/libavcodec/mpegvideo_xvmc.c

http://github.com/FFmpeg/FFmpeg · C · 376 lines · 263 code · 47 blank · 66 comment · 55 complexity · 964ee8ed28a6dcebc7b38ef652eaf311 MD5 · raw file

  1. /*
  2. * XVideo Motion Compensation
  3. * Copyright (c) 2003 Ivan Kalvachev
  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. #include <limits.h>
  22. #include <X11/extensions/XvMC.h>
  23. #include "avcodec.h"
  24. #include "mpegutils.h"
  25. #include "mpegvideo.h"
  26. #undef NDEBUG
  27. #include <assert.h>
  28. #include "xvmc.h"
  29. #include "xvmc_internal.h"
  30. #include "version.h"
  31. /**
  32. * Initialize the block field of the MpegEncContext pointer passed as
  33. * parameter after making sure that the data is not corrupted.
  34. * In order to implement something like direct rendering instead of decoding
  35. * coefficients in s->blocks and then copying them, copy them directly
  36. * into the data_blocks array provided by xvmc.
  37. */
  38. void ff_xvmc_init_block(MpegEncContext *s)
  39. {
  40. struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
  41. assert(render && render->xvmc_id == AV_XVMC_ID);
  42. s->block = (int16_t (*)[64])(render->data_blocks + render->next_free_data_block_num * 64);
  43. }
  44. static void exchange_uv(MpegEncContext *s)
  45. {
  46. int16_t (*tmp)[64];
  47. tmp = s->pblocks[4];
  48. s->pblocks[4] = s->pblocks[5];
  49. s->pblocks[5] = tmp;
  50. }
  51. /**
  52. * Fill individual block pointers, so there are no gaps in the data_block array
  53. * in case not all blocks in the macroblock are coded.
  54. */
  55. void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
  56. {
  57. int i, j = 0;
  58. const int mb_block_count = 4 + (1 << s->chroma_format);
  59. cbp <<= 12-mb_block_count;
  60. for (i = 0; i < mb_block_count; i++) {
  61. if (cbp & (1 << 11))
  62. s->pblocks[i] = &s->block[j++];
  63. else
  64. s->pblocks[i] = NULL;
  65. cbp += cbp;
  66. }
  67. if (s->swap_uv) {
  68. exchange_uv(s);
  69. }
  70. }
  71. /**
  72. * Find and store the surfaces that are used as reference frames.
  73. * This function should be called for every new field and/or frame.
  74. * It should be safe to call the function a few times for the same field.
  75. */
  76. static int ff_xvmc_field_start(AVCodecContext *avctx, const uint8_t *buf, uint32_t buf_size)
  77. {
  78. struct MpegEncContext *s = avctx->priv_data;
  79. struct xvmc_pix_fmt *last, *next, *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
  80. const int mb_block_count = 4 + (1 << s->chroma_format);
  81. assert(avctx);
  82. if (!render || render->xvmc_id != AV_XVMC_ID ||
  83. !render->data_blocks || !render->mv_blocks ||
  84. (unsigned int)render->allocated_mv_blocks > INT_MAX/(64*6) ||
  85. (unsigned int)render->allocated_data_blocks > INT_MAX/64 ||
  86. !render->p_surface) {
  87. av_log(avctx, AV_LOG_ERROR,
  88. "Render token doesn't look as expected.\n");
  89. return -1; // make sure that this is a render packet
  90. }
  91. if (render->filled_mv_blocks_num) {
  92. av_log(avctx, AV_LOG_ERROR,
  93. "Rendering surface contains %i unprocessed blocks.\n",
  94. render->filled_mv_blocks_num);
  95. return -1;
  96. }
  97. if (render->allocated_mv_blocks < 1 ||
  98. render->allocated_data_blocks < render->allocated_mv_blocks*mb_block_count ||
  99. render->start_mv_blocks_num >= render->allocated_mv_blocks ||
  100. render->next_free_data_block_num >
  101. render->allocated_data_blocks -
  102. mb_block_count*(render->allocated_mv_blocks-render->start_mv_blocks_num)) {
  103. av_log(avctx, AV_LOG_ERROR,
  104. "Rendering surface doesn't provide enough block structures to work with.\n");
  105. return -1;
  106. }
  107. render->picture_structure = s->picture_structure;
  108. render->flags = s->first_field ? 0 : XVMC_SECOND_FIELD;
  109. render->p_future_surface = NULL;
  110. render->p_past_surface = NULL;
  111. switch(s->pict_type) {
  112. case AV_PICTURE_TYPE_I:
  113. return 0; // no prediction from other frames
  114. case AV_PICTURE_TYPE_B:
  115. next = (struct xvmc_pix_fmt*)s->next_picture.f->data[2];
  116. if (!next)
  117. return -1;
  118. if (next->xvmc_id != AV_XVMC_ID)
  119. return -1;
  120. render->p_future_surface = next->p_surface;
  121. // no return here, going to set forward prediction
  122. case AV_PICTURE_TYPE_P:
  123. last = (struct xvmc_pix_fmt*)s->last_picture.f->data[2];
  124. if (!last)
  125. last = render; // predict second field from the first
  126. if (last->xvmc_id != AV_XVMC_ID)
  127. return -1;
  128. render->p_past_surface = last->p_surface;
  129. return 0;
  130. }
  131. return -1;
  132. }
  133. /**
  134. * Complete frame/field rendering by passing any remaining blocks.
