| 1 | /* |
| 2 | * Flash Screen Video Version 2 encoder |
| 3 | * Copyright (C) 2009 Joshua Warner |
| 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 | * Flash Screen Video Version 2 encoder |
| 25 | * @author Joshua Warner |
| 26 | */ |
| 27 | |
| 28 | /* Differences from version 1 stream: |
| 29 | * NOTE: Currently, the only player that supports version 2 streams is Adobe Flash Player itself. |
| 30 | * * Supports sending only a range of scanlines in a block, |
| 31 | * indicating a difference from the corresponding block in the last keyframe. |
| 32 | * * Supports initializing the zlib dictionary with data from the corresponding |
| 33 | * block in the last keyframe, to improve compression. |
| 34 | * * Supports a hybrid 15-bit rgb / 7-bit palette color space. |
| 35 | */ |
| 36 | |
| 37 | /* TODO: |
| 38 | * Don't keep Block structures for both current frame and keyframe. |
| 39 | * Make better heuristics for deciding stream parameters (optimum_* functions). Currently these return constants. |
| 40 | * Figure out how to encode palette information in the stream, choose an optimum palette at each keyframe. |
| 41 | * Figure out how the zlibPrimeCompressCurrent flag works, implement support. |
| 42 | * Find other sample files (that weren't generated here), develop a decoder. |
| 43 | */ |
| 44 | |
| 45 | #include <stdio.h> |
| 46 | #include <stdlib.h> |
| 47 | #include <zlib.h> |
| 48 | |
| 49 | #include "libavutil/imgutils.h" |
| 50 | #include "avcodec.h" |
| 51 | #include "internal.h" |
| 52 | #include "put_bits.h" |
| 53 | #include "bytestream.h" |
| 54 | |
| 55 | #define HAS_IFRAME_IMAGE 0x02 |
| 56 | #define HAS_PALLET_INFO 0x01 |
| 57 | |
| 58 | #define COLORSPACE_BGR 0x00 |
| 59 | #define COLORSPACE_15_7 0x10 |
| 60 | #define HAS_DIFF_BLOCKS 0x04 |
| 61 | #define ZLIB_PRIME_COMPRESS_CURRENT 0x02 |
| 62 | #define ZLIB_PRIME_COMPRESS_PREVIOUS 0x01 |
| 63 | |
| 64 | // Disables experimental "smart" parameter-choosing code, as well as the statistics that it depends on. |
| 65 | // At the moment, the "smart" code is a great example of how the parameters *shouldn't* be chosen. |
| 66 | #define FLASHSV2_DUMB |
| 67 | |
| 68 | typedef struct Block { |
| 69 | uint8_t *enc; |
| 70 | uint8_t *sl_begin, *sl_end; |
| 71 | int enc_size; |
| 72 | uint8_t *data; |
| 73 | unsigned long data_size; |
| 74 | |
| 75 | uint8_t start, len; |
| 76 | uint8_t dirty; |
| 77 | uint8_t col, row, width, height; |
| 78 | uint8_t flags; |
| 79 | } Block; |
| 80 | |
| 81 | typedef struct Palette { |
| 82 | unsigned colors[128]; |
| 83 | uint8_t index[1 << 15]; |
| 84 | } Palette; |
| 85 | |
| 86 | typedef struct FlashSV2Context { |
| 87 | AVCodecContext *avctx; |
| 88 | uint8_t *current_frame; |
| 89 | uint8_t *key_frame; |
| 90 | uint8_t *encbuffer; |
| 91 | uint8_t *keybuffer; |
| 92 | uint8_t *databuffer; |
| 93 | |
| 94 | uint8_t *blockbuffer; |
| 95 | int blockbuffer_size; |
| 96 | |
| 97 | Block *frame_blocks; |
| 98 | Block *key_blocks; |
| 99 | int frame_size; |
| 100 | int blocks_size; |
| 101 | |
| 102 | int use15_7, dist, comp; |
| 103 | |
| 104 | int rows, cols; |
| 105 | |
| 106 | int last_key_frame; |
| 107 | |
| 108 | int image_width, image_height; |
| 109 | int block_width, block_height; |
| 110 | uint8_t flags; |
| 111 | uint8_t use_custom_palette; |
| 112 | uint8_t palette_type; ///< 0=>default, 1=>custom - changed when palette regenerated. |
| 113 | Palette palette; |
| 114 | #ifndef FLASHSV2_DUMB |
| 115 | double tot_blocks; ///< blocks encoded since last keyframe |
| 116 | double diff_blocks; ///< blocks that were different since last keyframe |
| 117 | double tot_lines; ///< total scanlines in image since last keyframe |
| 118 | double diff_lines; ///< scanlines that were different since last keyframe |
| 119 | double raw_size; ///< size of raw frames since last keyframe |
| 120 | double comp_size; ///< size of compressed data since last keyframe |
| 121 | double uncomp_size; ///< size of uncompressed data since last keyframe |
| 122 | |
