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

updated patch to bring in the first functions with "transparent API"

Browse Source
This commit is contained in:
Vadim Pisarevsky
2013-11-18 11:48:00 -05:00
parent bb4bf7a1f9
commit d914f20a4c
64 changed files with 13355 additions and 318 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
modules/core/include/opencv2/core.hpp
View File

@@ -347,6 +347,10 @@ CV_EXPORTS_W void max(InputArray src1, InputArray src2, OutputArray dst);
CV_EXPORTS void min(const Mat& src1, const Mat& src2, Mat& dst);
//! computes per-element maximum of two arrays (dst = max(src1, src2))
CV_EXPORTS void max(const Mat& src1, const Mat& src2, Mat& dst);
//! computes per-element minimum of two arrays (dst = min(src1, src2))
CV_EXPORTS void min(const UMat& src1, const UMat& src2, UMat& dst);
//! computes per-element maximum of two arrays (dst = max(src1, src2))
CV_EXPORTS void max(const UMat& src1, const UMat& src2, UMat& dst);
//! computes square root of each matrix element (dst = src**0.5)
CV_EXPORTS_W void sqrt(InputArray src, OutputArray dst);

14
modules/core/include/opencv2/core/mat.hpp
View File

@@ -58,6 +58,8 @@ namespace cv
enum { ACCESS_READ=1<<24, ACCESS_WRITE=1<<25,
ACCESS_RW=3<<24, ACCESS_MASK=ACCESS_RW, ACCESS_FAST=1<<26 };
class CV_EXPORTS _OutputArray;
//////////////////////// Input/Output Array Arguments /////////////////////////////////
/*!
@@ -116,12 +118,22 @@ public:
void* getObj() const;
virtual int kind() const;
virtual int dims(int i=-1) const;
virtual Size size(int i=-1) const;
virtual int sizend(int* sz, int i=-1) const;
virtual bool sameSize(const _InputArray& arr) const;
virtual size_t total(int i=-1) const;
virtual int type(int i=-1) const;
virtual int depth(int i=-1) const;
virtual int channels(int i=-1) const;
virtual bool isContinuous(int i=-1) const;
virtual bool empty() const;
virtual void copyTo(const _OutputArray& arr) const;
bool isMat() const;
bool isUMat() const;
bool isMatVectot() const;
bool isUMatVector() const;
bool isMatx();
virtual ~_InputArray();
@@ -197,8 +209,10 @@ public:
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;
virtual void createSameSize(const _InputArray& arr, int mtype) const;
virtual void release() const;
virtual void clear() const;
virtual void setTo(const _InputArray& value) const;
};

6
modules/core/include/opencv2/core/mat.inl.hpp
View File

@@ -108,6 +108,12 @@ inline _InputArray::_InputArray(const cuda::CudaMem& cuda_mem)
inline _InputArray::~_InputArray() {}
inline bool _InputArray::isMat() const { return kind() == _InputArray::MAT; }
inline bool _InputArray::isUMat() const { return kind() == _InputArray::UMAT; }
inline bool _InputArray::isMatVectot() const { return kind() == _InputArray::STD_VECTOR_MAT; }
inline bool _InputArray::isUMatVector() const { return kind() == _InputArray::STD_VECTOR_UMAT; }
inline bool _InputArray::isMatx() { return kind() == _InputArray::MATX; }
////////////////////////////////////////////////////////////////////////////////////////
inline _OutputArray::_OutputArray() { init(ACCESS_WRITE, 0); }

133
modules/core/include/opencv2/core/ocl.hpp
View File

@@ -49,13 +49,13 @@ namespace cv { namespace ocl {
CV_EXPORTS bool haveOpenCL();
CV_EXPORTS bool useOpenCL();
CV_EXPORTS void setUseOpenCL(bool flag);
CV_EXPORTS void finish();
CV_EXPORTS void finish2();
class CV_EXPORTS Context;
class CV_EXPORTS Context2;
class CV_EXPORTS Device;
class CV_EXPORTS Kernel;
class CV_EXPORTS Program;
class CV_EXPORTS ProgramSource;
class CV_EXPORTS ProgramSource2;
class CV_EXPORTS Queue;
class CV_EXPORTS Device
@@ -199,22 +199,22 @@ protected:
};
class CV_EXPORTS Context
class CV_EXPORTS Context2
{
public:
Context();
explicit Context(int dtype);
~Context();
Context(const Context& c);
Context& operator = (const Context& c);
Context2();
explicit Context2(int dtype);
~Context2();
Context2(const Context2& c);
Context2& operator = (const Context2& c);
bool create(int dtype);
size_t ndevices() const;
const Device& device(size_t idx) const;
Program getProg(const ProgramSource& prog,
Program getProg(const ProgramSource2& prog,
const String& buildopt, String& errmsg);
static Context& getDefault();
static Context2& getDefault();
void* ptr() const;
protected:
struct Impl;
@@ -226,12 +226,12 @@ class CV_EXPORTS Queue
{
public:
Queue();
explicit Queue(const Context& c, const Device& d=Device());
explicit Queue(const Context2& c, const Device& d=Device());
~Queue();
Queue(const Queue& q);
Queue& operator = (const Queue& q);
bool create(const Context& c=Context(), const Device& d=Device());
bool create(const Context2& c=Context2(), const Device& d=Device());
void finish();
void* ptr() const;
static Queue& getDefault();
@@ -245,41 +245,55 @@ protected:
class CV_EXPORTS KernelArg
{
public:
enum { LOCAL=1, READ_ONLY=2, WRITE_ONLY=4, READ_WRITE=6, CONSTANT=8 };
KernelArg(int _flags, UMat* _m, void* _obj=0, size_t _sz=0);
enum { LOCAL=1, READ_ONLY=2, WRITE_ONLY=4, READ_WRITE=6, CONSTANT=8, NO_SIZE=256 };
KernelArg(int _flags, UMat* _m, int wscale=1, const void* _obj=0, size_t _sz=0);
KernelArg();
static KernelArg Local() { return KernelArg(LOCAL, 0); }
static KernelArg ReadOnly(const UMat& m) { return KernelArg(READ_ONLY, (UMat*)&m); }
static KernelArg WriteOnly(const UMat& m) { return KernelArg(WRITE_ONLY, (UMat*)&m); }
static KernelArg ReadWrite(const UMat& m, int wscale=1)
{ return KernelArg(READ_WRITE, (UMat*)&m, wscale); }
static KernelArg ReadWriteNoSize(const UMat& m, int wscale=1)
{ return KernelArg(READ_WRITE+NO_SIZE, (UMat*)&m, wscale); }
static KernelArg ReadOnly(const UMat& m, int wscale=1)
{ return KernelArg(READ_ONLY, (UMat*)&m, wscale); }
static KernelArg WriteOnly(const UMat& m, int wscale=1)
{ return KernelArg(WRITE_ONLY, (UMat*)&m, wscale); }
static KernelArg ReadOnlyNoSize(const UMat& m, int wscale=1)
{ return KernelArg(READ_ONLY+NO_SIZE, (UMat*)&m, wscale); }
static KernelArg WriteOnlyNoSize(const UMat& m, int wscale=1)
{ return KernelArg(WRITE_ONLY+NO_SIZE, (UMat*)&m, wscale); }
static KernelArg Constant(const Mat& m);
template<typename _Tp> static KernelArg Constant(const _Tp* arr, size_t n)
{ return KernelArg(CONSTANT, 0, (void*)arr, n); }
{ return KernelArg(CONSTANT, 0, 1, (void*)arr, n); }
int flags;
UMat* m;
void* obj;
const void* obj;
size_t sz;
int wscale;
};
class CV_EXPORTS Kernel
{
public:
Kernel();
Kernel(const char* kname, const Program& prog);
Kernel(const char* kname, const ProgramSource& prog,
const String& buildopts, String& errmsg);
Kernel(const char* kname, const ProgramSource2& prog,
const String& buildopts, String* errmsg=0);
~Kernel();
Kernel(const Kernel& k);
Kernel& operator = (const Kernel& k);
bool empty() const;
bool create(const char* kname, const Program& prog);
bool create(const char* kname, const ProgramSource& prog,
const String& buildopts, String& errmsg);
bool create(const char* kname, const ProgramSource2& prog,
const String& buildopts, String* errmsg=0);
void set(int i, const void* value, size_t sz);
void set(int i, const UMat& m);
void set(int i, const KernelArg& arg);
template<typename _Tp> void set(int i, const _Tp& value)
int set(int i, const void* value, size_t sz);
int set(int i, const UMat& m);
int set(int i, const KernelArg& arg);
template<typename _Tp> int set(int i, const _Tp& value)
{ return set(i, &value, sizeof(value)); }
template<typename _Tp0>
@@ -291,26 +305,27 @@ public:
template<typename _Tp0, typename _Tp1>
Kernel& args(const _Tp0& a0, const _Tp1& a1)
{
set(0, a0); set(1, a1); return *this;
int i = set(0, a0); set(i, a1); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2>
Kernel& args(const _Tp0& a0, const _Tp1& a1, const _Tp2& a2)
{
set(0, a0); set(1, a1); set(2, a2); return *this;
int i = set(0, a0); i = set(i, a1); set(i, a2); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2, typename _Tp3>
Kernel& args(const _Tp0& a0, const _Tp1& a1, const _Tp2& a2, const _Tp3& a3)
{
set(0, a0); set(1, a1); set(2, a2); set(3, a3); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2); i = set(i, a3); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2, typename _Tp3, typename _Tp4>
Kernel& args(const _Tp0& a0, const _Tp1& a1, const _Tp2& a2,
const _Tp3& a3, const _Tp4& a4)
{
set(0, a0); set(1, a1); set(2, a2); set(3, a3); set(4, a4); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2);
i = set(i, a3); set(i, a4); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2,
@@ -318,8 +333,8 @@ public:
Kernel& args(const _Tp0& a0, const _Tp1& a1, const _Tp2& a2,
const _Tp3& a3, const _Tp4& a4, const _Tp5& a5)
{
set(0, a0); set(1, a1); set(2, a2);
set(3, a3); set(4, a4); set(5, a5); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2);
i = set(i, a3); i = set(i, a4); set(i, a5); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2, typename _Tp3,
@@ -327,8 +342,8 @@ public:
Kernel& args(const _Tp0& a0, const _Tp1& a1, const _Tp2& a2, const _Tp3& a3,
const _Tp4& a4, const _Tp5& a5, const _Tp6& a6)
{
set(0, a0); set(1, a1); set(2, a2); set(3, a3);
set(4, a4); set(5, a5); set(6, a6); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2); i = set(i, a3);
i = set(i, a4); i = set(i, a5); set(i, a6); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2, typename _Tp3,
@@ -336,8 +351,8 @@ public:
Kernel& args(const _Tp0& a0, const _Tp1& a1, const _Tp2& a2, const _Tp3& a3,
const _Tp4& a4, const _Tp5& a5, const _Tp6& a6, const _Tp7& a7)
{
set(0, a0); set(1, a1); set(2, a2); set(3, a3);
set(4, a4); set(5, a5); set(6, a6); set(7, a7); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2); i = set(i, a3);
i = set(i, a4); i = set(i, a5); i = set(i, a6); set(i, a7); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2, typename _Tp3, typename _Tp4,
@@ -346,8 +361,8 @@ public:
const _Tp4& a4, const _Tp5& a5, const _Tp6& a6, const _Tp7& a7,
const _Tp8& a8)
{
set(0, a0); set(1, a1); set(2, a2); set(3, a3); set(4, a4);
set(5, a5); set(6, a6); set(7, a7); set(8, a8); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2); i = set(i, a3); i = set(i, a4);
i = set(i, a5); i = set(i, a6); i = set(i, a7); set(i, a8); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2, typename _Tp3, typename _Tp4,
@@ -356,8 +371,8 @@ public:
const _Tp4& a4, const _Tp5& a5, const _Tp6& a6, const _Tp7& a7,
const _Tp8& a8, const _Tp9& a9)
{
set(0, a0); set(1, a1); set(2, a2); set(3, a3); set(4, a4); set(5, a5);
set(6, a6); set(7, a7); set(8, a8); set(9, a9); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2); i = set(i, a3); i = set(i, a4); i = set(i, a5);
i = set(i, a6); i = set(i, a7); i = set(i, a8); set(i, a9); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2, typename _Tp3,
@@ -367,8 +382,8 @@ public:
const _Tp4& a4, const _Tp5& a5, const _Tp6& a6, const _Tp7& a7,
const _Tp8& a8, const _Tp9& a9, const _Tp10& a10)
{
set(0, a0); set(1, a1); set(2, a2); set(3, a3); set(4, a4); set(5, a5);
set(6, a6); set(7, a7); set(8, a8); set(9, a9); set(10, a10); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2); i = set(i, a3); i = set(i, a4); i = set(i, a5);
i = set(i, a6); i = set(i, a7); i = set(i, a8); i = set(i, a9); set(i, a10); return *this;
}
template<typename _Tp0, typename _Tp1, typename _Tp2, typename _Tp3,
@@ -378,13 +393,13 @@ public:
const _Tp4& a4, const _Tp5& a5, const _Tp6& a6, const _Tp7& a7,
const _Tp8& a8, const _Tp9& a9, const _Tp10& a10, const _Tp11& a11)
{
set(0, a0); set(1, a1); set(2, a2); set(3, a3); set(4, a4); set(5, a5);
set(6, a6); set(7, a7); set(8, a8); set(9, a9); set(10, a10); set(11, a11); return *this;
int i = set(0, a0); i = set(i, a1); i = set(i, a2); i = set(i, a3); i = set(i, a4); i = set(i, a5);
i = set(i, a6); i = set(i, a7); i = set(i, a8); i = set(i, a9); i = set(i, a10); set(i, a11); return *this;
}
void run(int dims, size_t offset[], size_t globalsize[],
bool run(int dims, size_t globalsize[],
size_t localsize[], bool sync, const Queue& q=Queue());
void runTask(bool sync, const Queue& q=Queue());
bool runTask(bool sync, const Queue& q=Queue());
size_t workGroupSize() const;
bool compileWorkGroupSize(size_t wsz[]) const;
@@ -401,7 +416,7 @@ class CV_EXPORTS Program
{
public:
Program();
Program(const ProgramSource& src,
Program(const ProgramSource2& src,
const String& buildflags, String& errmsg);
explicit Program(const String& buf);
Program(const Program& prog);
@@ -409,12 +424,12 @@ public:
Program& operator = (const Program& prog);
~Program();
bool create(const ProgramSource& src,
bool create(const ProgramSource2& src,
const String& buildflags, String& errmsg);
bool read(const String& buf, const String& buildflags);
bool write(String& buf) const;
const ProgramSource& source() const;
const ProgramSource2& source() const;
void* ptr() const;
String getPrefix() const;
@@ -426,17 +441,17 @@ protected:
};
class CV_EXPORTS ProgramSource
class CV_EXPORTS ProgramSource2
{
public:
typedef uint64 hash_t;
ProgramSource();
explicit ProgramSource(const String& prog);
explicit ProgramSource(const char* prog);
~ProgramSource();
ProgramSource(const ProgramSource& prog);
ProgramSource& operator = (const ProgramSource& prog);
ProgramSource2();
explicit ProgramSource2(const String& prog);
explicit ProgramSource2(const char* prog);
~ProgramSource2();
ProgramSource2(const ProgramSource2& prog);
ProgramSource2& operator = (const ProgramSource2& prog);
const String& source() const;
hash_t hash() const;
@@ -446,6 +461,10 @@ protected:
Impl* p;
};
CV_EXPORTS const char* convertTypeStr(int sdepth, int ddepth, int cn, char* buf);
CV_EXPORTS const char* typeToStr(int t);
CV_EXPORTS const char* memopTypeToStr(int t);
}}
#endif

