| @@ -0,0 +1,836 @@ |
| 1 | +/* |
| 2 | + * Apple ProRes encoder |
| 3 | + * |
| 4 | + * Copyright (c) 2012 Konstantin Shishkov |
| 5 | + * |
| 6 | + * This file is part of Libav. |
| 7 | + * |
| 8 | + * Libav is free software; you can redistribute it and/or |
| 9 | + * modify it under the terms of the GNU Lesser General Public |
| 10 | + * License as published by the Free Software Foundation; either |
| 11 | + * version 2.1 of the License, or (at your option) any later version. |
| 12 | + * |
| 13 | + * Libav is distributed in the hope that it will be useful, |
| 14 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | + * Lesser General Public License for more details. |
| 17 | + * |
| 18 | + * You should have received a copy of the GNU Lesser General Public |
| 19 | + * License along with Libav; if not, write to the Free Software |
| 20 | + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | + */ |
| 22 | + |
| 23 | +#include "libavutil/opt.h" |
| 24 | +#include "avcodec.h" |
| 25 | +#include "put_bits.h" |
| 26 | +#include "bytestream.h" |
| 27 | +#include "internal.h" |
| 28 | +#include "proresdsp.h" |
| 29 | +#include "proresdata.h" |
| 30 | + |
| 31 | +#define CFACTOR_Y422 2 |
| 32 | +#define CFACTOR_Y444 3 |
| 33 | + |
| 34 | +#define MAX_MBS_PER_SLICE 8 |
| 35 | + |
| 36 | +#define MAX_PLANES 3 // should be increased to 4 when there's PIX_FMT_YUV444AP10 |
| 37 | + |
| 38 | +enum { |
| 39 | + PRORES_PROFILE_PROXY = 0, |
| 40 | + PRORES_PROFILE_LT, |
| 41 | + PRORES_PROFILE_STANDARD, |
| 42 | + PRORES_PROFILE_HQ, |
| 43 | +}; |
| 44 | + |
| 45 | +#define NUM_MB_LIMITS 4 |
| 46 | +static const int prores_mb_limits[NUM_MB_LIMITS] = { |
| 47 | + 1620, // up to 720x576 |
| 48 | + 2700, // up to 960x720 |
| 49 | + 6075, // up to 1440x1080 |
| 50 | + 9216, // up to 2048x1152 |
| 51 | +}; |
| 52 | + |
| 53 | +static const struct prores_profile { |
| 54 | + const char *full_name; |
| 55 | + uint32_t tag; |
| 56 | + int min_quant; |
| 57 | + int max_quant; |
| 58 | + int br_tab[NUM_MB_LIMITS]; |
| 59 | + uint8_t quant[64]; |
| 60 | +} prores_profile_info[4] = { |
| 61 | + { |
| 62 | + .full_name = "proxy", |
| 63 | + .tag = MKTAG('a', 'p', 'c', 'o'), |
| 64 | + .min_quant = 4, |
| 65 | + .max_quant = 8, |
| 66 | + .br_tab = { 300, 242, 220, 194 }, |
| 67 | + .quant = { |
| 68 | + 4, 7, 9, 11, 13, 14, 15, 63, |
| 69 | + 7, 7, 11, 12, 14, 15, 63, 63, |
| 70 | + 9, 11, 13, 14, 15, 63, 63, 63, |
| 71 | + 11, 11, 13, 14, 63, 63, 63, 63, |
| 72 | + 11, 13, 14, 63, 63, 63, 63, 63, |
| 73 | + 13, 14, 63, 63, 63, 63, 63, 63, |
| 74 | + 13, 63, 63, 63, 63, 63, 63, 63, |
| 75 | + 63, 63, 63, 63, 63, 63, 63, 63, |
| 76 | + }, |
| 77 | + }, |
| 78 | + { |
| 79 | + .full_name = "LT", |
| 80 | + .tag = MKTAG('a', 'p', 'c', 's'), |
| 81 | + .min_quant = 1, |
| 82 | + .max_quant = 9, |
| 83 | + .br_tab = { 720, 560, 490, 440 }, |
| 84 | + .quant = { |
| 85 | + 4, 5, 6, 7, 9, 11, 13, 15, |
| 86 | + 5, 5, 7, 8, 11, 13, 15, 17, |
| 87 | + 6, 7, 9, 11, 13, 15, 15, 17, |
| 88 | + 7, 7, 9, 11, 13, 15, 17, 19, |
| 89 | + 7, 9, 11, 13, 14, 16, 19, 23, |
| 90 | + 9, 11, 13, 14, 16, 19, 23, 29, |
| 91 | + 9, 11, 13, 15, 17, 21, 28, 35, |
| 92 | + 11, 13, 16, 17, 21, 28, 35, 41, |
| 93 | + }, |
| 94 | + }, |
| 95 | + { |
| 96 | + .full_name = "standard", |
| 97 | + .tag = MKTAG('a', 'p', 'c', 'n'), |
| 98 | + .min_quant = 1, |
| 99 | + .max_quant = 6, |
| 100 | + .br_tab = { 1050, 808, 710, 632 }, |
| 101 | + .quant = { |
| 102 | + 4, 4, 5, 5, 6, 7, 7, 9, |
| 103 | + 4, 4, 5, 6, 7, 7, 9, 9, |
| 104 | + 5, 5, 6, 7, 7, 9, 9, 10, |
| 105 | + 5, 5, 6, 7, 7, 9, 9, 10, |
| 106 | + 5, 6, 7, 7, 8, 9, 10, 12, |
| 107 | + 6, 7, 7, 8, 9, 10, 12, 15, |
| 108 | + 6, 7, 7, 9, 10, 11, 14, 17, |
| 109 | + 7, 7, 9, 10, 11, 14, 17, 21, |
| 110 | + }, |
| 111 | + }, |
| 112 | + { |
| 113 | + .full_name = "high quality", |
| 114 | + .tag = MKTAG('a', 'p', 'c', 'h'), |
| 115 | + .min_quant = 1, |
