1 /*
2 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
3 * Copyright (c) 2006 Stefan Gehrer <stefan.gehrer@gmx.de>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
25 * @author Stefan Gehrer <stefan.gehrer@gmx.de>
26 */
27
28 #include "avcodec.h"
29 #include "golomb.h"
30 #include "h264chroma.h"
31 #include "idctdsp.h"
32 #include "mathops.h"
33 #include "qpeldsp.h"
34 #include "cavs.h"
35
36 static const uint8_t alpha_tab[64] = {
37 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3,
38 4, 4, 5, 5, 6, 7, 8, 9, 10, 11, 12, 13, 15, 16, 18, 20,
39 22, 24, 26, 28, 30, 33, 33, 35, 35, 36, 37, 37, 39, 39, 42, 44,
40 46, 48, 50, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64
41 };
42
43 static const uint8_t beta_tab[64] = {
44 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2,
45 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 6, 6,
46 6, 7, 7, 7, 8, 8, 8, 9, 9, 10, 10, 11, 11, 12, 13, 14,
47 15, 16, 17, 18, 19, 20, 21, 22, 23, 23, 24, 24, 25, 25, 26, 27
48 };
49
50 static const uint8_t tc_tab[64] = {
51 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
52 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
53 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4,
54 5, 5, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9, 9
55 };
56
57 /** mark block as unavailable, i.e. out of picture
58 * or not yet decoded */
59 static const cavs_vector un_mv = { 0, 0, 1, NOT_AVAIL };
60
61 static const int8_t left_modifier_l[8] = { 0, -1, 6, -1, -1, 7, 6, 7 };
62 static const int8_t top_modifier_l[8] = { -1, 1, 5, -1, -1, 5, 7, 7 };
63 static const int8_t left_modifier_c[7] = { 5, -1, 2, -1, 6, 5, 6 };
64 static const int8_t top_modifier_c[7] = { 4, 1, -1, -1, 4, 6, 6 };
65
66 /*****************************************************************************
67 *
68 * in-loop deblocking filter
69 *
70 ****************************************************************************/
71
get_bs(cavs_vector * mvP,cavs_vector * mvQ,int b)72 static inline int get_bs(cavs_vector *mvP, cavs_vector *mvQ, int b)
73 {
74 if ((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
75 return 2;
76 if((abs(mvP->x - mvQ->x) >= 4) ||
77 (abs(mvP->y - mvQ->y) >= 4) ||
78 (mvP->ref != mvQ->ref))
79 return 1;
80 if (b) {
81 mvP += MV_BWD_OFFS;
82 mvQ += MV_BWD_OFFS;
83 if((abs(mvP->x - mvQ->x) >= 4) ||
84 (abs(mvP->y - mvQ->y) >= 4) ||
85 (mvP->ref != mvQ->ref))
86 return 1;
87 }
88 return 0;
89 }
90
91 #define SET_PARAMS \
92 alpha = alpha_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)]; \
93 beta = beta_tab[av_clip_uintp2(qp_avg + h->beta_offset, 6)]; \
94 tc = tc_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)];
95
96 /**
97 * in-loop deblocking filter for a single macroblock
98 *
99 * boundary strength (bs) mapping:
100 *
101 * --4---5--
102 * 0 2 |
103 * | 6 | 7 |
104 * 1 3 |
105 * ---------
106 */
ff_cavs_filter(AVSContext * h,enum cavs_mb mb_type)107 void ff_cavs_filter(AVSContext *h, enum cavs_mb mb_type)
108 {
109 uint8_t bs[8];
110 int qp_avg, alpha, beta, tc;
111 int i;
112
113 /* save un-deblocked lines */
114 h->topleft_border_y = h->top_border_y[h->mbx * 16 + 15];
115 h->topleft_border_u = h->top_border_u[h->mbx * 10 + 8];
116 h->topleft_border_v = h->top_border_v[h->mbx * 10 + 8];
117 memcpy(&h->top_border_y[h->mbx * 16], h->cy + 15 * h->l_stride, 16);
118 memcpy(&h->top_border_u[h->mbx * 10 + 1], h->cu + 7 * h->c_stride, 8);
119 memcpy(&h->top_border_v[h->mbx * 10 + 1], h->cv + 7 * h->c_stride, 8);
120 for (i = 0; i < 8; i++) {
121 h->left_border_y[i * 2 + 1] = *(h->cy + 15 + (i * 2 + 0) * h->l_stride);
122 h->left_border_y[i * 2 + 2] = *(h->cy + 15 + (i * 2 + 1) * h->l_stride);
123 h->left_border_u[i + 1] = *(h->cu + 7 + i * h->c_stride);
124 h->left_border_v[i + 1] = *(h->cv + 7 + i * h->c_stride);