60
modules/core/include/opencv2/core/ocl_genbase.hpp Normal file
View File

@@ -0,0 +1,60 @@
/*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) 2013, OpenCV Foundation, 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 OpenCV Foundation 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_OPENCL_GENBASE_HPP__
#define __OPENCV_OPENCL_GENBASE_HPP__
namespace cv
{
namespace ocl
{
struct ProgramEntry
{
const char* name;
const char* programStr;
const char* programHash;
};
}
}
#endif

461
modules/core/src/arithm.cpp
View File

@@ -911,33 +911,112 @@ void convertAndUnrollScalar( const Mat& sc, int buftype, uchar* scbuf, size_t bl
scbuf[i] = scbuf[i - esz];
}
static void binary_op(InputArray _src1, InputArray _src2, OutputArray _dst,
InputArray _mask, const BinaryFunc* tab, bool bitwise)
enum { OCL_OP_ADD=0, OCL_OP_SUB=1, OCL_OP_RSUB=2, OCL_OP_ABSDIFF=3, OCL_OP_MUL=4,
OCL_OP_MUL_SCALE=5, OCL_OP_DIV_SCALE=6, OCL_OP_RECIP_SCALE=7, OCL_OP_ADDW=8,
OCL_OP_AND=9, OCL_OP_OR=10, OCL_OP_XOR=11, OCL_OP_NOT=12, OCL_OP_MIN=13, OCL_OP_MAX=14 };
static const char* oclop2str[] = { "OP_ADD", "OP_SUB", "OP_RSUB", "OP_ABSDIFF",
"OP_MUL", "OP_MUL_SCALE", "OP_DIV_SCALE", "OP_RECIP_SCALE",
"OP_ADDW", "OP_AND", "OP_OR", "OP_XOR", "OP_NOT", "OP_MIN", "OP_MAX", 0 };
static bool ocl_binary_op(InputArray _src1, InputArray _src2, OutputArray _dst,
InputArray _mask, bool bitwise, int oclop, bool haveScalar )
{
int kind1 = _src1.kind(), kind2 = _src2.kind();
Mat src1 = _src1.getMat(), src2 = _src2.getMat();
bool haveMask = !_mask.empty();
int srctype = _src1.type();
int srcdepth = CV_MAT_DEPTH(srctype);
int cn = CV_MAT_CN(srctype);
if( oclop < 0 || ((haveMask || haveScalar) && cn > 4) )
return false;
UMat src1 = _src1.getUMat(), src2;
UMat dst = _dst.getUMat(), mask = _mask.getUMat();
char opts[1024];
int kercn = haveMask || haveScalar ? cn : 1;
sprintf(opts, "-D %s%s -D %s -D dstT=%s",
(haveMask ? "MASK_" : ""), (haveScalar ? "UNARY_OP" : "BINARY_OP"), oclop2str[oclop],
bitwise ? ocl::memopTypeToStr(CV_MAKETYPE(srcdepth, kercn)) :
ocl::typeToStr(CV_MAKETYPE(srcdepth, kercn)));
ocl::Kernel k("KF", ocl::core::arithm_oclsrc, opts);
if( k.empty() )
return false;
int cscale = cn/kercn;
ocl::KernelArg src1arg = ocl::KernelArg::ReadOnlyNoSize(src1, cscale);
ocl::KernelArg dstarg = haveMask ? ocl::KernelArg::ReadWrite(dst, cscale) :
ocl::KernelArg::WriteOnly(dst, cscale);
ocl::KernelArg maskarg = ocl::KernelArg::ReadOnlyNoSize(mask, 1);
if( haveScalar )
{
size_t esz = CV_ELEM_SIZE(srctype);
double buf[4] = {0,0,0,0};
if( oclop != OCL_OP_NOT )
{
Mat src2sc = _src2.getMat();
convertAndUnrollScalar(src2sc, srctype, (uchar*)buf, 1);
}
ocl::KernelArg scalararg = ocl::KernelArg(0, 0, 0, buf, esz);
if( !haveMask )
k.args(src1arg, dstarg, scalararg);
else
k.args(src1arg, maskarg, dstarg, scalararg);
}
else
{
src2 = _src2.getUMat();
ocl::KernelArg src2arg = ocl::KernelArg::ReadOnlyNoSize(src2, cscale);
if( !haveMask )
k.args(src1arg, src2arg, dstarg);
else
k.args(src1arg, src2arg, maskarg, dstarg);
}
size_t globalsize[] = { src1.cols*(cn/kercn), src1.rows };
return k.run(2, globalsize, 0, false);
}
static void binary_op( InputArray _src1, InputArray _src2, OutputArray _dst,
InputArray _mask, const BinaryFunc* tab,
bool bitwise, int oclop )
{
const _InputArray *psrc1 = &_src1, *psrc2 = &_src2;
int kind1 = psrc1->kind(), kind2 = psrc2->kind();
int type1 = psrc1->type(), depth1 = CV_MAT_DEPTH(type1), cn = CV_MAT_CN(type1);
int type2 = psrc2->type(), depth2 = CV_MAT_DEPTH(type2), cn2 = CV_MAT_CN(type2);
int dims1 = psrc1->dims(), dims2 = psrc2->dims();
Size sz1 = dims1 <= 2 ? psrc1->size() : Size();
Size sz2 = dims2 <= 2 ? psrc2->size() : Size();
bool use_opencl = (kind1 == _InputArray::UMAT || kind2 == _InputArray::UMAT) &&
ocl::useOpenCL() && dims1 <= 2 && dims2 <= 2;
bool haveMask = !_mask.empty(), haveScalar = false;
BinaryFunc func;
int c;
if( src1.dims <= 2 && src2.dims <= 2 && kind1 == kind2 &&
src1.size() == src2.size() && src1.type() == src2.type() && !haveMask )
if( dims1 <= 2 && dims2 <= 2 && kind1 == kind2 && sz1 == sz2 && type1 == type2 && !haveMask )
{
_dst.create(src1.size(), src1.type());
Mat dst = _dst.getMat();
_dst.create(sz1, type1);
if( use_opencl && ocl_binary_op(*psrc1, *psrc2, _dst, _mask, bitwise, oclop, false) )
return;
if( bitwise )
{
func = *tab;
c = (int)src1.elemSize();
cn = (int)CV_ELEM_SIZE(type1);
}
else
{
func = tab[src1.depth()];
c = src1.channels();
}
func = tab[depth1];
Mat src1 = psrc1->getMat(), src2 = psrc2->getMat(), dst = _dst.getMat();
Size sz = getContinuousSize(src1, src2, dst);
size_t len = sz.width*(size_t)c;
size_t len = sz.width*(size_t)cn;
if( len == (size_t)(int)len )
{
sz.width = (int)len;
@@ -946,56 +1025,67 @@ static void binary_op(InputArray _src1, InputArray _src2, OutputArray _dst,
}
}
if( (kind1 == _InputArray::MATX) + (kind2 == _InputArray::MATX) == 1 ||
src1.size != src2.size || src1.type() != src2.type() )
if( oclop == OCL_OP_NOT )
haveScalar = true;
else if( (kind1 == _InputArray::MATX) + (kind2 == _InputArray::MATX) == 1 ||
!psrc1->sameSize(*psrc2) || type1 != type2 )
{
if( checkScalar(src1, src2.type(), kind1, kind2) )
if( checkScalar(*psrc1, type2, kind1, kind2) )
{
// src1 is a scalar; swap it with src2
swap(src1, src2);
else if( !checkScalar(src2, src1.type(), kind2, kind1) )
swap(psrc1, psrc2);
swap(type1, type2);
swap(depth1, depth2);
swap(cn, cn2);
swap(sz1, sz2);
}
else if( !checkScalar(*psrc2, type1, kind2, kind1) )
CV_Error( CV_StsUnmatchedSizes,
"The operation is neither 'array op array' (where arrays have the same size and type), "
"nor 'array op scalar', nor 'scalar op array'" );
haveScalar = true;
}
else
{
CV_Assert( psrc1->sameSize(*psrc2) && type1 == type2 );
}
size_t esz = src1.elemSize();
size_t esz = CV_ELEM_SIZE(type1);
size_t blocksize0 = (BLOCK_SIZE + esz-1)/esz;
int cn = src1.channels();
BinaryFunc copymask = 0;
Mat mask;
bool reallocate = false;
if( haveMask )
{
mask = _mask.getMat();
CV_Assert( (mask.type() == CV_8UC1 || mask.type() == CV_8SC1) );
CV_Assert( mask.size == src1.size );
int mtype = _mask.type();
CV_Assert( (mtype == CV_8U || mtype == CV_8S) && _mask.sameSize(*psrc1));
copymask = getCopyMaskFunc(esz);
Mat tdst = _dst.getMat();
reallocate = tdst.size != src1.size || tdst.type() != src1.type();
reallocate = !_dst.sameSize(*psrc1) || _dst.type() != type1;
}
AutoBuffer<uchar> _buf;
uchar *scbuf = 0, *maskbuf = 0;
_dst.create(src1.dims, src1.size, src1.type());
Mat dst = _dst.getMat();
_dst.createSameSize(*psrc1, type1);
// if this is mask operation and dst has been reallocated,
// we have to
// we have to clear the destination
if( haveMask && reallocate )
dst = Scalar::all(0);
_dst.setTo(0.);
if( use_opencl && ocl_binary_op(*psrc1, *psrc2, _dst, _mask, bitwise, oclop, haveScalar ))
return;
Mat src1 = psrc1->getMat(), src2 = psrc2->getMat();
Mat dst = _dst.getMat(), mask = _mask.getMat();
if( bitwise )
{
func = *tab;
c = (int)esz;
cn = (int)esz;
}
else
{
func = tab[src1.depth()];
c = cn;
func = tab[depth1];
}
if( !haveScalar )
@@ -1006,8 +1096,8 @@ static void binary_op(InputArray _src1, InputArray _src2, OutputArray _dst,
NAryMatIterator it(arrays, ptrs);
size_t total = it.size, blocksize = total;
if( blocksize*c > INT_MAX )
blocksize = INT_MAX/c;
if( blocksize*cn > INT_MAX )
blocksize = INT_MAX/cn;
if( haveMask )
{
@@ -1022,7 +1112,7 @@ static void binary_op(InputArray _src1, InputArray _src2, OutputArray _dst,
{
int bsz = (int)MIN(total - j, blocksize);
func( ptrs[0], 0, ptrs[1], 0, haveMask ? maskbuf : ptrs[2], 0, Size(bsz*c, 1), 0 );
func( ptrs[0], 0, ptrs[1], 0, haveMask ? maskbuf : ptrs[2], 0, Size(bsz*cn, 1), 0 );
if( haveMask )
{
copymask( maskbuf, 0, ptrs[3], 0, ptrs[2], 0, Size(bsz, 1), &esz );
@@ -1054,7 +1144,7 @@ static void binary_op(InputArray _src1, InputArray _src2, OutputArray _dst,
{
int bsz = (int)MIN(total - j, blocksize);
func( ptrs[0], 0, scbuf, 0, haveMask ? maskbuf : ptrs[1], 0, Size(bsz*c, 1), 0 );
func( ptrs[0], 0, scbuf, 0, haveMask ? maskbuf : ptrs[1], 0, Size(bsz*cn, 1), 0 );
if( haveMask )
{
copymask( maskbuf, 0, ptrs[2], 0, ptrs[1], 0, Size(bsz, 1), &esz );
@@ -1101,47 +1191,59 @@ static BinaryFunc* getMinTab()
void cv::bitwise_and(InputArray a, InputArray b, OutputArray c, InputArray mask)
{
BinaryFunc f = (BinaryFunc)GET_OPTIMIZED(and8u);
binary_op(a, b, c, mask, &f, true);
binary_op(a, b, c, mask, &f, true, OCL_OP_AND);
}
void cv::bitwise_or(InputArray a, InputArray b, OutputArray c, InputArray mask)
{
BinaryFunc f = (BinaryFunc)GET_OPTIMIZED(or8u);
binary_op(a, b, c, mask, &f, true);
binary_op(a, b, c, mask, &f, true, OCL_OP_OR);
}
void cv::bitwise_xor(InputArray a, InputArray b, OutputArray c, InputArray mask)
{
BinaryFunc f = (BinaryFunc)GET_OPTIMIZED(xor8u);
binary_op(a, b, c, mask, &f, true);
binary_op(a, b, c, mask, &f, true, OCL_OP_XOR);
}
void cv::bitwise_not(InputArray a, OutputArray c, InputArray mask)
{
BinaryFunc f = (BinaryFunc)GET_OPTIMIZED(not8u);
binary_op(a, a, c, mask, &f, true);
binary_op(a, a, c, mask, &f, true, OCL_OP_NOT);
}
void cv::max( InputArray src1, InputArray src2, OutputArray dst )
{
binary_op(src1, src2, dst, noArray(), getMaxTab(), false );
binary_op(src1, src2, dst, noArray(), getMaxTab(), false, OCL_OP_MAX );
}
void cv::min( InputArray src1, InputArray src2, OutputArray dst )
{
binary_op(src1, src2, dst, noArray(), getMinTab(), false );
binary_op(src1, src2, dst, noArray(), getMinTab(), false, OCL_OP_MIN );
}
void cv::max(const Mat& src1, const Mat& src2, Mat& dst)
{
OutputArray _dst(dst);
binary_op(src1, src2, _dst, noArray(), getMaxTab(), false );
binary_op(src1, src2, _dst, noArray(), getMaxTab(), false, OCL_OP_MAX );
}
void cv::min(const Mat& src1, const Mat& src2, Mat& dst)
{
OutputArray _dst(dst);
binary_op(src1, src2, _dst, noArray(), getMinTab(), false );
binary_op(src1, src2, _dst, noArray(), getMinTab(), false, OCL_OP_MIN );
}
void cv::max(const UMat& src1, const UMat& src2, UMat& dst)
{
OutputArray _dst(dst);
binary_op(src1, src2, _dst, noArray(), getMaxTab(), false, OCL_OP_MAX );
}
void cv::min(const UMat& src1, const UMat& src2, UMat& dst)
{
OutputArray _dst(dst);
binary_op(src1, src2, _dst, noArray(), getMinTab(), false, OCL_OP_MIN );
}
@@ -1171,73 +1273,213 @@ static int actualScalarDepth(const double* data, int len)
CV_32S;
}
static void arithm_op(InputArray _src1, InputArray _src2, OutputArray _dst,
InputArray _mask, int dtype, BinaryFunc* tab, bool muldiv=false, void* usrdata=0)
static bool ocl_arithm_op(InputArray _src1, InputArray _src2, OutputArray _dst,
InputArray _mask, int wtype,
void* usrdata, int oclop,
bool haveScalar )
{
int kind1 = _src1.kind(), kind2 = _src2.kind();
Mat src1 = _src1.getMat(), src2 = _src2.getMat();
int type1 = _src1.type(), depth1 = CV_MAT_DEPTH(type1), cn = CV_MAT_CN(type1);
bool haveMask = !_mask.empty();
if( (haveMask || haveScalar) && cn > 4 )
return false;
int dtype = _dst.type(), ddepth = CV_MAT_DEPTH(dtype), wdepth = CV_MAT_DEPTH(wtype);
wtype = CV_MAKETYPE(wdepth, cn);
int type2 = haveScalar ? _src2.type() : wtype, depth2 = CV_MAT_DEPTH(type2);
UMat src1 = _src1.getUMat(), src2;
UMat dst = _dst.getUMat(), mask = _mask.getUMat();
char opts[1024];
int kercn = haveMask || haveScalar ? cn : 1;
if( (depth1 == depth2 || haveScalar) && ddepth == depth1 && wdepth == depth1 )
{
const char* oclopstr = oclop2str[oclop];
if( wdepth <= CV_16S )
{
oclopstr = oclop == OCL_OP_ADD ? "OCL_OP_ADD_SAT" :
oclop == OCL_OP_SUB ? "OCL_OP_SUB_SAT" :
