-#define GET_NOTHEV(p1, p0, q0, q1, hev_thresh, 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 */\
-}
-
-#define GET_BASE_DELTA(p1, p0, q0, q1, o) { \
- const __m128i qp0 = _mm_subs_epi8(q0, p0); /* q0 - p0 */ \
- o = _mm_subs_epi8(p1, q1); /* p1 - q1 */ \
- o = _mm_adds_epi8(o, qp0); /* p1 - q1 + 1 * (q0 - p0) */ \
- o = _mm_adds_epi8(o, qp0); /* p1 - q1 + 2 * (q0 - p0) */ \
- o = _mm_adds_epi8(o, qp0); /* p1 - q1 + 3 * (q0 - p0) */ \
-}
-
-#define DO_SIMPLE_FILTER(p0, q0, fl) { \
- const __m128i three = _mm_set1_epi8(3); \
- const __m128i four = _mm_set1_epi8(4); \
- __m128i v3 = _mm_adds_epi8(fl, three); \
- __m128i v4 = _mm_adds_epi8(fl, four); \
- \
- /* Do +4 side */ \
- SIGNED_SHIFT_N(v4, 3); /* v4 >> 3 */ \
- q0 = _mm_subs_epi8(q0, v4); /* q0 -= v4 */ \
- \
- /* Now do +3 side */ \
- SIGNED_SHIFT_N(v3, 3); /* v3 >> 3 */ \
- p0 = _mm_adds_epi8(p0, v3); /* p0 += v3 */ \
+// 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 int8_t
+static WEBP_INLINE void GetBaseDelta(const __m128i* const p1,
+ const __m128i* const p0,
+ const __m128i* const q0,
+ const __m128i* const q1,
+ __m128i* const delta) {
+ // beware of addition order, for saturation!
+ const __m128i p1_q1 = _mm_subs_epi8(*p1, *q1); // p1 - q1
+ const __m128i q0_p0 = _mm_subs_epi8(*q0, *p0); // q0 - p0
+ const __m128i s1 = _mm_adds_epi8(p1_q1, q0_p0); // p1 - q1 + 1 * (q0 - p0)
+ const __m128i s2 = _mm_adds_epi8(q0_p0, s1); // p1 - q1 + 2 * (q0 - p0)
+ const __m128i s3 = _mm_adds_epi8(q0_p0, s2); // p1 - q1 + 3 * (q0 - p0)
+ *delta = s3;
+}
+
+// input and output are int8_t
+static WEBP_INLINE void DoSimpleFilter(__m128i* const p0, __m128i* const q0,
+ const __m128i* const fl) {
+ const __m128i k3 = _mm_set1_epi8(3);
+ const __m128i k4 = _mm_set1_epi8(4);
+ __m128i v3 = _mm_adds_epi8(*fl, k3);
+ __m128i v4 = _mm_adds_epi8(*fl, k4);
+
+ SignedShift8b(&v4); // v4 >> 3
+ SignedShift8b(&v3); // v3 >> 3
+ *q0 = _mm_subs_epi8(*q0, v4); // q0 -= v4
+ *p0 = _mm_adds_epi8(*p0, v3); // p0 += v3