yading@10: /*
yading@10:  * RoQ Video Encoder.
yading@10:  *
yading@10:  * Copyright (C) 2007 Vitor Sessak <vitor1001@gmail.com>
yading@10:  * Copyright (C) 2004-2007 Eric Lasota
yading@10:  *    Based on RoQ specs (C) 2001 Tim Ferguson
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: /**
yading@10:  * @file
yading@10:  * id RoQ encoder by Vitor. Based on the Switchblade3 library and the
yading@10:  * Switchblade3 FFmpeg glue by Eric Lasota.
yading@10:  */
yading@10: 
yading@10: /*
yading@10:  * COSTS:
yading@10:  * Level 1:
yading@10:  *  SKIP - 2 bits
yading@10:  *  MOTION - 2 + 8 bits
yading@10:  *  CODEBOOK - 2 + 8 bits
yading@10:  *  SUBDIVIDE - 2 + combined subcel cost
yading@10:  *
yading@10:  * Level 2:
yading@10:  *  SKIP - 2 bits
yading@10:  *  MOTION - 2 + 8 bits
yading@10:  *  CODEBOOK - 2 + 8 bits
yading@10:  *  SUBDIVIDE - 2 + 4*8 bits
yading@10:  *
yading@10:  * Maximum cost: 138 bits per cel
yading@10:  *
yading@10:  * Proper evaluation requires LCD fraction comparison, which requires
yading@10:  * Squared Error (SE) loss * savings increase
yading@10:  *
yading@10:  * Maximum savings increase: 136 bits
yading@10:  * Maximum SE loss without overflow: 31580641
yading@10:  * Components in 8x8 supercel: 192
yading@10:  * Maximum SE precision per component: 164482
yading@10:  *    >65025, so no truncation is needed (phew)
yading@10:  */
yading@10: 
yading@10: #include <string.h>
yading@10: 
yading@10: #include "roqvideo.h"
yading@10: #include "bytestream.h"
yading@10: #include "elbg.h"
yading@10: #include "internal.h"
yading@10: #include "mathops.h"
yading@10: 
yading@10: #define CHROMA_BIAS 1
yading@10: 
yading@10: /**
yading@10:  * Maximum number of generated 4x4 codebooks. Can't be 256 to workaround a
yading@10:  * Quake 3 bug.
yading@10:  */
yading@10: #define MAX_CBS_4x4 255
yading@10: 
yading@10: #define MAX_CBS_2x2 256 ///< Maximum number of 2x2 codebooks.
yading@10: 
yading@10: /* The cast is useful when multiplying it by INT_MAX */
yading@10: #define ROQ_LAMBDA_SCALE ((uint64_t) FF_LAMBDA_SCALE)
yading@10: 
yading@10: /* Macroblock support functions */
yading@10: static void unpack_roq_cell(roq_cell *cell, uint8_t u[4*3])
yading@10: {
yading@10:     memcpy(u  , cell->y, 4);
yading@10:     memset(u+4, cell->u, 4);
yading@10:     memset(u+8, cell->v, 4);
yading@10: }
yading@10: 
yading@10: static void unpack_roq_qcell(uint8_t cb2[], roq_qcell *qcell, uint8_t u[4*4*3])
yading@10: {
yading@10:     int i,cp;
yading@10:     static const int offsets[4] = {0, 2, 8, 10};
yading@10: 
yading@10:     for (cp=0; cp<3; cp++)
yading@10:         for (i=0; i<4; i++) {
yading@10:             u[4*4*cp + offsets[i]  ] = cb2[qcell->idx[i]*2*2*3 + 4*cp  ];
yading@10:             u[4*4*cp + offsets[i]+1] = cb2[qcell->idx[i]*2*2*3 + 4*cp+1];
yading@10:             u[4*4*cp + offsets[i]+4] = cb2[qcell->idx[i]*2*2*3 + 4*cp+2];
yading@10:             u[4*4*cp + offsets[i]+5] = cb2[qcell->idx[i]*2*2*3 + 4*cp+3];
yading@10:         }
yading@10: }
yading@10: 
yading@10: 
yading@10: static void enlarge_roq_mb4(uint8_t base[3*16], uint8_t u[3*64])
yading@10: {
yading@10:     int x,y,cp;
yading@10: 
yading@10:     for(cp=0; cp<3; cp++)
yading@10:         for(y=0; y<8; y++)
yading@10:             for(x=0; x<8; x++)
yading@10:                 *u++ = base[(y/2)*4 + (x/2) + 16*cp];
yading@10: }
yading@10: 
yading@10: static inline int square(int x)
yading@10: {
yading@10:     return x*x;
yading@10: }
yading@10: 
yading@10: static inline int eval_sse(const uint8_t *a, const uint8_t *b, int count)
yading@10: {
yading@10:     int diff=0;
yading@10: 
yading@10:     while(count--)
yading@10:         diff += square(*b++ - *a++);
yading@10: 
yading@10:     return diff;
yading@10: }
yading@10: 
yading@10: // FIXME Could use DSPContext.sse, but it is not so speed critical (used
yading@10: // just for motion estimation).