oclop == OCL_OP_RSUB ? "OCL_OP_RSUB_SAT" : oclopstr;
}
sprintf(opts, "-D %s%s -D %s -D dstT=%s",
(haveMask ? "MASK_" : ""), (haveScalar ? "UNARY_OP" : "BINARY_OP"),
oclop2str[oclop], ocl::typeToStr(CV_MAKETYPE(ddepth, kercn)));
}
else
{
char cvtstr[3][32];
sprintf(opts, "-D %s%s -D %s -D srcT1=%s -D srcT2=%s "
"-D dstT=%s -D workT=%s -D convertToWT1=%s "
"-D convertToWT2=%s -D convertToDT=%s",
(haveMask ? "MASK_" : ""), (haveScalar ? "UNARY_OP" : "BINARY_OP"),
oclop2str[oclop], ocl::typeToStr(CV_MAKETYPE(depth1, kercn)),
ocl::typeToStr(CV_MAKETYPE(depth2, kercn)),
ocl::typeToStr(CV_MAKETYPE(ddepth, kercn)),
ocl::typeToStr(CV_MAKETYPE(wdepth, kercn)),
ocl::convertTypeStr(depth1, wdepth, kercn, cvtstr[0]),
ocl::convertTypeStr(depth2, wdepth, kercn, cvtstr[1]),
ocl::convertTypeStr(wdepth, ddepth, kercn, cvtstr[2]));
}
const uchar* usrdata_p = (const uchar*)usrdata;
const double* usrdata_d = (const double*)usrdata;
float usrdata_f[3];
int i, n = oclop == OCL_OP_MUL_SCALE || oclop == OCL_OP_DIV_SCALE ||
oclop == OCL_OP_RECIP_SCALE ? 1 : oclop == OCL_OP_ADDW ? 3 : 0;
if( n > 0 && wdepth == CV_32F )
{
for( i = 0; i < n; i++ )
usrdata_f[i] = (float)usrdata_d[i];
usrdata_p = (const uchar*)usrdata_f;
}
size_t usrdata_esz = CV_ELEM_SIZE(wdepth);
ocl::Kernel k("KF", ocl::core::arithm_oclsrc, opts);
if( k.empty() )
return false;
int cscale = cn/kercn;
ocl::KernelArg src1arg = ocl::KernelArg::ReadOnlyNoSize(src1, cscale);
ocl::KernelArg dstarg = haveMask ? ocl::KernelArg::ReadWrite(dst, cscale) :
ocl::KernelArg::WriteOnly(dst, cscale);
ocl::KernelArg maskarg = ocl::KernelArg::ReadOnlyNoSize(mask, 1);
if( haveScalar )
{
size_t esz = CV_ELEM_SIZE(wtype);
double buf[4]={0,0,0,0};
Mat src2sc = _src2.getMat();
if( !src2sc.empty() )
{
convertAndUnrollScalar(src2sc, wtype, (uchar*)buf, 1);
}
ocl::KernelArg scalararg = ocl::KernelArg(0, 0, 0, buf, esz);
if( !haveMask )
k.args(src1arg, dstarg, scalararg);
else
k.args(src1arg, maskarg, dstarg, scalararg);
}
else
{
src2 = _src2.getUMat();
ocl::KernelArg src2arg = ocl::KernelArg::ReadOnlyNoSize(src2, cscale);
if( !haveMask )
{
if(n == 0)
k.args(src1arg, src2arg, dstarg);
else if(n == 1)
k.args(src1arg, src2arg, dstarg,
ocl::KernelArg(0, 0, 0, usrdata_p, usrdata_esz));
else if(n == 3)
k.args(src1arg, src2arg, dstarg,
ocl::KernelArg(0, 0, 0, usrdata_p, usrdata_esz),
ocl::KernelArg(0, 0, 0, usrdata_p + usrdata_esz, usrdata_esz),
ocl::KernelArg(0, 0, 0, usrdata_p + usrdata_esz*2, usrdata_esz));
else
CV_Error(Error::StsNotImplemented, "unsupported number of extra parameters");
}
else
{
k.args(src1arg, src2arg, maskarg, dstarg);
}
}
size_t globalsize[] = { src1.cols*(cn/kercn), src1.rows };
return k.run(2, globalsize, 0, false);
}
static void arithm_op(InputArray _src1, InputArray _src2, OutputArray _dst,
InputArray _mask, int dtype, BinaryFunc* tab, bool muldiv=false,
void* usrdata=0, int oclop=-1 )
{
const _InputArray *psrc1 = &_src1, *psrc2 = &_src2;
int kind1 = psrc1->kind(), kind2 = psrc2->kind();
bool haveMask = !_mask.empty();
bool reallocate = false;
int type1 = psrc1->type(), depth1 = CV_MAT_DEPTH(type1), cn = CV_MAT_CN(type1);
int type2 = psrc2->type(), depth2 = CV_MAT_DEPTH(type2), cn2 = CV_MAT_CN(type2);
int wtype, dims1 = psrc1->dims(), dims2 = psrc2->dims();
Size sz1 = dims1 <= 2 ? psrc1->size() : Size();
Size sz2 = dims2 <= 2 ? psrc2->size() : Size();
bool use_opencl = (kind1 == _InputArray::UMAT || kind2 == _InputArray::UMAT) &&
ocl::useOpenCL() && dims1 <= 2 && dims2 <= 2;
bool src1Scalar = checkScalar(*psrc1, type2, kind1, kind2);
bool src2Scalar = checkScalar(*psrc2, type1, kind2, kind1);
bool src1Scalar = checkScalar(src1, src2.type(), kind1, kind2);
bool src2Scalar = checkScalar(src2, src1.type(), kind2, kind1);
if( (kind1 == kind2 || src1.channels() == 1) && src1.dims <= 2 && src2.dims <= 2 &&
src1.size() == src2.size() && src1.type() == src2.type() &&
!haveMask && ((!_dst.fixedType() && (dtype < 0 || CV_MAT_DEPTH(dtype) == src1.depth())) ||
(_dst.fixedType() && _dst.type() == _src1.type())) &&
if( (kind1 == kind2 || cn == 1) && sz1 == sz2 && dims1 <= 2 && dims2 <= 2 && type1 == type2 &&
!haveMask && ((!_dst.fixedType() && (dtype < 0 || CV_MAT_DEPTH(dtype) == depth1)) ||
(_dst.fixedType() && _dst.type() == type1)) &&
((src1Scalar && src2Scalar) || (!src1Scalar && !src2Scalar)) )
{
_dst.create(src1.size(), src1.type());
Mat dst = _dst.getMat();
_dst.createSameSize(*psrc1, type1);
if( use_opencl &&
ocl_arithm_op(*psrc1, *psrc2, _dst, _mask,
(!usrdata ? type1 : std::max(depth1, CV_32F)),
usrdata, oclop, false))
return;
Mat src1 = psrc1->getMat(), src2 = psrc2->getMat(), dst = _dst.getMat();
Size sz = getContinuousSize(src1, src2, dst, src1.channels());
tab[src1.depth()](src1.data, src1.step, src2.data, src2.step, dst.data, dst.step, sz, usrdata);
tab[depth1](src1.data, src1.step, src2.data, src2.step, dst.data, dst.step, sz, usrdata);
return;
}
bool haveScalar = false, swapped12 = false;
int depth2 = src2.depth();
if( src1.size != src2.size || src1.channels() != src2.channels() ||
if( dims1 != dims2 || sz1 != sz2 || cn != cn2 ||
((kind1 == _InputArray::MATX || kind2 == _InputArray::MATX) &&
src1.cols == 1 && src2.rows == 4) )
(sz1 == Size(1,4) || sz2 == Size(1,4))) )
{
if( checkScalar(src1, src2.type(), kind1, kind2) )
if( checkScalar(*psrc1, type2, kind1, kind2) )
{
// src1 is a scalar; swap it with src2
swap(src1, src2);
swap(psrc1, psrc2);
swap(sz1, sz2);
swap(type1, type2);
swap(depth1, depth2);
swap(cn, cn2);
swap(dims1, dims2);
swapped12 = true;
if( oclop == OCL_OP_SUB )
oclop = OCL_OP_RSUB;
}
else if( !checkScalar(src2, src1.type(), kind2, kind1) )
else if( !checkScalar(*psrc2, type1, kind2, kind1) )
CV_Error( CV_StsUnmatchedSizes,
"The operation is neither 'array op array' (where arrays have the same size and the same number of channels), "
"The operation is neither 'array op array' "
"(where arrays have the same size and the same number of channels), "
"nor 'array op scalar', nor 'scalar op array'" );
haveScalar = true;
CV_Assert(src2.type() == CV_64F && (src2.rows == 4 || src2.rows == 1));
CV_Assert(type2 == CV_64F && (sz2.height == 1 || sz2.height == 4));
if (!muldiv)
{
depth2 = actualScalarDepth(src2.ptr<double>(), src1.channels());
if( depth2 == CV_64F && (src1.depth() < CV_32S || src1.depth() == CV_32F) )
Mat sc = psrc2->getMat();
depth2 = actualScalarDepth(sc.ptr<double>(), cn);
if( depth2 == CV_64F && (depth1 < CV_32S || depth1 == CV_32F) )
depth2 = CV_32F;
}
else
depth2 = CV_64F;
}
int cn = src1.channels(), depth1 = src1.depth(), wtype;
BinaryFunc cvtsrc1 = 0, cvtsrc2 = 0, cvtdst = 0;
if( dtype < 0 )
{
if( _dst.fixedType() )
dtype = _dst.type();
else
{
if( !haveScalar && src1.type() != src2.type() )
if( !haveScalar && type1 != type2 )
CV_Error(CV_StsBadArg,
"When the input arrays in add/subtract/multiply/divide functions have different types, "
"the output array type must be explicitly specified");
dtype = src1.type();
dtype = type1;
}
}
dtype = CV_MAT_DEPTH(dtype);
@@ -1262,39 +1504,41 @@ static void arithm_op(InputArray _src1, InputArray _src2, OutputArray _dst,
wtype = std::max(wtype, dtype);
}
cvtsrc1 = depth1 == wtype ? 0 : getConvertFunc(depth1, wtype);
cvtsrc2 = depth2 == depth1 ? cvtsrc1 : depth2 == wtype ? 0 : getConvertFunc(depth2, wtype);
cvtdst = dtype == wtype ? 0 : getConvertFunc(wtype, dtype);
dtype = CV_MAKETYPE(dtype, cn);
wtype = CV_MAKETYPE(wtype, cn);
size_t esz1 = src1.elemSize(), esz2 = src2.elemSize();
size_t dsz = CV_ELEM_SIZE(dtype), wsz = CV_ELEM_SIZE(wtype);
size_t blocksize0 = (size_t)(BLOCK_SIZE + wsz-1)/wsz;
BinaryFunc copymask = 0;
Mat mask;
if( haveMask )
{
mask = _mask.getMat();
CV_Assert( (mask.type() == CV_8UC1 || mask.type() == CV_8SC1) );
CV_Assert( mask.size == src1.size );
copymask = getCopyMaskFunc(dsz);
Mat tdst = _dst.getMat();
reallocate = tdst.size != src1.size || tdst.type() != dtype;
int mtype = _mask.type();
CV_Assert( (mtype == CV_8UC1 || mtype == CV_8SC1) && _mask.sameSize(*psrc1) );
reallocate = !_dst.sameSize(*psrc1) || _dst.type() != dtype;
}
_dst.createSameSize(*psrc1, dtype);
if( reallocate )
_dst.setTo(0.);
if( use_opencl &&
ocl_arithm_op(*psrc1, *psrc2, _dst, _mask, wtype,
usrdata, oclop, haveScalar))
return;
BinaryFunc cvtsrc1 = type1 == wtype ? 0 : getConvertFunc(type1, wtype);
BinaryFunc cvtsrc2 = type2 == type1 ? cvtsrc1 : type2 == wtype ? 0 : getConvertFunc(type2, wtype);
BinaryFunc cvtdst = dtype == wtype ? 0 : getConvertFunc(wtype, dtype);
size_t esz1 = CV_ELEM_SIZE(type1), esz2 = CV_ELEM_SIZE(type2);
size_t dsz = CV_ELEM_SIZE(dtype), wsz = CV_ELEM_SIZE(wtype);
size_t blocksize0 = (size_t)(BLOCK_SIZE + wsz-1)/wsz;
BinaryFunc copymask = getCopyMaskFunc(dsz);
Mat src1 = psrc1->getMat(), src2 = psrc2->getMat(), dst = _dst.getMat(), mask = _mask.getMat();
AutoBuffer<uchar> _buf;
uchar *buf, *maskbuf = 0, *buf1 = 0, *buf2 = 0, *wbuf = 0;
size_t bufesz = (cvtsrc1 ? wsz : 0) + (cvtsrc2 || haveScalar ? wsz : 0) + (cvtdst ? wsz : 0) + (haveMask ? dsz : 0);
_dst.create(src1.dims, src1.size, dtype);
Mat dst = _dst.getMat();
if( haveMask && reallocate )
dst = Scalar::all(0);
size_t bufesz = (cvtsrc1 ? wsz : 0) +
(cvtsrc2 || haveScalar ? wsz : 0) +
(cvtdst ? wsz : 0) +
(haveMask ? dsz : 0);
BinaryFunc func = tab[CV_MAT_DEPTH(wtype)];
if( !haveScalar )
@@ -1476,7 +1720,7 @@ static BinaryFunc* getAbsDiffTab()
void cv::add( InputArray src1, InputArray src2, OutputArray dst,
InputArray mask, int dtype )
{
arithm_op(src1, src2, dst, mask, dtype, getAddTab() );
arithm_op(src1, src2, dst, mask, dtype, getAddTab(), false, 0, OCL_OP_ADD );
}
void cv::subtract( InputArray src1, InputArray src2, OutputArray dst,
@@ -1511,12 +1755,12 @@ void cv::subtract( InputArray src1, InputArray src2, OutputArray dst,
}
}
#endif
arithm_op(src1, src2, dst, mask, dtype, getSubTab() );
arithm_op(src1, src2, dst, mask, dtype, getSubTab(), false, 0, OCL_OP_SUB );
}
void cv::absdiff( InputArray src1, InputArray src2, OutputArray dst )
{
arithm_op(src1, src2, dst, noArray(), -1, getAbsDiffTab());
arithm_op(src1, src2, dst, noArray(), -1, getAbsDiffTab(), false, 0, OCL_OP_ABSDIFF);
}
/****************************************************************************************\
@@ -1847,19 +2091,20 @@ static BinaryFunc* getRecipTab()
void cv::multiply(InputArray src1, InputArray src2,
OutputArray dst, double scale, int dtype)
{
arithm_op(src1, src2, dst, noArray(), dtype, getMulTab(), true, &scale);
arithm_op(src1, src2, dst, noArray(), dtype, getMulTab(),
true, &scale, scale == 1. ? OCL_OP_MUL : OCL_OP_MUL_SCALE);
}
void cv::divide(InputArray src1, InputArray src2,
OutputArray dst, double scale, int dtype)
{
arithm_op(src1, src2, dst, noArray(), dtype, getDivTab(), true, &scale);
arithm_op(src1, src2, dst, noArray(), dtype, getDivTab(), true, &scale, OCL_OP_DIV_SCALE);
}
void cv::divide(double scale, InputArray src2,
OutputArray dst, int dtype)
{
arithm_op(src2, src2, dst, noArray(), dtype, getRecipTab(), true, &scale);
arithm_op(src2, src2, dst, noArray(), dtype, getRecipTab(), true, &scale, OCL_OP_RECIP_SCALE);
}
/****************************************************************************************\
@@ -2020,7 +2265,7 @@ void cv::addWeighted( InputArray src1, double alpha, InputArray src2,
double beta, double gamma, OutputArray dst, int dtype )
{
double scalars[] = {alpha, beta, gamma};
arithm_op(src1, src2, dst, noArray(), dtype, getAddWeightedTab(), true, scalars);
arithm_op(src1, src2, dst, noArray(), dtype, getAddWeightedTab(), true, scalars, OCL_OP_ADDW);
}

