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

Merge branch 'work'

Browse Source
This commit is contained in:
Marina Kolpakova 2012-06-19 13:41:15 +00:00
parent 18918a5e44
commit 6620c5c92d
1 changed files with 14 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
modules/gpu/src/opencv2/gpu/device/warp_reduce.hpp
View File

@ -44,25 +44,25 @@
#ifndef OPENCV_GPU_WARP_REDUCE_HPP__ #ifndef OPENCV_GPU_WARP_REDUCE_HPP__
#define OPENCV_GPU_WARP_REDUCE_HPP__ #define OPENCV_GPU_WARP_REDUCE_HPP__
namespace cv { namespace gpu { namespace device namespace cv { namespace gpu { namespace device
{ {
template <class T> template <class T>
__device__ __forceinline__ T warp_reduce(volatile T *ptr , const unsigned int tid = threadIdx.x) __device__ __forceinline__ T warp_reduce(volatile T *ptr , const unsigned int tid = threadIdx.x)
{ {
const unsigned int lane = tid & 31; // index of thread in warp (0..31) const unsigned int lane = tid & 31; // index of thread in warp (0..31)
if (lane < 16)
{
T partial = ptr[tid];
ptr[tid] = partial = partial + ptr[tid + 16]; if (lane < 16)
ptr[tid] = partial = partial + ptr[tid + 8]; {
ptr[tid] = partial = partial + ptr[tid + 4]; T partial = ptr[tid];
ptr[tid] = partial = partial + ptr[tid + 2];
ptr[tid] = partial = partial + ptr[tid + 1];
}
return ptr[tid - lane]; ptr[tid] = partial = partial + ptr[tid + 16];
ptr[tid] = partial = partial + ptr[tid + 8];
ptr[tid] = partial = partial + ptr[tid + 4];
ptr[tid] = partial = partial + ptr[tid + 2];
ptr[tid] = partial = partial + ptr[tid + 1];
}
return ptr[tid - lane];
} }
}}} // namespace cv { namespace gpu { namespace device { }}} // namespace cv { namespace gpu { namespace device {