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 "libavutil/mem_internal.h"
29
30 #include "avcodec.h"
31 #include "get_bits.h"
32 #include "golomb.h"
33 #include "h264chroma.h"
34 #include "idctdsp.h"
35 #include "internal.h"
36 #include "mathops.h"
37 #include "qpeldsp.h"
38 #include "cavs.h"
39
40 static const uint8_t alpha_tab[64] = {
41 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3,
42 4, 4, 5, 5, 6, 7, 8, 9, 10, 11, 12, 13, 15, 16, 18, 20,
43 22, 24, 26, 28, 30, 33, 33, 35, 35, 36, 37, 37, 39, 39, 42, 44,
44 46, 48, 50, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64
45 };
46
47 static const uint8_t beta_tab[64] = {
48 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2,
49 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 6, 6,
50 6, 7, 7, 7, 8, 8, 8, 9, 9, 10, 10, 11, 11, 12, 13, 14,
51 15, 16, 17, 18, 19, 20, 21, 22, 23, 23, 24, 24, 25, 25, 26, 27
52 };
53
54 static const uint8_t tc_tab[64] = {
55 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
56 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
57 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4,
58 5, 5, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9, 9
59 };
60
61 /** mark block as unavailable, i.e. out of picture
62 * or not yet decoded */
63 static const cavs_vector un_mv = { 0, 0, 1, NOT_AVAIL };
64
65 static const int8_t left_modifier_l[8] = { 0, -1, 6, -1, -1, 7, 6, 7 };
66 static const int8_t top_modifier_l[8] = { -1, 1, 5, -1, -1, 5, 7, 7 };
67 static const int8_t left_modifier_c[7] = { 5, -1, 2, -1, 6, 5, 6 };
68 static const int8_t top_modifier_c[7] = { 4, 1, -1, -1, 4, 6, 6 };
69
70 /*****************************************************************************
71 *
72 * in-loop deblocking filter
73 *
74 ****************************************************************************/
75
get_bs(cavs_vector * mvP,cavs_vector * mvQ,int b)76 static inline int get_bs(cavs_vector *mvP, cavs_vector *mvQ, int b)
77 {
78 if ((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
79 return 2;
80 if((abs(mvP->x - mvQ->x) >= 4) ||
81 (abs(mvP->y - mvQ->y) >= 4) ||
82 (mvP->ref != mvQ->ref))
83 return 1;
84 if (b) {
85 mvP += MV_BWD_OFFS;
86 mvQ += MV_BWD_OFFS;
87 if((abs(mvP->x - mvQ->x) >= 4) ||
88 (abs(mvP->y - mvQ->y) >= 4) ||
89 (mvP->ref != mvQ->ref))
90 return 1;
91 }
92 return 0;
93 }
94
95 #define SET_PARAMS \
96 alpha = alpha_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)]; \
97 beta = beta_tab[av_clip_uintp2(qp_avg + h->beta_offset, 6)]; \
98 tc = tc_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)];
99
100 /**
101 * in-loop deblocking filter for a single macroblock
102 *
103 * boundary strength (bs) mapping:
104 *
105 * --4---5--
106 * 0 2 |
107 * | 6 | 7 |
108 * 1 3 |
109 * ---------
110 */
ff_cavs_filter(AVSContext * h,enum cavs_mb mb_type)111 void ff_cavs_filter(AVSContext *h, enum cavs_mb mb_type)
112 {
113 uint8_t bs[8];
114 int qp_avg, alpha, beta, tc;
115 int i;
116
117 /* save un-deblocked lines */
118 h->topleft_border_y = h->top_border_y[h->mbx * 16 + 15];
119 h->topleft_border_u = h->top_border_u[h->mbx * 10 + 8];
120 h->topleft_border_v = h->top_border_v[h->mbx * 10 + 8];
121 memcpy(&h->top_border_y[h->mbx * 16], h->cy + 15 * h->l_stride, 16);
122 memcpy(&h->top_border_u[h->mbx * 10 + 1], h->cu + 7 * h->c_stride, 8);
123 memcpy(&h->top_border_v[h->mbx * 10 + 1], h->cv + 7 * h->c_stride, 8);
124 for (i = 0; i < 8; i++) {
125 h->left_border_y[i * 2 + 1] = *(h->cy + 15 + (i * 2 + 0) * h->l_stride);
126 h->left_border_y[i * 2 + 2] = *(h->cy + 15 + (i * 2 + 1) * h->l_stride);
127 h->left_border_u[i + 1] = *(h->cu + 7 + i * h->c_stride);
128 h->left_border_v[i + 1] = *(h->cv + 7 + i * h->c_stride);
129 }
130 if (!h->loop_filter_disable) {