15
modules/core/src/copy.cpp
View File

@@ -220,6 +220,21 @@ void Mat::copyTo( OutputArray _dst ) const
return;
}
if( _dst.isUMat() )
{
_dst.create( dims, size.p, type() );
UMat dst = _dst.getUMat();
size_t i, sz[CV_MAX_DIM], dstofs[CV_MAX_DIM], esz = elemSize();
for( i = 0; i < (size_t)dims; i++ )
sz[i] = size.p[i];
sz[dims-1] *= esz;
dst.ndoffset(dstofs);
dstofs[dims-1] *= esz;
dst.u->currAllocator->upload(dst.u, data, dims, sz, dstofs, dst.step.p, step.p);
return;
}
if( dims <= 2 )
{
_dst.create( rows, cols, type() );

253
modules/core/src/matrix.cpp
View File

@@ -1436,6 +1436,181 @@ Size _InputArray::size(int i) const
}
}
int _InputArray::sizend(int* sz, int i) const
{
int j, d=0, k = kind();
if( k == NONE )
;
else if( k == MAT )
{
CV_Assert( i < 0 );
const Mat& m = *(const Mat*)obj;
d = m.dims;
if(sz)
for(j = 0; j < d; j++)
sz[j] = m.size.p[j];
}
else if( k == UMAT )
{
CV_Assert( i < 0 );
const UMat& m = *(const UMat*)obj;
d = m.dims;
if(sz)
for(j = 0; j < d; j++)
sz[j] = m.size.p[j];
}
else if( k == STD_VECTOR_MAT && i >= 0 )
{
const std::vector<Mat>& vv = *(const std::vector<Mat>*)obj;
CV_Assert( i < (int)vv.size() );
const Mat& m = vv[i];
d = m.dims;
if(sz)
for(j = 0; j < d; j++)
sz[j] = m.size.p[j];
}
else if( k == STD_VECTOR_UMAT && i >= 0 )
{
const std::vector<UMat>& vv = *(const std::vector<UMat>*)obj;
CV_Assert( i < (int)vv.size() );
const UMat& m = vv[i];
d = m.dims;
if(sz)
for(j = 0; j < d; j++)
sz[j] = m.size.p[j];
}
else
{
Size sz2d = size(i);
d = 2;
if(sz)
{
sz[0] = sz2d.height;
sz[1] = sz2d.width;
}
}
return d;
}
bool _InputArray::sameSize(const _InputArray& arr) const
{
int k1 = kind(), k2 = arr.kind();
Size sz1;
if( k1 == MAT )
{
const Mat* m = ((const Mat*)obj);
if( k2 == MAT )
return m->size == ((const Mat*)arr.obj)->size;
if( k2 == UMAT )
return m->size == ((const UMat*)arr.obj)->size;
if( m->dims > 2 )
return false;
sz1 = m->size();
}
else if( k1 == UMAT )
{
const UMat* m = ((const UMat*)obj);
if( k2 == MAT )
return m->size == ((const Mat*)arr.obj)->size;
if( k2 == UMAT )
return m->size == ((const UMat*)arr.obj)->size;
if( m->dims > 2 )
return false;
sz1 = m->size();
}
else
sz1 = size();
if( arr.dims() > 2 )
return false;
return sz1 == arr.size();
}
int _InputArray::dims(int i) const
{
int k = kind();
if( k == MAT )
{
CV_Assert( i < 0 );
return ((const Mat*)obj)->dims;
}
if( k == EXPR )
{
CV_Assert( i < 0 );
return ((const MatExpr*)obj)->a.dims;
}
if( k == UMAT )
{
CV_Assert( i < 0 );
return ((const UMat*)obj)->dims;
}
if( k == MATX )
{
CV_Assert( i < 0 );
return 2;
}
if( k == STD_VECTOR )
{
CV_Assert( i < 0 );
return 2;
}
if( k == NONE )
return 0;
if( k == STD_VECTOR_VECTOR )
{
const std::vector<std::vector<uchar> >& vv = *(const std::vector<std::vector<uchar> >*)obj;
if( i < 0 )
return 1;
CV_Assert( i < (int)vv.size() );
return 2;
}
if( k == STD_VECTOR_MAT )
{
const std::vector<Mat>& vv = *(const std::vector<Mat>*)obj;
if( i < 0 )
return 1;
CV_Assert( i < (int)vv.size() );
return vv[i].dims;
}
if( k == OPENGL_BUFFER )
{
CV_Assert( i < 0 );
return 2;
}
if( k == GPU_MAT )
{
CV_Assert( i < 0 );
return 2;
}
if( k == OCL_MAT )
{
return 2;
}
CV_Assert( k == CUDA_MEM );
//if( k == CUDA_MEM )
{
CV_Assert( i < 0 );
return 2;
}
}
size_t _InputArray::total(int i) const
{
int k = kind();
@@ -1570,6 +1745,61 @@ bool _InputArray::empty() const
return ((const cuda::CudaMem*)obj)->empty();
}
bool _InputArray::isContinuous(int i) const
{
int k = kind();
if( k == MAT )
return i < 0 ? ((const Mat*)obj)->isContinuous() : true;
if( k == UMAT )
return i < 0 ? ((const UMat*)obj)->isContinuous() : true;
if( k == EXPR || k == MATX || k == STD_VECTOR || k == NONE || k == STD_VECTOR_VECTOR)
return true;
if( k == STD_VECTOR_MAT )
{
const std::vector<Mat>& vv = *(const std::vector<Mat>*)obj;
CV_Assert((size_t)i < vv.size());
return vv[i].isContinuous();
}
if( k == STD_VECTOR_UMAT )
{
const std::vector<UMat>& vv = *(const std::vector<UMat>*)obj;
CV_Assert((size_t)i < vv.size());
return vv[i].isContinuous();
}
CV_Error(CV_StsNotImplemented, "This method is not implemented for oclMat yet");
return false;
}
void _InputArray::copyTo(const _OutputArray& arr) const
{
int k = kind();
if( k == NONE )
arr.release();
else if( k == MAT || k == MATX || k == STD_VECTOR )
{
Mat m = getMat();
m.copyTo(arr);
}
else if( k == EXPR )
{
const MatExpr& e = *((MatExpr*)obj);
if( arr.kind() == MAT )
arr.getMatRef() = e;
else
Mat(e).copyTo(arr);
}
else if( k == UMAT )
((UMat*)obj)->copyTo(arr);
else
CV_Error(Error::StsNotImplemented, "");
}
bool _OutputArray::fixedSize() const
{
@@ -1899,6 +2129,12 @@ void _OutputArray::create(int dims, const int* sizes, int mtype, int i,
CV_Error(Error::StsNotImplemented, "Unknown/unsupported array type");
}
void _OutputArray::createSameSize(const _InputArray& arr, int mtype) const
{
int sz[CV_MAX_DIM], d = arr.sizend(sz);
create(d, sz, mtype);
}
void _OutputArray::release() const
{
CV_Assert(!fixedSize());
@@ -2010,6 +2246,23 @@ cuda::CudaMem& _OutputArray::getCudaMemRef() const
return *(cuda::CudaMem*)obj;
}
void _OutputArray::setTo(const _InputArray& arr) const
{
int k = kind();
if( k == NONE )
;
else if( k == MAT || k == MATX || k == STD_VECTOR )
{
Mat m = getMat();
m.setTo(arr);
}
else if( k == UMAT )
((UMat*)obj)->setTo(arr);
else
CV_Error(Error::StsNotImplemented, "");
}
static _InputOutputArray _none;
InputOutputArray noArray() { return _none; }