| 116 | + .max_quant = 6, |
| 117 | + .br_tab = { 1566, 1216, 1070, 950 }, |
| 118 | + .quant = { |
| 119 | + 4, 4, 4, 4, 4, 4, 4, 4, |
| 120 | + 4, 4, 4, 4, 4, 4, 4, 4, |
| 121 | + 4, 4, 4, 4, 4, 4, 4, 4, |
| 122 | + 4, 4, 4, 4, 4, 4, 4, 5, |
| 123 | + 4, 4, 4, 4, 4, 4, 5, 5, |
| 124 | + 4, 4, 4, 4, 4, 5, 5, 6, |
| 125 | + 4, 4, 4, 4, 5, 5, 6, 7, |
| 126 | + 4, 4, 4, 4, 5, 6, 7, 7, |
| 127 | + }, |
| 128 | + } |
| 129 | +// for 4444 profile bitrate numbers are { 2350, 1828, 1600, 1425 } |
| 130 | +}; |
| 131 | + |
| 132 | +#define TRELLIS_WIDTH 16 |
| 133 | +#define SCORE_LIMIT INT_MAX / 2 |
| 134 | + |
| 135 | +struct TrellisNode { |
| 136 | + int prev_node; |
| 137 | + int quant; |
| 138 | + int bits; |
| 139 | + int score; |
| 140 | +}; |
| 141 | + |
| 142 | +typedef struct ProresContext { |
| 143 | + AVClass *class; |
| 144 | + DECLARE_ALIGNED(16, DCTELEM, blocks)[MAX_PLANES][64 * 4 * MAX_MBS_PER_SLICE]; |
| 145 | + DECLARE_ALIGNED(16, uint16_t, emu_buf)[16*16]; |
| 146 | + int16_t quants[16][64]; |
| 147 | + |
| 148 | + ProresDSPContext dsp; |
| 149 | + ScanTable scantable; |
| 150 | + |
| 151 | + int mb_width, mb_height; |
| 152 | + int mbs_per_slice; |
| 153 | + int num_chroma_blocks, chroma_factor; |
| 154 | + int slices_width; |
| 155 | + int num_slices; |
| 156 | + int num_planes; |
| 157 | + int bits_per_mb; |
| 158 | + |
| 159 | + int profile; |
| 160 | + const struct prores_profile *profile_info; |
| 161 | + |
| 162 | + struct TrellisNode *nodes; |
| 163 | + int *slice_q; |
| 164 | +} ProresContext; |
| 165 | + |
| 166 | +static void get_slice_data(ProresContext *ctx, const uint16_t *src, |
| 167 | + int linesize, int x, int y, int w, int h, |
| 168 | + DCTELEM *blocks, |
| 169 | + int mbs_per_slice, int blocks_per_mb) |
| 170 | +{ |
| 171 | + const uint16_t *esrc; |
| 172 | + const int mb_width = 4 * blocks_per_mb; |
| 173 | + int elinesize; |
| 174 | + int i, j, k; |
| 175 | + |
| 176 | + for (i = 0; i < mbs_per_slice; i++, src += mb_width) { |
| 177 | + if (x >= w) { |
| 178 | + memset(blocks, 0, 64 * (mbs_per_slice - i) * blocks_per_mb |
| 179 | + * sizeof(*blocks)); |
| 180 | + return; |
| 181 | + } |
| 182 | + if (x + mb_width <= w && y + 16 <= h) { |
| 183 | + esrc = src; |
| 184 | + elinesize = linesize; |
| 185 | + } else { |
| 186 | + int bw, bh, pix; |
| 187 | + const int estride = 16 / sizeof(*ctx->emu_buf); |
| 188 | + |
| 189 | + esrc = ctx->emu_buf; |
| 190 | + elinesize = 16; |
| 191 | + |
| 192 | + bw = FFMIN(w - x, mb_width); |
| 193 | + bh = FFMIN(h - y, 16); |
| 194 | + |
| 195 | + for (j = 0; j < bh; j++) { |
| 196 | + memcpy(ctx->emu_buf + j * estride, src + j * linesize, |
| 197 | + bw * sizeof(*src)); |
| 198 | + pix = ctx->emu_buf[j * estride + bw - 1]; |
| 199 | + for (k = bw; k < mb_width; k++) |
| 200 | + ctx->emu_buf[j * estride + k] = pix; |
| 201 | + } |
| 202 | + for (; j < 16; j++) |
| 203 | + memcpy(ctx->emu_buf + j * estride, |
| 204 | + ctx->emu_buf + (bh - 1) * estride, |
| 205 | + mb_width * sizeof(*ctx->emu_buf)); |
| 206 | + } |
| 207 | + ctx->dsp.fdct(esrc, elinesize, blocks); |
| 208 | + blocks += 64; |
| 209 | + if (blocks_per_mb > 2) { |
| 210 | + ctx->dsp.fdct(src + 8, linesize, blocks); |
| 211 | + blocks += 64; |
| 212 | + } |
| 213 | + ctx->dsp.fdct(src + linesize * 4, linesize, blocks); |
| 214 | + blocks += 64; |
| 215 | + if (blocks_per_mb > 2) { |
| 216 | + ctx->dsp.fdct(src + linesize * 4 + 8, linesize, blocks); |
| 217 | + blocks += 64; |
| 218 | + } |
| 219 | + |
| 220 | + x += mb_width; |
| 221 | + } |
| 222 | +} |
| 223 | + |
| 224 | +/** |
| 225 | + * Write an unsigned rice/exp golomb codeword. |
| 226 | + */ |
| 227 | +static inline void encode_vlc_codeword(PutBitContext *pb, uint8_t codebook, int val) |
| 228 | +{ |
| 229 | + unsigned int rice_order, exp_order, switch_bits, switch_val; |
| 230 | + int exponent; |
| 231 | + |
| 232 | + /* number of prefix bits to switch between Rice and expGolomb */ |