131 /* determine bs */
132 if (mb_type == I_8X8)
133 memset(bs, 2, 8);
134 else {
135 memset(bs, 0, 8);
136 if (ff_cavs_partition_flags[mb_type] & SPLITV) {
137 bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
138 bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
139 }
140 if (ff_cavs_partition_flags[mb_type] & SPLITH) {
141 bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
142 bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
143 }
144 bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
145 bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
146 bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
147 bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
148 }
149 if (AV_RN64(bs)) {
150 if (h->flags & A_AVAIL) {
151 qp_avg = (h->qp + h->left_qp + 1) >> 1;
152 SET_PARAMS;
153 h->cdsp.cavs_filter_lv(h->cy, h->l_stride, alpha, beta, tc, bs[0], bs[1]);
154 qp_avg = (ff_cavs_chroma_qp[h->qp] + ff_cavs_chroma_qp[h->left_qp] + 1) >> 1;
155 SET_PARAMS;
156 h->cdsp.cavs_filter_cv(h->cu, h->c_stride, alpha, beta, tc, bs[0], bs[1]);
157 h->cdsp.cavs_filter_cv(h->cv, h->c_stride, alpha, beta, tc, bs[0], bs[1]);
158 }
159 qp_avg = h->qp;
160 SET_PARAMS;
161 h->cdsp.cavs_filter_lv(h->cy + 8, h->l_stride, alpha, beta, tc, bs[2], bs[3]);
162 h->cdsp.cavs_filter_lh(h->cy + 8 * h->l_stride, h->l_stride, alpha, beta, tc, bs[6], bs[7]);
163
164 if (h->flags & B_AVAIL) {
165 qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
166 SET_PARAMS;
167 h->cdsp.cavs_filter_lh(h->cy, h->l_stride, alpha, beta, tc, bs[4], bs[5]);
168 qp_avg = (ff_cavs_chroma_qp[h->qp] + ff_cavs_chroma_qp[h->top_qp[h->mbx]] + 1) >> 1;
169 SET_PARAMS;
170 h->cdsp.cavs_filter_ch(h->cu, h->c_stride, alpha, beta, tc, bs[4], bs[5]);
171 h->cdsp.cavs_filter_ch(h->cv, h->c_stride, alpha, beta, tc, bs[4], bs[5]);
172 }
173 }
174 }
175 h->left_qp = h->qp;
176 h->top_qp[h->mbx] = h->qp;
177 }
178
179 #undef SET_PARAMS
180
181 /*****************************************************************************
182 *
183 * spatial intra prediction
184 *
185 ****************************************************************************/
186
ff_cavs_load_intra_pred_luma(AVSContext * h,uint8_t * top,uint8_t ** left,int block)187 void ff_cavs_load_intra_pred_luma(AVSContext *h, uint8_t *top,
188 uint8_t **left, int block)
189 {
190 int i;
191
192 switch (block) {
193 case 0:
194 *left = h->left_border_y;
195 h->left_border_y[0] = h->left_border_y[1];
196 memset(&h->left_border_y[17], h->left_border_y[16], 9);
197 memcpy(&top[1], &h->top_border_y[h->mbx * 16], 16);
198 top[17] = top[16];
199 top[0] = top[1];
200 if ((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
201 h->left_border_y[0] = top[0] = h->topleft_border_y;
202 break;
203 case 1:
204 *left = h->intern_border_y;
205 for (i = 0; i < 8; i++)
206 h->intern_border_y[i + 1] = *(h->cy + 7 + i * h->l_stride);
207 memset(&h->intern_border_y[9], h->intern_border_y[8], 9);
208 h->intern_border_y[0] = h->intern_border_y[1];
209 memcpy(&top[1], &h->top_border_y[h->mbx * 16 + 8], 8);
210 if (h->flags & C_AVAIL)
211 memcpy(&top[9], &h->top_border_y[(h->mbx + 1) * 16], 8);
212 else
213 memset(&top[9], top[8], 9);
214 top[17] = top[16];
215 top[0] = top[1];
216 if (h->flags & B_AVAIL)
217 h->intern_border_y[0] = top[0] = h->top_border_y[h->mbx * 16 + 7];
218 break;
219 case 2:
220 *left = &h->left_border_y[8];
221 memcpy(&top[1], h->cy + 7 * h->l_stride, 16);
222 top[17] = top[16];
223 top[0] = top[1];
224 if (h->flags & A_AVAIL)
225 top[0] = h->left_border_y[8];
226 break;
227 case 3:
228 *left = &h->intern_border_y[8];
229 for (i = 0; i < 8; i++)
230 h->intern_border_y[i + 9] = *(h->cy + 7 + (i + 8) * h->l_stride);
