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

Fixed GPU version of multi-band blending in stitching module

Browse Source
This commit is contained in:
Alexey Spizhevoy
2011-09-24 05:58:29 +00:00
parent afc894db9f
commit 48dec9c03a
4 changed files with 39 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
modules/gpu/src/cuda/element_operations.cu
View File

@@ -654,27 +654,18 @@ namespace cv { namespace gpu { namespace device
//////////////////////////////////////////////////////////////////////////
// multiply
struct add_16sc4 : binary_function<short4, short4, short4>
template <> struct TransformFunctorTraits< plus<short> > : DefaultTransformFunctorTraits< plus<short> >
{
__device__ __forceinline__ short4 operator ()(short4 a, short4 b) const
{
return make_short4(a.x + b.x, a.y + b.y, a.z + b.z, a.w + b.w);
}
};
template <> struct TransformFunctorTraits<add_16sc4> : DefaultTransformFunctorTraits<add_16sc4>
{
enum { smart_block_dim_x = 8 };
enum { smart_block_dim_y = 8 };
enum { smart_shift = 8 };
enum { smart_shift = 4 };
};
void add_gpu(const DevMem2D_<short4>& src1, const DevMem2D_<short4>& src2, const DevMem2D_<short4>& dst, cudaStream_t stream)
template <typename T> void add_gpu(const DevMem2D src1, const DevMem2D src2, DevMem2D dst, cudaStream_t stream)
{
transform(static_cast< DevMem2D_<short4> >(src1), static_cast< DevMem2D_<short4> >(src2),
static_cast< DevMem2D_<short4> >(dst), add_16sc4(), stream);
transform((DevMem2D_<T>)src1, (DevMem2D_<T>)src2, (DevMem2D_<T>)dst, plus<T>(), stream);
}
template void add_gpu<short>(const DevMem2D src1, const DevMem2D src2, DevMem2D dst, cudaStream_t stream);
//////////////////////////////////////////////////////////////////////////
// multiply

7
modules/gpu/src/element_operations.cpp
View File

@@ -177,16 +177,17 @@ namespace
namespace cv { namespace gpu { namespace device
{
void add_gpu(const DevMem2D_<short4>& src1, const DevMem2D_<short4>& src2, const DevMem2D_<short4>& dst, cudaStream_t stream);
template <typename T>
void add_gpu(const DevMem2D src1, const DevMem2D src2, DevMem2D dst, cudaStream_t stream);
}}}
void cv::gpu::add(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
{
if (src1.type() == CV_16SC4 && src2.type() == CV_16SC4)
if (src1.depth() == CV_16S && src2.depth() == CV_16S)
{
CV_Assert(src1.size() == src2.size());
dst.create(src1.size(), src1.type());
device::add_gpu(src1, src2, dst, StreamAccessor::getStream(stream));
device::add_gpu<short>(src1.reshape(1), src2.reshape(1), dst.reshape(1), StreamAccessor::getStream(stream));
}
else
nppArithmCaller(src1, src2, dst, nppiAdd_8u_C1RSfs, nppiAdd_8u_C4RSfs, nppiAdd_32s_C1R, nppiAdd_32f_C1R, StreamAccessor::getStream(stream));