMat _InputArray::getGpuMat() const
if (k == NONE)
return cuda::GpuMat();
CV_Error(cv::Error::StsNotImplemented, "getGpuMat is available only for cuda::GpuMat and cuda::CudaMem");
CV_Error(cv::Error::StsNotImplemented, "getGpuMat is available only for cuda::GpuMat and cuda::HostMem");
return cuda::GpuMat();
}
@@ -1520,20 +1520,22 @@ Size _InputArray::size(int i) const
return buf->size();
}
if( k == GPU_MAT )
if( k == CUDA_GPU_MAT )
{
CV_Assert( i < 0 );
const cuda::GpuMat* d_mat = (const cuda::GpuMat*)obj;
return d_mat->size();
}
CV_Assert( k == CUDA_MEM );
//if( k == CUDA_MEM )
if( k == CUDA_HOST_MEM )
{
CV_Assert( i < 0 );
const cuda::CudaMem* cuda_mem = (const cuda::CudaMem*)obj;
const cuda::HostMem* cuda_mem = (const cuda::HostMem*)obj;
return cuda_mem->size();
}
CV_Error(Error::StsNotImplemented, "Unknown/unsupported array type");
return Size();
}
int _InputArray::sizend(int* arrsz, int i) const
@@ -1700,18 +1702,20 @@ int _InputArray::dims(int i) const
return 2;
}
if( k == GPU_MAT )
if( k == CUDA_GPU_MAT )
{
CV_Assert( i < 0 );
return 2;
}
CV_Assert( k == CUDA_MEM );
//if( k == CUDA_MEM )
if( k == CUDA_HOST_MEM )
{
CV_Assert( i < 0 );
return 2;
}
CV_Error(Error::StsNotImplemented, "Unknown/unsupported array type");
return 0;
}
size_t _InputArray::total(int i) const
@@ -1799,12 +1803,14 @@ int _InputArray::type(int i) const
if( k == OPENGL_BUFFER )
return ((const ogl::Buffer*)obj)->type();
if( k == GPU_MAT )
if( k == CUDA_GPU_MAT )
return ((const cuda::GpuMat*)obj)->type();
CV_Assert( k == CUDA_MEM );
//if( k == CUDA_MEM )
return ((const cuda::CudaMem*)obj)->type();
if( k == CUDA_HOST_MEM )
return ((const cuda::HostMem*)obj)->type();
CV_Error(Error::StsNotImplemented, "Unknown/unsupported array type");
return 0;
}
int _InputArray::depth(int i) const
@@ -1863,12 +1869,14 @@ bool _InputArray::empty() const
if( k == OPENGL_BUFFER )
return ((const ogl::Buffer*)obj)->empty();
if( k == GPU_MAT )
if( k == CUDA_GPU_MAT )
return ((const cuda::GpuMat*)obj)->empty();
CV_Assert( k == CUDA_MEM );
//if( k == CUDA_MEM )
return ((const cuda::CudaMem*)obj)->empty();
if( k == CUDA_HOST_MEM )
return ((const cuda::HostMem*)obj)->empty();
CV_Error(Error::StsNotImplemented, "Unknown/unsupported array type");
return true;
}
bool _InputArray::isContinuous(int i) const
@@ -1970,7 +1978,7 @@ size_t _InputArray::offset(int i) const
return vv[i].offset;
}
if( k == GPU_MAT )
if( k == CUDA_GPU_MAT )
{
CV_Assert( i < 0 );
const cuda::GpuMat * const m = ((const cuda::GpuMat*)obj);
@@ -2016,7 +2024,7 @@ size_t _InputArray::step(int i) const
return vv[i].step;
}
if( k == GPU_MAT )
if( k == CUDA_GPU_MAT )
{
CV_Assert( i < 0 );
return ((const cuda::GpuMat*)obj)->step;
@@ -2095,7 +2103,7 @@ void _OutputArray::create(Size _sz, int mtype, int i, bool allowTransposed, int
((UMat*)obj)->create(_sz, mtype);
return;
}
if( k == GPU_MAT && i < 0 && !allowTransposed && fixedDepthMask == 0 )
if( k == CUDA_GPU_MAT && i < 0 && !allowTransposed && fixedDepthMask == 0 )
{
CV_Assert(!fixedSize() || ((cuda::GpuMat*)obj)->size() == _sz);
CV_Assert(!fixedType() || ((cuda::GpuMat*)obj)->type() == mtype);
@@ -2109,11 +2117,11 @@ void _OutputArray::create(Size _sz, int mtype, int i, bool allowTransposed, int
((ogl::Buffer*)obj)->create(_sz, mtype);
return;
}
if( k == CUDA_MEM && i < 0 && !allowTransposed && fixedDepthMask == 0 )
if( k == CUDA_HOST_MEM && i < 0 && !allowTransposed && fixedDepthMask == 0 )
