28 uint8_t *p3 = src - (stride << 2);
29 uint8_t *p2 = src - ((stride << 1) + stride);
30 uint8_t *p1 = src - (stride << 1);
34 uint8_t *q2 = src + (stride << 1);
35 uint8_t *q3 = src + (stride << 1) + stride;
37 int32_t dp00, dq00, dp30, dq30, d00, d30;
38 int32_t dp04, dq04, dp34, dq34, d04, d34;
39 int32_t tc0, p_is_pcm0, q_is_pcm0, beta30, beta20, tc250;
41 uint64_t dst_val0, dst_val1;
42 v16u8 dst0, dst1, dst2, dst3, dst4, dst5;
43 v2i64 cmp0, cmp1, cmp2, p_is_pcm_vec, q_is_pcm_vec;
47 v8i16 diff0, diff1, delta0, delta1, delta2, abs_delta0;
49 v8u16 p3_src, p2_src, p1_src, p0_src, q0_src, q1_src, q2_src, q3_src;
51 dp00 = abs(p2[0] - (p1[0] << 1) + p0[0]);
52 dq00 = abs(q2[0] - (q1[0] << 1) + q0[0]);
53 dp30 = abs(p2[3] - (p1[3] << 1) + p0[3]);
54 dq30 = abs(q2[3] - (q1[3] << 1) + q0[3]);
57 p_is_pcm0 = p_is_pcm[0];
58 q_is_pcm0 = q_is_pcm[0];
59 dp04 = abs(p2[4] - (p1[4] << 1) + p0[4]);
60 dq04 = abs(q2[4] - (q1[4] << 1) + q0[4]);
61 dp34 = abs(p2[7] - (p1[7] << 1) + p0[7]);
62 dq34 = abs(q2[7] - (q1[7] << 1) + q0[7]);
65 p_is_pcm4 = p_is_pcm[1];
66 q_is_pcm4 = q_is_pcm[1];
68 if (!p_is_pcm0 || !p_is_pcm4 || !q_is_pcm0 || !q_is_pcm4) {
69 if (!(d00 + d30 >= beta) || !(d04 + d34 >= beta)) {
82 tc250 = ((tc0 * 5 + 1) >> 1);
84 tc254 = ((tc4 * 5 + 1) >> 1);
86 flag0 = (abs(p3[0] - p0[0]) + abs(q3[0] - q0[0]) < beta30 &&
87 abs(p0[0] - q0[0]) < tc250 &&
88 abs(p3[3] - p0[3]) + abs(q3[3] - q0[3]) < beta30 &&
89 abs(p0[3] - q0[3]) < tc250 &&
90 (d00 << 1) < beta20 && (d30 << 1) < beta20);
91 cmp0 = __msa_fill_d(flag0);
93 flag1 = (abs(p3[4] - p0[4]) + abs(q3[4] - q0[4]) < beta30 &&
94 abs(p0[4] - q0[4]) < tc254 &&
95 abs(p3[7] - p0[7]) + abs(q3[7] - q0[7]) < beta30 &&
96 abs(p0[7] - q0[7]) < tc254 &&
97 (d04 << 1) < beta20 && (d34 << 1) < beta20);
98 cmp1 = __msa_fill_d(flag1);
99 cmp2 = __msa_ilvev_d(cmp1, cmp0);
100 cmp2 = __msa_ceqi_d(cmp2, 0);
102 ILVR_B8_UH(zero, p3_src, zero, p2_src, zero, p1_src, zero, p0_src,
103 zero, q0_src, zero, q1_src, zero, q2_src, zero, q3_src,
104 p3_src, p2_src, p1_src, p0_src, q0_src, q1_src, q2_src,
107 cmp0 = (v2i64) __msa_fill_h(tc0);
108 cmp1 = (v2i64) __msa_fill_h(tc4);
109 tc_pos = (v8i16) __msa_ilvev_d(cmp1, cmp0);
113 temp0 = (p1_src + p0_src + q0_src);
114 temp1 = ((p3_src + p2_src) << 1) + p2_src + temp0;
115 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 3);
116 temp2 = (v8i16) (temp1 - p2_src);
117 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
118 dst0 = (v16u8) (temp2 + (v8i16) p2_src);
120 temp1 = temp0 + p2_src;
121 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 2);
122 temp2 = (v8i16) (temp1 - p1_src);
123 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
124 dst1 = (v16u8) (temp2 + (v8i16) p1_src);
126 temp1 = (temp0 << 1) + p2_src + q1_src;
127 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 3);
128 temp2 = (v8i16) (temp1 - p0_src);
129 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
130 dst2 = (v16u8) (temp2 + (v8i16) p0_src);
132 cmp0 = __msa_fill_d(p_is_pcm0);
133 cmp1 = __msa_fill_d(p_is_pcm4);
134 p_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
135 p_is_pcm_vec = __msa_ceqi_d(p_is_pcm_vec, 0);
137 dst0 = __msa_bmz_v(dst0, (v16u8) p2_src, (v16u8) p_is_pcm_vec);
138 dst1 = __msa_bmz_v(dst1, (v16u8) p1_src, (v16u8) p_is_pcm_vec);
139 dst2 = __msa_bmz_v(dst2, (v16u8) p0_src, (v16u8) p_is_pcm_vec);
141 temp0 = (q1_src + p0_src + q0_src);
143 temp1 = ((q3_src + q2_src) << 1) + q2_src + temp0;
144 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 3);
145 temp2 = (v8i16) (temp1 - q2_src);
146 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
147 dst5 = (v16u8) (temp2 + (v8i16) q2_src);
149 temp1 = temp0 + q2_src;
150 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 2);
151 temp2 = (v8i16) (temp1 - q1_src);
152 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
153 dst4 = (v16u8) (temp2 + (v8i16) q1_src);
155 temp1 = (temp0 << 1) + p1_src + q2_src;
156 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 3);
157 temp2 = (v8i16) (temp1 - q0_src);
158 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
159 dst3 = (v16u8) (temp2 + (v8i16) q0_src);
161 cmp0 = __msa_fill_d(q_is_pcm0);
162 cmp1 = __msa_fill_d(q_is_pcm4);
163 q_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
164 q_is_pcm_vec = __msa_ceqi_d(q_is_pcm_vec, 0);
166 dst3 = __msa_bmz_v(dst3, (v16u8) q0_src, (v16u8) q_is_pcm_vec);
167 dst4 = __msa_bmz_v(dst4, (v16u8) q1_src, (v16u8) q_is_pcm_vec);
168 dst5 = __msa_bmz_v(dst5, (v16u8) q2_src, (v16u8) q_is_pcm_vec);
173 diff0 = (v8i16) (q0_src - p0_src);
174 diff1 = (v8i16) (q1_src - p1_src);
175 diff0 = (diff0 << 3) + diff0;
176 diff1 = (diff1 << 1) + diff1;
177 delta0 = diff0 - diff1;
178 delta0 = __msa_srari_h(delta0, 4);
180 temp1 = (v8u16) ((tc_pos << 3) + (tc_pos << 1));
181 abs_delta0 = __msa_add_a_h(delta0, (v8i16) zero);
182 abs_delta0 = (v8u16) abs_delta0 < temp1;
184 delta0 =
CLIP_SH(delta0, tc_neg, tc_pos);
186 temp0 = (v8u16) (delta0 + p0_src);
188 temp0 = (v8u16) __msa_bmz_v((v16u8) temp0, (v16u8) p0_src,
189 (v16u8) p_is_pcm_vec);
191 temp2 = (v8i16) (q0_src - delta0);
193 temp2 = (v8i16) __msa_bmz_v((v16u8) temp2, (v16u8) q0_src,
194 (v16u8) q_is_pcm_vec);
196 tmp = (beta + (beta >> 1)) >> 3;
197 cmp0 = __msa_fill_d(!p_is_pcm0 && ((dp00 + dp30) < tmp));
198 cmp1 = __msa_fill_d(!p_is_pcm4 && ((dp04 + dp34) < tmp));
199 p_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
200 p_is_pcm_vec = __msa_ceqi_d(p_is_pcm_vec, 0);
202 cmp0 = (v2i64) __msa_fill_h((!q_is_pcm0) && (dq00 + dq30 <
tmp));
203 cmp1 = (v2i64) __msa_fill_h((!q_is_pcm4) && (dq04 + dq34 <
tmp));
204 q_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
205 q_is_pcm_vec = __msa_ceqi_d(q_is_pcm_vec, 0);
210 delta1 = (v8i16) __msa_aver_u_h(p2_src, p0_src);
211 delta1 -= (v8i16) p1_src;
214 delta1 =
CLIP_SH(delta1, tc_neg, tc_pos);
215 delta1 = (v8i16) p1_src + (v8i16) delta1;
217 delta1 = (v8i16) __msa_bmnz_v((v16u8) delta1, (v16u8) p1_src,
218 (v16u8) p_is_pcm_vec);
220 delta2 = (v8i16) __msa_aver_u_h(q0_src, q2_src);
221 delta2 = delta2 - (v8i16) q1_src;
222 delta2 = delta2 - delta0;
223 delta2 = delta2 >> 1;
224 delta2 =
CLIP_SH(delta2, tc_neg, tc_pos);
225 delta2 = (v8i16) q1_src + (v8i16) delta2;
227 delta2 = (v8i16) __msa_bmnz_v((v16u8) delta2, (v16u8) q1_src,
228 (v16u8) q_is_pcm_vec);
230 delta1 = (v8i16) __msa_bmz_v((v16u8) delta1, (v16u8) p1_src,
232 temp0 = (v8u16) __msa_bmz_v((v16u8) temp0, (v16u8) p0_src,
234 temp2 = (v8i16) __msa_bmz_v((v16u8) temp2, (v16u8) q0_src,
236 delta2 = (v8i16) __msa_bmz_v((v16u8) delta2, (v16u8) q1_src,
239 dst2 = __msa_bmnz_v(dst2, (v16u8) temp0, (v16u8) cmp2);
240 dst3 = __msa_bmnz_v(dst3, (v16u8) temp2, (v16u8) cmp2);
241 dst1 = __msa_bmnz_v(dst1, (v16u8) delta1, (v16u8) cmp2);
242 dst4 = __msa_bmnz_v(dst4, (v16u8) delta2, (v16u8) cmp2);
243 dst0 = __msa_bmnz_v(dst0, (v16u8) p2_src, (v16u8) cmp2);
244 dst5 = __msa_bmnz_v(dst5, (v16u8) q2_src, (v16u8) cmp2);
246 cmp0 = __msa_fill_d(d00 + d30 >= beta);
247 cmp1 = __msa_fill_d(d04 + d34 >= beta);
248 cmp0 = __msa_ilvev_d(cmp1, cmp0);
249 cmp0 = __msa_ceqi_d(cmp0, 0);
251 dst0 = __msa_bmz_v(dst0, (v16u8) p2_src, (v16u8) cmp0);
252 dst1 = __msa_bmz_v(dst1, (v16u8) p1_src, (v16u8) cmp0);
253 dst2 = __msa_bmz_v(dst2, (v16u8) p0_src, (v16u8) cmp0);
254 dst3 = __msa_bmz_v(dst3, (v16u8) q0_src, (v16u8) cmp0);
255 dst4 = __msa_bmz_v(dst4, (v16u8) q1_src, (v16u8) cmp0);
256 dst5 = __msa_bmz_v(dst5, (v16u8) q2_src, (v16u8) cmp0);
259 dst2 = (v16u8) __msa_pckev_b((v16i8) dst5, (v16i8) dst4);
261 dst_val0 = __msa_copy_u_d((v2i64) dst2, 0);
262 dst_val1 = __msa_copy_u_d((v2i64) dst2, 1);
279 uint8_t *p1 = src + (stride << 2);
284 int32_t dp00, dq00, dp30, dq30, d00, d30;
285 int32_t dp04, dq04, dp34, dq34, d04, d34;
286 int32_t tc0, p_is_pcm0, q_is_pcm0, beta30, beta20, tc250;
287 int32_t tc4, p_is_pcm4, q_is_pcm4, tc254,
tmp;
288 v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
