You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1140 lines
41KB

  1. /*
  2. * Indeo Video v3 compatible decoder
  3. * Copyright (c) 2009 - 2011 Maxim Poliakovski
  4. *
  5. * This file is part of Libav.
  6. *
  7. * Libav 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. * Libav 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 Libav; if not, write to the Free Software
  19. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20. */
  21. /**
  22. * @file
  23. * This is a decoder for Intel Indeo Video v3.
  24. * It is based on vector quantization, run-length coding and motion compensation.
  25. * Known container formats: .avi and .mov
  26. * Known FOURCCs: 'IV31', 'IV32'
  27. *
  28. * @see http://wiki.multimedia.cx/index.php?title=Indeo_3
  29. */
  30. #include "libavutil/imgutils.h"
  31. #include "libavutil/intreadwrite.h"
  32. #include "avcodec.h"
  33. #include "bytestream.h"
  34. #include "get_bits.h"
  35. #include "hpeldsp.h"
  36. #include "internal.h"
  37. #include "indeo3data.h"
  38. /* RLE opcodes. */
  39. enum {
  40. RLE_ESC_F9 = 249, ///< same as RLE_ESC_FA + do the same with next block
  41. RLE_ESC_FA = 250, ///< INTRA: skip block, INTER: copy data from reference
  42. RLE_ESC_FB = 251, ///< apply null delta to N blocks / skip N blocks
  43. RLE_ESC_FC = 252, ///< same as RLE_ESC_FD + do the same with next block
  44. RLE_ESC_FD = 253, ///< apply null delta to all remaining lines of this block
  45. RLE_ESC_FE = 254, ///< apply null delta to all lines up to the 3rd line
  46. RLE_ESC_FF = 255 ///< apply null delta to all lines up to the 2nd line
  47. };
  48. /* Some constants for parsing frame bitstream flags. */
  49. #define BS_8BIT_PEL (1 << 1) ///< 8bit pixel bitdepth indicator
  50. #define BS_KEYFRAME (1 << 2) ///< intra frame indicator
  51. #define BS_MV_Y_HALF (1 << 4) ///< vertical mv halfpel resolution indicator
  52. #define BS_MV_X_HALF (1 << 5) ///< horizontal mv halfpel resolution indicator
  53. #define BS_NONREF (1 << 8) ///< nonref (discardable) frame indicator
  54. #define BS_BUFFER 9 ///< indicates which of two frame buffers should be used
  55. typedef struct Plane {
  56. uint8_t *buffers[2];
  57. uint8_t *pixels[2]; ///< pointer to the actual pixel data of the buffers above
  58. uint32_t width;
  59. uint32_t height;
  60. uint32_t pitch;
  61. } Plane;
  62. #define CELL_STACK_MAX 20
  63. typedef struct Cell {
  64. int16_t xpos; ///< cell coordinates in 4x4 blocks
  65. int16_t ypos;
  66. int16_t width; ///< cell width in 4x4 blocks
  67. int16_t height; ///< cell height in 4x4 blocks
  68. uint8_t tree; ///< tree id: 0- MC tree, 1 - VQ tree
  69. const int8_t *mv_ptr; ///< ptr to the motion vector if any
  70. } Cell;
  71. typedef struct Indeo3DecodeContext {
  72. AVCodecContext *avctx;
  73. HpelDSPContext hdsp;
  74. GetBitContext gb;
  75. int need_resync;
  76. int skip_bits;
  77. const uint8_t *next_cell_data;
  78. const uint8_t *last_byte;
  79. const int8_t *mc_vectors;
  80. unsigned num_vectors; ///< number of motion vectors in mc_vectors
  81. int16_t width, height;
  82. uint32_t frame_num; ///< current frame number (zero-based)
  83. uint32_t data_size; ///< size of the frame data in bytes
  84. uint16_t frame_flags; ///< frame properties
  85. uint8_t cb_offset; ///< needed for selecting VQ tables
  86. uint8_t buf_sel; ///< active frame buffer: 0 - primary, 1 -secondary
  87. const uint8_t *y_data_ptr;
  88. const uint8_t *v_data_ptr;
  89. const uint8_t *u_data_ptr;
  90. int32_t y_data_size;
  91. int32_t v_data_size;
  92. int32_t u_data_size;
  93. const uint8_t *alt_quant; ///< secondary VQ table set for the modes 1 and 4
  94. Plane planes[3];
  95. } Indeo3DecodeContext;
  96. static uint8_t requant_tab[8][128];
  97. /*
  98. * Build the static requantization table.
  99. * This table is used to remap pixel values according to a specific
  100. * quant index and thus avoid overflows while adding deltas.