125 }
126 if (!h->loop_filter_disable) {
127 /* determine bs */
128 if (mb_type == I_8X8)
129 memset(bs, 2, 8);
130 else {
131 memset(bs, 0, 8);
132 if (ff_cavs_partition_flags[mb_type] & SPLITV) {
133 bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
134 bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
135 }
136 if (ff_cavs_partition_flags[mb_type] & SPLITH) {
137 bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
138 bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
139 }
140 bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
141 bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
142 bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
143 bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
144 }
145 if (AV_RN64(bs)) {
146 if (h->flags & A_AVAIL) {
147 qp_avg = (h->qp + h->left_qp + 1) >> 1;
148 SET_PARAMS;
149 h->cdsp.cavs_filter_lv(h->cy, h->l_stride, alpha, beta, tc, bs[0], bs[1]);
150 qp_avg = (ff_cavs_chroma_qp[h->qp] + ff_cavs_chroma_qp[h->left_qp] + 1) >> 1;
151 SET_PARAMS;
152 h->cdsp.cavs_filter_cv(h->cu, h->c_stride, alpha, beta, tc, bs[0], bs[1]);
153 h->cdsp.cavs_filter_cv(h->cv, h->c_stride, alpha, beta, tc, bs[0], bs[1]);
154 }
155 qp_avg = h->qp;
156 SET_PARAMS;
157 h->cdsp.cavs_filter_lv(h->cy + 8, h->l_stride, alpha, beta, tc, bs[2], bs[3]);
158 h->cdsp.cavs_filter_lh(h->cy + 8 * h->l_stride, h->l_stride, alpha, beta, tc, bs[6], bs[7]);
159
160 if (h->flags & B_AVAIL) {
161 qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
162 SET_PARAMS;
163 h->cdsp.cavs_filter_lh(h->cy, h->l_stride, alpha, beta, tc, bs[4], bs[5]);
164 qp_avg = (ff_cavs_chroma_qp[h->qp] + ff_cavs_chroma_qp[h->top_qp[h->mbx]] + 1) >> 1;
165 SET_PARAMS;
166 h->cdsp.cavs_filter_ch(h->cu, h->c_stride, alpha, beta, tc, bs[4], bs[5]);
167 h->cdsp.cavs_filter_ch(h->cv, h->c_stride, alpha, beta, tc, bs[4], bs[5]);
168 }
169 }
170 }
171 h->left_qp = h->qp;
172 h->top_qp[h->mbx] = h->qp;
173 }
174
175 #undef SET_PARAMS
176
177 /*****************************************************************************
178 *
179 * spatial intra prediction
180 *
181 ****************************************************************************/
182
ff_cavs_load_intra_pred_luma(AVSContext * h,uint8_t * top,uint8_t ** left,int block)183 void ff_cavs_load_intra_pred_luma(AVSContext *h, uint8_t *top,
184 uint8_t **left, int block)
185 {
186 int i;
187
188 switch (block) {
189 case 0:
190 *left = h->left_border_y;
191 h->left_border_y[0] = h->left_border_y[1];
192 memset(&h->left_border_y[17], h->left_border_y[16], 9);
193 memcpy(&top[1], &h->top_border_y[h->mbx * 16], 16);
194 top[17] = top[16];
195 top[0] = top[1];
196 if ((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
197 h->left_border_y[0] = top[0] = h->topleft_border_y;
198 break;
199 case 1:
200 *left = h->intern_border_y;
201 for (i = 0; i < 8; i++)
202 h->intern_border_y[i + 1] = *(h->cy + 7 + i * h->l_stride);
203 memset(&h->intern_border_y[9], h->intern_border_y[8], 9);
204 h->intern_border_y[0] = h->intern_border_y[1];
205 memcpy(&top[1], &h->top_border_y[h->mbx * 16 + 8], 8);
206 if (h->flags & C_AVAIL)
207 memcpy(&top[9], &h->top_border_y[(h->mbx + 1) * 16], 8);
208 else
209 memset(&top[9], top[8], 9);
210 top[17] = top[16];
211 top[0] = top[1];
212 if (h->flags & B_AVAIL)
213 h->intern_border_y[0] = top[0] = h->top_border_y[h->mbx * 16 + 7];
214 break;
215 case 2:
216 *left = &h->left_border_y[8];
217 memcpy(&top[1], h->cy + 7 * h->l_stride, 16);
218 top[17] = top[16];
219 top[0] = top[1];
220 if (h->flags & A_AVAIL)
221 top[0] = h->left_border_y[8];
222 break;
223 case 3:
224 *left = &h->intern_border_y[8];
225 for (i = 0; i < 8; i++)
226 h->intern_border_y[i + 9] = *(h->cy + 7 + (i + 8) * h->l_stride);
227 memset(&h->intern_border_y[17], h->intern_border_y[16], 9);
