Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
6147 | serge | 1 | /* |
2 | * Apple ProRes encoder |
||
3 | * |
||
4 | * Copyright (c) 2012 Konstantin Shishkov |
||
5 | * |
||
6 | * This encoder appears to be based on Anatoliy Wassermans considering |
||
7 | * similarities in the bugs. |
||
8 | * |
||
9 | * This file is part of FFmpeg. |
||
10 | * |
||
11 | * FFmpeg is free software; you can redistribute it and/or |
||
12 | * modify it under the terms of the GNU Lesser General Public |
||
13 | * License as published by the Free Software Foundation; either |
||
14 | * version 2.1 of the License, or (at your option) any later version. |
||
15 | * |
||
16 | * FFmpeg is distributed in the hope that it will be useful, |
||
17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
||
18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
||
19 | * Lesser General Public License for more details. |
||
20 | * |
||
21 | * You should have received a copy of the GNU Lesser General Public |
||
22 | * License along with FFmpeg; if not, write to the Free Software |
||
23 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
||
24 | */ |
||
25 | |||
26 | #include "libavutil/opt.h" |
||
27 | #include "libavutil/pixdesc.h" |
||
28 | #include "avcodec.h" |
||
29 | #include "fdctdsp.h" |
||
30 | #include "put_bits.h" |
||
31 | #include "bytestream.h" |
||
32 | #include "internal.h" |
||
33 | #include "proresdata.h" |
||
34 | |||
35 | #define CFACTOR_Y422 2 |
||
36 | #define CFACTOR_Y444 3 |
||
37 | |||
38 | #define MAX_MBS_PER_SLICE 8 |
||
39 | |||
40 | #define MAX_PLANES 4 |
||
41 | |||
42 | enum { |
||
43 | PRORES_PROFILE_AUTO = -1, |
||
44 | PRORES_PROFILE_PROXY = 0, |
||
45 | PRORES_PROFILE_LT, |
||
46 | PRORES_PROFILE_STANDARD, |
||
47 | PRORES_PROFILE_HQ, |
||
48 | PRORES_PROFILE_4444, |
||
49 | }; |
||
50 | |||
51 | enum { |
||
52 | QUANT_MAT_PROXY = 0, |
||
53 | QUANT_MAT_LT, |
||
54 | QUANT_MAT_STANDARD, |
||
55 | QUANT_MAT_HQ, |
||
56 | QUANT_MAT_DEFAULT, |
||
57 | }; |
||
58 | |||
59 | static const uint8_t prores_quant_matrices[][64] = { |
||
60 | { // proxy |
||
61 | 4, 7, 9, 11, 13, 14, 15, 63, |
||
62 | 7, 7, 11, 12, 14, 15, 63, 63, |
||
63 | 9, 11, 13, 14, 15, 63, 63, 63, |
||
64 | 11, 11, 13, 14, 63, 63, 63, 63, |
||
65 | 11, 13, 14, 63, 63, 63, 63, 63, |
||
66 | 13, 14, 63, 63, 63, 63, 63, 63, |
||
67 | 13, 63, 63, 63, 63, 63, 63, 63, |
||
68 | 63, 63, 63, 63, 63, 63, 63, 63, |
||
69 | }, |
||
70 | { // LT |
||
71 | 4, 5, 6, 7, 9, 11, 13, 15, |
||
72 | 5, 5, 7, 8, 11, 13, 15, 17, |
||
73 | 6, 7, 9, 11, 13, 15, 15, 17, |
||
74 | 7, 7, 9, 11, 13, 15, 17, 19, |
||
75 | 7, 9, 11, 13, 14, 16, 19, 23, |
||
76 | 9, 11, 13, 14, 16, 19, 23, 29, |
||
77 | 9, 11, 13, 15, 17, 21, 28, 35, |
||
78 | 11, 13, 16, 17, 21, 28, 35, 41, |
||
79 | }, |
||
80 | { // standard |
||
81 | 4, 4, 5, 5, 6, 7, 7, 9, |
||
82 | 4, 4, 5, 6, 7, 7, 9, 9, |
||
83 | 5, 5, 6, 7, 7, 9, 9, 10, |
||
84 | 5, 5, 6, 7, 7, 9, 9, 10, |
||
85 | 5, 6, 7, 7, 8, 9, 10, 12, |
||
86 | 6, 7, 7, 8, 9, 10, 12, 15, |
||
87 | 6, 7, 7, 9, 10, 11, 14, 17, |
||
88 | 7, 7, 9, 10, 11, 14, 17, 21, |
||
89 | }, |
||
90 | { // high quality |
||
91 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
92 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
93 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
94 | 4, 4, 4, 4, 4, 4, 4, 5, |
||
95 | 4, 4, 4, 4, 4, 4, 5, 5, |
||
96 | 4, 4, 4, 4, 4, 5, 5, 6, |
||
97 | 4, 4, 4, 4, 5, 5, 6, 7, |
||
98 | 4, 4, 4, 4, 5, 6, 7, 7, |
||
99 | }, |
||
100 | { // codec default |
||
101 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
102 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
103 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
104 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
105 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
106 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
107 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
108 | 4, 4, 4, 4, 4, 4, 4, 4, |
||
109 | }, |
||
110 | }; |
||
111 | |||
112 | #define NUM_MB_LIMITS 4 |
||
113 | static const int prores_mb_limits[NUM_MB_LIMITS] = { |
||
114 | 1620, // up to 720x576 |
||
115 | 2700, // up to 960x720 |
||
116 | 6075, // up to 1440x1080 |
||
117 | 9216, // up to 2048x1152 |
||
118 | }; |
||
119 | |||
120 | static const struct prores_profile { |
||
121 | const char *full_name; |
||
122 | uint32_t tag; |
||
123 | int min_quant; |
||
124 | int max_quant; |
||
125 | int br_tab[NUM_MB_LIMITS]; |
||
126 | int quant; |
||
127 | } prores_profile_info[5] = { |
||
128 | { |
||
129 | .full_name = "proxy", |
||
130 | .tag = MKTAG('a', 'p', 'c', 'o'), |
||
131 | .min_quant = 4, |
||
132 | .max_quant = 8, |
||
133 | .br_tab = { 300, 242, 220, 194 }, |
||
134 | .quant = QUANT_MAT_PROXY, |
||
135 | }, |
||
136 | { |
||
137 | .full_name = "LT", |
||
138 | .tag = MKTAG('a', 'p', 'c', 's'), |
||
139 | .min_quant = 1, |
||
140 | .max_quant = 9, |
||
141 | .br_tab = { 720, 560, 490, 440 }, |
||
142 | .quant = QUANT_MAT_LT, |
||
143 | }, |
||
144 | { |
||
145 | .full_name = "standard", |
||
146 | .tag = MKTAG('a', 'p', 'c', 'n'), |
||
147 | .min_quant = 1, |
||
148 | .max_quant = 6, |
||
149 | .br_tab = { 1050, 808, 710, 632 }, |
||
150 | .quant = QUANT_MAT_STANDARD, |
||
151 | }, |
||
152 | { |
||
153 | .full_name = "high quality", |
||
154 | .tag = MKTAG('a', 'p', 'c', 'h'), |
||
155 | .min_quant = 1, |
||
156 | .max_quant = 6, |
||
157 | .br_tab = { 1566, 1216, 1070, 950 }, |
||
158 | .quant = QUANT_MAT_HQ, |
||
159 | }, |
||
160 | { |
||
161 | .full_name = "4444", |
||
162 | .tag = MKTAG('a', 'p', '4', 'h'), |
||
163 | .min_quant = 1, |
||
164 | .max_quant = 6, |
||
165 | .br_tab = { 2350, 1828, 1600, 1425 }, |
||
166 | .quant = QUANT_MAT_HQ, |
||
167 | } |
||
168 | }; |
||
169 | |||
170 | #define TRELLIS_WIDTH 16 |
||
171 | #define SCORE_LIMIT INT_MAX / 2 |
||
172 | |||
173 | struct TrellisNode { |
||
174 | int prev_node; |
||
175 | int quant; |
||
176 | int bits; |
||
177 | int score; |
||
178 | }; |
||
179 | |||
180 | #define MAX_STORED_Q 16 |
||
181 | |||
182 | typedef struct ProresThreadData { |
||
183 | DECLARE_ALIGNED(16, int16_t, blocks)[MAX_PLANES][64 * 4 * MAX_MBS_PER_SLICE]; |
||
184 | DECLARE_ALIGNED(16, uint16_t, emu_buf)[16 * 16]; |
||
185 | int16_t custom_q[64]; |
||
186 | struct TrellisNode *nodes; |
