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

fixed backward compatibility with less than 1.2 CUDA capability

Browse Source
This commit is contained in:
Marina Kolpakova
2012-07-07 21:48:53 +00:00
parent 3ac6272c3b
commit 09269b4cd8
2 changed files with 68 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
modules/gpu/src/cuda/lbp.cu
View File

@@ -86,8 +86,11 @@ namespace cv { namespace gpu { namespace device
rect.y = roundf(y * scale);
rect.z = roundf(clWidth);
rect.w = roundf(clHeight);
int res = atomicInc(n, 100);
#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
int res = __atomicInc(n, 100U);
#else
int res = atomicInc(n, 100U);
#endif
objects(0, res) = rect;
}
@@ -111,14 +114,24 @@ namespace cv { namespace gpu { namespace device
__syncthreads();
int cls = labels[tid];
#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
__atomicAdd((int*)(rrects + cls * 4 + 0), candidates[tid].x);
__atomicAdd((int*)(rrects + cls * 4 + 1), candidates[tid].y);
__atomicAdd((int*)(rrects + cls * 4 + 2), candidates[tid].z);
__atomicAdd((int*)(rrects + cls * 4 + 3), candidates[tid].w);
#else
atomicAdd((int*)(rrects + cls * 4 + 0), candidates[tid].x);
atomicAdd((int*)(rrects + cls * 4 + 1), candidates[tid].y);
atomicAdd((int*)(rrects + cls * 4 + 2), candidates[tid].z);
atomicAdd((int*)(rrects + cls * 4 + 3), candidates[tid].w);
#endif
labels[tid] = 0;
__syncthreads();
#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
__atomicInc((unsigned int*)labels + cls, n);
#else
atomicInc((unsigned int*)labels + cls, n);
#endif
*nclasses = 0;
int active = labels[tid];
@@ -154,7 +167,11 @@ namespace cv { namespace gpu { namespace device
}
if( j == n)
{
#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
objects[__atomicInc(nclasses, n)] = VecTraits<int4>::make(r1[0], r1[1], r1[2], r1[3]);
#else
objects[atomicInc(nclasses, n)] = VecTraits<int4>::make(r1[0], r1[1], r1[2], r1[3]);
#endif
}
}
}