yading@10: static int block_sse(uint8_t * const *buf1, uint8_t * const *buf2, int x1, int y1,
yading@10:                      int x2, int y2, const int *stride1, const int *stride2, int size)
yading@10: {
yading@10:     int i, k;
yading@10:     int sse=0;
yading@10: 
yading@10:     for (k=0; k<3; k++) {
yading@10:         int bias = (k ? CHROMA_BIAS : 4);
yading@10:         for (i=0; i<size; i++)
yading@10:             sse += bias*eval_sse(buf1[k] + (y1+i)*stride1[k] + x1,
yading@10:                                  buf2[k] + (y2+i)*stride2[k] + x2, size);
yading@10:     }
yading@10: 
yading@10:     return sse;
yading@10: }
yading@10: 
yading@10: static int eval_motion_dist(RoqContext *enc, int x, int y, motion_vect vect,
yading@10:                              int size)
yading@10: {
yading@10:     int mx=vect.d[0];
yading@10:     int my=vect.d[1];
yading@10: 
yading@10:     if (mx < -7 || mx > 7)
yading@10:         return INT_MAX;
yading@10: 
yading@10:     if (my < -7 || my > 7)
yading@10:         return INT_MAX;
yading@10: 
yading@10:     mx += x;
yading@10:     my += y;
yading@10: 
yading@10:     if ((unsigned) mx > enc->width-size || (unsigned) my > enc->height-size)
yading@10:         return INT_MAX;
yading@10: 
yading@10:     return block_sse(enc->frame_to_enc->data, enc->last_frame->data, x, y,
yading@10:                      mx, my,
yading@10:                      enc->frame_to_enc->linesize, enc->last_frame->linesize,
yading@10:                      size);
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * @return distortion between two macroblocks
yading@10:  */
yading@10: static inline int squared_diff_macroblock(uint8_t a[], uint8_t b[], int size)
yading@10: {
yading@10:     int cp, sdiff=0;
yading@10: 
yading@10:     for(cp=0;cp<3;cp++) {
yading@10:         int bias = (cp ? CHROMA_BIAS : 4);
yading@10:         sdiff += bias*eval_sse(a, b, size*size);
yading@10:         a += size*size;
yading@10:         b += size*size;
yading@10:     }
yading@10: 
yading@10:     return sdiff;
yading@10: }
yading@10: 
yading@10: typedef struct
yading@10: {
yading@10:     int eval_dist[4];
yading@10:     int best_bit_use;
yading@10:     int best_coding;
yading@10: 
yading@10:     int subCels[4];
yading@10:     motion_vect motion;
yading@10:     int cbEntry;
yading@10: } SubcelEvaluation;
yading@10: 
yading@10: typedef struct
yading@10: {
yading@10:     int eval_dist[4];
yading@10:     int best_coding;
yading@10: 
yading@10:     SubcelEvaluation subCels[4];
yading@10: 
yading@10:     motion_vect motion;
yading@10:     int cbEntry;
yading@10: 
yading@10:     int sourceX, sourceY;
yading@10: } CelEvaluation;
yading@10: 
yading@10: typedef struct
yading@10: {
yading@10:     int numCB4;
yading@10:     int numCB2;
yading@10:     int usedCB2[MAX_CBS_2x2];
yading@10:     int usedCB4[MAX_CBS_4x4];
yading@10:     uint8_t unpacked_cb2[MAX_CBS_2x2*2*2*3];
yading@10:     uint8_t unpacked_cb4[MAX_CBS_4x4*4*4*3];
yading@10:     uint8_t unpacked_cb4_enlarged[MAX_CBS_4x4*8*8*3];
yading@10: } RoqCodebooks;
yading@10: 
yading@10: /**
yading@10:  * Temporary vars
yading@10:  */
yading@10: typedef struct RoqTempData
yading@10: {
yading@10:     CelEvaluation *cel_evals;
yading@10: 
yading@10:     int f2i4[MAX_CBS_4x4];
yading@10:     int i2f4[MAX_CBS_4x4];
yading@10:     int f2i2[MAX_CBS_2x2];
yading@10:     int i2f2[MAX_CBS_2x2];
yading@10: 
yading@10:     int mainChunkSize;
yading@10: 
yading@10:     int numCB4;
yading@10:     int numCB2;
yading@10: 
yading@10:     RoqCodebooks codebooks;
yading@10: 
yading@10:     int *closest_cb2;
yading@10:     int used_option[4];
yading@10: } RoqTempdata;
yading@10: 
yading@10: /**
yading@10:  * Initialize cel evaluators and set their source coordinates
yading@10:  */
yading@10: static void create_cel_evals(RoqContext *enc, RoqTempdata *tempData)
yading@10: {
yading@10:     int n=0, x, y, i;
yading@10: 
yading@10:     tempData->cel_evals = av_malloc(enc->width*enc->height/64 * sizeof(CelEvaluation));
yading@10: 
yading@10:     /* Map to the ROQ quadtree order */
yading@10:     for (y=0; y<enc->height; y+=16)
yading@10:         for (x=0; x<enc->width; x+=16)
yading@10:             for(i=0; i<4; i++) {
yading@10:                 tempData->cel_evals[n  ].sourceX = x + (i&1)*8;
yading@10:                 tempData->cel_evals[n++].sourceY = y + (i&2)*4;
yading@10:             }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Get macroblocks from parts of the image
yading@10:  */
yading@10: static void get_frame_mb(const AVFrame *frame, int x, int y, uint8_t mb[], int dim)
yading@10: {
yading@10:     int i, j, cp;
yading@10: 
yading@10:     for (cp=0; cp<3; cp++) {
yading@10:         int stride = frame->linesize[cp];
yading@10:         for (i=0; i<dim; i++)
yading@10:             for (j=0; j<dim; j++)
yading@10:                 *mb++ = frame->data[cp][(y+i)*stride + x + j];
yading@10:     }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Find the codebook with the lowest distortion from an image
yading@10:  */
yading@10: static int index_mb(uint8_t cluster[], uint8_t cb[], int numCB,
yading@10:                     int *outIndex, int dim)
yading@10: {
yading@10:     int i, lDiff = INT_MAX, pick=0;
yading@10: 
yading@10:     /* Diff against the others */
yading@10:     for (i=0; i<numCB; i++) {
yading@10:         int diff = squared_diff_macroblock(cluster, cb + i*dim*dim*3, dim);
yading@10:         if (diff < lDiff) {
yading@10:             lDiff = diff;
yading@10:             pick = i;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     *outIndex = pick;
yading@10:     return lDiff;
yading@10: }
yading@10: 
yading@10: #define EVAL_MOTION(MOTION) \
