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

warpScanInclusive

Browse Source
This commit is contained in:
Vladislav Vinogradov
2012-11-16 12:49:43 +04:00
parent 76e8794e81
commit 572d2d6a84
3 changed files with 99 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
modules/gpu/include/opencv2/gpu/device/warp_shuffle.hpp
View File

@@ -54,6 +54,14 @@ namespace cv { namespace gpu { namespace device
return T();
#endif
}
__device__ __forceinline__ unsigned int shfl(unsigned int val, int srcLane, int width = warpSize)
{
#if __CUDA_ARCH__ >= 300
return (unsigned int) __shfl((int) val, srcLane, width);
#else
return 0;
#endif
}
__device__ __forceinline__ double shfl(double val, int srcLane, int width = warpSize)
{
#if __CUDA_ARCH__ >= 300
@@ -78,6 +86,14 @@ namespace cv { namespace gpu { namespace device
return T();
#endif
}
__device__ __forceinline__ unsigned int shfl_down(unsigned int val, unsigned int delta, int width = warpSize)
{
#if __CUDA_ARCH__ >= 300
return (unsigned int) __shfl_down((int) val, delta, width);
#else
return 0;
#endif
}
__device__ __forceinline__ double shfl_down(double val, unsigned int delta, int width = warpSize)
{
#if __CUDA_ARCH__ >= 300
@@ -92,6 +108,38 @@ namespace cv { namespace gpu { namespace device
return 0.0;
#endif
}
template <typename T>
__device__ __forceinline__ T shfl_up(T val, unsigned int delta, int width = warpSize)
{
#if __CUDA_ARCH__ >= 300
return __shfl_up(val, delta, width);
#else
return T();
#endif
}
__device__ __forceinline__ unsigned int shfl_up(unsigned int val, unsigned int delta, int width = warpSize)
{
#if __CUDA_ARCH__ >= 300
return (unsigned int) __shfl_up((int) val, delta, width);
#else
return 0;
#endif
}
__device__ __forceinline__ double shfl_up(double val, unsigned int delta, int width = warpSize)
{
#if __CUDA_ARCH__ >= 300
int lo = __double2loint(val);
int hi = __double2hiint(val);
lo = __shfl_up(lo, delta, width);
hi = __shfl_up(hi, delta, width);
return __hiloint2double(hi, lo);
#else
return 0.0;
#endif
}
}}}
#endif // __OPENCV_GPU_WARP_SHUFFLE_HPP__