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

refactored and fixed bugs in gpu warp functions (remap, resize, warpAffine, warpPerspective)

Browse Source 
wrote more complicated tests for them
implemented own version of warpAffine and warpPerspective for different border interpolation types
refactored some gpu tests
This commit is contained in:
Vladislav Vinogradov
2012-03-14 15:54:17 +00:00
parent 6e2507c197
commit ade7394e77
33 changed files with 6243 additions and 4545 deletions
Download Patch File Download Diff File Expand all files Collapse all files

319
modules/gpu/src/color.cpp
View File

File diff suppressed because it is too large Load Diff

56
modules/gpu/src/cuda/remap.cu
View File

@@ -47,10 +47,10 @@
#include "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/filters.hpp"
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
namespace imgproc
{
template <typename Ptr2D, typename T> __global__ void remap(const Ptr2D src, const PtrStepf mapx, const PtrStepf mapy, DevMem2D_<T> dst)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
@@ -67,11 +67,10 @@ namespace cv { namespace gpu { namespace device
template <template <typename> class Filter, template <typename> class B, typename T> struct RemapDispatcherStream
{
static void call(DevMem2D_<T> src, DevMem2Df mapx, DevMem2Df mapy, DevMem2D_<T> dst,
const float* borderValue, cudaStream_t stream, int)
static void call(DevMem2D_<T> src, DevMem2Df mapx, DevMem2Df mapy, DevMem2D_<T> dst, const float* borderValue, cudaStream_t stream, int)
{
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type work_type;
dim3 block(32, 8);
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
@@ -86,11 +85,10 @@ namespace cv { namespace gpu { namespace device
template <template <typename> class Filter, template <typename> class B, typename T> struct RemapDispatcherNonStream
{
static void call(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, DevMem2Df mapx, DevMem2Df mapy,
DevMem2D_<T> dst, const float* borderValue, int)
static void call(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, DevMem2Df mapx, DevMem2Df mapy, DevMem2D_<T> dst, const float* borderValue, int)
{
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type work_type;
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type work_type;
dim3 block(32, 8);
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
@@ -189,7 +187,7 @@ namespace cv { namespace gpu { namespace device
#undef OPENCV_GPU_IMPLEMENT_REMAP_TEX
template <template <typename> class Filter, template <typename> class B, typename T> struct RemapDispatcher
{
{
static void call(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, DevMem2Df mapx, DevMem2Df mapy,
DevMem2D_<T> dst, const float* borderValue, cudaStream_t stream, int cc)
{
@@ -208,26 +206,26 @@ namespace cv { namespace gpu { namespace device
static const caller_t callers[3][5] =
{
{
RemapDispatcher<PointFilter, BrdReflect101, T>::call,
RemapDispatcher<PointFilter, BrdReplicate, T>::call,
RemapDispatcher<PointFilter, BrdConstant, T>::call,
RemapDispatcher<PointFilter, BrdReflect, T>::call,
RemapDispatcher<PointFilter, BrdWrap, T>::call
{
RemapDispatcher<PointFilter, BrdReflect101, T>::call,
RemapDispatcher<PointFilter, BrdReplicate, T>::call,
RemapDispatcher<PointFilter, BrdConstant, T>::call,
RemapDispatcher<PointFilter, BrdReflect, T>::call,
RemapDispatcher<PointFilter, BrdWrap, T>::call
},
{
RemapDispatcher<LinearFilter, BrdReflect101, T>::call,
RemapDispatcher<LinearFilter, BrdReplicate, T>::call,
RemapDispatcher<LinearFilter, BrdConstant, T>::call,
RemapDispatcher<LinearFilter, BrdReflect, T>::call,
RemapDispatcher<LinearFilter, BrdWrap, T>::call
{
RemapDispatcher<LinearFilter, BrdReflect101, T>::call,
RemapDispatcher<LinearFilter, BrdReplicate, T>::call,
RemapDispatcher<LinearFilter, BrdConstant, T>::call,
RemapDispatcher<LinearFilter, BrdReflect, T>::call,
RemapDispatcher<LinearFilter, BrdWrap, T>::call
},
{
RemapDispatcher<CubicFilter, BrdReflect101, T>::call,
RemapDispatcher<CubicFilter, BrdReplicate, T>::call,
RemapDispatcher<CubicFilter, BrdConstant, T>::call,
RemapDispatcher<CubicFilter, BrdReflect, T>::call,
RemapDispatcher<CubicFilter, BrdWrap, T>::call
{
RemapDispatcher<CubicFilter, BrdReflect101, T>::call,
RemapDispatcher<CubicFilter, BrdReplicate, T>::call,
RemapDispatcher<CubicFilter, BrdConstant, T>::call,
RemapDispatcher<CubicFilter, BrdReflect, T>::call,
RemapDispatcher<CubicFilter, BrdWrap, T>::call
}
};