310
modules/core/src/ocl.cpp
View File

@@ -592,9 +592,16 @@ static void* initOpenCLAndLoad(const char* funcname)
{
if(!initialized)
{
handle = dlopen("/System/Library/Frameworks/OpenCL.framework/Versions/Current/OpenCL", RTLD_LAZY);
const char* oclpath = getenv("OPENCV_OPENCL_RUNTIME");
oclpath = oclpath && strlen(oclpath) > 0 ? oclpath :
"/System/Library/Frameworks/OpenCL.framework/Versions/Current/OpenCL";
handle = dlopen(oclpath, RTLD_LAZY);
initialized = true;
g_haveOpenCL = handle != 0 && dlsym(handle, oclFuncToCheck) != 0;
if( g_haveOpenCL )
fprintf(stderr, "Succesffuly loaded OpenCL v1.1+ runtime from %s\n", oclpath);
else
fprintf(stderr, "Failed to load OpenCL runtime\n");
}
if(!handle)
return 0;
@@ -1212,16 +1219,13 @@ namespace cv { namespace ocl {
struct UMat2D
{
UMat2D(const UMat& m, int accessFlags)
UMat2D(const UMat& m)
{
CV_Assert(m.dims == 2);
data = (cl_mem)m.handle(accessFlags);
offset = m.offset;
step = m.step;
rows = m.rows;
cols = m.cols;
}
cl_mem data;
size_t offset;
size_t step;
int rows;
@@ -1230,10 +1234,8 @@ struct UMat2D
struct UMat3D
{
UMat3D(const UMat& m, int accessFlags)
UMat3D(const UMat& m)
{
CV_Assert(m.dims == 3);
data = (cl_mem)m.handle(accessFlags);
offset = m.offset;
step = m.step.p[1];
slicestep = m.step.p[0];
@@ -1241,7 +1243,6 @@ struct UMat3D
rows = m.size.p[1];
cols = m.size.p[2];
}
cl_mem data;
size_t offset;
size_t slicestep;
size_t step;
@@ -1315,7 +1316,7 @@ void setUseOpenCL(bool flag)
}
}
void finish()
void finish2()
{
Queue::getDefault().finish();
}
@@ -1528,7 +1529,7 @@ String Device::OpenCLVersion() const
{ return p ? p->getStrProp(CL_DEVICE_EXTENSIONS) : String(); }
String Device::driverVersion() const
{ return p ? p->getStrProp(CL_DEVICE_EXTENSIONS) : String(); }
{ return p ? p->getStrProp(CL_DRIVER_VERSION) : String(); }
int Device::type() const
{ return p ? p->getProp<cl_device_type, int>(CL_DEVICE_TYPE) : 0; }
@@ -1705,14 +1706,14 @@ size_t Device::profilingTimerResolution() const
const Device& Device::getDefault()
{
const Context& ctx = Context::getDefault();
const Context2& ctx = Context2::getDefault();
int idx = TLSData::get()->device;
return ctx.device(idx);
}
/////////////////////////////////////////////////////////////////////////////////////////
struct Context::Impl
struct Context2::Impl
{
Impl(int dtype0)
{
@@ -1777,7 +1778,7 @@ struct Context::Impl
devices.clear();
}
Program getProg(const ProgramSource& src,
Program getProg(const ProgramSource2& src,
const String& buildflags, String& errmsg)
{
String prefix = Program::getPrefix(buildflags);
@@ -1787,7 +1788,8 @@ struct Context::Impl
return it->second;
//String filename = format("%08x%08x_%08x%08x.clb2",
Program prog(src, buildflags, errmsg);
phash.insert(std::pair<HashKey,Program>(k, prog));
if(prog.ptr())
phash.insert(std::pair<HashKey,Program>(k, prog));
return prog;
}
@@ -1797,7 +1799,7 @@ struct Context::Impl
std::vector<Device> devices;
bool initialized;
typedef ProgramSource::hash_t hash_t;
typedef ProgramSource2::hash_t hash_t;
struct HashKey
{
@@ -1812,18 +1814,18 @@ struct Context::Impl
};
Context::Context()
Context2::Context2()
{
p = 0;
}
Context::Context(int dtype)
Context2::Context2(int dtype)
{
p = 0;
create(dtype);
}
bool Context::create(int dtype0)
bool Context2::create(int dtype0)
{
if( !haveOpenCL() )
return false;
@@ -1838,19 +1840,19 @@ bool Context::create(int dtype0)
return p != 0;
}
Context::~Context()
Context2::~Context2()
{
p->release();
}
Context::Context(const Context& c)
Context2::Context2(const Context2& c)
{
p = (Impl*)c.p;
if(p)
p->addref();
}
Context& Context::operator = (const Context& c)
Context2& Context2::operator = (const Context2& c)
{
Impl* newp = (Impl*)c.p;
if(newp)
@@ -1861,30 +1863,30 @@ Context& Context::operator = (const Context& c)
return *this;
}
void* Context::ptr() const
void* Context2::ptr() const
{
return p->handle;
}
size_t Context::ndevices() const
size_t Context2::ndevices() const
{
return p ? p->devices.size() : 0;
}
const Device& Context::device(size_t idx) const
const Device& Context2::device(size_t idx) const
{
static Device dummy;
return !p || idx >= p->devices.size() ? dummy : p->devices[idx];
}
Context& Context::getDefault()
Context2& Context2::getDefault()
{
static Context ctx;
static Context2 ctx;
if( !ctx.p && haveOpenCL() )
{
// do not create new Context right away.
// do not create new Context2 right away.
// First, try to retrieve existing context of the same type.
// In its turn, Platform::getContext() may call Context::create()
// In its turn, Platform::getContext() may call Context2::create()
// if there is no such context.
ctx.create(Device::TYPE_ACCELERATOR);
if(!ctx.p)
@@ -1898,7 +1900,7 @@ Context& Context::getDefault()
return ctx;
}
Program Context::getProg(const ProgramSource& prog,
Program Context2::getProg(const ProgramSource2& prog,
const String& buildopts, String& errmsg)
{
return p ? p->getProg(prog, buildopts, errmsg) : Program();
@@ -1906,14 +1908,14 @@ Program Context::getProg(const ProgramSource& prog,
struct Queue::Impl
{
Impl(const Context& c, const Device& d)
Impl(const Context2& c, const Device& d)
{
refcount = 1;
const Context* pc = &c;
const Context2* pc = &c;
cl_context ch = (cl_context)pc->ptr();
if( !ch )
{
pc = &Context::getDefault();
pc = &Context2::getDefault();
ch = (cl_context)pc->ptr();
}
cl_device_id dh = (cl_device_id)d.ptr();
@@ -1943,7 +1945,7 @@ Queue::Queue()
p = 0;
}
Queue::Queue(const Context& c, const Device& d)
Queue::Queue(const Context2& c, const Device& d)
{
p = 0;
create(c, d);
@@ -1973,7 +1975,7 @@ Queue::~Queue()
p->release();
}
bool Queue::create(const Context& c, const Device& d)
bool Queue::create(const Context2& c, const Device& d)
{
if(p)
p->release();
@@ -1996,7 +1998,7 @@ Queue& Queue::getDefault()
{
Queue& q = TLSData::get()->oclQueue;
if( !q.p )
q.create(Context::getDefault());
q.create(Context2::getDefault());
return q;
}
@@ -2008,15 +2010,20 @@ static cl_command_queue getQueue(const Queue& q)
return qq;
}
KernelArg::KernelArg(int _flags, UMat* _m, void* _obj, size_t _sz)
: flags(_flags), m(_m), obj(_obj), sz(_sz)
KernelArg::KernelArg()
: flags(0), m(0), obj(0), sz(0), wscale(1)
{
}
KernelArg::KernelArg(int _flags, UMat* _m, int _wscale, const void* _obj, size_t _sz)
: flags(_flags), m(_m), obj(_obj), sz(_sz), wscale(_wscale)
{
}
KernelArg KernelArg::Constant(const Mat& m)
{
CV_Assert(m.isContinuous());
return KernelArg(CONSTANT, 0, m.data, m.total()*m.elemSize());
return KernelArg(CONSTANT, 0, 1, m.data, m.total()*m.elemSize());
}
@@ -2099,8 +2106,8 @@ Kernel::Kernel(const char* kname, const Program& prog)
create(kname, prog);
}
Kernel::Kernel(const char* kname, const ProgramSource& src,
const String& buildopts, String& errmsg)
Kernel::Kernel(const char* kname, const ProgramSource2& src,
const String& buildopts, String* errmsg)
{
p = 0;
create(kname, src, buildopts, errmsg);
@@ -2143,15 +2150,17 @@ bool Kernel::create(const char* kname, const Program& prog)
return p != 0;
}
bool Kernel::create(const char* kname, const ProgramSource& src,
const String& buildopts, String& errmsg)
bool Kernel::create(const char* kname, const ProgramSource2& src,
const String& buildopts, String* errmsg)
{
if(p)
{
p->release();
p = 0;
}
const Program& prog = Context::getDefault().getProg(src, buildopts, errmsg);
String tempmsg;
if( !errmsg ) errmsg = &tempmsg;
const Program& prog = Context2::getDefault().getProg(src, buildopts, *errmsg);
return create(kname, prog);
}
@@ -2160,55 +2169,91 @@ void* Kernel::ptr() const
return p ? p->handle : 0;
}
void Kernel::set(int i, const void* value, size_t sz)
bool Kernel::empty() const
{
CV_Assert( p && clSetKernelArg(p->handle, (cl_uint)i, sz, value) >= 0 );
if( i == 0 )
p->cleanupUMats();
return ptr() == 0;
}
void Kernel::set(int i, const UMat& m)
int Kernel::set(int i, const void* value, size_t sz)
{
set(i, KernelArg(KernelArg::READ_WRITE, (UMat*)&m, 0, 0));
}
void Kernel::set(int i, const KernelArg& arg)
{
CV_Assert( p && p->handle );
CV_Assert(i >= 0);
if( i == 0 )
p->cleanupUMats();
if( !p || !p->handle || clSetKernelArg(p->handle, (cl_uint)i, sz, value) < 0 )
return -1;
return i+1;
}
int Kernel::set(int i, const UMat& m)
{
return set(i, KernelArg(KernelArg::READ_WRITE, (UMat*)&m, 0, 0));
}
int Kernel::set(int i, const KernelArg& arg)
{
CV_Assert( i >= 0 );
if( i == 0 )
p->cleanupUMats();
if( !p || !p->handle )
return -1;
if( arg.m )
{
int accessFlags = ((arg.flags & KernelArg::READ_ONLY) ? ACCESS_READ : 0) +
((arg.flags & KernelArg::WRITE_ONLY) ? ACCESS_WRITE : 0);
cl_mem h = (cl_mem)arg.m->handle(accessFlags);
if( arg.m->dims <= 2 )
{
UMat2D u2d(*arg.m, accessFlags);
clSetKernelArg(p->handle, (cl_uint)i, sizeof(u2d), &u2d);
UMat2D u2d(*arg.m);
clSetKernelArg(p->handle, (cl_uint)i, sizeof(h), &h);
clSetKernelArg(p->handle, (cl_uint)(i+1), sizeof(u2d.step), &u2d.step);
clSetKernelArg(p->handle, (cl_uint)(i+2), sizeof(u2d.offset), &u2d.offset);
i += 3;
if( !(arg.flags & KernelArg::NO_SIZE) )
{
int cols = u2d.cols*arg.wscale;
clSetKernelArg(p->handle, (cl_uint)i, sizeof(u2d.rows), &u2d.rows);
clSetKernelArg(p->handle, (cl_uint)(i+1), sizeof(u2d.cols), &cols);
i += 2;
}
}
else
{
UMat3D u3d(*arg.m, accessFlags);
clSetKernelArg(p->handle, (cl_uint)i, sizeof(u3d), &u3d);
UMat3D u3d(*arg.m);