228 memcpy(&top[0], h->cy + 7 + 7 * h->l_stride, 9);
229 memset(&top[9], top[8], 9);
230 break;
231 }
232 }
233
ff_cavs_load_intra_pred_chroma(AVSContext * h)234 void ff_cavs_load_intra_pred_chroma(AVSContext *h)
235 {
236 /* extend borders by one pixel */
237 h->left_border_u[9] = h->left_border_u[8];
238 h->left_border_v[9] = h->left_border_v[8];
239 if(h->flags & C_AVAIL) {
240 h->top_border_u[h->mbx*10 + 9] = h->top_border_u[h->mbx*10 + 11];
241 h->top_border_v[h->mbx*10 + 9] = h->top_border_v[h->mbx*10 + 11];
242 } else {
243 h->top_border_u[h->mbx * 10 + 9] = h->top_border_u[h->mbx * 10 + 8];
244 h->top_border_v[h->mbx * 10 + 9] = h->top_border_v[h->mbx * 10 + 8];
245 }
246 if((h->flags & A_AVAIL) && (h->flags & B_AVAIL)) {
247 h->top_border_u[h->mbx * 10] = h->left_border_u[0] = h->topleft_border_u;
248 h->top_border_v[h->mbx * 10] = h->left_border_v[0] = h->topleft_border_v;
249 } else {
250 h->left_border_u[0] = h->left_border_u[1];
251 h->left_border_v[0] = h->left_border_v[1];
252 h->top_border_u[h->mbx * 10] = h->top_border_u[h->mbx * 10 + 1];
253 h->top_border_v[h->mbx * 10] = h->top_border_v[h->mbx * 10 + 1];
254 }
255 }
256
intra_pred_vert(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)257 static void intra_pred_vert(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
258 {
259 int y;
260 uint64_t a = AV_RN64(&top[1]);
261 for (y = 0; y < 8; y++)
262 *((uint64_t *)(d + y * stride)) = a;
263 }
264
intra_pred_horiz(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)265 static void intra_pred_horiz(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
266 {
267 int y;
268 uint64_t a;
269 for (y = 0; y < 8; y++) {
270 a = left[y + 1] * 0x0101010101010101ULL;
271 *((uint64_t *)(d + y * stride)) = a;
272 }
273 }
274
intra_pred_dc_128(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)275 static void intra_pred_dc_128(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
276 {
277 int y;
278 uint64_t a = 0x8080808080808080ULL;
279 for (y = 0; y < 8; y++)
280 *((uint64_t *)(d + y * stride)) = a;
281 }
282
intra_pred_plane(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)283 static void intra_pred_plane(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
284 {
285 int x, y, ia;
286 int ih = 0;
287 int iv = 0;
288 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
289
290 for (x = 0; x < 4; x++) {
291 ih += (x + 1) * (top[5 + x] - top[3 - x]);
292 iv += (x + 1) * (left[5 + x] - left[3 - x]);
293 }
294 ia = (top[8] + left[8]) << 4;
295 ih = (17 * ih + 16) >> 5;
296 iv = (17 * iv + 16) >> 5;
297 for (y = 0; y < 8; y++)
298 for (x = 0; x < 8; x++)
299 d[y * stride + x] = cm[(ia + (x - 3) * ih + (y - 3) * iv + 16) >> 5];
300 }
301
302 #define LOWPASS(ARRAY, INDEX) \
303 ((ARRAY[(INDEX) - 1] + 2 * ARRAY[(INDEX)] + ARRAY[(INDEX) + 1] + 2) >> 2)
304
intra_pred_lp(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)305 static void intra_pred_lp(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
306 {
307 int x, y;
308 for (y = 0; y < 8; y++)
309 for (x = 0; x < 8; x++)
310 d[y * stride + x] = (LOWPASS(top, x + 1) + LOWPASS(left, y + 1)) >> 1;
311 }
312
intra_pred_down_left(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)313 static void intra_pred_down_left(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
314 {
315 int x, y;
316 for (y = 0; y < 8; y++)
317 for (x = 0; x < 8; x++)
318 d[y * stride + x] = (LOWPASS(top, x + y + 2) + LOWPASS(left, x + y + 2)) >> 1;
319 }
320
intra_pred_down_right(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)321 static void intra_pred_down_right(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
322 {
323 int x, y;
324 for (y = 0; y < 8; y++)
325 for (x = 0; x < 8; x++)
326 if (x == y)
327 d[y * stride + x] = (left[1] + 2 * top[0] + top[1] + 2) >> 2;