121
modules/gpu/src/cuda/resize.cu
View File

@@ -47,10 +47,10 @@
#include "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/filters.hpp"
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
namespace imgproc
{
template <typename Ptr2D, typename T> __global__ void resize(const Ptr2D src, float fx, float fy, DevMem2D_<T> dst)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
@@ -58,52 +58,25 @@ namespace cv { namespace gpu { namespace device
if (x < dst.cols && y < dst.rows)
{
const float xcoo = x / fx;
const float ycoo = y / fy;
const float xcoo = x * fx;
const float ycoo = y * fy;
dst.ptr(y)[x] = saturate_cast<T>(src(ycoo, xcoo));
}
}
template <typename Ptr2D, typename T> __global__ void resizeNN(const Ptr2D src, float fx, float fy, DevMem2D_<T> dst)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < dst.cols && y < dst.rows)
{
const float xcoo = x / fx;
const float ycoo = y / fy;
dst.ptr(y)[x] = src(__float2int_rd(ycoo), __float2int_rd(xcoo));
dst(y, x) = saturate_cast<T>(src(ycoo, xcoo));
}
}
template <template <typename> class Filter, typename T> struct ResizeDispatcherStream
{
static void call(DevMem2D_<T> src, float fx, float fy, DevMem2D_<T> dst, cudaStream_t stream)
{
{
dim3 block(32, 8);
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
BrdReplicate<T> brd(src.rows, src.cols);
BorderReader< PtrStep<T>, BrdReplicate<T> > brdSrc(src, brd);
Filter< BorderReader< PtrStep<T>, BrdReplicate<T> > > filter_src(brdSrc);
Filter< BorderReader< PtrStep<T>, BrdReplicate<T> > > filteredSrc(brdSrc);
resize<<<grid, block, 0, stream>>>(filter_src, fx, fy, dst);
cudaSafeCall( cudaGetLastError() );
}
};
template <typename T> struct ResizeDispatcherStream<PointFilter, T>
{
static void call(DevMem2D_<T> src, float fx, float fy, DevMem2D_<T> dst, cudaStream_t stream)
{
dim3 block(32, 8);
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
BrdReplicate<T> brd(src.rows, src.cols);
BorderReader< PtrStep<T>, BrdReplicate<T> > brdSrc(src, brd);
resizeNN<<<grid, block, 0, stream>>>(brdSrc, fx, fy, dst);
resize<<<grid, block, 0, stream>>>(filteredSrc, fx, fy, dst);
cudaSafeCall( cudaGetLastError() );
}
};
@@ -111,31 +84,15 @@ namespace cv { namespace gpu { namespace device
template <template <typename> class Filter, typename T> struct ResizeDispatcherNonStream
{
static void call(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, float fx, float fy, DevMem2D_<T> dst)
{
{
dim3 block(32, 8);
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
BrdReplicate<T> brd(src.rows, src.cols);
BorderReader< PtrStep<T>, BrdReplicate<T> > brdSrc(src, brd);
Filter< BorderReader< PtrStep<T>, BrdReplicate<T> > > filter_src(brdSrc);
Filter< BorderReader< PtrStep<T>, BrdReplicate<T> > > filteredSrc(brdSrc);
resize<<<grid, block>>>(filter_src, fx, fy, dst);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template <typename T> struct ResizeDispatcherNonStream<PointFilter, T>
{
static void call(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, float fx, float fy, DevMem2D_<T> dst)
{
dim3 block(32, 8);
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
BrdReplicate<T> brd(src.rows, src.cols);
BorderReader< PtrStep<T>, BrdReplicate<T> > brdSrc(src, brd);
resizeNN<<<grid, block>>>(brdSrc, fx, fy, dst);
resize<<<grid, block>>>(filteredSrc, fx, fy, dst);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
@@ -148,73 +105,61 @@ namespace cv { namespace gpu { namespace device
{ \
typedef type elem_type; \
typedef int index_type; \
int xoff, yoff; \
tex_resize_ ## type ## _reader (int xoff_, int yoff_) : xoff(xoff_), yoff(yoff_) {} \
const int xoff; \
const int yoff; \
__host__ tex_resize_ ## type ## _reader(int xoff_, int yoff_) : xoff(xoff_), yoff(yoff_) {} \
__device__ __forceinline__ elem_type operator ()(index_type y, index_type x) const \
{ \
return tex2D(tex_resize_ ## type , x + xoff, y + yoff); \
return tex2D(tex_resize_ ## type, x + xoff, y + yoff); \
} \
}; \
template <template <typename> class Filter> struct ResizeDispatcherNonStream<Filter, type> \
template <template <typename> class Filter> struct ResizeDispatcherNonStream<Filter, type > \
{ \
static void call(DevMem2D_< type > src, DevMem2D_< type > srcWhole, int xoff, int yoff, float fx, float fy, DevMem2D_< type > dst) \
{ \
dim3 block(32, 8); \
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y)); \
bindTexture(&tex_resize_ ## type , srcWhole); \
tex_resize_ ## type ##_reader texSrc(xoff, yoff); \
BrdReplicate< type > brd(src.rows, src.cols); \
BorderReader< tex_resize_ ## type ##_reader , BrdReplicate< type > > brdSrc(texSrc, brd); \
Filter< BorderReader< tex_resize_ ## type ##_reader , BrdReplicate< type > > > filter_src(brdSrc); \
resize<<<grid, block>>>(filter_src, fx, fy, dst); \
cudaSafeCall( cudaGetLastError() ); \
cudaSafeCall( cudaDeviceSynchronize() ); \
} \
}; \
template <> struct ResizeDispatcherNonStream<PointFilter, type> \
{ \
static void call(DevMem2D_< type > src, DevMem2D_< type > srcWhole, int xoff, int yoff, float fx, float fy, DevMem2D_< type > dst) \
{ \
dim3 block(32, 8); \
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y)); \
bindTexture(&tex_resize_ ## type , srcWhole); \
tex_resize_ ## type ##_reader texSrc(xoff, yoff); \
BrdReplicate< type > brd(src.rows, src.cols); \
BorderReader< tex_resize_ ## type ##_reader , BrdReplicate< type > > brdSrc(texSrc, brd); \
resizeNN<<<grid, block>>>(brdSrc, fx, fy, dst); \
bindTexture(&tex_resize_ ## type, srcWhole); \
tex_resize_ ## type ## _reader texSrc(xoff, yoff); \
if (srcWhole.cols == src.cols && srcWhole.rows == src.rows) \
{ \
Filter<tex_resize_ ## type ## _reader> filteredSrc(texSrc); \
resize<<<grid, block>>>(filteredSrc, fx, fy, dst); \
} \
else \
{ \
BrdReplicate< type > brd(src.rows, src.cols); \
BorderReader<tex_resize_ ## type ## _reader, BrdReplicate< type > > brdSrc(texSrc, brd); \
Filter< BorderReader<tex_resize_ ## type ## _reader, BrdReplicate< type > > > filteredSrc(brdSrc); \
resize<<<grid, block>>>(filteredSrc, fx, fy, dst); \
} \
cudaSafeCall( cudaGetLastError() ); \
cudaSafeCall( cudaDeviceSynchronize() ); \
} \
};
OPENCV_GPU_IMPLEMENT_RESIZE_TEX(uchar)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(uchar2)
OPENCV_GPU_IMPLEMENT_RESIZE_TEX(uchar4)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(schar)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(char2)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(char4)
OPENCV_GPU_IMPLEMENT_RESIZE_TEX(ushort)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(ushort2)
OPENCV_GPU_IMPLEMENT_RESIZE_TEX(ushort4)
OPENCV_GPU_IMPLEMENT_RESIZE_TEX(short)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(short2)
OPENCV_GPU_IMPLEMENT_RESIZE_TEX(short4)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(int)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(int2)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(int4)
OPENCV_GPU_IMPLEMENT_RESIZE_TEX(float)
//OPENCV_GPU_IMPLEMENT_RESIZE_TEX(float2)
OPENCV_GPU_IMPLEMENT_RESIZE_TEX(float4)
#undef OPENCV_GPU_IMPLEMENT_RESIZE_TEX
template <template <typename> class Filter, typename T> struct ResizeDispatcher
{
{
static void call(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, float fx, float fy, DevMem2D_<T> dst, cudaStream_t stream)
{
if (stream == 0)