yading@10:     do { \
yading@10:         diff = eval_motion_dist(enc, j, i, MOTION, blocksize); \
yading@10:             \
yading@10:         if (diff < lowestdiff) { \
yading@10:             lowestdiff = diff; \
yading@10:             bestpick = MOTION; \
yading@10:         } \
yading@10:     } while(0)
yading@10: 
yading@10: static void motion_search(RoqContext *enc, int blocksize)
yading@10: {
yading@10:     static const motion_vect offsets[8] = {
yading@10:         {{ 0,-1}},
yading@10:         {{ 0, 1}},
yading@10:         {{-1, 0}},
yading@10:         {{ 1, 0}},
yading@10:         {{-1, 1}},
yading@10:         {{ 1,-1}},
yading@10:         {{-1,-1}},
yading@10:         {{ 1, 1}},
yading@10:     };
yading@10: 
yading@10:     int diff, lowestdiff, oldbest;
yading@10:     int off[3];
yading@10:     motion_vect bestpick = {{0,0}};
yading@10:     int i, j, k, offset;
yading@10: 
yading@10:     motion_vect *last_motion;
yading@10:     motion_vect *this_motion;
yading@10:     motion_vect vect, vect2;
yading@10: 
yading@10:     int max=(enc->width/blocksize)*enc->height/blocksize;
yading@10: 
yading@10:     if (blocksize == 4) {
yading@10:         last_motion = enc->last_motion4;
yading@10:         this_motion = enc->this_motion4;
yading@10:     } else {
yading@10:         last_motion = enc->last_motion8;
yading@10:         this_motion = enc->this_motion8;
yading@10:     }
yading@10: 
yading@10:     for (i=0; i<enc->height; i+=blocksize)
yading@10:         for (j=0; j<enc->width; j+=blocksize) {
yading@10:             lowestdiff = eval_motion_dist(enc, j, i, (motion_vect) {{0,0}},
yading@10:                                           blocksize);
yading@10:             bestpick.d[0] = 0;
yading@10:             bestpick.d[1] = 0;
yading@10: 
yading@10:             if (blocksize == 4)
yading@10:                 EVAL_MOTION(enc->this_motion8[(i/8)*(enc->width/8) + j/8]);
yading@10: 
yading@10:             offset = (i/blocksize)*enc->width/blocksize + j/blocksize;
yading@10:             if (offset < max && offset >= 0)
yading@10:                 EVAL_MOTION(last_motion[offset]);
yading@10: 
yading@10:             offset++;
yading@10:             if (offset < max && offset >= 0)
yading@10:                 EVAL_MOTION(last_motion[offset]);
yading@10: 
yading@10:             offset = (i/blocksize + 1)*enc->width/blocksize + j/blocksize;
yading@10:             if (offset < max && offset >= 0)
yading@10:                 EVAL_MOTION(last_motion[offset]);
yading@10: 
yading@10:             off[0]= (i/blocksize)*enc->width/blocksize + j/blocksize - 1;
yading@10:             off[1]= off[0] - enc->width/blocksize + 1;
yading@10:             off[2]= off[1] + 1;
yading@10: 
yading@10:             if (i) {
yading@10: 
yading@10:                 for(k=0; k<2; k++)
yading@10:                     vect.d[k]= mid_pred(this_motion[off[0]].d[k],
yading@10:                                         this_motion[off[1]].d[k],
yading@10:                                         this_motion[off[2]].d[k]);
yading@10: 
yading@10:                 EVAL_MOTION(vect);
yading@10:                 for(k=0; k<3; k++)
yading@10:                     EVAL_MOTION(this_motion[off[k]]);
yading@10:             } else if(j)
yading@10:                 EVAL_MOTION(this_motion[off[0]]);
yading@10: 
yading@10:             vect = bestpick;
yading@10: 
yading@10:             oldbest = -1;
yading@10:             while (oldbest != lowestdiff) {
yading@10:                 oldbest = lowestdiff;
yading@10:                 for (k=0; k<8; k++) {
yading@10:                     vect2 = vect;
yading@10:                     vect2.d[0] += offsets[k].d[0];
yading@10:                     vect2.d[1] += offsets[k].d[1];
yading@10:                     EVAL_MOTION(vect2);
yading@10:                 }
yading@10:                 vect = bestpick;
yading@10:             }
yading@10:             offset = (i/blocksize)*enc->width/blocksize + j/blocksize;
yading@10:             this_motion[offset] = bestpick;
yading@10:         }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Get distortion for all options available to a subcel
yading@10:  */
yading@10: static void gather_data_for_subcel(SubcelEvaluation *subcel, int x,
yading@10:                                    int y, RoqContext *enc, RoqTempdata *tempData)
yading@10: {
yading@10:     uint8_t mb4[4*4*3];
yading@10:     uint8_t mb2[2*2*3];
yading@10:     int cluster_index;
yading@10:     int i, best_dist;
yading@10: 
yading@10:     static const int bitsUsed[4] = {2, 10, 10, 34};
yading@10: 
yading@10:     if (enc->framesSinceKeyframe >= 1) {
yading@10:         subcel->motion = enc->this_motion4[y*enc->width/16 + x/4];
yading@10: 
yading@10:         subcel->eval_dist[RoQ_ID_FCC] =
yading@10:             eval_motion_dist(enc, x, y,
yading@10:                              enc->this_motion4[y*enc->width/16 + x/4], 4);
yading@10:     } else
yading@10:         subcel->eval_dist[RoQ_ID_FCC] = INT_MAX;
yading@10: 
yading@10:     if (enc->framesSinceKeyframe >= 2)
yading@10:         subcel->eval_dist[RoQ_ID_MOT] = block_sse(enc->frame_to_enc->data,
yading@10:                                                   enc->current_frame->data, x,
yading@10:                                                   y, x, y,
yading@10:                                                   enc->frame_to_enc->linesize,
yading@10:                                                   enc->current_frame->linesize,
yading@10:                                                   4);
yading@10:     else
yading@10:         subcel->eval_dist[RoQ_ID_MOT] = INT_MAX;
yading@10: 
yading@10:     cluster_index = y*enc->width/16 + x/4;
yading@10: 
