yading@10: /*
yading@10:  * XVideo Motion Compensation
yading@10:  * Copyright (c) 2003 Ivan Kalvachev
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include <limits.h>
yading@10: #include <X11/extensions/XvMC.h>
yading@10: 
yading@10: #include "avcodec.h"
yading@10: #include "mpegvideo.h"
yading@10: 
yading@10: #undef NDEBUG
yading@10: #include <assert.h>
yading@10: 
yading@10: #include "xvmc.h"
yading@10: #include "xvmc_internal.h"
yading@10: 
yading@10: /**
yading@10:  * Initialize the block field of the MpegEncContext pointer passed as
yading@10:  * parameter after making sure that the data is not corrupted.
yading@10:  * In order to implement something like direct rendering instead of decoding
yading@10:  * coefficients in s->blocks and then copying them, copy them directly
yading@10:  * into the data_blocks array provided by xvmc.
yading@10:  */
yading@10: void ff_xvmc_init_block(MpegEncContext *s)
yading@10: {
yading@10:     struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
yading@10:     assert(render && render->xvmc_id == AV_XVMC_ID);
yading@10: 
yading@10:     s->block = (int16_t (*)[64])(render->data_blocks + render->next_free_data_block_num * 64);
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Fill individual block pointers, so there are no gaps in the data_block array
yading@10:  * in case not all blocks in the macroblock are coded.
yading@10:  */
yading@10: void ff_xvmc_pack_pblocks(MpegEncContext *s, int cbp)
yading@10: {
yading@10:     int i, j = 0;
yading@10:     const int mb_block_count = 4 + (1 << s->chroma_format);
yading@10: 
yading@10:     cbp <<= 12-mb_block_count;
yading@10:     for (i = 0; i < mb_block_count; i++) {
yading@10:         if (cbp & (1 << 11))
yading@10:             s->pblocks[i] = &s->block[j++];
yading@10:         else
yading@10:             s->pblocks[i] = NULL;
yading@10:         cbp += cbp;
yading@10:     }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Find and store the surfaces that are used as reference frames.
yading@10:  * This function should be called for every new field and/or frame.
yading@10:  * It should be safe to call the function a few times for the same field.
yading@10:  */
yading@10: int ff_xvmc_field_start(MpegEncContext *s, AVCodecContext *avctx)
yading@10: {
yading@10:     struct xvmc_pix_fmt *last, *next, *render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
yading@10:     const int mb_block_count = 4 + (1 << s->chroma_format);
yading@10: 
yading@10:     assert(avctx);
yading@10:     if (!render || render->xvmc_id != AV_XVMC_ID ||
yading@10:         !render->data_blocks || !render->mv_blocks ||
yading@10:         (unsigned int)render->allocated_mv_blocks   > INT_MAX/(64*6) ||
yading@10:         (unsigned int)render->allocated_data_blocks > INT_MAX/64     ||
yading@10:         !render->p_surface) {
yading@10:         av_log(avctx, AV_LOG_ERROR,
yading@10:                "Render token doesn't look as expected.\n");
yading@10:         return -1; // make sure that this is a render packet
yading@10:     }
yading@10: 
yading@10:     if (render->filled_mv_blocks_num) {
yading@10:         av_log(avctx, AV_LOG_ERROR,
yading@10:                "Rendering surface contains %i unprocessed blocks.\n",
yading@10:                render->filled_mv_blocks_num);
yading@10:         return -1;
yading@10:     }
yading@10:     if (render->allocated_mv_blocks   < 1 ||
yading@10:         render->allocated_data_blocks <  render->allocated_mv_blocks*mb_block_count ||
yading@10:         render->start_mv_blocks_num   >= render->allocated_mv_blocks                ||