| 123 | double total_bits; ///< total bits written to stream so far |
| 124 | #endif |
| 125 | } FlashSV2Context; |
| 126 | |
| 127 | static av_cold void cleanup(FlashSV2Context * s) |
| 128 | { |
| 129 | av_freep(&s->encbuffer); |
| 130 | av_freep(&s->keybuffer); |
| 131 | av_freep(&s->databuffer); |
| 132 | av_freep(&s->blockbuffer); |
| 133 | av_freep(&s->current_frame); |
| 134 | av_freep(&s->key_frame); |
| 135 | |
| 136 | av_freep(&s->frame_blocks); |
| 137 | av_freep(&s->key_blocks); |
| 138 | } |
| 139 | |
| 140 | static void init_blocks(FlashSV2Context * s, Block * blocks, |
| 141 | uint8_t * encbuf, uint8_t * databuf) |
| 142 | { |
| 143 | int row, col; |
| 144 | Block *b; |
| 145 | for (col = 0; col < s->cols; col++) { |
| 146 | for (row = 0; row < s->rows; row++) { |
| 147 | b = blocks + (col + row * s->cols); |
| 148 | b->width = (col < s->cols - 1) ? |
| 149 | s->block_width : |
| 150 | s->image_width - col * s->block_width; |
| 151 | |
| 152 | b->height = (row < s->rows - 1) ? |
| 153 | s->block_height : |
| 154 | s->image_height - row * s->block_height; |
| 155 | |
| 156 | b->row = row; |
| 157 | b->col = col; |
| 158 | b->enc = encbuf; |
| 159 | b->data = databuf; |
| 160 | encbuf += b->width * b->height * 3; |
| 161 | databuf += !databuf ? 0 : b->width * b->height * 6; |
| 162 | } |
| 163 | } |
| 164 | } |
| 165 | |
| 166 | static void reset_stats(FlashSV2Context * s) |
| 167 | { |
| 168 | #ifndef FLASHSV2_DUMB |
| 169 | s->diff_blocks = 0.1; |
| 170 | s->tot_blocks = 1; |
| 171 | s->diff_lines = 0.1; |
| 172 | s->tot_lines = 1; |
| 173 | s->raw_size = s->comp_size = s->uncomp_size = 10; |
| 174 | #endif |
| 175 | } |
| 176 | |
| 177 | static av_cold int flashsv2_encode_init(AVCodecContext * avctx) |
| 178 | { |
| 179 | FlashSV2Context *s = avctx->priv_data; |
| 180 | |
| 181 | s->avctx = avctx; |
| 182 | |
| 183 | s->comp = avctx->compression_level; |
| 184 | if (s->comp == -1) |
| 185 | s->comp = 9; |
| 186 | if (s->comp < 0 || s->comp > 9) { |
| 187 | av_log(avctx, AV_LOG_ERROR, |
| 188 | "Compression level should be 0-9, not %d\n", s->comp); |
| 189 | return -1; |
| 190 | } |
| 191 | |
| 192 | |
| 193 | if ((avctx->width > 4095) || (avctx->height > 4095)) { |
| 194 | av_log(avctx, AV_LOG_ERROR, |
| 195 | "Input dimensions too large, input must be max 4095x4095 !\n"); |
| 196 | return -1; |
| 197 | } |
| 198 | if ((avctx->width < 16) || (avctx->height < 16)) { |
| 199 | av_log(avctx, AV_LOG_ERROR, |
| 200 | "Input dimensions too small, input must be at least 16x16 !\n"); |
| 201 | return -1; |
| 202 | } |
| 203 | |
| 204 | if (av_image_check_size(avctx->width, avctx->height, 0, avctx) < 0) |
| 205 | return -1; |
| 206 | |
| 207 | |
| 208 | s->last_key_frame = 0; |
| 209 | |
| 210 | s->image_width = avctx->width; |
| 211 | s->image_height = avctx->height; |
| 212 | |
| 213 | s->block_width = (s->image_width / 12) & ~15; |
| 214 | s->block_height = (s->image_height / 12) & ~15; |
| 215 | |
| 216 | if(!s->block_width) |
| 217 | s->block_width = 1; |
| 218 | if(!s->block_height) |
| 219 | s->block_height = 1; |
| 220 | |
| 221 | s->rows = (s->image_height + s->block_height - 1) / s->block_height; |
| 222 | s->cols = (s->image_width + s->block_width - 1) / s->block_width; |
| 223 | |
| 224 | s->frame_size = s->image_width * s->image_height * 3; |
| 225 | s->blocks_size = s->rows * s->cols * sizeof(Block); |
| 226 | |
| 227 | s->encbuffer = av_mallocz(s->frame_size); |
| 228 | s->keybuffer = av_mallocz(s->frame_size); |
| 229 | s->databuffer = av_mallocz(s->frame_size * 6); |
| 230 | s->current_frame = av_mallocz(s->frame_size); |
| 231 | s->key_frame = av_mallocz(s->frame_size); |
| 232 | s->frame_blocks = av_mallocz(s->blocks_size); |
| 233 | s->key_blocks = av_mallocz(s->blocks_size); |
| 234 | |
| 235 | s->blockbuffer = NULL; |
| 236 | s->blockbuffer_size = 0; |
| 237 | |
| 238 | init_blocks(s, s->frame_blocks, s->encbuffer, s->databuffer); |
| 239 | init_blocks(s, s->key_blocks, s->keybuffer, 0); |
| 240 | reset_stats(s); |
| 241 | #ifndef FLASHSV2_DUMB |
| 242 | s->total_bits = 1; |