380
modules/gpu/src/cuda/warp.cu Normal file
View File

@@ -0,0 +1,380 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "internal_shared.hpp"
#include "opencv2/gpu/device/border_interpolate.hpp"
#include "opencv2/gpu/device/vec_traits.hpp"
#include "opencv2/gpu/device/vec_math.hpp"
#include "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/filters.hpp"
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
__constant__ float c_warpMat[3 * 3];
struct AffineTransform
{
static __device__ __forceinline__ float2 calcCoord(int x, int y)
{
const float xcoo = c_warpMat[0] * x + c_warpMat[1] * y + c_warpMat[2];
const float ycoo = c_warpMat[3] * x + c_warpMat[4] * y + c_warpMat[5];
return make_float2(xcoo, ycoo);
}
};
struct PerspectiveTransform
{
static __device__ __forceinline__ float2 calcCoord(int x, int y)
{
const float coeff = 1.0f / (c_warpMat[6] * x + c_warpMat[7] * y + c_warpMat[8]);
const float xcoo = coeff * (c_warpMat[0] * x + c_warpMat[1] * y + c_warpMat[2]);
const float ycoo = coeff * (c_warpMat[3] * x + c_warpMat[4] * y + c_warpMat[5]);
return make_float2(xcoo, ycoo);
}
};
///////////////////////////////////////////////////////////////////
// Build Maps
template <class Transform> __global__ void buildWarpMaps(DevMem2Df xmap, PtrStepf ymap)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < xmap.cols && y < xmap.rows)
{
const float2 coord = Transform::calcCoord(x, y);
xmap(y, x) = coord.x;
ymap(y, x) = coord.y;
}
}
template <class Transform> void buildWarpMaps_caller(DevMem2Df xmap, DevMem2Df ymap, cudaStream_t stream)
{
dim3 block(32, 8);
dim3 grid(divUp(xmap.cols, block.x), divUp(xmap.rows, block.y));
buildWarpMaps<Transform><<<grid, block, 0, stream>>>(xmap, ymap);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
void buildWarpAffineMaps_gpu(float coeffs[2 * 3], DevMem2Df xmap, DevMem2Df ymap, cudaStream_t stream)
{
cudaSafeCall( cudaMemcpyToSymbol(c_warpMat, coeffs, 2 * 3 * sizeof(float)) );
buildWarpMaps_caller<AffineTransform>(xmap, ymap, stream);
}
void buildWarpPerspectiveMaps_gpu(float coeffs[3 * 3], DevMem2Df xmap, DevMem2Df ymap, cudaStream_t stream)
{
cudaSafeCall( cudaMemcpyToSymbol(c_warpMat, coeffs, 3 * 3 * sizeof(float)) );
buildWarpMaps_caller<PerspectiveTransform>(xmap, ymap, stream);
}
///////////////////////////////////////////////////////////////////
// Warp
template <class Transform, class Ptr2D, typename T> __global__ void warp(const Ptr2D src, DevMem2D_<T> dst)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < dst.cols && y < dst.rows)
{
const float2 coord = Transform::calcCoord(x, y);
dst.ptr(y)[x] = saturate_cast<T>(src(coord.y, coord.x));
}
}
template <class Transform, template <typename> class Filter, template <typename> class B, typename T> struct WarpDispatcherStream
{
static void call(DevMem2D_<T> src, DevMem2D_<T> dst, const float* borderValue, cudaStream_t stream, int)
{
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type work_type;
dim3 block(32, 8);
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
B<work_type> brd(src.rows, src.cols, VecTraits<work_type>::make(borderValue));
BorderReader< PtrStep<T>, B<work_type> > brdSrc(src, brd);
Filter< BorderReader< PtrStep<T>, B<work_type> > > filter_src(brdSrc);
warp<Transform><<<grid, block, 0, stream>>>(filter_src, dst);
cudaSafeCall( cudaGetLastError() );
}
};
template <class Transform, template <typename> class Filter, template <typename> class B, typename T> struct WarpDispatcherNonStream
{
static void call(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, DevMem2D_<T> dst, const float* borderValue, int)
{
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type work_type;
dim3 block(32, 8);
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
B<work_type> brd(src.rows, src.cols, VecTraits<work_type>::make(borderValue));
BorderReader< PtrStep<T>, B<work_type> > brdSrc(src, brd);
Filter< BorderReader< PtrStep<T>, B<work_type> > > filter_src(brdSrc);
warp<Transform><<<grid, block>>>(filter_src, dst);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
#define OPENCV_GPU_IMPLEMENT_WARP_TEX(type) \
texture< type , cudaTextureType2D > tex_warp_ ## type (0, cudaFilterModePoint, cudaAddressModeClamp); \
struct tex_warp_ ## type ## _reader \
{ \
typedef type elem_type; \
typedef int index_type; \
int xoff, yoff; \
tex_warp_ ## type ## _reader (int xoff_, int yoff_) : xoff(xoff_), yoff(yoff_) {} \
__device__ __forceinline__ elem_type operator ()(index_type y, index_type x) const \
{ \
return tex2D(tex_warp_ ## type , x + xoff, y + yoff); \
} \
}; \
template <class Transform, template <typename> class Filter, template <typename> class B> struct WarpDispatcherNonStream<Transform, Filter, B, type> \
{ \
static void call(DevMem2D_< type > src, DevMem2D_< type > srcWhole, int xoff, int yoff, DevMem2D_< type > dst, const float* borderValue, int cc) \
{ \
typedef typename TypeVec<float, VecTraits< type >::cn>::vec_type work_type; \
dim3 block(32, cc >= 20 ? 8 : 4); \
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y)); \
bindTexture(&tex_warp_ ## type , srcWhole); \
tex_warp_ ## type ##_reader texSrc(xoff, yoff); \
B<work_type> brd(src.rows, src.cols, VecTraits<work_type>::make(borderValue)); \
BorderReader< tex_warp_ ## type ##_reader, B<work_type> > brdSrc(texSrc, brd); \
Filter< BorderReader< tex_warp_ ## type ##_reader, B<work_type> > > filter_src(brdSrc); \
warp<Transform><<<grid, block>>>(filter_src, dst); \
cudaSafeCall( cudaGetLastError() ); \
cudaSafeCall( cudaDeviceSynchronize() ); \
} \
}; \
template <class Transform, template <typename> class Filter> struct WarpDispatcherNonStream<Transform, Filter, BrdReplicate, type> \
{ \
static void call(DevMem2D_< type > src, DevMem2D_< type > srcWhole, int xoff, int yoff, DevMem2D_< type > dst, const float*, int) \
{ \
dim3 block(32, 8); \
dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y)); \
bindTexture(&tex_warp_ ## type , srcWhole); \
tex_warp_ ## type ##_reader texSrc(xoff, yoff); \
if (srcWhole.cols == src.cols && srcWhole.rows == src.rows) \
{ \
Filter< tex_warp_ ## type ##_reader > filter_src(texSrc); \
warp<Transform><<<grid, block>>>(filter_src, dst); \
} \
else \
{ \
BrdReplicate<type> brd(src.rows, src.cols); \
BorderReader< tex_warp_ ## type ##_reader, BrdReplicate<type> > brdSrc(texSrc, brd); \
Filter< BorderReader< tex_warp_ ## type ##_reader, BrdReplicate<type> > > filter_src(brdSrc); \
warp<Transform><<<grid, block>>>(filter_src, dst); \
} \
cudaSafeCall( cudaGetLastError() ); \
cudaSafeCall( cudaDeviceSynchronize() ); \
} \
};
OPENCV_GPU_IMPLEMENT_WARP_TEX(uchar)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(uchar2)
OPENCV_GPU_IMPLEMENT_WARP_TEX(uchar4)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(schar)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(char2)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(char4)
OPENCV_GPU_IMPLEMENT_WARP_TEX(ushort)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(ushort2)
OPENCV_GPU_IMPLEMENT_WARP_TEX(ushort4)
OPENCV_GPU_IMPLEMENT_WARP_TEX(short)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(short2)
OPENCV_GPU_IMPLEMENT_WARP_TEX(short4)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(int)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(int2)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(int4)
OPENCV_GPU_IMPLEMENT_WARP_TEX(float)
//OPENCV_GPU_IMPLEMENT_WARP_TEX(float2)
OPENCV_GPU_IMPLEMENT_WARP_TEX(float4)
#undef OPENCV_GPU_IMPLEMENT_WARP_TEX
template <class Transform, template <typename> class Filter, template <typename> class B, typename T> struct WarpDispatcher
{
static void call(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, DevMem2D_<T> dst, const float* borderValue, cudaStream_t stream, int cc)
{
if (stream == 0)
WarpDispatcherNonStream<Transform, Filter, B, T>::call(src, srcWhole, xoff, yoff, dst, borderValue, cc);
else
WarpDispatcherStream<Transform, Filter, B, T>::call(src, dst, borderValue, stream, cc);
}
};
template <class Transform, typename T>
void warp_caller(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc)
{
typedef void (*func_t)(DevMem2D_<T> src, DevMem2D_<T> srcWhole, int xoff, int yoff, DevMem2D_<T> dst, const float* borderValue, cudaStream_t stream, int cc);
static const func_t funcs[3][5] =
{
{
WarpDispatcher<Transform, PointFilter, BrdReflect101, T>::call,
WarpDispatcher<Transform, PointFilter, BrdReplicate, T>::call,
WarpDispatcher<Transform, PointFilter, BrdConstant, T>::call,
WarpDispatcher<Transform, PointFilter, BrdReflect, T>::call,
WarpDispatcher<Transform, PointFilter, BrdWrap, T>::call
},
{
WarpDispatcher<Transform, LinearFilter, BrdReflect101, T>::call,
WarpDispatcher<Transform, LinearFilter, BrdReplicate, T>::call,
WarpDispatcher<Transform, LinearFilter, BrdConstant, T>::call,
WarpDispatcher<Transform, LinearFilter, BrdReflect, T>::call,
WarpDispatcher<Transform, LinearFilter, BrdWrap, T>::call
},
{
WarpDispatcher<Transform, CubicFilter, BrdReflect101, T>::call,
WarpDispatcher<Transform, CubicFilter, BrdReplicate, T>::call,
WarpDispatcher<Transform, CubicFilter, BrdConstant, T>::call,
WarpDispatcher<Transform, CubicFilter, BrdReflect, T>::call,
WarpDispatcher<Transform, CubicFilter, BrdWrap, T>::call
}
};
funcs[interpolation][borderMode](static_cast< DevMem2D_<T> >(src), static_cast< DevMem2D_<T> >(srcWhole), xoff, yoff,
static_cast< DevMem2D_<T> >(dst), borderValue, stream, cc);
}
template <typename T> void warpAffine_gpu(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc)
{
cudaSafeCall( cudaMemcpyToSymbol(c_warpMat, coeffs, 2 * 3 * sizeof(float)) );
warp_caller<AffineTransform, T>(src, srcWhole, xoff, yoff, dst, interpolation, borderMode, borderValue, stream, cc);
}
template void warpAffine_gpu<uchar >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<uchar2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<uchar3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<uchar4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<schar>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<char2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<char3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<char4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<ushort >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<ushort2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<ushort3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<ushort4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<short >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<short2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<short3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<short4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<int >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<int2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<int3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<int4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<float >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpAffine_gpu<float2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<float3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpAffine_gpu<float4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template <typename T> void warpPerspective_gpu(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc)
{
cudaSafeCall( cudaMemcpyToSymbol(c_warpMat, coeffs, 3 * 3 * sizeof(float)) );
warp_caller<PerspectiveTransform, T>(src, srcWhole, xoff, yoff, dst, interpolation, borderMode, borderValue, stream, cc);
}
template void warpPerspective_gpu<uchar >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<uchar2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<uchar3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<uchar4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<schar>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<char2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<char3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<char4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<ushort >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<ushort2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<ushort3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<ushort4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<short >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<short2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<short3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<short4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<int >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<int2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<int3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<int4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<float >(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
//template void warpPerspective_gpu<float2>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<float3>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
template void warpPerspective_gpu<float4>(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
} // namespace imgproc
}}} // namespace cv { namespace gpu { namespace device

