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.

332 lines
12KB

  1. /*
  2. * XVideo Motion Compensation
  3. * Copyright (c) 2003 Ivan Kalvachev
  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. #include <limits.h>
  22. #include <X11/extensions/XvMC.h>
  23. #include "avcodec.h"
  24. #include "mpegvideo.h"
  25. #undef NDEBUG
  26. #include <assert.h>
  27. #include "xvmc.h"
  28. #include "xvmc_internal.h"
  29. /**
  30. * Initialize the block field of the MpegEncContext pointer passed as
  31. * parameter after making sure that the data is not corrupted.
  32. * In order to implement something like direct rendering instead of decoding
  33. * coefficients in s->blocks and then copying them, copy them directly
  34. * into the data_blocks array provided by xvmc.
  35. */
  36. void ff_xvmc_init_block(MpegEncContext *s)
  37. {
  38. struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
  39. assert(render && render->xvmc_id == AV_XVMC_ID);
  40. s->block = (int16_t (*)[64])(render->data_blocks + render->next_free_data_block_num * 64);
  41. }
  42. /**
  43. * Fill individual block pointers, so there are no gaps in the data_block array
  44. * in case not all blocks in the macroblock are coded.
  45. */
  46. void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
  47. {
  48. int i, j = 0;
  49. const int mb_block_count = 4 + (1 << s->chroma_format);
  50. cbp <<= 12-mb_block_count;
  51. for (i = 0; i < mb_block_count; i++) {
  52. if (cbp & (1 << 11))
  53. s->pblocks[i] = &s->block[j++];
  54. else
  55. s->pblocks[i] = NULL;
  56. cbp += cbp;
  57. }
  58. }
  59. /**
  60. * Find and store the surfaces that are used as reference frames.
  61. * This function should be called for every new field and/or frame.
  62. * It should be safe to call the function a few times for the same field.
  63. */
  64. int ff_xvmc_field_start(MpegEncContext *s, AVCodecContext *avctx)
  65. {
  66. struct xvmc_pix_fmt *last, *next, *render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
  67. const int mb_block_count = 4 + (1 << s->chroma_format);
  68. assert(avctx);
  69. if (!render || render->xvmc_id != AV_XVMC_ID ||
  70. !render->data_blocks || !render->mv_blocks ||
  71. (unsigned int)render->allocated_mv_blocks > INT_MAX/(64*6) ||
  72. (unsigned int)render->allocated_data_blocks > INT_MAX/64 ||
  73. !render->p_surface) {
  74. av_log(avctx, AV_LOG_ERROR,
  75. "Render token doesn't look as expected.\n");
  76. return -1; // make sure that this is a render packet
  77. }
  78. if (render->filled_mv_blocks_num) {
  79. av_log(avctx, AV_LOG_ERROR,
  80. "Rendering surface contains %i unprocessed blocks.\n",
  81. render->filled_mv_blocks_num);
  82. return -1;
  83. }
  84. if (render->allocated_mv_blocks < 1 ||
  85. render->allocated_data_blocks < render->allocated_mv_blocks*mb_block_count ||
  86. render->start_mv_blocks_num >= render->allocated_mv_blocks ||
  87. render->next_free_data_block_num >
  88. render->allocated_data_blocks -
  89. mb_block_count*(render->allocated_mv_blocks-render->start_mv_blocks_num)) {
  90. av_log(avctx, AV_LOG_ERROR,
  91. "Rendering surface doesn't provide enough block structures to work with.\n");
  92. return -1;
  93. }
  94. render->picture_structure = s->picture_structure;
  95. render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD;
  96. render->p_future_surface = NULL;
  97. render->p_past_surface = NULL;
  98. switch(s->pict_type) {
  99. case AV_PICTURE_TYPE_I:
  100. return 0; // no prediction from other frames
  101. case AV_PICTURE_TYPE_B:
  102. next = (struct xvmc_pix_fmt*)s->next_picture.f.data[2];
  103. if (!next)
  104. return -1;
  105. if (next->xvmc_id != AV_XVMC_ID)
  106. return -1;
  107. render->p_future_surface = next->p_surface;
  108. // no return here, going to set forward prediction
  109. case AV_PICTURE_TYPE_P:
  110. last = (struct xvmc_pix_fmt*)s->last_picture.f.data[2];
  111. if (!last)
  112. last = render; // predict second field from the first
  113. if (last->xvmc_id != AV_XVMC_ID)
  114. return -1;
  115. render->p_past_surface = last->p_surface;
  116. return 0;
  117. }
  118. return -1;
  119. }
  120. /**
  121. * Complete frame/field rendering by passing any remaining blocks.