clSetKernelArg(p->handle, (cl_uint)i, sizeof(h), &h);
clSetKernelArg(p->handle, (cl_uint)(i+1), sizeof(u3d.slicestep), &u3d.slicestep);
clSetKernelArg(p->handle, (cl_uint)(i+2), sizeof(u3d.step), &u3d.step);
clSetKernelArg(p->handle, (cl_uint)(i+3), sizeof(u3d.offset), &u3d.offset);
i += 4;
if( !(arg.flags & KernelArg::NO_SIZE) )
{
int cols = u3d.cols*arg.wscale;
clSetKernelArg(p->handle, (cl_uint)i, sizeof(u3d.slices), &u3d.rows);
clSetKernelArg(p->handle, (cl_uint)(i+1), sizeof(u3d.rows), &u3d.rows);
clSetKernelArg(p->handle, (cl_uint)(i+2), sizeof(u3d.cols), &cols);
i += 3;
}
}
p->addUMat(*arg.m);
return i;
}
else
{
clSetKernelArg(p->handle, (cl_uint)i, arg.sz, arg.obj);
}
clSetKernelArg(p->handle, (cl_uint)i, arg.sz, arg.obj);
return i+1;
}
void Kernel::run(int dims, size_t offset[], size_t globalsize[], size_t localsize[],
bool Kernel::run(int dims, size_t globalsize[], size_t localsize[],
bool sync, const Queue& q)
{
CV_Assert(p && p->handle && p->e == 0);
if(!p || !p->handle || p->e != 0)
return false;
cl_command_queue qq = getQueue(q);
clEnqueueNDRangeKernel(qq, p->handle, (cl_uint)dims,
offset, globalsize, localsize, 0, 0,
sync ? 0 : &p->e);
if( sync )
size_t offset[CV_MAX_DIM] = {0};
cl_int retval = clEnqueueNDRangeKernel(qq, p->handle, (cl_uint)dims,
offset, globalsize, localsize, 0, 0,
sync ? 0 : &p->e);
if( sync || retval < 0 )
{
clFinish(qq);
p->cleanupUMats();
@@ -2218,14 +2263,17 @@ void Kernel::run(int dims, size_t offset[], size_t globalsize[], size_t localsiz
p->addref();
clSetEventCallback(p->e, CL_COMPLETE, oclCleanupCallback, p);
}
return retval >= 0;
}
void Kernel::runTask(bool sync, const Queue& q)
bool Kernel::runTask(bool sync, const Queue& q)
{
CV_Assert(p && p->handle && p->e == 0);
if(!p || !p->handle || p->e != 0)
return false;
cl_command_queue qq = getQueue(q);
clEnqueueTask(qq, p->handle, 0, 0, sync ? 0 : &p->e);
if( sync )
cl_int retval = clEnqueueTask(qq, p->handle, 0, 0, sync ? 0 : &p->e);
if( sync || retval < 0 )
{
clFinish(qq);
p->cleanupUMats();
@@ -2235,6 +2283,7 @@ void Kernel::runTask(bool sync, const Queue& q)
p->addref();
clSetEventCallback(p->e, CL_COMPLETE, oclCleanupCallback, p);
}
return retval >= 0;
}
@@ -2273,11 +2322,11 @@ size_t Kernel::localMemSize() const
struct Program::Impl
{
Impl(const ProgramSource& _src,
Impl(const ProgramSource2& _src,
const String& _buildflags, String& errmsg)
{
refcount = 1;
const Context& ctx = Context::getDefault();
const Context2& ctx = Context2::getDefault();
src = _src;
buildflags = _buildflags;
const String& srcstr = src.source();
@@ -2293,17 +2342,20 @@ struct Program::Impl
void** deviceList = deviceListBuf;
for( i = 0; i < n; i++ )
deviceList[i] = ctx.device(i).ptr();
printf("Building the OpenCL program ...\n");
retval = clBuildProgram(handle, n,
(const cl_device_id*)deviceList,
buildflags.c_str(), 0, 0);
if( retval == CL_BUILD_PROGRAM_FAILURE )
{
char buf[1024];
char buf[1<<16];
size_t retsz = 0;
clGetProgramBuildInfo(handle, (cl_device_id)deviceList[0], CL_PROGRAM_BUILD_LOG,
sizeof(buf)-16, buf, &retsz);
errmsg = String(buf);
CV_Error_(Error::StsAssert, ("OpenCL program can not be built: %s", errmsg.c_str()));
}
CV_Assert(retval >= 0);
}
}
@@ -2315,7 +2367,7 @@ struct Program::Impl
if(_buf.empty())
return;
String prefix0 = Program::getPrefix(buildflags);
const Context& ctx = Context::getDefault();
const Context2& ctx = Context2::getDefault();
const Device& dev = Device::getDefault();
const char* pos0 = _buf.c_str();
const char* pos1 = strchr(pos0, '\n');
@@ -2366,7 +2418,7 @@ struct Program::Impl
IMPLEMENT_REFCOUNTABLE();
ProgramSource src;
ProgramSource2 src;
String buildflags;
cl_program handle;
};
@@ -2374,7 +2426,7 @@ struct Program::Impl
Program::Program() { p = 0; }
Program::Program(const ProgramSource& src,
Program::Program(const ProgramSource2& src,
const String& buildflags, String& errmsg)
{
p = 0;
@@ -2405,7 +2457,7 @@ Program::~Program()
p->release();
}
bool Program::create(const ProgramSource& src,
bool Program::create(const ProgramSource2& src,
const String& buildflags, String& errmsg)
{
if(p)
@@ -2419,9 +2471,9 @@ bool Program::create(const ProgramSource& src,
return p != 0;
}
const ProgramSource& Program::source() const
const ProgramSource2& Program::source() const
{
static ProgramSource dummy;
static ProgramSource2 dummy;
return p ? p->src : dummy;
}
@@ -2455,7 +2507,7 @@ String Program::getPrefix() const
String Program::getPrefix(const String& buildflags)
{
const Context& ctx = Context::getDefault();
const Context2& ctx = Context2::getDefault();
const Device& dev = ctx.device(0);
return format("name=%s\ndriver=%s\nbuildflags=%s\n",
dev.name().c_str(), dev.driverVersion().c_str(), buildflags.c_str());
@@ -2463,7 +2515,7 @@ String Program::getPrefix(const String& buildflags)
////////////////////////////////////////////////////////////////////////////////////////
struct ProgramSource::Impl
struct ProgramSource2::Impl
{
Impl(const char* _src)
{
@@ -2482,39 +2534,39 @@ struct ProgramSource::Impl
IMPLEMENT_REFCOUNTABLE();
String src;
ProgramSource::hash_t h;
ProgramSource2::hash_t h;
};
ProgramSource::ProgramSource()
ProgramSource2::ProgramSource2()
{
p = 0;
}
ProgramSource::ProgramSource(const char* prog)
ProgramSource2::ProgramSource2(const char* prog)
{
p = new Impl(prog);
}
ProgramSource::ProgramSource(const String& prog)
ProgramSource2::ProgramSource2(const String& prog)
{
p = new Impl(prog);
}
ProgramSource::~ProgramSource()
ProgramSource2::~ProgramSource2()
{
if(p)
p->release();
}
ProgramSource::ProgramSource(const ProgramSource& prog)
ProgramSource2::ProgramSource2(const ProgramSource2& prog)
{
p = prog.p;
if(p)
p->addref();
}
ProgramSource& ProgramSource::operator = (const ProgramSource& prog)
ProgramSource2& ProgramSource2::operator = (const ProgramSource2& prog)
{
Impl* newp = (Impl*)prog.p;
if(newp)
@@ -2525,13 +2577,13 @@ ProgramSource& ProgramSource::operator = (const ProgramSource& prog)
return *this;
}
const String& ProgramSource::source() const
const String& ProgramSource2::source() const
{
static String dummy;
return p ? p->src : dummy;
}
ProgramSource::hash_t ProgramSource::hash() const
ProgramSource2::hash_t ProgramSource2::hash() const
{
return p ? p->h : 0;
}
@@ -2551,7 +2603,7 @@ public:
return u;
}
void getBestFlags(const Context& ctx, int& createFlags, int& flags0) const
void getBestFlags(const Context2& ctx, int& createFlags, int& flags0) const
{
const Device& dev = ctx.device(0);
createFlags = CL_MEM_READ_WRITE;
@@ -2574,7 +2626,7 @@ public:
total *= sizes[i];
}
Context& ctx = Context::getDefault();
Context2& ctx = Context2::getDefault();
int createFlags = 0, flags0 = 0;
getBestFlags(ctx, createFlags, flags0);
@@ -2603,7 +2655,7 @@ public:
if(u->handle == 0)
{
CV_Assert(u->origdata != 0);
Context& ctx = Context::getDefault();
Context2& ctx = Context2::getDefault();
int createFlags = 0, flags0 = 0;
getBestFlags(ctx, createFlags, flags0);
@@ -2848,7 +2900,6 @@ public:
new_srcofs, new_dstofs, new_sz, new_srcstep[0], new_srcstep[1],
new_dststep[0], new_dststep[1], dstptr, 0, 0, 0) >= 0 );
}
clFinish(q);
}
void upload(UMatData* u, const void* srcptr, int dims, const size_t sz[],
@@ -2890,6 +2941,9 @@ public:
if( iscontinuous )
{
int crc = 0;
for( size_t i = 0; i < total; i++ )
crc ^= ((uchar*)srcptr)[i];
CV_Assert( clEnqueueWriteBuffer(q, (cl_mem)u->handle,
CL_TRUE, dstrawofs, total, srcptr, 0, 0, 0) >= 0 );
}
@@ -2949,10 +3003,11 @@ public:
}
else
{
CV_Assert( clEnqueueCopyBufferRect(q, (cl_mem)src->handle, (cl_mem)dst->handle,
cl_int retval;
CV_Assert( (retval = clEnqueueCopyBufferRect(q, (cl_mem)src->handle, (cl_mem)dst->handle,
new_srcofs, new_dstofs, new_sz,
new_srcstep[0], new_srcstep[1], new_dststep[0], new_dststep[1],
0, 0, 0) >= 0 );
0, 0, 0)) >= 0 );
}
dst->markHostCopyObsolete(true);
@@ -2969,4 +3024,61 @@ MatAllocator* getOpenCLAllocator()
return &allocator;
}
const char* typeToStr(int t)
{
static const char* tab[]=
{
"uchar", "uchar2", "uchar3", "uchar4",
"char", "char2", "char3", "char4",
"ushort", "ushort2", "ushort3", "ushort4",
"short", "short2", "short3", "short4",
"int", "int2", "int3", "int4",
"float", "float2", "float3", "float4",
"double", "double2", "double3", "double4",
"?", "?", "?", "?"
};
int cn = CV_MAT_CN(t);
return cn >= 4 ? "?" : tab[CV_MAT_DEPTH(t)*4 + cn-1];
}
const char* memopTypeToStr(int t)
{
static const char* tab[]=
{
"uchar", "uchar2", "uchar3", "uchar4",
"uchar", "uchar2", "uchar3", "uchar4",
"ushort", "ushort2", "ushort3", "ushort4",
"ushort", "ushort2", "ushort3", "ushort4",
"int", "int2", "int3", "int4",
"int", "int2", "int3", "int4",
"long", "long2", "long3", "long4",
"?", "?", "?", "?"
};
int cn = CV_MAT_CN(t);
return cn >= 4 ? "?" : tab[CV_MAT_DEPTH(t)*4 + cn-1];
}
const char* convertTypeStr(int sdepth, int ddepth, int cn, char* buf)
{
if( sdepth == ddepth )
return "noconvert";
const char *typestr = typeToStr(CV_MAKETYPE(ddepth, cn));
if( ddepth >= CV_32F ||
(ddepth == CV_32S && sdepth < CV_32S) ||
(ddepth == CV_16S && sdepth <= CV_8S) ||
(ddepth == CV_16U && sdepth == CV_8U))
{
sprintf(buf, "convert_%s", typestr);
}
else if( sdepth >= CV_32F )
{
sprintf(buf, "convert_%s%s_rte", typestr, (ddepth < CV_32S ? "_sat" : ""));
}
else
{
sprintf(buf, "convert_%s_sat", typestr);
}
return buf;
}
}}