231 memset(&h->intern_border_y[17], h->intern_border_y[16], 9);
232 memcpy(&top[0], h->cy + 7 + 7 * h->l_stride, 9);
233 memset(&top[9], top[8], 9);
234 break;
235 }
236 }
237
ff_cavs_load_intra_pred_chroma(AVSContext * h)238 void ff_cavs_load_intra_pred_chroma(AVSContext *h)
239 {
240 /* extend borders by one pixel */
241 h->left_border_u[9] = h->left_border_u[8];
242 h->left_border_v[9] = h->left_border_v[8];
243 if(h->flags & C_AVAIL) {
244 h->top_border_u[h->mbx*10 + 9] = h->top_border_u[h->mbx*10 + 11];
245 h->top_border_v[h->mbx*10 + 9] = h->top_border_v[h->mbx*10 + 11];
246 } else {
247 h->top_border_u[h->mbx * 10 + 9] = h->top_border_u[h->mbx * 10 + 8];
248 h->top_border_v[h->mbx * 10 + 9] = h->top_border_v[h->mbx * 10 + 8];
249 }
250 if((h->flags & A_AVAIL) && (h->flags & B_AVAIL)) {
251 h->top_border_u[h->mbx * 10] = h->left_border_u[0] = h->topleft_border_u;
252 h->top_border_v[h->mbx * 10] = h->left_border_v[0] = h->topleft_border_v;
253 } else {
254 h->left_border_u[0] = h->left_border_u[1];
255 h->left_border_v[0] = h->left_border_v[1];
256 h->top_border_u[h->mbx * 10] = h->top_border_u[h->mbx * 10 + 1];
257 h->top_border_v[h->mbx * 10] = h->top_border_v[h->mbx * 10 + 1];
258 }
259 }
260
intra_pred_vert(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)261 static void intra_pred_vert(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
262 {
263 int y;
264 uint64_t a = AV_RN64(&top[1]);
265 for (y = 0; y < 8; y++)
266 *((uint64_t *)(d + y * stride)) = a;
267 }
268
intra_pred_horiz(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)269 static void intra_pred_horiz(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
270 {
271 int y;
272 uint64_t a;
273 for (y = 0; y < 8; y++) {
274 a = left[y + 1] * 0x0101010101010101ULL;
275 *((uint64_t *)(d + y * stride)) = a;
276 }
277 }
278
intra_pred_dc_128(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)279 static void intra_pred_dc_128(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
280 {
281 int y;
282 uint64_t a = 0x8080808080808080ULL;
283 for (y = 0; y < 8; y++)
284 *((uint64_t *)(d + y * stride)) = a;
285 }
286
intra_pred_plane(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)287 static void intra_pred_plane(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
288 {
289 int x, y, ia;
290 int ih = 0;
291 int iv = 0;
292 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
293
294 for (x = 0; x < 4; x++) {
295 ih += (x + 1) * (top[5 + x] - top[3 - x]);
296 iv += (x + 1) * (left[5 + x] - left[3 - x]);
297 }
298 ia = (top[8] + left[8]) << 4;
299 ih = (17 * ih + 16) >> 5;
300 iv = (17 * iv + 16) >> 5;
301 for (y = 0; y < 8; y++)
302 for (x = 0; x < 8; x++)
303 d[y * stride + x] = cm[(ia + (x - 3) * ih + (y - 3) * iv + 16) >> 5];
304 }
305
306 #define LOWPASS(ARRAY, INDEX) \
307 ((ARRAY[(INDEX) - 1] + 2 * ARRAY[(INDEX)] + ARRAY[(INDEX) + 1] + 2) >> 2)
308
intra_pred_lp(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)309 static void intra_pred_lp(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
310 {
311 int x, y;
312 for (y = 0; y < 8; y++)
313 for (x = 0; x < 8; x++)
314 d[y * stride + x] = (LOWPASS(top, x + 1) + LOWPASS(left, y + 1)) >> 1;
315 }
316
intra_pred_down_left(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)317 static void intra_pred_down_left(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
318 {
319 int x, y;
320 for (y = 0; y < 8; y++)
321 for (x = 0; x < 8; x++)
322 d[y * stride + x] = (LOWPASS(top, x + y + 2) + LOWPASS(left, x + y + 2)) >> 1;
323 }
324
intra_pred_down_right(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)325 static void intra_pred_down_right(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
326 {
327 int x, y;
328 for (y = 0; y < 8; y++)
