potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 23:33:05 +04:00
Code Packages Releases Wiki Activity

new reduce and reduceKeyVal implementation

Browse Source
This commit is contained in:
Vladislav Vinogradov
2012-11-12 10:55:32 +04:00
parent d47c112434
commit 7a1874b2cc
8 changed files with 1149 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/gpu/src/cuda/orb.cu
+7 -7
View File
@@ -109,9 +109,9 @@ namespace cv { namespace gpu { namespace device
c += Ix * Iy;
}
reduce<32>(srow, a, threadIdx.x, plus<volatile int>());
reduce<32>(srow, b, threadIdx.x, plus<volatile int>());
reduce<32>(srow, c, threadIdx.x, plus<volatile int>());
reduce_old<32>(srow, a, threadIdx.x, plus<volatile int>());
reduce_old<32>(srow, b, threadIdx.x, plus<volatile int>());
reduce_old<32>(srow, c, threadIdx.x, plus<volatile int>());
if (threadIdx.x == 0)
{
@@ -167,7 +167,7 @@ namespace cv { namespace gpu { namespace device
for (int u = threadIdx.x - half_k; u <= half_k; u += blockDim.x)
m_10 += u * image(loc.y, loc.x + u);
reduce<32>(srow, m_10, threadIdx.x, plus<volatile int>());
reduce_old<32>(srow, m_10, threadIdx.x, plus<volatile int>());
for (int v = 1; v <= half_k; ++v)
{
@@ -185,8 +185,8 @@ namespace cv { namespace gpu { namespace device
m_sum += u * (val_plus + val_minus);
}
reduce<32>(srow, v_sum, threadIdx.x, plus<volatile int>());
reduce<32>(srow, m_sum, threadIdx.x, plus<volatile int>());
reduce_old<32>(srow, v_sum, threadIdx.x, plus<volatile int>());
reduce_old<32>(srow, m_sum, threadIdx.x, plus<volatile int>());
m_10 += m_sum;
m_01 += v * v_sum;
@@ -419,4 +419,4 @@ namespace cv { namespace gpu { namespace device
}
}}}
#endif /* CUDA_DISABLER */
#endif /* CUDA_DISABLER */
modules/gpu/src/cuda/surf.cu
+2 -2
View File
@@ -599,8 +599,8 @@ namespace cv { namespace gpu { namespace device
sumy += s_Y[threadIdx.x + 96];
}
device::reduce<32>(s_sumx + threadIdx.y * 32, sumx, threadIdx.x, plus<volatile float>());
device::reduce<32>(s_sumy + threadIdx.y * 32, sumy, threadIdx.x, plus<volatile float>());
device::reduce_old<32>(s_sumx + threadIdx.y * 32, sumx, threadIdx.x, plus<volatile float>());
device::reduce_old<32>(s_sumy + threadIdx.y * 32, sumy, threadIdx.x, plus<volatile float>());
const float temp_mod = sumx * sumx + sumy * sumy;
if (temp_mod > best_mod)