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.

1138 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 "bitstream.h"
  34. #include "bytestream.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) ///< 8-bit 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. ptrdiff_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. BitstreamContext bc;
  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 const int8_t offsets[8] = { 1, 1, 2, -3, -3, 3, 4, 4 };
  105. static const 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_size, chroma_size;
  133. ptrdiff_t luma_pitch, chroma_pitch;
  134. luma_width = ctx->width;
  135. luma_height = ctx->height;
  136. if (luma_width < 16 || luma_width > 640 ||
  137. luma_height < 16 || luma_height > 480 ||
  138. luma_width & 3 || luma_height & 3) {
  139. av_log(avctx, AV_LOG_ERROR, "Invalid picture dimensions: %d x %d!\n",
  140. luma_width, luma_height);
  141. return AVERROR_INVALIDDATA;
  142. }
  143. chroma_width = FFALIGN(luma_width >> 2, 4);
  144. chroma_height = FFALIGN(luma_height >> 2, 4);
  145. luma_pitch = FFALIGN(luma_width, 16);
  146. chroma_pitch = FFALIGN(chroma_width, 16);
  147. /* Calculate size of the luminance plane. */
  148. /* Add one line more for INTRA prediction. */
  149. luma_size = luma_pitch * (luma_height + 1);
  150. /* Calculate size of a chrominance planes. */
  151. /* Add one line more for INTRA prediction. */
  152. chroma_size = chroma_pitch * (chroma_height + 1);
  153. /* allocate frame buffers */
  154. for (p = 0; p < 3; p++) {
  155. ctx->planes[p].pitch = !p ? luma_pitch : chroma_pitch;
  156. ctx->planes[p].width = !p ? luma_width : chroma_width;
  157. ctx->planes[p].height = !p ? luma_height : chroma_height;
  158. ctx->planes[p].buffers[0] = av_malloc(!p ? luma_size : chroma_size);
  159. ctx->planes[p].buffers[1] = av_malloc(!p ? luma_size : chroma_size);
  160. /* fill the INTRA prediction lines with the middle pixel value = 64 */
  161. memset(ctx->planes[p].buffers[0], 0x40, ctx->planes[p].pitch);
  162. memset(ctx->planes[p].buffers[1], 0x40, ctx->planes[p].pitch);
  163. /* set buffer pointers = buf_ptr + pitch and thus skip the INTRA prediction line */
  164. ctx->planes[p].pixels[0] = ctx->planes[p].buffers[0] + ctx->planes[p].pitch;
  165. ctx->planes[p].pixels[1] = ctx->planes[p].buffers[1] + ctx->planes[p].pitch;
  166. memset(ctx->planes[p].pixels[0], 0, ctx->planes[p].pitch * ctx->planes[p].height);
  167. memset(ctx->planes[p].pixels[1], 0, ctx->planes[p].pitch * ctx->planes[p].height);
  168. }
  169. return 0;
  170. }
  171. static av_cold void free_frame_buffers(Indeo3DecodeContext *ctx)
  172. {
  173. int p;
  174. for (p = 0; p < 3; p++) {
  175. av_freep(&ctx->planes[p].buffers[0]);
  176. av_freep(&ctx->planes[p].buffers[1]);
  177. ctx->planes[p].pixels[0] = ctx->planes[p].pixels[1] = 0;
  178. }
  179. }
  180. /**
  181. * Copy pixels of the cell(x + mv_x, y + mv_y) from the previous frame into
  182. * the cell(x, y) in the current frame.