329 for (x = 0; x < 8; x++)
330 if (x == y)
331 d[y * stride + x] = (left[1] + 2 * top[0] + top[1] + 2) >> 2;
332 else if (x > y)
333 d[y * stride + x] = LOWPASS(top, x - y);
334 else
335 d[y * stride + x] = LOWPASS(left, y - x);
336 }
337
intra_pred_lp_left(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)338 static void intra_pred_lp_left(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
339 {
340 int x, y;
341 for (y = 0; y < 8; y++)
342 for (x = 0; x < 8; x++)
343 d[y * stride + x] = LOWPASS(left, y + 1);
344 }
345
intra_pred_lp_top(uint8_t * d,uint8_t * top,uint8_t * left,ptrdiff_t stride)346 static void intra_pred_lp_top(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
347 {
348 int x, y;
349 for (y = 0; y < 8; y++)
350 for (x = 0; x < 8; x++)
351 d[y * stride + x] = LOWPASS(top, x + 1);
352 }
353
354 #undef LOWPASS
355
modify_pred(const int8_t * mod_table,int * mode)356 static inline void modify_pred(const int8_t *mod_table, int *mode)
357 {
358 *mode = mod_table[*mode];
359 if (*mode < 0) {
360 av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
361 *mode = 0;
362 }
363 }
364
ff_cavs_modify_mb_i(AVSContext * h,int * pred_mode_uv)365 void ff_cavs_modify_mb_i(AVSContext *h, int *pred_mode_uv)
366 {
367 /* save pred modes before they get modified */
368 h->pred_mode_Y[3] = h->pred_mode_Y[5];
369 h->pred_mode_Y[6] = h->pred_mode_Y[8];
370 h->top_pred_Y[h->mbx * 2 + 0] = h->pred_mode_Y[7];
371 h->top_pred_Y[h->mbx * 2 + 1] = h->pred_mode_Y[8];
372
373 /* modify pred modes according to availability of neighbour samples */
374 if (!(h->flags & A_AVAIL)) {
375 modify_pred(left_modifier_l, &h->pred_mode_Y[4]);
376 modify_pred(left_modifier_l, &h->pred_mode_Y[7]);
377 modify_pred(left_modifier_c, pred_mode_uv);
378 }
379 if (!(h->flags & B_AVAIL)) {
380 modify_pred(top_modifier_l, &h->pred_mode_Y[4]);
381 modify_pred(top_modifier_l, &h->pred_mode_Y[5]);
382 modify_pred(top_modifier_c, pred_mode_uv);
383 }
384 }
385
386 /*****************************************************************************
387 *
388 * motion compensation
389 *
390 ****************************************************************************/
391
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)392 static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height,
393 int delta, int list, uint8_t *dest_y,
394 uint8_t *dest_cb, uint8_t *dest_cr,
395 int src_x_offset, int src_y_offset,
396 qpel_mc_func *qpix_op,
397 h264_chroma_mc_func chroma_op, cavs_vector *mv)
398 {
399 const int mx = mv->x + src_x_offset * 8;
400 const int my = mv->y + src_y_offset * 8;
401 const int luma_xy = (mx & 3) + ((my & 3) << 2);
402 uint8_t *src_y = pic->data[0] + (mx >> 2) + (my >> 2) * h->l_stride;
403 uint8_t *src_cb = pic->data[1] + (mx >> 3) + (my >> 3) * h->c_stride;
404 uint8_t *src_cr = pic->data[2] + (mx >> 3) + (my >> 3) * h->c_stride;
405 int extra_width = 0;
406 int extra_height = extra_width;
407 const int full_mx = mx >> 2;
408 const int full_my = my >> 2;
409 const int pic_width = 16 * h->mb_width;
410 const int pic_height = 16 * h->mb_height;
411 int emu = 0;
412
413 if (!pic->data[0])
414 return;
415 if (mx & 7)
416 extra_width -= 3;
417 if (my & 7)
418 extra_height -= 3;
419
420 if (full_mx < 0 - extra_width ||
421 full_my < 0 - extra_height ||
422 full_mx + 16 /* FIXME */ > pic_width + extra_width ||
423 full_my + 16 /* FIXME */ > pic_height + extra_height) {
424 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
425 src_y - 2 - 2 * h->l_stride,
426 h->l_stride, h->l_stride,
427 16 + 5, 16 + 5 /* FIXME */,
428 full_mx - 2, full_my - 2,
429 pic_width, pic_height);
430 src_y = h->edge_emu_buffer + 2 + 2 * h->l_stride;
431 emu = 1;
432 }
433
434 // FIXME try variable height perhaps?