yading@10:     get_frame_mb(enc->frame_to_enc, x, y, mb4, 4);
yading@10: 
yading@10:     subcel->eval_dist[RoQ_ID_SLD] = index_mb(mb4,
yading@10:                                              tempData->codebooks.unpacked_cb4,
yading@10:                                              tempData->codebooks.numCB4,
yading@10:                                              &subcel->cbEntry, 4);
yading@10: 
yading@10:     subcel->eval_dist[RoQ_ID_CCC] = 0;
yading@10: 
yading@10:     for(i=0;i<4;i++) {
yading@10:         subcel->subCels[i] = tempData->closest_cb2[cluster_index*4+i];
yading@10: 
yading@10:         get_frame_mb(enc->frame_to_enc, x+2*(i&1),
yading@10:                      y+(i&2), mb2, 2);
yading@10: 
yading@10:         subcel->eval_dist[RoQ_ID_CCC] +=
yading@10:             squared_diff_macroblock(tempData->codebooks.unpacked_cb2 + subcel->subCels[i]*2*2*3, mb2, 2);
yading@10:     }
yading@10: 
yading@10:     best_dist = INT_MAX;
yading@10:     for (i=0; i<4; i++)
yading@10:         if (ROQ_LAMBDA_SCALE*subcel->eval_dist[i] + enc->lambda*bitsUsed[i] <
yading@10:             best_dist) {
yading@10:             subcel->best_coding = i;
yading@10:             subcel->best_bit_use = bitsUsed[i];
yading@10:             best_dist = ROQ_LAMBDA_SCALE*subcel->eval_dist[i] +
yading@10:                 enc->lambda*bitsUsed[i];
yading@10:         }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Get distortion for all options available to a cel
yading@10:  */
yading@10: static void gather_data_for_cel(CelEvaluation *cel, RoqContext *enc,
yading@10:                                 RoqTempdata *tempData)
yading@10: {
yading@10:     uint8_t mb8[8*8*3];
yading@10:     int index = cel->sourceY*enc->width/64 + cel->sourceX/8;
yading@10:     int i, j, best_dist, divide_bit_use;
yading@10: 
yading@10:     int bitsUsed[4] = {2, 10, 10, 0};
yading@10: 
yading@10:     if (enc->framesSinceKeyframe >= 1) {
yading@10:         cel->motion = enc->this_motion8[index];
yading@10: 
yading@10:         cel->eval_dist[RoQ_ID_FCC] =
yading@10:             eval_motion_dist(enc, cel->sourceX, cel->sourceY,
yading@10:                              enc->this_motion8[index], 8);
yading@10:     } else
yading@10:         cel->eval_dist[RoQ_ID_FCC] = INT_MAX;
yading@10: 
yading@10:     if (enc->framesSinceKeyframe >= 2)
yading@10:         cel->eval_dist[RoQ_ID_MOT] = block_sse(enc->frame_to_enc->data,
yading@10:                                                enc->current_frame->data,
yading@10:                                                cel->sourceX, cel->sourceY,
yading@10:                                                cel->sourceX, cel->sourceY,
yading@10:                                                enc->frame_to_enc->linesize,
yading@10:                                                enc->current_frame->linesize,8);
yading@10:     else
yading@10:         cel->eval_dist[RoQ_ID_MOT] = INT_MAX;
yading@10: 
yading@10:     get_frame_mb(enc->frame_to_enc, cel->sourceX, cel->sourceY, mb8, 8);
yading@10: 
yading@10:     cel->eval_dist[RoQ_ID_SLD] =
yading@10:         index_mb(mb8, tempData->codebooks.unpacked_cb4_enlarged,
yading@10:                  tempData->codebooks.numCB4, &cel->cbEntry, 8);
yading@10: 
yading@10:     gather_data_for_subcel(cel->subCels + 0, cel->sourceX+0, cel->sourceY+0, enc, tempData);
yading@10:     gather_data_for_subcel(cel->subCels + 1, cel->sourceX+4, cel->sourceY+0, enc, tempData);
yading@10:     gather_data_for_subcel(cel->subCels + 2, cel->sourceX+0, cel->sourceY+4, enc, tempData);
yading@10:     gather_data_for_subcel(cel->subCels + 3, cel->sourceX+4, cel->sourceY+4, enc, tempData);
yading@10: 
yading@10:     cel->eval_dist[RoQ_ID_CCC] = 0;
yading@10:     divide_bit_use = 0;
yading@10:     for (i=0; i<4; i++) {
yading@10:         cel->eval_dist[RoQ_ID_CCC] +=
yading@10:             cel->subCels[i].eval_dist[cel->subCels[i].best_coding];
yading@10:         divide_bit_use += cel->subCels[i].best_bit_use;
yading@10:     }
yading@10: 
yading@10:     best_dist = INT_MAX;
yading@10:     bitsUsed[3] = 2 + divide_bit_use;
yading@10: 
yading@10:     for (i=0; i<4; i++)
yading@10:         if (ROQ_LAMBDA_SCALE*cel->eval_dist[i] + enc->lambda*bitsUsed[i] <
yading@10:             best_dist) {
yading@10:             cel->best_coding = i;
yading@10:             best_dist = ROQ_LAMBDA_SCALE*cel->eval_dist[i] +
yading@10:                 enc->lambda*bitsUsed[i];
yading@10:         }
yading@10: 
yading@10:     tempData->used_option[cel->best_coding]++;
yading@10:     tempData->mainChunkSize += bitsUsed[cel->best_coding];
yading@10: 
yading@10:     if (cel->best_coding == RoQ_ID_SLD)
yading@10:         tempData->codebooks.usedCB4[cel->cbEntry]++;
yading@10: 
yading@10:     if (cel->best_coding == RoQ_ID_CCC)
yading@10:         for (i=0; i<4; i++) {
yading@10:             if (cel->subCels[i].best_coding == RoQ_ID_SLD)
yading@10:                 tempData->codebooks.usedCB4[cel->subCels[i].cbEntry]++;
yading@10:             else if (cel->subCels[i].best_coding == RoQ_ID_CCC)
yading@10:                 for (j=0; j<4; j++)
yading@10:                     tempData->codebooks.usedCB2[cel->subCels[i].subCels[j]]++;
yading@10:         }
yading@10: }
yading@10: 
yading@10: static void remap_codebooks(RoqContext *enc, RoqTempdata *tempData)
yading@10: {
yading@10:     int i, j, idx=0;
yading@10: 
yading@10:     /* Make remaps for the final codebook usage */
yading@10:     for (i=0; i<MAX_CBS_4x4; i++) {
yading@10:         if (tempData->codebooks.usedCB4[i]) {
yading@10:             tempData->i2f4[i] = idx;
yading@10:             tempData->f2i4[idx] = i;
yading@10:             for (j=0; j<4; j++)