| 233 | + switch_bits = (codebook & 3) + 1; |
| 234 | + rice_order = codebook >> 5; /* rice code order */ |
| 235 | + exp_order = (codebook >> 2) & 7; /* exp golomb code order */ |
| 236 | + |
| 237 | + switch_val = switch_bits << rice_order; |
| 238 | + |
| 239 | + if (val >= switch_val) { |
| 240 | + val -= switch_val - (1 << exp_order); |
| 241 | + exponent = av_log2(val); |
| 242 | + |
| 243 | + put_bits(pb, exponent - exp_order + switch_bits, 0); |
| 244 | + put_bits(pb, 1, 1); |
| 245 | + put_bits(pb, exponent, val); |
| 246 | + } else { |
| 247 | + exponent = val >> rice_order; |
| 248 | + |
| 249 | + if (exponent) |
| 250 | + put_bits(pb, exponent, 0); |
| 251 | + put_bits(pb, 1, 1); |
| 252 | + if (rice_order) |
| 253 | + put_sbits(pb, rice_order, val); |
| 254 | + } |
| 255 | +} |
| 256 | + |
| 257 | +#define GET_SIGN(x) ((x) >> 31) |
| 258 | +#define MAKE_CODE(x) (((x) << 1) ^ GET_SIGN(x)) |
| 259 | + |
| 260 | +static void encode_dcs(PutBitContext *pb, DCTELEM *blocks, |
| 261 | + int blocks_per_slice, int scale) |
| 262 | +{ |
| 263 | + int i; |
| 264 | + int codebook = 3, code, dc, prev_dc, delta, sign, new_sign; |
| 265 | + |
| 266 | + prev_dc = (blocks[0] - 0x4000) / scale; |
| 267 | + encode_vlc_codeword(pb, FIRST_DC_CB, MAKE_CODE(prev_dc)); |
| 268 | + codebook = 3; |
| 269 | + blocks += 64; |
| 270 | + |
| 271 | + for (i = 1; i < blocks_per_slice; i++, blocks += 64) { |
| 272 | + dc = (blocks[0] - 0x4000) / scale; |
| 273 | + delta = dc - prev_dc; |
| 274 | + new_sign = GET_SIGN(delta); |
| 275 | + delta = (delta ^ sign) - sign; |
| 276 | + code = MAKE_CODE(delta); |
| 277 | + encode_vlc_codeword(pb, ff_prores_dc_codebook[codebook], code); |
| 278 | + codebook = (code + (code & 1)) >> 1; |
| 279 | + codebook = FFMIN(codebook, 3); |
| 280 | + sign = new_sign; |
| 281 | + prev_dc = dc; |
| 282 | + } |
| 283 | +} |
| 284 | + |
| 285 | +static void encode_acs(PutBitContext *pb, DCTELEM *blocks, |
| 286 | + int blocks_per_slice, |
| 287 | + int plane_size_factor, |
| 288 | + const uint8_t *scan, const int16_t *qmat) |
| 289 | +{ |
| 290 | + int idx, i; |
| 291 | + int run, level, run_cb, lev_cb; |
| 292 | + int max_coeffs, abs_level; |
| 293 | + |
| 294 | + max_coeffs = blocks_per_slice << 6; |
| 295 | + run_cb = ff_prores_run_to_cb_index[4]; |
| 296 | + lev_cb = ff_prores_lev_to_cb_index[2]; |
| 297 | + run = 0; |
| 298 | + |
| 299 | + for (i = 1; i < 64; i++) { |
| 300 | + for (idx = scan[i]; idx < max_coeffs; idx += 64) { |
| 301 | + level = blocks[idx] / qmat[scan[i]]; |
| 302 | + if (level) { |
| 303 | + abs_level = FFABS(level); |
| 304 | + encode_vlc_codeword(pb, ff_prores_ac_codebook[run_cb], run); |
| 305 | + encode_vlc_codeword(pb, ff_prores_ac_codebook[lev_cb], |
| 306 | + abs_level - 1); |
| 307 | + put_sbits(pb, 1, GET_SIGN(level)); |
| 308 | + |
| 309 | + run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)]; |
| 310 | + lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)]; |
| 311 | + run = 0; |
| 312 | + } else { |
| 313 | + run++; |
| 314 | + } |
| 315 | + } |
| 316 | + } |
| 317 | +} |
| 318 | + |
| 319 | +static int encode_slice_plane(ProresContext *ctx, PutBitContext *pb, |
| 320 | + const uint16_t *src, int linesize, |
| 321 | + int mbs_per_slice, DCTELEM *blocks, |
| 322 | + int blocks_per_mb, int plane_size_factor, |
| 323 | + const int16_t *qmat) |
| 324 | +{ |
| 325 | + int blocks_per_slice, saved_pos; |
| 326 | + |
| 327 | + saved_pos = put_bits_count(pb); |
| 328 | + blocks_per_slice = mbs_per_slice * blocks_per_mb; |
| 329 | + |
| 330 | + encode_dcs(pb, blocks, blocks_per_slice, qmat[0]); |
| 331 | + encode_acs(pb, blocks, blocks_per_slice, plane_size_factor, |
| 332 | + ctx->scantable.permutated, qmat); |
| 333 | + flush_put_bits(pb); |
| 334 | + |
| 335 | + return (put_bits_count(pb) - saved_pos) >> 3; |
| 336 | +} |
| 337 | + |
