static void NeedsFilter(const __m128i* p1, const __m128i* p0, const __m128i* q0,
const __m128i* q1, int thresh, __m128i *mask) {
__m128i t1 = MM_ABS(*p1, *q1); // abs(p1 - q1)
*mask = _mm_set1_epi8(0xFE);
t1 = _mm_and_si128(t1, *mask); // set lsb of each byte to zero
t1 = _mm_srli_epi16(t1, 1); // abs(p1 - q1) / 2
*mask = MM_ABS(*p0, *q0); // abs(p0 - q0)
*mask = _mm_adds_epu8(*mask, *mask); // abs(p0 - q0) * 2
*mask = _mm_adds_epu8(*mask, t1); // abs(p0 - q0) * 2 + abs(p1 - q1) / 2
t1 = _mm_set1_epi8(thresh);
*mask = _mm_subs_epu8(*mask, t1); // mask <= thresh
*mask = _mm_cmpeq_epi8(*mask, _mm_setzero_si128());
}
// input/output is uint8_t
static WEBP_INLINE void GetNotHEV(const __m128i* const p1,
const __m128i* const p0,
const __m128i* const q0,
const __m128i* const q1,
int hev_thresh, __m128i* const not_hev) {
const __m128i zero = _mm_setzero_si128();
const __m128i t_1 = MM_ABS(*p1, *p0);
const __m128i t_2 = MM_ABS(*q1, *q0);
const __m128i h = _mm_set1_epi8(hev_thresh);
const __m128i t_3 = _mm_subs_epu8(t_1, h); // abs(p1 - p0) - hev_tresh
const __m128i t_4 = _mm_subs_epu8(t_2, h); // abs(q1 - q0) - hev_tresh
*not_hev = _mm_or_si128(t_3, t_4);
*not_hev = _mm_cmpeq_epi8(*not_hev, zero); // not_hev <= t1 && not_hev <= t2
}
// input pixels are uint8_t
static WEBP_INLINE void NeedsFilter(const __m128i* const p1,
const __m128i* const p0,
const __m128i* const q0,
const __m128i* const q1,
int thresh, __m128i* const mask) {
const __m128i m_thresh = _mm_set1_epi8(thresh);
const __m128i t1 = MM_ABS(*p1, *q1); // abs(p1 - q1)
const __m128i kFE = _mm_set1_epi8(0xFE);
const __m128i t2 = _mm_and_si128(t1, kFE); // set lsb of each byte to zero
const __m128i t3 = _mm_srli_epi16(t2, 1); // abs(p1 - q1) / 2
const __m128i t4 = MM_ABS(*p0, *q0); // abs(p0 - q0)
const __m128i t5 = _mm_adds_epu8(t4, t4); // abs(p0 - q0) * 2
const __m128i t6 = _mm_adds_epu8(t5, t3); // abs(p0-q0)*2 + abs(p1-q1)/2
const __m128i t7 = _mm_subs_epu8(t6, m_thresh); // mask <= m_thresh
*mask = _mm_cmpeq_epi8(t7, _mm_setzero_si128());
}
