static inline __m128i darken_byte_SSE2(const __m128i& sc, const __m128i& dc,
const __m128i& sa, const __m128i& da) {
__m128i sd = _mm_mullo_epi16(sc, da);
__m128i ds = _mm_mullo_epi16(dc, sa);
__m128i cmp = _mm_cmplt_epi32(sd, ds);
__m128i tmp = _mm_add_epi32(sc, dc);
__m128i ret1 = _mm_sub_epi32(tmp, SkDiv255Round_SSE2(ds));
__m128i ret2 = _mm_sub_epi32(tmp, SkDiv255Round_SSE2(sd));
__m128i ret = _mm_or_si128(_mm_and_si128(cmp, ret1),
_mm_andnot_si128(cmp, ret2));
return ret;
}
static inline __m128i difference_byte_SSE2(const __m128i& sc, const __m128i& dc,
const __m128i& sa, const __m128i& da) {
__m128i tmp1 = _mm_mullo_epi16(sc, da);
__m128i tmp2 = _mm_mullo_epi16(dc, sa);
__m128i tmp = SkMin32_SSE2(tmp1, tmp2);
__m128i ret1 = _mm_add_epi32(sc, dc);
__m128i ret2 = _mm_slli_epi32(SkDiv255Round_SSE2(tmp), 1);
__m128i ret = _mm_sub_epi32(ret1, ret2);
ret = clamp_signed_byte_SSE2(ret);
return ret;
}
static inline __m128i clamp_div255round_SSE2(const __m128i& prod) {
// test if > 0
__m128i cmp1 = _mm_cmpgt_epi32(prod, _mm_setzero_si128());
// test if < 255*255
__m128i cmp2 = _mm_cmplt_epi32(prod, _mm_set1_epi32(255*255));
__m128i ret = _mm_setzero_si128();
// if value >= 255*255, value = 255
ret = _mm_andnot_si128(cmp2, _mm_set1_epi32(255));
__m128i div = SkDiv255Round_SSE2(prod);
// test if > 0 && < 255*255
__m128i cmp = _mm_and_si128(cmp1, cmp2);
ret = _mm_or_si128(_mm_and_si128(cmp, div), _mm_andnot_si128(cmp, ret));
return ret;
}