||
187 | } ProresThreadData; |
||
188 | |||
189 | typedef struct ProresContext { |
||
190 | AVClass *class; |
||
191 | DECLARE_ALIGNED(16, int16_t, blocks)[MAX_PLANES][64 * 4 * MAX_MBS_PER_SLICE]; |
||
192 | DECLARE_ALIGNED(16, uint16_t, emu_buf)[16*16]; |
||
193 | int16_t quants[MAX_STORED_Q][64]; |
||
194 | int16_t custom_q[64]; |
||
195 | const uint8_t *quant_mat; |
||
196 | const uint8_t *scantable; |
||
197 | |||
198 | void (*fdct)(FDCTDSPContext *fdsp, const uint16_t *src, |
||
199 | int linesize, int16_t *block); |
||
200 | FDCTDSPContext fdsp; |
||
201 | |||
202 | const AVFrame *pic; |
||
203 | int mb_width, mb_height; |
||
204 | int mbs_per_slice; |
||
205 | int num_chroma_blocks, chroma_factor; |
||
206 | int slices_width; |
||
207 | int slices_per_picture; |
||
208 | int pictures_per_frame; // 1 for progressive, 2 for interlaced |
||
209 | int cur_picture_idx; |
||
210 | int num_planes; |
||
211 | int bits_per_mb; |
||
212 | int force_quant; |
||
213 | int alpha_bits; |
||
214 | int warn; |
||
215 | |||
216 | char *vendor; |
||
217 | int quant_sel; |
||
218 | |||
219 | int frame_size_upper_bound; |
||
220 | |||
221 | int profile; |
||
222 | const struct prores_profile *profile_info; |
||
223 | |||
224 | int *slice_q; |
||
225 | |||
226 | ProresThreadData *tdata; |
||
227 | } ProresContext; |
||
228 | |||
229 | static void get_slice_data(ProresContext *ctx, const uint16_t *src, |
||
230 | int linesize, int x, int y, int w, int h, |
||
231 | int16_t *blocks, uint16_t *emu_buf, |
||
232 | int mbs_per_slice, int blocks_per_mb, int is_chroma) |
||
233 | { |
||
234 | const uint16_t *esrc; |
||
235 | const int mb_width = 4 * blocks_per_mb; |
||
236 | int elinesize; |
||
237 | int i, j, k; |
||
238 | |||
239 | for (i = 0; i < mbs_per_slice; i++, src += mb_width) { |
||
240 | if (x >= w) { |
||
241 | memset(blocks, 0, 64 * (mbs_per_slice - i) * blocks_per_mb |
||
242 | * sizeof(*blocks)); |
||
243 | return; |
||
244 | } |
||
245 | if (x + mb_width <= w && y + 16 <= h) { |
||
246 | esrc = src; |
||
247 | elinesize = linesize; |
||
248 | } else { |
||
249 | int bw, bh, pix; |
||
250 | |||
251 | esrc = emu_buf; |
||
252 | elinesize = 16 * sizeof(*emu_buf); |
||
253 | |||
254 | bw = FFMIN(w - x, mb_width); |
||
255 | bh = FFMIN(h - y, 16); |
||
256 | |||
257 | for (j = 0; j < bh; j++) { |
||
258 | memcpy(emu_buf + j * 16, |
||
259 | (const uint8_t*)src + j * linesize, |
||
260 | bw * sizeof(*src)); |
||
261 | pix = emu_buf[j * 16 + bw - 1]; |
||
262 | for (k = bw; k < mb_width; k++) |
||
263 | emu_buf[j * 16 + k] = pix; |
||
264 | } |
||
265 | for (; j < 16; j++) |
||
266 | memcpy(emu_buf + j * 16, |
||
267 | emu_buf + (bh - 1) * 16, |
||
268 | mb_width * sizeof(*emu_buf)); |
||
269 | } |
||
270 | if (!is_chroma) { |
||
271 | ctx->fdct(&ctx->fdsp, esrc, elinesize, blocks); |
||
272 | blocks += 64; |
||
273 | if (blocks_per_mb > 2) { |
||
274 | ctx->fdct(&ctx->fdsp, esrc + 8, elinesize, blocks); |
||
275 | blocks += 64; |
||
276 | } |
||
277 | ctx->fdct(&ctx->fdsp, esrc + elinesize * 4, elinesize, blocks); |
||
278 | blocks += 64; |
||
279 | if (blocks_per_mb > 2) { |
||
280 | ctx->fdct(&ctx->fdsp, esrc + elinesize * 4 + 8, elinesize, blocks); |
||
281 | blocks += 64; |
||
282 | } |
||
283 | } else { |
||
284 | ctx->fdct(&ctx->fdsp, esrc, elinesize, blocks); |
||
285 | blocks += 64; |
||
286 | ctx->fdct(&ctx->fdsp, esrc + elinesize * 4, elinesize, blocks); |
||
287 | blocks += 64; |
||
288 | if (blocks_per_mb > 2) { |
||
289 | ctx->fdct(&ctx->fdsp, esrc + 8, elinesize, blocks); |
||
290 | blocks += 64; |
||
291 | ctx->fdct(&ctx->fdsp, esrc + elinesize * 4 + 8, elinesize, blocks); |
||
292 | blocks += 64; |
||
293 | } |
||
294 | } |
||
295 | |||
296 | x += mb_width; |
||
297 | } |
||
298 | } |
||
299 | |||
300 | static void get_alpha_data(ProresContext *ctx, const uint16_t *src, |
||
301 | int linesize, int x, int y, int w, int h, |
||
302 | int16_t *blocks, int mbs_per_slice, int abits) |
||
303 | { |
||
304 | const int slice_width = 16 * mbs_per_slice; |
||
305 | int i, j, copy_w, copy_h; |
||
306 | |||
307 | copy_w = FFMIN(w - x, slice_width); |
||
308 | copy_h = FFMIN(h - y, 16); |
||
309 | for (i = 0; i < copy_h; i++) { |
||
310 | memcpy(blocks, src, copy_w * sizeof(*src)); |
||
311 | if (abits == 8) |
||
312 | for (j = 0; j < copy_w; j++) |
||
313 | blocks[j] >>= 2; |
||
314 | else |
||
315 | for (j = 0; j < copy_w; j++) |
||
316 | blocks[j] = (blocks[j] << 6) | (blocks[j] >> 4); |
||
317 | for (j = copy_w; j < slice_width; j++) |
||
318 | blocks[j] = blocks[copy_w - 1]; |
||
319 | blocks += slice_width; |
||
320 | src += linesize >> 1; |
||
321 | } |
||
322 | for (; i < 16; i++) { |
||
323 | memcpy(blocks, blocks - slice_width, slice_width * sizeof(*blocks)); |
||
324 | blocks += slice_width; |
||
325 | } |
||
326 | } |
||
327 | |||
328 | /** |
||
329 | * Write an unsigned rice/exp golomb codeword. |
||
330 | */ |
||
331 | static inline void encode_vlc_codeword(PutBitContext *pb, unsigned codebook, int val) |
||
332 | { |
||
333 | unsigned int rice_order, exp_order, switch_bits, switch_val; |
||
334 | int exponent; |
||
335 | |||
336 | /* number of prefix bits to switch between Rice and expGolomb */ |
||
337 | switch_bits = (codebook & 3) + 1; |
||
338 | rice_order = codebook >> 5; /* rice code order */ |
||
339 | exp_order = (codebook >> 2) & 7; /* exp golomb code order */ |
||
340 | |||
341 | switch_val = switch_bits << rice_order; |
||
342 | |||
343 | if (val >= switch_val) { |
||
344 | val -= switch_val - (1 << exp_order); |
||
345 | exponent = av_log2(val); |
||
346 | |||
347 | put_bits(pb, exponent - exp_order + switch_bits, 0); |
||
348 | put_bits(pb, exponent + 1, val); |
||
349 | } else { |
||
350 | exponent = val >> rice_order; |
||
351 | |||
352 | if (exponent) |
||
353 | put_bits(pb, exponent, 0); |
||
354 | put_bits(pb, 1, 1); |
||
355 | if (rice_order) |
||
356 | put_sbits(pb, rice_order, val); |
||
357 | } |
||
358 | } |
||
359 | |||
360 | #define GET_SIGN(x) ((x) >> 31) |
||
361 | #define MAKE_CODE(x) (((x) << 1) ^ GET_SIGN(x)) |
||
362 | |||
363 | static void encode_dcs(PutBitContext *pb, int16_t *blocks, |
||
364 | int blocks_per_slice, int scale) |
||
365 | { |
||
366 | int i; |
||
367 | int codebook = 3, code, dc, prev_dc, delta, sign, new_sign; |
||
368 | |||