{
CV_Assert(!fixedSize() || ((cuda::CudaMem*)obj)->size() == _sz);
CV_Assert(!fixedType() || ((cuda::CudaMem*)obj)->type() == mtype);
((cuda::CudaMem*)obj)->create(_sz, mtype);
CV_Assert(!fixedSize() || ((cuda::HostMem*)obj)->size() == _sz);
CV_Assert(!fixedType() || ((cuda::HostMem*)obj)->type() == mtype);
((cuda::HostMem*)obj)->create(_sz, mtype);
return;
}
int sizes[] = {_sz.height, _sz.width};
@@ -2137,7 +2145,7 @@ void _OutputArray::create(int _rows, int _cols, int mtype, int i, bool allowTran
((UMat*)obj)->create(_rows, _cols, mtype);
return;
}
if( k == GPU_MAT && i < 0 && !allowTransposed && fixedDepthMask == 0 )
if( k == CUDA_GPU_MAT && i < 0 && !allowTransposed && fixedDepthMask == 0 )
{
CV_Assert(!fixedSize() || ((cuda::GpuMat*)obj)->size() == Size(_cols, _rows));
CV_Assert(!fixedType() || ((cuda::GpuMat*)obj)->type() == mtype);
@@ -2151,11 +2159,11 @@ void _OutputArray::create(int _rows, int _cols, int mtype, int i, bool allowTran
((ogl::Buffer*)obj)->create(_rows, _cols, mtype);
return;
}
if( k == CUDA_MEM && i < 0 && !allowTransposed && fixedDepthMask == 0 )
if( k == CUDA_HOST_MEM && i < 0 && !allowTransposed && fixedDepthMask == 0 )
{
CV_Assert(!fixedSize() || ((cuda::CudaMem*)obj)->size() == Size(_cols, _rows));
CV_Assert(!fixedType() || ((cuda::CudaMem*)obj)->type() == mtype);
((cuda::CudaMem*)obj)->create(_rows, _cols, mtype);
CV_Assert(!fixedSize() || ((cuda::HostMem*)obj)->size() == Size(_cols, _rows));
CV_Assert(!fixedType() || ((cuda::HostMem*)obj)->type() == mtype);
((cuda::HostMem*)obj)->create(_rows, _cols, mtype);
return;
}
int sizes[] = {_rows, _cols};
@@ -2479,15 +2487,15 @@ void _OutputArray::release() const
return;
}
if( k == GPU_MAT )
if( k == CUDA_GPU_MAT )
{
((cuda::GpuMat*)obj)->release();
return;
}
if( k == CUDA_MEM )
if( k == CUDA_HOST_MEM )
{
((cuda::CudaMem*)obj)->release();
((cuda::HostMem*)obj)->release();
return;
}
@@ -2583,7 +2591,7 @@ UMat& _OutputArray::getUMatRef(int i) const
cuda::GpuMat& _OutputArray::getGpuMatRef() const
{
int k = kind();
CV_Assert( k == GPU_MAT );
CV_Assert( k == CUDA_GPU_MAT );
return *(cuda::GpuMat*)obj;
}
@@ -2594,11 +2602,11 @@ ogl::Buffer& _OutputArray::getOGlBufferRef() const
return *(ogl::Buffer*)obj;
}
cuda::CudaMem& _OutputArray::getCudaMemRef() const
cuda::HostMem& _OutputArray::getHostMemRef() const
{
int k = kind();
CV_Assert( k == CUDA_MEM );
return *(cuda::CudaMem*)obj;
CV_Assert( k == CUDA_HOST_MEM );
return *(cuda::HostMem*)obj;
}
void _OutputArray::setTo(const _InputArray& arr, const _InputArray & mask) const
@@ -2614,10 +2622,10 @@ void _OutputArray::setTo(const _InputArray& arr, const _InputArray & mask) const
}
else if( k == UMAT )
((UMat*)obj)->setTo(arr, mask);
else if( k == GPU_MAT )
else if( k == CUDA_GPU_MAT )
{
Mat value = arr.getMat();
CV_Assert( checkScalar(value, type(), arr.kind(), _InputArray::GPU_MAT) );
CV_Assert( checkScalar(value, type(), arr.kind(), _InputArray::CUDA_GPU_MAT) );
((cuda::GpuMat*)obj)->setTo(Scalar(Vec<double, 4>(value.ptr<double>())), mask);
}
else

12
modules/core/src/opengl.cpp
View File

@@ -509,7 +509,7 @@ cv::ogl::Buffer::Buffer(InputArray arr, Target target, bool autoRelease) : rows_
switch (kind)
{
case _InputArray::OPENGL_BUFFER:
case _InputArray::GPU_MAT:
case _InputArray::CUDA_GPU_MAT:
copyFrom(arr, target, autoRelease);
break;
@@ -594,7 +594,7 @@ void cv::ogl::Buffer::copyFrom(InputArray arr, Target target, bool autoRelease)