  101. */
  102. static av_cold void build_requant_tab(void)
  103. {
  104. static int8_t offsets[8] = { 1, 1, 2, -3, -3, 3, 4, 4 };
  105. static int8_t deltas [8] = { 0, 1, 0, 4, 4, 1, 0, 1 };
  106. int i, j, step;
  107. for (i = 0; i < 8; i++) {
  108. step = i + 2;
  109. for (j = 0; j < 128; j++)
  110. requant_tab[i][j] = (j + offsets[i]) / step * step + deltas[i];
  111. }
  112. /* some last elements calculated above will have values >= 128 */
  113. /* pixel values shall never exceed 127 so set them to non-overflowing values */
  114. /* according with the quantization step of the respective section */
  115. requant_tab[0][127] = 126;
  116. requant_tab[1][119] = 118;
  117. requant_tab[1][120] = 118;
  118. requant_tab[2][126] = 124;
  119. requant_tab[2][127] = 124;
  120. requant_tab[6][124] = 120;
  121. requant_tab[6][125] = 120;
  122. requant_tab[6][126] = 120;
  123. requant_tab[6][127] = 120;
  124. /* Patch for compatibility with the Intel's binary decoders */
  125. requant_tab[1][7] = 10;
  126. requant_tab[4][8] = 10;
  127. }
  128. static av_cold int allocate_frame_buffers(Indeo3DecodeContext *ctx,
  129. AVCodecContext *avctx)
  130. {
  131. int p, luma_width, luma_height, chroma_width, chroma_height;
  132. int luma_pitch, chroma_pitch, luma_size, chroma_size;
  133. luma_width = ctx->width;
  134. luma_height = ctx->height;
  135. if (luma_width < 16 || luma_width > 640 ||
  136. luma_height < 16 || luma_height > 480 ||
  137. luma_width & 3 || luma_height & 3) {
  138. av_log(avctx, AV_LOG_ERROR, "Invalid picture dimensions: %d x %d!\n",
  139. luma_width, luma_height);
  140. return AVERROR_INVALIDDATA;
  141. }
  142. chroma_width = FFALIGN(luma_width >> 2, 4);
  143. chroma_height = FFALIGN(luma_height >> 2, 4);
  144. luma_pitch = FFALIGN(luma_width, 16);
  145. chroma_pitch = FFALIGN(chroma_width, 16);
  146. /* Calculate size of the luminance plane. */
  147. /* Add one line more for INTRA prediction. */
  148. luma_size = luma_pitch * (luma_height + 1);
  149. /* Calculate size of a chrominance planes. */
  150. /* Add one line more for INTRA prediction. */
  151. chroma_size = chroma_pitch * (chroma_height + 1);
  152. /* allocate frame buffers */
  153. for (p = 0; p < 3; p++) {
  154. ctx->planes[p].pitch = !p ? luma_pitch : chroma_pitch;
  155. ctx->planes[p].width = !p ? luma_width : chroma_width;
  156. ctx->planes[p].height = !p ? luma_height : chroma_height;
  157. ctx->planes[p].buffers[0] = av_malloc(!p ? luma_size : chroma_size);
  158. ctx->planes[p].buffers[1] = av_malloc(!p ? luma_size : chroma_size);
  159. /* fill the INTRA prediction lines with the middle pixel value = 64 */
  160. memset(ctx->planes[p].buffers[0], 0x40, ctx->planes[p].pitch);
  161. memset(ctx->planes[p].buffers[1], 0x40, ctx->planes[p].pitch);
  162. /* set buffer pointers = buf_ptr + pitch and thus skip the INTRA prediction line */
  163. ctx->planes[p].pixels[0] = ctx->planes[p].buffers[0] + ctx->planes[p].pitch;
  164. ctx->planes[p].pixels[1] = ctx->planes[p].buffers[1] + ctx->planes[p].pitch;
  165. memset(ctx->planes[p].pixels[0], 0, ctx->planes[p].pitch * ctx->planes[p].height);
  166. memset(ctx->planes[p].pixels[1], 0, ctx->planes[p].pitch * ctx->planes[p].height);
  167. }
  168. return 0;
  169. }
  170. static av_cold void free_frame_buffers(Indeo3DecodeContext *ctx)
  171. {
  172. int p;
  173. for (p = 0; p < 3; p++) {
  174. av_freep(&ctx->planes[p].buffers[0]);
  175. av_freep(&ctx->planes[p].buffers[1]);
  176. ctx->planes[p].pixels[0] = ctx->planes[p].pixels[1] = 0;
  177. }
  178. }
  179. /**
  180. * Copy pixels of the cell(x + mv_x, y + mv_y) from the previous frame into
  181. * the cell(x, y) in the current frame.
  182. *
  183. * @param ctx pointer to the decoder context
  184. * @param plane pointer to the plane descriptor
  185. * @param cell pointer to the cell descriptor
  186. */
  187. static int copy_cell(Indeo3DecodeContext *ctx, Plane *plane, Cell *cell)
  188. {
  189. int h, w, mv_x, mv_y, offset, offset_dst;
  190. uint8_t *src, *dst;
  191. /* setup output and reference pointers */
  192. offset_dst = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
  193. dst = plane->pixels[ctx->buf_sel] + offset_dst;
  194. mv_y = cell->mv_ptr[0];
  195. mv_x = cell->mv_ptr[1];
  196. /* -1 because there is an extra line on top for prediction */
  197. if ((cell->ypos << 2) + mv_y < -1 || (cell->xpos << 2) + mv_x < 0 ||
  198. ((cell->ypos + cell->height) << 2) + mv_y > plane->height ||
  199. ((cell->xpos + cell->width) << 2) + mv_x > plane->width) {
  200. av_log(ctx->avctx, AV_LOG_ERROR,
  201. "Motion vectors point out of the frame.\n");
  202. return AVERROR_INVALIDDATA;
  203. }
  204. offset = offset_dst + mv_y * plane->pitch + mv_x;
  205. src = plane->pixels[ctx->buf_sel ^ 1] + offset;
  206. h = cell->height << 2;