328 else if (x > y)
329 d[y * stride + x] = LOWPASS(top, x - y);
330 else
331 d[y * stride + x] = LOWPASS(left, y - x);
332 }
333
intra_pred_lp_left(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)334 static void intra_pred_lp_left(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
335 {
336 int x, y;
337 for (y = 0; y < 8; y++)
338 for (x = 0; x < 8; x++)
339 d[y * stride + x] = LOWPASS(left, y + 1);
340 }
341
intra_pred_lp_top(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)342 static void intra_pred_lp_top(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
343 {
344 int x, y;
345 for (y = 0; y < 8; y++)
346 for (x = 0; x < 8; x++)
347 d[y * stride + x] = LOWPASS(top, x + 1);
348 }
349
350 #undef LOWPASS
351
modify_pred(const int8_t * mod_table,int * mode)352 static inline void modify_pred(const int8_t *mod_table, int *mode)
353 {
354 *mode = mod_table[*mode];
355 if (*mode < 0) {
356 av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
357 *mode = 0;
358 }
359 }
360
ff_cavs_modify_mb_i(AVSContext * h,int * pred_mode_uv)361 void ff_cavs_modify_mb_i(AVSContext *h, int *pred_mode_uv)
362 {
363 /* save pred modes before they get modified */
364 h->pred_mode_Y[3] = h->pred_mode_Y[5];
365 h->pred_mode_Y[6] = h->pred_mode_Y[8];
366 h->top_pred_Y[h->mbx * 2 + 0] = h->pred_mode_Y[7];
367 h->top_pred_Y[h->mbx * 2 + 1] = h->pred_mode_Y[8];
368
369 /* modify pred modes according to availability of neighbour samples */
370 if (!(h->flags & A_AVAIL)) {
371 modify_pred(left_modifier_l, &h->pred_mode_Y[4]);
372 modify_pred(left_modifier_l, &h->pred_mode_Y[7]);
373 modify_pred(left_modifier_c, pred_mode_uv);
374 }
375 if (!(h->flags & B_AVAIL)) {
376 modify_pred(top_modifier_l, &h->pred_mode_Y[4]);
377 modify_pred(top_modifier_l, &h->pred_mode_Y[5]);
378 modify_pred(top_modifier_c, pred_mode_uv);
379 }
380 }
381
382 /*****************************************************************************
383 *
384 * motion compensation
385 *
386 ****************************************************************************/
387
mc_dir_part(AVSContext * h,AVFrame * pic,int chroma_height,int delta,int list,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int src_x_offset,int src_y_offset,qpel_mc_func * qpix_op,h264_chroma_mc_func chroma_op,cavs_vector * mv)388 static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height,
389 int delta, int list, uint8_t *dest_y,
390 uint8_t *dest_cb, uint8_t *dest_cr,
391 int src_x_offset, int src_y_offset,
392 qpel_mc_func *qpix_op,
393 h264_chroma_mc_func chroma_op, cavs_vector *mv)
394 {
395 const int mx = mv->x + src_x_offset * 8;
396 const int my = mv->y + src_y_offset * 8;
397 const int luma_xy = (mx & 3) + ((my & 3) << 2);
398 uint8_t *src_y = pic->data[0] + (mx >> 2) + (my >> 2) * h->l_stride;
399 uint8_t *src_cb = pic->data[1] + (mx >> 3) + (my >> 3) * h->c_stride;
400 uint8_t *src_cr = pic->data[2] + (mx >> 3) + (my >> 3) * h->c_stride;
401 int extra_width = 0;
402 int extra_height = extra_width;
403 const int full_mx = mx >> 2;
404 const int full_my = my >> 2;
405 const int pic_width = 16 * h->mb_width;
406 const int pic_height = 16 * h->mb_height;
407 int emu = 0;
408
409 if (!pic->data[0])
410 return;
411 if (mx & 7)
412 extra_width -= 3;
413 if (my & 7)
414 extra_height -= 3;
415
416 if (full_mx < 0 - extra_width ||
417 full_my < 0 - extra_height ||
418 full_mx + 16 /* FIXME */ > pic_width + extra_width ||
419 full_my + 16 /* FIXME */ > pic_height + extra_height) {
420 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
421 src_y - 2 - 2 * h->l_stride,
422 h->l_stride, h->l_stride,
423 16 + 5, 16 + 5 /* FIXME */,
424 full_mx - 2, full_my - 2,
425 pic_width, pic_height);
426 src_y = h->edge_emu_buffer + 2 + 2 * h->l_stride;
427 emu = 1;
428 }
429
430 // FIXME try variable height perhaps?
