libavcodec/mpegvideo_xvmc.c

   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
  22 #include <limits.h>
  23
  24 //avcodec include
  25 #include "avcodec.h"
  26 #include "dsputil.h"
  27 #include "mpegvideo.h"
  28
  29 #undef NDEBUG
  30 #include <assert.h>
  31
  32 #include "xvmc.h"
  33
  34 //set s->block
  35 void ff_xvmc_init_block(MpegEncContext *s)
  36 {
  37     struct xvmc_render_state * render;
  38     render = (struct xvmc_render_state*)s->current_picture.data[2];
  39     assert(render);
  40     if (!render || render->magic != AV_XVMC_RENDER_MAGIC) {
  41         assert(0);
  42         return; // make sure that this is a render packet
  43     }
  44     s->block = (DCTELEM *)(render->data_blocks + render->next_free_data_block_num * 64);
  45 }
  46
  47 void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
  48 {
  49     int i,j;
  50     const int mb_block_count = 4 + (1 << s->chroma_format);
  51
  52     j = 0;
  53     cbp <<= 12-mb_block_count;
  54     for (i = 0; i < mb_block_count; i++) {
  55         if (cbp & (1 << 11))
  56             s->pblocks[i] = (short *)(&s->block[j++]);
  57         else
  58             s->pblocks[i] = NULL;
  59         cbp+=cbp;
  60     }
  61 }
  62
  63 // These functions should be called on every new field and/or frame.
  64 // They should be safe if they are called a few times for the same field!
  65 int ff_xvmc_field_start(MpegEncContext*s, AVCodecContext *avctx)
  66 {
  67     struct xvmc_render_state * render, * last, * next;
  68
  69     assert(avctx);
  70
  71     render = (struct xvmc_render_state*)s->current_picture.data[2];
  72     assert(render);
  73     if (!render || render->magic != AV_XVMC_RENDER_MAGIC)
  74         return -1; // make sure that this is render packet
  75
  76     render->picture_structure = s->picture_structure;
  77     render->flags             = s->first_field ? 0 : XVMC_SECOND_FIELD;
  78
  79     // make sure that all data is drawn by XVMC_end_frame
  80     assert(render->filled_mv_blocks_num == 0);
  81
  82     render->p_future_surface = NULL;
  83     render->p_past_surface   = NULL;
  84
  85     switch(s->pict_type){
  86         case  FF_I_TYPE:
  87             return 0; // no prediction from other frames
  88         case  FF_B_TYPE:
  89             next = (struct xvmc_render_state*)s->next_picture.data[2];
  90             assert(next);
  91             if (!next)
  92                 return -1;
  93             if (next->magic != AV_XVMC_RENDER_MAGIC)
  94                 return -1;
  95             render->p_future_surface = next->p_surface;
  96             // no return here, going to set forward prediction
  97         case  FF_P_TYPE:
  98             last = (struct xvmc_render_state*)s->last_picture.data[2];
  99             if (!last) // && !s->first_field)
 100                 last = render; // predict second field from the first
 101             if (last->magic != AV_XVMC_RENDER_MAGIC)
 102                 return -1;
 103             render->p_past_surface = last->p_surface;
 104             return 0;
 105     }
 106
 107 return -1;
 108 }
 109
 110 void ff_xvmc_field_end(MpegEncContext *s)
 111 {
 112     struct xvmc_render_state * render;
 113     render = (struct xvmc_render_state*)s->current_picture.data[2];
 114     assert(render);
 115
 116     if (render->filled_mv_blocks_num > 0)
 117         ff_draw_horiz_band(s,0,0);
 118 }
 119
 120 void ff_xvmc_decode_mb(MpegEncContext *s)
 121 {
 122     XvMCMacroBlock * mv_block;
 123     struct xvmc_render_state * render;
 124     int i,cbp,blocks_per_mb;
 125
 126     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
 127
 128
 129     if (s->encoding) {
 130         av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
 131         return;
 132     }
 133
 134     // from MPV_decode_mb(), update DC predictors for P macroblocks
 135     if (!s->mb_intra) {
 136         s->last_dc[0] =
 137         s->last_dc[1] =
 138         s->last_dc[2] =  128 << s->intra_dc_precision;
 139     }
 140
 141     // MC doesn't skip blocks
 142     s->mb_skipped = 0;
 143
 144
 145     // Do I need to export quant when I could not perform postprocessing?
 146     // Anyway, it doesn't hurt.