289 v2i64 cmp0, cmp1, cmp2, p_is_pcm_vec, q_is_pcm_vec;
292 v8i16 tc_pos, tc_neg;
293 v8i16 diff0, diff1, delta0, delta1, delta2, abs_delta0;
295 v8u16 p3_src, p2_src, p1_src, p0_src, q0_src, q1_src, q2_src, q3_src;
297 dp00 = abs(p3[-3] - (p3[-2] << 1) + p3[-1]);
298 dq00 = abs(p3[2] - (p3[1] << 1) + p3[0]);
299 dp30 = abs(p2[-3] - (p2[-2] << 1) + p2[-1]);
300 dq30 = abs(p2[2] - (p2[1] << 1) + p2[0]);
303 p_is_pcm0 = p_is_pcm[0];
304 q_is_pcm0 = q_is_pcm[0];
306 dp04 = abs(p1[-3] - (p1[-2] << 1) + p1[-1]);
307 dq04 = abs(p1[2] - (p1[1] << 1) + p1[0]);
308 dp34 = abs(p0[-3] - (p0[-2] << 1) + p0[-1]);
309 dq34 = abs(p0[2] - (p0[1] << 1) + p0[0]);
312 p_is_pcm4 = p_is_pcm[1];
313 q_is_pcm4 = q_is_pcm[1];
315 if (!p_is_pcm0 || !p_is_pcm4 || !q_is_pcm0 || !q_is_pcm4) {
316 if (!(d00 + d30 >= beta) || !(d04 + d34 >= beta)) {
319 p3_src, p2_src, p1_src, p0_src, q0_src, q1_src, q2_src,
325 tc250 = ((tc0 * 5 + 1) >> 1);
328 tc254 = ((tc4 * 5 + 1) >> 1);
331 q2_src, q3_src, p3_src, p2_src, p1_src, p0_src,
332 q0_src, q1_src, q2_src, q3_src);
334 flag0 = (abs(p3[-4] - p3[-1]) + abs(p3[3] - p3[0]) < beta30 &&
335 abs(p3[-1] - p3[0]) < tc250 &&
336 abs(p2[-4] - p2[-1]) + abs(p2[3] - p2[0]) < beta30 &&
337 abs(p2[-1] - p2[0]) < tc250 &&
338 (d00 << 1) < beta20 && (d30 << 1) < beta20);
339 cmp0 = __msa_fill_d(flag0);
341 flag1 = (abs(p1[-4] - p1[-1]) + abs(p1[3] - p1[0]) < beta30 &&
342 abs(p1[-1] - p1[0]) < tc254 &&
343 abs(p0[-4] - p0[-1]) + abs(p0[3] - p0[0]) < beta30 &&
344 abs(p0[-1] - p0[0]) < tc254 &&
345 (d04 << 1) < beta20 && (d34 << 1) < beta20);
346 cmp1 = __msa_fill_d(flag1);
347 cmp2 = __msa_ilvev_d(cmp1, cmp0);
348 cmp2 = __msa_ceqi_d(cmp2, 0);
350 ILVR_B8_UH(zero, p3_src, zero, p2_src, zero, p1_src, zero, p0_src,
351 zero, q0_src, zero, q1_src, zero, q2_src, zero, q3_src,
352 p3_src, p2_src, p1_src, p0_src, q0_src, q1_src, q2_src,
355 cmp0 = (v2i64) __msa_fill_h(tc0 << 1);
356 cmp1 = (v2i64) __msa_fill_h(tc4 << 1);
357 tc_pos = (v8i16) __msa_ilvev_d(cmp1, cmp0);
360 temp0 = (p1_src + p0_src + q0_src);
362 temp1 = ((p3_src + p2_src) << 1) + p2_src + temp0;
363 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 3);
364 temp2 = (v8i16) (temp1 - p2_src);
365 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
366 dst0 = (v16u8) (temp2 + (v8i16) p2_src);
368 temp1 = temp0 + p2_src;
369 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 2);
370 temp2 = (v8i16) (temp1 - p1_src);
371 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
372 dst1 = (v16u8) (temp2 + (v8i16) p1_src);
374 temp1 = (temp0 << 1) + p2_src + q1_src;
375 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 3);
376 temp2 = (v8i16) (temp1 - p0_src);
377 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
378 dst2 = (v16u8) (temp2 + (v8i16) p0_src);
380 cmp0 = __msa_fill_d(p_is_pcm0);
381 cmp1 = __msa_fill_d(p_is_pcm4);
382 p_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
383 p_is_pcm_vec = __msa_ceqi_d(p_is_pcm_vec, 0);
385 dst0 = __msa_bmz_v(dst0, (v16u8) p2_src, (v16u8) p_is_pcm_vec);
386 dst1 = __msa_bmz_v(dst1, (v16u8) p1_src, (v16u8) p_is_pcm_vec);
387 dst2 = __msa_bmz_v(dst2, (v16u8) p0_src, (v16u8) p_is_pcm_vec);
389 temp0 = (q1_src + p0_src + q0_src);
390 temp1 = ((q3_src + q2_src) << 1) + q2_src + temp0;
391 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 3);
392 temp2 = (v8i16) (temp1 - q2_src);
393 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
394 dst5 = (v16u8) (temp2 + (v8i16) q2_src);
396 temp1 = temp0 + q2_src;
397 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 2);
398 temp2 = (v8i16) (temp1 - q1_src);
399 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
400 dst4 = (v16u8) (temp2 + (v8i16) q1_src);
402 temp1 = (temp0 << 1) + p1_src + q2_src;
403 temp1 = (v8u16) __msa_srari_h((v8i16) temp1, 3);
404 temp2 = (v8i16) (temp1 - q0_src);
405 temp2 =
CLIP_SH(temp2, tc_neg, tc_pos);
406 dst3 = (v16u8) (temp2 + (v8i16) q0_src);
408 cmp0 = __msa_fill_d(q_is_pcm0);
409 cmp1 = __msa_fill_d(q_is_pcm4);
410 q_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
411 q_is_pcm_vec = __msa_ceqi_d(q_is_pcm_vec, 0);
413 dst3 = __msa_bmz_v(dst3, (v16u8) q0_src, (v16u8) q_is_pcm_vec);
414 dst4 = __msa_bmz_v(dst4, (v16u8) q1_src, (v16u8) q_is_pcm_vec);
415 dst5 = __msa_bmz_v(dst5, (v16u8) q2_src, (v16u8) q_is_pcm_vec);
420 diff0 = (v8i16) (q0_src - p0_src);
421 diff1 = (v8i16) (q1_src - p1_src);
422 diff0 = (v8i16) (diff0 << 3) + diff0;
423 diff1 = (v8i16) (diff1 << 1) + diff1;
424 delta0 = diff0 - diff1;
425 delta0 = __msa_srari_h(delta0, 4);
427 temp1 = (v8u16) ((tc_pos << 3) + (tc_pos << 1));
428 abs_delta0 = __msa_add_a_h(delta0, (v8i16) zero);
429 abs_delta0 = (v8u16) abs_delta0 < temp1;
431 delta0 =
CLIP_SH(delta0, tc_neg, tc_pos);
432 temp0 = (v8u16) delta0 + p0_src;
434 temp0 = (v8u16) __msa_bmz_v((v16u8) temp0, (v16u8) p0_src,
435 (v16u8) p_is_pcm_vec);
437 temp2 = (v8i16) q0_src - delta0;
439 temp2 = (v8i16) __msa_bmz_v((v16u8) temp2, (v16u8) q0_src,
440 (v16u8) q_is_pcm_vec);
442 tmp = ((beta + (beta >> 1)) >> 3);
443 cmp0 = __msa_fill_d(!p_is_pcm0 && (dp00 + dp30 < tmp));
444 cmp1 = __msa_fill_d(!p_is_pcm4 && (dp04 + dp34 < tmp));
445 p_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
446 p_is_pcm_vec = __msa_ceqi_d(p_is_pcm_vec, 0);
448 cmp0 = (v2i64) __msa_fill_h((!q_is_pcm0) && (dq00 + dq30 <
tmp));
449 cmp1 = (v2i64) __msa_fill_h((!q_is_pcm4) && (dq04 + dq34 <
tmp));
450 q_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
451 q_is_pcm_vec = __msa_ceqi_d(q_is_pcm_vec, 0);
456 delta1 = (v8i16) __msa_aver_u_h(p2_src, p0_src);
457 delta1 -= (v8i16) p1_src;
460 delta1 =
CLIP_SH(delta1, tc_neg, tc_pos);
461 delta1 = (v8i16) p1_src + (v8i16) delta1;
463 delta1 = (v8i16) __msa_bmnz_v((v16u8) delta1, (v16u8) p1_src,
464 (v16u8) p_is_pcm_vec);
466 delta2 = (v8i16) __msa_aver_u_h(q0_src, q2_src);
467 delta2 = delta2 - (v8i16) q1_src;
468 delta2 = delta2 - delta0;
469 delta2 = delta2 >> 1;
470 delta2 =
CLIP_SH(delta2, tc_neg, tc_pos);
471 delta2 = (v8i16) q1_src + (v8i16) delta2;
473 delta2 = (v8i16) __msa_bmnz_v((v16u8) delta2, (v16u8) q1_src,
474 (v16u8) q_is_pcm_vec);
475 delta1 = (v8i16) __msa_bmz_v((v16u8) delta1, (v16u8) p1_src,
477 temp0 = (v8u16) __msa_bmz_v((v16u8) temp0, (v16u8) p0_src,
479 temp2 = (v8i16) __msa_bmz_v((v16u8) temp2, (v16u8) q0_src,
481 delta2 = (v8i16) __msa_bmz_v((v16u8) delta2, (v16u8) q1_src,
484 dst2 = __msa_bmnz_v(dst2, (v16u8) temp0, (v16u8) cmp2);
485 dst3 = __msa_bmnz_v(dst3, (v16u8) temp2, (v16u8) cmp2);
486 dst1 = __msa_bmnz_v(dst1, (v16u8) delta1, (v16u8) cmp2);
487 dst4 = __msa_bmnz_v(dst4, (v16u8) delta2, (v16u8) cmp2);
488 dst0 = __msa_bmnz_v(dst0, (v16u8) p2_src, (v16u8) cmp2);
489 dst5 = __msa_bmnz_v(dst5, (v16u8) q2_src, (v16u8) cmp2);
491 cmp0 = __msa_fill_d(d00 + d30 >= beta);
492 dst7 = (v16u8) __msa_fill_d(d04 + d34 >= beta);
493 cmp0 = __msa_ilvev_d((v2i64) dst7, cmp0);
494 dst6 = (v16u8) __msa_ceqi_d(cmp0, 0);
496 dst0 = __msa_bmz_v(dst0, (v16u8) p2_src, dst6);
497 dst1 = __msa_bmz_v(dst1, (v16u8) p1_src, dst6);
498 dst2 = __msa_bmz_v(dst2, (v16u8) p0_src, dst6);
499 dst3 = __msa_bmz_v(dst3, (v16u8) q0_src, dst6);
500 dst4 = __msa_bmz_v(dst4, (v16u8) q1_src, dst6);
501 dst5 = __msa_bmz_v(dst5, (v16u8) q2_src, dst6);
503 PCKEV_B4_UB(dst0, dst0, dst1, dst1, dst2, dst2, dst3, dst3,
504 dst0, dst1, dst2, dst3);
508 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
512 tmp2 = __msa_copy_u_w((v4i32) dst0, 0);
513 tmp0 = __msa_copy_u_h((v8i16) dst0, 2);
514 tmp3 = __msa_copy_u_w((v4i32) dst1, 0);
515 tmp1 = __msa_copy_u_h((v8i16) dst1, 2);
523 tmp2 = __msa_copy_u_w((v4i32) dst2, 0);
524 tmp0 = __msa_copy_u_h((v8i16) dst2, 2);
525 tmp3 = __msa_copy_u_w((v4i32) dst3, 0);
526 tmp1 = __msa_copy_u_h((v8i16) dst3, 2);
534 tmp2 = __msa_copy_u_w((v4i32) dst4, 0);
535 tmp0 = __msa_copy_u_h((v8i16) dst4, 2);
536 tmp3 = __msa_copy_u_w((v4i32) dst5, 0);
537 tmp1 = __msa_copy_u_h((v8i16) dst5, 2);
545 tmp2 = __msa_copy_u_w((v4i32) dst6, 0);