  207. for (w = cell->width; w > 0;) {
  208. /* copy using 16xH blocks */
  209. if (!((cell->xpos << 2) & 15) && w >= 4) {
  210. for (; w >= 4; src += 16, dst += 16, w -= 4)
  211. ctx->hdsp.put_pixels_tab[0][0](dst, src, plane->pitch, h);
  212. }
  213. /* copy using 8xH blocks */
  214. if (!((cell->xpos << 2) & 7) && w >= 2) {
  215. ctx->hdsp.put_pixels_tab[1][0](dst, src, plane->pitch, h);
  216. w -= 2;
  217. src += 8;
  218. dst += 8;
  219. }
  220. if (w >= 1) {
  221. ctx->hdsp.put_pixels_tab[2][0](dst, src, plane->pitch, h);
  222. w--;
  223. src += 4;
  224. dst += 4;
  225. }
  226. }
  227. return 0;
  228. }
  229. /* Average 4/8 pixels at once without rounding using SWAR */
  230. #define AVG_32(dst, src, ref) \
  231. AV_WN32A(dst, ((AV_RN32A(src) + AV_RN32A(ref)) >> 1) & 0x7F7F7F7FUL)
  232. #define AVG_64(dst, src, ref) \
  233. AV_WN64A(dst, ((AV_RN64A(src) + AV_RN64A(ref)) >> 1) & 0x7F7F7F7F7F7F7F7FULL)
  234. /*
  235. * Replicate each even pixel as follows:
  236. * ABCDEFGH -> AACCEEGG
  237. */
  238. static inline uint64_t replicate64(uint64_t a) {
  239. #if HAVE_BIGENDIAN
  240. a &= 0xFF00FF00FF00FF00ULL;
  241. a |= a >> 8;
  242. #else
  243. a &= 0x00FF00FF00FF00FFULL;
  244. a |= a << 8;
  245. #endif
  246. return a;
  247. }
  248. static inline uint32_t replicate32(uint32_t a) {
  249. #if HAVE_BIGENDIAN
  250. a &= 0xFF00FF00UL;
  251. a |= a >> 8;
  252. #else
  253. a &= 0x00FF00FFUL;
  254. a |= a << 8;
  255. #endif
  256. return a;
  257. }
  258. /* Fill n lines with 64bit pixel value pix */
  259. static inline void fill_64(uint8_t *dst, const uint64_t pix, int32_t n,
  260. int32_t row_offset)
  261. {
  262. for (; n > 0; dst += row_offset, n--)
  263. AV_WN64A(dst, pix);
  264. }
  265. /* Error codes for cell decoding. */
  266. enum {
  267. IV3_NOERR = 0,
  268. IV3_BAD_RLE = 1,
  269. IV3_BAD_DATA = 2,
  270. IV3_BAD_COUNTER = 3,
  271. IV3_UNSUPPORTED = 4,
  272. IV3_OUT_OF_DATA = 5
  273. };
  274. #define BUFFER_PRECHECK \
  275. if (*data_ptr >= last_ptr) \
  276. return IV3_OUT_OF_DATA; \
  277. #define RLE_BLOCK_COPY \
  278. if (cell->mv_ptr || !skip_flag) \
  279. ctx->hdsp.put_pixels_tab[2][0](dst, ref, row_offset, 4 << v_zoom)
  280. #define RLE_BLOCK_COPY_8 \
  281. pix64 = AV_RN64A(ref);\
  282. if (is_first_row) {/* special prediction case: top line of a cell */\
  283. pix64 = replicate64(pix64);\
  284. fill_64(dst + row_offset, pix64, 7, row_offset);\
  285. AVG_64(dst, ref, dst + row_offset);\
  286. } else \
  287. fill_64(dst, pix64, 8, row_offset)
  288. #define RLE_LINES_COPY \
  289. ctx->hdsp.put_pixels_tab[2][0](dst, ref, row_offset, num_lines << v_zoom)
  290. #define RLE_LINES_COPY_M10 \
  291. pix64 = AV_RN64A(ref);\
  292. if (is_top_of_cell) {\
  293. pix64 = replicate64(pix64);\
  294. fill_64(dst + row_offset, pix64, (num_lines << 1) - 1, row_offset);\
  295. AVG_64(dst, ref, dst + row_offset);\
  296. } else \
  297. fill_64(dst, pix64, num_lines << 1, row_offset)
  298. #define APPLY_DELTA_4 \
  299. AV_WN16A(dst + line_offset ,\
  300. (AV_RN16A(ref ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
  301. AV_WN16A(dst + line_offset + 2,\
  302. (AV_RN16A(ref + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
  303. if (mode >= 3) {\
  304. if (is_top_of_cell && !cell->ypos) {\
  305. AV_COPY32(dst, dst + row_offset);\
  306. } else {\
  307. AVG_32(dst, ref, dst + row_offset);\
  308. }\
  309. }
  310. #define APPLY_DELTA_8 \
  311. /* apply two 32-bit VQ deltas to next even line */\
  312. if (is_top_of_cell) { \
  313. AV_WN32A(dst + row_offset , \
  314. (replicate32(AV_RN32A(ref )) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
  315. AV_WN32A(dst + row_offset + 4, \
  316. (replicate32(AV_RN32A(ref + 4)) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
  317. } else { \
  318. AV_WN32A(dst + row_offset , \
  319. (AV_RN32A(ref ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
  320. AV_WN32A(dst + row_offset + 4, \
  321. (AV_RN32A(ref + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
  322. } \
  323. /* odd lines are not coded but rather interpolated/replicated */\
  324. /* first line of the cell on the top of image? - replicate */\
  325. /* otherwise - interpolate */\
  326. if (is_top_of_cell && !cell->ypos) {\
  327. AV_COPY64(dst, dst + row_offset);\
  328. } else \
  329. AVG_64(dst, ref, dst + row_offset);
  330. #define APPLY_DELTA_1011_INTER \
  331. if (mode == 10) { \
  332. AV_WN32A(dst , \
  333. (AV_RN32A(dst ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
  334. AV_WN32A(dst + 4 , \
  335. (AV_RN32A(dst + 4 ) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
  336. AV_WN32A(dst + row_offset , \
  337. (AV_RN32A(dst + row_offset ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
  338. AV_WN32A(dst + row_offset + 4, \
