Opencv二值化源码分析（SSE）

Opencv二值化可以说是最简单的算子了，只要设定阈值，再根据相应的规则对于像素赋值为0或255即可

普通的C++代码应该不用在解释，这里贴入Opencv的二值化带SSE代码

for( int i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
    for( j = 0; j <= roi.width - v_uint8::nlanes; j += v_uint8::nlanes)
    {
        v_uint8 v0;
        v0 = vx_load( src + j );
        v0 = thresh_u < v0;
        v0 = v0 & maxval16;
        v_store( dst + j, v0 );
    }
}

这里的v_uint8 在SSE中就是__m128i，vx_load就是_mm_loadu_si128，v_store 就是_mm_storeu_si128，并且还对等于，大于等做了重载

#define OPENCV_HAL_IMPL_SSE_INT_CMP_OP(_Tpuvec, _Tpsvec, suffix, sbit) \
inline _Tpuvec operator == (const _Tpuvec& a, const _Tpuvec& b) \
{ return _Tpuvec(_mm_cmpeq_##suffix(a.val, b.val)); } \
inline _Tpuvec operator != (const _Tpuvec& a, const _Tpuvec& b) \
{ \
    __m128i not_mask = _mm_set1_epi32(-1); \
    return _Tpuvec(_mm_xor_si128(_mm_cmpeq_##suffix(a.val, b.val), not_mask)); \
} \
inline _Tpsvec operator == (const _Tpsvec& a, const _Tpsvec& b) \
{ return _Tpsvec(_mm_cmpeq_##suffix(a.val, b.val)); } \
inline _Tpsvec operator != (const _Tpsvec& a, const _Tpsvec& b) \
{ \
    __m128i not_mask = _mm_set1_epi32(-1); \
    return _Tpsvec(_mm_xor_si128(_mm_cmpeq_##suffix(a.val, b.val), not_mask)); \
} \
inline _Tpuvec operator < (const _Tpuvec& a, const _Tpuvec& b) \
{ \
    __m128i smask = _mm_set1_##suffix(sbit); \
    return _Tpuvec(_mm_cmpgt_##suffix(_mm_xor_si128(b.val, smask), _mm_xor_si128(a.val, smask))); \
} \
inline _Tpuvec operator > (const _Tpuvec& a, const _Tpuvec& b) \
{ \
    __m128i smask = _mm_set1_##suffix(sbit); \
    return _Tpuvec(_mm_cmpgt_##suffix(_mm_xor_si128(a.val, smask), _mm_xor_si128(b.val, smask))); \
} \
inline _Tpuvec operator <= (const _Tpuvec& a, const _Tpuvec& b) \
{ \
    __m128i smask = _mm_set1_##suffix(sbit); \
    __m128i not_mask = _mm_set1_epi32(-1); \
    __m128i res = _mm_cmpgt_##suffix(_mm_xor_si128(a.val, smask), _mm_xor_si128(b.val, smask)); \
    return _Tpuvec(_mm_xor_si128(res, not_mask)); \
} \
inline _Tpuvec operator >= (const _Tpuvec& a, const _Tpuvec& b) \
{ \
    __m128i smask = _mm_set1_##suffix(sbit); \
    __m128i not_mask = _mm_set1_epi32(-1); \
    __m128i res = _mm_cmpgt_##suffix(_mm_xor_si128(b.val, smask), _mm_xor_si128(a.val, smask)); \
    return _Tpuvec(_mm_xor_si128(res, not_mask)); \
} \
inline _Tpsvec operator < (const _Tpsvec& a, const _Tpsvec& b) \
{ \
    return _Tpsvec(_mm_cmpgt_##suffix(b.val, a.val)); \
} \
inline _Tpsvec operator > (const _Tpsvec& a, const _Tpsvec& b) \
{ \
    return _Tpsvec(_mm_cmpgt_##suffix(a.val, b.val)); \
} \
inline _Tpsvec operator <= (const _Tpsvec& a, const _Tpsvec& b) \
{ \
    __m128i not_mask = _mm_set1_epi32(-1); \
    return _Tpsvec(_mm_xor_si128(_mm_cmpgt_##suffix(a.val, b.val), not_mask)); \
} \
inline _Tpsvec operator >= (const _Tpsvec& a, const _Tpsvec& b) \
{ \
    __m128i not_mask = _mm_set1_epi32(-1); \
    return _Tpsvec(_mm_xor_si128(_mm_cmpgt_##suffix(b.val, a.val), not_mask)); \
}