546 tmp0 = __msa_copy_u_h((v8i16) dst6, 2);
547 tmp3 = __msa_copy_u_w((v4i32) dst7, 0);
548 tmp1 = __msa_copy_u_h((v8i16) dst7, 2);
562 uint8_t *p1_ptr = src - (stride << 1);
566 v2i64 cmp0, cmp1, p_is_pcm_vec, q_is_pcm_vec;
567 v8u16 p1, p0,
q0,
q1;
568 v8i16 tc_pos, tc_neg;
570 v8i16 temp0, temp1,
delta;
572 if (!(tc[0] <= 0) || !(tc[1] <= 0)) {
573 cmp0 = (v2i64) __msa_fill_h(tc[0]);
574 cmp1 = (v2i64) __msa_fill_h(tc[1]);
575 tc_pos = (v8i16) __msa_ilvev_d(cmp1, cmp0);
578 cmp0 = __msa_fill_d(p_is_pcm[0]);
579 cmp1 = __msa_fill_d(p_is_pcm[1]);
580 p_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
581 p_is_pcm_vec = __msa_ceqi_d(p_is_pcm_vec, 0);
583 cmp0 = __msa_fill_d(q_is_pcm[0]);
584 cmp1 = __msa_fill_d(q_is_pcm[1]);
585 q_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
586 q_is_pcm_vec = __msa_ceqi_d(q_is_pcm_vec, 0);
593 ILVR_B4_UH(zero, p1, zero, p0, zero, q0, zero, q1, p1, p0, q0, q1);
595 temp0 = (v8i16) (q0 - p0);
596 temp1 = (v8i16) (p1 - q1);
599 delta = __msa_srari_h((v8i16) temp0, 3);
600 delta =
CLIP_SH(delta, tc_neg, tc_pos);
602 temp0 = (v8i16) ((v8i16) p0 +
delta);
604 temp0 = (v8i16) __msa_bmz_v((v16u8) temp0, (v16u8) p0,
605 (v16u8) p_is_pcm_vec);
607 temp1 = (v8i16) ((v8i16) q0 -
delta);
609 temp1 = (v8i16) __msa_bmz_v((v16u8) temp1, (v16u8) q0,
610 (v16u8) q_is_pcm_vec);
612 tc_pos = (v8i16) __msa_clei_s_d((v2i64) tc_pos, 0);
613 temp0 = (v8i16) __msa_bmnz_v((v16u8) temp0, (v16u8) p0, (v16u8) tc_pos);
614 temp1 = (v8i16) __msa_bmnz_v((v16u8) temp1, (v16u8) q0, (v16u8) tc_pos);
616 temp0 = (v8i16) __msa_pckev_b((v16i8) temp1, (v16i8) temp0);
625 v2i64 cmp0, cmp1, p_is_pcm_vec, q_is_pcm_vec;
626 v16u8
src0,
src1, src2, src3, src4, src5, src6, src7;
627 v8u16 p1, p0,
q0,
q1;
628 v8i16 tc_pos, tc_neg;
630 v8i16 temp0, temp1,
delta;
632 if (!(tc[0] <= 0) || !(tc[1] <= 0)) {
633 cmp0 = (v2i64) __msa_fill_h(tc[0]);
634 cmp1 = (v2i64) __msa_fill_h(tc[1]);
635 tc_pos = (v8i16) __msa_ilvev_d(cmp1, cmp0);
638 cmp0 = __msa_fill_d(p_is_pcm[0]);
639 cmp1 = __msa_fill_d(p_is_pcm[1]);
640 p_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
641 p_is_pcm_vec = __msa_ceqi_d(p_is_pcm_vec, 0);
643 cmp0 = __msa_fill_d(q_is_pcm[0]);
644 cmp1 = __msa_fill_d(q_is_pcm[1]);
645 q_is_pcm_vec = __msa_ilvev_d(cmp1, cmp0);
646 q_is_pcm_vec = __msa_ceqi_d(q_is_pcm_vec, 0);
649 LD_UB8(src, stride, src0, src1, src2, src3, src4, src5, src6, src7);
652 ILVR_B4_UH(zero, p1, zero, p0, zero, q0, zero, q1, p1, p0, q0, q1);
654 temp0 = (v8i16) (q0 - p0);
655 temp1 = (v8i16) (p1 - q1);
658 delta = __msa_srari_h((v8i16) temp0, 3);
659 delta =
CLIP_SH(delta, tc_neg, tc_pos);
661 temp0 = (v8i16) ((v8i16) p0 +
delta);
663 temp0 = (v8i16) __msa_bmz_v((v16u8) temp0, (v16u8) p0,
664 (v16u8) p_is_pcm_vec);
666 temp1 = (v8i16) ((v8i16) q0 -
delta);
668 temp1 = (v8i16) __msa_bmz_v((v16u8) temp1, (v16u8) q0,
669 (v16u8) q_is_pcm_vec);
671 tc_pos = (v8i16) __msa_clei_s_d((v2i64) tc_pos, 0);
672 temp0 = (v8i16) __msa_bmnz_v((v16u8) temp0, (v16u8) p0, (v16u8) tc_pos);
673 temp1 = (v8i16) __msa_bmnz_v((v16u8) temp1, (v16u8) q0, (v16u8) tc_pos);
675 temp0 = (v8i16) __msa_ilvev_b((v16i8) temp1, (v16i8) temp0);
687 int16_t *sao_offset_val,
692 v16i8 src0_r, src1_r;
694 v16i8 offset0 = { 0 };
695 v16i8 offset1 = { 0 };
697 v8i16 temp0, temp1, dst0, dst1;
699 offset_val =
LD_SB(sao_offset_val + 1);
700 offset_val = (v16i8) __msa_pckev_d((v2i64) offset_val, (v2i64) offset_val);
702 offset_val = __msa_pckev_b(offset_val, offset_val);
703 offset1 = (v16i8) __msa_insve_w((v4i32) offset1, 3, (v4i32) offset_val);
704 offset0 = __msa_sld_b(offset1, offset0, 28 - ((sao_left_class) & 31));
705 offset1 = __msa_sld_b(zero, offset1, 28 - ((sao_left_class) & 31));
707 if (!((sao_left_class > 12) & (sao_left_class < 29))) {
708 SWAP(offset0, offset1);
711 for (h_cnt = height >> 2; h_cnt--;) {
712 LD_UB4(src, src_stride, src0, src1, src2, src3);
713 src += (4 * src_stride);
715 ILVEV_D2_SB(src0, src1, src2, src3, src0_r, src1_r);
717 src0_r = (v16i8) __msa_pckev_w((v4i32) src1_r, (v4i32) src0_r);
718 mask = __msa_srli_b(src0_r, 3);
719 offset = __msa_vshf_b(mask, offset1, offset0);
723 ADD2(dst0, temp0, dst1, temp1, dst0, dst1);
725 dst0 = (v8i16) __msa_pckev_b((v16i8) dst1, (v16i8) dst0);
726 ST4x4_UB(dst0, dst0, 0, 1, 2, 3, dst, dst_stride);
727 dst += (4 * dst_stride);
734 int16_t *sao_offset_val,
739 v16i8 src0_r, src1_r, mask0, mask1;
741 v16i8 offset0 = { 0 };
742 v16i8 offset1 = { 0 };
744 v8i16 dst0, dst1, dst2, dst3;
745 v8i16 temp0, temp1, temp2, temp3;
747 offset_val =
LD_SB(sao_offset_val + 1);
748 offset_val = (v16i8) __msa_pckev_d((v2i64) offset_val, (v2i64) offset_val);
749 offset_val = __msa_pckev_b(offset_val, offset_val);
750 offset1 = (v16i8) __msa_insve_w((v4i32) offset1, 3, (v4i32) offset_val);
751 offset0 = __msa_sld_b(offset1, offset0, 28 - ((sao_left_class) & 31));
752 offset1 = __msa_sld_b(zero, offset1, 28 - ((sao_left_class) & 31));
754 if (!((sao_left_class > 12) & (sao_left_class < 29))) {
755 SWAP(offset0, offset1);
758 for (h_cnt = height >> 2; h_cnt--;) {
759 LD_UB4(src, src_stride, src0, src1, src2, src3);
760 src += (4 * src_stride);
762 ILVR_D2_SB(src1, src0, src3, src2, src0_r, src1_r);
764 mask0 = __msa_srli_b(src0_r, 3);
765 mask1 = __msa_srli_b(src1_r, 3);
767 offset = __msa_vshf_b(mask0, offset1, offset0);
770 offset = __msa_vshf_b(mask1, offset1, offset0);
775 ADD4(dst0, temp0, dst1, temp1, dst2, temp2, dst3, temp3,
776 dst0, dst1, dst2, dst3);
779 ST8x4_UB(dst0, dst2, dst, dst_stride);
780 dst += (4 * dst_stride);
789 int16_t *sao_offset_val,
794 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
795 v16i8 out0, out1, out2, out3;
796 v16i8 mask0, mask1, mask2, mask3;
797 v16i8 tmp0, tmp1, tmp2, tmp3, offset_val;
798 v16i8 offset0 = { 0 };
799 v16i8 offset1 = { 0 };
801 v8i16 temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
803 offset_val =
LD_SB(sao_offset_val + 1);
804 offset_val = (v16i8) __msa_pckev_d((v2i64) offset_val, (v2i64) offset_val);
805 offset_val = __msa_pckev_b(offset_val, offset_val);
806 offset1 = (v16i8) __msa_insve_w((v4i32) offset1, 3, (v4i32) offset_val);
807 offset0 = __msa_sld_b(offset1, offset0, 28 - ((sao_left_class) & 31));
808 offset1 = __msa_sld_b(zero, offset1, 28 - ((sao_left_class) & 31));
810 if (!((sao_left_class > 12) & (sao_left_class < 29))) {
811 SWAP(offset0, offset1);
814 for (h_cnt = height >> 2; h_cnt--;) {
815 for (w_cnt = 0; w_cnt < (width >> 4); w_cnt++) {
816 LD_UB4(src + w_cnt * 16, src_stride, src0, src1, src2, src3);
818 mask0 = __msa_srli_b((v16i8) src0, 3);
819 mask1 = __msa_srli_b((v16i8) src1, 3);
820 mask2 = __msa_srli_b((v16i8) src2, 3);
821 mask3 = __msa_srli_b((v16i8) src3, 3);
823 VSHF_B2_SB(offset0, offset1, offset0, offset1, mask0, mask1,
825 VSHF_B2_SB(offset0, offset1, offset0, offset1, mask2, mask3,
835 ADD4(dst0, temp0, dst1, temp1, dst2, temp2, dst3, temp3,
836 dst0, dst1, dst2, dst3);
837 ADD4(dst4, temp4, dst5, temp5, dst6, temp6, dst7, temp7,
838 dst4, dst5, dst6, dst7);
841 PCKEV_B4_SB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6,
842 out0, out1, out2, out3);
843 ST_SB4(out0, out1, out2, out3, dst + w_cnt * 16, dst_stride);
846 src += src_stride << 2;
847 dst += dst_stride << 2;
855 int16_t *sao_offset_val,
859 uint32_t dst_val0, dst_val1;
860 v8i16 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0 };
861 v16u8 const1 = (v16u8) __msa_ldi_b(1);
862 v16u8 cmp_minus10, diff_minus10, cmp_minus11, diff_minus11;
863 v16u8 src_minus10, src_minus11;
865 v16i8 src_zero0, src_zero1, src_plus10, src_plus11, dst0;
866 v8i16 offset_mask0, offset_mask1;
867 v8i16 sao_offset, src00, src01;
869 sao_offset =
LD_SH(sao_offset_val);
872 for (h_cnt = (height >> 1); h_cnt--;) {
873 LD_UB2(src, src_stride, src_minus10, src_minus11);
874 src += (2 * src_stride);
876 SLDI_B2_0_SB(src_minus10, src_minus11, src_zero0, src_zero1, 1);
877 SLDI_B2_0_SB(src_minus10, src_minus11, src_plus10, src_plus11, 2);