  122. * Normally ff_draw_horiz_band() is called for each slice, however,
  123. * some leftover blocks, for example from error_resilience(), may remain.
  124. * It should be safe to call the function a few times for the same field.
  125. */
  126. void ff_xvmc_field_end(MpegEncContext *s)
  127. {
  128. struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
  129. assert(render);
  130. if (render->filled_mv_blocks_num > 0)
  131. ff_mpeg_draw_horiz_band(s, 0, 0);
  132. }
  133. /**
  134. * Synthesize the data needed by XvMC to render one macroblock of data.
  135. * Fill all relevant fields, if necessary do IDCT.
  136. */
  137. void ff_xvmc_decode_mb(MpegEncContext *s)
  138. {
  139. XvMCMacroBlock *mv_block;
  140. struct xvmc_pix_fmt *render;
  141. int i, cbp, blocks_per_mb;
  142. const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
  143. if (s->encoding) {
  144. av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
  145. return;
  146. }
  147. // from MPV_decode_mb(), update DC predictors for P macroblocks
  148. if (!s->mb_intra) {
  149. s->last_dc[0] =
  150. s->last_dc[1] =
  151. s->last_dc[2] = 128 << s->intra_dc_precision;
  152. }
  153. // MC doesn't skip blocks
  154. s->mb_skipped = 0;
  155. // Do I need to export quant when I could not perform postprocessing?
  156. // Anyway, it doesn't hurt.
  157. s->current_picture.qscale_table[mb_xy] = s->qscale;
  158. // start of XVMC-specific code
  159. render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
  160. assert(render);
  161. assert(render->xvmc_id == AV_XVMC_ID);
  162. assert(render->mv_blocks);
  163. // take the next free macroblock
  164. mv_block = &render->mv_blocks[render->start_mv_blocks_num +
  165. render->filled_mv_blocks_num];
  166. mv_block->x = s->mb_x;
  167. mv_block->y = s->mb_y;
  168. mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
  169. if (s->mb_intra) {
  170. mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
  171. } else {
  172. mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
  173. if (s->mv_dir & MV_DIR_FORWARD) {
  174. mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
  175. // PMV[n][dir][xy] = mv[dir][n][xy]
  176. mv_block->PMV[0][0][0] = s->mv[0][0][0];
  177. mv_block->PMV[0][0][1] = s->mv[0][0][1];
  178. mv_block->PMV[1][0][0] = s->mv[0][1][0];
  179. mv_block->PMV[1][0][1] = s->mv[0][1][1];
  180. }
  181. if (s->mv_dir & MV_DIR_BACKWARD) {
  182. mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
  183. mv_block->PMV[0][1][0] = s->mv[1][0][0];
  184. mv_block->PMV[0][1][1] = s->mv[1][0][1];
  185. mv_block->PMV[1][1][0] = s->mv[1][1][0];
  186. mv_block->PMV[1][1][1] = s->mv[1][1][1];
  187. }
  188. switch(s->mv_type) {
  189. case MV_TYPE_16X16:
  190. mv_block->motion_type = XVMC_PREDICTION_FRAME;
  191. break;
  192. case MV_TYPE_16X8:
  193. mv_block->motion_type = XVMC_PREDICTION_16x8;
  194. break;
  195. case MV_TYPE_FIELD:
  196. mv_block->motion_type = XVMC_PREDICTION_FIELD;
  197. if (s->picture_structure == PICT_FRAME) {
  198. mv_block->PMV[0][0][1] <<= 1;
  199. mv_block->PMV[1][0][1] <<= 1;
  200. mv_block->PMV[0][1][1] <<= 1;
  201. mv_block->PMV[1][1][1] <<= 1;
  202. }
  203. break;
  204. case MV_TYPE_DMV:
  205. mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
  206. if (s->picture_structure == PICT_FRAME) {
  207. mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top
