X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fsvq1.c;h=55595b7baaebb1e7f304383043b5f01964d41895;hb=6aaf3abd6d7a5690effa7ae298ba54fd62df7525;hp=48b25ec4767146937012d0dc3302c12f6d93ce2b;hpb=fa2ae822063a8e41ef3c0713a6872899153c1027;p=ffmpeg diff --git a/libavcodec/svq1.c b/libavcodec/svq1.c index 48b25ec4767..55595b7baae 100644 --- a/libavcodec/svq1.c +++ b/libavcodec/svq1.c @@ -1,21 +1,23 @@ /* - * + * * Copyright (C) 2002 the xine project * Copyright (C) 2002 the ffmpeg project - * - * This library is free software; you can redistribute it and/or + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. + * version 2.1 of the License, or (at your option) any later version. * - * This library is distributed in the hope that it will be useful, + * FFmpeg is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA * * (SVQ1 Decoder) * Ported to mplayer by Arpi @@ -48,6 +50,8 @@ #undef NDEBUG #include +extern const uint8_t mvtab[33][2]; + static VLC svq1_block_type; static VLC svq1_motion_component; static VLC svq1_intra_multistage[6]; @@ -55,21 +59,21 @@ static VLC svq1_inter_multistage[6]; static VLC svq1_intra_mean; static VLC svq1_inter_mean; -#define MEDIAN(a,b,c) (((a < b) != (b >= c)) ? b : (((a < c) != (c > b)) ? c : a)) - -#define SVQ1_BLOCK_SKIP 0 -#define SVQ1_BLOCK_INTER 1 -#define SVQ1_BLOCK_INTER_4V 2 -#define SVQ1_BLOCK_INTRA 3 +#define SVQ1_BLOCK_SKIP 0 +#define SVQ1_BLOCK_INTER 1 +#define SVQ1_BLOCK_INTER_4V 2 +#define SVQ1_BLOCK_INTRA 3 typedef struct SVQ1Context { - + MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independent of MpegEncContext, so this will be removed then (FIXME/XXX) AVCodecContext *avctx; DSPContext dsp; AVFrame picture; + AVFrame current_picture; + AVFrame last_picture; PutBitContext pb; GetBitContext gb; - + PutBitContext reorder_pb[6]; //why ooh why this sick breadth first order, everything is slower and more complex int frame_width; @@ -83,14 +87,18 @@ typedef struct SVQ1Context { int c_block_width; int c_block_height; - unsigned char *c_plane; + uint16_t *mb_type; + uint32_t *dummy; + int16_t (*motion_val8[3])[2]; + int16_t (*motion_val16[3])[2]; + int64_t rd_total; } SVQ1Context; /* motion vector (prediction) */ typedef struct svq1_pmv_s { - int x; - int y; + int x; + int y; } svq1_pmv_t; #include "svq1_cb.h" @@ -170,65 +178,65 @@ static const uint8_t string_table[256] = { for (; level > 0; i++) {\ /* process next depth */\ if (i == m) {\ - m = n;\ - if (--level == 0)\ - break;\ + m = n;\ + if (--level == 0)\ + break;\ }\ /* divide block if next bit set */\ if (get_bits (bitbuf, 1) == 0)\ - break;\ + break;\ /* add child nodes */\ list[n++] = list[i];\ list[n++] = list[i] + (((level & 1) ? pitch : 1) << ((level / 2) + 1));\ } #define SVQ1_ADD_CODEBOOK()\ - /* add codebook entries to vector */\ - for (j=0; j < stages; j++) {\ - n3 = codebook[entries[j]] ^ 0x80808080;\ - n1 += ((n3 & 0xFF00FF00) >> 8);\ - n2 += (n3 & 0x00FF00FF);\ - }\ + /* add codebook entries to vector */\ + for (j=0; j < stages; j++) {\ + n3 = codebook[entries[j]] ^ 0x80808080;\ + n1 += ((n3 & 0xFF00FF00) >> 8);\ + n2 += (n3 & 0x00FF00FF);\ + }\ \ - /* clip to [0..255] */\ - if (n1 & 0xFF00FF00) {\ - n3 = ((( n1 >> 15) & 0x00010001) | 0x01000100) - 0x00010001;\ - n1 += 0x7F007F00;\ - n1 |= (((~n1 >> 15) & 0x00010001) | 0x01000100) - 0x00010001;\ - n1 &= (n3 & 0x00FF00FF);\ - }\ + /* clip to [0..255] */\ + if (n1 & 0xFF00FF00) {\ + n3 = ((( n1 >> 15) & 0x00010001) | 0x01000100) - 0x00010001;\ + n1 += 0x7F007F00;\ + n1 |= (((~n1 >> 15) & 0x00010001) | 0x01000100) - 0x00010001;\ + n1 &= (n3 & 0x00FF00FF);\ + }\ \ - if (n2 & 0xFF00FF00) {\ - n3 = ((( n2 >> 15) & 0x00010001) | 0x01000100) - 0x00010001;\ - n2 += 0x7F007F00;\ - n2 |= (((~n2 >> 15) & 0x00010001) | 0x01000100) - 0x00010001;\ - n2 &= (n3 & 0x00FF00FF);\ - } + if (n2 & 0xFF00FF00) {\ + n3 = ((( n2 >> 15) & 0x00010001) | 0x01000100) - 0x00010001;\ + n2 += 0x7F007F00;\ + n2 |= (((~n2 >> 15) & 0x00010001) | 0x01000100) - 0x00010001;\ + n2 &= (n3 & 0x00FF00FF);\ + } #define SVQ1_DO_CODEBOOK_INTRA()\ for (y=0; y < height; y++) {\ - for (x=0; x < (width / 