369 | prev_dc = (blocks[0] - 0x4000) / scale; |
||
370 | encode_vlc_codeword(pb, FIRST_DC_CB, MAKE_CODE(prev_dc)); |
||
371 | sign = 0; |
||
372 | codebook = 3; |
||
373 | blocks += 64; |
||
374 | |||
375 | for (i = 1; i < blocks_per_slice; i++, blocks += 64) { |
||
376 | dc = (blocks[0] - 0x4000) / scale; |
||
377 | delta = dc - prev_dc; |
||
378 | new_sign = GET_SIGN(delta); |
||
379 | delta = (delta ^ sign) - sign; |
||
380 | code = MAKE_CODE(delta); |
||
381 | encode_vlc_codeword(pb, ff_prores_dc_codebook[codebook], code); |
||
382 | codebook = (code + (code & 1)) >> 1; |
||
383 | codebook = FFMIN(codebook, 3); |
||
384 | sign = new_sign; |
||
385 | prev_dc = dc; |
||
386 | } |
||
387 | } |
||
388 | |||
389 | static void encode_acs(PutBitContext *pb, int16_t *blocks, |
||
390 | int blocks_per_slice, |
||
391 | int plane_size_factor, |
||
392 | const uint8_t *scan, const int16_t *qmat) |
||
393 | { |
||
394 | int idx, i; |
||
395 | int run, level, run_cb, lev_cb; |
||
396 | int max_coeffs, abs_level; |
||
397 | |||
398 | max_coeffs = blocks_per_slice << 6; |
||
399 | run_cb = ff_prores_run_to_cb_index[4]; |
||
400 | lev_cb = ff_prores_lev_to_cb_index[2]; |
||
401 | run = 0; |
||
402 | |||
403 | for (i = 1; i < 64; i++) { |
||
404 | for (idx = scan[i]; idx < max_coeffs; idx += 64) { |
||
405 | level = blocks[idx] / qmat[scan[i]]; |
||
406 | if (level) { |
||
407 | abs_level = FFABS(level); |
||
408 | encode_vlc_codeword(pb, ff_prores_ac_codebook[run_cb], run); |
||
409 | encode_vlc_codeword(pb, ff_prores_ac_codebook[lev_cb], |
||
410 | abs_level - 1); |
||
411 | put_sbits(pb, 1, GET_SIGN(level)); |
||
412 | |||
413 | run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)]; |
||
414 | lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)]; |
||
415 | run = 0; |
||
416 | } else { |
||
417 | run++; |
||
418 | } |
||
419 | } |
||
420 | } |
||
421 | } |
||
422 | |||
423 | static int encode_slice_plane(ProresContext *ctx, PutBitContext *pb, |
||
424 | const uint16_t *src, int linesize, |
||
425 | int mbs_per_slice, int16_t *blocks, |
||
426 | int blocks_per_mb, int plane_size_factor, |
||
427 | const int16_t *qmat) |
||
428 | { |
||
429 | int blocks_per_slice, saved_pos; |
||
430 | |||
431 | saved_pos = put_bits_count(pb); |
||
432 | blocks_per_slice = mbs_per_slice * blocks_per_mb; |
||
433 | |||
434 | encode_dcs(pb, blocks, blocks_per_slice, qmat[0]); |
||
435 | encode_acs(pb, blocks, blocks_per_slice, plane_size_factor, |
||
436 | ctx->scantable, qmat); |
||
437 | flush_put_bits(pb); |
||
438 | |||
439 | return (put_bits_count(pb) - saved_pos) >> 3; |
||
440 | } |
||
441 | |||
442 | static void put_alpha_diff(PutBitContext *pb, int cur, int prev, int abits) |
||
443 | { |
||
444 | const int dbits = (abits == 8) ? 4 : 7; |
||
445 | const int dsize = 1 << dbits - 1; |
||
446 | int diff = cur - prev; |
||
447 | |||
448 | diff = av_mod_uintp2(diff, abits); |
||
449 | if (diff >= (1 << abits) - dsize) |
||
450 | diff -= 1 << abits; |
||
451 | if (diff < -dsize || diff > dsize || !diff) { |
||
452 | put_bits(pb, 1, 1); |
||
453 | put_bits(pb, abits, diff); |
||
454 | } else { |
||
455 | put_bits(pb, 1, 0); |
||
456 | put_bits(pb, dbits - 1, FFABS(diff) - 1); |
||
457 | put_bits(pb, 1, diff < 0); |
||
458 | } |
||
459 | } |
||
460 | |||
461 | static void put_alpha_run(PutBitContext *pb, int run) |
||
462 | { |
||
463 | if (run) { |
||
464 | put_bits(pb, 1, 0); |
||
465 | if (run < 0x10) |
||
466 | put_bits(pb, 4, run); |
||
467 | else |
||
468 | put_bits(pb, 15, run); |
||
469 | } else { |
||
470 | put_bits(pb, 1, 1); |
||
471 | } |
||
472 | } |
||
473 | |||
474 | // todo alpha quantisation for high quants |
||
475 | static int encode_alpha_plane(ProresContext *ctx, PutBitContext *pb, |
||
476 | int mbs_per_slice, uint16_t *blocks, |
||
477 | int quant) |
||
478 | { |
||
479 | const int abits = ctx->alpha_bits; |
||
480 | const int mask = (1 << abits) - 1; |
||
481 | const int num_coeffs = mbs_per_slice * 256; |
||
482 | int saved_pos = put_bits_count(pb); |
||
483 | int prev = mask, cur; |
||
484 | int idx = 0; |
||
485 | int run = 0; |
||
486 | |||
487 | cur = blocks[idx++]; |
||
488 | put_alpha_diff(pb, cur, prev, abits); |
||
489 | prev = cur; |
||
490 | do { |
||
491 | cur = blocks[idx++]; |
||
492 | if (cur != prev) { |
||
493 | put_alpha_run (pb, run); |
||
494 | put_alpha_diff(pb, cur, prev, abits); |
||
495 | prev = cur; |
||
496 | run = 0; |
||
497 | } else { |
||
498 | run++; |
||
499 | } |
||
500 | } while (idx < num_coeffs); |
||
501 | if (run) |
||
502 | put_alpha_run(pb, run); |
||
503 | flush_put_bits(pb); |
||
504 | return (put_bits_count(pb) - saved_pos) >> 3; |
||
505 | } |
||
506 | |||
507 | static int encode_slice(AVCodecContext *avctx, const AVFrame *pic, |
||
508 | PutBitContext *pb, |
||
509 | int sizes[4], int x, int y, int quant, |
||
510 | int mbs_per_slice) |
||
511 | { |
||
512 | ProresContext *ctx = avctx->priv_data; |
||
513 | int i, xp, yp; |
||
514 | int total_size = 0; |
||
515 | const uint16_t *src; |
||
516 | int slice_width_factor = av_log2(mbs_per_slice); |
||
517 | int num_cblocks, pwidth, linesize, line_add; |
||
518 | int plane_factor, is_chroma; |
||
519 | uint16_t *qmat; |
||
520 | |||
521 | if (ctx->pictures_per_frame == 1) |
||
522 | line_add = 0; |
||
523 | else |
||
524 | line_add = ctx->cur_picture_idx ^ !pic->top_field_first; |
||
525 | |||
526 | if (ctx->force_quant) { |
||
527 | qmat = ctx->quants[0]; |
||
528 | } else if (quant < MAX_STORED_Q) { |
||
529 | qmat = ctx->quants[quant]; |
||
530 | } else { |
||
531 | qmat = ctx->custom_q; |
||
532 | for (i = 0; i < 64; i++) |
||
533 | qmat[i] = ctx->quant_mat[i] * quant; |
||
534 | } |
||
535 | |||
536 | for (i = 0; i < ctx->num_planes; i++) { |
||
537 | is_chroma = (i == 1 || i == 2); |
||
538 | plane_factor = slice_width_factor + 2; |
||
539 | if (is_chroma) |
||
540 | plane_factor += ctx->chroma_factor - 3; |
||
541 | if (!is_chroma || ctx->chroma_factor == CFACTOR_Y444) { |
||
542 | xp = x << 4; |
||
543 | yp = y << 4; |
||
544 | num_cblocks = 4; |
||
545 | pwidth = avctx->width; |
||
546 | } else { |
||
547 | xp = x << 3; |
||
548 | yp = y << 4; |
||
549 | num_cblocks = 2; |
||
550 | pwidth = avctx->width >> 1; |
||
551 | } |
||
552 | |||
553 | linesize = pic->linesize[i] * ctx->pictures_per_frame; |