  339. (AV_RN32A(dst + row_offset + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
  340. } else { \
  341. AV_WN16A(dst , \
  342. (AV_RN16A(dst ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
  343. AV_WN16A(dst + 2 , \
  344. (AV_RN16A(dst + 2 ) + delta_tab->deltas[dyad2]) & 0x7F7F);\
  345. AV_WN16A(dst + row_offset , \
  346. (AV_RN16A(dst + row_offset ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
  347. AV_WN16A(dst + row_offset + 2, \
  348. (AV_RN16A(dst + row_offset + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
  349. }
  350. static int decode_cell_data(Indeo3DecodeContext *ctx, Cell *cell,
  351. uint8_t *block, uint8_t *ref_block,
  352. int pitch, int h_zoom, int v_zoom, int mode,
  353. const vqEntry *delta[2], int swap_quads[2],
  354. const uint8_t **data_ptr, const uint8_t *last_ptr)
  355. {
  356. int x, y, line, num_lines;
  357. int rle_blocks = 0;
  358. uint8_t code, *dst, *ref;
  359. const vqEntry *delta_tab;
  360. unsigned int dyad1, dyad2;
  361. uint64_t pix64;
  362. int skip_flag = 0, is_top_of_cell, is_first_row = 1;
  363. int row_offset, blk_row_offset, line_offset;
  364. row_offset = pitch;
  365. blk_row_offset = (row_offset << (2 + v_zoom)) - (cell->width << 2);
  366. line_offset = v_zoom ? row_offset : 0;
  367. if (cell->height & v_zoom || cell->width & h_zoom)
  368. return IV3_BAD_DATA;
  369. for (y = 0; y < cell->height; is_first_row = 0, y += 1 + v_zoom) {
  370. for (x = 0; x < cell->width; x += 1 + h_zoom) {
  371. ref = ref_block;
  372. dst = block;
  373. if (rle_blocks > 0) {
  374. if (mode <= 4) {
  375. RLE_BLOCK_COPY;
  376. } else if (mode == 10 && !cell->mv_ptr) {
  377. RLE_BLOCK_COPY_8;
  378. }
  379. rle_blocks--;
  380. } else {
  381. for (line = 0; line < 4;) {
  382. num_lines = 1;
  383. is_top_of_cell = is_first_row && !line;
  384. /* select primary VQ table for odd, secondary for even lines */
  385. if (mode <= 4)
  386. delta_tab = delta[line & 1];
  387. else
  388. delta_tab = delta[1];
  389. BUFFER_PRECHECK;
  390. code = bytestream_get_byte(data_ptr);
  391. if (code < 248) {
  392. if (code < delta_tab->num_dyads) {
  393. BUFFER_PRECHECK;
  394. dyad1 = bytestream_get_byte(data_ptr);
  395. dyad2 = code;
  396. if (dyad1 >= delta_tab->num_dyads || dyad1 >= 248)
  397. return IV3_BAD_DATA;
  398. } else {
  399. /* process QUADS */
  400. code -= delta_tab->num_dyads;
  401. dyad1 = code / delta_tab->quad_exp;
  402. dyad2 = code % delta_tab->quad_exp;
  403. if (swap_quads[line & 1])
  404. FFSWAP(unsigned int, dyad1, dyad2);
  405. }
  406. if (mode <= 4) {
  407. APPLY_DELTA_4;
  408. } else if (mode == 10 && !cell->mv_ptr) {
  409. APPLY_DELTA_8;
  410. } else {
  411. APPLY_DELTA_1011_INTER;
  412. }
  413. } else {
  414. /* process RLE codes */
  415. switch (code) {
  416. case RLE_ESC_FC:
  417. skip_flag = 0;
  418. rle_blocks = 1;
  419. code = 253;
  420. /* FALLTHROUGH */
  421. case RLE_ESC_FF:
  422. case RLE_ESC_FE:
  423. case RLE_ESC_FD:
  424. num_lines = 257 - code - line;
  425. if (num_lines <= 0)
  426. return IV3_BAD_RLE;
  427. if (mode <= 4) {
  428. RLE_LINES_COPY;
  429. } else if (mode == 10 && !cell->mv_ptr) {
  430. RLE_LINES_COPY_M10;
  431. }
  432. break;
  433. case RLE_ESC_FB:
  434. BUFFER_PRECHECK;
  435. code = bytestream_get_byte(data_ptr);
  436. rle_blocks = (code & 0x1F) - 1; /* set block counter */
  437. if (code >= 64 || rle_blocks < 0)
  438. return IV3_BAD_COUNTER;
  439. skip_flag = code & 0x20;
  440. num_lines = 4 - line; /* enforce next block processing */
  441. if (mode >= 10 || (cell->mv_ptr || !skip_flag)) {
  442. if (mode <= 4) {
  443. RLE_LINES_COPY;
  444. } else if (mode == 10 && !cell->mv_ptr) {
  445. RLE_LINES_COPY_M10;
  446. }
  447. }
  448. break;
  449. case RLE_ESC_F9:
  450. skip_flag = 1;
  451. rle_blocks = 1;
  452. /* FALLTHROUGH */
  453. case RLE_ESC_FA:
  454. if (line)
  455. return IV3_BAD_RLE;
  456. num_lines = 4; /* enforce next block processing */
  457. if (cell->mv_ptr) {
  458. if (mode <= 4) {
  459. RLE_LINES_COPY;
  460. } else if (mode == 10 && !cell->mv_ptr) {
  461. RLE_LINES_COPY_M10;
  462. }
  463. }
  464. break;
  465. default:
  466. return IV3_UNSUPPORTED;
  467. }
  468. }
  469. line += num_lines;
  470. ref += row_offset * (num_lines << v_zoom);
  471. dst += row_offset * (num_lines << v_zoom);
  472. }
  473. }
  474. /* move to next horizontal block */
  475. block += 4 << h_zoom;
  476. ref_block += 4 << h_zoom;
  477. }
  478. /* move to next line of blocks */
  479. ref_block += blk_row_offset;
  480. block += blk_row_offset;
  481. }
  482. return IV3_NOERR;
  483. }
  484. /**
  485. * Decode a vector-quantized cell.
  486. * It consists of several routines, each of which handles one or more "modes"
  487. * with which a cell can be encoded.