878 ILVR_B2_UB(src_plus10, src_minus10, src_plus11, src_minus11,
879 src_minus10, src_minus11);
880 ILVR_B2_SB(src_zero0, src_zero0, src_zero1, src_zero1, src_zero0,
883 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
884 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
885 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
886 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
888 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
889 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
890 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
891 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
893 offset_mask0 = (v8i16) (__msa_hadd_u_h(diff_minus10, diff_minus10) + 2);
894 offset_mask1 = (v8i16) (__msa_hadd_u_h(diff_minus11, diff_minus11) + 2);
896 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
897 offset_mask0, offset_mask0, offset_mask0);
898 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
899 offset_mask1, offset_mask1, offset_mask1);
900 ILVEV_B2_SH(src_zero0, zero, src_zero1, zero, src00, src01);
901 ADD2(offset_mask0, src00, offset_mask1, src01, offset_mask0,
905 dst0 = __msa_pckev_b((v16i8) offset_mask1, (v16i8) offset_mask0);
906 dst_val0 = __msa_copy_u_w((v4i32) dst0, 0);
907 dst_val1 = __msa_copy_u_w((v4i32) dst0, 2);
919 int16_t *sao_offset_val,
924 uint64_t dst_val0, dst_val1;
925 v8i16 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0 };
926 v16u8 const1 = (v16u8) __msa_ldi_b(1);
929 v16u8 cmp_minus10, diff_minus10, cmp_minus11, diff_minus11;
930 v16u8 src_minus10, src_minus11;
931 v16i8 src_zero0, src_plus10, src_zero1, src_plus11;
932 v8i16 sao_offset, src00, offset_mask0, src01, offset_mask1;
934 sao_offset =
LD_SH(sao_offset_val);
936 for (h_cnt = (height >> 1); h_cnt--;) {
937 src_minus1 = src - 1;
938 LD_UB2(src_minus1, src_stride, src_minus10, src_minus11);
940 SLDI_B2_0_SB(src_minus10, src_minus11, src_zero0, src_zero1, 1);
941 SLDI_B2_0_SB(src_minus10, src_minus11, src_plus10, src_plus11, 2);
942 ILVR_B2_UB(src_plus10, src_minus10, src_plus11, src_minus11,
943 src_minus10, src_minus11);
944 ILVR_B2_SB(src_zero0, src_zero0, src_zero1, src_zero1,
945 src_zero0, src_zero1);
947 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
948 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
949 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
950 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
952 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
953 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
954 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
955 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
957 offset_mask0 = (v8i16) (__msa_hadd_u_h(diff_minus10, diff_minus10) + 2);
958 offset_mask1 = (v8i16) (__msa_hadd_u_h(diff_minus11, diff_minus11) + 2);
960 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
961 offset_mask0, offset_mask0, offset_mask0);
962 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
963 offset_mask1, offset_mask1, offset_mask1);
964 ILVEV_B2_SH(src_zero0, zero, src_zero1, zero, src00, src01);
966 ADD2(offset_mask0, src00, offset_mask1, src01, offset_mask0,
969 PCKEV_B2_SB(offset_mask0, offset_mask0, offset_mask1, offset_mask1,
972 dst_val0 = __msa_copy_u_d((v2i64) dst0, 0);
973 dst_val1 = __msa_copy_u_d((v2i64) dst1, 0);
978 src += (src_stride << 1);
986 int16_t *sao_offset_val,
992 v16i8 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
993 v16u8 const1 = (v16u8) __msa_ldi_b(1);
995 v16u8 cmp_minus10, cmp_plus10, diff_minus10, diff_plus10, cmp_minus11;
996 v16u8 cmp_plus11, diff_minus11, diff_plus11, cmp_minus12, cmp_plus12;
997 v16u8 diff_minus12, diff_plus12, cmp_minus13, cmp_plus13, diff_minus13;
999 v16u8 src10, src11, src12, src13, dst0, dst1, dst2, dst3;
1000 v16u8 src_minus10, src_minus11, src_minus12, src_minus13;
1001 v16i8 offset_mask0, offset_mask1, offset_mask2, offset_mask3;
1002 v16i8 src_zero0, src_zero1, src_zero2, src_zero3;
1003 v16i8 src_plus10, src_plus11, src_plus12, src_plus13;
1004 v8i16
src0,
src1, src2, src3, src4, src5, src6, src7;
1005 v8i16 temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
1007 sao_offset =
LD_SB(sao_offset_val);
1008 sao_offset = __msa_pckev_b(sao_offset, sao_offset);
1010 for (h_cnt = (height >> 2); h_cnt--;) {
1011 src_minus1 = src - 1;
1012 LD_UB4(src_minus1, src_stride,
1013 src_minus10, src_minus11, src_minus12, src_minus13);
1015 for (v_cnt = 0; v_cnt < (width >> 4); v_cnt++) {
1017 dst_ptr = dst + (v_cnt << 4);
1018 LD_UB4(src_minus1, src_stride, src10, src11, src12, src13);
1020 SLDI_B2_SB(src10, src11, src_minus10, src_minus11, src_zero0,
1022 SLDI_B2_SB(src12, src13, src_minus12, src_minus13, src_zero2,
1024 SLDI_B2_SB(src10, src11, src_minus10, src_minus11, src_plus10,
1026 SLDI_B2_SB(src12, src13, src_minus12, src_minus13, src_plus12,
1029 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1030 cmp_plus10 = ((v16u8) src_zero0 == (v16u8) src_plus10);
1031 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1032 cmp_plus11 = ((v16u8) src_zero1 == (v16u8) src_plus11);
1033 cmp_minus12 = ((v16u8) src_zero2 == src_minus12);
1034 cmp_plus12 = ((v16u8) src_zero2 == (v16u8) src_plus12);
1035 cmp_minus13 = ((v16u8) src_zero3 == src_minus13);
1036 cmp_plus13 = ((v16u8) src_zero3 == (v16u8) src_plus13);
1038 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1039 diff_plus10 = __msa_nor_v(cmp_plus10, cmp_plus10);
1040 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1041 diff_plus11 = __msa_nor_v(cmp_plus11, cmp_plus11);
1042 diff_minus12 = __msa_nor_v(cmp_minus12, cmp_minus12);
1043 diff_plus12 = __msa_nor_v(cmp_plus12, cmp_plus12);
1044 diff_minus13 = __msa_nor_v(cmp_minus13, cmp_minus13);
1045 diff_plus13 = __msa_nor_v(cmp_plus13, cmp_plus13);
1047 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1048 cmp_plus10 = ((v16u8) src_plus10 < (v16u8) src_zero0);
1049 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1050 cmp_plus11 = ((v16u8) src_plus11 < (v16u8) src_zero1);
1051 cmp_minus12 = (src_minus12 < (v16u8) src_zero2);
1052 cmp_plus12 = ((v16u8) src_plus12 < (v16u8) src_zero2);
1053 cmp_minus13 = (src_minus13 < (v16u8) src_zero3);
1054 cmp_plus13 = ((v16u8) src_plus13 < (v16u8) src_zero3);
1056 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1057 diff_plus10 = __msa_bmnz_v(diff_plus10, const1, cmp_plus10);
1058 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1059 diff_plus11 = __msa_bmnz_v(diff_plus11, const1, cmp_plus11);
1060 diff_minus12 = __msa_bmnz_v(diff_minus12, const1, cmp_minus12);
1061 diff_plus12 = __msa_bmnz_v(diff_plus12, const1, cmp_plus12);
1062 diff_minus13 = __msa_bmnz_v(diff_minus13, const1, cmp_minus13);
1063 diff_plus13 = __msa_bmnz_v(diff_plus13, const1, cmp_plus13);
1065 offset_mask0 = 2 + (v16i8) diff_minus10 + (v16i8) diff_plus10;
1066 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
1067 offset_mask0, offset_mask0, offset_mask0);
1068 offset_mask1 = 2 + (v16i8) diff_minus11 + (v16i8) diff_plus11;
1069 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
1070 offset_mask1, offset_mask1, offset_mask1);
1071 offset_mask2 = 2 + (v16i8) diff_minus12 + (v16i8) diff_plus12;
1072 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask2,
1073 offset_mask2, offset_mask2, offset_mask2);
1074 offset_mask3 = 2 + (v16i8) diff_minus13 + (v16i8) diff_plus13;
1075 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask3,
1076 offset_mask3, offset_mask3, offset_mask3);
1086 ADD4(temp0, src0, temp1, src1, temp2, src2, temp3, src3, temp0,
1087 temp1, temp2, temp3);
1088 ADD4(temp4, src4, temp5, src5, temp6, src6, temp7, src7, temp4,
1089 temp5, temp6, temp7);
1092 PCKEV_B4_UB(temp1, temp0, temp3, temp2, temp5, temp4, temp7, temp6,
1093 dst0, dst1, dst2, dst3);
1095 src_minus10 = src10;
1096 ST_UB(dst0, dst_ptr);
1097 src_minus11 = src11;
1098 ST_UB(dst1, dst_ptr + dst_stride);
1099 src_minus12 = src12;
1100 ST_UB(dst2, dst_ptr + (dst_stride << 1));