 147     s->current_picture.qscale_table[mb_xy] = s->qscale;
 148
 149     // START OF XVMC specific code
 150     render = (struct xvmc_render_state*)s->current_picture.data[2];
 151     assert(render);
 152     assert(render->magic==AV_XVMC_RENDER_MAGIC);
 153     assert(render->mv_blocks);
 154
 155     // take the next free macroblock
 156     mv_block = &render->mv_blocks[render->start_mv_blocks_num +
 157                                   render->filled_mv_blocks_num ];
 158
 159     mv_block->x        = s->mb_x;
 160     mv_block->y        = s->mb_y;
 161     mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
 162     if (s->mb_intra) {
 163         mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
 164     } else {
 165         mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
 166
 167         if (s->mv_dir & MV_DIR_FORWARD) {
 168             mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
 169             //pmv[n][dir][xy]=mv[dir][n][xy]
 170             mv_block->PMV[0][0][0] = s->mv[0][0][0];
 171             mv_block->PMV[0][0][1] = s->mv[0][0][1];
 172             mv_block->PMV[1][0][0] = s->mv[0][1][0];
 173             mv_block->PMV[1][0][1] = s->mv[0][1][1];
 174         }
 175         if (s->mv_dir & MV_DIR_BACKWARD) {
 176             mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
 177             mv_block->PMV[0][1][0] = s->mv[1][0][0];
 178             mv_block->PMV[0][1][1] = s->mv[1][0][1];
 179             mv_block->PMV[1][1][0] = s->mv[1][1][0];
 180             mv_block->PMV[1][1][1] = s->mv[1][1][1];
 181         }
 182
 183         switch(s->mv_type){
 184             case  MV_TYPE_16X16:
 185                 mv_block->motion_type = XVMC_PREDICTION_FRAME;
 186                 break;
 187             case  MV_TYPE_16X8:
 188                 mv_block->motion_type = XVMC_PREDICTION_16x8;
 189                 break;
 190             case  MV_TYPE_FIELD:
 191                 mv_block->motion_type = XVMC_PREDICTION_FIELD;
 192                 if (s->picture_structure == PICT_FRAME) {
 193                     mv_block->PMV[0][0][1] <<= 1;
 194                     mv_block->PMV[1][0][1] <<= 1;
 195                     mv_block->PMV[0][1][1] <<= 1;
 196                     mv_block->PMV[1][1][1] <<= 1;
 197                 }
 198                 break;
 199             case  MV_TYPE_DMV:
 200                 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
 201                 if (s->picture_structure == PICT_FRAME) {
 202
 203                     mv_block->PMV[0][0][0] = s->mv[0][0][0]; // top from top
 204                     mv_block->PMV[0][0][1] = s->mv[0][0][1]<<1;
 205
 206                     mv_block->PMV[0][1][0] = s->mv[0][0][0]; // bottom from bottom
 207                     mv_block->PMV[0][1][1] = s->mv[0][0][1]<<1;
 208
 209                     mv_block->PMV[1][0][0] = s->mv[0][2][0]; // dmv00, top from bottom
 210                     mv_block->PMV[1][0][1] = s->mv[0][2][1]<<1; // dmv01
 211
 212                     mv_block->PMV[1][1][0] = s->mv[0][3][0]; // dmv10, bottom from top
 213                     mv_block->PMV[1][1][1] = s->mv[0][3][1]<<1; // dmv11
 214
 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
 224         mv_block->motion_vertical_field_select = 0;
 225
 226         // set correct field references
 227         if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
 228             mv_block->motion_vertical_field_select |= s->field_select[0][0];
 229             mv_block->motion_vertical_field_select |= s->field_select[1][0]<<1;
 230             mv_block->motion_vertical_field_select |= s->field_select[0][1]<<2;
 231             mv_block->motion_vertical_field_select |= s->field_select[1][1]<<3;
 232         }
 233     } // !intra
 234     // time to handle data blocks;
 235     mv_block->index = render->next_free_data_block_num;
 236
 237     blocks_per_mb = 6;
 238     if (s->chroma_format >= 2) {
 239         blocks_per_mb = 4 + (1 << s->chroma_format);
 240     }
 241
 242     // calculate cbp
 243     cbp = 0;
 244     for (i = 0; i < blocks_per_mb; i++) {
 245         cbp += cbp;
 246         if (s->block_last_index[i] >= 0)
 247             cbp++;
 248     }
 249
 250     if (s->flags & CODEC_FLAG_GRAY) {
 251         if (s->mb_intra) { // intra frames are always full chroma block
 252             for (i = 4; i < blocks_per_mb; i++) {
 253                 memset(s->pblocks[i],0,sizeof(short)*8*8); // so we need to clear them
 254                 if (!render->unsigned_intra)
 255                     s->pblocks[i][0] = 1 << 10;
 256             }
 257         } else {
 258             cbp &= 0xf << (blocks_per_mb - 4);
 259             blocks_per_mb = 4; // luminance blocks only
 260         }
 261     }
 262     mv_block->coded_block_pattern = cbp;
 263     if (cbp == 0)
 264         mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
 265
 266     for (i = 0; i < blocks_per_mb; i++) {
 267         if (s->block_last_index[i] >= 0) {
 268             // I do not have unsigned_intra MOCO to test, hope it is OK.
 269             if (s->mb_intra && (render->idct || (!render->idct && !render->unsigned_intra)))
 270                 s->pblocks[i][0] -= 1 << 10;
 271             if (!render->idct) {
 272                 s->dsp.idct(s->pblocks[i]);
 273                 /* It is unclear if MC hardware requires pixel diff values to be in
 274                  * range [-255;255]. TODO cliping if such hardware is ever found.
 275                  * As of now it would only be unnecessery slowdown. */
 276             }
 277             // copy blocks only if the codec doesn't support pblocks reordering
 278             if (s->avctx->xvmc_acceleration == 1) {
 279                 memcpy(&render->data_blocks[render->next_free_data_block_num*64],
 280                        s->pblocks[i],sizeof(short)*8*8);
 281             }
 282             render->next_free_data_block_num++;
 283         }
 284     }
 285     render->filled_mv_blocks_num++;
 286
 287     assert(render->filled_mv_blocks_num     <= render->total_number_of_mv_blocks);
 288     assert(render->next_free_data_block_num <= render->total_number_of_data_blocks);
 289
 290
 291     if (render->filled_mv_blocks_num >= render->total_number_of_mv_blocks)
 292         ff_draw_horiz_band(s,0,0);
 293 }