break;
}
case _InputArray::GPU_MAT:
case _InputArray::CUDA_GPU_MAT:
{
#ifndef HAVE_CUDA
throw_no_cuda();
@@ -657,7 +657,7 @@ void cv::ogl::Buffer::copyTo(OutputArray arr) const
break;
}
case _InputArray::GPU_MAT:
case _InputArray::CUDA_GPU_MAT:
{
#ifndef HAVE_CUDA
throw_no_cuda();
@@ -1018,7 +1018,7 @@ cv::ogl::Texture2D::Texture2D(InputArray arr, bool autoRelease) : rows_(0), cols
break;
}
case _InputArray::GPU_MAT:
case _InputArray::CUDA_GPU_MAT:
{
#ifndef HAVE_CUDA
throw_no_cuda();
@@ -1132,7 +1132,7 @@ void cv::ogl::Texture2D::copyFrom(InputArray arr, bool autoRelease)
break;
}
case _InputArray::GPU_MAT:
case _InputArray::CUDA_GPU_MAT:
{
#ifndef HAVE_CUDA
throw_no_cuda();
@@ -1184,7 +1184,7 @@ void cv::ogl::Texture2D::copyTo(OutputArray arr, int ddepth, bool autoRelease) c
break;
}
case _InputArray::GPU_MAT:
case _InputArray::CUDA_GPU_MAT:
{
#ifndef HAVE_CUDA
throw_no_cuda();

119
modules/core/test/cuda/test_buffer_pool.cpp Normal file
View File

@@ -0,0 +1,119 @@
/*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 "../test_precomp.hpp"
#ifdef HAVE_CUDA
#include "opencv2/core/cuda.hpp"
#include "opencv2/core/private.cuda.hpp"
#include "opencv2/ts/cuda_test.hpp"
using namespace testing;
using namespace cv;
using namespace cv::cuda;
struct BufferPoolTest : TestWithParam<DeviceInfo>
{
void RunSimpleTest(Stream& stream, HostMem& dst_1, HostMem& dst_2)
{
BufferPool pool(stream);
{
GpuMat buf0 = pool.getBuffer(Size(640, 480), CV_8UC1);
EXPECT_FALSE( buf0.empty() );
buf0.setTo(Scalar::all(0), stream);
GpuMat buf1 = pool.getBuffer(Size(640, 480), CV_8UC1);
EXPECT_FALSE( buf1.empty() );
buf0.convertTo(buf1, buf1.type(), 1.0, 1.0, stream);
buf1.download(dst_1, stream);
}
{
GpuMat buf2 = pool.getBuffer(Size(1280, 1024), CV_32SC1);
EXPECT_FALSE( buf2.empty() );
buf2.setTo(Scalar::all(2), stream);
buf2.download(dst_2, stream);
}
}
void CheckSimpleTest(HostMem& dst_1, HostMem& dst_2)
{
EXPECT_MAT_NEAR(Mat(Size(640, 480), CV_8UC1, Scalar::all(1)), dst_1, 0.0);
EXPECT_MAT_NEAR(Mat(Size(1280, 1024), CV_32SC1, Scalar::all(2)), dst_2, 0.0);
}
};
CUDA_TEST_P(BufferPoolTest, FromNullStream)
{
HostMem dst_1, dst_2;
RunSimpleTest(Stream::Null(), dst_1, dst_2);
CheckSimpleTest(dst_1, dst_2);
}
CUDA_TEST_P(BufferPoolTest, From2Streams)
{
HostMem dst1_1, dst1_2;
HostMem dst2_1, dst2_2;
Stream stream1, stream2;
RunSimpleTest(stream1, dst1_1, dst1_2);
RunSimpleTest(stream2, dst2_1, dst2_2);
stream1.waitForCompletion();
stream2.waitForCompletion();
CheckSimpleTest(dst1_1, dst1_2);
CheckSimpleTest(dst2_1, dst2_2);
}
INSTANTIATE_TEST_CASE_P(CUDA_Stream, BufferPoolTest, ALL_DEVICES);
#endif // HAVE_CUDA

364
modules/core/test/cuda/test_gpumat.cpp Normal file
View File

@@ -0,0 +1,364 @@
/*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 "../test_precomp.hpp"
#ifdef HAVE_CUDA
#include "opencv2/core/cuda.hpp"
#include "opencv2/ts/cuda_test.hpp"
using namespace cvtest;
////////////////////////////////////////////////////////////////////////////////
// SetTo
PARAM_TEST_CASE(GpuMat_SetTo, cv::cuda::DeviceInfo, cv::Size, MatType, UseRoi)
{
cv::cuda::DeviceInfo devInfo;
cv::Size size;
int type;
bool useRoi;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
type = GET_PARAM(2);
useRoi = GET_PARAM(3);
cv::cuda::setDevice(devInfo.deviceID());
}
};
CUDA_TEST_P(GpuMat_SetTo, Zero)
{
cv::Scalar zero = cv::Scalar::all(0);