435 qpix_op[luma_xy](dest_y, src_y, h->l_stride);
436
437 if (emu) {
438 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb,
439 h->c_stride, h->c_stride,
440 9, 9 /* FIXME */,
441 mx >> 3, my >> 3,
442 pic_width >> 1, pic_height >> 1);
443 src_cb = h->edge_emu_buffer;
444 }
445 chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx & 7, my & 7);
446
447 if (emu) {
448 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr,
449 h->c_stride, h->c_stride,
450 9, 9 /* FIXME */,
451 mx >> 3, my >> 3,
452 pic_width >> 1, pic_height >> 1);
453 src_cr = h->edge_emu_buffer;
454 }
455 chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx & 7, my & 7);
456 }
457
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)458 static inline void mc_part_std(AVSContext *h, int chroma_height, int delta,
459 uint8_t *dest_y,
460 uint8_t *dest_cb,
461 uint8_t *dest_cr,
462 int x_offset, int y_offset,
463 qpel_mc_func *qpix_put,
464 h264_chroma_mc_func chroma_put,
465 qpel_mc_func *qpix_avg,
466 h264_chroma_mc_func chroma_avg,
467 cavs_vector *mv)
468 {
469 qpel_mc_func *qpix_op = qpix_put;
470 h264_chroma_mc_func chroma_op = chroma_put;
471
472 dest_y += x_offset * 2 + y_offset * h->l_stride * 2;
473 dest_cb += x_offset + y_offset * h->c_stride;
474 dest_cr += x_offset + y_offset * h->c_stride;
475 x_offset += 8 * h->mbx;
476 y_offset += 8 * h->mby;
477
478 if (mv->ref >= 0) {
479 AVFrame *ref = h->DPB[mv->ref].f;
480 mc_dir_part(h, ref, chroma_height, delta, 0,
481 dest_y, dest_cb, dest_cr, x_offset, y_offset,
482 qpix_op, chroma_op, mv);
483
484 qpix_op = qpix_avg;
485 chroma_op = chroma_avg;
486 }
487
488 if ((mv + MV_BWD_OFFS)->ref >= 0) {
489 AVFrame *ref = h->DPB[0].f;
490 mc_dir_part(h, ref, chroma_height, delta, 1,
491 dest_y, dest_cb, dest_cr, x_offset, y_offset,
492 qpix_op, chroma_op, mv + MV_BWD_OFFS);
493 }
494 }
495
ff_cavs_inter(AVSContext * h,enum cavs_mb mb_type)496 void ff_cavs_inter(AVSContext *h, enum cavs_mb mb_type)
497 {
498 if (ff_cavs_partition_flags[mb_type] == 0) { // 16x16
499 mc_part_std(h, 8, 0, h->cy, h->cu, h->cv, 0, 0,
500 h->cdsp.put_cavs_qpel_pixels_tab[0],
501 h->h264chroma.put_h264_chroma_pixels_tab[0],
502 h->cdsp.avg_cavs_qpel_pixels_tab[0],
503 h->h264chroma.avg_h264_chroma_pixels_tab[0],
504 &h->mv[MV_FWD_X0]);
505 } else {
506 mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 0,
507 h->cdsp.put_cavs_qpel_pixels_tab[1],
508 h->h264chroma.put_h264_chroma_pixels_tab[1],
509 h->cdsp.avg_cavs_qpel_pixels_tab[1],
510 h->h264chroma.avg_h264_chroma_pixels_tab[1],
511 &h->mv[MV_FWD_X0]);
512 mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 0,
513 h->cdsp.put_cavs_qpel_pixels_tab[1],
514 h->h264chroma.put_h264_chroma_pixels_tab[1],
515 h->cdsp.avg_cavs_qpel_pixels_tab[1],
516 h->h264chroma.avg_h264_chroma_pixels_tab[1],
517 &h->mv[MV_FWD_X1]);
518 mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 4,
519 h->cdsp.put_cavs_qpel_pixels_tab[1],
520 h->h264chroma.put_h264_chroma_pixels_tab[1],
521 h->cdsp.avg_cavs_qpel_pixels_tab[1],
522 h->h264chroma.avg_h264_chroma_pixels_tab[1],
523 &h->mv[MV_FWD_X2]);
524 mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 4,
525 h->cdsp.put_cavs_qpel_pixels_tab[1],
526 h->h264chroma.put_h264_chroma_pixels_tab[1],
527 h->cdsp.avg_cavs_qpel_pixels_tab[1],
528 h->h264chroma.avg_h264_chroma_pixels_tab[1],
529 &h->mv[MV_FWD_X3]);
530 }
531 }
532
533 /*****************************************************************************
534 *
535 * motion vector prediction
536 *
537 ****************************************************************************/
538