| 243 | #endif |
| 244 | |
| 245 | s->use_custom_palette = 0; |
| 246 | s->palette_type = -1; // so that the palette will be generated in reconfigure_at_keyframe |
| 247 | |
| 248 | if (!s->encbuffer || !s->keybuffer || !s->databuffer |
| 249 | || !s->current_frame || !s->key_frame || !s->key_blocks |
| 250 | || !s->frame_blocks) { |
| 251 | av_log(avctx, AV_LOG_ERROR, "Memory allocation failed.\n"); |
| 252 | cleanup(s); |
| 253 | return -1; |
| 254 | } |
| 255 | |
| 256 | return 0; |
| 257 | } |
| 258 | |
| 259 | static int new_key_frame(FlashSV2Context * s) |
| 260 | { |
| 261 | int i; |
| 262 | memcpy(s->key_blocks, s->frame_blocks, s->blocks_size); |
| 263 | memcpy(s->key_frame, s->current_frame, s->frame_size); |
| 264 | |
| 265 | for (i = 0; i < s->rows * s->cols; i++) { |
| 266 | s->key_blocks[i].enc += (s->keybuffer - s->encbuffer); |
| 267 | s->key_blocks[i].sl_begin = 0; |
| 268 | s->key_blocks[i].sl_end = 0; |
| 269 | s->key_blocks[i].data = 0; |
| 270 | } |
| 271 | memcpy(s->keybuffer, s->encbuffer, s->frame_size); |
| 272 | |
| 273 | return 0; |
| 274 | } |
| 275 | |
| 276 | static int write_palette(FlashSV2Context * s, uint8_t * buf, int buf_size) |
| 277 | { |
| 278 | //this isn't implemented yet! Default palette only! |
| 279 | return -1; |
| 280 | } |
| 281 | |
| 282 | static int write_header(FlashSV2Context * s, uint8_t * buf, int buf_size) |
| 283 | { |
| 284 | PutBitContext pb; |
| 285 | int buf_pos, len; |
| 286 | |
| 287 | if (buf_size < 5) |
| 288 | return -1; |
| 289 | |
| 290 | init_put_bits(&pb, buf, buf_size * 8); |
| 291 | |
| 292 | put_bits(&pb, 4, (s->block_width >> 4) - 1); |
| 293 | put_bits(&pb, 12, s->image_width); |
| 294 | put_bits(&pb, 4, (s->block_height >> 4) - 1); |
| 295 | put_bits(&pb, 12, s->image_height); |
| 296 | |
| 297 | flush_put_bits(&pb); |
| 298 | buf_pos = 4; |
| 299 | |
| 300 | buf[buf_pos++] = s->flags; |
| 301 | |
| 302 | if (s->flags & HAS_PALLET_INFO) { |
| 303 | len = write_palette(s, buf + buf_pos, buf_size - buf_pos); |
| 304 | if (len < 0) |
| 305 | return -1; |
| 306 | buf_pos += len; |
| 307 | } |
| 308 | |
| 309 | return buf_pos; |
| 310 | } |
| 311 | |
| 312 | static int write_block(Block * b, uint8_t * buf, int buf_size) |
| 313 | { |
| 314 | int buf_pos = 0; |
| 315 | unsigned block_size = b->data_size; |
| 316 | |
| 317 | if (b->flags & HAS_DIFF_BLOCKS) |
| 318 | block_size += 2; |
| 319 | if (b->flags & ZLIB_PRIME_COMPRESS_CURRENT) |
| 320 | block_size += 2; |
| 321 | if (block_size > 0) |
| 322 | block_size += 1; |
| 323 | if (buf_size < block_size + 2) |
| 324 | return -1; |
| 325 | |
| 326 | buf[buf_pos++] = block_size >> 8; |
| 327 | buf[buf_pos++] = block_size; |
| 328 | |
| 329 | if (block_size == 0) |
| 330 | return buf_pos; |
| 331 | |
| 332 | buf[buf_pos++] = b->flags; |
| 333 | |
| 334 | if (b->flags & HAS_DIFF_BLOCKS) { |
| 335 | buf[buf_pos++] = (b->start); |
| 336 | buf[buf_pos++] = (b->len); |
| 337 | } |
| 338 | |
| 339 | if (b->flags & ZLIB_PRIME_COMPRESS_CURRENT) { |
| 340 | //This feature of the format is poorly understood, and as of now, unused. |
| 341 | buf[buf_pos++] = (b->col); |
| 342 | buf[buf_pos++] = (b->row); |
| 343 | } |
| 344 | |
| 345 | memcpy(buf + buf_pos, b->data, b->data_size); |
| 346 | |
| 347 | buf_pos += b->data_size; |
| 348 | |
| 349 | return buf_pos; |
| 350 | } |
| 351 | |
| 352 | static int encode_zlib(Block * b, uint8_t * buf, unsigned long *buf_size, int comp) |
| 353 | { |
| 354 | int res = compress2(buf, buf_size, b->sl_begin, b->sl_end - b->sl_begin, comp); |
| 355 | return res == Z_OK ? 0 : -1; |
| 356 | } |
| 357 | |
| 358 | static int encode_zlibprime(Block * b, Block * prime, uint8_t * buf, |
| 359 | int *buf_size, int comp) |
| 360 | { |
| 361 | z_stream s; |
| 362 | int res; |
| 363 | s.zalloc = NULL; |
| 364 | s.zfree = NULL; |
| 365 | s.opaque = NULL; |
| 366 | res = deflateInit(&s, comp); |
| 367 | if (res < 0) |
| 368 | return -1; |
| 369 | |
| 370 | s.next_in = prime->enc; |
| 371 | s.avail_in = prime->enc_size; |
| 372 | while (s.avail_in > 0) { |
