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

Revert "Revert "Merge pull request #836 from jet47:gpu-modules""

Browse Source
This commit is contained in:
Vladislav Vinogradov
2013-06-04 13:32:35 +04:00
parent 10340fe234
commit 3eeaa9189c
472 changed files with 29894 additions and 23019 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
modules/core/include/opencv2/core/cuda/border_interpolate.hpp
View File

@@ -73,8 +73,8 @@ namespace cv { namespace gpu { namespace cudev
return (x >= 0 && x < width) ? saturate_cast<D>(data[x]) : val;
}
const int width;
const D val;
int width;
D val;
};
template <typename D> struct BrdColConstant
@@ -98,8 +98,8 @@ namespace cv { namespace gpu { namespace cudev
return (y >= 0 && y < height) ? saturate_cast<D>(*(const T*)((const char*)data + y * step)) : val;
}
const int height;
const D val;
int height;
D val;
};
template <typename D> struct BrdConstant
@@ -120,9 +120,9 @@ namespace cv { namespace gpu { namespace cudev
return (x >= 0 && x < width && y >= 0 && y < height) ? saturate_cast<D>(src(y, x)) : val;
}
const int height;
const int width;
const D val;
int height;
int width;
D val;
};
//////////////////////////////////////////////////////////////
@@ -165,7 +165,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(data[idx_col(x)]);
}
const int last_col;
int last_col;
};
template <typename D> struct BrdColReplicate
@@ -205,7 +205,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(*(const T*)((const char*)data + idx_row(y) * step));
}
const int last_row;
int last_row;
};
template <typename D> struct BrdReplicate
@@ -255,8 +255,8 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(src(idx_row(y), idx_col(x)));
}
const int last_row;
const int last_col;
int last_row;
int last_col;
};
//////////////////////////////////////////////////////////////
@@ -299,7 +299,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(data[idx_col(x)]);
}
const int last_col;
int last_col;
};
template <typename D> struct BrdColReflect101
@@ -339,7 +339,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(*(const D*)((const char*)data + idx_row(y) * step));
}
const int last_row;
int last_row;
};
template <typename D> struct BrdReflect101
@@ -389,8 +389,8 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(src(idx_row(y), idx_col(x)));
}
const int last_row;
const int last_col;
int last_row;
int last_col;
};
//////////////////////////////////////////////////////////////
@@ -433,7 +433,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(data[idx_col(x)]);
}
const int last_col;
int last_col;
};
template <typename D> struct BrdColReflect
@@ -473,7 +473,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(*(const D*)((const char*)data + idx_row(y) * step));
}
const int last_row;
int last_row;
};
template <typename D> struct BrdReflect
@@ -523,8 +523,8 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(src(idx_row(y), idx_col(x)));
}
const int last_row;
const int last_col;
int last_row;
int last_col;
};
//////////////////////////////////////////////////////////////
@@ -567,7 +567,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(data[idx_col(x)]);
}
const int width;
int width;
};
template <typename D> struct BrdColWrap
@@ -607,7 +607,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(*(const D*)((const char*)data + idx_row(y) * step));
}
const int height;
int height;
};
template <typename D> struct BrdWrap
@@ -664,8 +664,8 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<D>(src(idx_row(y), idx_col(x)));
}
const int height;
const int width;
int height;
int width;
};
//////////////////////////////////////////////////////////////
@@ -683,8 +683,8 @@ namespace cv { namespace gpu { namespace cudev
return b.at(y, x, ptr);
}
const Ptr2D ptr;
const B b;
Ptr2D ptr;
B b;
};
// under win32 there is some bug with templated types that passed as kernel parameters
@@ -704,10 +704,10 @@ namespace cv { namespace gpu { namespace cudev
return (x >= 0 && x < width && y >= 0 && y < height) ? saturate_cast<D>(src(y, x)) : val;
}
const Ptr2D src;
const int height;
const int width;
const D val;
Ptr2D src;
int height;
int width;
D val;
};
}}} // namespace cv { namespace gpu { namespace cudev

9
modules/core/include/opencv2/core/cuda/common.hpp
View File

@@ -87,15 +87,6 @@ namespace cv { namespace gpu
namespace cv { namespace gpu
{
enum
{
BORDER_REFLECT101_GPU = 0,
BORDER_REPLICATE_GPU,
BORDER_CONSTANT_GPU,
BORDER_REFLECT_GPU,
BORDER_WRAP_GPU
};
namespace cudev
{
__host__ __device__ __forceinline__ int divUp(int total, int grain)