scale_mv(AVSContext * h,int * d_x,int * d_y,cavs_vector * src,int distp)539 static inline void scale_mv(AVSContext *h, int *d_x, int *d_y,
540 cavs_vector *src, int distp)
541 {
542 int64_t den = h->scale_den[FFMAX(src->ref, 0)];
543 *d_x = (src->x * distp * den + 256 + FF_SIGNBIT(src->x)) >> 9;
544 *d_y = (src->y * distp * den + 256 + FF_SIGNBIT(src->y)) >> 9;
545 }
546
mv_pred_median(AVSContext * h,cavs_vector * mvP,cavs_vector * mvA,cavs_vector * mvB,cavs_vector * mvC)547 static inline void mv_pred_median(AVSContext *h,
548 cavs_vector *mvP,
549 cavs_vector *mvA,
550 cavs_vector *mvB,
551 cavs_vector *mvC)
552 {
553 int ax, ay, bx, by, cx, cy;
554 int len_ab, len_bc, len_ca, len_mid;
555
556 /* scale candidates according to their temporal span */
557 scale_mv(h, &ax, &ay, mvA, mvP->dist);
558 scale_mv(h, &bx, &by, mvB, mvP->dist);
559 scale_mv(h, &cx, &cy, mvC, mvP->dist);
560 /* find the geometrical median of the three candidates */
561 len_ab = abs(ax - bx) + abs(ay - by);
562 len_bc = abs(bx - cx) + abs(by - cy);
563 len_ca = abs(cx - ax) + abs(cy - ay);
564 len_mid = mid_pred(len_ab, len_bc, len_ca);
565 if (len_mid == len_ab) {
566 mvP->x = cx;
567 mvP->y = cy;
568 } else if (len_mid == len_bc) {
569 mvP->x = ax;
570 mvP->y = ay;
571 } else {
572 mvP->x = bx;
573 mvP->y = by;
574 }
575 }
576
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)577 void ff_cavs_mv(AVSContext *h, enum cavs_mv_loc nP, enum cavs_mv_loc nC,
578 enum cavs_mv_pred mode, enum cavs_block size, int ref)
579 {
580 cavs_vector *mvP = &h->mv[nP];
581 cavs_vector *mvA = &h->mv[nP-1];
582 cavs_vector *mvB = &h->mv[nP-4];
583 cavs_vector *mvC = &h->mv[nC];
584 const cavs_vector *mvP2 = NULL;
585
586 mvP->ref = ref;
587 mvP->dist = h->dist[mvP->ref];
588 if (mvC->ref == NOT_AVAIL || (nP == MV_FWD_X3) || (nP == MV_BWD_X3 ))
589 mvC = &h->mv[nP - 5]; // set to top-left (mvD)
590 if (mode == MV_PRED_PSKIP &&
591 (mvA->ref == NOT_AVAIL ||
592 mvB->ref == NOT_AVAIL ||
593 (mvA->x | mvA->y | mvA->ref) == 0 ||
594 (mvB->x | mvB->y | mvB->ref) == 0)) {
595 mvP2 = &un_mv;
596 /* if there is only one suitable candidate, take it */
597 } else if (mvA->ref >= 0 && mvB->ref < 0 && mvC->ref < 0) {
598 mvP2 = mvA;
599 } else if (mvA->ref < 0 && mvB->ref >= 0 && mvC->ref < 0) {
600 mvP2 = mvB;
601 } else if (mvA->ref < 0 && mvB->ref < 0 && mvC->ref >= 0) {
602 mvP2 = mvC;
603 } else if (mode == MV_PRED_LEFT && mvA->ref == ref) {
604 mvP2 = mvA;
605 } else if (mode == MV_PRED_TOP && mvB->ref == ref) {
606 mvP2 = mvB;
607 } else if (mode == MV_PRED_TOPRIGHT && mvC->ref == ref) {
608 mvP2 = mvC;
609 }
610 if (mvP2) {
611 mvP->x = mvP2->x;
612 mvP->y = mvP2->y;
613 } else
614 mv_pred_median(h, mvP, mvA, mvB, mvC);
615
616 if (mode < MV_PRED_PSKIP) {
617 int mx = get_se_golomb(&h->gb) + (unsigned)mvP->x;
618 int my = get_se_golomb(&h->gb) + (unsigned)mvP->y;
619
620 if (mx != (int16_t)mx || my != (int16_t)my) {
621 av_log(h->avctx, AV_LOG_ERROR, "MV %d %d out of supported range\n", mx, my);
622 } else {
623 mvP->x = mx;
624 mvP->y = my;
625 }
626 }
627 set_mvs(mvP, size);
628 }
629
630 /*****************************************************************************
631 *
632 * macroblock level
633 *
634 ****************************************************************************/
635
636 /**
637 * initialise predictors for motion vectors and intra prediction
638 */
ff_cavs_init_mb(AVSContext * h)639 void ff_cavs_init_mb(AVSContext *h)
640 {
641 int i;
642
643 /* copy predictors from top line (MB B and C) into cache */
644 for (i = 0; i < 3; i++) {
645 h->mv[MV_FWD_B2 + i] = h->top_mv[0][h->mbx * 2 + i];