  208. mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1;
  209. mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom
  210. mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1;
  211. mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom
  212. mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01
  213. mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top
  214. mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11
  215. } else {
  216. mv_block->PMV[0][1][0] = s->mv[0][2][0]; // dmv00
  217. mv_block->PMV[0][1][1] = s->mv[0][2][1]; // dmv01
  218. }
  219. break;
  220. default:
  221. assert(0);
  222. }
  223. mv_block->motion_vertical_field_select = 0;
  224. // set correct field references
  225. if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
  226. mv_block->motion_vertical_field_select |= s->field_select[0][0];
  227. mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
  228. mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
  229. mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3;
  230. }
  231. } // !intra
  232. // time to handle data blocks
  233. mv_block->index = render->next_free_data_block_num;
  234. blocks_per_mb = 6;
  235. if (s->chroma_format >= 2) {
  236. blocks_per_mb = 4 + (1 << s->chroma_format);
  237. }
  238. // calculate cbp
  239. cbp = 0;
  240. for (i = 0; i < blocks_per_mb; i++) {
  241. cbp += cbp;
  242. if (s->block_last_index[i] >= 0)
  243. cbp++;
  244. }
  245. if (s->flags & CODEC_FLAG_GRAY) {
  246. if (s->mb_intra) { // intra frames are always full chroma blocks
  247. for (i = 4; i < blocks_per_mb; i++) {
  248. memset(s->pblocks[i], 0, sizeof(*s->pblocks[i])); // so we need to clear them
  249. if (!render->unsigned_intra)
  250. *s->pblocks[i][0] = 1 << 10;
  251. }
  252. } else {
  253. cbp &= 0xf << (blocks_per_mb - 4);
  254. blocks_per_mb = 4; // luminance blocks only
  255. }
  256. }
  257. mv_block->coded_block_pattern = cbp;
  258. if (cbp == 0)
  259. mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
  260. for (i = 0; i < blocks_per_mb; i++) {
  261. if (s->block_last_index[i] >= 0) {
  262. // I do not have unsigned_intra MOCO to test, hope it is OK.
  263. if (s->mb_intra && (render->idct || !render->unsigned_intra))
  264. *s->pblocks[i][0] -= 1 << 10;
  265. if (!render->idct) {
  266. s->dsp.idct(*s->pblocks[i]);
  267. /* It is unclear if MC hardware requires pixel diff values to be
  268. * in the range [-255;255]. TODO: Clipping if such hardware is
  269. * ever found. As of now it would only be an unnecessary
  270. * slowdown. */
  271. }
  272. // copy blocks only if the codec doesn't support pblocks reordering
  273. if (s->avctx->xvmc_acceleration == 1) {
  274. memcpy(&render->data_blocks[render->next_free_data_block_num*64],
  275. s->pblocks[i], sizeof(*s->pblocks[i]));
  276. }
  277. render->next_free_data_block_num++;
  278. }
  279. }
  280. render->filled_mv_blocks_num++;
  281. assert(render->filled_mv_blocks_num <= render->allocated_mv_blocks);
  282. assert(render->next_free_data_block_num <= render->allocated_data_blocks);
  283. /* The above conditions should not be able to fail as long as this function
  284. * is used and the following 'if ()' automatically calls a callback to free
  285. * blocks. */
  286. if (render->filled_mv_blocks_num == render->allocated_mv_blocks)
  287. ff_mpeg_draw_horiz_band(s, 0, 0);
  288. }