4); x++, codebook++) {\ - n1 = n4;\ - n2 = n4;\ - SVQ1_ADD_CODEBOOK()\ - /* store result */\ - dst[x] = (n1 << 8) | n2;\ - }\ - dst += (pitch / 4);\ + for (x=0; x < (width / 4); x++, codebook++) {\ + n1 = n4;\ + n2 = n4;\ + SVQ1_ADD_CODEBOOK()\ + /* store result */\ + dst[x] = (n1 << 8) | n2;\ + }\ + dst += (pitch / 4);\ } #define SVQ1_DO_CODEBOOK_NONINTRA()\ for (y=0; y < height; y++) {\ - for (x=0; x < (width / 4); x++, codebook++) {\ - n3 = dst[x];\ - /* add mean value to vector */\ - n1 = ((n3 & 0xFF00FF00) >> 8) + n4;\ - n2 = (n3 & 0x00FF00FF) + n4;\ - SVQ1_ADD_CODEBOOK()\ - /* store result */\ - dst[x] = (n1 << 8) | n2;\ - }\ - dst += (pitch / 4);\ + for (x=0; x < (width / 4); x++, codebook++) {\ + n3 = dst[x];\ + /* add mean value to vector */\ + n1 = ((n3 & 0xFF00FF00) >> 8) + n4;\ + n2 = (n3 & 0x00FF00FF) + n4;\ + SVQ1_ADD_CODEBOOK()\ + /* store result */\ + dst[x] = (n1 << 8) | n2;\ + }\ + dst += (pitch / 4);\ } #define SVQ1_CALC_CODEBOOK_ENTRIES(cbook)\ @@ -236,7 +244,7 @@ static const uint8_t string_table[256] = { bit_cache = get_bits (bitbuf, 4*stages);\ /* calculate codebook entries for this vector */\ for (j=0; j < stages; j++) {\ - entries[j] = (((bit_cache >> (4*(stages - j - 1))) & 0xF) + 16*j) << (level + 1);\ + entries[j] = (((bit_cache >> (4*(stages - j - 1))) & 0xF) + 16*j) << (level + 1);\ }\ mean -= (stages * 128);\ n4 = ((mean + (mean >> 31)) << 16) | (mean & 0xFFFF); @@ -246,9 +254,9 @@ static int svq1_decode_block_intra (GetBitContext *bitbuf, uint8_t *pixels, int uint8_t *list[63]; uint32_t *dst; const uint32_t *codebook; - int entries[6]; - int i, j, m, n; - int mean, stages; + int entries[6]; + int i, j, m, n; + int mean, stages; unsigned x, y, width, height, level; uint32_t n1, n2, n3, n4; @@ -268,24 +276,24 @@ static int svq1_decode_block_intra (GetBitContext *bitbuf, uint8_t *pixels, int stages = get_vlc2(bitbuf, svq1_intra_multistage[level].table, 3, 3) - 1; if (stages == -1) { - for (y=0; y < height; y++) { - memset (&dst[y*(pitch / 4)], 0, width); - } - continue; /* skip vector */ + for (y=0; y < height; y++) { + memset (&dst[y*(pitch / 4)], 0, width); + } + continue; /* skip vector */ } if ((stages > 0) && (level >= 4)) { #ifdef DEBUG_SVQ1 av_log(s->avctx, AV_LOG_INFO, "Error (svq1_decode_block_intra): invalid vector: stages=%i level=%i\n",stages,level); #endif - return -1; /* invalid vector */ + return -1; /* invalid vector */ } mean = get_vlc2(bitbuf, svq1_intra_mean.table, 8, 3); if (stages == 0) { for (y=0; y < height; y++) { - memset (&dst[y*(pitch / 4)], mean, width); + memset (&dst[y*(pitch / 4)], mean, width); } } else { SVQ1_CALC_CODEBOOK_ENTRIES(svq1_intra_codebooks); @@ -301,10 +309,10 @@ static int svq1_decode_block_non_intra (GetBitContext *bitbuf, uint8_t *pixels, uint8_t *list[63]; uint32_t *dst; const uint32_t *codebook; - int entries[6]; - int i, j, m, n; - int mean, stages; - int x, y, width, height, level; + int entries[6]; + int i, j, m, n; + int mean, stages; + int x, y, width, height, level; uint32_t n1, n2, n3, n4; /* initialize list for breadth first processing of vectors */ @@ -322,13 +330,13 @@ static int svq1_decode_block_non_intra (GetBitContext *bitbuf, uint8_t *pixels, /* get number of stages (-1 skips vector, 0 for mean only) */ stages = get_vlc2(bitbuf, svq1_inter_multistage[level].table, 3, 2) - 1; - if (stages == -1) continue; /* skip vector */ + if (stages == -1) continue; /* skip vector */ if ((stages > 0) && (level >= 4)) { #ifdef DEBUG_SVQ1 av_log(s->avctx, AV_LOG_INFO, "Error (svq1_decode_block_non_intra): invalid vector: stages=%i level=%i\n",stages,level); #endif - return -1; /* invalid vector */ + return -1; /* invalid vector */ } mean = get_vlc2(bitbuf, svq1_inter_mean.table, 9, 3) - 256; @@ -340,19 +348,24 @@ static int svq1_decode_block_non_intra (GetBitContext *bitbuf, uint8_t *pixels, } static int svq1_decode_motion_vector (GetBitContext *bitbuf, svq1_pmv_t *mv, svq1_pmv_t **pmv) { - int diff; - int i; + int diff; + int i; for (i=0; i < 2; i++) { /* get motion code */ - diff = get_vlc2(bitbuf, svq1_motion_component.table, 7, 2) - 32; + diff = get_vlc2(bitbuf, svq1_motion_component.table, 7, 2); + if(diff<0) + return -1; + else if(diff){ + if(get_bits1(bitbuf)) diff= -diff; + } /* add median of motion vector predictors and clip result */ if (i == 1) - mv->y = ((diff + MEDIAN(pmv[0]->y, pmv[1]->y, pmv[2]->y)) << 26) >> 26; + mv->y = ((diff + mid_pred(pmv[0]->y, pmv[1]->y, pmv[2]->y)) << 26) >> 26; else - mv->x = ((diff + MEDIAN(pmv[0]->x, pmv[1]->x, pmv[2]->x)) << 26) >> 26; + mv->x = ((diff + mid_pred(pmv[0]->x, pmv[1]->x, pmv[2]->x)) << 26) >> 26; } return 0; @@ -361,7 +374,7 @@ static int svq1_decode_motion_vector (GetBitContext *bitbuf, svq1_pmv_t *mv, svq static void svq1_skip_block (uint8_t *current, uint8_t *previous, int pitch, int x, int y) { uint8_t *src; uint8_t *dst; - int i; + int i; src = &previous[x + y*pitch]; dst = current; @@ -374,13 +387,13 @@ static void svq1_skip_block (uint8_t *current, uint8_t *previous, int pitch, int } static int svq1_motion_inter_block (MpegEncContext *s, GetBitContext *bitbuf, - uint8_t *current, uint8_t *previous, int pitch, - svq1_pmv_t *motion, int x, int y) { + uint8_t *current, uint8_t *previous, int pitch, + svq1_pmv_t *motion, int x, int y) { uint8_t *src; uint8_t *dst; svq1_pmv_t mv; svq1_pmv_t *pmv[3]; - int result; + int result; /* predict and decode motion vector */ pmv[0] = &motion[0]; @@ -398,13 +411,13 @@ static int svq1_motion_inter_block (MpegEncContext *s, GetBitContext *bitbuf, if (result != 0) return result; - motion[0].x = - motion[(x / 8) + 2].x = - motion[(x / 8) + 3].x = mv.x; - motion[0].y = - motion[(x / 8) + 2].y = - motion[(x / 8) + 3].y = mv.y; - + motion[0].x = + motion[(x / 8) + 2].x = + motion[(x / 8) + 3].x = mv.x; + motion[0].y = + motion[(x / 8) + 2].y = + motion[(x / 8) + 3].y = mv.y; + if(y + (mv.y >> 1)<0) mv.y= 0; if(x + (mv.x >> 1)<0) @@ -416,7 +429,7 @@ static int svq1_motion_inter_block (MpegEncContext *s, GetBitContext *bitbuf, if(x + (mv.x >> 1)<0 || y + (mv.y >> 1)<0 || x + (mv.x >> 1) + 16 > w || y + (mv.y >> 1) + 16> h) av_log(s->avctx, AV_LOG_INFO, "%d %d %d %d\n", x, y, x + (mv.x >> 1), y + (mv.y >> 1)); #endif - + src = &previous[(x + (mv.x >> 1)) + (y + (mv.y >> 1))*pitch]; dst = current; @@ -426,13 +439,13 @@ static int svq1_motion_inter_block (MpegEncContext *s, GetBitContext *bitbuf, } static int svq1_motion_inter_4v_block (MpegEncContext *s, GetBitContext *bitbuf, - uint8_t *current, uint8_t *previous, int pitch, - svq1_pmv_t *motion,int x, int y) { + uint8_t *current, uint8_t *previous, int pitch, + svq1_pmv_t *motion,int x, int y) { uint8_t *src; uint8_t *dst; svq1_pmv_t mv; svq1_pmv_t *pmv[4]; - int i, result; + int i, result; /* predict and decode motion vector (0) */ pmv[0] = &motion[0]; @@ -486,7 +499,7 @@ static int svq1_motion_inter_4v_block (MpegEncContext *s, GetBitContext *bitbuf, for (i=0; i < 4; i++) { int mvx= pmv[i]->x + (i&1)*16; int mvy= pmv[i]->y + (i>>1)*16; - + ///XXX /FIXME cliping or padding? if(y + (mvy >> 1)<0) mvy= 0; @@ -501,7 +514,7 @@ static int svq1_motion_inter_4v_block (MpegEncContext *s, GetBitContext *bitbuf, #endif src = &previous[(x + (mvx >> 1)) + (y + (mvy >> 1))*pitch]; dst = current; - + s->dsp.put_pixels_tab[1][((mvy & 1) << 1) | (mvx & 1)](dst,src,pitch,8); /* select next block */ @@ -516,18 +529,18 @@ static int svq1_motion_inter_4v_block (MpegEncContext *s, GetBitContext *bitbuf, } static int svq1_decode_delta_block (MpegEncContext *s, GetBitContext *bitbuf, - uint8_t *current, uint8_t *previous, int pitch, - svq1_pmv_t *motion, int x, int y) { + uint8_t *current, uint8_t *previous, int pitch, + svq1_pmv_t *motion, int x, int y) { uint32_t block_type; - int result = 0; + int result = 0; /* get block type */ block_type = get_vlc2(bitbuf, svq1_block_type.table, 2, 2); /* reset motion vectors */ if (block_type == SVQ1_BLOCK_SKIP || block_type == SVQ1_BLOCK_INTRA) { - motion[0].x = - motion[0].y = + motion[0].x = + motion[0].y = motion[(x / 8) + 2].x = motion[(x / 8) + 2].y = motion[(x / 8) + 3].x = @@ -589,6 +602,7 @@ static uint16_t svq1_packet_checksum (uint8_t *data, int length, int value) { return value; } +#if 0 /* unused, remove? */ static