  488. *
  489. * @param ctx pointer to the decoder context
  490. * @param avctx ptr to the AVCodecContext
  491. * @param plane pointer to the plane descriptor
  492. * @param cell pointer to the cell descriptor
  493. * @param data_ptr pointer to the compressed data
  494. * @param last_ptr pointer to the last byte to catch reads past end of buffer
  495. * @return number of consumed bytes or negative number in case of error
  496. */
  497. static int decode_cell(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
  498. Plane *plane, Cell *cell, const uint8_t *data_ptr,
  499. const uint8_t *last_ptr)
  500. {
  501. int x, mv_x, mv_y, mode, vq_index, prim_indx, second_indx;
  502. int zoom_fac;
  503. int offset, error = 0, swap_quads[2];
  504. uint8_t code, *block, *ref_block = 0;
  505. const vqEntry *delta[2];
  506. const uint8_t *data_start = data_ptr;
  507. /* get coding mode and VQ table index from the VQ descriptor byte */
  508. code = *data_ptr++;
  509. mode = code >> 4;
  510. vq_index = code & 0xF;
  511. /* setup output and reference pointers */
  512. offset = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
  513. block = plane->pixels[ctx->buf_sel] + offset;
  514. if (!cell->mv_ptr) {
  515. /* use previous line as reference for INTRA cells */
  516. ref_block = block - plane->pitch;
  517. } else if (mode >= 10) {
  518. /* for mode 10 and 11 INTER first copy the predicted cell into the current one */
  519. /* so we don't need to do data copying for each RLE code later */
  520. int ret = copy_cell(ctx, plane, cell);
  521. if (ret < 0)
  522. return ret;
  523. } else {
  524. /* set the pointer to the reference pixels for modes 0-4 INTER */
  525. mv_y = cell->mv_ptr[0];
  526. mv_x = cell->mv_ptr[1];
  527. /* -1 because there is an extra line on top for prediction */
  528. if ((cell->ypos << 2) + mv_y < -1 || (cell->xpos << 2) + mv_x < 0 ||
  529. ((cell->ypos + cell->height) << 2) + mv_y > plane->height ||
  530. ((cell->xpos + cell->width) << 2) + mv_x > plane->width) {
  531. av_log(ctx->avctx, AV_LOG_ERROR,
  532. "Motion vectors point out of the frame.\n");
  533. return AVERROR_INVALIDDATA;
  534. }
  535. offset += mv_y * plane->pitch + mv_x;
  536. ref_block = plane->pixels[ctx->buf_sel ^ 1] + offset;
  537. }
  538. /* select VQ tables as follows: */
  539. /* modes 0 and 3 use only the primary table for all lines in a block */
  540. /* while modes 1 and 4 switch between primary and secondary tables on alternate lines */
  541. if (mode == 1 || mode == 4) {
  542. code = ctx->alt_quant[vq_index];
  543. prim_indx = (code >> 4) + ctx->cb_offset;
  544. second_indx = (code & 0xF) + ctx->cb_offset;
  545. } else {
  546. vq_index += ctx->cb_offset;
  547. prim_indx = second_indx = vq_index;
  548. }
  549. if (prim_indx >= 24 || second_indx >= 24) {
  550. av_log(avctx, AV_LOG_ERROR, "Invalid VQ table indexes! Primary: %d, secondary: %d!\n",
  551. prim_indx, second_indx);
  552. return AVERROR_INVALIDDATA;
  553. }
  554. delta[0] = &vq_tab[second_indx];
  555. delta[1] = &vq_tab[prim_indx];
  556. swap_quads[0] = second_indx >= 16;
  557. swap_quads[1] = prim_indx >= 16;
  558. /* requantize the prediction if VQ index of this cell differs from VQ index */
  559. /* of the predicted cell in order to avoid overflows. */
  560. if (vq_index >= 8 && ref_block) {
  561. for (x = 0; x < cell->width << 2; x++)
  562. ref_block[x] = requant_tab[vq_index & 7][ref_block[x]];
  563. }
  564. error = IV3_NOERR;
  565. switch (mode) {
  566. case 0: /*------------------ MODES 0 & 1 (4x4 block processing) --------------------*/
  567. case 1:
  568. case 3: /*------------------ MODES 3 & 4 (4x8 block processing) --------------------*/
  569. case 4:
  570. if (mode >= 3 && cell->mv_ptr) {
  571. av_log(avctx, AV_LOG_ERROR, "Attempt to apply Mode 3/4 to an INTER cell!\n");
  572. return AVERROR_INVALIDDATA;
  573. }
  574. zoom_fac = mode >= 3;
  575. error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
  576. 0, zoom_fac, mode, delta, swap_quads,
  577. &data_ptr, last_ptr);
  578. break;
  579. case 10: /*-------------------- MODE 10 (8x8 block processing) ---------------------*/
  580. case 11: /*----------------- MODE 11 (4x8 INTER block processing) ------------------*/
  581. if (mode == 10 && !cell->mv_ptr) { /* MODE 10 INTRA processing */
  582. error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
  583. 1, 1, mode, delta, swap_quads,
  584. &data_ptr, last_ptr);
  585. } else { /* mode 10 and 11 INTER processing */
  586. if (mode == 11 && !cell->mv_ptr) {
  587. av_log(avctx, AV_LOG_ERROR, "Attempt to use Mode 11 for an INTRA cell!\n");
  588. return AVERROR_INVALIDDATA;
  589. }
  590. zoom_fac = mode == 10;
  591. error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
  592. zoom_fac, 1, mode, delta, swap_quads,
  593. &data_ptr, last_ptr);
  594. }
  595. break;
  596. default:
  597. av_log(avctx, AV_LOG_ERROR, "Unsupported coding mode: %d\n", mode);
  598. return AVERROR_INVALIDDATA;
  599. }//switch mode
  600. switch (error) {
  601. case IV3_BAD_RLE:
  602. av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE code %X is not allowed at the current line\n",
  603. mode, data_ptr[-1]);