1101 src_minus13 = src13;
1102 ST_UB(dst3, dst_ptr + (dst_stride * 3));
1105 src += (src_stride << 2);
1106 dst += (dst_stride << 2);
1114 int16_t *sao_offset_val,
1118 uint32_t dst_val0, dst_val1;
1119 v8i16 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0 };
1120 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1123 v16u8 cmp_minus10, diff_minus10, cmp_minus11, diff_minus11;
1124 v16u8 src_minus10, src_minus11, src10, src11;
1125 v16i8 src_zero0, src_zero1;
1126 v8i16 sao_offset, src00, src01, offset_mask0, offset_mask1;
1128 sao_offset =
LD_SH(sao_offset_val);
1130 LD_UB2(src - src_stride, src_stride, src_minus10, src_minus11);
1132 for (h_cnt = (height >> 1); h_cnt--;) {
1133 LD_UB2(src + src_stride, src_stride, src10, src11);
1135 src_minus10 = (v16u8) __msa_ilvr_b((v16i8) src10, (v16i8) src_minus10);
1136 src_zero0 = __msa_ilvr_b((v16i8) src_minus11, (v16i8) src_minus11);
1137 src_minus11 = (v16u8) __msa_ilvr_b((v16i8) src11, (v16i8) src_minus11);
1138 src_zero1 = __msa_ilvr_b((v16i8) src10, (v16i8) src10);
1140 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1141 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1142 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1143 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1145 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1146 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1147 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1148 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1150 offset_mask0 = (v8i16) (__msa_hadd_u_h(diff_minus10, diff_minus10) + 2);
1151 offset_mask1 = (v8i16) (__msa_hadd_u_h(diff_minus11, diff_minus11) + 2);
1153 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
1154 offset_mask0, offset_mask0, offset_mask0);
1155 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
1156 offset_mask1, offset_mask1, offset_mask1);
1157 ILVEV_B2_SH(src_zero0, zero, src_zero1, zero, src00, src01);
1158 ADD2(offset_mask0, src00, offset_mask1, src01, offset_mask0,
1161 dst0 = __msa_pckev_b((v16i8) offset_mask1, (v16i8) offset_mask0);
1163 src_minus10 = src10;
1164 src_minus11 = src11;
1166 dst_val0 = __msa_copy_u_w((v4i32) dst0, 0);
1167 dst_val1 = __msa_copy_u_w((v4i32) dst0, 2);
1173 src += (src_stride << 1);
1181 int16_t *sao_offset_val,
1185 uint64_t dst_val0, dst_val1;
1186 v8i16 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0 };
1187 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1189 v16i8 src_zero0, src_zero1, dst0, dst1;
1190 v16u8 cmp_minus10, diff_minus10, cmp_minus11, diff_minus11;
1191 v16u8 src_minus10, src_minus11, src10, src11;
1192 v8i16 sao_offset, src00, offset_mask0, src01, offset_mask1;
1194 sao_offset =
LD_SH(sao_offset_val);
1196 LD_UB2(src - src_stride, src_stride, src_minus10, src_minus11);
1198 for (h_cnt = (height >> 1); h_cnt--;) {
1199 LD_UB2(src + src_stride, src_stride, src10, src11);
1201 src_minus10 = (v16u8) __msa_ilvr_b((v16i8) src10, (v16i8) src_minus10);
1202 src_zero0 = __msa_ilvr_b((v16i8) src_minus11, (v16i8) src_minus11);
1203 src_minus11 = (v16u8) __msa_ilvr_b((v16i8) src11, (v16i8) src_minus11);
1204 src_zero1 = __msa_ilvr_b((v16i8) src10, (v16i8) src10);
1206 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1207 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1208 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1209 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1211 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1212 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1213 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1214 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1216 offset_mask0 = (v8i16) (__msa_hadd_u_h(diff_minus10, diff_minus10) + 2);
1217 offset_mask1 = (v8i16) (__msa_hadd_u_h(diff_minus11, diff_minus11) + 2);
1219 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
1220 offset_mask0, offset_mask0, offset_mask0);
1221 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
1222 offset_mask1, offset_mask1, offset_mask1);
1223 ILVEV_B2_SH(src_zero0, zero, src_zero1, zero, src00, src01);
1224 ADD2(offset_mask0, src00, offset_mask1, src01, offset_mask0,
1227 PCKEV_B2_SB(offset_mask0, offset_mask0, offset_mask1, offset_mask1,
1230 src_minus10 = src10;
1231 src_minus11 = src11;
1233 dst_val0 = __msa_copy_u_d((v2i64) dst0, 0);
1234 dst_val1 = __msa_copy_u_d((v2i64) dst1, 0);
1239 src += (src_stride << 1);
1255 v16i8 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
1256 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1257 v16u8 cmp_minus10, cmp_plus10, diff_minus10, diff_plus10, cmp_minus11;
1258 v16u8 cmp_plus11, diff_minus11, diff_plus11, cmp_minus12, cmp_plus12;
1259 v16u8 diff_minus12, diff_plus12, cmp_minus13, cmp_plus13, diff_minus13;
1261 v16u8 src10, src_minus10, dst0, src11, src_minus11, dst1;
1262 v16u8 src12, dst2, src13, dst3;
1263 v16i8 offset_mask0, offset_mask1, offset_mask2, offset_mask3, sao_offset;
1264 v8i16
src0,
src1, src2, src3, src4, src5, src6, src7;
1265 v8i16 temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
1267 sao_offset =
LD_SB(sao_offset_val);
1268 sao_offset = __msa_pckev_b(sao_offset, sao_offset);
1270 for (v_cnt = 0; v_cnt < (width >> 4); v_cnt++) {
1271 src = src_orig + (v_cnt << 4);
1272 dst = dst_orig + (v_cnt << 4);
1274 LD_UB2(src - src_stride, src_stride, src_minus10, src_minus11);
1276 for (h_cnt = (height >> 2); h_cnt--;) {
1277 LD_UB4(src + src_stride, src_stride, src10, src11, src12, src13);
1279 cmp_minus10 = (src_minus11 == src_minus10);
1280 cmp_plus10 = (src_minus11 == src10);
1281 cmp_minus11 = (src10 == src_minus11);
1282 cmp_plus11 = (src10 == src11);
1283 cmp_minus12 = (src11 == src10);
1284 cmp_plus12 = (src11 == src12);
1285 cmp_minus13 = (src12 == src11);
1286 cmp_plus13 = (src12 == src13);
1288 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1289 diff_plus10 = __msa_nor_v(cmp_plus10, cmp_plus10);
1290 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1291 diff_plus11 = __msa_nor_v(cmp_plus11, cmp_plus11);
1292 diff_minus12 = __msa_nor_v(cmp_minus12, cmp_minus12);
1293 diff_plus12 = __msa_nor_v(cmp_plus12, cmp_plus12);
1294 diff_minus13 = __msa_nor_v(cmp_minus13, cmp_minus13);
1295 diff_plus13 = __msa_nor_v(cmp_plus13, cmp_plus13);
1297 cmp_minus10 = (src_minus10 < src_minus11);
1298 cmp_plus10 = (src10 < src_minus11);
1299 cmp_minus11 = (src_minus11 < src10);
1300 cmp_plus11 = (src11 < src10);
1301 cmp_minus12 = (src10 < src11);
1302 cmp_plus12 = (src12 < src11);
1303 cmp_minus13 = (src11 < src12);
1304 cmp_plus13 = (src13 < src12);
1306 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1307 diff_plus10 = __msa_bmnz_v(diff_plus10, const1, cmp_plus10);
1308 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1309 diff_plus11 = __msa_bmnz_v(diff_plus11, const1, cmp_plus11);
1310 diff_minus12 = __msa_bmnz_v(diff_minus12, const1, cmp_minus12);
1311 diff_plus12 = __msa_bmnz_v(diff_plus12, const1, cmp_plus12);
1312 diff_minus13 = __msa_bmnz_v(diff_minus13, const1, cmp_minus13);
1313 diff_plus13 = __msa_bmnz_v(diff_plus13, const1, cmp_plus13);
1315 offset_mask0 = 2 + (v16i8) diff_minus10 + (v16i8) diff_plus10;
1316 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1317 offset_mask0, offset_mask0, offset_mask0, offset_mask0);
1318 offset_mask1 = 2 + (v16i8) diff_minus11 + (v16i8) diff_plus11;
1319 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1320 offset_mask1, offset_mask1, offset_mask1, offset_mask1);
1321 offset_mask2 = 2 + (v16i8) diff_minus12 + (v16i8) diff_plus12;
1322 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1323 offset_mask2, offset_mask2, offset_mask2, offset_mask2);
1324 offset_mask3 = 2 + (v16i8) diff_minus13 + (v16i8) diff_plus13;
1325 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1326 offset_mask3, offset_mask3, offset_mask3, offset_mask3);
1336 ADD4(temp0, src0, temp1, src1, temp2, src2, temp3, src3, temp0,
1337 temp1, temp2, temp3);