yading@10:         render->next_free_data_block_num >
yading@10:                         render->allocated_data_blocks -
yading@10:                         mb_block_count*(render->allocated_mv_blocks-render->start_mv_blocks_num)) {
yading@10:         av_log(avctx, AV_LOG_ERROR,
yading@10:                "Rendering surface doesn't provide enough block structures to work with.\n");
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     render->picture_structure = s->picture_structure;
yading@10:     render->flags             = s->first_field ? 0 : XVMC_SECOND_FIELD;
yading@10:     render->p_future_surface  = NULL;
yading@10:     render->p_past_surface    = NULL;
yading@10: 
yading@10:     switch(s->pict_type) {
yading@10:         case  AV_PICTURE_TYPE_I:
yading@10:             return 0; // no prediction from other frames
yading@10:         case  AV_PICTURE_TYPE_B:
yading@10:             next = (struct xvmc_pix_fmt*)s->next_picture.f.data[2];
yading@10:             if (!next)
yading@10:                 return -1;
yading@10:             if (next->xvmc_id != AV_XVMC_ID)
yading@10:                 return -1;
yading@10:             render->p_future_surface = next->p_surface;
yading@10:             // no return here, going to set forward prediction
yading@10:         case  AV_PICTURE_TYPE_P:
yading@10:             last = (struct xvmc_pix_fmt*)s->last_picture.f.data[2];
yading@10:             if (!last)
yading@10:                 last = render; // predict second field from the first
yading@10:             if (last->xvmc_id != AV_XVMC_ID)
yading@10:                 return -1;
yading@10:             render->p_past_surface = last->p_surface;
yading@10:             return 0;
yading@10:     }
yading@10: 
yading@10: return -1;
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Complete frame/field rendering by passing any remaining blocks.
yading@10:  * Normally ff_draw_horiz_band() is called for each slice, however,
yading@10:  * some leftover blocks, for example from error_resilience(), may remain.
yading@10:  * It should be safe to call the function a few times for the same field.
yading@10:  */
yading@10: void ff_xvmc_field_end(MpegEncContext *s)
yading@10: {
yading@10:     struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
yading@10:     assert(render);
yading@10: 
yading@10:     if (render->filled_mv_blocks_num > 0)
yading@10:         ff_mpeg_draw_horiz_band(s, 0, 0);
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Synthesize the data needed by XvMC to render one macroblock of data.
yading@10:  * Fill all relevant fields, if necessary do IDCT.
yading@10:  */
yading@10: void ff_xvmc_decode_mb(MpegEncContext *s)
yading@10: {
yading@10:     XvMCMacroBlock *mv_block;
yading@10:     struct xvmc_pix_fmt *render;
yading@10:     int i, cbp, blocks_per_mb;
yading@10: 
yading@10:     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
yading@10: 
yading@10: 
yading@10:     if (s->encoding) {
yading@10:         av_log(s->avctx, AV_LOG_ERROR, "XVMC doesn't support encoding!!!\n");
yading@10:         return;
yading@10:     }
yading@10: 
yading@10:     // from MPV_decode_mb(), update DC predictors for P macroblocks
yading@10:     if (!s->mb_intra) {
yading@10:         s->last_dc[0] =
yading@10:         s->last_dc[1] =
yading@10:         s->last_dc[2] =  128 << s->intra_dc_precision;
yading@10:     }
yading@10: 
yading@10:     // MC doesn't skip blocks
yading@10:     s->mb_skipped = 0;
yading@10: 
yading@10: 
yading@10:     // Do I need to export quant when I could not perform postprocessing?
yading@10:     // Anyway, it doesn't hurt.