  135. * Normally ff_draw_horiz_band() is called for each slice, however,
  136. * some leftover blocks, for example from error_resilience(), may remain.
  137. * It should be safe to call the function a few times for the same field.
  138. */
  139. static int ff_xvmc_field_end(AVCodecContext *avctx)
  140. {
  141. struct MpegEncContext *s = avctx->priv_data;
  142. struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
  143. assert(render);
  144. if (render->filled_mv_blocks_num > 0)
  145. ff_mpeg_draw_horiz_band(s, 0, 0);
  146. return 0;
  147. }
  148. /**
  149. * Synthesize the data needed by XvMC to render one macroblock of data.
  150. * Fill all relevant fields, if necessary do IDCT.
  151. */
  152. static void ff_xvmc_decode_mb(struct MpegEncContext *s)
  153. {
  154. XvMCMacroBlock *mv_block;
  155. struct xvmc_pix_fmt *render;
  156. int i, cbp, blocks_per_mb;
  157. const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
  158. if (s->encoding) {
  159. av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
  160. return;
  161. }
  162. // from ff_mpv_reconstruct_mb(), update DC predictors for P macroblocks
  163. if (!s->mb_intra) {
  164. s->last_dc[0] =
  165. s->last_dc[1] =
  166. s->last_dc[2] = 128 << s->intra_dc_precision;
  167. }
  168. // MC doesn't skip blocks
  169. s->mb_skipped = 0;
  170. // Do I need to export quant when I could not perform postprocessing?
  171. // Anyway, it doesn't hurt.
  172. s->current_picture.qscale_table[mb_xy] = s->qscale;
  173. // start of XVMC-specific code
  174. render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
  175. assert(render);
  176. assert(render->xvmc_id == AV_XVMC_ID);
  177. assert(render->mv_blocks);
  178. // take the next free macroblock
  179. mv_block = &render->mv_blocks[render->start_mv_blocks_num +
  180. render->filled_mv_blocks_num];
  181. mv_block->x = s->mb_x;
  182. mv_block->y = s->mb_y;
  183. mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
  184. if (s->mb_intra) {
  185. mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
  186. } else {
  187. mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
  188. if (s->mv_dir & MV_DIR_FORWARD) {
  189. mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
  190. // PMV[n][dir][xy] = mv[dir][n][xy]
  191. mv_block->PMV[0][0][0] = s->mv[0][0][0];
  192. mv_block->PMV[0][0][1] = s->mv[0][0][1];
  193. mv_block->PMV[1][0][0] = s->mv[0][1][0];
  194. mv_block->PMV[1][0][1] = s->mv[0][1][1];
  195. }
  196. if (s->mv_dir & MV_DIR_BACKWARD) {
  197. mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
  198. mv_block->PMV[0][1][0] = s->mv[1][0][0];
  199. mv_block->PMV[0][1][1] = s->mv[1][0][1];
  200. mv_block->PMV[1][1][0] = s->mv[1][1][0];
  201. mv_block->PMV[1][1][1] = s->mv[1][1][1];
  202. }
  203. switch(s->mv_type) {
  204. case MV_TYPE_16X16:
  205. mv_block->motion_type = XVMC_PREDICTION_FRAME;
  206. break;
  207. case MV_TYPE_16X8:
  208. mv_block->motion_type = XVMC_PREDICTION_16x8;
  209. break;
  210. case MV_TYPE_FIELD:
  211. mv_block->motion_type = XVMC_PREDICTION_FIELD;
  212. if (s->picture_structure == PICT_FRAME) {
  213. mv_block->PMV[0][0][1] <<= 1;
  214. mv_block->PMV[1][0][1] <<= 1;
  215. mv_block->PMV[0][1][1] <<= 1;
  216. mv_block->PMV[1][1][1] <<= 1;
  217. }
  218. break;
  219. case MV_TYPE_DMV:
  220. mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
  221. if (s->picture_structure == PICT_FRAME) {
  222. mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top
  223. mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1;
  224. mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom
  225. mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1;
  226. mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom
  227. mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01
  228. mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top