cv::cuda::GpuMat mat = createMat(size, type, useRoi);
mat.setTo(zero);
EXPECT_MAT_NEAR(cv::Mat::zeros(size, type), mat, 0.0);
}
CUDA_TEST_P(GpuMat_SetTo, SameVal)
{
cv::Scalar val = cv::Scalar::all(randomDouble(0.0, 255.0));
if (CV_MAT_DEPTH(type) == CV_64F && !supportFeature(devInfo, cv::cuda::NATIVE_DOUBLE))
{
try
{
cv::cuda::GpuMat mat = createMat(size, type, useRoi);
mat.setTo(val);
}
catch (const cv::Exception& e)
{
ASSERT_EQ(cv::Error::StsUnsupportedFormat, e.code);
}
}
else
{
cv::cuda::GpuMat mat = createMat(size, type, useRoi);
mat.setTo(val);
EXPECT_MAT_NEAR(cv::Mat(size, type, val), mat, 0.0);
}
}
CUDA_TEST_P(GpuMat_SetTo, DifferentVal)
{
cv::Scalar val = randomScalar(0.0, 255.0);
if (CV_MAT_DEPTH(type) == CV_64F && !supportFeature(devInfo, cv::cuda::NATIVE_DOUBLE))
{
try
{
cv::cuda::GpuMat mat = createMat(size, type, useRoi);
mat.setTo(val);
}
catch (const cv::Exception& e)
{
ASSERT_EQ(cv::Error::StsUnsupportedFormat, e.code);
}
}
else
{
cv::cuda::GpuMat mat = createMat(size, type, useRoi);
mat.setTo(val);
EXPECT_MAT_NEAR(cv::Mat(size, type, val), mat, 0.0);
}
}
CUDA_TEST_P(GpuMat_SetTo, Masked)
{
cv::Scalar val = randomScalar(0.0, 255.0);
cv::Mat mat_gold = randomMat(size, type);
cv::Mat mask = randomMat(size, CV_8UC1, 0.0, 2.0);
if (CV_MAT_DEPTH(type) == CV_64F && !supportFeature(devInfo, cv::cuda::NATIVE_DOUBLE))
{
try
{
cv::cuda::GpuMat mat = createMat(size, type, useRoi);
mat.setTo(val, loadMat(mask));
}
catch (const cv::Exception& e)
{
ASSERT_EQ(cv::Error::StsUnsupportedFormat, e.code);
}
}
else
{
cv::cuda::GpuMat mat = loadMat(mat_gold, useRoi);
mat.setTo(val, loadMat(mask, useRoi));
mat_gold.setTo(val, mask);
EXPECT_MAT_NEAR(mat_gold, mat, 0.0);
}
}
INSTANTIATE_TEST_CASE_P(CUDA, GpuMat_SetTo, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
ALL_TYPES,
WHOLE_SUBMAT));
////////////////////////////////////////////////////////////////////////////////
// CopyTo
PARAM_TEST_CASE(GpuMat_CopyTo, cv::cuda::DeviceInfo, cv::Size, MatType, UseRoi)
{
cv::cuda::DeviceInfo devInfo;
cv::Size size;
int type;
bool useRoi;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
type = GET_PARAM(2);
useRoi = GET_PARAM(3);
cv::cuda::setDevice(devInfo.deviceID());
}
};
CUDA_TEST_P(GpuMat_CopyTo, WithOutMask)
{
cv::Mat src = randomMat(size, type);
cv::cuda::GpuMat d_src = loadMat(src, useRoi);
cv::cuda::GpuMat dst = createMat(size, type, useRoi);
d_src.copyTo(dst);
EXPECT_MAT_NEAR(src, dst, 0.0);
}
CUDA_TEST_P(GpuMat_CopyTo, Masked)
{
cv::Mat src = randomMat(size, type);
cv::Mat mask = randomMat(size, CV_8UC1, 0.0, 2.0);
if (CV_MAT_DEPTH(type) == CV_64F && !supportFeature(devInfo, cv::cuda::NATIVE_DOUBLE))
{
try
{
cv::cuda::GpuMat d_src = loadMat(src);
cv::cuda::GpuMat dst;
d_src.copyTo(dst, loadMat(mask, useRoi));
}
catch (const cv::Exception& e)
{
ASSERT_EQ(cv::Error::StsUnsupportedFormat, e.code);
}
}
else
{
cv::cuda::GpuMat d_src = loadMat(src, useRoi);
cv::cuda::GpuMat dst = loadMat(cv::Mat::zeros(size, type), useRoi);
d_src.copyTo(dst, loadMat(mask, useRoi));
cv::Mat dst_gold = cv::Mat::zeros(size, type);
src.copyTo(dst_gold, mask);
EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
}
}
INSTANTIATE_TEST_CASE_P(CUDA, GpuMat_CopyTo, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
ALL_TYPES,
WHOLE_SUBMAT));
////////////////////////////////////////////////////////////////////////////////
// ConvertTo
