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

optimize blend using universal intrinsic

Browse Source 
- add more channels/depth performance test for blend
This commit is contained in:
Tomoaki Teshima
2017-02-20 19:09:26 +09:00
parent 9053839282
commit 64cf206fb5
3 changed files with 286 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/hal/intrin_sse.hpp
+27
View File
@@ -1368,6 +1368,24 @@ OPENCV_HAL_IMPL_SSE_TRANSPOSE4x4(v_int32x4, epi32, OPENCV_HAL_NOP, OPENCV_HAL_NO
OPENCV_HAL_IMPL_SSE_TRANSPOSE4x4(v_float32x4, ps, _mm_castps_si128, _mm_castsi128_ps)
// adopted from sse_utils.hpp
inline void v_load_deinterleave(const uchar* ptr, v_uint8x16& a, v_uint8x16& b)
{
__m128i t00 = _mm_loadu_si128((const __m128i*)ptr);
__m128i t01 = _mm_loadu_si128((const __m128i*)(ptr + 16));
__m128i t10 = _mm_unpacklo_epi8(t00, t01);
__m128i t11 = _mm_unpackhi_epi8(t00, t01);
__m128i t20 = _mm_unpacklo_epi8(t10, t11);
__m128i t21 = _mm_unpackhi_epi8(t10, t11);
__m128i t30 = _mm_unpacklo_epi8(t20, t21);
__m128i t31 = _mm_unpackhi_epi8(t20, t21);
a.val = _mm_unpacklo_epi8(t30, t31);
b.val = _mm_unpackhi_epi8(t30, t31);
}
inline void v_load_deinterleave(const uchar* ptr, v_uint8x16& a, v_uint8x16& b, v_uint8x16& c)
{
__m128i t00 = _mm_loadu_si128((const __m128i*)ptr);
@@ -1507,6 +1525,15 @@ inline void v_store_interleave( short* ptr, const v_int16x8& a, const v_int16x8&
_mm_storeu_si128((__m128i*)(ptr + 8), t1);
}
inline void v_store_interleave( uchar* ptr, const v_uint8x16& a, const v_uint8x16& b)
{
__m128i v0 = _mm_unpacklo_epi8(a.val, b.val);
__m128i v1 = _mm_unpackhi_epi8(a.val, b.val);
_mm_storeu_si128((__m128i*)(ptr), v0);
_mm_storeu_si128((__m128i*)(ptr + 16), v1);
}
inline void v_store_interleave( uchar* ptr, const v_uint8x16& a, const v_uint8x16& b,
const v_uint8x16& c )
{