||
554 | src = (const uint16_t*)(pic->data[i] + yp * linesize + |
||
555 | line_add * pic->linesize[i]) + xp; |
||
556 | |||
557 | if (i < 3) { |
||
558 | get_slice_data(ctx, src, linesize, xp, yp, |
||
559 | pwidth, avctx->height / ctx->pictures_per_frame, |
||
560 | ctx->blocks[0], ctx->emu_buf, |
||
561 | mbs_per_slice, num_cblocks, is_chroma); |
||
562 | sizes[i] = encode_slice_plane(ctx, pb, src, linesize, |
||
563 | mbs_per_slice, ctx->blocks[0], |
||
564 | num_cblocks, plane_factor, |
||
565 | qmat); |
||
566 | } else { |
||
567 | get_alpha_data(ctx, src, linesize, xp, yp, |
||
568 | pwidth, avctx->height / ctx->pictures_per_frame, |
||
569 | ctx->blocks[0], mbs_per_slice, ctx->alpha_bits); |
||
570 | sizes[i] = encode_alpha_plane(ctx, pb, mbs_per_slice, |
||
571 | ctx->blocks[0], quant); |
||
572 | } |
||
573 | total_size += sizes[i]; |
||
574 | if (put_bits_left(pb) < 0) { |
||
575 | av_log(avctx, AV_LOG_ERROR, |
||
576 | "Underestimated required buffer size.\n"); |
||
577 | return AVERROR_BUG; |
||
578 | } |
||
579 | } |
||
580 | return total_size; |
||
581 | } |
||
582 | |||
583 | static inline int estimate_vlc(unsigned codebook, int val) |
||
584 | { |
||
585 | unsigned int rice_order, exp_order, switch_bits, switch_val; |
||
586 | int exponent; |
||
587 | |||
588 | /* number of prefix bits to switch between Rice and expGolomb */ |
||
589 | switch_bits = (codebook & 3) + 1; |
||
590 | rice_order = codebook >> 5; /* rice code order */ |
||
591 | exp_order = (codebook >> 2) & 7; /* exp golomb code order */ |
||
592 | |||
593 | switch_val = switch_bits << rice_order; |
||
594 | |||
595 | if (val >= switch_val) { |
||
596 | val -= switch_val - (1 << exp_order); |
||
597 | exponent = av_log2(val); |
||
598 | |||
599 | return exponent * 2 - exp_order + switch_bits + 1; |
||
600 | } else { |
||
601 | return (val >> rice_order) + rice_order + 1; |
||
602 | } |
||
603 | } |
||
604 | |||
605 | static int estimate_dcs(int *error, int16_t *blocks, int blocks_per_slice, |
||
606 | int scale) |
||
607 | { |
||
608 | int i; |
||
609 | int codebook = 3, code, dc, prev_dc, delta, sign, new_sign; |
||
610 | int bits; |
||
611 | |||
612 | prev_dc = (blocks[0] - 0x4000) / scale; |
||
613 | bits = estimate_vlc(FIRST_DC_CB, MAKE_CODE(prev_dc)); |
||
614 | sign = 0; |
||
615 | codebook = 3; |
||
616 | blocks += 64; |
||
617 | *error += FFABS(blocks[0] - 0x4000) % scale; |
||
618 | |||
619 | for (i = 1; i < blocks_per_slice; i++, blocks += 64) { |
||
620 | dc = (blocks[0] - 0x4000) / scale; |
||
621 | *error += FFABS(blocks[0] - 0x4000) % scale; |
||
622 | delta = dc - prev_dc; |
||
623 | new_sign = GET_SIGN(delta); |
||
624 | delta = (delta ^ sign) - sign; |
||
625 | code = MAKE_CODE(delta); |
||
626 | bits += estimate_vlc(ff_prores_dc_codebook[codebook], code); |
||
627 | codebook = (code + (code & 1)) >> 1; |
||
628 | codebook = FFMIN(codebook, 3); |
||
629 | sign = new_sign; |
||
630 | prev_dc = dc; |
||
631 | } |
||
632 | |||
633 | return bits; |
||
634 | } |
||
635 | |||
636 | static int estimate_acs(int *error, int16_t *blocks, int blocks_per_slice, |
||
637 | int plane_size_factor, |
||
638 | const uint8_t *scan, const int16_t *qmat) |
||
639 | { |
||
640 | int idx, i; |
||
641 | int run, level, run_cb, lev_cb; |
||
642 | int max_coeffs, abs_level; |
||
643 | int bits = 0; |
||
644 | |||
645 | max_coeffs = blocks_per_slice << 6; |
||
646 | run_cb = ff_prores_run_to_cb_index[4]; |
||
647 | lev_cb = ff_prores_lev_to_cb_index[2]; |
||
648 | run = 0; |
||
649 | |||
650 | for (i = 1; i < 64; i++) { |
||
651 | for (idx = scan[i]; idx < max_coeffs; idx += 64) { |
||
652 | level = blocks[idx] / qmat[scan[i]]; |
||
653 | *error += FFABS(blocks[idx]) % qmat[scan[i]]; |
||
654 | if (level) { |
||
655 | abs_level = FFABS(level); |
||
656 | bits += estimate_vlc(ff_prores_ac_codebook[run_cb], run); |
||
657 | bits += estimate_vlc(ff_prores_ac_codebook[lev_cb], |
||
658 | abs_level - 1) + 1; |
||
659 | |||
660 | run_cb = ff_prores_run_to_cb_index[FFMIN(run, 15)]; |
||
661 | lev_cb = ff_prores_lev_to_cb_index[FFMIN(abs_level, 9)]; |
||
662 | run = 0; |
||
663 | } else { |
||
664 | run++; |
||
665 | } |
||
666 | } |
||
667 | } |
||
668 | |||
669 | return bits; |
||
670 | } |
||
671 | |||
672 | static int estimate_slice_plane(ProresContext *ctx, int *error, int plane, |
||
673 | const uint16_t *src, int linesize, |
||
674 | int mbs_per_slice, |
||
675 | int blocks_per_mb, int plane_size_factor, |
||
676 | const int16_t *qmat, ProresThreadData *td) |
||
677 | { |
||
678 | int blocks_per_slice; |
||
679 | int bits; |
||
680 | |||
681 | blocks_per_slice = mbs_per_slice * blocks_per_mb; |
||
682 | |||
683 | bits = estimate_dcs(error, td->blocks[plane], blocks_per_slice, qmat[0]); |
||
684 | bits += estimate_acs(error, td->blocks[plane], blocks_per_slice, |
||
685 | plane_size_factor, ctx->scantable, qmat); |
||
686 | |||
687 | return FFALIGN(bits, 8); |
||
688 | } |
||
689 | |||
690 | static int est_alpha_diff(int cur, int prev, int abits) |
||
691 | { |
||
692 | const int dbits = (abits == 8) ? 4 : 7; |
||
693 | const int dsize = 1 << dbits - 1; |
||
694 | int diff = cur - prev; |
||
695 | |||
696 | diff = av_mod_uintp2(diff, abits); |
||
697 | if (diff >= (1 << abits) - dsize) |
||
698 | diff -= 1 << abits; |
||
699 | if (diff < -dsize || diff > dsize || !diff) |
||
700 | return abits + 1; |
||
701 | else |
||
702 | return dbits + 1; |
||
703 | } |
||
704 | |||
705 | static int estimate_alpha_plane(ProresContext *ctx, int *error, |
||
706 | const uint16_t *src, int linesize, |
||
707 | int mbs_per_slice, int quant, |
||
708 | int16_t *blocks) |
||
709 | { |
||
710 | const int abits = ctx->alpha_bits; |
||
711 | const int mask = (1 << abits) - 1; |
||
712 | const int num_coeffs = mbs_per_slice * 256; |
||
713 | int prev = mask, cur; |
||
714 | int idx = 0; |
||
715 | int run = 0; |
||
716 | int bits; |
||
717 | |||
718 | *error = 0; |
||
719 | cur = blocks[idx++]; |
||
720 | bits = est_alpha_diff(cur, prev, abits); |
||
721 | prev = cur; |
||
722 | do { |
||
723 | cur = blocks[idx++]; |
||
724 | if (cur != prev) { |
||
725 | if (!run) |
||
726 | bits++; |
||
727 | else if (run < 0x10) |
||
728 | bits += 4; |
||
729 | else |
||
730 | bits += 15; |
||
731 | bits += est_alpha_diff(cur, prev, abits); |