PARAM_TEST_CASE(GpuMat_ConvertTo, cv::cuda::DeviceInfo, cv::Size, MatDepth, MatDepth, UseRoi)
{
cv::cuda::DeviceInfo devInfo;
cv::Size size;
int depth1;
int depth2;
bool useRoi;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
depth1 = GET_PARAM(2);
depth2 = GET_PARAM(3);
useRoi = GET_PARAM(4);
cv::cuda::setDevice(devInfo.deviceID());
}
};
CUDA_TEST_P(GpuMat_ConvertTo, WithOutScaling)
{
cv::Mat src = randomMat(size, depth1);
if ((depth1 == CV_64F || depth2 == CV_64F) && !supportFeature(devInfo, cv::cuda::NATIVE_DOUBLE))
{
try
{
cv::cuda::GpuMat d_src = loadMat(src);
cv::cuda::GpuMat dst;
d_src.convertTo(dst, depth2);
}
catch (const cv::Exception& e)
{
ASSERT_EQ(cv::Error::StsUnsupportedFormat, e.code);
}
}
else
{
cv::cuda::GpuMat d_src = loadMat(src, useRoi);
cv::cuda::GpuMat dst = createMat(size, depth2, useRoi);
d_src.convertTo(dst, depth2);
cv::Mat dst_gold;
src.convertTo(dst_gold, depth2);
EXPECT_MAT_NEAR(dst_gold, dst, depth2 < CV_32F ? 1.0 : 1e-4);
}
}
CUDA_TEST_P(GpuMat_ConvertTo, WithScaling)
{
cv::Mat src = randomMat(size, depth1);
double a = randomDouble(0.0, 1.0);
double b = randomDouble(-10.0, 10.0);
if ((depth1 == CV_64F || depth2 == CV_64F) && !supportFeature(devInfo, cv::cuda::NATIVE_DOUBLE))
{
try
{
cv::cuda::GpuMat d_src = loadMat(src);
cv::cuda::GpuMat dst;
d_src.convertTo(dst, depth2, a, b);
}
catch (const cv::Exception& e)
{
ASSERT_EQ(cv::Error::StsUnsupportedFormat, e.code);
}
}
else
{
cv::cuda::GpuMat d_src = loadMat(src, useRoi);
cv::cuda::GpuMat dst = createMat(size, depth2, useRoi);
d_src.convertTo(dst, depth2, a, b);
cv::Mat dst_gold;
src.convertTo(dst_gold, depth2, a, b);
EXPECT_MAT_NEAR(dst_gold, dst, depth2 < CV_32F ? 1.0 : 1e-4);
}
}
INSTANTIATE_TEST_CASE_P(CUDA, GpuMat_ConvertTo, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
ALL_DEPTH,
ALL_DEPTH,
WHOLE_SUBMAT));
////////////////////////////////////////////////////////////////////////////////
// ensureSizeIsEnough
struct EnsureSizeIsEnough : testing::TestWithParam<cv::cuda::DeviceInfo>
{
virtual void SetUp()
{
cv::cuda::DeviceInfo devInfo = GetParam();
cv::cuda::setDevice(devInfo.deviceID());
}
};
CUDA_TEST_P(EnsureSizeIsEnough, BufferReuse)
{
cv::cuda::GpuMat buffer(100, 100, CV_8U);
cv::cuda::GpuMat old = buffer;
// don't reallocate memory
cv::cuda::ensureSizeIsEnough(10, 20, CV_8U, buffer);
EXPECT_EQ(10, buffer.rows);
EXPECT_EQ(20, buffer.cols);
EXPECT_EQ(CV_8UC1, buffer.type());
EXPECT_EQ(reinterpret_cast<intptr_t>(old.data), reinterpret_cast<intptr_t>(buffer.data));
// don't reallocate memory
cv::cuda::ensureSizeIsEnough(20, 30, CV_8U, buffer);
EXPECT_EQ(20, buffer.rows);
EXPECT_EQ(30, buffer.cols);
EXPECT_EQ(CV_8UC1, buffer.type());
EXPECT_EQ(reinterpret_cast<intptr_t>(old.data), reinterpret_cast<intptr_t>(buffer.data));
}
INSTANTIATE_TEST_CASE_P(CUDA, EnsureSizeIsEnough, ALL_DEVICES);
#endif // HAVE_CUDA

456
modules/core/test/cuda/test_opengl.cpp Normal file
View File

@@ -0,0 +1,456 @@
/*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 "../test_precomp.hpp"
#if defined(HAVE_CUDA) && defined(HAVE_OPENGL)
#include "opencv2/core/cuda.hpp"
#include "opencv2/core/opengl.hpp"
#include "opencv2/ts/cuda_test.hpp"
using namespace cvtest;
/////////////////////////////////////////////
// Buffer
PARAM_TEST_CASE(Buffer, cv::Size, MatType)
{
static void SetUpTestCase()
{
cv::namedWindow("test", cv::WINDOW_OPENGL);
}
static void TearDownTestCase()
{
cv::destroyAllWindows();