  183. *
  184. * @param ctx pointer to the decoder context
  185. * @param plane pointer to the plane descriptor
  186. * @param cell pointer to the cell descriptor
  187. */
  188. static int copy_cell(Indeo3DecodeContext *ctx, Plane *plane, Cell *cell)
  189. {
  190. int h, w, mv_x, mv_y, offset, offset_dst;
  191. uint8_t *src, *dst;
  192. /* setup output and reference pointers */
  193. offset_dst = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
  194. dst = plane->pixels[ctx->buf_sel] + offset_dst;
  195. mv_y = cell->mv_ptr[0];
  196. mv_x = cell->mv_ptr[1];
  197. /* -1 because there is an extra line on top for prediction */
  198. if ((cell->ypos << 2) + mv_y < -1 || (cell->xpos << 2) + mv_x < 0 ||
  199. ((cell->ypos + cell->height) << 2) + mv_y > plane->height ||
  200. ((cell->xpos + cell->width) << 2) + mv_x > plane->width) {
  201. av_log(ctx->avctx, AV_LOG_ERROR,
  202. "Motion vectors point out of the frame.\n");
  203. return AVERROR_INVALIDDATA;
  204. }
  205. offset = offset_dst + mv_y * plane->pitch + mv_x;
  206. src = plane->pixels[ctx->buf_sel ^ 1] + offset;
  207. h = cell->height << 2;
  208. for (w = cell->width; w > 0;) {
  209. /* copy using 16xH blocks */
  210. if (!((cell->xpos << 2) & 15) && w >= 4) {
  211. for (; w >= 4; src += 16, dst += 16, w -= 4)
  212. ctx->hdsp.put_pixels_tab[0][0](dst, src, plane->pitch, h);
  213. }
  214. /* copy using 8xH blocks */
  215. if (!((cell->xpos << 2) & 7) && w >= 2) {
  216. ctx->hdsp.put_pixels_tab[1][0](dst, src, plane->pitch, h);
  217. w -= 2;
  218. src += 8;
  219. dst += 8;
  220. } else 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_RN32(src) + AV_RN32(ref)) >> 1) & 0x7F7F7F7FUL)
  232. #define AVG_64(dst, src, ref) \
  233. AV_WN64A(dst, ((AV_RN64(src) + AV_RN64(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 64-bit 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_RN64(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_RN64(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_RN16(ref ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
  301. AV_WN16A(dst + line_offset + 2,\
  302. (AV_RN16(ref + 2) + delta_tab->deltas[dyad2]) & 0x7F7F);\
  303. if (mode >= 3) {\
  304. if (is_top_of_cell && !cell->ypos) {\
  305. AV_COPY32U(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_RN32(ref )) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
  315. AV_WN32A(dst + row_offset + 4, \
  316. (replicate32(AV_RN32(ref + 4)) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
  317. } else { \
  318. AV_WN32A(dst + row_offset , \
  319. (AV_RN32(ref ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
  320. AV_WN32A(dst + row_offset + 4, \
  321. (AV_RN32(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_COPY64U(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_RN32(dst ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
  334. AV_WN32A(dst + 4 , \
  335. (AV_RN32(dst + 4 ) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
  336. AV_WN32A(dst + row_offset , \
  337. (AV_RN32(dst + row_offset ) + delta_tab->deltas_m10[dyad1]) & 0x7F7F7F7F);\
  338. AV_WN32A(dst + row_offset + 4, \
  339. (AV_RN32(dst + row_offset + 4) + delta_tab->deltas_m10[dyad2]) & 0x7F7F7F7F);\
  340. } else { \
  341. AV_WN16A(dst , \
  342. (AV_RN16(dst ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
  343. AV_WN16A(dst + 2 , \
  344. (AV_RN16(dst + 2 ) + delta_tab->deltas[dyad2]) & 0x7F7F);\
  345. AV_WN16A(dst + row_offset , \
  346. (AV_RN16(dst + row_offset ) + delta_tab->deltas[dyad1]) & 0x7F7F);\
  347. AV_WN16A(dst + row_offset + 2, \
  348. (AV_RN16(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. ptrdiff_t row_offset, 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 blk_row_offset, line_offset;
  364. blk_row_offset = (row_offset << (2 + v_zoom)) - (cell->width << 2);
  365. line_offset = v_zoom ? row_offset : 0;