431 qpix_op[luma_xy](dest_y, src_y, h->l_stride);
432
433 if (emu) {
434 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb,
435 h->c_stride, h->c_stride,
436 9, 9 /* FIXME */,
437 mx >> 3, my >> 3,
438 pic_width >> 1, pic_height >> 1);
439 src_cb = h->edge_emu_buffer;
440 }
441 chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx & 7, my & 7);
442
443 if (emu) {
444 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr,
445 h->c_stride, h->c_stride,
446 9, 9 /* FIXME */,
447 mx >> 3, my >> 3,
448 pic_width >> 1, pic_height >> 1);
449 src_cr = h->edge_emu_buffer;
450 }
451 chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx & 7, my & 7);
452 }
453
mc_part_std(AVSContext * h,int chroma_height,int delta,uint8_t * dest_y,uint8_t * dest_cb,uint8_t * dest_cr,int x_offset,int y_offset,qpel_mc_func * qpix_put,h264_chroma_mc_func chroma_put,qpel_mc_func * qpix_avg,h264_chroma_mc_func chroma_avg,cavs_vector * mv)454 static inline void mc_part_std(AVSContext *h, int chroma_height, int delta,
455 uint8_t *dest_y,
456 uint8_t *dest_cb,
457 uint8_t *dest_cr,
458 int x_offset, int y_offset,
459 qpel_mc_func *qpix_put,
460 h264_chroma_mc_func chroma_put,
461 qpel_mc_func *qpix_avg,
462 h264_chroma_mc_func chroma_avg,
463 cavs_vector *mv)
464 {
465 qpel_mc_func *qpix_op = qpix_put;
466 h264_chroma_mc_func chroma_op = chroma_put;
467
468 dest_y += x_offset * 2 + y_offset * h->l_stride * 2;
469 dest_cb += x_offset + y_offset * h->c_stride;
470 dest_cr += x_offset + y_offset * h->c_stride;
471 x_offset += 8 * h->mbx;
472 y_offset += 8 * h->mby;
473
474 if (mv->ref >= 0) {
475 AVFrame *ref = h->DPB[mv->ref].f;
476 mc_dir_part(h, ref, chroma_height, delta, 0,
477 dest_y, dest_cb, dest_cr, x_offset, y_offset,
478 qpix_op, chroma_op, mv);
479
480 qpix_op = qpix_avg;
481 chroma_op = chroma_avg;
482 }
483
484 if ((mv + MV_BWD_OFFS)->ref >= 0) {
485 AVFrame *ref = h->DPB[0].f;
486 mc_dir_part(h, ref, chroma_height, delta, 1,
487 dest_y, dest_cb, dest_cr, x_offset, y_offset,
488 qpix_op, chroma_op, mv + MV_BWD_OFFS);
489 }
490 }
491
ff_cavs_inter(AVSContext * h,enum cavs_mb mb_type)492 void ff_cavs_inter(AVSContext *h, enum cavs_mb mb_type)
493 {
494 if (ff_cavs_partition_flags[mb_type] == 0) { // 16x16
495 mc_part_std(h, 8, 0, h->cy, h->cu, h->cv, 0, 0,
496 h->cdsp.put_cavs_qpel_pixels_tab[0],
497 h->h264chroma.put_h264_chroma_pixels_tab[0],
498 h->cdsp.avg_cavs_qpel_pixels_tab[0],
499 h->h264chroma.avg_h264_chroma_pixels_tab[0],
500 &h->mv[MV_FWD_X0]);
501 } else {
502 mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 0,
503 h->cdsp.put_cavs_qpel_pixels_tab[1],
504 h->h264chroma.put_h264_chroma_pixels_tab[1],
505 h->cdsp.avg_cavs_qpel_pixels_tab[1],
506 h->h264chroma.avg_h264_chroma_pixels_tab[1],
507 &h->mv[MV_FWD_X0]);
508 mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 0,
509 h->cdsp.put_cavs_qpel_pixels_tab[1],
510 h->h264chroma.put_h264_chroma_pixels_tab[1],
511 h->cdsp.avg_cavs_qpel_pixels_tab[1],
512 h->h264chroma.avg_h264_chroma_pixels_tab[1],
513 &h->mv[MV_FWD_X1]);
514 mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 4,
515 h->cdsp.put_cavs_qpel_pixels_tab[1],
516 h->h264chroma.put_h264_chroma_pixels_tab[1],
517 h->cdsp.avg_cavs_qpel_pixels_tab[1],
518 h->h264chroma.avg_h264_chroma_pixels_tab[1],
519 &h->mv[MV_FWD_X2]);
520 mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 4,
521 h->cdsp.put_cavs_qpel_pixels_tab[1],
522 h->h264chroma.put_h264_chroma_pixels_tab[1],
523 h->cdsp.avg_cavs_qpel_pixels_tab[1],
524 h->h264chroma.avg_h264_chroma_pixels_tab[1],
525 &h->mv[MV_FWD_X3]);
526 }
527 }
528
529 /*****************************************************************************
530 *