}
cv::Size size;
int type;
virtual void SetUp()
{
size = GET_PARAM(0);
type = GET_PARAM(1);
}
};
CUDA_TEST_P(Buffer, Constructor1)
{
cv::ogl::Buffer buf(size.height, size.width, type, cv::ogl::Buffer::ARRAY_BUFFER, true);
EXPECT_EQ(size.height, buf.rows());
EXPECT_EQ(size.width, buf.cols());
EXPECT_EQ(type, buf.type());
}
CUDA_TEST_P(Buffer, Constructor2)
{
cv::ogl::Buffer buf(size, type, cv::ogl::Buffer::ARRAY_BUFFER, true);
EXPECT_EQ(size.height, buf.rows());
EXPECT_EQ(size.width, buf.cols());
EXPECT_EQ(type, buf.type());
}
CUDA_TEST_P(Buffer, ConstructorFromMat)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::Mat bufData;
buf.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 0);
}
CUDA_TEST_P(Buffer, ConstructorFromGpuMat)
{
cv::Mat gold = randomMat(size, type);
cv::cuda::GpuMat d_gold(gold);
cv::ogl::Buffer buf(d_gold, cv::ogl::Buffer::ARRAY_BUFFER);
cv::Mat bufData;
buf.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 0);
}
CUDA_TEST_P(Buffer, ConstructorFromBuffer)
{
cv::ogl::Buffer buf_gold(size, type, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::ogl::Buffer buf(buf_gold);
EXPECT_EQ(buf_gold.bufId(), buf.bufId());
EXPECT_EQ(buf_gold.rows(), buf.rows());
EXPECT_EQ(buf_gold.cols(), buf.cols());
EXPECT_EQ(buf_gold.type(), buf.type());
}
CUDA_TEST_P(Buffer, Create)
{
cv::ogl::Buffer buf;
buf.create(size.height, size.width, type, cv::ogl::Buffer::ARRAY_BUFFER, true);
EXPECT_EQ(size.height, buf.rows());
EXPECT_EQ(size.width, buf.cols());
EXPECT_EQ(type, buf.type());
}
CUDA_TEST_P(Buffer, CopyFromMat)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf;
buf.copyFrom(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::Mat bufData;
buf.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 0);
}
CUDA_TEST_P(Buffer, CopyFromGpuMat)
{
cv::Mat gold = randomMat(size, type);
cv::cuda::GpuMat d_gold(gold);
cv::ogl::Buffer buf;
buf.copyFrom(d_gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::Mat bufData;
buf.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 0);
}
CUDA_TEST_P(Buffer, CopyFromBuffer)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf_gold(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::ogl::Buffer buf;
buf.copyFrom(buf_gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
EXPECT_NE(buf_gold.bufId(), buf.bufId());
cv::Mat bufData;
buf.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 0);
}
CUDA_TEST_P(Buffer, CopyToGpuMat)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::cuda::GpuMat dst;
buf.copyTo(dst);
EXPECT_MAT_NEAR(gold, dst, 0);
}
CUDA_TEST_P(Buffer, CopyToBuffer)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::ogl::Buffer dst;
buf.copyTo(dst);
dst.setAutoRelease(true);
EXPECT_NE(buf.bufId(), dst.bufId());
cv::Mat bufData;
dst.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 0);
}
CUDA_TEST_P(Buffer, Clone)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::ogl::Buffer dst = buf.clone(cv::ogl::Buffer::ARRAY_BUFFER, true);
EXPECT_NE(buf.bufId(), dst.bufId());
cv::Mat bufData;
dst.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 0);
}
CUDA_TEST_P(Buffer, MapHostRead)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::Mat dst = buf.mapHost(cv::ogl::Buffer::READ_ONLY);
EXPECT_MAT_NEAR(gold, dst, 0);
buf.unmapHost();
}
CUDA_TEST_P(Buffer, MapHostWrite)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf(size, type, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::Mat dst = buf.mapHost(cv::ogl::Buffer::WRITE_ONLY);