||
732 | prev = cur; |
||
733 | run = 0; |
||
734 | } else { |
||
735 | run++; |
||
736 | } |
||
737 | } while (idx < num_coeffs); |
||
738 | |||
739 | if (run) { |
||
740 | if (run < 0x10) |
||
741 | bits += 4; |
||
742 | else |
||
743 | bits += 15; |
||
744 | } |
||
745 | |||
746 | return bits; |
||
747 | } |
||
748 | |||
749 | static int find_slice_quant(AVCodecContext *avctx, |
||
750 | int trellis_node, int x, int y, int mbs_per_slice, |
||
751 | ProresThreadData *td) |
||
752 | { |
||
753 | ProresContext *ctx = avctx->priv_data; |
||
754 | int i, q, pq, xp, yp; |
||
755 | const uint16_t *src; |
||
756 | int slice_width_factor = av_log2(mbs_per_slice); |
||
757 | int num_cblocks[MAX_PLANES], pwidth; |
||
758 | int plane_factor[MAX_PLANES], is_chroma[MAX_PLANES]; |
||
759 | const int min_quant = ctx->profile_info->min_quant; |
||
760 | const int max_quant = ctx->profile_info->max_quant; |
||
761 | int error, bits, bits_limit; |
||
762 | int mbs, prev, cur, new_score; |
||
763 | int slice_bits[TRELLIS_WIDTH], slice_score[TRELLIS_WIDTH]; |
||
764 | int overquant; |
||
765 | uint16_t *qmat; |
||
766 | int linesize[4], line_add; |
||
767 | |||
768 | if (ctx->pictures_per_frame == 1) |
||
769 | line_add = 0; |
||
770 | else |
||
771 | line_add = ctx->cur_picture_idx ^ !ctx->pic->top_field_first; |
||
772 | mbs = x + mbs_per_slice; |
||
773 | |||
774 | for (i = 0; i < ctx->num_planes; i++) { |
||
775 | is_chroma[i] = (i == 1 || i == 2); |
||
776 | plane_factor[i] = slice_width_factor + 2; |
||
777 | if (is_chroma[i]) |
||
778 | plane_factor[i] += ctx->chroma_factor - 3; |
||
779 | if (!is_chroma[i] || ctx->chroma_factor == CFACTOR_Y444) { |
||
780 | xp = x << 4; |
||
781 | yp = y << 4; |
||
782 | num_cblocks[i] = 4; |
||
783 | pwidth = avctx->width; |
||
784 | } else { |
||
785 | xp = x << 3; |
||
786 | yp = y << 4; |
||
787 | num_cblocks[i] = 2; |
||
788 | pwidth = avctx->width >> 1; |
||
789 | } |
||
790 | |||
791 | linesize[i] = ctx->pic->linesize[i] * ctx->pictures_per_frame; |
||
792 | src = (const uint16_t *)(ctx->pic->data[i] + yp * linesize[i] + |
||
793 | line_add * ctx->pic->linesize[i]) + xp; |
||
794 | |||
795 | if (i < 3) { |
||
796 | get_slice_data(ctx, src, linesize[i], xp, yp, |
||
797 | pwidth, avctx->height / ctx->pictures_per_frame, |
||
798 | td->blocks[i], td->emu_buf, |
||
799 | mbs_per_slice, num_cblocks[i], is_chroma[i]); |
||
800 | } else { |
||
801 | get_alpha_data(ctx, src, linesize[i], xp, yp, |
||
802 | pwidth, avctx->height / ctx->pictures_per_frame, |
||
803 | td->blocks[i], mbs_per_slice, ctx->alpha_bits); |
||
804 | } |
||
805 | } |
||
806 | |||
807 | for (q = min_quant; q < max_quant + 2; q++) { |
||
808 | td->nodes[trellis_node + q].prev_node = -1; |
||
809 | td->nodes[trellis_node + q].quant = q; |
||
810 | } |
||
811 | |||
812 | // todo: maybe perform coarser quantising to fit into frame size when needed |
||
813 | for (q = min_quant; q <= max_quant; q++) { |
||
814 | bits = 0; |
||
815 | error = 0; |
||
816 | for (i = 0; i < ctx->num_planes - !!ctx->alpha_bits; i++) { |
||
817 | bits += estimate_slice_plane(ctx, &error, i, |
||
818 | src, linesize[i], |
||
819 | mbs_per_slice, |
||
820 | num_cblocks[i], plane_factor[i], |
||
821 | ctx->quants[q], td); |
||
822 | } |
||
823 | if (ctx->alpha_bits) |
||
824 | bits += estimate_alpha_plane(ctx, &error, src, linesize[3], |
||
825 | mbs_per_slice, q, td->blocks[3]); |
||
826 | if (bits > 65000 * 8) |
||
827 | error = SCORE_LIMIT; |
||
828 | |||
829 | slice_bits[q] = bits; |
||
830 | slice_score[q] = error; |
||
831 | } |
||
832 | if (slice_bits[max_quant] <= ctx->bits_per_mb * mbs_per_slice) { |
||
833 | slice_bits[max_quant + 1] = slice_bits[max_quant]; |
||
834 | slice_score[max_quant + 1] = slice_score[max_quant] + 1; |
||
835 | overquant = max_quant; |
||
836 | } else { |
||
837 | for (q = max_quant + 1; q < 128; q++) { |
||
838 | bits = 0; |
||
839 | error = 0; |
||
840 | if (q < MAX_STORED_Q) { |
||
841 | qmat = ctx->quants[q]; |
||
842 | } else { |
||
843 | qmat = td->custom_q; |
||
844 | for (i = 0; i < 64; i++) |
||
845 | qmat[i] = ctx->quant_mat[i] * q; |
||
846 | } |
||
847 | for (i = 0; i < ctx->num_planes - !!ctx->alpha_bits; i++) { |
||
848 | bits += estimate_slice_plane(ctx, &error, i, |
||
849 | src, linesize[i], |
||
850 | mbs_per_slice, |
||
851 | num_cblocks[i], plane_factor[i], |
||
852 | qmat, td); |
||
853 | } |
||
854 | if (ctx->alpha_bits) |
||
855 | bits += estimate_alpha_plane(ctx, &error, src, linesize[3], |
||
856 | mbs_per_slice, q, td->blocks[3]); |
||
857 | if (bits <= ctx->bits_per_mb * mbs_per_slice) |
||
858 | break; |
||
859 | } |
||
860 | |||
861 | slice_bits[max_quant + 1] = bits; |
||
862 | slice_score[max_quant + 1] = error; |
||
863 | overquant = q; |
||
864 | } |
||
865 | td->nodes[trellis_node + max_quant + 1].quant = overquant; |
||
866 | |||
867 | bits_limit = mbs * ctx->bits_per_mb; |
||
868 | for (pq = min_quant; pq < max_quant + 2; pq++) { |
||
869 | prev = trellis_node - TRELLIS_WIDTH + pq; |
||
870 | |||
871 | for (q = min_quant; q < max_quant + 2; q++) { |
||
872 | cur = trellis_node + q; |
||
873 | |||
874 | bits = td->nodes[prev].bits + slice_bits[q]; |
||
875 | error = slice_score[q]; |
||
876 | if (bits > bits_limit) |
||
877 | error = SCORE_LIMIT; |
||
878 | |||
879 | if (td->nodes[prev].score < SCORE_LIMIT && error < SCORE_LIMIT) |
||
880 | new_score = td->nodes[prev].score + error; |
||
881 | else |
||
882 | new_score = SCORE_LIMIT; |
||
883 | if (td->nodes[cur].prev_node == -1 || |
||
884 | td->nodes[cur].score >= new_score) { |
||
885 | |||
886 | td->nodes[cur].bits = bits; |
||
887 | td->nodes[cur].score = new_score; |
||
888 | td->nodes[cur].prev_node = prev; |
||
889 | } |
||
890 | } |
||
891 | } |
||
892 | |||
893 | error = td->nodes[trellis_node + min_quant].score; |
||
894 | pq = trellis_node + min_quant; |
||
895 | for (q = min_quant + 1; q < max_quant + 2; q++) { |
||
896 | if (td->nodes[trellis_node + q].score <= error) { |
||
897 | error = td->nodes[trellis_node + q].score; |
||
898 | pq = trellis_node + q; |
||
899 | } |
||
900 | } |
||
901 | |||
902 | return pq; |
||
903 | } |
||
904 | |||
905 | static int find_quant_thread(AVCodecContext *avctx, void *arg, |
||
906 | int jobnr, int threadnr) |
||
907 | { |
||