yading@10:                 tempData->codebooks.usedCB2[enc->cb4x4[i].idx[j]]++;
yading@10:             idx++;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     tempData->numCB4 = idx;
yading@10: 
yading@10:     idx = 0;
yading@10:     for (i=0; i<MAX_CBS_2x2; i++) {
yading@10:         if (tempData->codebooks.usedCB2[i]) {
yading@10:             tempData->i2f2[i] = idx;
yading@10:             tempData->f2i2[idx] = i;
yading@10:             idx++;
yading@10:         }
yading@10:     }
yading@10:     tempData->numCB2 = idx;
yading@10: 
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Write codebook chunk
yading@10:  */
yading@10: static void write_codebooks(RoqContext *enc, RoqTempdata *tempData)
yading@10: {
yading@10:     int i, j;
yading@10:     uint8_t **outp= &enc->out_buf;
yading@10: 
yading@10:     if (tempData->numCB2) {
yading@10:         bytestream_put_le16(outp, RoQ_QUAD_CODEBOOK);
yading@10:         bytestream_put_le32(outp, tempData->numCB2*6 + tempData->numCB4*4);
yading@10:         bytestream_put_byte(outp, tempData->numCB4);
yading@10:         bytestream_put_byte(outp, tempData->numCB2);
yading@10: 
yading@10:         for (i=0; i<tempData->numCB2; i++) {
yading@10:             bytestream_put_buffer(outp, enc->cb2x2[tempData->f2i2[i]].y, 4);
yading@10:             bytestream_put_byte(outp, enc->cb2x2[tempData->f2i2[i]].u);
yading@10:             bytestream_put_byte(outp, enc->cb2x2[tempData->f2i2[i]].v);
yading@10:         }
yading@10: 
yading@10:         for (i=0; i<tempData->numCB4; i++)
yading@10:             for (j=0; j<4; j++)
yading@10:                 bytestream_put_byte(outp, tempData->i2f2[enc->cb4x4[tempData->f2i4[i]].idx[j]]);
yading@10: 
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline uint8_t motion_arg(motion_vect mot)
yading@10: {
yading@10:     uint8_t ax = 8 - ((uint8_t) mot.d[0]);
yading@10:     uint8_t ay = 8 - ((uint8_t) mot.d[1]);
yading@10:     return ((ax&15)<<4) | (ay&15);
yading@10: }
yading@10: 
yading@10: typedef struct
yading@10: {
yading@10:     int typeSpool;
yading@10:     int typeSpoolLength;
yading@10:     uint8_t argumentSpool[64];
yading@10:     uint8_t *args;
yading@10:     uint8_t **pout;
yading@10: } CodingSpool;
yading@10: 
yading@10: /* NOTE: Typecodes must be spooled AFTER arguments!! */
yading@10: static void write_typecode(CodingSpool *s, uint8_t type)
yading@10: {
yading@10:     s->typeSpool |= (type & 3) << (14 - s->typeSpoolLength);
yading@10:     s->typeSpoolLength += 2;
yading@10:     if (s->typeSpoolLength == 16) {
yading@10:         bytestream_put_le16(s->pout, s->typeSpool);
yading@10:         bytestream_put_buffer(s->pout, s->argumentSpool,
yading@10:                               s->args - s->argumentSpool);
yading@10:         s->typeSpoolLength = 0;
yading@10:         s->typeSpool = 0;
yading@10:         s->args = s->argumentSpool;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void reconstruct_and_encode_image(RoqContext *enc, RoqTempdata *tempData, int w, int h, int numBlocks)
yading@10: {
yading@10:     int i, j, k;
yading@10:     int x, y;
yading@10:     int subX, subY;
yading@10:     int dist=0;
yading@10: 
yading@10:     roq_qcell *qcell;
yading@10:     CelEvaluation *eval;
yading@10: 
yading@10:     CodingSpool spool;
yading@10: 
yading@10:     spool.typeSpool=0;
yading@10:     spool.typeSpoolLength=0;
yading@10:     spool.args = spool.argumentSpool;
yading@10:     spool.pout = &enc->out_buf;
yading@10: 
yading@10:     if (tempData->used_option[RoQ_ID_CCC]%2)
yading@10:         tempData->mainChunkSize+=8; //FIXME
yading@10: 
yading@10:     /* Write the video chunk header */
yading@10:     bytestream_put_le16(&enc->out_buf, RoQ_QUAD_VQ);
yading@10:     bytestream_put_le32(&enc->out_buf, tempData->mainChunkSize/8);
yading@10:     bytestream_put_byte(&enc->out_buf, 0x0);
yading@10:     bytestream_put_byte(&enc->out_buf, 0x0);
yading@10: 
yading@10:     for (i=0; i<numBlocks; i++) {
yading@10:         eval = tempData->cel_evals + i;
yading@10: 
yading@10:         x = eval->sourceX;
yading@10:         y = eval->sourceY;
yading@10:         dist += eval->eval_dist[eval->best_coding];
yading@10: 
yading@10:         switch (eval->best_coding) {
yading@10:         case RoQ_ID_MOT:
yading@10:             write_typecode(&spool, RoQ_ID_MOT);
yading@10:             break;
yading@10: 
yading@10:         case RoQ_ID_FCC:
yading@10:             bytestream_put_byte(&spool.args, motion_arg(eval->motion));
yading@10: 
yading@10:             write_typecode(&spool, RoQ_ID_FCC);
yading@10:             ff_apply_motion_8x8(enc, x, y,
yading@10:                                 eval->motion.d[0], eval->motion.d[1]);
yading@10:             break;
yading@10: 
yading@10:         case RoQ_ID_SLD:
yading@10:             bytestream_put_byte(&spool.args, tempData->i2f4[eval->cbEntry]);
yading@10:             write_typecode(&spool, RoQ_ID_SLD);
yading@10: 
yading@10:             qcell = enc->cb4x4 + eval->cbEntry;
yading@10:             ff_apply_vector_4x4(enc, x  , y  , enc->cb2x2 + qcell->idx[0]);
yading@10:             ff_apply_vector_4x4(enc, x+4, y  , enc->cb2x2 + qcell->idx[1]);
yading@10:             ff_apply_vector_4x4(enc, x  , y+4, enc->cb2x2 + qcell->idx[2]);
yading@10:             ff_apply_vector_4x4(enc, x+4, y+4, enc->cb2x2 + qcell->idx[3]);
yading@10:             break;
yading@10: 
yading@10:         case RoQ_ID_CCC:
yading@10:             write_typecode(&spool, RoQ_ID_CCC);
yading@10: 
yading@10:             for (j=0; j<4; j++) {
yading@10:                 subX = x + 4*(j&1);
yading@10:                 subY = y + 2*(j&2);
yading@10: 
yading@10:                 switch(eval->subCels[j].best_coding) {
yading@10:                 case RoQ_ID_MOT:
yading@10:                     break;
yading@10: 
yading@10:                 case RoQ_ID_FCC:
yading@10:                     bytestream_put_byte(&spool.args,
yading@10:                                         motion_arg(eval->subCels[j].motion));
yading@10: 
yading@10:                     ff_apply_motion_4x4(enc, subX, subY,
yading@10:                                         eval->subCels[j].motion.d[0],
yading@10:                                         eval->subCels[j].motion.d[1]);
yading@10:                     break;
yading@10: 
yading@10:                 case RoQ_ID_SLD:
yading@10:                     bytestream_put_byte(&spool.args,
yading@10:                                         tempData->i2f4[eval->subCels[j].cbEntry]);
yading@10: 
yading@10:                     qcell = enc->cb4x4 + eval->subCels[j].cbEntry;
yading@10: 
yading@10:                     ff_apply_vector_2x2(enc, subX  , subY  ,
yading@10:                                         enc->cb2x2 + qcell->idx[0]);
yading@10:                     ff_apply_vector_2x2(enc, subX+2, subY  ,
yading@10:                                         enc->cb2x2 + qcell->idx[1]);
yading@10:                     ff_apply_vector_2x2(enc, subX  , subY+2,
yading@10:                                         enc->cb2x2 + qcell->idx[2]);
yading@10:                     ff_apply_vector_2x2(enc, subX+2, subY+2,
yading@10:                                         enc->cb2x2 + qcell->idx[3]);
yading@10:                     break;
yading@10: 
yading@10:                 case RoQ_ID_CCC:
yading@10:                     for (k=0; k<4; k++) {
yading@10:                         int cb_idx = eval->subCels[j].subCels[k];
yading@10:                         bytestream_put_byte(&spool.args,
yading@10:                                             tempData->i2f2[cb_idx]);
yading@10: 
yading@10:                         ff_apply_vector_2x2(enc, subX + 2*(k&1), subY + (k&2),
yading@10:                                             enc->cb2x2 + cb_idx);
yading@10:                     }
yading@10:                     break;
yading@10:                 }
yading@10:                 write_typecode(&spool, eval->subCels[j].best_coding);
yading@10:             }
yading@10:             break;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     /* Flush the remainder of the argument/type spool */
yading@10:     while (spool.typeSpoolLength)
yading@10:         write_typecode(&spool, 0x0);
yading@10: 
yading@10: #if 0
yading@10:     uint8_t *fdata[3] = {enc->frame_to_enc->data[0],
yading@10:                            enc->frame_to_enc->data[1],
yading@10:                            enc->frame_to_enc->data[2]};
yading@10:     uint8_t *cdata[3] = {enc->current_frame->data[0],
yading@10:                            enc->current_frame->data[1],
yading@10:                            enc->current_frame->data[2]};
yading@10:     av_log(enc->avctx, AV_LOG_ERROR, "Expected distortion: %i Actual: %i\n",
yading@10:            dist,
yading@10:            block_sse(fdata, cdata, 0, 0, 0, 0,
yading@10:                      enc->frame_to_enc->linesize,
yading@10:                      enc->current_frame->linesize,
yading@10:                      enc->width));  //WARNING: Square dimensions implied...
yading@10: #endif
yading@10: }
yading@10: 
yading@10: 
yading@10: /**
yading@10:  * Create a single YUV cell from a 2x2 section of the image
yading@10:  */
yading@10: static inline void frame_block_to_cell(uint8_t *block, uint8_t * const *data,
yading@10:                                        int top, int left, const int *stride)
yading@10: {
yading@10:     int i, j, u=0, v=0;
yading@10: 
yading@10:     for (i=0; i<2; i++)
yading@10:         for (j=0; j<2; j++) {
yading@10:             int x = (top+i)*stride[0] + left + j;
yading@10:             *block++ = data[0][x];
yading@10:             x = (top+i)*stride[1] + left + j;
yading@10:             u       += data[1][x];
yading@10:             v       += data[2][x];
yading@10:         }
yading@10: 
yading@10:     *block++ = (u+2)/4;
yading@10:     *block++ = (v+2)/4;
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Create YUV clusters for the entire image
yading@10:  */
yading@10: static void create_clusters(const AVFrame *frame, int w, int h, uint8_t *yuvClusters)
yading@10: {
yading@10:     int i, j, k, l;
yading@10: 
yading@10:     for (i=0; i<h; i+=4)
yading@10:         for (j=0; j<w; j+=4) {
yading@10:             for (k=0; k < 2; k++)
yading@10:                 for (l=0; l < 2; l++)
yading@10:                     frame_block_to_cell(yuvClusters + (l + 2*k)*6, frame->data,
yading@10:                                         i+2*k, j+2*l, frame->linesize);
yading@10:             yuvClusters += 24;
yading@10:         }
yading@10: }
yading@10: 
yading@10: static void generate_codebook(RoqContext *enc, RoqTempdata *tempdata,
yading@10:                               int *points, int inputCount, roq_cell *results,
yading@10:                               int size, int cbsize)
yading@10: {
yading@10:     int i, j, k;
yading@10:     int c_size = size*size/4;
yading@10:     int *buf;
yading@10:     int *codebook = av_malloc(6*c_size*cbsize*sizeof(int));
yading@10:     int *closest_cb;
yading@10: 
yading@10:     if (size == 4)
yading@10:         closest_cb = av_malloc(6*c_size*inputCount*sizeof(int));
yading@10:     else
yading@10:         closest_cb = tempdata->closest_cb2;
yading@10: 
yading@10:     ff_init_elbg(points, 6*c_size, inputCount, codebook, cbsize, 1, closest_cb, &enc->randctx);
yading@10:     ff_do_elbg(points, 6*c_size, inputCount, codebook, cbsize, 1, closest_cb, &enc->randctx);
yading@10: 
yading@10:     if (size == 4)