125
modules/core/include/opencv2/core/cuda/emulation.hpp
View File

@@ -43,6 +43,7 @@
#ifndef OPENCV_GPU_EMULATION_HPP_
#define OPENCV_GPU_EMULATION_HPP_
#include "common.hpp"
#include "warp_reduce.hpp"
namespace cv { namespace gpu { namespace cudev
@@ -131,8 +132,130 @@ namespace cv { namespace gpu { namespace cudev
return ::atomicMin(address, val);
#endif
}
}; // struct cmem
struct glob
{
static __device__ __forceinline__ int atomicAdd(int* address, int val)
{
return ::atomicAdd(address, val);
}
static __device__ __forceinline__ unsigned int atomicAdd(unsigned int* address, unsigned int val)
{
return ::atomicAdd(address, val);
}
static __device__ __forceinline__ float atomicAdd(float* address, float val)
{
#if __CUDA_ARCH__ >= 200
return ::atomicAdd(address, val);
#else
int* address_as_i = (int*) address;
int old = *address_as_i, assumed;
do {
assumed = old;
old = ::atomicCAS(address_as_i, assumed,
__float_as_int(val + __int_as_float(assumed)));
} while (assumed != old);
return __int_as_float(old);
#endif
}
static __device__ __forceinline__ double atomicAdd(double* address, double val)
{
#if __CUDA_ARCH__ >= 130
unsigned long long int* address_as_ull = (unsigned long long int*) address;
unsigned long long int old = *address_as_ull, assumed;
do {
assumed = old;
old = ::atomicCAS(address_as_ull, assumed,
__double_as_longlong(val + __longlong_as_double(assumed)));
} while (assumed != old);
return __longlong_as_double(old);
#else
(void) address;
(void) val;
return 0.0;
#endif
}
static __device__ __forceinline__ int atomicMin(int* address, int val)
{
return ::atomicMin(address, val);
}
static __device__ __forceinline__ float atomicMin(float* address, float val)
{
#if __CUDA_ARCH__ >= 120
int* address_as_i = (int*) address;
int old = *address_as_i, assumed;
do {
assumed = old;
old = ::atomicCAS(address_as_i, assumed,
__float_as_int(::fminf(val, __int_as_float(assumed))));
} while (assumed != old);
return __int_as_float(old);
#else
(void) address;
(void) val;
return 0.0f;
#endif
}
static __device__ __forceinline__ double atomicMin(double* address, double val)
{
#if __CUDA_ARCH__ >= 130
unsigned long long int* address_as_ull = (unsigned long long int*) address;
unsigned long long int old = *address_as_ull, assumed;
do {
assumed = old;
old = ::atomicCAS(address_as_ull, assumed,
__double_as_longlong(::fmin(val, __longlong_as_double(assumed))));
} while (assumed != old);
return __longlong_as_double(old);
#else
(void) address;
(void) val;
return 0.0;
#endif
}
static __device__ __forceinline__ int atomicMax(int* address, int val)
{
return ::atomicMax(address, val);
}
static __device__ __forceinline__ float atomicMax(float* address, float val)
{
#if __CUDA_ARCH__ >= 120
int* address_as_i = (int*) address;
int old = *address_as_i, assumed;
do {
assumed = old;
old = ::atomicCAS(address_as_i, assumed,
__float_as_int(::fmaxf(val, __int_as_float(assumed))));
} while (assumed != old);
return __int_as_float(old);
#else
(void) address;
(void) val;
return 0.0f;
#endif
}
static __device__ __forceinline__ double atomicMax(double* address, double val)
{
#if __CUDA_ARCH__ >= 130
unsigned long long int* address_as_ull = (unsigned long long int*) address;
unsigned long long int old = *address_as_ull, assumed;
do {
assumed = old;
old = ::atomicCAS(address_as_ull, assumed,
__double_as_longlong(::fmax(val, __longlong_as_double(assumed))));
} while (assumed != old);
return __longlong_as_double(old);
#else
(void) address;
(void) val;
return 0.0;
#endif
}
};
};
}; //struct Emulation
}}} // namespace cv { namespace gpu { namespace cudev
#endif /* OPENCV_GPU_EMULATION_HPP_ */