908 | ProresContext *ctx = avctx->priv_data; |
||
909 | ProresThreadData *td = ctx->tdata + threadnr; |
||
910 | int mbs_per_slice = ctx->mbs_per_slice; |
||
911 | int x, y = jobnr, mb, q = 0; |
||
912 | |||
913 | for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) { |
||
914 | while (ctx->mb_width - x < mbs_per_slice) |
||
915 | mbs_per_slice >>= 1; |
||
916 | q = find_slice_quant(avctx, |
||
917 | (mb + 1) * TRELLIS_WIDTH, x, y, |
||
918 | mbs_per_slice, td); |
||
919 | } |
||
920 | |||
921 | for (x = ctx->slices_width - 1; x >= 0; x--) { |
||
922 | ctx->slice_q[x + y * ctx->slices_width] = td->nodes[q].quant; |
||
923 | q = td->nodes[q].prev_node; |
||
924 | } |
||
925 | |||
926 | return 0; |
||
927 | } |
||
928 | |||
929 | static int encode_frame(AVCodecContext *avctx, AVPacket *pkt, |
||
930 | const AVFrame *pic, int *got_packet) |
||
931 | { |
||
932 | ProresContext *ctx = avctx->priv_data; |
||
933 | uint8_t *orig_buf, *buf, *slice_hdr, *slice_sizes, *tmp; |
||
934 | uint8_t *picture_size_pos; |
||
935 | PutBitContext pb; |
||
936 | int x, y, i, mb, q = 0; |
||
937 | int sizes[4] = { 0 }; |
||
938 | int slice_hdr_size = 2 + 2 * (ctx->num_planes - 1); |
||
939 | int frame_size, picture_size, slice_size; |
||
940 | int pkt_size, ret; |
||
941 | int max_slice_size = (ctx->frame_size_upper_bound - 200) / (ctx->pictures_per_frame * ctx->slices_per_picture + 1); |
||
942 | uint8_t frame_flags; |
||
943 | |||
944 | ctx->pic = pic; |
||
945 | pkt_size = ctx->frame_size_upper_bound; |
||
946 | |||
947 | if ((ret = ff_alloc_packet2(avctx, pkt, pkt_size + AV_INPUT_BUFFER_MIN_SIZE, 0)) < 0) |
||
948 | return ret; |
||
949 | |||
950 | orig_buf = pkt->data; |
||
951 | |||
952 | // frame atom |
||
953 | orig_buf += 4; // frame size |
||
954 | bytestream_put_be32 (&orig_buf, FRAME_ID); // frame container ID |
||
955 | buf = orig_buf; |
||
956 | |||
957 | // frame header |
||
958 | tmp = buf; |
||
959 | buf += 2; // frame header size will be stored here |
||
960 | bytestream_put_be16 (&buf, 0); // version 1 |
||
961 | bytestream_put_buffer(&buf, ctx->vendor, 4); |
||
962 | bytestream_put_be16 (&buf, avctx->width); |
||
963 | bytestream_put_be16 (&buf, avctx->height); |
||
964 | |||
965 | frame_flags = ctx->chroma_factor << 6; |
||
966 | if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) |
||
967 | frame_flags |= pic->top_field_first ? 0x04 : 0x08; |
||
968 | bytestream_put_byte (&buf, frame_flags); |
||
969 | |||
970 | bytestream_put_byte (&buf, 0); // reserved |
||
971 | bytestream_put_byte (&buf, avctx->color_primaries); |
||
972 | bytestream_put_byte (&buf, avctx->color_trc); |
||
973 | bytestream_put_byte (&buf, avctx->colorspace); |
||
974 | bytestream_put_byte (&buf, 0x40 | (ctx->alpha_bits >> 3)); |
||
975 | bytestream_put_byte (&buf, 0); // reserved |
||
976 | if (ctx->quant_sel != QUANT_MAT_DEFAULT) { |
||
977 | bytestream_put_byte (&buf, 0x03); // matrix flags - both matrices are present |
||
978 | // luma quantisation matrix |
||
979 | for (i = 0; i < 64; i++) |
||
980 | bytestream_put_byte(&buf, ctx->quant_mat[i]); |
||
981 | // chroma quantisation matrix |
||
982 | for (i = 0; i < 64; i++) |
||
983 | bytestream_put_byte(&buf, ctx->quant_mat[i]); |
||
984 | } else { |
||
985 | bytestream_put_byte (&buf, 0x00); // matrix flags - default matrices are used |
||
986 | } |
||
987 | bytestream_put_be16 (&tmp, buf - orig_buf); // write back frame header size |
||
988 | |||
989 | for (ctx->cur_picture_idx = 0; |
||
990 | ctx->cur_picture_idx < ctx->pictures_per_frame; |
||
991 | ctx->cur_picture_idx++) { |
||
992 | // picture header |
||
993 | picture_size_pos = buf + 1; |
||
994 | bytestream_put_byte (&buf, 0x40); // picture header size (in bits) |
||
995 | buf += 4; // picture data size will be stored here |
||
996 | bytestream_put_be16 (&buf, ctx->slices_per_picture); |
||
997 | bytestream_put_byte (&buf, av_log2(ctx->mbs_per_slice) << 4); // slice width and height in MBs |
||
998 | |||
999 | // seek table - will be filled during slice encoding |
||
1000 | slice_sizes = buf; |
||
1001 | buf += ctx->slices_per_picture * 2; |
||
1002 | |||
1003 | // slices |
||
1004 | if (!ctx->force_quant) { |
||
1005 | ret = avctx->execute2(avctx, find_quant_thread, (void*)pic, NULL, |
||
1006 | ctx->mb_height); |
||
1007 | if (ret) |
||
1008 | return ret; |
||
1009 | } |
||
1010 | |||
1011 | for (y = 0; y < ctx->mb_height; y++) { |
||
1012 | int mbs_per_slice = ctx->mbs_per_slice; |
||
1013 | for (x = mb = 0; x < ctx->mb_width; x += mbs_per_slice, mb++) { |
||
1014 | q = ctx->force_quant ? ctx->force_quant |
||
1015 | : ctx->slice_q[mb + y * ctx->slices_width]; |
||
1016 | |||
1017 | while (ctx->mb_width - x < mbs_per_slice) |
||
1018 | mbs_per_slice >>= 1; |
||
1019 | |||
1020 | bytestream_put_byte(&buf, slice_hdr_size << 3); |
||
1021 | slice_hdr = buf; |
||
1022 | buf += slice_hdr_size - 1; |
||
1023 | if (pkt_size <= buf - orig_buf + 2 * max_slice_size) { |
||
1024 | uint8_t *start = pkt->data; |
||
1025 | // Recompute new size according to max_slice_size |
||
1026 | // and deduce delta |
||
1027 | int delta = 200 + (ctx->pictures_per_frame * |
||
1028 | ctx->slices_per_picture + 1) * |
||
1029 | max_slice_size - pkt_size; |
||
1030 | |||
1031 | delta = FFMAX(delta, 2 * max_slice_size); |
||
1032 | ctx->frame_size_upper_bound += delta; |
||
1033 | |||
1034 | if (!ctx->warn) { |
||
1035 | avpriv_request_sample(avctx, |
||
1036 | "Packet too small: is %i," |
||
1037 | " needs %i (slice: %i). " |
||
1038 | "Correct allocation", |
||
1039 | pkt_size, delta, max_slice_size); |
||
1040 | ctx->warn = 1; |
||
1041 | } |
||
1042 | |||
1043 | ret = av_grow_packet(pkt, delta); |
||
1044 | if (ret < 0) |
||
1045 | return ret; |
||
1046 | |||
1047 | pkt_size += delta; |
||
1048 | // restore pointers |
||
1049 | orig_buf = pkt->data + (orig_buf - start); |
||
1050 | buf = pkt->data + (buf - start); |
||
1051 | picture_size_pos = pkt->data + (picture_size_pos - start); |
||
1052 | slice_sizes = pkt->data + (slice_sizes - start); |
||
1053 | slice_hdr = pkt->data + (slice_hdr - start); |
||
1054 | tmp = pkt->data + (tmp - start); |
||
1055 | } |
||
1056 | init_put_bits(&pb, buf, (pkt_size - (buf - orig_buf))); |
||
1057 | ret = encode_slice(avctx, pic, &pb, sizes, x, y, q, |
||
1058 | mbs_per_slice); |
||