531 * motion vector prediction
532 *
533 ****************************************************************************/
534
scale_mv(AVSContext * h,int * d_x,int * d_y,cavs_vector * src,int distp)535 static inline void scale_mv(AVSContext *h, int *d_x, int *d_y,
536 cavs_vector *src, int distp)
537 {
538 int64_t den = h->scale_den[FFMAX(src->ref, 0)];
539 *d_x = (src->x * distp * den + 256 + FF_SIGNBIT(src->x)) >> 9;
540 *d_y = (src->y * distp * den + 256 + FF_SIGNBIT(src->y)) >> 9;
541 }
542
mv_pred_median(AVSContext * h,cavs_vector * mvP,cavs_vector * mvA,cavs_vector * mvB,cavs_vector * mvC)543 static inline void mv_pred_median(AVSContext *h,
544 cavs_vector *mvP,
545 cavs_vector *mvA,
546 cavs_vector *mvB,
547 cavs_vector *mvC)
548 {
549 int ax, ay, bx, by, cx, cy;
550 int len_ab, len_bc, len_ca, len_mid;
551
552 /* scale candidates according to their temporal span */
553 scale_mv(h, &ax, &ay, mvA, mvP->dist);
554 scale_mv(h, &bx, &by, mvB, mvP->dist);
555 scale_mv(h, &cx, &cy, mvC, mvP->dist);
556 /* find the geometrical median of the three candidates */
557 len_ab = abs(ax - bx) + abs(ay - by);
558 len_bc = abs(bx - cx) + abs(by - cy);
559 len_ca = abs(cx - ax) + abs(cy - ay);
560 len_mid = mid_pred(len_ab, len_bc, len_ca);
561 if (len_mid == len_ab) {
562 mvP->x = cx;
563 mvP->y = cy;
564 } else if (len_mid == len_bc) {
565 mvP->x = ax;
566 mvP->y = ay;
567 } else {
568 mvP->x = bx;
569 mvP->y = by;
570 }
571 }
572
ff_cavs_mv(AVSContext * h,enum cavs_mv_loc nP,enum cavs_mv_loc nC,enum cavs_mv_pred mode,enum cavs_block size,int ref)573 void ff_cavs_mv(AVSContext *h, enum cavs_mv_loc nP, enum cavs_mv_loc nC,
574 enum cavs_mv_pred mode, enum cavs_block size, int ref)
575 {
576 cavs_vector *mvP = &h->mv[nP];
577 cavs_vector *mvA = &h->mv[nP-1];
578 cavs_vector *mvB = &h->mv[nP-4];
579 cavs_vector *mvC = &h->mv[nC];
580 const cavs_vector *mvP2 = NULL;
581
582 mvP->ref = ref;
583 mvP->dist = h->dist[mvP->ref];
584 if (mvC->ref == NOT_AVAIL || (nP == MV_FWD_X3) || (nP == MV_BWD_X3 ))
585 mvC = &h->mv[nP - 5]; // set to top-left (mvD)
586 if (mode == MV_PRED_PSKIP &&
587 (mvA->ref == NOT_AVAIL ||
588 mvB->ref == NOT_AVAIL ||
589 (mvA->x | mvA->y | mvA->ref) == 0 ||
590 (mvB->x | mvB->y | mvB->ref) == 0)) {
591 mvP2 = &un_mv;
592 /* if there is only one suitable candidate, take it */
593 } else if (mvA->ref >= 0 && mvB->ref < 0 && mvC->ref < 0) {
594 mvP2 = mvA;
595 } else if (mvA->ref < 0 && mvB->ref >= 0 && mvC->ref < 0) {
596 mvP2 = mvB;
597 } else if (mvA->ref < 0 && mvB->ref < 0 && mvC->ref >= 0) {
598 mvP2 = mvC;
599 } else if (mode == MV_PRED_LEFT && mvA->ref == ref) {
600 mvP2 = mvA;
601 } else if (mode == MV_PRED_TOP && mvB->ref == ref) {
602 mvP2 = mvB;
603 } else if (mode == MV_PRED_TOPRIGHT && mvC->ref == ref) {
604 mvP2 = mvC;
605 }
606 if (mvP2) {
607 mvP->x = mvP2->x;
608 mvP->y = mvP2->y;
609 } else
610 mv_pred_median(h, mvP, mvA, mvB, mvC);
611
612 if (mode < MV_PRED_PSKIP) {
613 int mx = get_se_golomb(&h->gb) + (unsigned)mvP->x;
614 int my = get_se_golomb(&h->gb) + (unsigned)mvP->y;
615
616 if (mx != (int16_t)mx || my != (int16_t)my) {
617 av_log(h->avctx, AV_LOG_ERROR, "MV %d %d out of supported range\n", mx, my);
618 } else {
619 mvP->x = mx;
620 mvP->y = my;
621 }
622 }
623 set_mvs(mvP, size);
624 }
625
626 /*****************************************************************************
627 *
628 * macroblock level
629 *
630 ****************************************************************************/
631
632 /**
633 * initialise predictors for motion vectors and intra prediction
634 */
ff_cavs_init_mb(AVSContext * h)635 void ff_cavs_init_mb(AVSContext *h)
636 {
637 int i;
638