uint16_t svq1_component_checksum (uint16_t *pixels, int pitch, int width, int height, int value) { int x, y; @@ -603,7 +617,9 @@ static uint16_t svq1_component_checksum (uint16_t *pixels, int pitch, return value; } +#endif +#ifdef CONFIG_DECODERS static void svq1_parse_string (GetBitContext *bitbuf, uint8_t *out) { uint8_t seed; int i; @@ -626,9 +642,9 @@ static int svq1_decode_frame_header (GetBitContext *bitbuf,MpegEncContext *s) { /* frame type */ s->pict_type= get_bits (bitbuf, 2)+1; - if(s->pict_type==4) + if(s->pict_type==4) return -1; - + if (s->pict_type == I_TYPE) { /* unknown fields */ @@ -642,9 +658,9 @@ static int svq1_decode_frame_header (GetBitContext *bitbuf,MpegEncContext *s) { } if ((s->f_code ^ 0x10) >= 0x50) { - char msg[256]; + uint8_t msg[256]; - svq1_parse_string (bitbuf, (char *) msg); + svq1_parse_string (bitbuf, msg); av_log(s->avctx, AV_LOG_INFO, "embedded message: \"%s\"\n", (char *) msg); } @@ -689,18 +705,18 @@ static int svq1_decode_frame_header (GetBitContext *bitbuf,MpegEncContext *s) { skip_bits (bitbuf, 8); } } - + return 0; } -static int svq1_decode_frame(AVCodecContext *avctx, +static int svq1_decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size) { MpegEncContext *s=avctx->priv_data; - uint8_t *current, *previous; - int result, i, x, y, width, height; - AVFrame *pict = data; + uint8_t *current, *previous; + int result, i, x, y, width, height; + AVFrame *pict = data; /* initialize bit buffer */ init_get_bits(&s->gb,buf,buf_size*8); @@ -729,12 +745,16 @@ static int svq1_decode_frame(AVCodecContext *avctx, #endif return result; } - + //FIXME this avoids some confusion for "B frames" without 2 references //this should be removed after libavcodec can handle more flexible picture types & ordering if(s->pict_type==B_TYPE && s->last_picture_ptr==NULL) return buf_size; - + if(avctx->hurry_up && s->pict_type==B_TYPE) return buf_size; + if( (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type==B_TYPE) + ||(avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type!=I_TYPE) + || avctx->skip_frame >= AVDISCARD_ALL) + return buf_size; if(MPV_frame_start(s, avctx) < 0) return -1; @@ -764,17 +784,17 @@ static int svq1_decode_frame(AVCodecContext *avctx, if (s->pict_type == I_TYPE) { /* keyframe */ for (y=0; y < height; y+=16) { - for (x=0; x < width; x+=16) { - result = svq1_decode_block_intra (&s->gb, ¤t[x], linesize); - if (result != 0) - { + for (x=0; x < width; x+=16) { + result = svq1_decode_block_intra (&s->gb, ¤t[x], linesize); + if (result != 0) + { //#ifdef DEBUG_SVQ1 - av_log(s->avctx, AV_LOG_INFO, "Error in svq1_decode_block %i (keyframe)\n",result); + av_log(s->avctx, AV_LOG_INFO, "Error in svq1_decode_block %i (keyframe)\n",result); //#endif - return result; - } - } - current += 16*linesize; + return result; + } + } + current += 16*linesize; } } else { svq1_pmv_t pmv[width/8+3]; @@ -782,31 +802,31 @@ static int svq1_decode_frame(AVCodecContext *avctx, memset (pmv, 0, ((width / 8) + 3) * sizeof(svq1_pmv_t)); for (y=0; y < height; y+=16) { - for (x=0; x < width; x+=16) { - result = svq1_decode_delta_block (s, &s->gb, ¤t[x], previous, - linesize, pmv, x, y); - if (result != 0) - { + for (x=0; x < width; x+=16) { + result = svq1_decode_delta_block (s, &s->gb, ¤t[x], previous, + linesize, pmv, x, y); + if (result != 0) + { #ifdef DEBUG_SVQ1 av_log(s->avctx, AV_LOG_INFO, "Error in svq1_decode_delta_block %i\n",result); #endif - return result; - } - } + return result; + } + } - pmv[0].x = - pmv[0].y = 0; + pmv[0].x = + pmv[0].y = 0; - current += 16*linesize; + current += 16*linesize; } } } - + *pict = *(AVFrame*)&s->current_picture; MPV_frame_end(s); - + *data_size=sizeof(AVFrame); return buf_size; } @@ -829,28 +849,28 @@ static int svq1_decode_init(AVCodecContext *avctx) init_vlc(&svq1_block_type, 2, 4, &svq1_block_type_vlc[0][1], 2, 1, - &svq1_block_type_vlc[0][0], 2, 1); + &svq1_block_type_vlc[0][0], 2, 1, 1); - init_vlc(&svq1_motion_component, 7, 65, - &svq1_motion_component_vlc[0][1], 4, 2, - &svq1_motion_component_vlc[0][0], 4, 2); + init_vlc(&svq1_motion_component, 7, 33, + &mvtab[0][1], 