gold.copyTo(dst);
buf.unmapHost();
dst.release();
cv::Mat bufData;
buf.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 0);
}
CUDA_TEST_P(Buffer, MapDevice)
{
cv::Mat gold = randomMat(size, type);
cv::ogl::Buffer buf(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
cv::cuda::GpuMat dst = buf.mapDevice();
EXPECT_MAT_NEAR(gold, dst, 0);
buf.unmapDevice();
}
INSTANTIATE_TEST_CASE_P(OpenGL, Buffer, testing::Combine(DIFFERENT_SIZES, ALL_TYPES));
/////////////////////////////////////////////
// Texture2D
PARAM_TEST_CASE(Texture2D, cv::Size, MatType)
{
static void SetUpTestCase()
{
cv::namedWindow("test", cv::WINDOW_OPENGL);
}
static void TearDownTestCase()
{
cv::destroyAllWindows();
}
cv::Size size;
int type;
int depth;
int cn;
cv::ogl::Texture2D::Format format;
virtual void SetUp()
{
size = GET_PARAM(0);
type = GET_PARAM(1);
depth = CV_MAT_DEPTH(type);
cn = CV_MAT_CN(type);
format = cn == 1 ? cv::ogl::Texture2D::DEPTH_COMPONENT : cn == 3 ? cv::ogl::Texture2D::RGB : cn == 4 ? cv::ogl::Texture2D::RGBA : cv::ogl::Texture2D::NONE;
}
};
CUDA_TEST_P(Texture2D, Constructor1)
{
cv::ogl::Texture2D tex(size.height, size.width, format, true);
EXPECT_EQ(size.height, tex.rows());
EXPECT_EQ(size.width, tex.cols());
EXPECT_EQ(format, tex.format());
}
CUDA_TEST_P(Texture2D, Constructor2)
{
cv::ogl::Texture2D tex(size, format, true);
EXPECT_EQ(size.height, tex.rows());
EXPECT_EQ(size.width, tex.cols());
EXPECT_EQ(format, tex.format());
}
CUDA_TEST_P(Texture2D, ConstructorFromMat)
{
cv::Mat gold = randomMat(size, type, 0, depth == CV_8U ? 255 : 1);
cv::ogl::Texture2D tex(gold, true);
cv::Mat texData;
tex.copyTo(texData, depth);
EXPECT_MAT_NEAR(gold, texData, 1e-2);
}
CUDA_TEST_P(Texture2D, ConstructorFromGpuMat)
{
cv::Mat gold = randomMat(size, type, 0, depth == CV_8U ? 255 : 1);
cv::cuda::GpuMat d_gold(gold);
cv::ogl::Texture2D tex(d_gold, true);
cv::Mat texData;
tex.copyTo(texData, depth);
EXPECT_MAT_NEAR(gold, texData, 1e-2);
}
CUDA_TEST_P(Texture2D, ConstructorFromBuffer)
{
cv::Mat gold = randomMat(size, type, 0, depth == CV_8U ? 255 : 1);
cv::ogl::Buffer buf_gold(gold, cv::ogl::Buffer::PIXEL_UNPACK_BUFFER, true);
cv::ogl::Texture2D tex(buf_gold, true);
cv::Mat texData;
tex.copyTo(texData, depth);
EXPECT_MAT_NEAR(gold, texData, 1e-2);
}
CUDA_TEST_P(Texture2D, ConstructorFromTexture2D)
{
cv::ogl::Texture2D tex_gold(size, format, true);
cv::ogl::Texture2D tex(tex_gold);
EXPECT_EQ(tex_gold.texId(), tex.texId());
EXPECT_EQ(tex_gold.rows(), tex.rows());
EXPECT_EQ(tex_gold.cols(), tex.cols());
EXPECT_EQ(tex_gold.format(), tex.format());
}
CUDA_TEST_P(Texture2D, Create)
{
cv::ogl::Texture2D tex;
tex.create(size.height, size.width, format, true);
EXPECT_EQ(size.height, tex.rows());
EXPECT_EQ(size.width, tex.cols());
EXPECT_EQ(format, tex.format());
}
CUDA_TEST_P(Texture2D, CopyFromMat)
{
cv::Mat gold = randomMat(size, type, 0, depth == CV_8U ? 255 : 1);
cv::ogl::Texture2D tex;
tex.copyFrom(gold, true);
cv::Mat texData;
tex.copyTo(texData, depth);
EXPECT_MAT_NEAR(gold, texData, 1e-2);
}
CUDA_TEST_P(Texture2D, CopyFromGpuMat)
{
cv::Mat gold = randomMat(size, type, 0, depth == CV_8U ? 255 : 1);
cv::cuda::GpuMat d_gold(gold);
cv::ogl::Texture2D tex;
tex.copyFrom(d_gold, true);
cv::Mat texData;
tex.copyTo(texData, depth);
EXPECT_MAT_NEAR(gold, texData, 1e-2);
}