| 373 | s.next_out = buf; |
| 374 | s.avail_out = *buf_size; |
| 375 | res = deflate(&s, Z_SYNC_FLUSH); |
| 376 | if (res < 0) |
| 377 | return -1; |
| 378 | } |
| 379 | |
| 380 | s.next_in = b->sl_begin; |
| 381 | s.avail_in = b->sl_end - b->sl_begin; |
| 382 | s.next_out = buf; |
| 383 | s.avail_out = *buf_size; |
| 384 | res = deflate(&s, Z_FINISH); |
| 385 | deflateEnd(&s); |
| 386 | *buf_size -= s.avail_out; |
| 387 | if (res != Z_STREAM_END) |
| 388 | return -1; |
| 389 | return 0; |
| 390 | } |
| 391 | |
| 392 | static int encode_bgr(Block * b, const uint8_t * src, int stride) |
| 393 | { |
| 394 | int i; |
| 395 | uint8_t *ptr = b->enc; |
| 396 | for (i = 0; i < b->start; i++) |
| 397 | memcpy(ptr + i * b->width * 3, src + i * stride, b->width * 3); |
| 398 | b->sl_begin = ptr + i * b->width * 3; |
| 399 | for (; i < b->start + b->len; i++) |
| 400 | memcpy(ptr + i * b->width * 3, src + i * stride, b->width * 3); |
| 401 | b->sl_end = ptr + i * b->width * 3; |
| 402 | for (; i < b->height; i++) |
| 403 | memcpy(ptr + i * b->width * 3, src + i * stride, b->width * 3); |
| 404 | b->enc_size = ptr + i * b->width * 3 - b->enc; |
| 405 | return b->enc_size; |
| 406 | } |
| 407 | |
| 408 | static inline unsigned pixel_color15(const uint8_t * src) |
| 409 | { |
| 410 | return (src[0] >> 3) | ((src[1] & 0xf8) << 2) | ((src[2] & 0xf8) << 7); |
| 411 | } |
| 412 | |
| 413 | static inline unsigned int chroma_diff(unsigned int c1, unsigned int c2) |
| 414 | { |
| 415 | unsigned int t1 = (c1 & 0x000000ff) + ((c1 & 0x0000ff00) >> 8) + ((c1 & 0x00ff0000) >> 16); |
| 416 | unsigned int t2 = (c2 & 0x000000ff) + ((c2 & 0x0000ff00) >> 8) + ((c2 & 0x00ff0000) >> 16); |
| 417 | |
| 418 | return abs(t1 - t2) + abs((c1 & 0x000000ff) - (c2 & 0x000000ff)) + |
| 419 | abs(((c1 & 0x0000ff00) >> 8) - ((c2 & 0x0000ff00) >> 8)) + |
| 420 | abs(((c1 & 0x00ff0000) >> 16) - ((c2 & 0x00ff0000) >> 16)); |
| 421 | } |
| 422 | |
| 423 | static inline int pixel_color7_fast(Palette * palette, unsigned c15) |
| 424 | { |
| 425 | return palette->index[c15]; |
| 426 | } |
| 427 | |
| 428 | static int pixel_color7_slow(Palette * palette, unsigned color) |
| 429 | { |
| 430 | int i, min = 0x7fffffff; |
| 431 | int minc = -1; |
| 432 | for (i = 0; i < 128; i++) { |
| 433 | int c1 = palette->colors[i]; |
| 434 | int diff = chroma_diff(c1, color); |
| 435 | if (diff < min) { |
| 436 | min = diff; |
| 437 | minc = i; |
| 438 | } |
| 439 | } |
| 440 | return minc; |
| 441 | } |
| 442 | |
| 443 | static inline unsigned pixel_bgr(const uint8_t * src) |
| 444 | { |
| 445 | return (src[0]) | (src[1] << 8) | (src[2] << 16); |
| 446 | } |
| 447 | |
| 448 | static int write_pixel_15_7(Palette * palette, uint8_t * dest, const uint8_t * src, |
| 449 | int dist) |
| 450 | { |
| 451 | unsigned c15 = pixel_color15(src); |
| 452 | unsigned color = pixel_bgr(src); |
| 453 | int d15 = chroma_diff(color, color & 0x00f8f8f8); |
| 454 | int c7 = pixel_color7_fast(palette, c15); |
| 455 | int d7 = chroma_diff(color, palette->colors[c7]); |
| 456 | if (dist + d15 >= d7) { |
| 457 | dest[0] = c7; |
| 458 | return 1; |
| 459 | } else { |
| 460 | dest[0] = 0x80 | (c15 >> 8); |
| 461 | dest[1] = c15 & 0xff; |
| 462 | return 2; |
| 463 | } |
| 464 | } |
| 465 | |
| 466 | static int update_palette_index(Palette * palette) |
| 467 | { |
| 468 | int r, g, b; |
| 469 | unsigned int bgr, c15, index; |
| 470 | for (r = 4; r < 256; r += 8) { |
| 471 | for (g = 4; g < 256; g += 8) { |
| 472 | for (b = 4; b < 256; b += 8) { |
| 473 | bgr = b | (g << 8) | (r << 16); |
| 474 | c15 = (b >> 3) | ((g & 0xf8) << 2) | ((r & 0xf8) << 7); |
| 475 | index = pixel_color7_slow(palette, bgr); |
| 476 | |
| 477 | palette->index[c15] = index; |
| 478 | } |
| 479 | } |
| 480 | } |
| 481 | return 0; |
| 482 | } |
| 483 | |
| 484 | static const unsigned int default_screen_video_v2_palette[128] = { |
| 485 | 0x00000000, 0x00333333, 0x00666666, 0x00999999, 0x00CCCCCC, 0x00FFFFFF, |
| 486 | 0x00330000, 0x00660000, 0x00990000, 0x00CC0000, 0x00FF0000, 0x00003300, |