1338 ADD4(temp4, src4, temp5, src5, temp6, src6, temp7, src7, temp4,
1339 temp5, temp6, temp7);
1342 PCKEV_B4_UB(temp1, temp0, temp3, temp2, temp5, temp4, temp7, temp6,
1343 dst0, dst1, dst2, dst3);
1345 src_minus10 = src12;
1346 src_minus11 = src13;
1348 ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
1350 src += (src_stride << 2);
1351 dst += (dst_stride << 2);
1360 int16_t *sao_offset_val,
1365 uint32_t dst_val0, dst_val1;
1366 v8i16 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0 };
1367 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1369 v16u8 cmp_minus10, diff_minus10, src_minus10, cmp_minus11, diff_minus11;
1370 v16u8 src_minus11, src10, src11;
1371 v16i8 src_plus0, src_zero0, src_plus1, src_zero1, dst0;
1372 v8i16 sao_offset, src00, src01, offset_mask0, offset_mask1;
1374 sao_offset =
LD_SH(sao_offset_val);
1377 LD_UB2(src_orig - src_stride, src_stride, src_minus10, src_minus11);
1379 for (h_cnt = (height >> 1); h_cnt--;) {
1380 LD_UB2(src_orig + src_stride, src_stride, src10, src11);
1382 SLDI_B2_0_SB(src_minus11, src10, src_zero0, src_zero1, 1);
1385 ILVR_B2_UB(src_plus0, src_minus10, src_plus1, src_minus11, src_minus10,
1387 ILVR_B2_SB(src_zero0, src_zero0, src_zero1, src_zero1, src_zero0,
1390 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1391 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1392 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1393 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1395 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1396 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1397 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1398 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1400 offset_mask0 = (v8i16) (__msa_hadd_u_h(diff_minus10, diff_minus10) + 2);
1401 offset_mask1 = (v8i16) (__msa_hadd_u_h(diff_minus11, diff_minus11) + 2);
1403 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
1404 offset_mask0, offset_mask0, offset_mask0);
1405 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
1406 offset_mask1, offset_mask1, offset_mask1);
1407 ILVEV_B2_SH(src_zero0, zero, src_zero1, zero, src00, src01);
1408 ADD2(offset_mask0, src00, offset_mask1, src01, offset_mask0,
1412 dst0 = __msa_pckev_b((v16i8) offset_mask1, (v16i8) offset_mask0);
1414 src_minus10 = src10;
1415 src_minus11 = src11;
1417 dst_val0 = __msa_copy_u_w((v4i32) dst0, 0);
1418 dst_val1 = __msa_copy_u_w((v4i32) dst0, 2);
1424 src_orig += (src_stride << 1);
1432 int16_t *sao_offset_val,
1437 uint64_t dst_val0, dst_val1;
1438 v8i16 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0 };
1439 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1441 v16u8 cmp_minus10, diff_minus10, cmp_minus11, diff_minus11;
1442 v16u8 src_minus10, src10, src_minus11, src11;
1443 v16i8 src_zero0, src_plus10, src_zero1, src_plus11, dst0, dst1;
1444 v8i16 sao_offset, src00, offset_mask0, src01, offset_mask1;
1446 sao_offset =
LD_SH(sao_offset_val);
1449 LD_UB2(src_orig - src_stride, src_stride, src_minus10, src_minus11);
1451 for (h_cnt = (height >> 1); h_cnt--;) {
1452 LD_UB2(src_orig + src_stride, src_stride, src10, src11);
1454 SLDI_B2_0_SB(src_minus11, src10, src_zero0, src_zero1, 1);
1457 ILVR_B2_UB(src_plus10, src_minus10, src_plus11, src_minus11,
1458 src_minus10, src_minus11);
1459 ILVR_B2_SB(src_zero0, src_zero0, src_zero1, src_zero1,
1460 src_zero0, src_zero1);
1462 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1463 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1464 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1465 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1467 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1468 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1469 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1470 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1472 offset_mask0 = (v8i16) (__msa_hadd_u_h(diff_minus10, diff_minus10) + 2);
1473 offset_mask1 = (v8i16) (__msa_hadd_u_h(diff_minus11, diff_minus11) + 2);
1475 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
1476 offset_mask0, offset_mask0, offset_mask0);
1477 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
1478 offset_mask1, offset_mask1, offset_mask1);
1479 ILVEV_B2_SH(src_zero0, zero, src_zero1, zero, src00, src01);
1480 ADD2(offset_mask0, src00, offset_mask1, src01, offset_mask0,
1483 PCKEV_B2_SB(offset_mask0, offset_mask0, offset_mask1, offset_mask1,
1486 src_minus10 = src10;
1487 src_minus11 = src11;
1489 dst_val0 = __msa_copy_u_d((v2i64) dst0, 0);
1490 dst_val1 = __msa_copy_u_d((v2i64) dst1, 0);
1496 src_orig += (src_stride << 1);
1512 v16i8 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
1513 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1514 v16u8 cmp_minus10, cmp_plus10, diff_minus10, diff_plus10, cmp_minus11;
1515 v16u8 cmp_plus11, diff_minus11, diff_plus11, cmp_minus12, cmp_plus12;
1516 v16u8 diff_minus12, diff_plus12, cmp_minus13, cmp_plus13, diff_minus13;
1517 v16u8 diff_plus13, src_minus14, src_plus13;
1518 v16i8 offset_mask0, offset_mask1, offset_mask2, offset_mask3;
1519 v16u8 src10, src_minus10, dst0, src11, src_minus11, dst1;
1520 v16u8 src12, src_minus12, dst2, src13, src_minus13, dst3;
1521 v16i8 src_zero0, src_plus10, src_zero1, src_plus11, src_zero2, src_plus12;
1522 v16i8 src_zero3, sao_offset;
1523 v8i16
src0,
src1, src2, src3, src4, src5, src6, src7;
1524 v8i16 temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
1526 sao_offset =
LD_SB(sao_offset_val);
1527 sao_offset = __msa_pckev_b(sao_offset, sao_offset);
1529 for (h_cnt = (height >> 2); h_cnt--;) {
1532 LD_UB4(src_orig, src_stride,
1533 src_minus11, src_minus12, src_minus13, src_minus14);
1535 for (v_cnt = 0; v_cnt < (width >> 4); v_cnt++) {
1536 src_minus10 =
LD_UB(src_orig - src_stride);
1537 LD_UB4(src_orig + 16, src_stride, src10, src11, src12, src13);
1538 src_plus13 =
LD_UB(src + 1 + (v_cnt << 4) + (src_stride << 2));
1541 SLDI_B2_SB(src10, src11, src_minus11, src_minus12, src_zero0,
1543 SLDI_B2_SB(src12, src13, src_minus13, src_minus14, src_zero2,
1545 SLDI_B2_SB(src11, src12, src_minus12, src_minus13, src_plus10,
1548 src_plus12 = __msa_sldi_b((v16i8) src13, (v16i8) src_minus14, 2);
1550 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1551 cmp_plus10 = ((v16u8) src_zero0 == (v16u8) src_plus10);
1552 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1553 cmp_plus11 = ((v16u8) src_zero1 == (v16u8) src_plus11);
1554 cmp_minus12 = ((v16u8) src_zero2 == src_minus12);
1555 cmp_plus12 = ((v16u8) src_zero2 == (v16u8) src_plus12);
1556 cmp_minus13 = ((v16u8) src_zero3 == src_minus13);
1557 cmp_plus13 = ((v16u8) src_zero3 == src_plus13);
1559 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1560 diff_plus10 = __msa_nor_v(cmp_plus10, cmp_plus10);
1561 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1562 diff_plus11 = __msa_nor_v(cmp_plus11, cmp_plus11);
1563 diff_minus12 = __msa_nor_v(cmp_minus12, cmp_minus12);
1564 diff_plus12 = __msa_nor_v(cmp_plus12, cmp_plus12);
1565 diff_minus13 = __msa_nor_v(cmp_minus13, cmp_minus13);
1566 diff_plus13 = __msa_nor_v(cmp_plus13, cmp_plus13);
1568 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1569 cmp_plus10 = ((v16u8) src_plus10 < (v16u8) src_zero0);
1570 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1571 cmp_plus11 = ((v16u8) src_plus11 < (v16u8) src_zero1);
1572 cmp_minus12 = (src_minus12 < (v16u8) src_zero2);
1573 cmp_plus12 = ((v16u8) src_plus12 < (v16u8) src_zero2);
1574 cmp_minus13 = (src_minus13 < (v16u8) src_zero3);
1575 cmp_plus13 = (src_plus13 < (v16u8) src_zero3);
1577 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1578 diff_plus10 = __msa_bmnz_v(diff_plus10, const1, cmp_plus10);
1579 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1580 diff_plus11 = __msa_bmnz_v(diff_plus11, const1, cmp_plus11);
1581 diff_minus12 = __msa_bmnz_v(diff_minus12, const1, cmp_minus12);
1582 diff_plus12 = __msa_bmnz_v(diff_plus12, const1, cmp_plus12);