  229. mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11
  230. } else {
  231. mv_block->PMV[0][1][0] = s->mv[0][2][0]; // dmv00
  232. mv_block->PMV[0][1][1] = s->mv[0][2][1]; // dmv01
  233. }
  234. break;
  235. default:
  236. assert(0);
  237. }
  238. mv_block->motion_vertical_field_select = 0;
  239. // set correct field references
  240. if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
  241. mv_block->motion_vertical_field_select |= s->field_select[0][0];
  242. mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
  243. mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
  244. mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3;
  245. }
  246. } // !intra
  247. // time to handle data blocks
  248. mv_block->index = render->next_free_data_block_num;
  249. blocks_per_mb = 6;
  250. if (s->chroma_format >= 2) {
  251. blocks_per_mb = 4 + (1 << s->chroma_format);
  252. }
  253. // calculate cbp
  254. cbp = 0;
  255. for (i = 0; i < blocks_per_mb; i++) {
  256. cbp += cbp;
  257. if (s->block_last_index[i] >= 0)
  258. cbp++;
  259. }
  260. if (s->avctx->flags & AV_CODEC_FLAG_GRAY) {
  261. if (s->mb_intra) { // intra frames are always full chroma blocks
  262. for (i = 4; i < blocks_per_mb; i++) {
  263. memset(s->pblocks[i], 0, sizeof(*s->pblocks[i])); // so we need to clear them
  264. if (!render->unsigned_intra)
  265. *s->pblocks[i][0] = 1 << 10;
  266. }
  267. } else {
  268. cbp &= 0xf << (blocks_per_mb - 4);
  269. blocks_per_mb = 4; // luminance blocks only
  270. }
  271. }
  272. mv_block->coded_block_pattern = cbp;
  273. if (cbp == 0)
  274. mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
  275. for (i = 0; i < blocks_per_mb; i++) {
  276. if (s->block_last_index[i] >= 0) {
  277. // I do not have unsigned_intra MOCO to test, hope it is OK.
  278. if (s->mb_intra && (render->idct || !render->unsigned_intra))
  279. *s->pblocks[i][0] -= 1 << 10;
  280. if (!render->idct) {
  281. s->idsp.idct(*s->pblocks[i]);
  282. /* It is unclear if MC hardware requires pixel diff values to be
  283. * in the range [-255;255]. TODO: Clipping if such hardware is
  284. * ever found. As of now it would only be an unnecessary
  285. * slowdown. */
  286. }
  287. // copy blocks only if the codec doesn't support pblocks reordering
  288. if (!s->pack_pblocks) {
  289. memcpy(&render->data_blocks[render->next_free_data_block_num*64],
  290. s->pblocks[i], sizeof(*s->pblocks[i]));
  291. }
  292. render->next_free_data_block_num++;
  293. }
  294. }
  295. render->filled_mv_blocks_num++;
  296. assert(render->filled_mv_blocks_num <= render->allocated_mv_blocks);
  297. assert(render->next_free_data_block_num <= render->allocated_data_blocks);
  298. /* The above conditions should not be able to fail as long as this function
  299. * is used and the following 'if ()' automatically calls a callback to free
  300. * blocks. */
  301. if (render->filled_mv_blocks_num == render->allocated_mv_blocks)
  302. ff_mpeg_draw_horiz_band(s, 0, 0);
  303. }
  304. #if CONFIG_MPEG1_XVMC_HWACCEL
  305. const AVHWAccel ff_mpeg1_xvmc_hwaccel = {
  306. .name = "mpeg1_xvmc",
  307. .type = AVMEDIA_TYPE_VIDEO,
  308. .id = AV_CODEC_ID_MPEG1VIDEO,
  309. .pix_fmt = AV_PIX_FMT_XVMC,
  310. .start_frame = ff_xvmc_field_start,
  311. .end_frame = ff_xvmc_field_end,
  312. .decode_slice = NULL,
  313. .decode_mb = ff_xvmc_decode_mb,
  314. .priv_data_size = 0,
  315. };
  316. #endif
  317. #if CONFIG_MPEG2_XVMC_HWACCEL
  318. const AVHWAccel ff_mpeg2_xvmc_hwaccel = {
  319. .name = "mpeg2_xvmc",
  320. .type = AVMEDIA_TYPE_VIDEO,
  321. .id = AV_CODEC_ID_MPEG2VIDEO,
  322. .pix_fmt = AV_PIX_FMT_XVMC,
  323. .start_frame = ff_xvmc_field_start,
  324. .end_frame = ff_xvmc_field_end,
  325. .decode_slice = NULL,
  326. .decode_mb = ff_xvmc_decode_mb,
  327. .priv_data_size = 0,
  328. };
  329. #endif