  366. if (cell->height & v_zoom || cell->width & h_zoom)
  367. return IV3_BAD_DATA;
  368. for (y = 0; y < cell->height; is_first_row = 0, y += 1 + v_zoom) {
  369. for (x = 0; x < cell->width; x += 1 + h_zoom) {
  370. ref = ref_block;
  371. dst = block;
  372. if (rle_blocks > 0) {
  373. if (mode <= 4) {
  374. RLE_BLOCK_COPY;
  375. } else if (mode == 10 && !cell->mv_ptr) {
  376. RLE_BLOCK_COPY_8;
  377. }
  378. rle_blocks--;
  379. } else {
  380. for (line = 0; line < 4;) {
  381. num_lines = 1;
  382. is_top_of_cell = is_first_row && !line;
  383. /* select primary VQ table for odd, secondary for even lines */
  384. if (mode <= 4)
  385. delta_tab = delta[line & 1];
  386. else
  387. delta_tab = delta[1];
  388. BUFFER_PRECHECK;
  389. code = bytestream_get_byte(data_ptr);
  390. if (code < 248) {
  391. if (code < delta_tab->num_dyads) {
  392. BUFFER_PRECHECK;
  393. dyad1 = bytestream_get_byte(data_ptr);
  394. dyad2 = code;
  395. if (dyad1 >= delta_tab->num_dyads || dyad1 >= 248)
  396. return IV3_BAD_DATA;
  397. } else {
  398. /* process QUADS */
  399. code -= delta_tab->num_dyads;
  400. dyad1 = code / delta_tab->quad_exp;
  401. dyad2 = code % delta_tab->quad_exp;
  402. if (swap_quads[line & 1])
  403. FFSWAP(unsigned int, dyad1, dyad2);
  404. }
  405. if (mode <= 4) {
  406. APPLY_DELTA_4;
  407. } else if (mode == 10 && !cell->mv_ptr) {
  408. APPLY_DELTA_8;
  409. } else {
  410. APPLY_DELTA_1011_INTER;
  411. }
  412. } else {
  413. /* process RLE codes */
  414. switch (code) {
  415. case RLE_ESC_FC:
  416. skip_flag = 0;
  417. rle_blocks = 1;
  418. code = 253;
  419. /* FALLTHROUGH */
  420. case RLE_ESC_FF:
  421. case RLE_ESC_FE:
  422. case RLE_ESC_FD:
  423. num_lines = 257 - code - line;
  424. if (num_lines <= 0)
  425. return IV3_BAD_RLE;
  426. if (mode <= 4) {
  427. RLE_LINES_COPY;
  428. } else if (mode == 10 && !cell->mv_ptr) {
  429. RLE_LINES_COPY_M10;
  430. }
  431. break;
  432. case RLE_ESC_FB:
  433. BUFFER_PRECHECK;
  434. code = bytestream_get_byte(data_ptr);
  435. rle_blocks = (code & 0x1F) - 1; /* set block counter */
  436. if (code >= 64 || rle_blocks < 0)
  437. return IV3_BAD_COUNTER;
  438. skip_flag = code & 0x20;
  439. num_lines = 4 - line; /* enforce next block processing */
  440. if (mode >= 10 || (cell->mv_ptr || !skip_flag)) {
  441. if (mode <= 4) {
  442. RLE_LINES_COPY;
  443. } else if (mode == 10 && !cell->mv_ptr) {
  444. RLE_LINES_COPY_M10;
  445. }
  446. }
  447. break;
  448. case RLE_ESC_F9:
  449. skip_flag = 1;
  450. rle_blocks = 1;
  451. /* FALLTHROUGH */
  452. case RLE_ESC_FA:
  453. if (line)
  454. return IV3_BAD_RLE;
  455. num_lines = 4; /* enforce next block processing */
  456. if (cell->mv_ptr) {
  457. if (mode <= 4) {
  458. RLE_LINES_COPY;
  459. } else if (mode == 10 && !cell->mv_ptr) {
  460. RLE_LINES_COPY_M10;
  461. }
  462. }
  463. break;
  464. default:
  465. return IV3_UNSUPPORTED;
  466. }
  467. }
  468. line += num_lines;
  469. ref += row_offset * (num_lines << v_zoom);
  470. dst += row_offset * (num_lines << v_zoom);
  471. }
  472. }
  473. /* move to next horizontal block */
  474. block += 4 << h_zoom;
  475. ref_block += 4 << h_zoom;
  476. }
  477. /* move to next line of blocks */
  478. ref_block += blk_row_offset;
  479. block += blk_row_offset;
  480. }
  481. return IV3_NOERR;
  482. }
  483. /**
  484. * Decode a vector-quantized cell.
  485. * It consists of several routines, each of which handles one or more "modes"
  486. * with which a cell can be encoded.
  487. *
  488. * @param ctx pointer to the decoder context
  489. * @param avctx ptr to the AVCodecContext
  490. * @param plane pointer to the plane descriptor
  491. * @param cell pointer to the cell descriptor
  492. * @param data_ptr pointer to the compressed data
  493. * @param last_ptr pointer to the last byte to catch reads past end of buffer