331
modules/gpu/src/imgproc.cpp
View File

@@ -47,15 +47,11 @@ using namespace cv::gpu;
#if !defined (HAVE_CUDA)
void cv::gpu::remap(const GpuMat&, GpuMat&, const GpuMat&, const GpuMat&, int, int, const Scalar&, Stream&){ throw_nogpu(); }
void cv::gpu::meanShiftFiltering(const GpuMat&, GpuMat&, int, int, TermCriteria, Stream&) { throw_nogpu(); }
void cv::gpu::meanShiftProc(const GpuMat&, GpuMat&, GpuMat&, int, int, TermCriteria, Stream&) { throw_nogpu(); }
void cv::gpu::drawColorDisp(const GpuMat&, GpuMat&, int, Stream&) { throw_nogpu(); }
void cv::gpu::reprojectImageTo3D(const GpuMat&, GpuMat&, const Mat&, Stream&) { throw_nogpu(); }
void cv::gpu::resize(const GpuMat&, GpuMat&, Size, double, double, int, Stream&) { throw_nogpu(); }
void cv::gpu::copyMakeBorder(const GpuMat&, GpuMat&, int, int, int, int, int, const Scalar&, Stream&) { throw_nogpu(); }
void cv::gpu::warpAffine(const GpuMat&, GpuMat&, const Mat&, Size, int, Stream&) { throw_nogpu(); }
void cv::gpu::warpPerspective(const GpuMat&, GpuMat&, const Mat&, Size, int, Stream&) { throw_nogpu(); }
void cv::gpu::buildWarpPlaneMaps(Size, Rect, const Mat&, const Mat&, const Mat&, float, GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::buildWarpCylindricalMaps(Size, Rect, const Mat&, const Mat&, float, GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::buildWarpSphericalMaps(Size, Rect, const Mat&, const Mat&, float, GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
@@ -105,64 +101,6 @@ void cv::gpu::ImagePyramid::getLayer(GpuMat&, Size, Stream&) const { throw_nogpu
#else /* !defined (HAVE_CUDA) */
////////////////////////////////////////////////////////////////////////
// remap
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
template <typename T>
void remap_gpu(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, DevMem2Df xmap, DevMem2Df ymap, DevMem2Db dst,
int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
}
}}}
void cv::gpu::remap(const GpuMat& src, GpuMat& dst, const GpuMat& xmap, const GpuMat& ymap, int interpolation, int borderMode, const Scalar& borderValue, Stream& stream)
{
using namespace ::cv::gpu::device::imgproc;
typedef void (*caller_t)(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, DevMem2Df xmap, DevMem2Df ymap, DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
static const caller_t callers[6][4] =
{
{remap_gpu<uchar>, 0/*remap_gpu<uchar2>*/, remap_gpu<uchar3>, remap_gpu<uchar4>},
{0/*remap_gpu<schar>*/, 0/*remap_gpu<char2>*/, 0/*remap_gpu<char3>*/, 0/*remap_gpu<char4>*/},
{remap_gpu<ushort>, 0/*remap_gpu<ushort2>*/, remap_gpu<ushort3>, remap_gpu<ushort4>},
{remap_gpu<short>, 0/*remap_gpu<short2>*/, remap_gpu<short3>, remap_gpu<short4>},
{0/*remap_gpu<int>*/, 0/*remap_gpu<int2>*/, 0/*remap_gpu<int3>*/, 0/*remap_gpu<int4>*/},
{remap_gpu<float>, 0/*remap_gpu<float2>*/, remap_gpu<float3>, remap_gpu<float4>}
};
CV_Assert(src.depth() <= CV_32F && src.channels() <= 4);
CV_Assert(xmap.type() == CV_32F && ymap.type() == CV_32F && xmap.size() == ymap.size());
caller_t func = callers[src.depth()][src.channels() - 1];
CV_Assert(func != 0);
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
CV_Assert(borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP);
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderMode, gpuBorderType));
dst.create(xmap.size(), src.type());
Scalar_<float> borderValueFloat;
borderValueFloat = borderValue;
DeviceInfo info;
int cc = info.majorVersion() * 10 + info.minorVersion();
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
func(src, DevMem2Db(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, xmap, ymap,
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(stream), cc);
}
////////////////////////////////////////////////////////////////////////
// meanShiftFiltering_GPU
@@ -308,106 +246,6 @@ void cv::gpu::reprojectImageTo3D(const GpuMat& disp, GpuMat& xyzw, const Mat& Q,
reprojectImageTo3D_callers[disp.type()](disp, xyzw, Q, StreamAccessor::getStream(stream));
}
////////////////////////////////////////////////////////////////////////
// resize
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
template <typename T> void resize_gpu(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float fx, float fy,
DevMem2Db dst, int interpolation, cudaStream_t stream);
}
}}}
void cv::gpu::resize(const GpuMat& src, GpuMat& dst, Size dsize, double fx, double fy, int interpolation, Stream& s)
{
CV_Assert( src.depth() <= CV_32F && src.channels() <= 4 );
CV_Assert( interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC );
CV_Assert( !(dsize == Size()) || (fx > 0 && fy > 0) );
if( dsize == Size() )
{
dsize = Size(saturate_cast<int>(src.cols * fx), saturate_cast<int>(src.rows * fy));
}
else
{
fx = (double)dsize.width / src.cols;
fy = (double)dsize.height / src.rows;
}
dst.create(dsize, src.type());
if (dsize == src.size())
{
if (s)
s.enqueueCopy(src, dst);
else
src.copyTo(dst);
return;
}
cudaStream_t stream = StreamAccessor::getStream(s);
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
if ((src.type() == CV_8UC1 || src.type() == CV_8UC4) && (interpolation == INTER_NEAREST || interpolation == INTER_LINEAR))
{
static const int npp_inter[] = {NPPI_INTER_NN, NPPI_INTER_LINEAR, NPPI_INTER_CUBIC, 0, NPPI_INTER_LANCZOS};
NppiSize srcsz;
srcsz.width = wholeSize.width;
srcsz.height = wholeSize.height;
NppiRect srcrect;
srcrect.x = ofs.x;
srcrect.y = ofs.y;
srcrect.width = src.cols;
srcrect.height = src.rows;
NppiSize dstsz;
dstsz.width = dst.cols;
dstsz.height = dst.rows;
NppStreamHandler h(stream);
if (src.type() == CV_8UC1)
{
nppSafeCall( nppiResize_8u_C1R(src.datastart, srcsz, static_cast<int>(src.step), srcrect,
dst.ptr<Npp8u>(), static_cast<int>(dst.step), dstsz, fx, fy, npp_inter[interpolation]) );
}
else
{
nppSafeCall( nppiResize_8u_C4R(src.datastart, srcsz, static_cast<int>(src.step), srcrect,
dst.ptr<Npp8u>(), static_cast<int>(dst.step), dstsz, fx, fy, npp_inter[interpolation]) );
}
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
else
{
using namespace ::cv::gpu::device::imgproc;
typedef void (*caller_t)(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float fx, float fy, DevMem2Db dst, int interpolation, cudaStream_t stream);
static const caller_t callers[6][4] =
{
{resize_gpu<uchar>, 0/*resize_gpu<uchar2>*/, resize_gpu<uchar3>, resize_gpu<uchar4>},
{0/*resize_gpu<schar>*/, 0/*resize_gpu<char2>*/, 0/*resize_gpu<char3>*/, 0/*resize_gpu<char4>*/},
{resize_gpu<ushort>, 0/*resize_gpu<ushort2>*/, resize_gpu<ushort3>, resize_gpu<ushort4>},
{resize_gpu<short>, 0/*resize_gpu<short2>*/, resize_gpu<short3>, resize_gpu<short4>},
{0/*resize_gpu<int>*/, 0/*resize_gpu<int2>*/, 0/*resize_gpu<int3>*/, 0/*resize_gpu<int4>*/},
{resize_gpu<float>, 0/*resize_gpu<float2>*/, resize_gpu<float3>, resize_gpu<float4>}
};
callers[src.depth()][src.channels() - 1](src, DevMem2Db(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y,
static_cast<float>(fx), static_cast<float>(fy), dst, interpolation, stream);
}
}
////////////////////////////////////////////////////////////////////////
// copyMakeBorder
@@ -511,175 +349,6 @@ void cv::gpu::copyMakeBorder(const GpuMat& src, GpuMat& dst, int top, int bottom
}
}
////////////////////////////////////////////////////////////////////////
// warp
namespace
{
typedef NppStatus (*npp_warp_8u_t)(const Npp8u* pSrc, NppiSize srcSize, int srcStep, NppiRect srcRoi, Npp8u* pDst,
int dstStep, NppiRect dstRoi, const double coeffs[][3],
int interpolation);
typedef NppStatus (*npp_warp_16u_t)(const Npp16u* pSrc, NppiSize srcSize, int srcStep, NppiRect srcRoi, Npp16u* pDst,
int dstStep, NppiRect dstRoi, const double coeffs[][3],
int interpolation);
typedef NppStatus (*npp_warp_32s_t)(const Npp32s* pSrc, NppiSize srcSize, int srcStep, NppiRect srcRoi, Npp32s* pDst,
int dstStep, NppiRect dstRoi, const double coeffs[][3],
int interpolation);
typedef NppStatus (*npp_warp_32f_t)(const Npp32f* pSrc, NppiSize srcSize, int srcStep, NppiRect srcRoi, Npp32f* pDst,
int dstStep, NppiRect dstRoi, const double coeffs[][3],
int interpolation);
void nppWarpCaller(const GpuMat& src, GpuMat& dst, double coeffs[][3], const Size& dsize, int flags,
npp_warp_8u_t npp_warp_8u[][2], npp_warp_16u_t npp_warp_16u[][2],
npp_warp_32s_t npp_warp_32s[][2], npp_warp_32f_t npp_warp_32f[][2], cudaStream_t stream)
{
static const int npp_inter[] = {NPPI_INTER_NN, NPPI_INTER_LINEAR, NPPI_INTER_CUBIC};
int interpolation = flags & INTER_MAX;
CV_Assert((src.depth() == CV_8U || src.depth() == CV_16U || src.depth() == CV_32S || src.depth() == CV_32F) && src.channels() != 2);
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
dst.create(dsize, src.type());
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
NppiSize srcsz;
srcsz.height = wholeSize.height;
srcsz.width = wholeSize.width;
NppiRect srcroi;
srcroi.x = ofs.x;
srcroi.y = ofs.y;
srcroi.height = src.rows;
srcroi.width = src.cols;
NppiRect dstroi;
dstroi.x = dstroi.y = 0;
dstroi.height = dst.rows;
dstroi.width = dst.cols;
int warpInd = (flags & WARP_INVERSE_MAP) >> 4;
NppStreamHandler h(stream);
switch (src.depth())
{
case CV_8U:
nppSafeCall( npp_warp_8u[src.channels()][warpInd]((Npp8u*)src.datastart, srcsz, static_cast<int>(src.step), srcroi,
dst.ptr<Npp8u>(), static_cast<int>(dst.step), dstroi, coeffs, npp_inter[interpolation]) );
break;
case CV_16U:
nppSafeCall( npp_warp_16u[src.channels()][warpInd]((Npp16u*)src.datastart, srcsz, static_cast<int>(src.step), srcroi,
dst.ptr<Npp16u>(), static_cast<int>(dst.step), dstroi, coeffs, npp_inter[interpolation]) );
break;
case CV_32S:
nppSafeCall( npp_warp_32s[src.channels()][warpInd]((Npp32s*)src.datastart, srcsz, static_cast<int>(src.step), srcroi,
dst.ptr<Npp32s>(), static_cast<int>(dst.step), dstroi, coeffs, npp_inter[interpolation]) );
break;
case CV_32F:
nppSafeCall( npp_warp_32f[src.channels()][warpInd]((Npp32f*)src.datastart, srcsz, static_cast<int>(src.step), srcroi,
dst.ptr<Npp32f>(), static_cast<int>(dst.step), dstroi, coeffs, npp_inter[interpolation]) );
break;
default:
CV_Assert(!"Unsupported source type");
}
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsize, int flags, Stream& s)
{
static npp_warp_8u_t npp_warpAffine_8u[][2] =
{
{0, 0},
{nppiWarpAffine_8u_C1R, nppiWarpAffineBack_8u_C1R},
{0, 0},
{nppiWarpAffine_8u_C3R, nppiWarpAffineBack_8u_C3R},
{nppiWarpAffine_8u_C4R, nppiWarpAffineBack_8u_C4R}
};
static npp_warp_16u_t npp_warpAffine_16u[][2] =
{
{0, 0},
{nppiWarpAffine_16u_C1R, nppiWarpAffineBack_16u_C1R},
{0, 0},
{nppiWarpAffine_16u_C3R, nppiWarpAffineBack_16u_C3R},
{nppiWarpAffine_16u_C4R, nppiWarpAffineBack_16u_C4R}
};
static npp_warp_32s_t npp_warpAffine_32s[][2] =
{
{0, 0},
{nppiWarpAffine_32s_C1R, nppiWarpAffineBack_32s_C1R},
{0, 0},
{nppiWarpAffine_32s_C3R, nppiWarpAffineBack_32s_C3R},
{nppiWarpAffine_32s_C4R, nppiWarpAffineBack_32s_C4R}
};
static npp_warp_32f_t npp_warpAffine_32f[][2] =
{
{0, 0},
{nppiWarpAffine_32f_C1R, nppiWarpAffineBack_32f_C1R},
{0, 0},
{nppiWarpAffine_32f_C3R, nppiWarpAffineBack_32f_C3R},
{nppiWarpAffine_32f_C4R, nppiWarpAffineBack_32f_C4R}
};
CV_Assert(M.rows == 2 && M.cols == 3);
double coeffs[2][3];
Mat coeffsMat(2, 3, CV_64F, (void*)coeffs);
M.convertTo(coeffsMat, coeffsMat.type());
nppWarpCaller(src, dst, coeffs, dsize, flags, npp_warpAffine_8u, npp_warpAffine_16u, npp_warpAffine_32s, npp_warpAffine_32f, StreamAccessor::getStream(s));
}
void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsize, int flags, Stream& s)
{
static npp_warp_8u_t npp_warpPerspective_8u[][2] =
{
{0, 0},
{nppiWarpPerspective_8u_C1R, nppiWarpPerspectiveBack_8u_C1R},
{0, 0},
{nppiWarpPerspective_8u_C3R, nppiWarpPerspectiveBack_8u_C3R},
{nppiWarpPerspective_8u_C4R, nppiWarpPerspectiveBack_8u_C4R}
};
static npp_warp_16u_t npp_warpPerspective_16u[][2] =
{
{0, 0},
{nppiWarpPerspective_16u_C1R, nppiWarpPerspectiveBack_16u_C1R},
{0, 0},
{nppiWarpPerspective_16u_C3R, nppiWarpPerspectiveBack_16u_C3R},
{nppiWarpPerspective_16u_C4R, nppiWarpPerspectiveBack_16u_C4R}
};
static npp_warp_32s_t npp_warpPerspective_32s[][2] =
{
{0, 0},
{nppiWarpPerspective_32s_C1R, nppiWarpPerspectiveBack_32s_C1R},
{0, 0},
{nppiWarpPerspective_32s_C3R, nppiWarpPerspectiveBack_32s_C3R},
{nppiWarpPerspective_32s_C4R, nppiWarpPerspectiveBack_32s_C4R}
};
static npp_warp_32f_t npp_warpPerspective_32f[][2] =
{
{0, 0},
{nppiWarpPerspective_32f_C1R, nppiWarpPerspectiveBack_32f_C1R},
{0, 0},
{nppiWarpPerspective_32f_C3R, nppiWarpPerspectiveBack_32f_C3R},
{nppiWarpPerspective_32f_C4R, nppiWarpPerspectiveBack_32f_C4R}
};
CV_Assert(M.rows == 3 && M.cols == 3);
double coeffs[3][3];
Mat coeffsMat(3, 3, CV_64F, (void*)coeffs);
M.convertTo(coeffsMat, coeffsMat.type());
nppWarpCaller(src, dst, coeffs, dsize, flags, npp_warpPerspective_8u, npp_warpPerspective_16u, npp_warpPerspective_32s, npp_warpPerspective_32f, StreamAccessor::getStream(s));
}
//////////////////////////////////////////////////////////////////////////////
// buildWarpPlaneMaps