CUDA_TEST_P(Texture2D, CopyFromBuffer)
{
cv::Mat gold = randomMat(size, type, 0, depth == CV_8U ? 255 : 1);
cv::ogl::Buffer buf_gold(gold, cv::ogl::Buffer::PIXEL_UNPACK_BUFFER, true);
cv::ogl::Texture2D tex;
tex.copyFrom(buf_gold, true);
cv::Mat texData;
tex.copyTo(texData, depth);
EXPECT_MAT_NEAR(gold, texData, 1e-2);
}
CUDA_TEST_P(Texture2D, CopyToGpuMat)
{
cv::Mat gold = randomMat(size, type, 0, depth == CV_8U ? 255 : 1);
cv::ogl::Texture2D tex(gold, true);
cv::cuda::GpuMat dst;
tex.copyTo(dst, depth);
EXPECT_MAT_NEAR(gold, dst, 1e-2);
}
CUDA_TEST_P(Texture2D, CopyToBuffer)
{
cv::Mat gold = randomMat(size, type, 0, depth == CV_8U ? 255 : 1);
cv::ogl::Texture2D tex(gold, true);
cv::ogl::Buffer dst;
tex.copyTo(dst, depth, true);
cv::Mat bufData;
dst.copyTo(bufData);
EXPECT_MAT_NEAR(gold, bufData, 1e-2);
}
INSTANTIATE_TEST_CASE_P(OpenGL, Texture2D, testing::Combine(DIFFERENT_SIZES, testing::Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC3, CV_32FC4)));
#endif

153
modules/core/test/cuda/test_stream.cpp Normal file
View File

@@ -0,0 +1,153 @@
/*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 "../test_precomp.hpp"
#ifdef HAVE_CUDA
#include <cuda_runtime.h>
#include "opencv2/core/cuda.hpp"
#include "opencv2/ts/cuda_test.hpp"
using namespace cvtest;
struct Async : testing::TestWithParam<cv::cuda::DeviceInfo>
{
cv::cuda::HostMem src;
cv::cuda::GpuMat d_src;
cv::cuda::HostMem dst;
cv::cuda::GpuMat d_dst;
virtual void SetUp()
{
cv::cuda::DeviceInfo devInfo = GetParam();
cv::cuda::setDevice(devInfo.deviceID());
src = cv::cuda::HostMem(cv::cuda::HostMem::PAGE_LOCKED);
cv::Mat m = randomMat(cv::Size(128, 128), CV_8UC1);
m.copyTo(src);
}
};
void checkMemSet(int status, void* userData)
{
ASSERT_EQ(cudaSuccess, status);
Async* test = reinterpret_cast<Async*>(userData);
cv::cuda::HostMem src = test->src;
cv::cuda::HostMem dst = test->dst;
cv::Mat dst_gold = cv::Mat::zeros(src.size(), src.type());
ASSERT_MAT_NEAR(dst_gold, dst, 0);
}
CUDA_TEST_P(Async, MemSet)
{
cv::cuda::Stream stream;
d_dst.upload(src);
d_dst.setTo(cv::Scalar::all(0), stream);
d_dst.download(dst, stream);
Async* test = this;
stream.enqueueHostCallback(checkMemSet, test);
stream.waitForCompletion();
}
void checkConvert(int status, void* userData)
{
ASSERT_EQ(cudaSuccess, status);
Async* test = reinterpret_cast<Async*>(userData);
cv::cuda::HostMem src = test->src;
cv::cuda::HostMem dst = test->dst;
cv::Mat dst_gold;
src.createMatHeader().convertTo(dst_gold, CV_32S);
ASSERT_MAT_NEAR(dst_gold, dst, 0);
}
CUDA_TEST_P(Async, Convert)
{
cv::cuda::Stream stream;
d_src.upload(src, stream);
d_src.convertTo(d_dst, CV_32S, stream);
d_dst.download(dst, stream);
Async* test = this;
stream.enqueueHostCallback(checkConvert, test);
stream.waitForCompletion();
}
CUDA_TEST_P(Async, HostMemAllocator)
{
cv::cuda::Stream stream;
cv::Mat h_dst;
h_dst.allocator = cv::cuda::HostMem::getAllocator();
d_src.upload(src, stream);
d_src.convertTo(d_dst, CV_32S, stream);
d_dst.download(h_dst, stream);
stream.waitForCompletion();
cv::Mat dst_gold;
src.createMatHeader().convertTo(dst_gold, CV_32S);
ASSERT_MAT_NEAR(dst_gold, h_dst, 0);
}
INSTANTIATE_TEST_CASE_P(CUDA_Stream, Async, ALL_DEVICES);
#endif // HAVE_CUDA

10
modules/core/test/test_main.cpp
View File

@@ -7,4 +7,14 @@
#include "test_precomp.hpp"
#ifndef HAVE_CUDA
CV_TEST_MAIN("cv")
#else
#include "opencv2/ts/cuda_test.hpp"
CV_CUDA_TEST_MAIN("cv")
#endif