| 338 | +static int encode_slice(AVCodecContext *avctx, const AVFrame *pic, |
| 339 | + PutBitContext *pb, |
| 340 | + int sizes[4], int x, int y, int quant, |
| 341 | + int mbs_per_slice) |
| 342 | +{ |
| 343 | + ProresContext *ctx = avctx->priv_data; |
| 344 | + int i, xp, yp; |
| 345 | + int total_size = 0; |
| 346 | + const uint16_t *src; |
| 347 | + int slice_width_factor = av_log2(mbs_per_slice); |
| 348 | + int num_cblocks, pwidth; |
| 349 | + int plane_factor, is_chroma; |
| 350 | + |
| 351 | + for (i = 0; i < ctx->num_planes; i++) { |
| 352 | + is_chroma = (i == 1 || i == 2); |
| 353 | + plane_factor = slice_width_factor + 2; |
| 354 | + if (is_chroma) |
| 355 | + plane_factor += ctx->chroma_factor - 3; |
| 356 | + if (!is_chroma || ctx->chroma_factor == CFACTOR_Y444) { |
| 357 | + xp = x << 4; |
| 358 | + yp = y << 4; |
| 359 | + num_cblocks = 4; |
| 360 | + pwidth = avctx->width; |
| 361 | + } else { |
| 362 | + xp = x << 3; |
| 363 | + yp = y << 4; |
| 364 | + num_cblocks = 2; |
| 365 | + pwidth = avctx->width >> 1; |
| 366 | + } |
| 367 | + src = (const uint16_t*)(pic->data[i] + yp * pic->linesize[i]) + xp; |
| 368 | + |
| 369 | + get_slice_data(ctx, src, pic->linesize[i], xp, yp, |
| 370 | + pwidth, avctx->height, ctx->blocks[0], |
| 371 | + mbs_per_slice, num_cblocks); |
| 372 | + sizes[i] = encode_slice_plane(ctx, pb, src, pic->linesize[i], |
| 373 | + mbs_per_slice, ctx->blocks[0], |
| 374 | + num_cblocks, plane_factor, |
| 375 | + ctx->quants[quant]); |
| 376 | + total_size += sizes[i]; |
| 377 | + } |
| 378 | + return total_size; |
| 379 | +} |
| 380 | + |
| 381 | +static inline int estimate_vlc(uint8_t codebook, int val) |
| 382 | +{ |
| 383 | + unsigned int rice_order, exp_order, switch_bits, switch_val; |
| 384 | + int exponent; |
| 385 | + |
| 386 | + /* number of prefix bits to switch between Rice and expGolomb */ |
| 387 | + switch_bits = (codebook & 3) + 1; |
| 388 | + rice_order = codebook >> 5; /* rice code order */ |
| 389 | + exp_order = (codebook >> 2) & 7; /* exp golomb code order */ |
| 390 | + |
| 391 | + switch_val = switch_bits << rice_order; |
| 392 | + |
| 393 | + if (val >= switch_val) { |
| 394 | + val -= switch_val - (1 << exp_order); |
| 395 | + exponent = av_log2(val); |
| 396 | + |
| 397 | + return exponent * 2 - exp_order + switch_bits + 1; |
| 398 | + } else { |
| 399 | + return (val >> rice_order) + rice_order + 1; |
| 400 | + } |
| 401 | +} |
| 402 | + |
| 403 | +static int estimate_dcs(int *error, DCTELEM *blocks, int blocks_per_slice, |
| 404 | + int scale) |
| 405 | +{ |
| 406 | + int i; |
| 407 | + int codebook = 3, code, dc, prev_dc, delta, sign, new_sign; |
| 408 | + int bits; |
| 409 | + |
| 410 | + prev_dc = (blocks[0] - 0x4000) / scale; |
| 411 | + bits = estimate_vlc(FIRST_DC_CB, MAKE_CODE(prev_dc)); |
| 412 | + codebook = 3; |
| 413 | + blocks += 64; |
| 414 | + *error += FFABS(blocks[0] - 0x4000) % scale; |
| 415 | + |
| 416 | + for (i = 1; i < blocks_per_slice; i++, blocks += 64) { |
| 417 | + dc = (blocks[0] - 0x4000) / scale; |
| 418 | + *error += FFABS(blocks[0] - 0x4000) % scale; |
| 419 | + delta = dc - prev_dc; |
| 420 | + new_sign = GET_SIGN(delta); |
| 421 | + delta = (delta ^ sign) - sign; |
| 422 | + code = MAKE_CODE(delta); |
| 423 | + bits += estimate_vlc(ff_prores_dc_codebook[codebook], code); |
| 424 | + codebook = (code + (code & 1)) >> 1; |
| 425 | + codebook = FFMIN(codebook, 3); |
| 426 | + sign = new_sign; |
| 427 | + prev_dc = dc; |
| 428 | + } |
| 429 | + |
| 430 | + return bits; |
| 431 | +} |
| 432 | + |
| 433 | +static int estimate_acs(int *error, DCTELEM *blocks, int blocks_per_slice, |
| 434 | + int plane_size_factor, |
| 435 | + const uint8_t *scan, const int16_t *qmat) |
| 436 | +{ |
| 437 | + int idx, i; |
| 438 | + int run, level, run_cb, lev_cb; |
| 439 | + int max_coeffs, abs_level; |
| 440 | + int bits = 0; |
| 441 | + |
| 442 | + max_coeffs = blocks_per_slice << 6; |