307
modules/core/src/opencl/arithm.cl Normal file
View File

@@ -0,0 +1,307 @@
/*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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jia Haipeng, jiahaipeng95@gmail.com
//
//
// 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 copyright holders 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*/
/*
Usage:
after compiling this program user gets a single kernel called KF.
the following flags should be passed:
1) one of "-D BINARY_OP", "-D UNARY_OP", "-D MASK_BINARY_OP" or "-D MASK_UNARY_OP"
2) the actual operation performed, one of "-D OP_...", see below the list of operations.
2a) "-D dstDepth=<destination depth> [-D cn=<num channels]"
for some operations, like min/max/and/or/xor it's enough
2b) "-D srcDepth1=<source1 depth> -D srcDepth2=<source2 depth> -D dstDepth=<destination depth>
-D workDepth=<work depth> [-D cn=<num channels>]" - for mixed-type operations
*/
#if defined (DOUBLE_SUPPORT)
#ifdef cl_khr_fp64
#pragma OPENCL EXTENSION cl_khr_fp64:enable
#elif defined (cl_amd_fp64)
#pragma OPENCL EXTENSION cl_amd_fp64:enable
#endif
#endif
#define CV_32S 4
#define CV_32F 5
#define dstelem *(dstT*)(dstptr + dst_index)
#define noconvert(x) x
#ifndef workT
#define srcT1 dstT
#define srcT2 dstT
#define workT dstT
#define srcelem1 *(dstT*)(srcptr1 + src1_index)
#define srcelem2 *(dstT*)(srcptr2 + src2_index)
#define convertToDT noconvert
#else
#define srcelem1 convertToWT1(*(srcT1*)(srcptr1 + src1_index))
#define srcelem2 convertToWT2(*(srcT2*)(srcptr2 + src2_index))
#endif
#define EXTRA_PARAMS
#if defined OP_ADD_SAT
#define PROCESS_ELEM dstelem = add_sat(srcelem1, srcelem2)
#elif defined OP_ADD
#define PROCESS_ELEM dstelem = convertToDT(srcelem1 + srcelem2)
#elif defined OP_SUB_SAT
#define PROCESS_ELEM dstelem = sub_sat(srcelem1, srcelem2)
#elif defined OP_SUB
#define PROCESS_ELEM dstelem = convertToDT(srcelem1 - srcelem2)
#elif defined OP_RSUB_SAT
#define PROCESS_ELEM dstelem = sub_sat(srcelem2, srcelem1)
#elif defined OP_RSUB
#define PROCESS_ELEM dstelem = convertToDT(srcelem2 - srcelem1)
#elif defined OP_ABSDIFF
#define PROCESS_ELEM dstelem = abs_diff(srcelem1, srcelem2)
#elif defined OP_AND
#define PROCESS_ELEM dstelem = srcelem1 & srcelem2
#elif defined OP_OR
#define PROCESS_ELEM dstelem = srcelem1 | srcelem2
#elif defined OP_XOR
#define PROCESS_ELEM dstelem = srcelem1 ^ srcelem2
#elif defined OP_NOT
#define PROCESS_ELEM dstelem = ~srcelem1
#elif defined OP_MIN
#define PROCESS_ELEM dstelem = min(srcelem1, srcelem2)
#elif defined OP_MAX
#define PROCESS_ELEM dstelem = max(srcelem1, srcelem2)
#elif defined OP_MUL
#define PROCESS_ELEM dstelem = convertToDT(srcelem1 * srcelem2)
#elif defined OP_MUL_SCALE
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT scale
#define PROCESS_ELEM dstelem = convertToDT(srcelem1 * srcelem2 * scale)
#elif defined OP_DIV
#define PROCESS_ELEM \
workT e2 = srcelem2, zero = (workT)(0); \
dstelem = convertToDT(e2 != zero ? srcelem1 / e2 : zero)
#elif defined OP_DIV_SCALE
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT scale
#define PROCESS_ELEM \
workT e2 = srcelem2, zero = (workT)(0); \
dstelem = convertToDT(e2 != zero ? srcelem1 * scale / e2 : zero)
#elif defined OP_RECIP_SCALE
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT scale
#define PROCESS_ELEM \
workT e1 = srcelem1, zero = (workT)(0); \
dstelem = convertToDT(e1 != zero ? scale / e1 : zero)
#elif defined OP_ADDW
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT alpha, workT beta, workT gamma
#define PROCESS_ELEM dstelem = convertToDT(srcelem1*alpha + srcelem2*beta + gamma)
#elif defined OP_MAG
#define PROCESS_ELEM dstelem = hypot(srcelem1, srcelem2)
#elif defined OP_PHASE_RADIANS
#define PROCESS_ELEM \
workT tmp = atan2(srcelem2, srcelem1); \
if(tmp < 0) tmp += 6.283185307179586232; \
dstelem = tmp
#elif defined OP_PHASE_DEGREES
#define PROCESS_ELEM \
workT tmp = atan2(srcelem2, srcelem1)*57.29577951308232286465; \
if(tmp < 0) tmp += 360; \
dstelem = tmp
#elif defined OP_EXP
#define PROCESS_ELEM dstelem = exp(srcelem1)
#elif defined OP_SQRT
#define PROCESS_ELEM dstelem = sqrt(srcelem1)
#elif defined OP_LOG
#define PROCESS_ELEM dstelem = log(abs(srcelem1))
#elif defined OP_CMP
#define PROCESS_ELEM dstelem = convert_uchar(srcelem1 CMP_OPERATOR srcelem2 ? 255 : 0)
#elif defined OP_CONVERT
#define PROCESS_ELEM dstelem = convertToDT(srcelem1)
#elif defined OP_CONVERT_SCALE
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT alpha, workT beta
#define PROCESS_ELEM dstelem = convertToDT(srcelem1*alpha + beta)
#else
#error "unknown op type"
#endif
#if defined UNARY_OP || defined MASK_UNARY_OP
#undef srcelem2
#if defined OP_AND || defined OP_OR || defined OP_XOR || defined OP_ADD || defined OP_SAT_ADD || \
defined OP_SUB || defined OP_SAT_SUB || defined OP_RSUB || defined OP_SAT_RSUB || \
defined OP_ABSDIFF || defined OP_CMP || defined OP_MIN || defined OP_MAX
#undef EXTRA_PARAMS
#define EXTRA_PARAMS , workT srcelem2
#endif
#endif
#if defined BINARY_OP
__kernel void KF(__global const uchar* srcptr1, int srcstep1, int srcoffset1,
__global const uchar* srcptr2, int srcstep2, int srcoffset2,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols EXTRA_PARAMS )
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, srcstep1, x*sizeof(srcT1) + srcoffset1);
int src2_index = mad24(y, srcstep2, x*sizeof(srcT2) + srcoffset2);
int dst_index = mad24(y, dststep, x*sizeof(dstT) + dstoffset);
PROCESS_ELEM;
//printf("(x=%d, y=%d). %d, %d, %d\n", x, y, (int)srcelem1, (int)srcelem2, (int)dstelem);
}
}
#elif defined MASK_BINARY_OP
__kernel void KF(__global const uchar* srcptr1, int srcstep1, int srcoffset1,
__global const uchar* srcptr2, int srcstep2, int srcoffset2,
__global const uchar* mask, int maskstep, int maskoffset,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols EXTRA_PARAMS )
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int mask_index = mad24(y, maskstep, x + maskoffset);
if( mask[mask_index] )
{
int src1_index = mad24(y, srcstep1, x*sizeof(srcT1) + srcoffset1);
int src2_index = mad24(y, srcstep2, x*sizeof(srcT2) + srcoffset2);
int dst_index = mad24(y, dststep, x*sizeof(dstT) + dstoffset);
PROCESS_ELEM;
}
}
}
#elif defined UNARY_OP
__kernel void KF(__global const uchar* srcptr1, int srcstep1, int srcoffset1,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols EXTRA_PARAMS )
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, srcstep1, x*sizeof(srcT1) + srcoffset1);
int dst_index = mad24(y, dststep, x*sizeof(dstT) + dstoffset);
PROCESS_ELEM;
}
}
#elif defined MASK_UNARY_OP
__kernel void KF(__global const uchar* srcptr1, int srcstep1, int srcoffset1,
__global const uchar* mask, int maskstep, int maskoffset,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols EXTRA_PARAMS )
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int mask_index = mad24(y, maskstep, x + maskoffset);
if( mask[mask_index] )
{
int src1_index = mad24(y, srcstep1, x*sizeof(srcT1) + srcoffset1);
int dst_index = mad24(y, dststep, x*sizeof(dstT) + dstoffset);
PROCESS_ELEM;
}
}
}
#else
#error "Unknown operation type"
#endif

74
modules/core/src/opencl/copyset.cl Normal file
View File

@@ -0,0 +1,74 @@
/*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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, 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 copyright holders 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*/
__kernel void setMask(__global const uchar* mask, int maskstep, int maskoffset,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols, dstT value )
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int mask_index = mad24(y, maskstep, x + maskoffset);
if( mask[mask_index] )
{
int dst_index = mad24(y, dststep, x*sizeof(dstT) + dstoffset);
*(dstT*)(dstptr + dst_index) = value;
}
}
}
__kernel void set(__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols, dstT value )
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int dst_index = mad24(y, dststep, x*sizeof(dstT) + dstoffset);
*(dstT*)(dstptr + dst_index) = value;
}
}