10
modules/core/include/opencv2/core/cuda/filters.hpp
View File

@@ -67,7 +67,7 @@ namespace cv { namespace gpu { namespace cudev
return src(__float2int_rz(y), __float2int_rz(x));
}
const Ptr2D src;
Ptr2D src;
};
template <typename Ptr2D> struct LinearFilter
@@ -107,7 +107,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<elem_type>(out);
}
const Ptr2D src;
Ptr2D src;
};
template <typename Ptr2D> struct CubicFilter
@@ -166,7 +166,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<elem_type>(res);
}
const Ptr2D src;
Ptr2D src;
};
// for integer scaling
template <typename Ptr2D> struct IntegerAreaFilter
@@ -203,7 +203,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<elem_type>(out);
}
const Ptr2D src;
Ptr2D src;
float scale_x, scale_y ,scale;
};
@@ -269,7 +269,7 @@ namespace cv { namespace gpu { namespace cudev
return saturate_cast<elem_type>(out);
}
const Ptr2D src;
Ptr2D src;
float scale_x, scale_y;
int width, haight;
};

26
modules/core/include/opencv2/core/cuda/functional.hpp
View File

@@ -554,8 +554,8 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ thresh_binary_func(const thresh_binary_func& other)
: thresh(other.thresh), maxVal(other.maxVal) {}
const T thresh;
const T maxVal;
T thresh;
T maxVal;
};
template <typename T> struct thresh_binary_inv_func : unary_function<T, T>
@@ -571,8 +571,8 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ thresh_binary_inv_func(const thresh_binary_inv_func& other)
: thresh(other.thresh), maxVal(other.maxVal) {}
const T thresh;
const T maxVal;
T thresh;
T maxVal;
};
template <typename T> struct thresh_trunc_func : unary_function<T, T>
@@ -588,7 +588,7 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ thresh_trunc_func(const thresh_trunc_func& other)
: thresh(other.thresh) {}
const T thresh;
T thresh;
};
template <typename T> struct thresh_to_zero_func : unary_function<T, T>
@@ -604,7 +604,7 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ thresh_to_zero_func(const thresh_to_zero_func& other)
: thresh(other.thresh) {}
const T thresh;
T thresh;
};
template <typename T> struct thresh_to_zero_inv_func : unary_function<T, T>
@@ -620,7 +620,7 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ thresh_to_zero_inv_func(const thresh_to_zero_inv_func& other)
: thresh(other.thresh) {}
const T thresh;
T thresh;
};
// Function Object Adaptors
@@ -636,7 +636,7 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ unary_negate() {}
__host__ __device__ __forceinline__ unary_negate(const unary_negate& other) : pred(other.pred) {}
const Predicate pred;
Predicate pred;
};
template <typename Predicate> __host__ __device__ __forceinline__ unary_negate<Predicate> not1(const Predicate& pred)
@@ -657,7 +657,7 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ binary_negate() {}
__host__ __device__ __forceinline__ binary_negate(const binary_negate& other) : pred(other.pred) {}
const Predicate pred;
Predicate pred;
};
template <typename BinaryPredicate> __host__ __device__ __forceinline__ binary_negate<BinaryPredicate> not2(const BinaryPredicate& pred)
@@ -677,8 +677,8 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ binder1st() {}
__host__ __device__ __forceinline__ binder1st(const binder1st& other) : op(other.op), arg1(other.arg1) {}
const Op op;
const typename Op::first_argument_type arg1;
Op op;
typename Op::first_argument_type arg1;
};
template <typename Op, typename T> __host__ __device__ __forceinline__ binder1st<Op> bind1st(const Op& op, const T& x)
@@ -698,8 +698,8 @@ namespace cv { namespace gpu { namespace cudev
__host__ __device__ __forceinline__ binder2nd() {}
__host__ __device__ __forceinline__ binder2nd(const binder2nd& other) : op(other.op), arg2(other.arg2) {}
const Op op;
const typename Op::second_argument_type arg2;
Op op;
typename Op::second_argument_type arg2;
};
template <typename Op, typename T> __host__ __device__ __forceinline__ binder2nd<Op> bind2nd(const Op& op, const T& x)

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

@@ -74,10 +74,6 @@
namespace cv { namespace gpu {
CV_EXPORTS cv::String getNppErrorMessage(int code);
CV_EXPORTS cv::String getCudaDriverApiErrorMessage(int code);
// Converts CPU border extrapolation mode into GPU internal analogue.
// Returns true if the GPU analogue exists, false otherwise.
CV_EXPORTS bool tryConvertToGpuBorderType(int cpuBorderType, int& gpuBorderType);
}}
#ifndef HAVE_CUDA