| 443 | + run_cb = ff_prores_run_to_cb_index[4]; |
| 444 | + lev_cb = ff_prores_lev_to_cb_index[2]; |
| 445 | + run = 0; |
| 446 | + |
| 447 | + for (i = 1; i < 64; i++) { |
| 448 | + for (idx = scan[i]; idx < max_coeffs; idx += 64) { |
| 449 | + level = blocks[idx] / qmat[scan[i]]; |
| 450 | + *error += FFABS(blocks[idx]) % qmat[scan[i]]; |
| 451 | + if (level) { |
| 452 | + abs_level = FFABS(level); |
| 453 | + bits += estimate_vlc(ff_prores_ac_codebook[run_cb], run); |
| 454 | + bits += estimate_vlc(ff_prores_ac_codebook[lev_cb], |
| 455 | + abs_level - 1) + 1; |
| 456 | + |
| 457 | + run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)]; |
| 458 | + lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)]; |
| 459 | + run = 0; |
| 460 | + } else { |
| 461 | + run++; |
| 462 | + } |
| 463 | + } |
| 464 | + } |
| 465 | + |
| 466 | + return bits; |
| 467 | +} |
| 468 | + |
| 469 | +static int estimate_slice_plane(ProresContext *ctx, int *error, int plane, |
| 470 | + const uint16_t *src, int linesize, |
| 471 | + int mbs_per_slice, |
| 472 | + int blocks_per_mb, int plane_size_factor, |
| 473 | + const int16_t *qmat) |
| 474 | +{ |
| 475 | + int blocks_per_slice; |
| 476 | + int bits; |
| 477 | + |
| 478 | + blocks_per_slice = mbs_per_slice * blocks_per_mb; |
| 479 | + |
| 480 | + bits = estimate_dcs(error, ctx->blocks[plane], blocks_per_slice, qmat[0]); |
| 481 | + bits += estimate_acs(error, ctx->blocks[plane], blocks_per_slice, |
| 482 | + plane_size_factor, ctx->scantable.permutated, qmat); |
| 483 | + |
| 484 | + return FFALIGN(bits, 8); |
| 485 | +} |
| 486 | + |
| 487 | +static int find_slice_quant(AVCodecContext *avctx, const AVFrame *pic, |
| 488 | + int trellis_node, int x, int y, int mbs_per_slice) |
| 489 | +{ |
| 490 | + ProresContext *ctx = avctx->priv_data; |
| 491 | + int i, q, pq, xp, yp; |
| 492 | + const uint16_t *src; |
| 493 | + int slice_width_factor = av_log2(mbs_per_slice); |
| 494 | + int num_cblocks[MAX_PLANES], pwidth; |
| 495 | + int plane_factor[MAX_PLANES], is_chroma[MAX_PLANES]; |
| 496 | + const int min_quant = ctx->profile_info->min_quant; |
| 497 | + const int max_quant = ctx->profile_info->max_quant; |
| 498 | + int error, bits, bits_limit; |
| 499 | + int mbs, prev, cur, new_score; |
| 500 | + int slice_bits[TRELLIS_WIDTH], slice_score[TRELLIS_WIDTH]; |
| 501 | + |
| 502 | + mbs = x + mbs_per_slice; |
| 503 | + |
| 504 | + for (i = 0; i < ctx->num_planes; i++) { |
| 505 | + is_chroma[i] = (i == 1 || i == 2); |
| 506 | + plane_factor[i] = slice_width_factor + 2; |
| 507 | + if (is_chroma[i]) |
| 508 | + plane_factor[i] += ctx->chroma_factor - 3; |
| 509 | + if (!is_chroma[i] || ctx->chroma_factor == CFACTOR_Y444) { |
| 510 | + xp = x << 4; |
| 511 | + yp = y << 4; |
| 512 | + num_cblocks[i] = 4; |
| 513 | + pwidth = avctx->width; |
| 514 | + } else { |
| 515 | + xp = x << 3; |
| 516 | + yp = y << 4; |
| 517 | + num_cblocks[i] = 2; |
| 518 | + pwidth = avctx->width >> 1; |
| 519 | + } |
| 520 | + src = (const uint16_t*)(pic->data[i] + yp * pic->linesize[i]) + xp; |
| 521 | + |
| 522 | + get_slice_data(ctx, src, pic->linesize[i], xp, yp, |
| 523 | + pwidth, avctx->height, ctx->blocks[i], |
| 524 | + mbs_per_slice, num_cblocks[i]); |
| 525 | + } |
| 526 | + |
| 527 | + for (q = min_quant; q <= max_quant; q++) { |
| 528 | + ctx->nodes[trellis_node + q].prev_node = -1; |
| 529 | + ctx->nodes[trellis_node + q].quant = q; |
| 530 | + } |
| 531 | + |
| 532 | + // todo: maybe perform coarser quantising to fit into frame size when needed |
| 533 | + for (q = min_quant; q <= max_quant; q++) { |
| 534 | + bits = 0; |
| 535 | + error = 0; |
| 536 | + for (i = 0; i < ctx->num_planes; i++) { |
| 537 | + bits += estimate_slice_plane(ctx, &error, i, |
| 538 | + src, pic->linesize[i], |
| 539 | + mbs_per_slice, |
| 540 | + num_cblocks[i], plane_factor[i], |
| 541 | + ctx->quants[q]); |
| 542 | + } |
| 543 | + if (bits > 65000 * 8) { |