2, 1, + &mvtab[0][0], 2, 1, 1); for (i = 0; i < 6; i++) { init_vlc(&svq1_intra_multistage[i], 3, 8, &svq1_intra_multistage_vlc[i][0][1], 2, 1, - &svq1_intra_multistage_vlc[i][0][0], 2, 1); + &svq1_intra_multistage_vlc[i][0][0], 2, 1, 1); init_vlc(&svq1_inter_multistage[i], 3, 8, &svq1_inter_multistage_vlc[i][0][1], 2, 1, - &svq1_inter_multistage_vlc[i][0][0], 2, 1); + &svq1_inter_multistage_vlc[i][0][0], 2, 1, 1); } init_vlc(&svq1_intra_mean, 8, 256, &svq1_intra_mean_vlc[0][1], 4, 2, - &svq1_intra_mean_vlc[0][0], 4, 2); + &svq1_intra_mean_vlc[0][0], 4, 2, 1); init_vlc(&svq1_inter_mean, 9, 512, &svq1_inter_mean_vlc[0][1], 4, 2, - &svq1_inter_mean_vlc[0][0], 4, 2); + &svq1_inter_mean_vlc[0][0], 4, 2, 1); return 0; } @@ -862,9 +882,13 @@ static int svq1_decode_end(AVCodecContext *avctx) MPV_common_end(s); return 0; } +#endif /* CONFIG_DECODERS */ +#ifdef CONFIG_ENCODERS static void svq1_write_header(SVQ1Context *s, int frame_type) { + int i; + /* frame code */ put_bits(&s->pb, 22, 0x20); @@ -881,435 +905,39 @@ static void svq1_write_header(SVQ1Context *s, int frame_type) /* no embedded string either */ /* output 5 unknown bits (2 + 2 + 1) */ - put_bits(&s->pb, 5, 0); - - /* forget about matching up resolutions, just use the free-form - * resolution code (7) for now */ - put_bits(&s->pb, 3, 7); - put_bits(&s->pb, 12, s->frame_width); - put_bits(&s->pb, 12, s->frame_height); - - } - - /* no checksum or extra data (next 2 bits get 0) */ - put_bits(&s->pb, 2, 0); -} - -int level_sizes[6] = { 8, 16, 32, 64, 128, 256 }; -int level_log2_sizes[6] = { 3, 4, 5, 6, 7, 8 }; - -#define IABS(x) ((x < 0) ? (-(x)) : x) - - - -//#define USE_MAD_ALGORITHM - -#ifdef USE_MAD_ALGORITHM - -#define QUALITY_THRESHOLD 100 -#define THRESHOLD_MULTIPLIER 0.6 - -/* This function calculates vector differences using mean absolute - * difference (MAD). */ - -static int encode_vector(SVQ1Context *s, unsigned char *vector, - unsigned int level, int threshold) -{ - int i, j, k; - int mean; - signed short work_vector[256]; - int best_codebook; - int best_score; - int multistage_codebooks[6]; - int number_of_stages = 0; - int8_t *current_codebook; - int total_deviation; - int ret; - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " ** recursive entry point: encoding level %d vector at threshold %d\n", - level, threshold); -#endif - if (level > 5) { - av_log(s->avctx, AV_LOG_INFO, " help! level %d > 5\n", level); - return 0; - } - -#ifdef DEBUG_SVQ1 -for (i = 0; i < level_sizes[level]; i++) - av_log(s->avctx, AV_LOG_INFO, " %02X", vector[i]); -av_log(s->avctx, AV_LOG_INFO, "\n"); -#endif - - /* calculate the mean */ - mean = 0; - for (i = 0; i < level_sizes[level]; i++) - mean += vector[i]; - mean >>= level_log2_sizes[level]; - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " vector mean = 0x%02X\n", mean); -#endif - - /* remove the mean from the vector */ - total_deviation = 0; - for (i = 0; i < level_sizes[level]; i++) { - work_vector[i] = (signed short)vector[i] - mean; - total_deviation += IABS(work_vector[i]); -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " %d", work_vector[i]); -#endif - } - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, "\n total deviation = %d\n", total_deviation); -#endif - - if (total_deviation < threshold) { - -#ifdef DEBUG_SVQ1 - av_log(s->avctx, AV_LOG_INFO, " mean-only encoding found for level %d vector, mean = %d\n", - level, mean); -#endif - - /* indicate that this is the end of the subdivisions */ - if (level > 0) - put_bits(&s->pb, 1, 0); - - /* index 1 in the table indicates mean-only encoding */ - put_bits(&s->pb, svq1_intra_multistage_vlc[level][1][1], - svq1_intra_multistage_vlc[level][1][0]); - put_bits(&s->pb, svq1_intra_mean_vlc[mean][1], - svq1_intra_mean_vlc[mean][0]); - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " mean-only L%d, VLC = (0x%X, %d), mean = %d (0x%X, %d)\n", - level, - svq1_intra_multistage_vlc[level][1 + number_of_stages][0], - svq1_intra_multistage_vlc[level][1 + number_of_stages][1], - mean, - svq1_intra_mean_vlc[mean][0], - svq1_intra_mean_vlc[mean][1]); -#endif - - ret = 0; - - } else { - - if (level <= 3) { - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " multistage VQ search...