1583 diff_minus13 = __msa_bmnz_v(diff_minus13, const1, cmp_minus13);
1584 diff_plus13 = __msa_bmnz_v(diff_plus13, const1, cmp_plus13);
1586 offset_mask0 = 2 + (v16i8) diff_minus10 + (v16i8) diff_plus10;
1587 offset_mask1 = 2 + (v16i8) diff_minus11 + (v16i8) diff_plus11;
1588 offset_mask2 = 2 + (v16i8) diff_minus12 + (v16i8) diff_plus12;
1589 offset_mask3 = 2 + (v16i8) diff_minus13 + (v16i8) diff_plus13;
1591 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1592 offset_mask0, offset_mask0, offset_mask0, offset_mask0);
1593 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1594 offset_mask1, offset_mask1, offset_mask1, offset_mask1);
1595 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1596 offset_mask2, offset_mask2, offset_mask2, offset_mask2);
1597 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1598 offset_mask3, offset_mask3, offset_mask3, offset_mask3);
1608 ADD4(temp0, src0, temp1, src1, temp2, src2, temp3, src3, temp0,
1609 temp1, temp2, temp3);
1610 ADD4(temp4, src4, temp5, src5, temp6, src6, temp7, src7, temp4,
1611 temp5, temp6, temp7);
1614 PCKEV_B4_UB(temp1, temp0, temp3, temp2, temp5, temp4,
1615 temp7, temp6, dst0, dst1, dst2, dst3);
1617 src_minus11 = src10;
1618 src_minus12 = src11;
1619 src_minus13 = src12;
1620 src_minus14 = src13;
1622 ST_UB4(dst0, dst1, dst2, dst3, dst_orig, dst_stride);
1626 src += (src_stride << 2);
1627 dst += (dst_stride << 2);
1635 int16_t *sao_offset_val,
1640 uint32_t dst_val0, dst_val1;
1641 v8i16 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0 };
1642 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1644 v16i8 src_zero0, src_zero1, dst0;
1645 v16u8 cmp_minus10, diff_minus10, cmp_minus11, diff_minus11;
1646 v16u8 src_minus10, src10, src_minus11, src11;
1647 v8i16 offset_mask0, offset_mask1, sao_offset, src00, src01;
1649 sao_offset =
LD_SH(sao_offset_val);
1652 LD_UB2(src_orig - src_stride, src_stride, src_minus10, src_minus11);
1654 for (h_cnt = (height >> 1); h_cnt--;) {
1655 LD_UB2(src_orig + src_stride, src_stride, src10, src11);
1657 SLDI_B2_0_SB(src_minus11, src10, src_zero0, src_zero1, 1);
1658 SLDI_B2_0_UB(src_minus10, src_minus11, src_minus10, src_minus11, 2);
1660 ILVR_B2_UB(src10, src_minus10, src11, src_minus11, src_minus10,
1662 ILVR_B2_SB(src_zero0, src_zero0, src_zero1, src_zero1, src_zero0,
1665 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1666 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1667 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1668 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1670 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1671 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1672 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1673 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1675 offset_mask0 = (v8i16) (__msa_hadd_u_h(diff_minus10, diff_minus10) + 2);
1676 offset_mask1 = (v8i16) (__msa_hadd_u_h(diff_minus11, diff_minus11) + 2);
1678 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
1679 offset_mask0, offset_mask0, offset_mask0);
1680 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
1681 offset_mask1, offset_mask1, offset_mask1);
1682 ILVEV_B2_SH(src_zero0, zero, src_zero1, zero, src00, src01);
1683 ADD2(offset_mask0, src00, offset_mask1, src01, offset_mask0,
1686 dst0 = __msa_pckev_b((v16i8) offset_mask1, (v16i8) offset_mask0);
1688 src_minus10 = src10;
1689 src_minus11 = src11;
1691 dst_val0 = __msa_copy_u_w((v4i32) dst0, 0);
1692 dst_val1 = __msa_copy_u_w((v4i32) dst0, 2);
1699 src_orig += (src_stride << 1);
1707 int16_t *sao_offset_val,
1712 uint64_t dst_val0, dst_val1;
1713 v8i16 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0 };
1714 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1716 v16i8 src_zero0, src_zero1, dst0, dst1;
1717 v16u8 cmp_minus10, diff_minus10, cmp_minus11, diff_minus11;
1718 v16u8 src_minus10, src10, src_minus11, src11;
1719 v8i16 sao_offset, src00, offset_mask0, src01, offset_mask1;
1721 sao_offset =
LD_SH(sao_offset_val);
1724 LD_UB2(src_orig - src_stride, src_stride, src_minus10, src_minus11);
1726 for (h_cnt = (height >> 1); h_cnt--;) {
1727 LD_UB2(src_orig + src_stride, src_stride, src10, src11);
1729 SLDI_B2_0_SB(src_minus11, src10, src_zero0, src_zero1, 1);
1730 SLDI_B2_0_UB(src_minus10, src_minus11, src_minus10, src_minus11, 2);
1731 ILVR_B2_UB(src10, src_minus10, src11, src_minus11, src_minus10,
1733 ILVR_B2_SB(src_zero0, src_zero0, src_zero1, src_zero1, src_zero0,
1736 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1737 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1738 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1739 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1741 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1742 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1743 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1744 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1746 offset_mask0 = (v8i16) (__msa_hadd_u_h(diff_minus10, diff_minus10) + 2);
1747 offset_mask1 = (v8i16) (__msa_hadd_u_h(diff_minus11, diff_minus11) + 2);
1749 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask0,
1750 offset_mask0, offset_mask0, offset_mask0);
1751 VSHF_H2_SH(edge_idx, edge_idx, sao_offset, sao_offset, offset_mask1,
1752 offset_mask1, offset_mask1, offset_mask1);
1753 ILVEV_B2_SH(src_zero0, zero, src_zero1, zero, src00, src01);
1754 ADD2(offset_mask0, src00, offset_mask1, src01, offset_mask0,
1757 PCKEV_B2_SB(offset_mask0, offset_mask0, offset_mask1, offset_mask1,
1760 src_minus10 = src10;
1761 src_minus11 = src11;
1763 dst_val0 = __msa_copy_u_d((v2i64) dst0, 0);
1764 dst_val1 = __msa_copy_u_d((v2i64) dst1, 0);
1771 src_orig += (src_stride << 1);
1786 v16i8 edge_idx = { 1, 2, 0, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
1787 v16u8 const1 = (v16u8) __msa_ldi_b(1);
1788 v16u8 dst0, dst1, dst2, dst3;
1789 v16u8 cmp_minus10, cmp_minus11, cmp_minus12, cmp_minus13, cmp_plus10;
1790 v16u8 cmp_plus11, cmp_plus12, cmp_plus13, diff_minus10, diff_minus11;
1791 v16u8 diff_minus12, diff_minus13, diff_plus10, diff_plus11, diff_plus12;
1792 v16u8 diff_plus13, src10, src11, src12, src13, src_minus10, src_minus11;
1793 v16u8 src_plus10, src_plus11, src_plus12, src_plus13;
1794 v16i8 src_minus12, src_minus13, src_zero0, src_zero1, src_zero2, src_zero3;
1795 v16i8 offset_mask0, offset_mask1, offset_mask2, offset_mask3, sao_offset;
1796 v8i16
src0,
src1, src2, src3, src4, src5, src6, src7;
1797 v8i16 temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
1799 sao_offset =
LD_SB(sao_offset_val);
1800 sao_offset = __msa_pckev_b(sao_offset, sao_offset);
1802 for (h_cnt = (height >> 2); h_cnt--;) {
1806 LD_UB4(src_orig, src_stride,
1807 src_minus11, src_plus10, src_plus11, src_plus12);
1809 for (v_cnt = 0; v_cnt < (width >> 4); v_cnt++) {
1810 src_minus10 =
LD_UB(src_orig + 2 - src_stride);
1811 LD_UB4(src_orig + 16, src_stride, src10, src11, src12, src13);
1812 src_plus13 =
LD_UB(src_orig + (src_stride << 2));
1815 src_zero0 = __msa_sldi_b((v16i8) src10, (v16i8) src_minus11, 1);
1816 cmp_minus10 = ((v16u8) src_zero0 == src_minus10);
1817 cmp_plus10 = ((v16u8) src_zero0 == src_plus10);
1819 src_zero1 = __msa_sldi_b((v16i8) src11, (v16i8) src_plus10, 1);
1820 src_minus11 = (v16u8) __msa_sldi_b((v16i8) src10,
1821 (v16i8) src_minus11, 2);
1822 cmp_minus11 = ((v16u8) src_zero1 == src_minus11);
1823 cmp_plus11 = ((v16u8) src_zero1 == src_plus11);
1825 src_zero2 = __msa_sldi_b((v16i8) src12, (v16i8) src_plus11, 1);
1826 src_minus12 = __msa_sldi_b((v16i8) src11, (v16i8) src_plus10, 2);
1827 cmp_minus12 = ((v16u8) src_zero2 == (v16u8) src_minus12);
1828 cmp_plus12 = ((v16u8) src_zero2 == src_plus12);
1830 src_zero3 = __msa_sldi_b((v16i8) src13, (v16i8) src_plus12, 1);
1831 src_minus13 = __msa_sldi_b((v16i8) src12, (v16i8) src_plus11, 2);
1832 cmp_minus13 = ((v16u8) src_zero3 == (v16u8) src_minus13);
1833 cmp_plus13 = ((v16u8) src_zero3 == src_plus13);