646 h->mv[MV_BWD_B2 + i] = h->top_mv[1][h->mbx * 2 + i];
647 }
648 h->pred_mode_Y[1] = h->top_pred_Y[h->mbx * 2 + 0];
649 h->pred_mode_Y[2] = h->top_pred_Y[h->mbx * 2 + 1];
650 /* clear top predictors if MB B is not available */
651 if (!(h->flags & B_AVAIL)) {
652 h->mv[MV_FWD_B2] = un_mv;
653 h->mv[MV_FWD_B3] = un_mv;
654 h->mv[MV_BWD_B2] = un_mv;
655 h->mv[MV_BWD_B3] = un_mv;
656 h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
657 h->flags &= ~(C_AVAIL | D_AVAIL);
658 } else if (h->mbx) {
659 h->flags |= D_AVAIL;
660 }
661 if (h->mbx == h->mb_width - 1) // MB C not available
662 h->flags &= ~C_AVAIL;
663 /* clear top-right predictors if MB C is not available */
664 if (!(h->flags & C_AVAIL)) {
665 h->mv[MV_FWD_C2] = un_mv;
666 h->mv[MV_BWD_C2] = un_mv;
667 }
668 /* clear top-left predictors if MB D is not available */
669 if (!(h->flags & D_AVAIL)) {
670 h->mv[MV_FWD_D3] = un_mv;
671 h->mv[MV_BWD_D3] = un_mv;
672 }
673 }
674
675 /**
676 * save predictors for later macroblocks and increase
677 * macroblock address
678 * @return 0 if end of frame is reached, 1 otherwise
679 */
ff_cavs_next_mb(AVSContext * h)680 int ff_cavs_next_mb(AVSContext *h)
681 {
682 int i;
683
684 h->flags |= A_AVAIL;
685 h->cy += 16;
686 h->cu += 8;
687 h->cv += 8;
688 /* copy mvs as predictors to the left */
689 for (i = 0; i <= 20; i += 4)
690 h->mv[i] = h->mv[i + 2];
691 /* copy bottom mvs from cache to top line */
692 h->top_mv[0][h->mbx * 2 + 0] = h->mv[MV_FWD_X2];
693 h->top_mv[0][h->mbx * 2 + 1] = h->mv[MV_FWD_X3];
694 h->top_mv[1][h->mbx * 2 + 0] = h->mv[MV_BWD_X2];
695 h->top_mv[1][h->mbx * 2 + 1] = h->mv[MV_BWD_X3];
696 /* next MB address */
697 h->mbidx++;
698 h->mbx++;
699 if (h->mbx == h->mb_width) { // New mb line
700 h->flags = B_AVAIL | C_AVAIL;
701 /* clear left pred_modes */
702 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
703 /* clear left mv predictors */
704 for (i = 0; i <= 20; i += 4)
705 h->mv[i] = un_mv;
706 h->mbx = 0;
707 h->mby++;
708 /* re-calculate sample pointers */
709 h->cy = h->cur.f->data[0] + h->mby * 16 * h->l_stride;
710 h->cu = h->cur.f->data[1] + h->mby * 8 * h->c_stride;
711 h->cv = h->cur.f->data[2] + h->mby * 8 * h->c_stride;
712 if (h->mby == h->mb_height) { // Frame end
713 return 0;
714 }
715 }
716 return 1;
717 }
718
719 /*****************************************************************************
720 *
721 * frame level
722 *
723 ****************************************************************************/
724
ff_cavs_init_pic(AVSContext * h)725 int ff_cavs_init_pic(AVSContext *h)
726 {
727 int i;
728
729 /* clear some predictors */
730 for (i = 0; i <= 20; i += 4)
731 h->mv[i] = un_mv;
732 h->mv[MV_BWD_X0] = ff_cavs_dir_mv;
733 set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
734 h->mv[MV_FWD_X0] = ff_cavs_dir_mv;
735 set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
736 h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
737 h->cy = h->cur.f->data[0];
738 h->cu = h->cur.f->data[1];
739 h->cv = h->cur.f->data[2];
740 h->l_stride = h->cur.f->linesize[0];
741 h->c_stride = h->cur.f->linesize[1];
742 h->luma_scan[2] = 8 * h->l_stride;
743 h->luma_scan[3] = 8 * h->l_stride + 8;
744 h->mbx = h->mby = h->mbidx = 0;
745 h->flags = 0;
746
747 return 0;
748 }
749
750 /*****************************************************************************
751 *
752 * headers and interface
753 *
754 ****************************************************************************/
755
756 /**
757 * some predictions require data from the top-neighbouring macroblock.
