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

SymmColumnSmallVec_32s16s [3, 10, 3] Scharr

Browse Source 
NEON speedup: 2.04x
Auto-vect speedup: 1x
This commit is contained in:
orestis 2014-12-19 22:35:52 +02:00
parent 61a7f48bf4
commit 33dfeb85be
1 changed files with 18 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
modules/imgproc/src/filter.cpp
View File

@ -2672,7 +2672,24 @@ struct SymmColumnSmallVec_32s16s
} }
else if( ky[0] == 10 && ky[1] == 3 ) else if( ky[0] == 10 && ky[1] == 3 )
{ {
return 0; for( ; i <= width - 4; i += 4 )
{
int32x4_t x0, x1, x2, x3;
x0 = vld1q_s32((int32_t const *)(S0 + i));
x1 = vld1q_s32((int32_t const *)(S1 + i));
x2 = vld1q_s32((int32_t const *)(S2 + i));
x3 = vaddq_s32(x0, x2);
int32x4_t y0;
y0 = vmlaq_n_s32(d4, x1, 10);
y0 = vmlaq_n_s32(y0, x3, 3);
int16x4_t t;
t = vqmovn_s32(y0);
vst1_s16((int16_t *)(dst + i), t);
}
} }
else else
{ {