yading@10:         av_free(closest_cb);
yading@10: 
yading@10:     buf = codebook;
yading@10:     for (i=0; i<cbsize; i++)
yading@10:         for (k=0; k<c_size; k++) {
yading@10:             for(j=0; j<4; j++)
yading@10:                 results->y[j] = *buf++;
yading@10: 
yading@10:             results->u =    (*buf++ + CHROMA_BIAS/2)/CHROMA_BIAS;
yading@10:             results->v =    (*buf++ + CHROMA_BIAS/2)/CHROMA_BIAS;
yading@10:             results++;
yading@10:         }
yading@10: 
yading@10:     av_free(codebook);
yading@10: }
yading@10: 
yading@10: static void generate_new_codebooks(RoqContext *enc, RoqTempdata *tempData)
yading@10: {
yading@10:     int i,j;
yading@10:     RoqCodebooks *codebooks = &tempData->codebooks;
yading@10:     int max = enc->width*enc->height/16;
yading@10:     uint8_t mb2[3*4];
yading@10:     roq_cell *results4 = av_malloc(sizeof(roq_cell)*MAX_CBS_4x4*4);
yading@10:     uint8_t *yuvClusters=av_malloc(sizeof(int)*max*6*4);
yading@10:     int *points = av_malloc(max*6*4*sizeof(int));
yading@10:     int bias;
yading@10: 
yading@10:     /* Subsample YUV data */
yading@10:     create_clusters(enc->frame_to_enc, enc->width, enc->height, yuvClusters);
yading@10: 
yading@10:     /* Cast to integer and apply chroma bias */
yading@10:     for (i=0; i<max*24; i++) {
yading@10:         bias = ((i%6)<4) ? 1 : CHROMA_BIAS;
yading@10:         points[i] = bias*yuvClusters[i];
yading@10:     }
yading@10: 
yading@10:     /* Create 4x4 codebooks */
yading@10:     generate_codebook(enc, tempData, points, max, results4, 4, MAX_CBS_4x4);
yading@10: 
yading@10:     codebooks->numCB4 = MAX_CBS_4x4;
yading@10: 
yading@10:     tempData->closest_cb2 = av_malloc(max*4*sizeof(int));
yading@10: 
yading@10:     /* Create 2x2 codebooks */
yading@10:     generate_codebook(enc, tempData, points, max*4, enc->cb2x2, 2, MAX_CBS_2x2);
yading@10: 
yading@10:     codebooks->numCB2 = MAX_CBS_2x2;
yading@10: 
yading@10:     /* Unpack 2x2 codebook clusters */
yading@10:     for (i=0; i<codebooks->numCB2; i++)
yading@10:         unpack_roq_cell(enc->cb2x2 + i, codebooks->unpacked_cb2 + i*2*2*3);
yading@10: 
yading@10:     /* Index all 4x4 entries to the 2x2 entries, unpack, and enlarge */
yading@10:     for (i=0; i<codebooks->numCB4; i++) {
yading@10:         for (j=0; j<4; j++) {
yading@10:             unpack_roq_cell(&results4[4*i + j], mb2);
yading@10:             index_mb(mb2, codebooks->unpacked_cb2, codebooks->numCB2,
yading@10:                      &enc->cb4x4[i].idx[j], 2);
yading@10:         }
yading@10:         unpack_roq_qcell(codebooks->unpacked_cb2, enc->cb4x4 + i,
yading@10:                          codebooks->unpacked_cb4 + i*4*4*3);
yading@10:         enlarge_roq_mb4(codebooks->unpacked_cb4 + i*4*4*3,
yading@10:                         codebooks->unpacked_cb4_enlarged + i*8*8*3);
yading@10:     }
yading@10: 
yading@10:     av_free(yuvClusters);
yading@10:     av_free(points);
yading@10:     av_free(results4);
yading@10: }
yading@10: 
yading@10: static void roq_encode_video(RoqContext *enc)
yading@10: {
yading@10:     RoqTempdata *tempData = enc->tmpData;
yading@10:     int i;
yading@10: 
yading@10:     memset(tempData, 0, sizeof(*tempData));
yading@10: 
yading@10:     create_cel_evals(enc, tempData);
yading@10: 
yading@10:     generate_new_codebooks(enc, tempData);
yading@10: 
yading@10:     if (enc->framesSinceKeyframe >= 1) {
yading@10:         motion_search(enc, 8);
yading@10:         motion_search(enc, 4);
yading@10:     }
yading@10: 
yading@10:  retry_encode:
yading@10:     for (i=0; i<enc->width*enc->height/64; i++)
yading@10:         gather_data_for_cel(tempData->cel_evals + i, enc, tempData);
yading@10: 
yading@10:     /* Quake 3 can't handle chunks bigger than 65535 bytes */
yading@10:     if (tempData->mainChunkSize/8 > 65535) {
yading@10:         av_log(enc->avctx, AV_LOG_ERROR,
yading@10:                "Warning, generated a frame too big (%d > 65535), "
yading@10:                "try using a smaller qscale value.\n",
yading@10:                tempData->mainChunkSize/8);
yading@10:         enc->lambda *= 1.5;
yading@10:         tempData->mainChunkSize = 0;
yading@10:         memset(tempData->used_option, 0, sizeof(tempData->used_option));
yading@10:         memset(tempData->codebooks.usedCB4, 0,
yading@10:                sizeof(tempData->codebooks.usedCB4));
yading@10:         memset(tempData->codebooks.usedCB2, 0,
yading@10:                sizeof(tempData->codebooks.usedCB2));
yading@10: 
yading@10:         goto retry_encode;
yading@10:     }
yading@10: 
yading@10:     remap_codebooks(enc, tempData);
yading@10: 
yading@10:     write_codebooks(enc, tempData);
yading@10: 
yading@10:     reconstruct_and_encode_image(enc, tempData, enc->width, enc->height,
yading@10:                                  enc->width*enc->height/64);
yading@10: 
yading@10:     enc->avctx->coded_frame = enc->current_frame;
yading@10: 
yading@10:     /* Rotate frame history */
yading@10:     FFSWAP(AVFrame *, enc->current_frame, enc->last_frame);
yading@10:     FFSWAP(motion_vect *, enc->last_motion4, enc->this_motion4);
yading@10:     FFSWAP(motion_vect *, enc->last_motion8, enc->this_motion8);
yading@10: 
yading@10:     av_free(tempData->cel_evals);
yading@10:     av_free(tempData->closest_cb2);
yading@10: 
yading@10:     enc->framesSinceKeyframe++;
yading@10: }
yading@10: 
yading@10: static int roq_encode_end(AVCodecContext *avctx)
yading@10: {
yading@10:     RoqContext *enc = avctx->priv_data;
yading@10: 
yading@10:     av_frame_free(&enc->current_frame);
yading@10:     av_frame_free(&enc->last_frame);
yading@10: 
yading@10:     av_free(enc->tmpData);