  604. return AVERROR_INVALIDDATA;
  605. case IV3_BAD_DATA:
  606. av_log(avctx, AV_LOG_ERROR, "Mode %d: invalid VQ data\n", mode);
  607. return AVERROR_INVALIDDATA;
  608. case IV3_BAD_COUNTER:
  609. av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE-FB invalid counter: %d\n", mode, code);
  610. return AVERROR_INVALIDDATA;
  611. case IV3_UNSUPPORTED:
  612. av_log(avctx, AV_LOG_ERROR, "Mode %d: unsupported RLE code: %X\n", mode, data_ptr[-1]);
  613. return AVERROR_INVALIDDATA;
  614. case IV3_OUT_OF_DATA:
  615. av_log(avctx, AV_LOG_ERROR, "Mode %d: attempt to read past end of buffer\n", mode);
  616. return AVERROR_INVALIDDATA;
  617. }
  618. return data_ptr - data_start; /* report number of bytes consumed from the input buffer */
  619. }
  620. /* Binary tree codes. */
  621. enum {
  622. H_SPLIT = 0,
  623. V_SPLIT = 1,
  624. INTRA_NULL = 2,
  625. INTER_DATA = 3
  626. };
  627. #define SPLIT_CELL(size, new_size) (new_size) = ((size) > 2) ? ((((size) + 2) >> 2) << 1) : 1
  628. #define UPDATE_BITPOS(n) \
  629. ctx->skip_bits += (n); \
  630. ctx->need_resync = 1
  631. #define RESYNC_BITSTREAM \
  632. if (ctx->need_resync && !(get_bits_count(&ctx->gb) & 7)) { \
  633. skip_bits_long(&ctx->gb, ctx->skip_bits); \
  634. ctx->skip_bits = 0; \
  635. ctx->need_resync = 0; \
  636. }
  637. #define CHECK_CELL \
  638. if (curr_cell.xpos + curr_cell.width > (plane->width >> 2) || \
  639. curr_cell.ypos + curr_cell.height > (plane->height >> 2)) { \
  640. av_log(avctx, AV_LOG_ERROR, "Invalid cell: x=%d, y=%d, w=%d, h=%d\n", \
  641. curr_cell.xpos, curr_cell.ypos, curr_cell.width, curr_cell.height); \
  642. return AVERROR_INVALIDDATA; \
  643. }
  644. static int parse_bintree(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
  645. Plane *plane, int code, Cell *ref_cell,
  646. const int depth, const int strip_width)
  647. {
  648. Cell curr_cell;
  649. int bytes_used, ret;
  650. if (depth <= 0) {
  651. av_log(avctx, AV_LOG_ERROR, "Stack overflow (corrupted binary tree)!\n");
  652. return AVERROR_INVALIDDATA; // unwind recursion
  653. }
  654. curr_cell = *ref_cell; // clone parent cell
  655. if (code == H_SPLIT) {
  656. SPLIT_CELL(ref_cell->height, curr_cell.height);
  657. ref_cell->ypos += curr_cell.height;
  658. ref_cell->height -= curr_cell.height;
  659. if (ref_cell->height <= 0 || curr_cell.height <= 0)
  660. return AVERROR_INVALIDDATA;
  661. } else if (code == V_SPLIT) {
  662. if (curr_cell.width > strip_width) {
  663. /* split strip */
  664. curr_cell.width = (curr_cell.width <= (strip_width << 1) ? 1 : 2) * strip_width;
  665. } else
  666. SPLIT_CELL(ref_cell->width, curr_cell.width);
  667. ref_cell->xpos += curr_cell.width;
  668. ref_cell->width -= curr_cell.width;
  669. if (ref_cell->width <= 0 || curr_cell.width <= 0)
  670. return AVERROR_INVALIDDATA;
  671. }
  672. while (1) { /* loop until return */
  673. RESYNC_BITSTREAM;
  674. switch (code = get_bits(&ctx->gb, 2)) {
  675. case H_SPLIT:
  676. case V_SPLIT:
  677. if (parse_bintree(ctx, avctx, plane, code, &curr_cell, depth - 1, strip_width))
  678. return AVERROR_INVALIDDATA;
  679. break;
  680. case INTRA_NULL:
  681. if (!curr_cell.tree) { /* MC tree INTRA code */
  682. curr_cell.mv_ptr = 0; /* mark the current strip as INTRA */
  683. curr_cell.tree = 1; /* enter the VQ tree */
  684. } else { /* VQ tree NULL code */
  685. RESYNC_BITSTREAM;
  686. code = get_bits(&ctx->gb, 2);
  687. if (code >= 2) {
  688. av_log(avctx, AV_LOG_ERROR, "Invalid VQ_NULL code: %d\n", code);
  689. return AVERROR_INVALIDDATA;
  690. }
  691. if (code == 1)
  692. av_log(avctx, AV_LOG_ERROR, "SkipCell procedure not implemented yet!\n");
  693. CHECK_CELL
  694. if (!curr_cell.mv_ptr)
  695. return AVERROR_INVALIDDATA;
  696. ret = copy_cell(ctx, plane, &curr_cell);
  697. return ret;
  698. }
  699. break;
  700. case INTER_DATA:
  701. if (!curr_cell.tree) { /* MC tree INTER code */
  702. unsigned mv_idx;
  703. /* get motion vector index and setup the pointer to the mv set */
  704. if (!ctx->need_resync)
  705. ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
  706. mv_idx = *(ctx->next_cell_data++);
  707. if (mv_idx >= ctx->num_vectors) {
  708. av_log(avctx, AV_LOG_ERROR, "motion vector index out of range\n");
  709. return AVERROR_INVALIDDATA;
  710. }
  711. curr_cell.mv_ptr = &ctx->mc_vectors[mv_idx << 1];
  712. curr_cell.tree = 1; /* enter the VQ tree */
  713. UPDATE_BITPOS(8);
  714. } else { /* VQ tree DATA code */
  715. if (!ctx->need_resync)
  716. ctx->next_cell_data = &ctx->gb.buffer[(get_bits_count(&ctx->gb) + 7) >> 3];
  717. CHECK_CELL
  718. bytes_used = decode_cell(ctx, avctx, plane, &curr_cell,
  719. ctx->next_cell_data, ctx->last_byte);
  720. if (bytes_used < 0)
  721. return AVERROR_INVALIDDATA;
  722. UPDATE_BITPOS(bytes_used << 3);
  723. ctx->next_cell_data += bytes_used;
  724. return 0;
  725. }
  726. break;
  727. }
  728. }//while
  729. return 0;
  730. }
  731. static int decode_plane(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
  732. Plane *plane, const uint8_t *data, int32_t data_size,
  733. int32_t strip_width)
  734. {
  735. Cell curr_cell;
  736. unsigned num_vectors;
  737. /* each plane data starts with mc_vector_count field, */
