potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-31 00:03:03 +04:00
Code Packages Releases Wiki Activity

core(hal): improve v_select() SSE4.1+

Browse Source 
v_select 'mask' is restricted to these values only: 0 or ~0 (0xff/0xffff/etc)
mask in accuracy test is updated.
This commit is contained in:
Alexander Alekhin
2018-04-17 17:50:23 +03:00
parent 0477284b0a
commit 65726e4244
5 changed files with 50 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/src/copy.cpp
+2 -19
View File
@@ -91,11 +91,7 @@ copyMask_<uchar>(const uchar* _src, size_t sstep, const uchar* mask, size_t mste
uchar* dst = (uchar*)_dst;
int x = 0;
#if CV_SIMD128
if( hasSIMD128()
#if CV_SSE4_2
&& USE_SSE4_2
#endif
) {
{
v_uint8x16 v_zero = v_setzero_u8();
for( ; x <= size.width - 16; x += 16 )
@@ -104,11 +100,7 @@ copyMask_<uchar>(const uchar* _src, size_t sstep, const uchar* mask, size_t mste
v_dst = v_load(dst + x),
v_nmask = v_load(mask + x) == v_zero;
#if CV_SSE4_2
v_dst = v_uint8x16(_mm_blendv_epi8(v_src.val, v_dst.val, v_nmask.val));
#else
v_dst = v_select(v_nmask, v_dst, v_src);
#endif
v_store(dst + x, v_dst);
}
}
@@ -130,11 +122,7 @@ copyMask_<ushort>(const uchar* _src, size_t sstep, const uchar* mask, size_t mst
ushort* dst = (ushort*)_dst;
int x = 0;
#if CV_SIMD128
if( hasSIMD128()
#if CV_SSE4_2
&& USE_SSE4_2
#endif
) {
{
v_uint8x16 v_zero = v_setzero_u8();
for( ; x <= size.width - 16; x += 16 )
@@ -146,13 +134,8 @@ copyMask_<ushort>(const uchar* _src, size_t sstep, const uchar* mask, size_t mst
v_uint8x16 v_nmask = v_load(mask + x) == v_zero;
v_zip(v_nmask, v_nmask, v_nmask1, v_nmask2);
#if CV_SSE4_2
v_dst1 = v_uint16x8(_mm_blendv_epi8(v_src1.val, v_dst1.val, v_nmask1.val));
v_dst2 = v_uint16x8(_mm_blendv_epi8(v_src2.val, v_dst2.val, v_nmask2.val));
#else
v_dst1 = v_select(v_reinterpret_as_u16(v_nmask1), v_dst1, v_src1);
v_dst2 = v_select(v_reinterpret_as_u16(v_nmask2), v_dst2, v_src2);
#endif
v_store(dst + x, v_dst1);
v_store(dst + x + 8, v_dst2);
}