| 544 | + error = SCORE_LIMIT; |
| 545 | + break; |
| 546 | + } |
| 547 | + slice_bits[q] = bits; |
| 548 | + slice_score[q] = error; |
| 549 | + } |
| 550 | + |
| 551 | + bits_limit = mbs * ctx->bits_per_mb; |
| 552 | + for (pq = min_quant; pq <= max_quant; pq++) { |
| 553 | + prev = trellis_node - TRELLIS_WIDTH + pq; |
| 554 | + |
| 555 | + for (q = min_quant; q <= max_quant; q++) { |
| 556 | + cur = trellis_node + q; |
| 557 | + |
| 558 | + bits = ctx->nodes[prev].bits + slice_bits[q]; |
| 559 | + error = slice_score[q]; |
| 560 | + if (bits > bits_limit) |
| 561 | + error = SCORE_LIMIT; |
| 562 | + |
| 563 | + if (ctx->nodes[prev].score < SCORE_LIMIT && error < SCORE_LIMIT) |
| 564 | + new_score = ctx->nodes[prev].score + error; |
| 565 | + else |
| 566 | + new_score = SCORE_LIMIT; |
| 567 | + if (ctx->nodes[cur].prev_node == -1 || |
| 568 | + ctx->nodes[cur].score >= new_score) { |
| 569 | + |
| 570 | + ctx->nodes[cur].bits = bits; |
| 571 | + ctx->nodes[cur].score = new_score; |
| 572 | + ctx->nodes[cur].prev_node = prev; |
| 573 | + } |
| 574 | + } |
| 575 | + } |
| 576 | + |
| 577 | + error = ctx->nodes[trellis_node + min_quant].score; |
| 578 | + pq = trellis_node + min_quant; |
| 579 | + for (q = min_quant + 1; q <= max_quant; q++) { |
| 580 | + if (ctx->nodes[trellis_node + q].score <= error) { |
| 581 | + error = ctx->nodes[trellis_node + q].score; |
| 582 | + pq = trellis_node + q; |
| 583 | + } |
| 584 | + } |
| 585 | + |
| 586 | + return pq; |
| 587 | +} |
| 588 | + |
| 589 | +static int encode_frame(AVCodecContext *avctx, AVPacket *pkt, |
| 590 | + const AVFrame *pic, int *got_packet) |
| 591 | +{ |
| 592 | + ProresContext *ctx = avctx->priv_data; |
| 593 | + uint8_t *orig_buf, *buf, *slice_hdr, *slice_sizes, *tmp; |
| 594 | + uint8_t *picture_size_pos; |
| 595 | + PutBitContext pb; |
| 596 | + int x, y, i, mb, q = 0; |
| 597 | + int sizes[4] = { 0 }; |
| 598 | + int slice_hdr_size = 2 + 2 * (ctx->num_planes - 1); |
| 599 | + int frame_size, picture_size, slice_size; |
| 600 | + int mbs_per_slice = ctx->mbs_per_slice; |
| 601 | + int pkt_size, ret; |
| 602 | + |
| 603 | + *avctx->coded_frame = *pic; |
| 604 | + avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I; |
| 605 | + avctx->coded_frame->key_frame = 1; |
| 606 | + |
| 607 | + pkt_size = ctx->mb_width * ctx->mb_height * 64 * 3 * 12 |
| 608 | + + ctx->num_slices * 2 + 200 + FF_MIN_BUFFER_SIZE; |
| 609 | + |
| 610 | + if ((ret = ff_alloc_packet(pkt, pkt_size)) < 0) { |
| 611 | + av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n"); |
| 612 | + return ret; |
| 613 | + } |
| 614 | + |
| 615 | + orig_buf = pkt->data; |
| 616 | + |
| 617 | + // frame atom |
| 618 | + orig_buf += 4; // frame size |
| 619 | + bytestream_put_be32 (&orig_buf, FRAME_ID); // frame container ID |
| 620 | + buf = orig_buf; |
| 621 | + |
| 622 | + // frame header |
| 623 | + tmp = buf; |
| 624 | + buf += 2; // frame header size will be stored here |
| 625 | + bytestream_put_be16 (&buf, 0); // version 1 |
| 626 | + bytestream_put_buffer(&buf, "Lavc", 4); // creator |
| 627 | + bytestream_put_be16 (&buf, avctx->width); |
| 628 | + bytestream_put_be16 (&buf, avctx->height); |
| 629 | + bytestream_put_byte (&buf, ctx->chroma_factor << 6); // frame flags |
| 630 | + bytestream_put_byte (&buf, 0); // reserved |
| 631 | + bytestream_put_byte (&buf, 0); // primaries |
| 632 | + bytestream_put_byte (&buf, 0); // transfer function |
| 633 | + bytestream_put_byte (&buf, 6); // colour matrix - ITU-R BT.601-4 |
| 634 | + bytestream_put_byte (&buf, 0x40); // source format and alpha information |
| 635 | + bytestream_put_byte (&buf, 0); // reserved |
| 636 | + bytestream_put_byte (&buf, 0x03); // matrix flags - both matrices are present |
| 637 | + // luma quantisation matrix |
| 638 | + for (i = 0; i < 64; i++) |
| 639 | + bytestream_put_byte(&buf, ctx->profile_info->quant[i]); |
| 640 | + // chroma quantisation matrix |