639 /* copy predictors from top line (MB B and C) into cache */
640 for (i = 0; i < 3; i++) {
641 h->mv[MV_FWD_B2 + i] = h->top_mv[0][h->mbx * 2 + i];
642 h->mv[MV_BWD_B2 + i] = h->top_mv[1][h->mbx * 2 + i];
643 }
644 h->pred_mode_Y[1] = h->top_pred_Y[h->mbx * 2 + 0];
645 h->pred_mode_Y[2] = h->top_pred_Y[h->mbx * 2 + 1];
646 /* clear top predictors if MB B is not available */
647 if (!(h->flags & B_AVAIL)) {
648 h->mv[MV_FWD_B2] = un_mv;
649 h->mv[MV_FWD_B3] = un_mv;
650 h->mv[MV_BWD_B2] = un_mv;
651 h->mv[MV_BWD_B3] = un_mv;
652 h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
653 h->flags &= ~(C_AVAIL | D_AVAIL);
654 } else if (h->mbx) {
655 h->flags |= D_AVAIL;
656 }
657 if (h->mbx == h->mb_width - 1) // MB C not available
658 h->flags &= ~C_AVAIL;
659 /* clear top-right predictors if MB C is not available */
660 if (!(h->flags & C_AVAIL)) {
661 h->mv[MV_FWD_C2] = un_mv;
662 h->mv[MV_BWD_C2] = un_mv;
663 }
664 /* clear top-left predictors if MB D is not available */
665 if (!(h->flags & D_AVAIL)) {
666 h->mv[MV_FWD_D3] = un_mv;
667 h->mv[MV_BWD_D3] = un_mv;
668 }
669 }
670
671 /**
672 * save predictors for later macroblocks and increase
673 * macroblock address
674 * @return 0 if end of frame is reached, 1 otherwise
675 */
ff_cavs_next_mb(AVSContext * h)676 int ff_cavs_next_mb(AVSContext *h)
677 {
678 int i;
679
680 h->flags |= A_AVAIL;
681 h->cy += 16;
682 h->cu += 8;
683 h->cv += 8;
684 /* copy mvs as predictors to the left */
685 for (i = 0; i <= 20; i += 4)
686 h->mv[i] = h->mv[i + 2];
687 /* copy bottom mvs from cache to top line */
688 h->top_mv[0][h->mbx * 2 + 0] = h->mv[MV_FWD_X2];
689 h->top_mv[0][h->mbx * 2 + 1] = h->mv[MV_FWD_X3];
690 h->top_mv[1][h->mbx * 2 + 0] = h->mv[MV_BWD_X2];
691 h->top_mv[1][h->mbx * 2 + 1] = h->mv[MV_BWD_X3];
692 /* next MB address */
693 h->mbidx++;
694 h->mbx++;
695 if (h->mbx == h->mb_width) { // New mb line
696 h->flags = B_AVAIL | C_AVAIL;
697 /* clear left pred_modes */
698 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
699 /* clear left mv predictors */
700 for (i = 0; i <= 20; i += 4)
701 h->mv[i] = un_mv;
702 h->mbx = 0;
703 h->mby++;
704 /* re-calculate sample pointers */
705 h->cy = h->cur.f->data[0] + h->mby * 16 * h->l_stride;
706 h->cu = h->cur.f->data[1] + h->mby * 8 * h->c_stride;
707 h->cv = h->cur.f->data[2] + h->mby * 8 * h->c_stride;
708 if (h->mby == h->mb_height) { // Frame end
709 return 0;
710 }
711 }
712 return 1;
713 }
714
715 /*****************************************************************************
716 *
717 * frame level
718 *
719 ****************************************************************************/
720
ff_cavs_init_pic(AVSContext * h)721 int ff_cavs_init_pic(AVSContext *h)
722 {
723 int i;
724
725 /* clear some predictors */
726 for (i = 0; i <= 20; i += 4)
727 h->mv[i] = un_mv;
728 h->mv[MV_BWD_X0] = ff_cavs_dir_mv;
729 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
730 h->mv[MV_FWD_X0] = ff_cavs_dir_mv;
731 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
732 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
733 h->cy = h->cur.f->data[0];
734 h->cu = h->cur.f->data[1];
735 h->cv = h->cur.f->data[2];
736 h->l_stride = h->cur.f->linesize[0];
737 h->c_stride = h->cur.f->linesize[1];
738 h->luma_scan[2] = 8 * h->l_stride;
739 h->luma_scan[3] = 8 * h->l_stride + 8;
740 h->mbx = h->mby = h->mbidx = 0;
741 h->flags = 0;
742
743 return 0;
744 }
745
746 /*****************************************************************************
747 *
748 * headers and interface
749 *
750 ****************************************************************************/
751
752 /**
753 * some predictions require data from the top-neighbouring macroblock.
