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

Neon optimization of cv::scaleAdd (CV_32F)

Browse Source
This commit is contained in:
Ilya Lavrenov 2014-09-23 21:14:42 +04:00
parent ecbec7235f
commit 5d018c090f
1 changed files with 10 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
modules/core/src/matmul.cpp
View File

@ -2102,6 +2102,16 @@ static void scaleAdd_32f(const float* src1, const float* src2, float* dst,
} }
} }
else else
#elif CV_NEON
if (true)
{
for ( ; i <= len - 4; i += 4)
{
float32x4_t v_src1 = vld1q_f32(src1 + i), v_src2 = vld1q_f32(src2 + i);
vst1q_f32(dst + i, vaddq_f32(vmulq_n_f32(v_src1, alpha), v_src2));
}
}
else
#endif #endif
//vz why do we need unroll here? //vz why do we need unroll here?
for( ; i <= len - 4; i += 4 ) for( ; i <= len - 4; i += 4 )