| 641 | + for (i = 0; i < 64; i++) |
| 642 | + bytestream_put_byte(&buf, ctx->profile_info->quant[i]); |
| 643 | + bytestream_put_be16 (&tmp, buf - orig_buf); // write back frame header size |
| 644 | + |
| 645 | + // picture header |
| 646 | + picture_size_pos = buf + 1; |
| 647 | + bytestream_put_byte (&buf, 0x40); // picture header size (in bits) |
| 648 | + buf += 4; // picture data size will be stored here |
| 649 | + bytestream_put_be16 (&buf, ctx->num_slices); // total number of slices |
| 650 | + bytestream_put_byte (&buf, av_log2(ctx->mbs_per_slice) << 4); // slice width and height in MBs |
| 651 | + |
| 652 | + // seek table - will be filled during slice encoding |
| 653 | + slice_sizes = buf; |
| 654 | + buf += ctx->num_slices * 2; |
| 655 | + |
| 656 | + // slices |
| 657 | + for (y = 0; y < ctx->mb_height; y++) { |
| 658 | + mbs_per_slice = ctx->mbs_per_slice; |
| 659 | + for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) { |
| 660 | + while (ctx->mb_width - x < mbs_per_slice) |
| 661 | + mbs_per_slice >>= 1; |
| 662 | + q = find_slice_quant(avctx, pic, (mb + 1) * TRELLIS_WIDTH, x, y, |
| 663 | + mbs_per_slice); |
| 664 | + } |
| 665 | + |
| 666 | + for (x = ctx->slices_width - 1; x >= 0; x--) { |
| 667 | + ctx->slice_q[x] = ctx->nodes[q].quant; |
| 668 | + q = ctx->nodes[q].prev_node; |
| 669 | + } |
| 670 | + |
| 671 | + mbs_per_slice = ctx->mbs_per_slice; |
| 672 | + for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) { |
| 673 | + q = ctx->slice_q[mb]; |
| 674 | + |
| 675 | + while (ctx->mb_width - x < mbs_per_slice) |
| 676 | + mbs_per_slice >>= 1; |
| 677 | + |
| 678 | + bytestream_put_byte(&buf, slice_hdr_size << 3); |
| 679 | + slice_hdr = buf; |
| 680 | + buf += slice_hdr_size - 1; |
| 681 | + init_put_bits(&pb, buf, (pkt_size - (buf - orig_buf)) * 8); |
| 682 | + encode_slice(avctx, pic, &pb, sizes, x, y, q, mbs_per_slice); |
| 683 | + |
| 684 | + bytestream_put_byte(&slice_hdr, q); |
| 685 | + slice_size = slice_hdr_size + sizes[ctx->num_planes - 1]; |
| 686 | + for (i = 0; i < ctx->num_planes - 1; i++) { |
| 687 | + bytestream_put_be16(&slice_hdr, sizes[i]); |
| 688 | + slice_size += sizes[i]; |
| 689 | + } |
| 690 | + bytestream_put_be16(&slice_sizes, slice_size); |
| 691 | + buf += slice_size - slice_hdr_size; |
| 692 | + } |
| 693 | + } |
| 694 | + |
| 695 | + orig_buf -= 8; |
| 696 | + frame_size = buf - orig_buf; |
| 697 | + picture_size = buf - picture_size_pos - 6; |
| 698 | + bytestream_put_be32(&orig_buf, frame_size); |
| 699 | + bytestream_put_be32(&picture_size_pos, picture_size); |
| 700 | + |
| 701 | + pkt->size = frame_size; |
| 702 | + pkt->flags |= AV_PKT_FLAG_KEY; |
| 703 | + *got_packet = 1; |
| 704 | + |
| 705 | + return 0; |
| 706 | +} |
| 707 | + |
| 708 | +static av_cold int encode_close(AVCodecContext *avctx) |
| 709 | +{ |
| 710 | + ProresContext *ctx = avctx->priv_data; |
| 711 | + |
| 712 | + if (avctx->coded_frame->data[0]) |
| 713 | + avctx->release_buffer(avctx, avctx->coded_frame); |
| 714 | + |
| 715 | + av_freep(&avctx->coded_frame); |
| 716 | + |
| 717 | + av_freep(&ctx->nodes); |
| 718 | + av_freep(&ctx->slice_q); |
| 719 | + |
| 720 | + return 0; |
| 721 | +} |
| 722 | + |
| 723 | +static av_cold int encode_init(AVCodecContext *avctx) |
| 724 | +{ |
| 725 | + ProresContext *ctx = avctx->priv_data; |
| 726 | + int mps; |
| 727 | + int i, j; |
| 728 | + int min_quant, max_quant; |
| 729 | + |
| 730 | + avctx->bits_per_raw_sample = 10; |
| 731 | + avctx->coded_frame = avcodec_alloc_frame(); |
| 732 | + if (!avctx->coded_frame) |
| 733 | + return AVERROR(ENOMEM); |
| 734 | + |
| 735 | + ff_proresdsp_init(&ctx->dsp); |
| 736 | + ff_init_scantable(ctx->dsp.dct_permutation, &ctx->scantable, |
| 737 | + ff_prores_progressive_scan); |
| 738 | + |
| 739 | + mps = ctx->mbs_per_slice; |
| 740 | + if (mps & (mps - 1)) { |
| 741 | + av_log(avctx, AV_LOG_ERROR, |