758 * this data has to be stored for one complete row of macroblocks
759 * and this storage space is allocated here
760 */
ff_cavs_init_top_lines(AVSContext * h)761 int ff_cavs_init_top_lines(AVSContext *h)
762 {
763 /* alloc top line of predictors */
764 h->top_qp = av_mallocz(h->mb_width);
765 h->top_mv[0] = av_mallocz_array(h->mb_width * 2 + 1, sizeof(cavs_vector));
766 h->top_mv[1] = av_mallocz_array(h->mb_width * 2 + 1, sizeof(cavs_vector));
767 h->top_pred_Y = av_mallocz_array(h->mb_width * 2, sizeof(*h->top_pred_Y));
768 h->top_border_y = av_mallocz_array(h->mb_width + 1, 16);
769 h->top_border_u = av_mallocz_array(h->mb_width, 10);
770 h->top_border_v = av_mallocz_array(h->mb_width, 10);
771
772 /* alloc space for co-located MVs and types */
773 h->col_mv = av_mallocz_array(h->mb_width * h->mb_height,
774 4 * sizeof(cavs_vector));
775 h->col_type_base = av_mallocz(h->mb_width * h->mb_height);
776 h->block = av_mallocz(64 * sizeof(int16_t));
777
778 if (!h->top_qp || !h->top_mv[0] || !h->top_mv[1] || !h->top_pred_Y ||
779 !h->top_border_y || !h->top_border_u || !h->top_border_v ||
780 !h->col_mv || !h->col_type_base || !h->block) {
781 av_freep(&h->top_qp);
782 av_freep(&h->top_mv[0]);
783 av_freep(&h->top_mv[1]);
784 av_freep(&h->top_pred_Y);
785 av_freep(&h->top_border_y);
786 av_freep(&h->top_border_u);
787 av_freep(&h->top_border_v);
788 av_freep(&h->col_mv);
789 av_freep(&h->col_type_base);
790 av_freep(&h->block);
791 return AVERROR(ENOMEM);
792 }
793 return 0;
794 }
795
ff_cavs_init(AVCodecContext * avctx)796 av_cold int ff_cavs_init(AVCodecContext *avctx)
797 {
798 AVSContext *h = avctx->priv_data;
799
800 ff_blockdsp_init(&h->bdsp, avctx);
801 ff_h264chroma_init(&h->h264chroma, 8);
802 ff_idctdsp_init(&h->idsp, avctx);
803 ff_videodsp_init(&h->vdsp, 8);
804 ff_cavsdsp_init(&h->cdsp, avctx);
805 ff_init_scantable_permutation(h->idsp.idct_permutation,
806 h->cdsp.idct_perm);
807 ff_init_scantable(h->idsp.idct_permutation, &h->scantable, ff_zigzag_direct);
808
809 h->avctx = avctx;
810 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
811
812 h->cur.f = av_frame_alloc();
813 h->DPB[0].f = av_frame_alloc();
814 h->DPB[1].f = av_frame_alloc();
815 if (!h->cur.f || !h->DPB[0].f || !h->DPB[1].f) {
816 ff_cavs_end(avctx);
817 return AVERROR(ENOMEM);
818 }
819
820 h->luma_scan[0] = 0;
821 h->luma_scan[1] = 8;
822 h->intra_pred_l[INTRA_L_VERT] = intra_pred_vert;
823 h->intra_pred_l[INTRA_L_HORIZ] = intra_pred_horiz;
824 h->intra_pred_l[INTRA_L_LP] = intra_pred_lp;
825 h->intra_pred_l[INTRA_L_DOWN_LEFT] = intra_pred_down_left;
826 h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
827 h->intra_pred_l[INTRA_L_LP_LEFT] = intra_pred_lp_left;
828 h->intra_pred_l[INTRA_L_LP_TOP] = intra_pred_lp_top;
829 h->intra_pred_l[INTRA_L_DC_128] = intra_pred_dc_128;
830 h->intra_pred_c[INTRA_C_LP] = intra_pred_lp;
831 h->intra_pred_c[INTRA_C_HORIZ] = intra_pred_horiz;
832 h->intra_pred_c[INTRA_C_VERT] = intra_pred_vert;
833 h->intra_pred_c[INTRA_C_PLANE] = intra_pred_plane;
834 h->intra_pred_c[INTRA_C_LP_LEFT] = intra_pred_lp_left;
835 h->intra_pred_c[INTRA_C_LP_TOP] = intra_pred_lp_top;
836 h->intra_pred_c[INTRA_C_DC_128] = intra_pred_dc_128;
837 h->mv[7] = un_mv;
838 h->mv[19] = un_mv;
839 return 0;
840 }
841
ff_cavs_end(AVCodecContext * avctx)842 av_cold int ff_cavs_end(AVCodecContext *avctx)
843 {
844 AVSContext *h = avctx->priv_data;
845
846 av_frame_free(&h->cur.f);
847 av_frame_free(&h->DPB[0].f);
848 av_frame_free(&h->DPB[1].f);
849
850 av_freep(&h->top_qp);
851 av_freep(&h->top_mv[0]);
852 av_freep(&h->top_mv[1]);
853 av_freep(&h->top_pred_Y);
854 av_freep(&h->top_border_y);
855 av_freep(&h->top_border_u);
856 av_freep(&h->top_border_v);
857 av_freep(&h->col_mv);
858 av_freep(&h->col_type_base);
859 av_freep(&h->block);
860 av_freep(&h->edge_emu_buffer);
861 return 0;
862 }
863