\n"); -#endif - /* conduct multistage VQ search, for each stage... */ - for (i = 0; i < 6; i++) { - - best_codebook = 0; - best_score = 0x7FFFFFFF; - /* for each codebook in stage */ - for (j = 0; j < 16; j++) { - - total_deviation = 0; - current_codebook = - &svq1_intra_codebooks[level] - [i * level_sizes[level] * 16 + j * level_sizes[level]]; - /* calculate the total deviation for the vector */ - for (k = 0; k < level_sizes[level]; k++) { - total_deviation += - IABS(work_vector[k] - current_codebook[k]); - } - - /* lowest score so far? */ - if (total_deviation < best_score) { - best_score = total_deviation; - best_codebook = j; - } -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " after %d, %d, best codebook is %d with a score of %d (score was %d)\n", - i, j, best_codebook, best_score, total_deviation); -#endif - } - - /* apply the winning codebook to the work vector and check if - * the vector meets the quality threshold */ - total_deviation = 0; - current_codebook = - &svq1_intra_codebooks[level] - [i * level_sizes[level] * 16 + j * level_sizes[level]]; - multistage_codebooks[number_of_stages++] = best_codebook; - for (j = 0; j < level_sizes[level]; j++) { - work_vector[j] = work_vector[j] - current_codebook[j]; - total_deviation += IABS(work_vector[j]); - } - - /* do not go forward with the rest of the search if an acceptable - * codebook combination has been found */ - if (total_deviation < threshold) - break; + put_bits(&s->pb, 5, 2); /* 2 needed by quicktime decoder */ + + for (i = 0; i < 7; i++) + { + if ((svq1_frame_size_table[i].width == s->frame_width) && + (svq1_frame_size_table[i].height == s->frame_height)) + { + put_bits(&s->pb, 3, i); + break; } } - if ((total_deviation < threshold) || (level == 0)) { -#ifdef DEBUG_SVQ1 - av_log(s->avctx, AV_LOG_INFO, " level %d VQ encoding found using mean %d and codebooks", level, mean); - for (i = 0; i < number_of_stages; i++) - av_log(s->avctx, AV_LOG_INFO, " %d", multistage_codebooks[i]); - av_log(s->avctx, AV_LOG_INFO, "\n"); -#endif - - /* indicate that this is the end of the subdivisions */ - if (level > 0) - put_bits(&s->pb, 1, 0); - - /* output the encoding */ - put_bits(&s->pb, - svq1_intra_multistage_vlc[level][1 + number_of_stages][1], - svq1_intra_multistage_vlc[level][1 + number_of_stages][0]); - put_bits(&s->pb, svq1_intra_mean_vlc[mean][1], - svq1_intra_mean_vlc[mean][0]); -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " L%d: multistage = %d (0x%X, %d), mean = %d (0x%X, %d), codebooks = ", - level, - number_of_stages, - svq1_intra_multistage_vlc[level][1 + number_of_stages][0], - svq1_intra_multistage_vlc[level][1 + number_of_stages][1], - mean, - svq1_intra_mean_vlc[mean][0], - svq1_intra_mean_vlc[mean][1]); -#endif - - for (i = 0; i < number_of_stages; i++) -{ -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, "%d ", multistage_codebooks[i]); -#endif - put_bits(&s->pb, 4, multistage_codebooks[i]); -} -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, "\n"); -#endif - - ret = 0; - - } else { - - /* output a subdivision bit to the encoded stream and signal to - * the calling function that this vector could not be - * coded at the requested threshold and needs to be subdivided */ - put_bits(&s->pb, 1, 1); - ret = 1; + if (i == 7) + { + put_bits(&s->pb, 3, 7); + put_bits(&s->pb, 12, s->frame_width); + put_bits(&s->pb, 12, s->frame_height); } } - return ret; + /* no checksum or extra data (next 2 bits get 0) */ + put_bits(&s->pb, 2, 0); } -#else #define QUALITY_THRESHOLD 100 #define THRESHOLD_MULTIPLIER 0.6 -/* This function calculates vector differences using mean square - * error (MSE). */ - -static int encode_vector(SVQ1Context *s, unsigned char *vector, - unsigned int level, int threshold) -{ - int i, j, k; - int mean; - signed short work_vector[256]; - int best_codebook; - int best_score; - int multistage_codebooks[6]; - int number_of_stages = 0; - int8_t *current_codebook; - int mse; - int diff; - int