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