| 487 | 0x00006600, 0x00009900, 0x0000CC00, 0x0000FF00, 0x00000033, 0x00000066, |
| 488 | 0x00000099, 0x000000CC, 0x000000FF, 0x00333300, 0x00666600, 0x00999900, |
| 489 | 0x00CCCC00, 0x00FFFF00, 0x00003333, 0x00006666, 0x00009999, 0x0000CCCC, |
| 490 | 0x0000FFFF, 0x00330033, 0x00660066, 0x00990099, 0x00CC00CC, 0x00FF00FF, |
| 491 | 0x00FFFF33, 0x00FFFF66, 0x00FFFF99, 0x00FFFFCC, 0x00FF33FF, 0x00FF66FF, |
| 492 | 0x00FF99FF, 0x00FFCCFF, 0x0033FFFF, 0x0066FFFF, 0x0099FFFF, 0x00CCFFFF, |
| 493 | 0x00CCCC33, 0x00CCCC66, 0x00CCCC99, 0x00CCCCFF, 0x00CC33CC, 0x00CC66CC, |
| 494 | 0x00CC99CC, 0x00CCFFCC, 0x0033CCCC, 0x0066CCCC, 0x0099CCCC, 0x00FFCCCC, |
| 495 | 0x00999933, 0x00999966, 0x009999CC, 0x009999FF, 0x00993399, 0x00996699, |
| 496 | 0x0099CC99, 0x0099FF99, 0x00339999, 0x00669999, 0x00CC9999, 0x00FF9999, |
| 497 | 0x00666633, 0x00666699, 0x006666CC, 0x006666FF, 0x00663366, 0x00669966, |
| 498 | 0x0066CC66, 0x0066FF66, 0x00336666, 0x00996666, 0x00CC6666, 0x00FF6666, |
| 499 | 0x00333366, 0x00333399, 0x003333CC, 0x003333FF, 0x00336633, 0x00339933, |
| 500 | 0x0033CC33, 0x0033FF33, 0x00663333, 0x00993333, 0x00CC3333, 0x00FF3333, |
| 501 | 0x00003366, 0x00336600, 0x00660033, 0x00006633, 0x00330066, 0x00663300, |
| 502 | 0x00336699, 0x00669933, 0x00993366, 0x00339966, 0x00663399, 0x00996633, |
| 503 | 0x006699CC, 0x0099CC66, 0x00CC6699, 0x0066CC99, 0x009966CC, 0x00CC9966, |
| 504 | 0x0099CCFF, 0x00CCFF99, 0x00FF99CC, 0x0099FFCC, 0x00CC99FF, 0x00FFCC99, |
| 505 | 0x00111111, 0x00222222, 0x00444444, 0x00555555, 0x00AAAAAA, 0x00BBBBBB, |
| 506 | 0x00DDDDDD, 0x00EEEEEE |
| 507 | }; |
| 508 | |
| 509 | static int generate_default_palette(Palette * palette) |
| 510 | { |
| 511 | memcpy(palette->colors, default_screen_video_v2_palette, |
| 512 | sizeof(default_screen_video_v2_palette)); |
| 513 | |
| 514 | return update_palette_index(palette); |
| 515 | } |
| 516 | |
| 517 | static int generate_optimum_palette(Palette * palette, const uint8_t * image, |
| 518 | int width, int height, int stride) |
| 519 | { |
| 520 | //this isn't implemented yet! Default palette only! |
| 521 | return -1; |
| 522 | } |
| 523 | |
| 524 | static inline int encode_15_7_sl(Palette * palette, uint8_t * dest, |
| 525 | const uint8_t * src, int width, int dist) |
| 526 | { |
| 527 | int len = 0, x; |
| 528 | for (x = 0; x < width; x++) { |
| 529 | len += write_pixel_15_7(palette, dest + len, src + 3 * x, dist); |
| 530 | } |
| 531 | return len; |
| 532 | } |
| 533 | |
| 534 | static int encode_15_7(Palette * palette, Block * b, const uint8_t * src, |
| 535 | int stride, int dist) |
| 536 | { |
| 537 | int i; |
| 538 | uint8_t *ptr = b->enc; |
| 539 | for (i = 0; i < b->start; i++) |
| 540 | ptr += encode_15_7_sl(palette, ptr, src + i * stride, b->width, dist); |
| 541 | b->sl_begin = ptr; |
| 542 | for (; i < b->start + b->len; i++) |
| 543 | ptr += encode_15_7_sl(palette, ptr, src + i * stride, b->width, dist); |
| 544 | b->sl_end = ptr; |
| 545 | for (; i < b->height; i++) |
| 546 | ptr += encode_15_7_sl(palette, ptr, src + i * stride, b->width, dist); |
| 547 | b->enc_size = ptr - b->enc; |
| 548 | return b->enc_size; |
| 549 | } |
| 550 | |
| 551 | static int encode_block(FlashSV2Context *s, Palette * palette, Block * b, |
| 552 | Block * prev, const uint8_t * src, int stride, int comp, |
| 553 | int dist, int keyframe) |
| 554 | { |
| 555 | unsigned buf_size = b->width * b->height * 6; |
| 556 | uint8_t *buf = s->blockbuffer; |
| 557 | int res; |
| 558 | |
| 559 | if (b->flags & COLORSPACE_15_7) { |
| 560 | encode_15_7(palette, b, src, stride, dist); |
| 561 | } else { |
| 562 | encode_bgr(b, src, stride); |
| 563 | } |
| 564 | |
| 565 | if (b->len > 0) { |
| 566 | b->data_size = buf_size; |
| 567 | res = encode_zlib(b, b->data, &b->data_size, comp); |
| 568 | if (res) |
| 569 | return res; |
| 570 | |
| 571 | if (!keyframe) { |
| 572 | res = encode_zlibprime(b, prev, buf, &buf_size, comp); |
| 573 | if (res) |
| 574 | return res; |
| 575 | |
| 576 | if (buf_size < b->data_size) { |