| 742 | + "there should be an integer power of two MBs per slice\n"); |
| 743 | + return AVERROR(EINVAL); |
| 744 | + } |
| 745 | + |
| 746 | + ctx->chroma_factor = avctx->pix_fmt == PIX_FMT_YUV422P10 |
| 747 | + ? CFACTOR_Y422 |
| 748 | + : CFACTOR_Y444; |
| 749 | + ctx->profile_info = prores_profile_info + ctx->profile; |
| 750 | + ctx->num_planes = 3; |
| 751 | + |
| 752 | + ctx->mb_width = FFALIGN(avctx->width, 16) >> 4; |
| 753 | + ctx->mb_height = FFALIGN(avctx->height, 16) >> 4; |
| 754 | + ctx->slices_width = ctx->mb_width / mps; |
| 755 | + ctx->slices_width += av_popcount(ctx->mb_width - ctx->slices_width * mps); |
| 756 | + ctx->num_slices = ctx->mb_height * ctx->slices_width; |
| 757 | + |
| 758 | + for (i = 0; i < NUM_MB_LIMITS - 1; i++) |
| 759 | + if (prores_mb_limits[i] >= ctx->mb_width * ctx->mb_height) |
| 760 | + break; |
| 761 | + ctx->bits_per_mb = ctx->profile_info->br_tab[i]; |
| 762 | + |
| 763 | + min_quant = ctx->profile_info->min_quant; |
| 764 | + max_quant = ctx->profile_info->max_quant; |
| 765 | + for (i = min_quant; i <= max_quant; i++) { |
| 766 | + for (j = 0; j < 64; j++) |
| 767 | + ctx->quants[i][j] = ctx->profile_info->quant[j] * i; |
| 768 | + } |
| 769 | + |
| 770 | + avctx->codec_tag = ctx->profile_info->tag; |
| 771 | + |
| 772 | + av_log(avctx, AV_LOG_DEBUG, "profile %d, %d slices, %d bits per MB\n", |
| 773 | + ctx->profile, ctx->num_slices, ctx->bits_per_mb); |
| 774 | + |
| 775 | + ctx->nodes = av_malloc((ctx->slices_width + 1) * TRELLIS_WIDTH |
| 776 | + * sizeof(*ctx->nodes)); |
| 777 | + if (!ctx->nodes) { |
| 778 | + encode_close(avctx); |
| 779 | + return AVERROR(ENOMEM); |
| 780 | + } |
| 781 | + for (i = min_quant; i <= max_quant; i++) { |
| 782 | + ctx->nodes[i].prev_node = -1; |
| 783 | + ctx->nodes[i].bits = 0; |
| 784 | + ctx->nodes[i].score = 0; |
| 785 | + } |
| 786 | + |
| 787 | + ctx->slice_q = av_malloc(ctx->slices_width * sizeof(*ctx->slice_q)); |
| 788 | + if (!ctx->slice_q) { |
| 789 | + encode_close(avctx); |
| 790 | + return AVERROR(ENOMEM); |
| 791 | + } |
| 792 | + |
| 793 | + return 0; |
| 794 | +} |
| 795 | + |
| 796 | +#define OFFSET(x) offsetof(ProresContext, x) |
| 797 | +#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM |
| 798 | + |
| 799 | +static const AVOption options[] = { |
| 800 | + { "mbs_per_slice", "macroblocks per slice", OFFSET(mbs_per_slice), |
| 801 | + AV_OPT_TYPE_INT, { 8 }, 1, MAX_MBS_PER_SLICE, VE }, |
| 802 | + { "profile", NULL, OFFSET(profile), AV_OPT_TYPE_INT, |
| 803 | + { PRORES_PROFILE_STANDARD }, |
| 804 | + PRORES_PROFILE_PROXY, PRORES_PROFILE_HQ, VE, "profile" }, |
| 805 | + { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_PROXY }, |
| 806 | + 0, 0, VE, "profile" }, |
| 807 | + { "lt", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_LT }, |
| 808 | + 0, 0, VE, "profile" }, |
| 809 | + { "standard", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_STANDARD }, |
| 810 | + 0, 0, VE, "profile" }, |
| 811 | + { "hq", NULL, 0, AV_OPT_TYPE_CONST, { PRORES_PROFILE_HQ }, |
| 812 | + 0, 0, VE, "profile" }, |
| 813 | + { NULL } |
| 814 | +}; |
| 815 | + |
| 816 | +static const AVClass proresenc_class = { |
| 817 | + .class_name = "ProRes encoder", |
| 818 | + .item_name = av_default_item_name, |
| 819 | + .option = options, |
| 820 | + .version = LIBAVUTIL_VERSION_INT, |
| 821 | +}; |
| 822 | + |
| 823 | +AVCodec ff_prores_encoder = { |
| 824 | + .name = "prores", |
| 825 | + .type = AVMEDIA_TYPE_VIDEO, |
| 826 | + .id = CODEC_ID_PRORES, |
| 827 | + .priv_data_size = sizeof(ProresContext), |
| 828 | + .init = encode_init, |
| 829 | + .close = encode_close, |
| 830 | + .encode2 = encode_frame, |
| 831 | + .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)"), |
| 832 | + .pix_fmts = (const enum PixelFormat[]) { |
| 833 | + PIX_FMT_YUV422P10, PIX_FMT_YUV444P10, PIX_FMT_NONE |
| 834 | + }, |
| 835 | + .priv_class = &proresenc_class, |
| 836 | +}; |