1835 diff_minus10 = __msa_nor_v(cmp_minus10, cmp_minus10);
1836 diff_plus10 = __msa_nor_v(cmp_plus10, cmp_plus10);
1837 diff_minus11 = __msa_nor_v(cmp_minus11, cmp_minus11);
1838 diff_plus11 = __msa_nor_v(cmp_plus11, cmp_plus11);
1839 diff_minus12 = __msa_nor_v(cmp_minus12, cmp_minus12);
1840 diff_plus12 = __msa_nor_v(cmp_plus12, cmp_plus12);
1841 diff_minus13 = __msa_nor_v(cmp_minus13, cmp_minus13);
1842 diff_plus13 = __msa_nor_v(cmp_plus13, cmp_plus13);
1844 cmp_minus10 = (src_minus10 < (v16u8) src_zero0);
1845 cmp_plus10 = (src_plus10 < (v16u8) src_zero0);
1846 cmp_minus11 = (src_minus11 < (v16u8) src_zero1);
1847 cmp_plus11 = (src_plus11 < (v16u8) src_zero1);
1848 cmp_minus12 = ((v16u8) src_minus12 < (v16u8) src_zero2);
1849 cmp_plus12 = (src_plus12 < (v16u8) src_zero2);
1850 cmp_minus13 = ((v16u8) src_minus13 < (v16u8) src_zero3);
1851 cmp_plus13 = (src_plus13 < (v16u8) src_zero3);
1853 diff_minus10 = __msa_bmnz_v(diff_minus10, const1, cmp_minus10);
1854 diff_plus10 = __msa_bmnz_v(diff_plus10, const1, cmp_plus10);
1855 diff_minus11 = __msa_bmnz_v(diff_minus11, const1, cmp_minus11);
1856 diff_plus11 = __msa_bmnz_v(diff_plus11, const1, cmp_plus11);
1857 diff_minus12 = __msa_bmnz_v(diff_minus12, const1, cmp_minus12);
1858 diff_plus12 = __msa_bmnz_v(diff_plus12, const1, cmp_plus12);
1859 diff_minus13 = __msa_bmnz_v(diff_minus13, const1, cmp_minus13);
1860 diff_plus13 = __msa_bmnz_v(diff_plus13, const1, cmp_plus13);
1862 offset_mask0 = 2 + (v16i8) diff_minus10 + (v16i8) diff_plus10;
1863 offset_mask1 = 2 + (v16i8) diff_minus11 + (v16i8) diff_plus11;
1864 offset_mask2 = 2 + (v16i8) diff_minus12 + (v16i8) diff_plus12;
1865 offset_mask3 = 2 + (v16i8) diff_minus13 + (v16i8) diff_plus13;
1867 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1868 offset_mask0, offset_mask0, offset_mask0, offset_mask0);
1869 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1870 offset_mask1, offset_mask1, offset_mask1, offset_mask1);
1871 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1872 offset_mask2, offset_mask2, offset_mask2, offset_mask2);
1873 VSHF_B2_SB(edge_idx, edge_idx, sao_offset, sao_offset,
1874 offset_mask3, offset_mask3, offset_mask3, offset_mask3);
1885 ADD4(temp0, src0, temp1, src1, temp2, src2, temp3, src3, temp0,
1886 temp1, temp2, temp3);
1887 ADD4(temp4, src4, temp5, src5, temp6, src6, temp7, src7, temp4,
1888 temp5, temp6, temp7);
1891 PCKEV_B4_UB(temp1, temp0, temp3, temp2, temp5, temp4, temp7, temp6,
1892 dst0, dst1, dst2, dst3);
1894 src_minus11 = src10;
1899 ST_UB4(dst0, dst1, dst2, dst3, dst_orig, dst_stride);
1903 src += (src_stride << 2);
1904 dst += (dst_stride << 2);
1909 ptrdiff_t src_stride,
1917 ptrdiff_t src_stride,
1925 ptrdiff_t src_stride,
1933 ptrdiff_t src_stride,
1941 ptrdiff_t stride_dst, ptrdiff_t stride_src,
1942 int16_t *sao_offset_val,
int sao_left_class,
1947 sao_left_class, sao_offset_val,
1948 width - (width % 16), height);
1949 dst += width - (width % 16);
1950 src += width - (width % 16);
1956 sao_left_class, sao_offset_val, height);
1964 sao_left_class, sao_offset_val, height);
1969 ptrdiff_t stride_dst,
1970 int16_t *sao_offset_val,
1973 ptrdiff_t stride_src = (2 * 64 + 32) /
sizeof(
uint8_t);
1981 width - (width % 16),
1983 dst += width - (width % 16);
1984 src += width - (width % 16);
1991 sao_offset_val, height);
2000 sao_offset_val, height);
2009 width - (width % 16),
2011 dst += width - (width % 16);
2012 src += width - (width % 16);
2019 sao_offset_val, height);
2028 sao_offset_val, height);
2037 width - (width % 16),
2039 dst += width - (width % 16);
2040 src += width - (width % 16);
2047 sao_offset_val, height);
2056 sao_offset_val, height);
2065 width - (width % 16),
2067 dst += width - (width % 16);
2068 src += width - (width % 16);
2075 sao_offset_val, height);
2084 sao_offset_val, height);
static void hevc_sao_band_filter_16multiple_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int32_t sao_left_class, int16_t *sao_offset_val, int32_t width, int32_t height)
static void hevc_sao_edge_filter_0degree_8width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t height)
#define TRANSPOSE8x4_UB_UH(...)
static void hevc_sao_band_filter_8width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int32_t sao_left_class, int16_t *sao_offset_val, int32_t height)
#define UNPCK_SB_SH(in, out0, out1)
void ff_hevc_loop_filter_chroma_h_8_msa(uint8_t *src, ptrdiff_t src_stride, int32_t *tc, uint8_t *no_p, uint8_t *no_q)
static void hevc_loopfilter_luma_hor_msa(uint8_t *src, int32_t stride, int32_t beta, int32_t *tc, uint8_t *p_is_pcm, uint8_t *q_is_pcm)
void ff_hevc_loop_filter_luma_v_8_msa(uint8_t *src, ptrdiff_t src_stride, int32_t beta, int32_t *tc, uint8_t *no_p, uint8_t *no_q)
static const uint8_t q1[256]
#define TRANSPOSE8x8_UB_UH(...)
#define ST4x4_UB(in0, in1, idx0, idx1, idx2, idx3, pdst, stride)
void ff_hevc_sao_band_filter_0_8_msa(uint8_t *dst, uint8_t *src, ptrdiff_t stride_dst, ptrdiff_t stride_src, int16_t *sao_offset_val, int sao_left_class, int width, int height)
#define UNPCK_UB_SH(in, out0, out1)
#define CLIP_SH_0_255(in)
#define CLIP_SH2_0_255(in0, in1)
#define ADD4(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2, out3)
static void hevc_sao_edge_filter_0degree_16multiple_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t width, int32_t height)
#define ST8x2_UB(in, pdst, stride)
static const uint16_t mask[17]
static void hevc_sao_band_filter_4width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int32_t sao_left_class, int16_t *sao_offset_val, int32_t height)
static void hevc_sao_edge_filter_90degree_4width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t height)
static const uint8_t offset[127][2]
static const uint8_t q0[256]
static void hevc_sao_edge_filter_0degree_4width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t height)
#define CLIP_SH4_0_255(in0, in1, in2, in3)
#define SLDI_B2_0_UB(...)
static void hevc_sao_edge_filter_45degree_4width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t height)
#define SLDI_B2_0_SB(...)
#define ST2x4_UB(in, stidx, pdst, stride)
static void hevc_sao_edge_filter_45degree_16multiple_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t width, int32_t height)
static void hevc_sao_edge_filter_135degree_4width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t height)
void ff_hevc_loop_filter_chroma_v_8_msa(uint8_t *src, ptrdiff_t src_stride, int32_t *tc, uint8_t *no_p, uint8_t *no_q)
#define TRANSPOSE8x8_UB_UB(...)
static void hevc_loopfilter_chroma_hor_msa(uint8_t *src, int32_t stride, int32_t *tc, uint8_t *p_is_pcm, uint8_t *q_is_pcm)
#define CLIP_SH(in, min, max)
static void hevc_sao_edge_filter_90degree_16multiple_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t width, int32_t height)
#define ADD2(in0, in1, in2, in3, out0, out1)
static void hevc_loopfilter_luma_ver_msa(uint8_t *src, int32_t stride, int32_t beta, int32_t *tc, uint8_t *p_is_pcm, uint8_t *q_is_pcm)
static void hevc_sao_edge_filter_90degree_8width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t height)
static void hevc_sao_edge_filter_45degree_8width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t height)
GLint GLenum GLboolean GLsizei stride
static void hevc_loopfilter_chroma_ver_msa(uint8_t *src, int32_t stride, int32_t *tc, uint8_t *p_is_pcm, uint8_t *q_is_pcm)
void ff_hevc_sao_edge_filter_8_msa(uint8_t *dst, uint8_t *src, ptrdiff_t stride_dst, int16_t *sao_offset_val, int eo, int width, int height)
#define ST8x4_UB(in0, in1, pdst, stride)
void ff_hevc_loop_filter_luma_h_8_msa(uint8_t *src, ptrdiff_t src_stride, int32_t beta, int32_t *tc, uint8_t *no_p, uint8_t *no_q)
static void hevc_sao_edge_filter_135degree_8width_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t height)
static void hevc_sao_edge_filter_135degree_16multiple_msa(uint8_t *dst, int32_t dst_stride, uint8_t *src, int32_t src_stride, int16_t *sao_offset_val, int32_t width, int32_t height)