  738. /* an optional array of motion vectors followed by the vq data */
  739. num_vectors = bytestream_get_le32(&data);
  740. if (num_vectors > 256) {
  741. av_log(ctx->avctx, AV_LOG_ERROR,
  742. "Read invalid number of motion vectors %d\n", num_vectors);
  743. return AVERROR_INVALIDDATA;
  744. }
  745. if (num_vectors * 2 >= data_size)
  746. return AVERROR_INVALIDDATA;
  747. ctx->num_vectors = num_vectors;
  748. ctx->mc_vectors = num_vectors ? data : 0;
  749. /* init the bitreader */
  750. init_get_bits(&ctx->gb, &data[num_vectors * 2], (data_size - num_vectors * 2) << 3);
  751. ctx->skip_bits = 0;
  752. ctx->need_resync = 0;
  753. ctx->last_byte = data + data_size - 1;
  754. /* initialize the 1st cell and set its dimensions to whole plane */
  755. curr_cell.xpos = curr_cell.ypos = 0;
  756. curr_cell.width = plane->width >> 2;
  757. curr_cell.height = plane->height >> 2;
  758. curr_cell.tree = 0; // we are in the MC tree now
  759. curr_cell.mv_ptr = 0; // no motion vector = INTRA cell
  760. return parse_bintree(ctx, avctx, plane, INTRA_NULL, &curr_cell, CELL_STACK_MAX, strip_width);
  761. }
  762. #define OS_HDR_ID MKBETAG('F', 'R', 'M', 'H')
  763. static int decode_frame_headers(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
  764. const uint8_t *buf, int buf_size)
  765. {
  766. GetByteContext gb;
  767. const uint8_t *bs_hdr;
  768. uint32_t frame_num, word2, check_sum, data_size;
  769. uint32_t y_offset, u_offset, v_offset, starts[3], ends[3];
  770. uint16_t height, width;
  771. int i, j;
  772. bytestream2_init(&gb, buf, buf_size);
  773. /* parse and check the OS header */
  774. frame_num = bytestream2_get_le32(&gb);
  775. word2 = bytestream2_get_le32(&gb);
  776. check_sum = bytestream2_get_le32(&gb);
  777. data_size = bytestream2_get_le32(&gb);
  778. if ((frame_num ^ word2 ^ data_size ^ OS_HDR_ID) != check_sum) {
  779. av_log(avctx, AV_LOG_ERROR, "OS header checksum mismatch!\n");
  780. return AVERROR_INVALIDDATA;
  781. }
  782. /* parse the bitstream header */
  783. bs_hdr = gb.buffer;
  784. if (bytestream2_get_le16(&gb) != 32) {
  785. av_log(avctx, AV_LOG_ERROR, "Unsupported codec version!\n");
  786. return AVERROR_INVALIDDATA;
  787. }
  788. ctx->frame_num = frame_num;
  789. ctx->frame_flags = bytestream2_get_le16(&gb);
  790. ctx->data_size = (bytestream2_get_le32(&gb) + 7) >> 3;
  791. ctx->cb_offset = bytestream2_get_byte(&gb);
  792. if (ctx->data_size == 16)
  793. return 4;
  794. ctx->data_size = FFMIN(ctx->data_size, buf_size - 16);
  795. bytestream2_skip(&gb, 3); // skip reserved byte and checksum
  796. /* check frame dimensions */
  797. height = bytestream2_get_le16(&gb);
  798. width = bytestream2_get_le16(&gb);
  799. if (av_image_check_size(width, height, 0, avctx))
  800. return AVERROR_INVALIDDATA;
  801. if (width != ctx->width || height != ctx->height) {
  802. int res;
  803. av_dlog(avctx, "Frame dimensions changed!\n");
  804. if (width < 16 || width > 640 ||
  805. height < 16 || height > 480 ||
  806. width & 3 || height & 3) {
  807. av_log(avctx, AV_LOG_ERROR,
  808. "Invalid picture dimensions: %d x %d!\n", width, height);
  809. return AVERROR_INVALIDDATA;
  810. }
  811. ctx->width = width;
  812. ctx->height = height;
  813. free_frame_buffers(ctx);
  814. if ((res = allocate_frame_buffers(ctx, avctx)) < 0)
  815. return res;
  816. avcodec_set_dimensions(avctx, width, height);
  817. }
  818. y_offset = bytestream2_get_le32(&gb);
  819. v_offset = bytestream2_get_le32(&gb);
  820. u_offset = bytestream2_get_le32(&gb);
  821. bytestream2_skip(&gb, 4);
  822. /* unfortunately there is no common order of planes in the buffer */
  823. /* so we use that sorting algo for determining planes data sizes */
  824. starts[0] = y_offset;
  825. starts[1] = v_offset;
  826. starts[2] = u_offset;
  827. for (j = 0; j < 3; j++) {
  828. ends[j] = ctx->data_size;
  829. for (i = 2; i >= 0; i--)
  830. if (starts[i] < ends[j] && starts[i] > starts[j])
  831. ends[j] = starts[i];
  832. }
  833. ctx->y_data_size = ends[0] - starts[0];
  834. ctx->v_data_size = ends[1] - starts[1];
  835. ctx->u_data_size = ends[2] - starts[2];
  836. if (FFMAX3(y_offset, v_offset, u_offset) >= ctx->data_size - 16 ||
  837. FFMIN3(y_offset, v_offset, u_offset) < gb.buffer - bs_hdr + 16 ||
  838. FFMIN3(ctx->y_data_size, ctx->v_data_size, ctx->u_data_size) <= 0) {
  839. av_log(avctx, AV_LOG_ERROR, "One of the y/u/v offsets is invalid\n");
  840. return AVERROR_INVALIDDATA;
  841. }
  842. ctx->y_data_ptr = bs_hdr + y_offset;
  843. ctx->v_data_ptr = bs_hdr + v_offset;
  844. ctx->u_data_ptr = bs_hdr + u_offset;
  845. ctx->alt_quant = gb.buffer;
  846. if (ctx->data_size == 16) {
  847. av_log(avctx, AV_LOG_DEBUG, "Sync frame encountered!\n");
  848. return 16;
  849. }
  850. if (ctx->frame_flags & BS_8BIT_PEL) {
  851. avpriv_request_sample(avctx, "8-bit pixel format");
  852. return AVERROR_PATCHWELCOME;
  853. }
  854. if (ctx->frame_flags & BS_MV_X_HALF || ctx->frame_flags & BS_MV_Y_HALF) {
  855. avpriv_request_sample(avctx, "Halfpel motion vectors");
  856. return AVERROR_PATCHWELCOME;
  857. }
  858. return 0;
  859. }
  860. /**
  861. * Convert and output the current plane.