  494. * @return number of consumed bytes or negative number in case of error
  495. */
  496. static int decode_cell(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
  497. Plane *plane, Cell *cell, const uint8_t *data_ptr,
  498. const uint8_t *last_ptr)
  499. {
  500. int x, mv_x, mv_y, mode, vq_index, prim_indx, second_indx;
  501. int zoom_fac;
  502. int offset, error = 0, swap_quads[2];
  503. uint8_t code, *block, *ref_block = 0;
  504. const vqEntry *delta[2];
  505. const uint8_t *data_start = data_ptr;
  506. /* get coding mode and VQ table index from the VQ descriptor byte */
  507. code = *data_ptr++;
  508. mode = code >> 4;
  509. vq_index = code & 0xF;
  510. /* setup output and reference pointers */
  511. offset = (cell->ypos << 2) * plane->pitch + (cell->xpos << 2);
  512. block = plane->pixels[ctx->buf_sel] + offset;
  513. if (!cell->mv_ptr) {
  514. /* use previous line as reference for INTRA cells */
  515. ref_block = block - plane->pitch;
  516. } else if (mode >= 10) {
  517. /* for mode 10 and 11 INTER first copy the predicted cell into the current one */
  518. /* so we don't need to do data copying for each RLE code later */
  519. int ret = copy_cell(ctx, plane, cell);
  520. if (ret < 0)
  521. return ret;
  522. } else {
  523. /* set the pointer to the reference pixels for modes 0-4 INTER */
  524. mv_y = cell->mv_ptr[0];
  525. mv_x = cell->mv_ptr[1];
  526. /* -1 because there is an extra line on top for prediction */
  527. if ((cell->ypos << 2) + mv_y < -1 || (cell->xpos << 2) + mv_x < 0 ||
  528. ((cell->ypos + cell->height) << 2) + mv_y > plane->height ||
  529. ((cell->xpos + cell->width) << 2) + mv_x > plane->width) {
  530. av_log(ctx->avctx, AV_LOG_ERROR,
  531. "Motion vectors point out of the frame.\n");
  532. return AVERROR_INVALIDDATA;
  533. }
  534. offset += mv_y * plane->pitch + mv_x;
  535. ref_block = plane->pixels[ctx->buf_sel ^ 1] + offset;
  536. }
  537. /* select VQ tables as follows: */
  538. /* modes 0 and 3 use only the primary table for all lines in a block */
  539. /* while modes 1 and 4 switch between primary and secondary tables on alternate lines */
  540. if (mode == 1 || mode == 4) {
  541. code = ctx->alt_quant[vq_index];
  542. prim_indx = (code >> 4) + ctx->cb_offset;
  543. second_indx = (code & 0xF) + ctx->cb_offset;
  544. } else {
  545. vq_index += ctx->cb_offset;
  546. prim_indx = second_indx = vq_index;
  547. }
  548. if (prim_indx >= 24 || second_indx >= 24) {
  549. av_log(avctx, AV_LOG_ERROR, "Invalid VQ table indexes! Primary: %d, secondary: %d!\n",
  550. prim_indx, second_indx);
  551. return AVERROR_INVALIDDATA;
  552. }
  553. delta[0] = &vq_tab[second_indx];
  554. delta[1] = &vq_tab[prim_indx];
  555. swap_quads[0] = second_indx >= 16;
  556. swap_quads[1] = prim_indx >= 16;
  557. /* requantize the prediction if VQ index of this cell differs from VQ index */
  558. /* of the predicted cell in order to avoid overflows. */
  559. if (vq_index >= 8 && ref_block) {
  560. for (x = 0; x < cell->width << 2; x++)
  561. ref_block[x] = requant_tab[vq_index & 7][ref_block[x]];
  562. }
  563. error = IV3_NOERR;
  564. switch (mode) {
  565. case 0: /*------------------ MODES 0 & 1 (4x4 block processing) --------------------*/
  566. case 1:
  567. case 3: /*------------------ MODES 3 & 4 (4x8 block processing) --------------------*/
  568. case 4:
  569. if (mode >= 3 && cell->mv_ptr) {
  570. av_log(avctx, AV_LOG_ERROR, "Attempt to apply Mode 3/4 to an INTER cell!\n");
  571. return AVERROR_INVALIDDATA;
  572. }
  573. zoom_fac = mode >= 3;
  574. error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
  575. 0, zoom_fac, mode, delta, swap_quads,
  576. &data_ptr, last_ptr);
  577. break;
  578. case 10: /*-------------------- MODE 10 (8x8 block processing) ---------------------*/