| 577 | b->data_size = buf_size; |
| 578 | memcpy(b->data, buf, buf_size); |
| 579 | b->flags |= ZLIB_PRIME_COMPRESS_PREVIOUS; |
| 580 | } |
| 581 | } |
| 582 | } else { |
| 583 | b->data_size = 0; |
| 584 | } |
| 585 | return 0; |
| 586 | } |
| 587 | |
| 588 | static int compare_sl(FlashSV2Context * s, Block * b, const uint8_t * src, |
| 589 | uint8_t * frame, uint8_t * key, int y, int keyframe) |
| 590 | { |
| 591 | if (memcmp(src, frame, b->width * 3) != 0) { |
| 592 | b->dirty = 1; |
| 593 | memcpy(frame, src, b->width * 3); |
| 594 | #ifndef FLASHSV2_DUMB |
| 595 | s->diff_lines++; |
| 596 | #endif |
| 597 | } |
| 598 | if (memcmp(src, key, b->width * 3) != 0) { |
| 599 | if (b->len == 0) |
| 600 | b->start = y; |
| 601 | b->len = y + 1 - b->start; |
| 602 | } |
| 603 | return 0; |
| 604 | } |
| 605 | |
| 606 | static int mark_all_blocks(FlashSV2Context * s, const uint8_t * src, int stride, |
| 607 | int keyframe) |
| 608 | { |
| 609 | int sl, rsl, col, pos, possl; |
| 610 | Block *b; |
| 611 | for (sl = s->image_height - 1; sl >= 0; sl--) { |
| 612 | for (col = 0; col < s->cols; col++) { |
| 613 | rsl = s->image_height - sl - 1; |
| 614 | b = s->frame_blocks + col + rsl / s->block_height * s->cols; |
| 615 | possl = stride * sl + col * s->block_width * 3; |
| 616 | pos = s->image_width * rsl * 3 + col * s->block_width * 3; |
| 617 | compare_sl(s, b, src + possl, s->current_frame + pos, |
| 618 | s->key_frame + pos, rsl % s->block_height, keyframe); |
| 619 | } |
| 620 | } |
| 621 | #ifndef FLASHSV2_DUMB |
| 622 | s->tot_lines += s->image_height * s->cols; |
| 623 | #endif |
| 624 | return 0; |
| 625 | } |
| 626 | |
| 627 | static int encode_all_blocks(FlashSV2Context * s, int keyframe) |
| 628 | { |
| 629 | int row, col, res; |
| 630 | uint8_t *data; |
| 631 | Block *b, *prev; |
| 632 | for (row = 0; row < s->rows; row++) { |
| 633 | for (col = 0; col < s->cols; col++) { |
| 634 | b = s->frame_blocks + (row * s->cols + col); |
| 635 | prev = s->key_blocks + (row * s->cols + col); |
| 636 | b->flags = s->use15_7 ? COLORSPACE_15_7 : 0; |
| 637 | if (keyframe) { |
| 638 | b->start = 0; |
| 639 | b->len = b->height; |
| 640 | } else if (!b->dirty) { |
| 641 | b->start = 0; |
| 642 | b->len = 0; |
| 643 | b->data_size = 0; |
| 644 | continue; |
| 645 | } else if (b->start != 0 || b->len != b->height) { |
| 646 | b->flags |= HAS_DIFF_BLOCKS; |
| 647 | } |
| 648 | data = s->current_frame + s->image_width * 3 * s->block_height * row + s->block_width * col * 3; |
| 649 | res = encode_block(s, &s->palette, b, prev, data, s->image_width * 3, s->comp, s->dist, keyframe); |
| 650 | #ifndef FLASHSV2_DUMB |
| 651 | if (b->dirty) |
| 652 | s->diff_blocks++; |
| 653 | s->comp_size += b->data_size; |
| 654 | s->uncomp_size += b->enc_size; |
| 655 | #endif |
| 656 | if (res) |
| 657 | return res; |
| 658 | } |
| 659 | } |
| 660 | #ifndef FLASHSV2_DUMB |
| 661 | s->raw_size += s->image_width * s->image_height * 3; |
| 662 | s->tot_blocks += s->rows * s->cols; |
| 663 | #endif |
| 664 | return 0; |
| 665 | } |
| 666 | |
| 667 | static int write_all_blocks(FlashSV2Context * s, uint8_t * buf, |
| 668 | int buf_size) |
| 669 | { |
| 670 | int row, col, buf_pos = 0, len; |
| 671 | Block *b; |
| 672 | for (row = 0; row < s->rows; row++) { |
| 673 | for (col = 0; col < s->cols; col++) { |
| 674 | b = s->frame_blocks + row * s->cols + col; |
| 675 | len = write_block(b, buf + buf_pos, buf_size - buf_pos); |
| 676 | b->start = b->len = b->dirty = 0; |
| 677 | if (len < 0) |
| 678 | return len; |
| 679 | buf_pos += len; |
| 680 | } |
| 681 | } |
| 682 | return buf_pos; |
| 683 | } |
| 684 | |
| 685 | static int write_bitstream(FlashSV2Context * s, const uint8_t * src, int stride, |
| 686 | uint8_t * buf, int buf_size, int keyframe) |
| 687 | { |
| 688 | int buf_pos, res; |
| 689 | |
| 690 | res = mark_all_blocks(s, src, stride, keyframe); |
| 691 | if (res) |
| 692 | return res; |
| 693 | res = encode_all_blocks(s, keyframe); |
| 694 | if (res) |
| 695 | return res; |
| 696 | |