yading@10:     s->current_picture.qscale_table[mb_xy] = s->qscale;
yading@10: 
yading@10:     // start of XVMC-specific code
yading@10:     render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
yading@10:     assert(render);
yading@10:     assert(render->xvmc_id == AV_XVMC_ID);
yading@10:     assert(render->mv_blocks);
yading@10: 
yading@10:     // take the next free macroblock
yading@10:     mv_block = &render->mv_blocks[render->start_mv_blocks_num +
yading@10:                                   render->filled_mv_blocks_num];
yading@10: 
yading@10:     mv_block->x        = s->mb_x;
yading@10:     mv_block->y        = s->mb_y;
yading@10:     mv_block->dct_type = s->interlaced_dct; // XVMC_DCT_TYPE_FRAME/FIELD;
yading@10:     if (s->mb_intra) {
yading@10:         mv_block->macroblock_type = XVMC_MB_TYPE_INTRA; // no MC, all done
yading@10:     } else {
yading@10:         mv_block->macroblock_type = XVMC_MB_TYPE_PATTERN;
yading@10: 
yading@10:         if (s->mv_dir & MV_DIR_FORWARD) {
yading@10:             mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_FORWARD;
yading@10:             // PMV[n][dir][xy] = mv[dir][n][xy]
yading@10:             mv_block->PMV[0][0][0] = s->mv[0][0][0];
yading@10:             mv_block->PMV[0][0][1] = s->mv[0][0][1];
yading@10:             mv_block->PMV[1][0][0] = s->mv[0][1][0];
yading@10:             mv_block->PMV[1][0][1] = s->mv[0][1][1];
yading@10:         }
yading@10:         if (s->mv_dir & MV_DIR_BACKWARD) {
yading@10:             mv_block->macroblock_type |= XVMC_MB_TYPE_MOTION_BACKWARD;
yading@10:             mv_block->PMV[0][1][0] = s->mv[1][0][0];
yading@10:             mv_block->PMV[0][1][1] = s->mv[1][0][1];
yading@10:             mv_block->PMV[1][1][0] = s->mv[1][1][0];
yading@10:             mv_block->PMV[1][1][1] = s->mv[1][1][1];
yading@10:         }
yading@10: 
yading@10:         switch(s->mv_type) {
yading@10:             case  MV_TYPE_16X16:
yading@10:                 mv_block->motion_type = XVMC_PREDICTION_FRAME;
yading@10:                 break;
yading@10:             case  MV_TYPE_16X8:
yading@10:                 mv_block->motion_type = XVMC_PREDICTION_16x8;
yading@10:                 break;
yading@10:             case  MV_TYPE_FIELD:
yading@10:                 mv_block->motion_type = XVMC_PREDICTION_FIELD;
yading@10:                 if (s->picture_structure == PICT_FRAME) {
yading@10:                     mv_block->PMV[0][0][1] <<= 1;
yading@10:                     mv_block->PMV[1][0][1] <<= 1;
yading@10:                     mv_block->PMV[0][1][1] <<= 1;
yading@10:                     mv_block->PMV[1][1][1] <<= 1;
yading@10:                 }
yading@10:                 break;
yading@10:             case  MV_TYPE_DMV:
yading@10:                 mv_block->motion_type = XVMC_PREDICTION_DUAL_PRIME;
yading@10:                 if (s->picture_structure == PICT_FRAME) {
yading@10: 
yading@10:                     mv_block->PMV[0][0][0] = s->mv[0][0][0];      // top from top
yading@10:                     mv_block->PMV[0][0][1] = s->mv[0][0][1] << 1;
yading@10: 
yading@10:                     mv_block->PMV[0][1][0] = s->mv[0][0][0];      // bottom from bottom
yading@10:                     mv_block->PMV[0][1][1] = s->mv[0][0][1] << 1;
yading@10: 
yading@10:                     mv_block->PMV[1][0][0] = s->mv[0][2][0];      // dmv00, top from bottom
yading@10:                     mv_block->PMV[1][0][1] = s->mv[0][2][1] << 1; // dmv01
yading@10: 
yading@10:                     mv_block->PMV[1][1][0] = s->mv[0][3][0];      // dmv10, bottom from top
yading@10:                     mv_block->PMV[1][1][1] = s->mv[0][3][1] << 1; // dmv11
yading@10: 
yading@10:                 } else {
yading@10:                     mv_block->PMV[0][1][0] = s->mv[0][2][0];      // dmv00
yading@10:                     mv_block->PMV[0][1][1] = s->mv[0][2][1];      // dmv01
yading@10:                 }
yading@10:                 break;