  579. case 11: /*----------------- MODE 11 (4x8 INTER block processing) ------------------*/
  580. if (mode == 10 && !cell->mv_ptr) { /* MODE 10 INTRA processing */
  581. error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
  582. 1, 1, mode, delta, swap_quads,
  583. &data_ptr, last_ptr);
  584. } else { /* mode 10 and 11 INTER processing */
  585. if (mode == 11 && !cell->mv_ptr) {
  586. av_log(avctx, AV_LOG_ERROR, "Attempt to use Mode 11 for an INTRA cell!\n");
  587. return AVERROR_INVALIDDATA;
  588. }
  589. zoom_fac = mode == 10;
  590. error = decode_cell_data(ctx, cell, block, ref_block, plane->pitch,
  591. zoom_fac, 1, mode, delta, swap_quads,
  592. &data_ptr, last_ptr);
  593. }
  594. break;
  595. default:
  596. av_log(avctx, AV_LOG_ERROR, "Unsupported coding mode: %d\n", mode);
  597. return AVERROR_INVALIDDATA;
  598. }//switch mode
  599. switch (error) {
  600. case IV3_BAD_RLE:
  601. av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE code %X is not allowed at the current line\n",
  602. mode, data_ptr[-1]);
  603. return AVERROR_INVALIDDATA;
  604. case IV3_BAD_DATA:
  605. av_log(avctx, AV_LOG_ERROR, "Mode %d: invalid VQ data\n", mode);
  606. return AVERROR_INVALIDDATA;
  607. case IV3_BAD_COUNTER:
  608. av_log(avctx, AV_LOG_ERROR, "Mode %d: RLE-FB invalid counter: %d\n", mode, code);
  609. return AVERROR_INVALIDDATA;
  610. case IV3_UNSUPPORTED:
  611. av_log(avctx, AV_LOG_ERROR, "Mode %d: unsupported RLE code: %X\n", mode, data_ptr[-1]);
  612. return AVERROR_INVALIDDATA;
  613. case IV3_OUT_OF_DATA:
  614. av_log(avctx, AV_LOG_ERROR, "Mode %d: attempt to read past end of buffer\n", mode);
  615. return AVERROR_INVALIDDATA;
  616. }
  617. return data_ptr - data_start; /* report number of bytes consumed from the input buffer */
  618. }
  619. /* Binary tree codes. */
  620. enum {
  621. H_SPLIT = 0,
  622. V_SPLIT = 1,
  623. INTRA_NULL = 2,
  624. INTER_DATA = 3
  625. };
  626. #define SPLIT_CELL(size, new_size) (new_size) = ((size) > 2) ? ((((size) + 2) >> 2) << 1) : 1
  627. #define UPDATE_BITPOS(n) \
  628. ctx->skip_bits += (n); \
  629. ctx->need_resync = 1
  630. #define RESYNC_BITSTREAM \
  631. if (ctx->need_resync && !(bitstream_tell(&ctx->bc) & 7)) { \
  632. bitstream_skip(&ctx->bc, ctx->skip_bits); \
  633. ctx->skip_bits = 0; \
  634. ctx->need_resync = 0; \
  635. }
  636. #define CHECK_CELL \
  637. if (curr_cell.xpos + curr_cell.width > (plane->width >> 2) || \
  638. curr_cell.ypos + curr_cell.height > (plane->height >> 2)) { \
  639. av_log(avctx, AV_LOG_ERROR, "Invalid cell: x=%d, y=%d, w=%d, h=%d\n", \
  640. curr_cell.xpos, curr_cell.ypos, curr_cell.width, curr_cell.height); \
  641. return AVERROR_INVALIDDATA; \
  642. }
  643. static int parse_bintree(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
  644. Plane *plane, int code, Cell *ref_cell,
  645. const int depth, const int strip_width)
  646. {
  647. Cell curr_cell;
  648. int bytes_used, ret;
  649. if (depth <= 0) {
  650. av_log(avctx, AV_LOG_ERROR, "Stack overflow (corrupted binary tree)!\n");
  651. return AVERROR_INVALIDDATA; // unwind recursion
  652. }
  653. curr_cell = *ref_cell; // clone parent cell
  654. if (code == H_SPLIT) {
  655. SPLIT_CELL(ref_cell->height, curr_cell.height);
  656. ref_cell->ypos += curr_cell.height;
  657. ref_cell->height -= curr_cell.height;
  658. if (ref_cell->height <= 0 || curr_cell.height <= 0)
  659. return AVERROR_INVALIDDATA;
  660. } else if (code == V_SPLIT) {
  661. if (curr_cell.width > strip_width) {
  662. /* split strip */
  663. curr_cell.width = (curr_cell.width <= (strip_width << 1) ? 1 : 2) * strip_width;
  664. } else
  665. SPLIT_CELL(ref_cell->width, curr_cell.width);
  666. ref_cell->xpos += curr_cell.width;
  667. ref_cell->width -= curr_cell.width;
  668. if (ref_cell->width <= 0 || curr_cell.width <= 0)
  669. return AVERROR_INVALIDDATA;