1059 | if (ret < 0) |
||
1060 | return ret; |
||
1061 | |||
1062 | bytestream_put_byte(&slice_hdr, q); |
||
1063 | slice_size = slice_hdr_size + sizes[ctx->num_planes - 1]; |
||
1064 | for (i = 0; i < ctx->num_planes - 1; i++) { |
||
1065 | bytestream_put_be16(&slice_hdr, sizes[i]); |
||
1066 | slice_size += sizes[i]; |
||
1067 | } |
||
1068 | bytestream_put_be16(&slice_sizes, slice_size); |
||
1069 | buf += slice_size - slice_hdr_size; |
||
1070 | if (max_slice_size < slice_size) |
||
1071 | max_slice_size = slice_size; |
||
1072 | } |
||
1073 | } |
||
1074 | |||
1075 | picture_size = buf - (picture_size_pos - 1); |
||
1076 | bytestream_put_be32(&picture_size_pos, picture_size); |
||
1077 | } |
||
1078 | |||
1079 | orig_buf -= 8; |
||
1080 | frame_size = buf - orig_buf; |
||
1081 | bytestream_put_be32(&orig_buf, frame_size); |
||
1082 | |||
1083 | pkt->size = frame_size; |
||
1084 | pkt->flags |= AV_PKT_FLAG_KEY; |
||
1085 | *got_packet = 1; |
||
1086 | |||
1087 | return 0; |
||
1088 | } |
||
1089 | |||
1090 | static av_cold int encode_close(AVCodecContext *avctx) |
||
1091 | { |
||
1092 | ProresContext *ctx = avctx->priv_data; |
||
1093 | int i; |
||
1094 | |||
1095 | if (ctx->tdata) { |
||
1096 | for (i = 0; i < avctx->thread_count; i++) |
||
1097 | av_freep(&ctx->tdata[i].nodes); |
||
1098 | } |
||
1099 | av_freep(&ctx->tdata); |
||
1100 | av_freep(&ctx->slice_q); |
||
1101 | |||
1102 | return 0; |
||
1103 | } |
||
1104 | |||
1105 | static void prores_fdct(FDCTDSPContext *fdsp, const uint16_t *src, |
||
1106 | int linesize, int16_t *block) |
||
1107 | { |
||
1108 | int x, y; |
||
1109 | const uint16_t *tsrc = src; |
||
1110 | |||
1111 | for (y = 0; y < 8; y++) { |
||
1112 | for (x = 0; x < 8; x++) |
||
1113 | block[y * 8 + x] = tsrc[x]; |
||
1114 | tsrc += linesize >> 1; |
||
1115 | } |
||
1116 | fdsp->fdct(block); |
||
1117 | } |
||
1118 | |||
1119 | static av_cold int encode_init(AVCodecContext *avctx) |
||
1120 | { |
||
1121 | ProresContext *ctx = avctx->priv_data; |
||
1122 | int mps; |
||
1123 | int i, j; |
||
1124 | int min_quant, max_quant; |
||
1125 | int interlaced = !!(avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT); |
||
1126 | |||
1127 | avctx->bits_per_raw_sample = 10; |
||
1128 | #if FF_API_CODED_FRAME |
||
1129 | FF_DISABLE_DEPRECATION_WARNINGS |
||
1130 | avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I; |
||
1131 | avctx->coded_frame->key_frame = 1; |
||
1132 | FF_ENABLE_DEPRECATION_WARNINGS |
||
1133 | #endif |
||
1134 | |||
1135 | ctx->fdct = prores_fdct; |
||
1136 | ctx->scantable = interlaced ? ff_prores_interlaced_scan |
||
1137 | : ff_prores_progressive_scan; |
||
1138 | ff_fdctdsp_init(&ctx->fdsp, avctx); |
||
1139 | |||
1140 | mps = ctx->mbs_per_slice; |
||
1141 | if (mps & (mps - 1)) { |
||
1142 | av_log(avctx, AV_LOG_ERROR, |
||
1143 | "there should be an integer power of two MBs per slice\n"); |
||
1144 | return AVERROR(EINVAL); |
||
1145 | } |
||
1146 | if (ctx->profile == PRORES_PROFILE_AUTO) { |
||
1147 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt); |
||
1148 | ctx->profile = (desc->flags & AV_PIX_FMT_FLAG_ALPHA || |
||
1149 | !(desc->log2_chroma_w + desc->log2_chroma_h)) |
||
1150 | ? PRORES_PROFILE_4444 : PRORES_PROFILE_HQ; |
||
1151 | av_log(avctx, AV_LOG_INFO, "Autoselected %s. It can be overridden " |
||
1152 | "through -profile option.\n", ctx->profile == PRORES_PROFILE_4444 |
||
1153 | ? "4:4:4:4 profile because of the used input colorspace" |
||
1154 | : "HQ profile to keep best quality"); |
||
1155 | } |
||
1156 | if (av_pix_fmt_desc_get(avctx->pix_fmt)->flags & AV_PIX_FMT_FLAG_ALPHA) { |
||
1157 | if (ctx->profile != PRORES_PROFILE_4444) { |
||
1158 | // force alpha and warn |
||
1159 | av_log(avctx, AV_LOG_WARNING, "Profile selected will not " |
||
1160 | "encode alpha. Override with -profile if needed.\n"); |
||
1161 | ctx->alpha_bits = 0; |
||
1162 | } |
||
1163 | if (ctx->alpha_bits & 7) { |
||
1164 | av_log(avctx, AV_LOG_ERROR, "alpha bits should be 0, 8 or 16\n"); |
||
1165 | return AVERROR(EINVAL); |
||
1166 | } |
||
1167 | avctx->bits_per_coded_sample = 32; |
||
1168 | } else { |
||
1169 | ctx->alpha_bits = 0; |
||
1170 | } |
||
1171 | |||
1172 | ctx->chroma_factor = avctx->pix_fmt == AV_PIX_FMT_YUV422P10 |
||
1173 | ? CFACTOR_Y422 |
||
1174 | : CFACTOR_Y444; |
||
1175 | ctx->profile_info = prores_profile_info + ctx->profile; |
||
1176 | ctx->num_planes = 3 + !!ctx->alpha_bits; |
||
1177 | |||
1178 | ctx->mb_width = FFALIGN(avctx->width, 16) >> 4; |
||
1179 | |||
1180 | if (interlaced) |
||
1181 | ctx->mb_height = FFALIGN(avctx->height, 32) >> 5; |
||
1182 | else |
||
1183 | ctx->mb_height = FFALIGN(avctx->height, 16) >> 4; |
||
1184 | |||
1185 | ctx->slices_width = ctx->mb_width / mps; |
||
1186 | ctx->slices_width += av_popcount(ctx->mb_width - ctx->slices_width * mps); |
||
1187 | ctx->slices_per_picture = ctx->mb_height * ctx->slices_width; |
||
1188 | ctx->pictures_per_frame = 1 + interlaced; |
||
1189 | |||
1190 | if (ctx->quant_sel == -1) |
||
1191 | ctx->quant_mat = prores_quant_matrices[ctx->profile_info->quant]; |
||
1192 | else |
||
1193 | ctx->quant_mat = prores_quant_matrices[ctx->quant_sel]; |
||
1194 | |||
1195 | if (strlen(ctx->vendor) != 4) { |
||
1196 | av_log(avctx, AV_LOG_ERROR, "vendor ID should be 4 bytes\n"); |
||
1197 | return AVERROR_INVALIDDATA; |
||
1198 | } |
||
1199 | |||
1200 | ctx->force_quant = avctx->global_quality / FF_QP2LAMBDA; |
||
1201 | if (!ctx->force_quant) { |
||
1202 | if (!ctx->bits_per_mb) { |
||
1203 | for (i = 0; i < NUM_MB_LIMITS - 1; i++) |
||
1204 | if (prores_mb_limits[i] >= ctx->mb_width * ctx->mb_height * |
||
1205 | ctx->pictures_per_frame) |
||
1206 | break; |
||
1207 | ctx->bits_per_mb = ctx->profile_info->br_tab[i]; |
||
1208 | } else if (ctx->bits_per_mb < 128) { |
||
1209 | av_log(avctx, AV_LOG_ERROR, "too few bits per MB, please set at least 128\n"); |
||
1210 | return AVERROR_INVALIDDATA; |
||
1211 | } |
||
1212 | |||
1213 | min_quant = ctx->profile_info->min_quant; |
||
1214 | max_quant = ctx->profile_info->max_quant; |
||
1215 | for (i = min_quant; i < MAX_STORED_Q; i++) { |
||
1216 | for (j = 0; j < 64; j++) |
||
1217 | ctx->quants[i][j] = ctx->quant_mat[j] * i; |
||
1218 | } |
||
1219 | |||
1220 | ctx->slice_q = av_malloc(ctx->slices_per_picture * sizeof(*ctx->slice_q)); |