| 697 | res = write_header(s, buf, buf_size); |
| 698 | if (res < 0) { |
| 699 | return res; |
| 700 | } else { |
| 701 | buf_pos = res; |
| 702 | } |
| 703 | res = write_all_blocks(s, buf + buf_pos, buf_size - buf_pos); |
| 704 | if (res < 0) |
| 705 | return res; |
| 706 | buf_pos += res; |
| 707 | #ifndef FLASHSV2_DUMB |
| 708 | s->total_bits += ((double) buf_pos) * 8.0; |
| 709 | #endif |
| 710 | |
| 711 | return buf_pos; |
| 712 | } |
| 713 | |
| 714 | static void recommend_keyframe(FlashSV2Context * s, int *keyframe) |
| 715 | { |
| 716 | #ifndef FLASHSV2_DUMB |
| 717 | double block_ratio, line_ratio, enc_ratio, comp_ratio, data_ratio; |
| 718 | if (s->avctx->gop_size > 0) { |
| 719 | block_ratio = s->diff_blocks / s->tot_blocks; |
| 720 | line_ratio = s->diff_lines / s->tot_lines; |
| 721 | enc_ratio = s->uncomp_size / s->raw_size; |
| 722 | comp_ratio = s->comp_size / s->uncomp_size; |
| 723 | data_ratio = s->comp_size / s->raw_size; |
| 724 | |
| 725 | if ((block_ratio >= 0.5 && line_ratio / block_ratio <= 0.5) || line_ratio >= 0.95) { |
| 726 | *keyframe = 1; |
| 727 | return; |
| 728 | } |
| 729 | } |
| 730 | #else |
| 731 | return; |
| 732 | #endif |
| 733 | } |
| 734 | |
| 735 | #ifndef FLASHSV2_DUMB |
| 736 | static const double block_size_fraction = 1.0 / 300; |
| 737 | static const double use15_7_threshold = 8192; |
| 738 | static const double color15_7_factor = 100; |
| 739 | #endif |
| 740 | static int optimum_block_width(FlashSV2Context * s) |
| 741 | { |
| 742 | #ifndef FLASHSV2_DUMB |
| 743 | double save = (1-pow(s->diff_lines/s->diff_blocks/s->block_height, 0.5)) * s->comp_size/s->tot_blocks; |
| 744 | double width = block_size_fraction * sqrt(0.5 * save * s->rows * s->cols) * s->image_width; |
| 745 | int pwidth = ((int) width); |
| 746 | return FFCLIP(pwidth & ~15, 256, 16); |
| 747 | #else |
| 748 | return 64; |
| 749 | #endif |
| 750 | } |
| 751 | |
| 752 | static int optimum_block_height(FlashSV2Context * s) |
| 753 | { |
| 754 | #ifndef FLASHSV2_DUMB |
| 755 | double save = (1-pow(s->diff_lines/s->diff_blocks/s->block_height, 0.5)) * s->comp_size/s->tot_blocks; |
| 756 | double height = block_size_fraction * sqrt(0.5 * save * s->rows * s->cols) * s->image_height; |
| 757 | int pheight = ((int) height); |
| 758 | return FFCLIP(pheight & ~15, 256, 16); |
| 759 | #else |
| 760 | return 64; |
| 761 | #endif |
| 762 | } |
| 763 | |
| 764 | static int optimum_use15_7(FlashSV2Context * s) |
| 765 | { |
| 766 | #ifndef FLASHSV2_DUMB |
| 767 | double ideal = ((double)(s->avctx->bit_rate * s->avctx->time_base.den * s->avctx->ticks_per_frame)) / |
| 768 | ((double) s->avctx->time_base.num) * s->avctx->frame_number; |
| 769 | if (ideal + use15_7_threshold < s->total_bits) { |
| 770 | return 1; |
| 771 | } else { |
| 772 | return 0; |
| 773 | } |
| 774 | #else |
| 775 | return s->avctx->global_quality == 0; |
| 776 | #endif |
| 777 | } |
| 778 | |
| 779 | static int optimum_dist(FlashSV2Context * s) |
| 780 | { |
| 781 | #ifndef FLASHSV2_DUMB |
| 782 | double ideal = |
| 783 | s->avctx->bit_rate * s->avctx->time_base.den * |
| 784 | s->avctx->ticks_per_frame; |
| 785 | int dist = pow((s->total_bits / ideal) * color15_7_factor, 3); |
| 786 | av_log(s->avctx, AV_LOG_DEBUG, "dist: %d\n", dist); |
| 787 | return dist; |
| 788 | #else |
| 789 | return 15; |
| 790 | #endif |
| 791 | } |
| 792 | |
| 793 | |
| 794 | static int reconfigure_at_keyframe(FlashSV2Context * s, const uint8_t * image, |
| 795 | int stride) |
| 796 | { |
| 797 | int update_palette = 0; |
| 798 | int res; |
| 799 | int block_width = optimum_block_width (s); |
| 800 | int block_height = optimum_block_height(s); |
| 801 | |
| 802 | s->rows = (s->image_height + block_height - 1) / block_height; |
| 803 | s->cols = (s->image_width + block_width - 1) / block_width; |
| 804 | |
| 805 | if (block_width != s->block_width || block_height != s->block_height) { |
| 806 | s->block_width = block_width; |
| 807 | s->block_height = block_height; |
| 808 | if (s->rows * s->cols > s->blocks_size / sizeof(Block)) { |