  670. }
  671. while (1) { /* loop until return */
  672. RESYNC_BITSTREAM;
  673. switch (code = bitstream_read(&ctx->bc, 2)) {
  674. case H_SPLIT:
  675. case V_SPLIT:
  676. if (parse_bintree(ctx, avctx, plane, code, &curr_cell, depth - 1, strip_width))
  677. return AVERROR_INVALIDDATA;
  678. break;
  679. case INTRA_NULL:
  680. if (!curr_cell.tree) { /* MC tree INTRA code */
  681. curr_cell.mv_ptr = 0; /* mark the current strip as INTRA */
  682. curr_cell.tree = 1; /* enter the VQ tree */
  683. } else { /* VQ tree NULL code */
  684. RESYNC_BITSTREAM;
  685. code = bitstream_read(&ctx->bc, 2);
  686. if (code >= 2) {
  687. av_log(avctx, AV_LOG_ERROR, "Invalid VQ_NULL code: %d\n", code);
  688. return AVERROR_INVALIDDATA;
  689. }
  690. if (code == 1)
  691. av_log(avctx, AV_LOG_ERROR, "SkipCell procedure not implemented yet!\n");
  692. CHECK_CELL
  693. if (!curr_cell.mv_ptr)
  694. return AVERROR_INVALIDDATA;
  695. ret = copy_cell(ctx, plane, &curr_cell);
  696. return ret;
  697. }
  698. break;
  699. case INTER_DATA:
  700. if (!curr_cell.tree) { /* MC tree INTER code */
  701. unsigned mv_idx;
  702. /* get motion vector index and setup the pointer to the mv set */
  703. if (!ctx->need_resync)
  704. ctx->next_cell_data = &ctx->bc.buffer[(bitstream_tell(&ctx->bc) + 7) >> 3];
  705. mv_idx = *(ctx->next_cell_data++);
  706. if (mv_idx >= ctx->num_vectors) {
  707. av_log(avctx, AV_LOG_ERROR, "motion vector index out of range\n");
  708. return AVERROR_INVALIDDATA;
  709. }
  710. curr_cell.mv_ptr = &ctx->mc_vectors[mv_idx << 1];
  711. curr_cell.tree = 1; /* enter the VQ tree */
  712. UPDATE_BITPOS(8);
  713. } else { /* VQ tree DATA code */
  714. if (!ctx->need_resync)
  715. ctx->next_cell_data = &ctx->bc.buffer[(bitstream_tell(&ctx->bc) + 7) >> 3];
  716. CHECK_CELL
  717. bytes_used = decode_cell(ctx, avctx, plane, &curr_cell,
  718. ctx->next_cell_data, ctx->last_byte);
  719. if (bytes_used < 0)
  720. return AVERROR_INVALIDDATA;
  721. UPDATE_BITPOS(bytes_used << 3);
  722. ctx->next_cell_data += bytes_used;
  723. return 0;
  724. }
  725. break;
  726. }
  727. }//while
  728. }
  729. static int decode_plane(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
  730. Plane *plane, const uint8_t *data, int32_t data_size,
  731. int32_t strip_width)
  732. {
  733. Cell curr_cell;
  734. unsigned num_vectors;
  735. /* each plane data starts with mc_vector_count field, */
  736. /* an optional array of motion vectors followed by the vq data */
  737. num_vectors = bytestream_get_le32(&data);
  738. if (num_vectors > 256) {
  739. av_log(ctx->avctx, AV_LOG_ERROR,
  740. "Read invalid number of motion vectors %d\n", num_vectors);
  741. return AVERROR_INVALIDDATA;
  742. }
  743. if (num_vectors * 2 >= data_size)
  744. return AVERROR_INVALIDDATA;
  745. ctx->num_vectors = num_vectors;
  746. ctx->mc_vectors = num_vectors ? data : 0;
  747. /* init the bitreader */
  748. bitstream_init8(&ctx->bc, &data[num_vectors * 2], data_size - num_vectors * 2);
  749. ctx->skip_bits = 0;
  750. ctx->need_resync = 0;
  751. ctx->last_byte = data + data_size - 1;
  752. /* initialize the 1st cell and set its dimensions to whole plane */
  753. curr_cell.xpos = curr_cell.ypos = 0;
  754. curr_cell.width = plane->width >> 2;
  755. curr_cell.height = plane->height >> 2;
  756. curr_cell.tree = 0; // we are in the MC tree now
  757. curr_cell.mv_ptr = 0; // no motion vector = INTRA cell
  758. return parse_bintree(ctx, avctx, plane, INTRA_NULL, &curr_cell, CELL_STACK_MAX, strip_width);
  759. }
  760. #define OS_HDR_ID MKBETAG('F', 'R', 'M', 'H')
  761. static int decode_frame_headers(Indeo3DecodeContext *ctx, AVCodecContext *avctx,
  762. const uint8_t *buf, int buf_size)
  763. {
  764. GetByteContext gb;
  765. const uint8_t *bs_hdr;
  766. uint32_t frame_num, word2, check_sum, data_size;