yading@10:     av_free(enc->this_motion4);
yading@10:     av_free(enc->last_motion4);
yading@10:     av_free(enc->this_motion8);
yading@10:     av_free(enc->last_motion8);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int roq_encode_init(AVCodecContext *avctx)
yading@10: {
yading@10:     RoqContext *enc = avctx->priv_data;
yading@10: 
yading@10:     av_lfg_init(&enc->randctx, 1);
yading@10: 
yading@10:     enc->framesSinceKeyframe = 0;
yading@10:     if ((avctx->width & 0xf) || (avctx->height & 0xf)) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Dimensions must be divisible by 16\n");
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     if (((avctx->width)&(avctx->width-1))||((avctx->height)&(avctx->height-1)))
yading@10:         av_log(avctx, AV_LOG_ERROR, "Warning: dimensions not power of two\n");
yading@10: 
yading@10:     enc->width = avctx->width;
yading@10:     enc->height = avctx->height;
yading@10: 
yading@10:     enc->framesSinceKeyframe = 0;
yading@10:     enc->first_frame = 1;
yading@10: 
yading@10:     enc->last_frame    = av_frame_alloc();
yading@10:     enc->current_frame = av_frame_alloc();
yading@10:     if (!enc->last_frame || !enc->current_frame) {
yading@10:         roq_encode_end(avctx);
yading@10:         return AVERROR(ENOMEM);
yading@10:     }
yading@10: 
yading@10:     enc->tmpData      = av_malloc(sizeof(RoqTempdata));
yading@10: 
yading@10:     enc->this_motion4 =
yading@10:         av_mallocz((enc->width*enc->height/16)*sizeof(motion_vect));
yading@10: 
yading@10:     enc->last_motion4 =
yading@10:         av_malloc ((enc->width*enc->height/16)*sizeof(motion_vect));
yading@10: 
yading@10:     enc->this_motion8 =
yading@10:         av_mallocz((enc->width*enc->height/64)*sizeof(motion_vect));
yading@10: 
yading@10:     enc->last_motion8 =
yading@10:         av_malloc ((enc->width*enc->height/64)*sizeof(motion_vect));
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static void roq_write_video_info_chunk(RoqContext *enc)
yading@10: {
yading@10:     /* ROQ info chunk */
yading@10:     bytestream_put_le16(&enc->out_buf, RoQ_INFO);
yading@10: 
yading@10:     /* Size: 8 bytes */
yading@10:     bytestream_put_le32(&enc->out_buf, 8);
yading@10: 
yading@10:     /* Unused argument */
yading@10:     bytestream_put_byte(&enc->out_buf, 0x00);
yading@10:     bytestream_put_byte(&enc->out_buf, 0x00);
yading@10: 
yading@10:     /* Width */
yading@10:     bytestream_put_le16(&enc->out_buf, enc->width);
yading@10: 
yading@10:     /* Height */
yading@10:     bytestream_put_le16(&enc->out_buf, enc->height);
yading@10: 
yading@10:     /* Unused in Quake 3, mimics the output of the real encoder */
yading@10:     bytestream_put_byte(&enc->out_buf, 0x08);
yading@10:     bytestream_put_byte(&enc->out_buf, 0x00);
yading@10:     bytestream_put_byte(&enc->out_buf, 0x04);
yading@10:     bytestream_put_byte(&enc->out_buf, 0x00);
yading@10: }
yading@10: 
yading@10: static int roq_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
yading@10:                             const AVFrame *frame, int *got_packet)
yading@10: {
yading@10:     RoqContext *enc = avctx->priv_data;
yading@10:     int size, ret;
yading@10: 
yading@10:     enc->avctx = avctx;
yading@10: 
yading@10:     enc->frame_to_enc = frame;
yading@10: 
yading@10:     if (frame->quality)
yading@10:         enc->lambda = frame->quality - 1;
yading@10:     else
yading@10:         enc->lambda = 2*ROQ_LAMBDA_SCALE;
yading@10: 
yading@10:     /* 138 bits max per 8x8 block +
yading@10:      *     256 codebooks*(6 bytes 2x2 + 4 bytes 4x4) + 8 bytes frame header */
yading@10:     size = ((enc->width * enc->height / 64) * 138 + 7) / 8 + 256 * (6 + 4) + 8;
yading@10:     if ((ret = ff_alloc_packet2(avctx, pkt, size)) < 0)
yading@10:         return ret;
yading@10:     enc->out_buf = pkt->data;
yading@10: 
yading@10:     /* Check for I frame */
yading@10:     if (enc->framesSinceKeyframe == avctx->gop_size)
yading@10:         enc->framesSinceKeyframe = 0;
yading@10: 
yading@10:     if (enc->first_frame) {
yading@10:         /* Alloc memory for the reconstruction data (we must know the stride
yading@10:          for that) */
yading@10:         if ((ret = ff_get_buffer(avctx, enc->current_frame, 0)) < 0 ||
yading@10:             (ret = ff_get_buffer(avctx, enc->last_frame,    0)) < 0)
yading@10:             return ret;
yading@10: 
yading@10:         /* Before the first video frame, write a "video info" chunk */
yading@10:         roq_write_video_info_chunk(enc);
yading@10: 
yading@10:         enc->first_frame = 0;
yading@10:     }
yading@10: 
yading@10:     /* Encode the actual frame */
yading@10:     roq_encode_video(enc);
yading@10: 
yading@10:     pkt->size   = enc->out_buf - pkt->data;
yading@10:     if (enc->framesSinceKeyframe == 1)
yading@10:         pkt->flags |= AV_PKT_FLAG_KEY;
yading@10:     *got_packet = 1;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: AVCodec ff_roq_encoder = {
yading@10:     .name                 = "roqvideo",
yading@10:     .type                 = AVMEDIA_TYPE_VIDEO,
yading@10:     .id                   = AV_CODEC_ID_ROQ,
yading@10:     .priv_data_size       = sizeof(RoqContext),
yading@10:     .init                 = roq_encode_init,
yading@10:     .encode2              = roq_encode_frame,
yading@10:     .close                = roq_encode_end,
yading@10:     .supported_framerates = (const AVRational[]){ {30,1}, {0,0} },
yading@10:     .pix_fmts             = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV444P,
yading@10:                                                         AV_PIX_FMT_NONE },
yading@10:     .long_name            = NULL_IF_CONFIG_SMALL("id RoQ video"),
yading@10: };