yading@10:             default:
yading@10:                 assert(0);
yading@10:         }
yading@10: 
yading@10:         mv_block->motion_vertical_field_select = 0;
yading@10: 
yading@10:         // set correct field references
yading@10:         if (s->mv_type == MV_TYPE_FIELD || s->mv_type == MV_TYPE_16X8) {
yading@10:             mv_block->motion_vertical_field_select |= s->field_select[0][0];
yading@10:             mv_block->motion_vertical_field_select |= s->field_select[1][0] << 1;
yading@10:             mv_block->motion_vertical_field_select |= s->field_select[0][1] << 2;
yading@10:             mv_block->motion_vertical_field_select |= s->field_select[1][1] << 3;
yading@10:         }
yading@10:     } // !intra
yading@10:     // time to handle data blocks
yading@10:     mv_block->index = render->next_free_data_block_num;
yading@10: 
yading@10:     blocks_per_mb = 6;
yading@10:     if (s->chroma_format >= 2) {
yading@10:         blocks_per_mb = 4 + (1 << s->chroma_format);
yading@10:     }
yading@10: 
yading@10:     // calculate cbp
yading@10:     cbp = 0;
yading@10:     for (i = 0; i < blocks_per_mb; i++) {
yading@10:         cbp += cbp;
yading@10:         if (s->block_last_index[i] >= 0)
yading@10:             cbp++;
yading@10:     }
yading@10: 
yading@10:     if (s->flags & CODEC_FLAG_GRAY) {
yading@10:         if (s->mb_intra) {                                   // intra frames are always full chroma blocks
yading@10:             for (i = 4; i < blocks_per_mb; i++) {
yading@10:                 memset(s->pblocks[i], 0, sizeof(*s->pblocks[i]));  // so we need to clear them
yading@10:                 if (!render->unsigned_intra)
yading@10:                     *s->pblocks[i][0] = 1 << 10;
yading@10:             }
yading@10:         } else {
yading@10:             cbp &= 0xf << (blocks_per_mb - 4);
yading@10:             blocks_per_mb = 4;                               // luminance blocks only
yading@10:         }
yading@10:     }
yading@10:     mv_block->coded_block_pattern = cbp;
yading@10:     if (cbp == 0)
yading@10:         mv_block->macroblock_type &= ~XVMC_MB_TYPE_PATTERN;
yading@10: 
yading@10:     for (i = 0; i < blocks_per_mb; i++) {
yading@10:         if (s->block_last_index[i] >= 0) {
yading@10:             // I do not have unsigned_intra MOCO to test, hope it is OK.
yading@10:             if (s->mb_intra && (render->idct || !render->unsigned_intra))
yading@10:                 *s->pblocks[i][0] -= 1 << 10;
yading@10:             if (!render->idct) {
yading@10:                 s->dsp.idct(*s->pblocks[i]);
yading@10:                 /* It is unclear if MC hardware requires pixel diff values to be
yading@10:                  * in the range [-255;255]. TODO: Clipping if such hardware is
yading@10:                  * ever found. As of now it would only be an unnecessary
yading@10:                  * slowdown. */
yading@10:             }
yading@10:             // copy blocks only if the codec doesn't support pblocks reordering
yading@10:             if (s->avctx->xvmc_acceleration == 1) {
yading@10:                 memcpy(&render->data_blocks[render->next_free_data_block_num*64],
yading@10:                        s->pblocks[i], sizeof(*s->pblocks[i]));
yading@10:             }
yading@10:             render->next_free_data_block_num++;
yading@10:         }
yading@10:     }
yading@10:     render->filled_mv_blocks_num++;
yading@10: 
yading@10:     assert(render->filled_mv_blocks_num     <= render->allocated_mv_blocks);
yading@10:     assert(render->next_free_data_block_num <= render->allocated_data_blocks);
yading@10:     /* The above conditions should not be able to fail as long as this function
yading@10:      * is used and the following 'if ()' automatically calls a callback to free
yading@10:      * blocks. */
yading@10: 
yading@10: 
yading@10:     if (render->filled_mv_blocks_num == render->allocated_mv_blocks)
yading@10:         ff_mpeg_draw_horiz_band(s, 0, 0);
yading@10: }