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

optimized cmp and cvtscale(16s->16s) using SSE2 (thanks to Victoria)

Browse Source
This commit is contained in:
Vadim Pisarevsky
2011-12-15 21:15:51 +00:00
parent 677fc3a09f
commit eef900e46a
2 changed files with 214 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
modules/core/src/convert.cpp
View File

@@ -606,6 +606,50 @@ cvtScale_( const T* src, size_t sstep,
}
}
//vz optimized template specialization
template<> static void
cvtScale_<short, short, float>( const short* src, size_t sstep,
short* dst, size_t dstep, Size size,
float scale, float shift )
{
sstep /= sizeof(src[0]);
dstep /= sizeof(dst[0]);
#if CV_SSE2
__m128 scale128, shift128;
if(USE_SSE2){
scale128 = _mm_set1_ps (scale);
shift128 = _mm_set1_ps (shift);
}
#endif
for( ; size.height--; src += sstep, dst += dstep )
{
int x = 0;
#if CV_SSE2
if(USE_SSE2)
{
for(; x <= size.width - 8; x += 8 )
{
__m128i r0 = _mm_loadl_epi64((const __m128i*)(src + x));
__m128i r1 = _mm_loadl_epi64((const __m128i*)(src + x + 4));
__m128 rf0 =_mm_cvtepi32_ps(_mm_srai_epi32(_mm_unpacklo_epi16(r0, r0), 16));
__m128 rf1 =_mm_cvtepi32_ps(_mm_srai_epi32(_mm_unpacklo_epi16(r1, r1), 16));
rf0 = _mm_add_ps(_mm_mul_ps(rf0, scale128), shift128);
rf1 = _mm_add_ps(_mm_mul_ps(rf1, scale128), shift128);
r0 = _mm_cvtps_epi32(rf0);
r1 = _mm_cvtps_epi32(rf1);
r0 = _mm_packs_epi32(r0, r1);
_mm_storeu_si128((__m128i*)(dst + x), r0);
}
}
#endif
for(; x < size.width; x++ )
dst[x] = saturate_cast<short>(src[x]*scale + shift);
}
}
template<typename T, typename DT> static void
cvt_( const T* src, size_t sstep,