||
1221 | if (!ctx->slice_q) { |
||
1222 | encode_close(avctx); |
||
1223 | return AVERROR(ENOMEM); |
||
1224 | } |
||
1225 | |||
1226 | ctx->tdata = av_mallocz(avctx->thread_count * sizeof(*ctx->tdata)); |
||
1227 | if (!ctx->tdata) { |
||
1228 | encode_close(avctx); |
||
1229 | return AVERROR(ENOMEM); |
||
1230 | } |
||
1231 | |||
1232 | for (j = 0; j < avctx->thread_count; j++) { |
||
1233 | ctx->tdata[j].nodes = av_malloc((ctx->slices_width + 1) |
||
1234 | * TRELLIS_WIDTH |
||
1235 | * sizeof(*ctx->tdata->nodes)); |
||
1236 | if (!ctx->tdata[j].nodes) { |
||
1237 | encode_close(avctx); |
||
1238 | return AVERROR(ENOMEM); |
||
1239 | } |
||
1240 | for (i = min_quant; i < max_quant + 2; i++) { |
||
1241 | ctx->tdata[j].nodes[i].prev_node = -1; |
||
1242 | ctx->tdata[j].nodes[i].bits = 0; |
||
1243 | ctx->tdata[j].nodes[i].score = 0; |
||
1244 | } |
||
1245 | } |
||
1246 | } else { |
||
1247 | int ls = 0; |
||
1248 | |||
1249 | if (ctx->force_quant > 64) { |
||
1250 | av_log(avctx, AV_LOG_ERROR, "too large quantiser, maximum is 64\n"); |
||
1251 | return AVERROR_INVALIDDATA; |
||
1252 | } |
||
1253 | |||
1254 | for (j = 0; j < 64; j++) { |
||
1255 | ctx->quants[0][j] = ctx->quant_mat[j] * ctx->force_quant; |
||
1256 | ls += av_log2((1 << 11) / ctx->quants[0][j]) * 2 + 1; |
||
1257 | } |
||
1258 | |||
1259 | ctx->bits_per_mb = ls * 8; |
||
1260 | if (ctx->chroma_factor == CFACTOR_Y444) |
||
1261 | ctx->bits_per_mb += ls * 4; |
||
1262 | } |
||
1263 | |||
1264 | ctx->frame_size_upper_bound = (ctx->pictures_per_frame * |
||
1265 | ctx->slices_per_picture + 1) * |
||
1266 | (2 + 2 * ctx->num_planes + |
||
1267 | (mps * ctx->bits_per_mb) / 8) |
||
1268 | + 200; |
||
1269 | |||
1270 | if (ctx->alpha_bits) { |
||
1271 | // The alpha plane is run-coded and might exceed the bit budget. |
||
1272 | ctx->frame_size_upper_bound += (ctx->pictures_per_frame * |
||
1273 | ctx->slices_per_picture + 1) * |
||
1274 | /* num pixels per slice */ (ctx->mbs_per_slice * 256 * |
||
1275 | /* bits per pixel */ (1 + ctx->alpha_bits + 1) + 7 >> 3); |
||
1276 | } |
||
1277 | |||
1278 | avctx->codec_tag = ctx->profile_info->tag; |
||
1279 | |||
1280 | av_log(avctx, AV_LOG_DEBUG, |
||
1281 | "profile %d, %d slices, interlacing: %s, %d bits per MB\n", |
||
1282 | ctx->profile, ctx->slices_per_picture * ctx->pictures_per_frame, |
||
1283 | interlaced ? "yes" : "no", ctx->bits_per_mb); |
||
1284 | av_log(avctx, AV_LOG_DEBUG, "frame size upper bound: %d\n", |
||
1285 | ctx->frame_size_upper_bound); |
||
1286 | |||
1287 | return 0; |
||
1288 | } |
||
1289 | |||
1290 | #define OFFSET(x) offsetof(ProresContext, x) |
||
1291 | #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM |
||
1292 | |||
1293 | static const AVOption options[] = { |
||
1294 | { "mbs_per_slice", "macroblocks per slice", OFFSET(mbs_per_slice), |
||
1295 | AV_OPT_TYPE_INT, { .i64 = 8 }, 1, MAX_MBS_PER_SLICE, VE }, |
||
1296 | { "profile", NULL, OFFSET(profile), AV_OPT_TYPE_INT, |
||
1297 | { .i64 = PRORES_PROFILE_AUTO }, |
||
1298 | PRORES_PROFILE_AUTO, PRORES_PROFILE_4444, VE, "profile" }, |
||
1299 | { "auto", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_AUTO }, |
||
1300 | 0, 0, VE, "profile" }, |
||
1301 | { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_PROXY }, |
||
1302 | 0, 0, VE, "profile" }, |
||
1303 | { "lt", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_LT }, |
||
1304 | 0, 0, VE, "profile" }, |
||
1305 | { "standard", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_STANDARD }, |
||
1306 | 0, 0, VE, "profile" }, |
||
1307 | { "hq", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_HQ }, |
||
1308 | 0, 0, VE, "profile" }, |
||
1309 | { "4444", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_4444 }, |
||
1310 | 0, 0, VE, "profile" }, |
||
1311 | { "vendor", "vendor ID", OFFSET(vendor), |
||
1312 | AV_OPT_TYPE_STRING, { .str = "Lavc" }, CHAR_MIN, CHAR_MAX, VE }, |
||
1313 | { "bits_per_mb", "desired bits per macroblock", OFFSET(bits_per_mb), |
||
1314 | AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 8192, VE }, |
||
1315 | { "quant_mat", "quantiser matrix", OFFSET(quant_sel), AV_OPT_TYPE_INT, |
||
1316 | { .i64 = -1 }, -1, QUANT_MAT_DEFAULT, VE, "quant_mat" }, |
||
1317 | { "auto", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = -1 }, |
||
1318 | 0, 0, VE, "quant_mat" }, |
||
1319 | { "proxy", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = QUANT_MAT_PROXY }, |
||
1320 | 0, 0, VE, "quant_mat" }, |
||
1321 | { "lt", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = QUANT_MAT_LT }, |
||
1322 | 0, 0, VE, "quant_mat" }, |
||
1323 | { "standard", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = QUANT_MAT_STANDARD }, |
||
1324 | 0, 0, VE, "quant_mat" }, |
||
1325 | { "hq", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = QUANT_MAT_HQ }, |
||
1326 | 0, 0, VE, "quant_mat" }, |
||
1327 | { "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = QUANT_MAT_DEFAULT }, |
||
1328 | 0, 0, VE, "quant_mat" }, |
||
1329 | { "alpha_bits", "bits for alpha plane", OFFSET(alpha_bits), AV_OPT_TYPE_INT, |
||
1330 | { .i64 = 16 }, 0, 16, VE }, |
||
1331 | { NULL } |
||
1332 | }; |
||
1333 | |||
1334 | static const AVClass proresenc_class = { |
||
1335 | .class_name = "ProRes encoder", |
||
1336 | .item_name = av_default_item_name, |
||
1337 | .option = options, |
||
1338 | .version = LIBAVUTIL_VERSION_INT, |
||
1339 | }; |
||
1340 | |||
1341 | AVCodec ff_prores_ks_encoder = { |
||
1342 | .name = "prores_ks", |
||
1343 | .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)"), |
||
1344 | .type = AVMEDIA_TYPE_VIDEO, |
||
1345 | .id = AV_CODEC_ID_PRORES, |
||
1346 | .priv_data_size = sizeof(ProresContext), |
||
1347 | .init = encode_init, |
||
1348 | .close = encode_close, |
||
1349 | .encode2 = encode_frame, |
||
1350 | .capabilities = AV_CODEC_CAP_SLICE_THREADS, |
||
1351 | .pix_fmts = (const enum AVPixelFormat[]) { |
||
1352 | AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10, |
||
1353 | AV_PIX_FMT_YUVA444P10, AV_PIX_FMT_NONE |
||
1354 | }, |
||
1355 | .priv_class = &proresenc_class, |
||
1356 | };><>>>>>>>>>>>>>>>=>><>>>>><>>>>><>>>>=>>>>>>=>>>>>=>>=>>>><>><>><>><>>>>>><>>><>><>><>>>><>>><>><>>>><>><>><>><>>>>>><>>>>><>><>><>>>><>>><>><>><>>>><>>>>>>>=>=>> |