30
modules/gpu/src/opencv2/gpu/device/filters.hpp
View File

@@ -46,8 +46,9 @@
#include "saturate_cast.hpp"
#include "vec_traits.hpp"
#include "vec_math.hpp"
#include "type_traits.hpp"
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
template <typename Ptr2D> struct PointFilter
{
@@ -55,10 +56,10 @@ namespace cv { namespace gpu { namespace device
typedef float index_type;
explicit __host__ __device__ __forceinline__ PointFilter(const Ptr2D& src_) : src(src_) {}
__device__ __forceinline__ elem_type operator ()(float y, float x) const
{
return src(__float2int_rn(y), __float2int_rn(x));
return src(__float2int_rd(y), __float2int_rd(x));
}
const Ptr2D src;
@@ -77,6 +78,9 @@ namespace cv { namespace gpu { namespace device
work_type out = VecTraits<work_type>::all(0);
x -= 0.5f;
y -= 0.5f;
const int x1 = __float2int_rd(x);
const int y1 = __float2int_rd(y);
const int x2 = x1 + 1;
@@ -107,8 +111,8 @@ namespace cv { namespace gpu { namespace device
typedef typename TypeVec<float, VecTraits<elem_type>::cn>::vec_type work_type;
explicit __host__ __device__ __forceinline__ CubicFilter(const Ptr2D& src_) : src(src_) {}
static __device__ __forceinline__ work_type cubicInterpolate(const work_type& p0, const work_type& p1, const work_type& p2, const work_type& p3, float x)
static __device__ __forceinline__ work_type cubicInterpolate(typename TypeTraits<work_type>::ParameterType p0, typename TypeTraits<work_type>::ParameterType p1, typename TypeTraits<work_type>::ParameterType p2, typename TypeTraits<work_type>::ParameterType p3, float x)
{
return p1 + 0.5f * x * (p2 - p0 + x * (2.0f * p0 - 5.0f * p1 + 4.0f * p2 - p3 + x * (3.0f * (p1 - p2) + p3 - p0)));
}
@@ -117,15 +121,15 @@ namespace cv { namespace gpu { namespace device
{
const int xi = __float2int_rn(x);
const int yi = __float2int_rn(y);
work_type arr[4];
arr[0] = cubicInterpolate(saturate_cast<work_type>(src(yi - 1, xi - 1)), saturate_cast<work_type>(src(yi - 1, xi)), saturate_cast<work_type>(src(yi - 1, xi + 1)), saturate_cast<work_type>(src(yi - 1, xi + 2)), x - xi);
arr[1] = cubicInterpolate(saturate_cast<work_type>(src(yi , xi - 1)), saturate_cast<work_type>(src(yi , xi)), saturate_cast<work_type>(src(yi , xi + 1)), saturate_cast<work_type>(src(yi , xi + 2)), x - xi);
arr[2] = cubicInterpolate(saturate_cast<work_type>(src(yi + 1, xi - 1)), saturate_cast<work_type>(src(yi + 1, xi)), saturate_cast<work_type>(src(yi + 1, xi + 1)), saturate_cast<work_type>(src(yi + 1, xi + 2)), x - xi);
arr[3] = cubicInterpolate(saturate_cast<work_type>(src(yi + 2, xi - 1)), saturate_cast<work_type>(src(yi + 2, xi)), saturate_cast<work_type>(src(yi + 2, xi + 1)), saturate_cast<work_type>(src(yi + 2, xi + 2)), x - xi);
return saturate_cast<elem_type>(cubicInterpolate(arr[0], arr[1], arr[2], arr[3], y - yi));
arr[0] = cubicInterpolate(saturate_cast<work_type>(src(yi - 2, xi - 2)), saturate_cast<work_type>(src(yi - 2, xi - 1)), saturate_cast<work_type>(src(yi - 2, xi)), saturate_cast<work_type>(src(yi - 2, xi + 1)), (x - xi + 2.0f) / 4.0f);
arr[1] = cubicInterpolate(saturate_cast<work_type>(src(yi - 1, xi - 2)), saturate_cast<work_type>(src(yi - 1, xi - 1)), saturate_cast<work_type>(src(yi - 1, xi)), saturate_cast<work_type>(src(yi - 1, xi + 1)), (x - xi + 2.0f) / 4.0f);
arr[2] = cubicInterpolate(saturate_cast<work_type>(src(yi , xi - 2)), saturate_cast<work_type>(src(yi , xi - 1)), saturate_cast<work_type>(src(yi , xi)), saturate_cast<work_type>(src(yi , xi + 1)), (x - xi + 2.0f) / 4.0f);
arr[3] = cubicInterpolate(saturate_cast<work_type>(src(yi + 1, xi - 2)), saturate_cast<work_type>(src(yi + 1, xi - 1)), saturate_cast<work_type>(src(yi + 1, xi)), saturate_cast<work_type>(src(yi + 1, xi + 1)), (x - xi + 2.0f) / 4.0f);
return saturate_cast<elem_type>(cubicInterpolate(arr[0], arr[1], arr[2], arr[3], (y - yi + 2.0f) / 4.0f));
}
const Ptr2D src;