| 809 | s->frame_blocks = av_realloc(s->frame_blocks, s->rows * s->cols * sizeof(Block)); |
| 810 | s->key_blocks = av_realloc(s->key_blocks, s->cols * s->rows * sizeof(Block)); |
| 811 | if (!s->frame_blocks || !s->key_blocks) { |
| 812 | av_log(s->avctx, AV_LOG_ERROR, "Memory allocation failed.\n"); |
| 813 | return -1; |
| 814 | } |
| 815 | s->blocks_size = s->rows * s->cols * sizeof(Block); |
| 816 | } |
| 817 | init_blocks(s, s->frame_blocks, s->encbuffer, s->databuffer); |
| 818 | init_blocks(s, s->key_blocks, s->keybuffer, 0); |
| 819 | |
| 820 | av_fast_malloc(&s->blockbuffer, &s->blockbuffer_size, block_width * block_height * 6); |
| 821 | if (!s->blockbuffer) { |
| 822 | av_log(s->avctx, AV_LOG_ERROR, "Could not allocate block buffer.\n"); |
| 823 | return AVERROR(ENOMEM); |
| 824 | } |
| 825 | } |
| 826 | |
| 827 | s->use15_7 = optimum_use15_7(s); |
| 828 | if (s->use15_7) { |
| 829 | if ((s->use_custom_palette && s->palette_type != 1) || update_palette) { |
| 830 | res = generate_optimum_palette(&s->palette, image, s->image_width, s->image_height, stride); |
| 831 | if (res) |
| 832 | return res; |
| 833 | s->palette_type = 1; |
| 834 | av_log(s->avctx, AV_LOG_DEBUG, "Generated optimum palette\n"); |
| 835 | } else if (!s->use_custom_palette && s->palette_type != 0) { |
| 836 | res = generate_default_palette(&s->palette); |
| 837 | if (res) |
| 838 | return res; |
| 839 | s->palette_type = 0; |
| 840 | av_log(s->avctx, AV_LOG_DEBUG, "Generated default palette\n"); |
| 841 | } |
| 842 | } |
| 843 | |
| 844 | |
| 845 | reset_stats(s); |
| 846 | |
| 847 | return 0; |
| 848 | } |
| 849 | |
| 850 | static int flashsv2_encode_frame(AVCodecContext *avctx, AVPacket *pkt, |
| 851 | const AVFrame *p, int *got_packet) |
| 852 | { |
| 853 | FlashSV2Context *const s = avctx->priv_data; |
| 854 | int res; |
| 855 | int keyframe = 0; |
| 856 | |
| 857 | if ((res = ff_alloc_packet2(avctx, pkt, s->frame_size + FF_MIN_BUFFER_SIZE)) < 0) |
| 858 | return res; |
| 859 | |
| 860 | /* First frame needs to be a keyframe */ |
| 861 | if (avctx->frame_number == 0) |
| 862 | keyframe = 1; |
| 863 | |
| 864 | /* Check the placement of keyframes */ |
| 865 | if (avctx->gop_size > 0) { |
| 866 | if (avctx->frame_number >= s->last_key_frame + avctx->gop_size) |
| 867 | keyframe = 1; |
| 868 | } |
| 869 | |
| 870 | if (!keyframe |
| 871 | && avctx->frame_number > s->last_key_frame + avctx->keyint_min) { |
| 872 | recommend_keyframe(s, &keyframe); |
| 873 | if (keyframe) |
| 874 | av_log(avctx, AV_LOG_DEBUG, "Recommending key frame at frame %d\n", avctx->frame_number); |
| 875 | } |
| 876 | |
| 877 | if (keyframe) { |
| 878 | res = reconfigure_at_keyframe(s, p->data[0], p->linesize[0]); |
| 879 | if (res) |
| 880 | return res; |
| 881 | } |
| 882 | |
| 883 | if (s->use15_7) |
| 884 | s->dist = optimum_dist(s); |
| 885 | |
| 886 | res = write_bitstream(s, p->data[0], p->linesize[0], pkt->data, pkt->size, keyframe); |
| 887 | |
| 888 | if (keyframe) { |
| 889 | new_key_frame(s); |
| 890 | s->last_key_frame = avctx->frame_number; |
| 891 | pkt->flags |= AV_PKT_FLAG_KEY; |
| 892 | av_log(avctx, AV_LOG_DEBUG, "Inserting key frame at frame %d\n", avctx->frame_number); |
| 893 | } |
| 894 | |
| 895 | pkt->size = res; |
| 896 | *got_packet = 1; |
| 897 | |
| 898 | return 0; |
| 899 | } |
| 900 | |
| 901 | static av_cold int flashsv2_encode_end(AVCodecContext * avctx) |
| 902 | { |
| 903 | FlashSV2Context *s = avctx->priv_data; |
| 904 | |
| 905 | cleanup(s); |
| 906 | |
| 907 | return 0; |
| 908 | } |
| 909 | |
| 910 | AVCodec ff_flashsv2_encoder = { |
| 911 | .name = "flashsv2", |
| 912 | .long_name = NULL_IF_CONFIG_SMALL("Flash Screen Video Version 2"), |
| 913 | .type = AVMEDIA_TYPE_VIDEO, |
| 914 | .id = AV_CODEC_ID_FLASHSV2, |
| 915 | .priv_data_size = sizeof(FlashSV2Context), |
| 916 | .init = flashsv2_encode_init, |
| 917 | .encode2 = flashsv2_encode_frame, |
| 918 | .close = flashsv2_encode_end, |
| 919 | .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_BGR24, AV_PIX_FMT_NONE }, |
| 920 | }; |