  767. uint32_t y_offset, u_offset, v_offset, starts[3], ends[3];
  768. uint16_t height, width;
  769. int i, j;
  770. bytestream2_init(&gb, buf, buf_size);
  771. /* parse and check the OS header */
  772. frame_num = bytestream2_get_le32(&gb);
  773. word2 = bytestream2_get_le32(&gb);
  774. check_sum = bytestream2_get_le32(&gb);
  775. data_size = bytestream2_get_le32(&gb);
  776. if ((frame_num ^ word2 ^ data_size ^ OS_HDR_ID) != check_sum) {
  777. av_log(avctx, AV_LOG_ERROR, "OS header checksum mismatch!\n");
  778. return AVERROR_INVALIDDATA;
  779. }
  780. /* parse the bitstream header */
  781. bs_hdr = gb.buffer;
  782. if (bytestream2_get_le16(&gb) != 32) {
  783. av_log(avctx, AV_LOG_ERROR, "Unsupported codec version!\n");
  784. return AVERROR_INVALIDDATA;
  785. }
  786. ctx->frame_num = frame_num;
  787. ctx->frame_flags = bytestream2_get_le16(&gb);
  788. ctx->data_size = (bytestream2_get_le32(&gb) + 7) >> 3;
  789. ctx->cb_offset = bytestream2_get_byte(&gb);
  790. if (ctx->data_size == 16)
  791. return 4;
  792. ctx->data_size = FFMIN(ctx->data_size, buf_size - 16);
  793. bytestream2_skip(&gb, 3); // skip reserved byte and checksum
  794. /* check frame dimensions */
  795. height = bytestream2_get_le16(&gb);
  796. width = bytestream2_get_le16(&gb);
  797. if (av_image_check_size(width, height, 0, avctx))
  798. return AVERROR_INVALIDDATA;
  799. if (width != ctx->width || height != ctx->height) {
  800. int res;
  801. ff_dlog(avctx, "Frame dimensions changed!\n");
  802. if (width < 16 || width > 640 ||
  803. height < 16 || height > 480 ||
  804. width & 3 || height & 3) {
  805. av_log(avctx, AV_LOG_ERROR,
  806. "Invalid picture dimensions: %d x %d!\n", width, height);
  807. return AVERROR_INVALIDDATA;
  808. }
  809. ctx->width = width;
  810. ctx->height = height;
  811. free_frame_buffers(ctx);
  812. if ((res = allocate_frame_buffers(ctx, avctx)) < 0)
  813. return res;
  814. if ((res = ff_set_dimensions(avctx, width, height)) < 0)
  815. return res;
  816. }
  817. y_offset = bytestream2_get_le32(&gb);
  818. v_offset = bytestream2_get_le32(&gb);
  819. u_offset = bytestream2_get_le32(&gb);
  820. bytestream2_skip(&gb, 4);
  821. /* unfortunately there is no common order of planes in the buffer */
  822. /* so we use that sorting algo for determining planes data sizes */
  823. starts[0] = y_offset;
  824. starts[1] = v_offset;
  825. starts[2] = u_offset;
  826. for (j = 0; j < 3; j++) {
  827. ends[j] = ctx->data_size;
  828. for (i = 2; i >= 0; i--)
  829. if (starts[i] < ends[j] && starts[i] > starts[j])
  830. ends[j] = starts[i];
  831. }
  832. ctx->y_data_size = ends[0] - starts[0];
  833. ctx->v_data_size = ends[1] - starts[1];
  834. ctx->u_data_size = ends[2] - starts[2];
  835. if (FFMAX3(y_offset, v_offset, u_offset) >= ctx->data_size - 16 ||
  836. FFMIN3(y_offset, v_offset, u_offset) < gb.buffer - bs_hdr + 16 ||
  837. FFMIN3(ctx->y_data_size, ctx->v_data_size, ctx->u_data_size) <= 0) {
  838. av_log(avctx, AV_LOG_ERROR, "One of the y/u/v offsets is invalid\n");
  839. return AVERROR_INVALIDDATA;
  840. }
  841. ctx->y_data_ptr = bs_hdr + y_offset;
  842. ctx->v_data_ptr = bs_hdr + v_offset;
  843. ctx->u_data_ptr = bs_hdr + u_offset;
  844. ctx->alt_quant = gb.buffer;
  845. if (ctx->data_size == 16) {
  846. av_log(avctx, AV_LOG_DEBUG, "Sync frame encountered!\n");
  847. return 16;
  848. }
  849. if (ctx->frame_flags & BS_8BIT_PEL) {
  850. avpriv_request_sample(avctx, "8-bit pixel format");
  851. return AVERROR_PATCHWELCOME;
  852. }
  853. if (ctx->frame_flags & BS_MV_X_HALF || ctx->frame_flags & BS_MV_Y_HALF) {
  854. avpriv_request_sample(avctx, "Halfpel motion vectors");
  855. return AVERROR_PATCHWELCOME;
  856. }
  857. return 0;
  858. }
  859. /**
  860. * Convert and output the current plane.
  861. * All pixel values will be upsampled by shifting right by one bit.