754 * this data has to be stored for one complete row of macroblocks
755 * and this storage space is allocated here
756 */
ff_cavs_init_top_lines(AVSContext * h)757 int ff_cavs_init_top_lines(AVSContext *h)
758 {
759 /* alloc top line of predictors */
760 h->top_qp = av_mallocz(h->mb_width);
761 h->top_mv[0] = av_calloc(h->mb_width * 2 + 1, sizeof(cavs_vector));
762 h->top_mv[1] = av_calloc(h->mb_width * 2 + 1, sizeof(cavs_vector));
763 h->top_pred_Y = av_calloc(h->mb_width * 2, sizeof(*h->top_pred_Y));
764 h->top_border_y = av_calloc(h->mb_width + 1, 16);
765 h->top_border_u = av_calloc(h->mb_width, 10);
766 h->top_border_v = av_calloc(h->mb_width, 10);
767
768 /* alloc space for co-located MVs and types */
769 h->col_mv = av_calloc(h->mb_width * h->mb_height,
770 4 * sizeof(*h->col_mv));
771 h->col_type_base = av_mallocz(h->mb_width * h->mb_height);
772 h->block = av_mallocz(64 * sizeof(int16_t));
773
774 if (!h->top_qp || !h->top_mv[0] || !h->top_mv[1] || !h->top_pred_Y ||
775 !h->top_border_y || !h->top_border_u || !h->top_border_v ||
776 !h->col_mv || !h->col_type_base || !h->block) {
777 av_freep(&h->top_qp);
778 av_freep(&h->top_mv[0]);
779 av_freep(&h->top_mv[1]);
780 av_freep(&h->top_pred_Y);
781 av_freep(&h->top_border_y);
782 av_freep(&h->top_border_u);
783 av_freep(&h->top_border_v);
784 av_freep(&h->col_mv);
785 av_freep(&h->col_type_base);
786 av_freep(&h->block);
787 return AVERROR(ENOMEM);
788 }
789 return 0;
790 }
791
ff_cavs_init(AVCodecContext * avctx)792 av_cold int ff_cavs_init(AVCodecContext *avctx)
793 {
794 AVSContext *h = avctx->priv_data;
795
796 ff_blockdsp_init(&h->bdsp, avctx);
797 ff_h264chroma_init(&h->h264chroma, 8);
798 ff_idctdsp_init(&h->idsp, avctx);
799 ff_videodsp_init(&h->vdsp, 8);
800 ff_cavsdsp_init(&h->cdsp, avctx);
801 ff_init_scantable_permutation(h->idsp.idct_permutation,
802 h->cdsp.idct_perm);
803 ff_init_scantable(h->idsp.idct_permutation, &h->scantable, ff_zigzag_direct);
804
805 h->avctx = avctx;
806 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
807
808 h->cur.f = av_frame_alloc();
809 h->DPB[0].f = av_frame_alloc();
810 h->DPB[1].f = av_frame_alloc();
811 if (!h->cur.f || !h->DPB[0].f || !h->DPB[1].f)
812 return AVERROR(ENOMEM);
813
814 h->luma_scan[0] = 0;
815 h->luma_scan[1] = 8;
816 h->intra_pred_l[INTRA_L_VERT] = intra_pred_vert;
817 h->intra_pred_l[INTRA_L_HORIZ] = intra_pred_horiz;
818 h->intra_pred_l[INTRA_L_LP] = intra_pred_lp;
819 h->intra_pred_l[INTRA_L_DOWN_LEFT] = intra_pred_down_left;
820 h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
821 h->intra_pred_l[INTRA_L_LP_LEFT] = intra_pred_lp_left;
822 h->intra_pred_l[INTRA_L_LP_TOP] = intra_pred_lp_top;
823 h->intra_pred_l[INTRA_L_DC_128] = intra_pred_dc_128;
824 h->intra_pred_c[INTRA_C_LP] = intra_pred_lp;
825 h->intra_pred_c[INTRA_C_HORIZ] = intra_pred_horiz;
826 h->intra_pred_c[INTRA_C_VERT] = intra_pred_vert;
827 h->intra_pred_c[INTRA_C_PLANE] = intra_pred_plane;
828 h->intra_pred_c[INTRA_C_LP_LEFT] = intra_pred_lp_left;
829 h->intra_pred_c[INTRA_C_LP_TOP] = intra_pred_lp_top;
830 h->intra_pred_c[INTRA_C_DC_128] = intra_pred_dc_128;
831 h->mv[7] = un_mv;
832 h->mv[19] = un_mv;
833 return 0;
834 }
835
ff_cavs_end(AVCodecContext * avctx)836 av_cold int ff_cavs_end(AVCodecContext *avctx)
837 {
838 AVSContext *h = avctx->priv_data;
839
840 av_frame_free(&h->cur.f);
841 av_frame_free(&h->DPB[0].f);
842 av_frame_free(&h->DPB[1].f);
843
844 av_freep(&h->top_qp);
845 av_freep(&h->top_mv[0]);
846 av_freep(&h->top_mv[1]);
847 av_freep(&h->top_pred_Y);
848 av_freep(&h->top_border_y);
849 av_freep(&h->top_border_u);
850 av_freep(&h->top_border_v);
851 av_freep(&h->col_mv);
852 av_freep(&h->col_type_base);
853 av_freep(&h->block);
854 av_freep(&h->edge_emu_buffer);
855 return 0;
856 }
857