105
modules/gpu/src/remap.cpp Normal file
View File

@@ -0,0 +1,105 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#ifndef HAVE_CUDA
void cv::gpu::remap(const GpuMat&, GpuMat&, const GpuMat&, const GpuMat&, int, int, Scalar, Stream&){ throw_nogpu(); }
#else // HAVE_CUDA
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
template <typename T>
void remap_gpu(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, DevMem2Df xmap, DevMem2Df ymap, DevMem2Db dst,
int interpolation, int borderMode, const float* borderValue, cudaStream_t stream, int cc);
}
}}}
void cv::gpu::remap(const GpuMat& src, GpuMat& dst, const GpuMat& xmap, const GpuMat& ymap, int interpolation, int borderMode, Scalar borderValue, Stream& stream)
{
using namespace cv::gpu::device::imgproc;
typedef void (*func_t)(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, DevMem2Df xmap, DevMem2Df ymap, DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
static const func_t funcs[6][4] =
{
{remap_gpu<uchar> , 0 /*remap_gpu<uchar2>*/ , remap_gpu<uchar3> , remap_gpu<uchar4> },
{0 /*remap_gpu<schar>*/, 0 /*remap_gpu<char2>*/ , 0 /*remap_gpu<char3>*/, 0 /*remap_gpu<char4>*/},
{remap_gpu<ushort> , 0 /*remap_gpu<ushort2>*/, remap_gpu<ushort3> , remap_gpu<ushort4> },
{remap_gpu<short> , 0 /*remap_gpu<short2>*/ , remap_gpu<short3> , remap_gpu<short4> },
{0 /*remap_gpu<int>*/ , 0 /*remap_gpu<int2>*/ , 0 /*remap_gpu<int3>*/ , 0 /*remap_gpu<int4>*/ },
{remap_gpu<float> , 0 /*remap_gpu<float2>*/ , remap_gpu<float3> , remap_gpu<float4> }
};
CV_Assert(src.depth() <= CV_32F && src.channels() <= 4);
CV_Assert(xmap.type() == CV_32F && ymap.type() == CV_32F && xmap.size() == ymap.size());
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
CV_Assert(borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP);
const func_t func = funcs[src.depth()][src.channels() - 1];
CV_Assert(func != 0);
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderMode, gpuBorderType));
dst.create(xmap.size(), src.type());
Scalar_<float> borderValueFloat;
borderValueFloat = borderValue;
DeviceInfo info;
int cc = info.majorVersion() * 10 + info.minorVersion();
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
func(src, DevMem2Db(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, xmap, ymap,
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(stream), cc);
}
#endif // HAVE_CUDA

150
modules/gpu/src/resize.cpp Normal file
View File

@@ -0,0 +1,150 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#ifndef HAVE_CUDA
void cv::gpu::resize(const GpuMat&, GpuMat&, Size, double, double, int, Stream&) { throw_nogpu(); }
#else // HAVE_CUDA
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
template <typename T>
void resize_gpu(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float fx, float fy,
DevMem2Db dst, int interpolation, cudaStream_t stream);
}
}}}
void cv::gpu::resize(const GpuMat& src, GpuMat& dst, Size dsize, double fx, double fy, int interpolation, Stream& s)
{
CV_Assert(src.depth() <= CV_32F && src.channels() <= 4);
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
CV_Assert(!(dsize == Size()) || (fx > 0 && fy > 0));
if (dsize == Size())
dsize = Size(saturate_cast<int>(src.cols * fx), saturate_cast<int>(src.rows * fy));
else
{
fx = static_cast<double>(dsize.width) / src.cols;
fy = static_cast<double>(dsize.height) / src.rows;
}
dst.create(dsize, src.type());
if (dsize == src.size())
{
if (s)
s.enqueueCopy(src, dst);
else
src.copyTo(dst);
return;
}
cudaStream_t stream = StreamAccessor::getStream(s);
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
bool useNpp = (src.type() == CV_8UC1 || src.type() == CV_8UC4);
useNpp = useNpp && (interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || src.type() == CV_8UC4);
if (useNpp)
{
typedef NppStatus (*func_t)(const Npp8u * pSrc, NppiSize oSrcSize, int nSrcStep, NppiRect oSrcROI, Npp8u * pDst, int nDstStep, NppiSize dstROISize,
double xFactor, double yFactor, int eInterpolation);
const func_t funcs[4] = { nppiResize_8u_C1R, 0, 0, nppiResize_8u_C4R };
static const int npp_inter[] = {NPPI_INTER_NN, NPPI_INTER_LINEAR, NPPI_INTER_CUBIC, 0, NPPI_INTER_LANCZOS};
NppiSize srcsz;
srcsz.width = wholeSize.width;
srcsz.height = wholeSize.height;
NppiRect srcrect;
srcrect.x = ofs.x;
srcrect.y = ofs.y;
srcrect.width = src.cols;
srcrect.height = src.rows;
NppiSize dstsz;
dstsz.width = dst.cols;
dstsz.height = dst.rows;
NppStreamHandler h(stream);
nppSafeCall( funcs[src.channels() - 1](src.datastart, srcsz, static_cast<int>(src.step), srcrect,
dst.ptr<Npp8u>(), static_cast<int>(dst.step), dstsz, fx, fy, npp_inter[interpolation]) );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
else
{
using namespace ::cv::gpu::device::imgproc;
typedef void (*func_t)(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float fx, float fy, DevMem2Db dst, int interpolation, cudaStream_t stream);
static const func_t funcs[6][4] =
{
{resize_gpu<uchar> , 0 /*resize_gpu<uchar2>*/ , resize_gpu<uchar3> , resize_gpu<uchar4> },
{0 /*resize_gpu<schar>*/, 0 /*resize_gpu<char2>*/ , 0 /*resize_gpu<char3>*/, 0 /*resize_gpu<char4>*/},
{resize_gpu<ushort> , 0 /*resize_gpu<ushort2>*/, resize_gpu<ushort3> , resize_gpu<ushort4> },
{resize_gpu<short> , 0 /*resize_gpu<short2>*/ , resize_gpu<short3> , resize_gpu<short4> },
{0 /*resize_gpu<int>*/ , 0 /*resize_gpu<int2>*/ , 0 /*resize_gpu<int3>*/ , 0 /*resize_gpu<int4>*/ },
{resize_gpu<float> , 0 /*resize_gpu<float2>*/ , resize_gpu<float3> , resize_gpu<float4> }
};
const func_t func = funcs[src.depth()][src.channels() - 1];
CV_Assert(func != 0);
func(src, DevMem2Db(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y,
static_cast<float>(1.0 / fx), static_cast<float>(1.0 / fy), dst, interpolation, stream);
}
}
#endif // HAVE_CUDA