96
modules/core/src/opencl/mulspectrums.cl Normal file
View File

@@ -0,0 +1,96 @@
/*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) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Peng Xiao, pengxiao@multicorewareinc.com
//
// 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 oclMaterials 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 uintel 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 uinterruption) 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*/
typedef float2 cfloat;
inline cfloat cmulf(cfloat a, cfloat b)
{
return (cfloat)( a.x*b.x - a.y*b.y, a.x*b.y + a.y*b.x);
}
inline cfloat conjf(cfloat a)
{
return (cfloat)( a.x, - a.y );
}
__kernel void
mulAndScaleSpectrumsKernel(
__global const cfloat* a,
__global const cfloat* b,
float scale,
__global cfloat* dst,
uint cols,
uint rows,
uint mstep
)
{
const uint x = get_global_id(0);
const uint y = get_global_id(1);
const uint idx = mad24(y, mstep / sizeof(cfloat), x);
if (x < cols && y < rows)
{
cfloat v = cmulf(a[idx], b[idx]);
dst[idx] = (cfloat)( v.x * scale, v.y * scale );
}
}
__kernel void
mulAndScaleSpectrumsKernel_CONJ(
__global const cfloat* a,
__global const cfloat* b,
float scale,
__global cfloat* dst,
uint cols,
uint rows,
uint mstep
)
{
const uint x = get_global_id(0);
const uint y = get_global_id(1);
const uint idx = mad24(y, mstep / sizeof(cfloat), x);
if (x < cols && y < rows)
{
cfloat v = cmulf(a[idx], conjf(b[idx]));
dst[idx] = (cfloat)( v.x * scale, v.y * scale );
}
}

73
modules/core/src/opencl/polarcart.cl Normal file
View File

@@ -0,0 +1,73 @@
/*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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, 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 copyright holders 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*/
__kernel void polarToCart(__global const uchar* mask, int maskstep, int maskoffset,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols, dstT value )
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int mask_index = mad24(y, maskstep, x + maskoffset);
if( mask[mask_index] )
{
int dst_index = mad24(y, dststep, x*sizeof(dstT) + dstoffset);
*(dstT*)(dstptr + dst_index) = value;
}
}
}
__kernel void cartToPolar(__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols, dstT value )
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int dst_index = mad24(y, dststep, x*sizeof(dstT) + dstoffset);
*(dstT*)(dstptr + dst_index) = value;
}
}

104
modules/core/src/opencl/reductions.cl Normal file
View File

@@ -0,0 +1,104 @@
/*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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Shengen Yan,yanshengen@gmail.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if defined (DOUBLE_SUPPORT)
#ifdef cl_khr_fp64
#pragma OPENCL EXTENSION cl_khr_fp64:enable
#elif defined (cl_amd_fp64)
#pragma OPENCL EXTENSION cl_amd_fp64:enable
#endif
#endif
#if FUNC_SUM
#define FUNC(a, b) b += a;
#elif FUNC_ABS_SUM
#define FUNC(a, b) b += a >= (dstT)(0) ? a : -a;
#elif FUNC_SQR_SUM
#define FUNC(a, b) b += a * a;
#else
#error No sum function
#endif
/**************************************Array buffer SUM**************************************/
__kernel void arithm_op_sum(int cols,int invalid_cols,int offset,int elemnum,int groupnum,
__global srcT *src, __global dstT *dst)
{
unsigned int lid = get_local_id(0);
unsigned int gid = get_group_id(0);
unsigned int id = get_global_id(0);
unsigned int idx = offset + id + (id / cols) * invalid_cols;
__local dstT localmem_sum[128];
dstT sum = (dstT)(0), temp;
for (int grainSize = groupnum << 8; id < elemnum; id += grainSize)
{
idx = offset + id + (id / cols) * invalid_cols;
temp = convertToDstT(src[idx]);
FUNC(temp, sum);
}
if (lid > 127)
localmem_sum[lid - 128] = sum;
barrier(CLK_LOCAL_MEM_FENCE);
if (lid < 128)
localmem_sum[lid] = sum + localmem_sum[lid];
barrier(CLK_LOCAL_MEM_FENCE);
for (int lsize = 64; lsize > 0; lsize >>= 1)
{
if (lid < lsize)
{
int lid2 = lsize + lid;
localmem_sum[lid] = localmem_sum[lid] + localmem_sum[lid2];
}
barrier(CLK_LOCAL_MEM_FENCE);
}
if (lid == 0)
dst[gid] = localmem_sum[0];
}

30
modules/core/src/precomp.hpp
View File

@@ -67,6 +67,8 @@
#define GET_OPTIMIZED(func) (func)
#endif
#include "opencl_kernels.hpp"
namespace cv
{
@@ -205,13 +207,30 @@ enum { BLOCK_SIZE = 1024 };
inline bool checkScalar(const Mat& sc, int atype, int sckind, int akind)
{
if( sc.dims > 2 || (sc.cols != 1 && sc.rows != 1) || !sc.isContinuous() )
if( sc.dims > 2 || !sc.isContinuous() )
return false;
Size sz = sc.size();
if(sz.width != 1 && sz.height != 1)
return false;
int cn = CV_MAT_CN(atype);
if( akind == _InputArray::MATX && sckind != _InputArray::MATX )
return false;
return sc.size() == Size(1, 1) || sc.size() == Size(1, cn) || sc.size() == Size(cn, 1) ||
(sc.size() == Size(1, 4) && sc.type() == CV_64F && cn <= 4);
return sz == Size(1, 1) || sz == Size(1, cn) || sz == Size(cn, 1) ||
(sz == Size(1, 4) && sc.type() == CV_64F && cn <= 4);
}
inline bool checkScalar(InputArray sc, int atype, int sckind, int akind)
{
if( sc.dims() > 2 || !sc.isContinuous() )
return false;
Size sz = sc.size();
if(sz.width != 1 && sz.height != 1)
return false;
int cn = CV_MAT_CN(atype);
if( akind == _InputArray::MATX && sckind != _InputArray::MATX )
return false;
return sz == Size(1, 1) || sz == Size(1, cn) || sz == Size(cn, 1) ||
(sz == Size(1, 4) && sc.type() == CV_64F && cn <= 4);
}
void convertAndUnrollScalar( const Mat& sc, int buftype, uchar* scbuf, size_t blocksize );
@@ -227,7 +246,10 @@ struct TLSData
static TLSData* get();
};
namespace ocl { MatAllocator* getOpenCLAllocator(); }
namespace ocl
{
MatAllocator* getOpenCLAllocator();
}
}

50
modules/core/src/umatrix.cpp
View File

@@ -197,6 +197,7 @@ UMat Mat::getUMat(int accessFlags) const
if(!u)
return hdr;
UMat::getStdAllocator()->allocate(u, accessFlags);
hdr.flags = flags;
setSize(hdr, dims, size.p, step.p);
finalizeHdr(hdr);
hdr.u = u;
@@ -548,7 +549,8 @@ Mat UMat::getMat(int accessFlags) const
CV_Assert(u->data != 0);
Mat hdr(dims, size.p, type(), u->data + offset, step.p);
hdr.u = u;
hdr.datastart = hdr.data = u->data;
hdr.datastart = u->data;
hdr.data = hdr.datastart + offset;
hdr.datalimit = hdr.dataend = u->data + u->size;
CV_XADD(&hdr.u->refcount, 1);
return hdr;
@@ -617,7 +619,7 @@ void UMat::copyTo(OutputArray _dst) const
void* dsthandle = dst.handle(ACCESS_WRITE);
if( srchandle == dsthandle && dst.offset == offset )
return;
ndoffset(dstofs);
dst.ndoffset(dstofs);
CV_Assert(u->currAllocator == dst.u->currAllocator);
u->currAllocator->copy(u, dst.u, dims, sz, srcofs, step.p, dstofs, dst.step.p, false);
}
@@ -633,6 +635,50 @@ void UMat::convertTo(OutputArray, int, double, double) const
CV_Error(Error::StsNotImplemented, "");
}
UMat& UMat::setTo(InputArray _value, InputArray _mask)
{
bool haveMask = !_mask.empty();
int t = type(), cn = CV_MAT_CN(t);
if( dims <= 2 && cn <= 4 && ocl::useOpenCL() )
{
Mat value = _value.getMat();
CV_Assert( checkScalar(value, type(), _value.kind(), _InputArray::UMAT) );
double buf[4];
convertAndUnrollScalar(value, t, (uchar*)buf, 1);
char opts[1024];
sprintf(opts, "-D dstT=%s", ocl::memopTypeToStr(t));
ocl::Kernel setK(haveMask ? "setMask" : "set", ocl::core::copyset_oclsrc, opts);
if( !setK.empty() )
{
ocl::KernelArg scalararg(0, 0, 0, buf, CV_ELEM_SIZE(t));
UMat mask;
if( haveMask )
{
mask = _mask.getUMat();
CV_Assert( mask.size() == size() && mask.type() == CV_8U );
ocl::KernelArg maskarg = ocl::KernelArg::ReadOnlyNoSize(mask);
ocl::KernelArg dstarg = ocl::KernelArg::ReadWrite(*this);
setK.args(maskarg, dstarg, scalararg);
}
else
{
ocl::KernelArg dstarg = ocl::KernelArg::WriteOnly(*this);
setK.args(dstarg, scalararg);
}
size_t globalsize[] = { cols, rows };
if( setK.run(2, globalsize, 0, false) )
return *this;
}
}
Mat m = getMat(haveMask ? ACCESS_RW : ACCESS_WRITE);
m.setTo(_value, _mask);
return *this;
}
UMat& UMat::operator = (const Scalar&)
{
CV_Error(Error::StsNotImplemented, "");

73
modules/core/test/test_umat.cpp
View File

@@ -91,11 +91,11 @@ bool CV_UMatTest::TestUMat()
{
try
{
Mat a(100, 100, CV_16S), b;
Mat a(100, 100, CV_16SC2), b, c;
randu(a, Scalar::all(-100), Scalar::all(100));
Rect roi(1, 3, 10, 20);
Mat ra(a, roi), rb;
UMat ua, ura;
Rect roi(1, 3, 5, 4);
Mat ra(a, roi), rb, rc, rc0;
UMat ua, ura, ub, urb, uc, urc;
a.copyTo(ua);
ua.copyTo(b);
CHECK_DIFF(a, b);
@@ -112,6 +112,71 @@ bool CV_UMatTest::TestUMat()
}
ra.copyTo(rb);
CHECK_DIFF(ra, rb);
b = a.clone();
ra = a(roi);
rb = b(roi);
randu(b, Scalar::all(-100), Scalar::all(100));
b.copyTo(ub);
urb = ub(roi);
/*std::cout << "==============================================\nbefore op (CPU):\n";
std::cout << "ra: " << ra << std::endl;
std::cout << "rb: " << rb << std::endl;*/
ra.copyTo(ura);
rb.copyTo(urb);
ra.release();
rb.release();
ura.copyTo(ra);
urb.copyTo(rb);
/*std::cout << "==============================================\nbefore op (GPU):\n";
std::cout << "ra: " << ra << std::endl;
std::cout << "rb: " << rb << std::endl;*/
cv::max(ra, rb, rc);
cv::max(ura, urb, urc);
urc.copyTo(rc0);
/*std::cout << "==============================================\nafter op:\n";
std::cout << "rc: " << rc << std::endl;
std::cout << "rc0: " << rc0 << std::endl;*/
CHECK_DIFF(rc0, rc);
{
UMat tmp = rc0.getUMat(ACCESS_WRITE);
cv::max(ura, urb, tmp);
}
CHECK_DIFF(rc0, rc);
ura.copyTo(urc);
cv::max(urc, urb, urc);
urc.copyTo(rc0);
CHECK_DIFF(rc0, rc);
rc = ra ^ rb;
cv::bitwise_xor(ura, urb, urc);
urc.copyTo(rc0);
/*std::cout << "==============================================\nafter op:\n";
std::cout << "ra: " << rc0 << std::endl;
std::cout << "rc: " << rc << std::endl;*/
CHECK_DIFF(rc0, rc);
rc = ra + rb;
cv::add(ura, urb, urc);
urc.copyTo(rc0);
CHECK_DIFF(rc0, rc);
cv::subtract(ra, Scalar::all(5), rc);
cv::subtract(ura, Scalar::all(5), urc);
urc.copyTo(rc0);
CHECK_DIFF(rc0, rc);
}
catch (const test_excep& e)
{