  862. *
  863. * @param[in] plane pointer to the descriptor of the plane being processed
  864. * @param[in] buf_sel indicates which frame buffer the input data stored in
  865. * @param[out] dst pointer to the buffer receiving converted pixels
  866. * @param[in] dst_pitch pitch for moving to the next y line
  867. * @param[in] dst_height output plane height
  868. */
  869. static void output_plane(const Plane *plane, int buf_sel, uint8_t *dst,
  870. ptrdiff_t dst_pitch, int dst_height)
  871. {
  872. int x,y;
  873. const uint8_t *src = plane->pixels[buf_sel];
  874. ptrdiff_t pitch = plane->pitch;
  875. dst_height = FFMIN(dst_height, plane->height);
  876. for (y = 0; y < dst_height; y++) {
  877. /* convert four pixels at once using SWAR */
  878. for (x = 0; x < plane->width >> 2; x++) {
  879. AV_WN32A(dst, (AV_RN32A(src) & 0x7F7F7F7F) << 1);
  880. src += 4;
  881. dst += 4;
  882. }
  883. for (x <<= 2; x < plane->width; x++)
  884. *dst++ = *src++ << 1;
  885. src += pitch - plane->width;
  886. dst += dst_pitch - plane->width;
  887. }
  888. }
  889. static av_cold int decode_init(AVCodecContext *avctx)
  890. {
  891. Indeo3DecodeContext *ctx = avctx->priv_data;
  892. ctx->avctx = avctx;
  893. ctx->width = avctx->width;
  894. ctx->height = avctx->height;
  895. avctx->pix_fmt = AV_PIX_FMT_YUV410P;
  896. build_requant_tab();
  897. ff_hpeldsp_init(&ctx->hdsp, avctx->flags);
  898. allocate_frame_buffers(ctx, avctx);
  899. return 0;
  900. }
  901. static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
  902. AVPacket *avpkt)
  903. {
  904. Indeo3DecodeContext *ctx = avctx->priv_data;
  905. const uint8_t *buf = avpkt->data;
  906. int buf_size = avpkt->size;
  907. AVFrame *frame = data;
  908. int res;
  909. res = decode_frame_headers(ctx, avctx, buf, buf_size);
  910. if (res < 0)
  911. return res;
  912. /* skip sync(null) frames */
  913. if (res) {
  914. // we have processed 16 bytes but no data was decoded
  915. *got_frame = 0;
  916. return buf_size;
  917. }
  918. /* skip droppable INTER frames if requested */
  919. if (ctx->frame_flags & BS_NONREF &&
  920. (avctx->skip_frame >= AVDISCARD_NONREF))
  921. return 0;
  922. /* skip INTER frames if requested */
  923. if (!(ctx->frame_flags & BS_KEYFRAME) && avctx->skip_frame >= AVDISCARD_NONKEY)
  924. return 0;
  925. /* use BS_BUFFER flag for buffer switching */
  926. ctx->buf_sel = (ctx->frame_flags >> BS_BUFFER) & 1;
  927. /* decode luma plane */
  928. if ((res = decode_plane(ctx, avctx, ctx->planes, ctx->y_data_ptr, ctx->y_data_size, 40)))
  929. return res;
  930. /* decode chroma planes */
  931. if ((res = decode_plane(ctx, avctx, &ctx->planes[1], ctx->u_data_ptr, ctx->u_data_size, 10)))
  932. return res;
  933. if ((res = decode_plane(ctx, avctx, &ctx->planes[2], ctx->v_data_ptr, ctx->v_data_size, 10)))
  934. return res;
  935. if ((res = ff_get_buffer(avctx, frame, 0)) < 0) {
  936. av_log(ctx->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
  937. return res;
  938. }
  939. output_plane(&ctx->planes[0], ctx->buf_sel,
  940. frame->data[0], frame->linesize[0],
  941. avctx->height);
  942. output_plane(&ctx->planes[1], ctx->buf_sel,
  943. frame->data[1], frame->linesize[1],
  944. (avctx->height + 3) >> 2);
  945. output_plane(&ctx->planes[2], ctx->buf_sel,
  946. frame->data[2], frame->linesize[2],
  947. (avctx->height + 3) >> 2);
  948. *got_frame = 1;
  949. return buf_size;
  950. }
  951. static av_cold int decode_close(AVCodecContext *avctx)
  952. {
  953. free_frame_buffers(avctx->priv_data);
  954. return 0;
  955. }
  956. AVCodec ff_indeo3_decoder = {
  957. .name = "indeo3",
  958. .long_name = NULL_IF_CONFIG_SMALL("Intel Indeo 3"),
  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 = AV_CODEC_CAP_DR1,
  966. };