463
modules/gpu/src/warp.cpp Normal file
View File

@@ -0,0 +1,463 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#ifndef HAVE_CUDA
void cv::gpu::warpAffine(const GpuMat&, GpuMat&, const Mat&, Size, int, int, Scalar, Stream&) { throw_nogpu(); }
void cv::gpu::buildWarpAffineMaps(const Mat&, bool, Size, GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::warpPerspective(const GpuMat&, GpuMat&, const Mat&, Size, int, int, Scalar, Stream&) { throw_nogpu(); }
void cv::gpu::buildWarpPerspectiveMaps(const Mat&, bool, Size, GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
#else // HAVE_CUDA
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
void buildWarpAffineMaps_gpu(float coeffs[2 * 3], DevMem2Df xmap, DevMem2Df ymap, cudaStream_t stream);
template <typename T>
void warpAffine_gpu(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
void buildWarpPerspectiveMaps_gpu(float coeffs[3 * 3], DevMem2Df xmap, DevMem2Df ymap, cudaStream_t stream);
template <typename T>
void warpPerspective_gpu(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[3 * 3], DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
}
}}}
void cv::gpu::buildWarpAffineMaps(const Mat& M, bool inverse, Size dsize, GpuMat& xmap, GpuMat& ymap, Stream& stream)
{
using namespace cv::gpu::device::imgproc;
CV_Assert(M.rows == 2 && M.cols == 3);
xmap.create(dsize, CV_32FC1);
ymap.create(dsize, CV_32FC1);
float coeffs[2 * 3];
Mat coeffsMat(2, 3, CV_32F, (void*)coeffs);
if (inverse)
M.convertTo(coeffsMat, coeffsMat.type());
else
{
cv::Mat iM;
invertAffineTransform(M, iM);
iM.convertTo(coeffsMat, coeffsMat.type());
}
buildWarpAffineMaps_gpu(coeffs, xmap, ymap, StreamAccessor::getStream(stream));
}
void cv::gpu::buildWarpPerspectiveMaps(const Mat& M, bool inverse, Size dsize, GpuMat& xmap, GpuMat& ymap, Stream& stream)
{
using namespace cv::gpu::device::imgproc;
CV_Assert(M.rows == 3 && M.cols == 3);
xmap.create(dsize, CV_32FC1);
ymap.create(dsize, CV_32FC1);
float coeffs[3 * 3];
Mat coeffsMat(3, 3, CV_32F, (void*)coeffs);
if (inverse)
M.convertTo(coeffsMat, coeffsMat.type());
else
{
cv::Mat iM;
invert(M, iM);
iM.convertTo(coeffsMat, coeffsMat.type());
}
buildWarpPerspectiveMaps_gpu(coeffs, xmap, ymap, StreamAccessor::getStream(stream));
}
namespace
{
template<int DEPTH> struct NppTypeTraits;
template<> struct NppTypeTraits<CV_8U> { typedef Npp8u npp_t; };
template<> struct NppTypeTraits<CV_8S> { typedef Npp8s npp_t; };
template<> struct NppTypeTraits<CV_16U> { typedef Npp16u npp_t; };
template<> struct NppTypeTraits<CV_16S> { typedef Npp16s npp_t; typedef Npp16sc npp_complex_type; };
template<> struct NppTypeTraits<CV_32S> { typedef Npp32s npp_t; typedef Npp32sc npp_complex_type; };
template<> struct NppTypeTraits<CV_32F> { typedef Npp32f npp_t; typedef Npp32fc npp_complex_type; };
template<> struct NppTypeTraits<CV_64F> { typedef Npp64f npp_t; typedef Npp64fc npp_complex_type; };
template <int DEPTH> struct NppWarpFunc
{
typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
typedef NppStatus (*func_t)(const npp_t* pSrc, NppiSize srcSize, int srcStep, NppiRect srcRoi, npp_t* pDst,
int dstStep, NppiRect dstRoi, const double coeffs[][3],
int interpolation);
};
template <int DEPTH, typename NppWarpFunc<DEPTH>::func_t func> struct NppWarp
{
typedef typename NppWarpFunc<DEPTH>::npp_t npp_t;
static void call(const cv::gpu::GpuMat& src, cv::Size wholeSize, cv::Point ofs, cv::gpu::GpuMat& dst,
double coeffs[][3], cv::Size dsize, int interpolation, cudaStream_t stream)
{
static const int npp_inter[] = {NPPI_INTER_NN, NPPI_INTER_LINEAR, NPPI_INTER_CUBIC};
dst.create(dsize, src.type());
dst.setTo(cv::Scalar::all(0));
NppiSize srcsz;
srcsz.height = wholeSize.height;
srcsz.width = wholeSize.width;
NppiRect srcroi;
srcroi.x = ofs.x;
srcroi.y = ofs.y;
srcroi.height = src.rows;
srcroi.width = src.cols;
NppiRect dstroi;
dstroi.x = dstroi.y = 0;
dstroi.height = dst.rows;
dstroi.width = dst.cols;
cv::gpu::NppStreamHandler h(stream);
nppSafeCall( func((npp_t*)src.datastart, srcsz, static_cast<int>(src.step), srcroi,
dst.ptr<npp_t>(), static_cast<int>(dst.step), dstroi, coeffs, npp_inter[interpolation]) );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
};
}
void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsize, int flags, int borderMode, Scalar borderValue, Stream& s)
{
CV_Assert(M.rows == 2 && M.cols == 3);
int interpolation = flags & INTER_MAX;
CV_Assert(src.depth() <= CV_32F && src.channels() <= 4);
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
CV_Assert(borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP);
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
static const bool useNppTab[6][4][3] =
{
{
{false, false, true},
{false, false, false},
{false, true, true},
{false, false, false}
},
{
{false, false, false},
{false, false, false},
{false, false, false},
{false, false, false}
},
{
{false, true, true},
{false, false, false},
{false, true, true},
{false, false, false}
},
{
{false, false, false},
{false, false, false},
{false, false, false},
{false, false, false}
},
{
{false, true, true},
{false, false, false},
{false, true, true},
{false, false, true}
},
{
{false, true, true},
{false, false, false},
{false, true, true},
{false, false, true}
}
};
bool useNpp = borderMode == BORDER_CONSTANT;
useNpp = useNpp && useNppTab[src.depth()][src.channels() - 1][interpolation];
#ifdef linux
// NPP bug on float data
useNpp = useNpp && src.depth() != CV_32F;
#endif
if (useNpp)
{
typedef void (*func_t)(const cv::gpu::GpuMat& src, cv::Size wholeSize, cv::Point ofs, cv::gpu::GpuMat& dst, double coeffs[][3], cv::Size dsize, int flags, cudaStream_t stream);
static const func_t funcs[2][6][4] =
{
{
{NppWarp<CV_8U, nppiWarpAffine_8u_C1R>::call, 0, NppWarp<CV_8U, nppiWarpAffine_8u_C3R>::call, NppWarp<CV_8U, nppiWarpAffine_8u_C4R>::call},
{0, 0, 0, 0},
{NppWarp<CV_16U, nppiWarpAffine_16u_C1R>::call, 0, NppWarp<CV_16U, nppiWarpAffine_16u_C3R>::call, NppWarp<CV_16U, nppiWarpAffine_16u_C4R>::call},
{0, 0, 0, 0},
{NppWarp<CV_32S, nppiWarpAffine_32s_C1R>::call, 0, NppWarp<CV_32S, nppiWarpAffine_32s_C3R>::call, NppWarp<CV_32S, nppiWarpAffine_32s_C4R>::call},
{NppWarp<CV_32F, nppiWarpAffine_32f_C1R>::call, 0, NppWarp<CV_32F, nppiWarpAffine_32f_C3R>::call, NppWarp<CV_32F, nppiWarpAffine_32f_C4R>::call}
},
{
{NppWarp<CV_8U, nppiWarpAffineBack_8u_C1R>::call, 0, NppWarp<CV_8U, nppiWarpAffineBack_8u_C3R>::call, NppWarp<CV_8U, nppiWarpAffineBack_8u_C4R>::call},
{0, 0, 0, 0},
{NppWarp<CV_16U, nppiWarpAffineBack_16u_C1R>::call, 0, NppWarp<CV_16U, nppiWarpAffineBack_16u_C3R>::call, NppWarp<CV_16U, nppiWarpAffineBack_16u_C4R>::call},
{0, 0, 0, 0},
{NppWarp<CV_32S, nppiWarpAffineBack_32s_C1R>::call, 0, NppWarp<CV_32S, nppiWarpAffineBack_32s_C3R>::call, NppWarp<CV_32S, nppiWarpAffineBack_32s_C4R>::call},
{NppWarp<CV_32F, nppiWarpAffineBack_32f_C1R>::call, 0, NppWarp<CV_32F, nppiWarpAffineBack_32f_C3R>::call, NppWarp<CV_32F, nppiWarpAffineBack_32f_C4R>::call}