  862. * All pixel values will be upsampled by shifting right by one bit.
  863. *
  864. * @param[in] plane pointer to the descriptor of the plane being processed
  865. * @param[in] buf_sel indicates which frame buffer the input data stored in
  866. * @param[out] dst pointer to the buffer receiving converted pixels
  867. * @param[in] dst_pitch pitch for moving to the next y line
  868. * @param[in] dst_height output plane height
  869. */
  870. static void output_plane(const Plane *plane, int buf_sel, uint8_t *dst,
  871. int dst_pitch, int dst_height)
  872. {
  873. int x,y;
  874. const uint8_t *src = plane->pixels[buf_sel];
  875. uint32_t pitch = plane->pitch;
  876. dst_height = FFMIN(dst_height, plane->height);
  877. for (y = 0; y < dst_height; y++) {
  878. /* convert four pixels at once using SWAR */
  879. for (x = 0; x < plane->width >> 2; x++) {
  880. AV_WN32A(dst, (AV_RN32A(src) & 0x7F7F7F7F) << 1);
  881. src += 4;
  882. dst += 4;
  883. }
  884. for (x <<= 2; x < plane->width; x++)
  885. *dst++ = *src++ << 1;
  886. src += pitch - plane->width;
  887. dst += dst_pitch - plane->width;
  888. }
  889. }
  890. static av_cold int decode_init(AVCodecContext *avctx)
  891. {
  892. Indeo3DecodeContext *ctx = avctx->priv_data;
  893. ctx->avctx = avctx;
  894. ctx->width = avctx->width;
  895. ctx->height = avctx->height;
  896. avctx->pix_fmt = AV_PIX_FMT_YUV410P;
  897. build_requant_tab();
  898. ff_hpeldsp_init(&ctx->hdsp, avctx->flags);
  899. allocate_frame_buffers(ctx, avctx);
  900. return 0;
  901. }
  902. static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
  903. AVPacket *avpkt)
  904. {
  905. Indeo3DecodeContext *ctx = avctx->priv_data;
  906. const uint8_t *buf = avpkt->data;
  907. int buf_size = avpkt->size;
  908. AVFrame *frame = data;
  909. int res;
  910. res = decode_frame_headers(ctx, avctx, buf, buf_size);
  911. if (res < 0)
  912. return res;
  913. /* skip sync(null) frames */
  914. if (res) {
  915. // we have processed 16 bytes but no data was decoded
  916. *got_frame = 0;
  917. return buf_size;
  918. }
  919. /* skip droppable INTER frames if requested */
  920. if (ctx->frame_flags & BS_NONREF &&
  921. (avctx->skip_frame >= AVDISCARD_NONREF))
  922. return 0;
  923. /* skip INTER frames if requested */
  924. if (!(ctx->frame_flags & BS_KEYFRAME) && avctx->skip_frame >= AVDISCARD_NONKEY)
  925. return 0;
  926. /* use BS_BUFFER flag for buffer switching */
  927. ctx->buf_sel = (ctx->frame_flags >> BS_BUFFER) & 1;
  928. /* decode luma plane */
  929. if ((res = decode_plane(ctx, avctx, ctx->planes, ctx->y_data_ptr, ctx->y_data_size, 40)))
  930. return res;
  931. /* decode chroma planes */
  932. if ((res = decode_plane(ctx, avctx, &ctx->planes[1], ctx->u_data_ptr, ctx->u_data_size, 10)))
  933. return res;
  934. if ((res = decode_plane(ctx, avctx, &ctx->planes[2], ctx->v_data_ptr, ctx->v_data_size, 10)))
  935. return res;
  936. if ((res = ff_get_buffer(avctx, frame, 0)) < 0) {
  937. av_log(ctx->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
  938. return res;
  939. }
  940. output_plane(&ctx->planes[0], ctx->buf_sel,
  941. frame->data[0], frame->linesize[0],
  942. avctx->height);
  943. output_plane(&ctx->planes[1], ctx->buf_sel,
  944. frame->data[1], frame->linesize[1],
  945. (avctx->height + 3) >> 2);
  946. output_plane(&ctx->planes[2], ctx->buf_sel,
  947. frame->data[2], frame->linesize[2],
  948. (avctx->height + 3) >> 2);
  949. *got_frame = 1;
  950. return buf_size;
  951. }
  952. static av_cold int decode_close(AVCodecContext *avctx)
  953. {
  954. free_frame_buffers(avctx->priv_data);
  955. return 0;
  956. }
  957. AVCodec ff_indeo3_decoder = {
  958. .name = "indeo3",
  959. .type = AVMEDIA_TYPE_VIDEO,
  960. .id = AV_CODEC_ID_INDEO3,
  961. .priv_data_size = sizeof(Indeo3DecodeContext),
  962. .init = decode_init,
  963. .close = decode_close,
  964. .decode = decode_frame,
  965. .capabilities = CODEC_CAP_DR1,
  966. .long_name = NULL_IF_CONFIG_SMALL("Intel Indeo 3"),
  967. };