}
};
double coeffs[2][3];
Mat coeffsMat(2, 3, CV_64F, (void*)coeffs);
M.convertTo(coeffsMat, coeffsMat.type());
const func_t func = funcs[(flags & WARP_INVERSE_MAP) != 0][src.depth()][src.channels() - 1];
CV_Assert(func != 0);
func(src, wholeSize, ofs, dst, coeffs, dsize, interpolation, StreamAccessor::getStream(s));
}
else
{
using namespace cv::gpu::device::imgproc;
typedef void (*func_t)(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
static const func_t funcs[6][4] =
{
{warpAffine_gpu<uchar> , 0 /*warpAffine_gpu<uchar2>*/ , warpAffine_gpu<uchar3> , warpAffine_gpu<uchar4> },
{0 /*warpAffine_gpu<schar>*/, 0 /*warpAffine_gpu<char2>*/ , 0 /*warpAffine_gpu<char3>*/, 0 /*warpAffine_gpu<char4>*/},
{warpAffine_gpu<ushort> , 0 /*warpAffine_gpu<ushort2>*/, warpAffine_gpu<ushort3> , warpAffine_gpu<ushort4> },
{warpAffine_gpu<short> , 0 /*warpAffine_gpu<short2>*/ , warpAffine_gpu<short3> , warpAffine_gpu<short4> },
{0 /*warpAffine_gpu<int>*/ , 0 /*warpAffine_gpu<int2>*/ , 0 /*warpAffine_gpu<int3>*/ , 0 /*warpAffine_gpu<int4>*/ },
{warpAffine_gpu<float> , 0 /*warpAffine_gpu<float2>*/ , warpAffine_gpu<float3> , warpAffine_gpu<float4> }
};
const func_t func = funcs[src.depth()][src.channels() - 1];
CV_Assert(func != 0);
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderMode, gpuBorderType));
dst.create(dsize, src.type());
float coeffs[2 * 3];
Mat coeffsMat(2, 3, CV_32F, (void*)coeffs);
if (flags & WARP_INVERSE_MAP)
M.convertTo(coeffsMat, coeffsMat.type());
else
{
cv::Mat iM;
invertAffineTransform(M, iM);
iM.convertTo(coeffsMat, coeffsMat.type());
}
Scalar_<float> borderValueFloat;
borderValueFloat = borderValue;
DeviceInfo info;
int cc = info.majorVersion() * 10 + info.minorVersion();
func(src, DevMem2Db(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, coeffs,
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(s), cc);
}
}
void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsize, int flags, int borderMode, Scalar borderValue, Stream& s)
{
CV_Assert(M.rows == 3 && M.cols == 3);
int interpolation = flags & INTER_MAX;
CV_Assert(src.depth() <= CV_32F && src.channels() <= 4);
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
CV_Assert(borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP);
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
static const bool useNppTab[6][4][3] =
{
{
{false, false, true},
{false, false, false},
{false, true, true},
{false, false, false}
},
{
{false, false, false},
{false, false, false},
{false, false, false},
{false, false, false}
},
{
{false, true, true},
{false, false, false},
{false, true, true},
{false, false, false}
},
{
{false, false, false},
{false, false, false},
{false, false, false},
{false, false, false}
},
{
{false, true, true},
{false, false, false},
{false, true, true},
{false, false, true}
},
{
{false, true, true},
{false, false, false},
{false, true, true},
{false, false, true}
}
};
bool useNpp = borderMode == BORDER_CONSTANT;
useNpp = useNpp && useNppTab[src.depth()][src.channels() - 1][interpolation];
#ifdef linux
// NPP bug on float data
useNpp = useNpp && src.depth() != CV_32F;
#endif
if (useNpp)
{
typedef void (*func_t)(const cv::gpu::GpuMat& src, cv::Size wholeSize, cv::Point ofs, cv::gpu::GpuMat& dst, double coeffs[][3], cv::Size dsize, int flags, cudaStream_t stream);
static const func_t funcs[2][6][4] =
{
{
{NppWarp<CV_8U, nppiWarpPerspective_8u_C1R>::call, 0, NppWarp<CV_8U, nppiWarpPerspective_8u_C3R>::call, NppWarp<CV_8U, nppiWarpPerspective_8u_C4R>::call},
{0, 0, 0, 0},
{NppWarp<CV_16U, nppiWarpPerspective_16u_C1R>::call, 0, NppWarp<CV_16U, nppiWarpPerspective_16u_C3R>::call, NppWarp<CV_16U, nppiWarpPerspective_16u_C4R>::call},
{0, 0, 0, 0},
{NppWarp<CV_32S, nppiWarpPerspective_32s_C1R>::call, 0, NppWarp<CV_32S, nppiWarpPerspective_32s_C3R>::call, NppWarp<CV_32S, nppiWarpPerspective_32s_C4R>::call},
{NppWarp<CV_32F, nppiWarpPerspective_32f_C1R>::call, 0, NppWarp<CV_32F, nppiWarpPerspective_32f_C3R>::call, NppWarp<CV_32F, nppiWarpPerspective_32f_C4R>::call}
},
{
{NppWarp<CV_8U, nppiWarpPerspectiveBack_8u_C1R>::call, 0, NppWarp<CV_8U, nppiWarpPerspectiveBack_8u_C3R>::call, NppWarp<CV_8U, nppiWarpPerspectiveBack_8u_C4R>::call},
{0, 0, 0, 0},
{NppWarp<CV_16U, nppiWarpPerspectiveBack_16u_C1R>::call, 0, NppWarp<CV_16U, nppiWarpPerspectiveBack_16u_C3R>::call, NppWarp<CV_16U, nppiWarpPerspectiveBack_16u_C4R>::call},
{0, 0, 0, 0},
{NppWarp<CV_32S, nppiWarpPerspectiveBack_32s_C1R>::call, 0, NppWarp<CV_32S, nppiWarpPerspectiveBack_32s_C3R>::call, NppWarp<CV_32S, nppiWarpPerspectiveBack_32s_C4R>::call},
{NppWarp<CV_32F, nppiWarpPerspectiveBack_32f_C1R>::call, 0, NppWarp<CV_32F, nppiWarpPerspectiveBack_32f_C3R>::call, NppWarp<CV_32F, nppiWarpPerspectiveBack_32f_C4R>::call}
}
};
double coeffs[3][3];
Mat coeffsMat(3, 3, CV_64F, (void*)coeffs);
M.convertTo(coeffsMat, coeffsMat.type());
const func_t func = funcs[(flags & WARP_INVERSE_MAP) != 0][src.depth()][src.channels() - 1];
CV_Assert(func != 0);
func(src, wholeSize, ofs, dst, coeffs, dsize, interpolation, StreamAccessor::getStream(s));
}
else
{
using namespace cv::gpu::device::imgproc;
typedef void (*func_t)(DevMem2Db src, DevMem2Db srcWhole, int xoff, int yoff, float coeffs[2 * 3], DevMem2Db dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
static const func_t funcs[6][4] =
{
{warpPerspective_gpu<uchar> , 0 /*warpPerspective_gpu<uchar2>*/ , warpPerspective_gpu<uchar3> , warpPerspective_gpu<uchar4> },
{0 /*warpPerspective_gpu<schar>*/, 0 /*warpPerspective_gpu<char2>*/ , 0 /*warpPerspective_gpu<char3>*/, 0 /*warpPerspective_gpu<char4>*/},
{warpPerspective_gpu<ushort> , 0 /*warpPerspective_gpu<ushort2>*/, warpPerspective_gpu<ushort3> , warpPerspective_gpu<ushort4> },
{warpPerspective_gpu<short> , 0 /*warpPerspective_gpu<short2>*/ , warpPerspective_gpu<short3> , warpPerspective_gpu<short4> },
{0 /*warpPerspective_gpu<int>*/ , 0 /*warpPerspective_gpu<int2>*/ , 0 /*warpPerspective_gpu<int3>*/ , 0 /*warpPerspective_gpu<int4>*/ },
{warpPerspective_gpu<float> , 0 /*warpPerspective_gpu<float2>*/ , warpPerspective_gpu<float3> , warpPerspective_gpu<float4> }
};
const func_t func = funcs[src.depth()][src.channels() - 1];
CV_Assert(func != 0);
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderMode, gpuBorderType));
dst.create(dsize, src.type());
float coeffs[3 * 3];
Mat coeffsMat(3, 3, CV_32F, (void*)coeffs);
if (flags & WARP_INVERSE_MAP)
M.convertTo(coeffsMat, coeffsMat.type());
else
{
cv::Mat iM;
invert(M, iM);
iM.convertTo(coeffsMat, coeffsMat.type());
}
Scalar_<float> borderValueFloat;
borderValueFloat = borderValue;
DeviceInfo info;
int cc = info.majorVersion() * 10 + info.minorVersion();
func(src, DevMem2Db(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, coeffs,
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(s), cc);
}
}
#endif // HAVE_CUDA