ret; - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " ** recursive entry point: encoding level %d vector at threshold %d\n", - level, threshold); -#endif - if (level > 5) { - av_log(s->avctx, AV_LOG_INFO, " help! level %d > 5\n", level); - return 0; - } - -#ifdef DEBUG_SVQ1 -for (i = 0; i < level_sizes[level]; i++) - av_log(s->avctx, AV_LOG_INFO, " %02X", vector[i]); -av_log(s->avctx, AV_LOG_INFO, "\n"); -#endif - - /* calculate the mean */ - mean = 0; - for (i = 0; i < level_sizes[level]; i++) - mean += vector[i]; - mean >>= level_log2_sizes[level]; - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " vector mean = 0x%02X\n", mean); -#endif - - /* remove the mean from the vector and compute the resulting MSE */ - mse = 0; - for (i = 0; i < level_sizes[level]; i++) { - work_vector[i] = (signed short)vector[i] - mean; - mse += (work_vector[i] * work_vector[i]); -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " %d", work_vector[i]); -#endif - } - mse >>= level_log2_sizes[level]; - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, "\n MSE = %d\n", mse); -#endif - - if (mse < threshold) { - -#ifdef DEBUG_SVQ1 - av_log(s->avctx, AV_LOG_INFO, " mean-only encoding found for level %d vector, mean = %d\n", - level, mean); -#endif - - /* indicate that this is the end of the subdivisions */ - if (level > 0) - put_bits(&s->pb, 1, 0); - - /* index 1 in the table indicates mean-only encoding */ - put_bits(&s->pb, svq1_intra_multistage_vlc[level][1][1], - svq1_intra_multistage_vlc[level][1][0]); - put_bits(&s->pb, svq1_intra_mean_vlc[mean][1], - svq1_intra_mean_vlc[mean][0]); - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " mean-only L%d, VLC = (0x%X, %d), mean = %d (0x%X, %d)\n", - level, - svq1_intra_multistage_vlc[level][1 + number_of_stages][0], - svq1_intra_multistage_vlc[level][1 + number_of_stages][1], - mean, - svq1_intra_mean_vlc[mean][0], - svq1_intra_mean_vlc[mean][1]); +#if defined(HAVE_ALTIVEC) +#undef vector #endif - ret = 0; - - } else { - - if (level <= 3) { - -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " multistage VQ search...\n"); -#endif - /* conduct multistage VQ search, for each stage... */ - for (i = 0; i < 6; i++) { - - best_codebook = 0; - best_score = 0x7FFFFFFF; - /* for each codebook in stage */ - for (j = 0; j < 16; j++) { - - mse = 0; - current_codebook = - &svq1_intra_codebooks[level] - [i * level_sizes[level] * 16 + j * level_sizes[level]]; - /* calculate the MSE for this vector */ - for (k = 0; k < level_sizes[level]; k++) { - diff = work_vector[k] - current_codebook[k]; - mse += (diff * diff); - } - mse >>= level_log2_sizes[level]; - - /* lowest score so far? */ - if (mse < best_score) { - best_score = mse; - best_codebook = j; - } -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " after %d, %d, best codebook is %d with a score of %d (score was %d)\n", - i, j, best_codebook, best_score, mse); -#endif - } - - /* apply the winning codebook to the work vector and check if - * the vector meets the quality threshold */ - mse = 0; - current_codebook = - &svq1_intra_codebooks[level] - [i * level_sizes[level] * 16 + j * level_sizes[level]]; - multistage_codebooks[number_of_stages++] = best_codebook; - for (j = 0; j < level_sizes[level]; j++) { - work_vector[j] = work_vector[j] - current_codebook[j]; - mse += (work_vector[j] * work_vector[j]); - } - mse >>= level_log2_sizes[level]; - - /* do not go forward with the rest of the search if an acceptable - * codebook combination has been found */ - if (mse < threshold) - break; - } - } - - if ((mse < threshold) || (level == 0)) { -#ifdef DEBUG_SVQ1 - av_log(s->avctx, AV_LOG_INFO, " level %d VQ encoding found using mean %d and codebooks", level, mean); - for (i = 0; i < number_of_stages; i++) - av_log(s->avctx, AV_LOG_INFO, " %d", multistage_codebooks[i]); - av_log(s->avctx, AV_LOG_INFO, "\n"); -#endif - - /* indicate that this is the end of the subdivisions */ - if (level > 0) - put_bits(&s->pb, 1, 0); - - /* output the encoding */ - put_bits(&s->pb, - svq1_intra_multistage_vlc[level][1 + number_of_stages][1], - svq1_intra_multistage_vlc[level][1 + number_of_stages][0]); - put_bits(&s->pb, svq1_intra_mean_vlc[mean][1], - svq1_intra_mean_vlc[mean][0]); -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " L%d: multistage = %d (0x%X, %d), mean = %d (0x%X, %d), codebooks = ", - level, - number_of_stages, - svq1_intra_multistage_vlc[level][1 + number_of_stages][0], - svq1_intra_multistage_vlc[level][1 + number_of_stages][1], - mean, - svq1_intra_mean_vlc[mean][0], - svq1_intra_mean_vlc[mean][1]); -#endif - - for (i = 0; i < number_of_stages; i++) -{ -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, "%d ", multistage_codebooks[i]); -#endif - put_bits(&s->pb, 4, multistage_codebooks[i]); -} -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, "\n"); -#endif - - ret = 0; - - } else { - - /* output a subdivision bit to the encoded stream and signal to - * the calling function that this vector could not be - * coded at the requested threshold and needs to be subdivided */ - put_bits(&s->pb, 1, 1); - ret = 1; - } - } - - return ret; -} -#endif - -static int encode_block(SVQ1Context *s, uint8_t *src, int stride, int level, int threshold, int lambda){ +static int encode_block(SVQ1Context *s, uint8_t *src, uint8_t *ref, uint8_t *decoded, int stride, int level, int threshold, int lambda, int intra){ int count, y, x, i, j, split, best_mean, best_score, best_count; int best_vector[6]; int block_sum[7]= {0, 0, 0, 0, 0, 0}; @@ -1317,14 +945,37 @@ static int encode_block(SVQ1Context *s, uint8_t *src, int stride, int level, int int h= 2<<((level+1)>>1); int size=w*h; int16_t block[7][256]; - + const int8_t *codebook_sum, *codebook; + const uint16_t (*mean_vlc)[2]; + const uint8_t (*multistage_vlc)[2]; + best_score=0; - for(y=0; y>1)) >> (level+3); - assert(mean >-50 && mean<300); - mean= clip(mean, 0, 255); + assert(mean >-300 && mean<300); + if(intra) mean= av_clip(mean, 0, 255); + else mean= av_clip(mean, -256, 255); score= sqr - ((diff*(int64_t)diff)>>(level+3)); //FIXME 64bit slooow if(score < best_vector_score){ best_vector_score= score; @@ -1363,17 +1014,17 @@ static int encode_block(SVQ1Context *s, uint8_t *src, int stride, int level, int best_vector_mean= mean; } } - assert(best_vector_mean != -99); - vector= svq1_intra_codebooks[level] + stage*size*16 + best_vector[stage]*size; + assert(best_vector_mean != -999); + vector= codebook + stage*size*16 + best_vector[stage]*size; for(j=0; j threshold && level){ int score=0; @@ -1391,10 +1042,10 @@ static int encode_block(SVQ1Context *s, uint8_t *src, int stride, int level, int for(i=level-1; i>=0; i--){ backup[i]= s->reorder_pb[i]; } - score += encode_block(s, src , stride, level-1, threshold>>1, lambda); - score += encode_block(s, src + offset, stride, level-1, threshold>>1, lambda); + score += encode_block(s, src , ref , decoded , stride, level-1, threshold>>1, lambda, intra); + score += encode_block(s, src + offset, ref + offset, decoded + offset, stride, level-1, threshold>>1, lambda, intra); score += lambda; - + if(score < best_score){ best_score= score; split=1; @@ -1408,48 +1059,43 @@ static int encode_block(SVQ1Context *s, uint8_t *src, int stride, int level, int put_bits(&s->reorder_pb[level], 1, split); if(!split){ - assert(best_mean >= 0 && best_mean<256); + assert((best_mean >= 0 && best_mean<256) || !intra); + assert(best_mean >= -256 && best_mean<256); assert(best_count >=0 && best_count<7); assert(level<4 || best_count==0); - + /* output the encoding */ - put_bits(&s->reorder_pb[level], - svq1_intra_multistage_vlc[level][1 + best_count][1], - svq1_intra_multistage_vlc[level][1 + best_count][0]); - put_bits(&s->reorder_pb[level], svq1_intra_mean_vlc[best_mean][1], - svq1_intra_mean_vlc[best_mean][0]); + put_bits(&s->reorder_pb[level], + multistage_vlc[1 + best_count][1], + multistage_vlc[1 + best_count][0]); + put_bits(&s->reorder_pb[level], mean_vlc[best_mean][1], + mean_vlc[best_mean][0]); for (i = 0; i < best_count; i++){ assert(best_vector[i]>=0 && best_vector[i]<16); put_bits(&s->reorder_pb[level], 4, best_vector[i]); } + + for(y=0; yavctx, AV_LOG_INFO, "********* frame #%d\n", frame++); -#endif + const int lambda= (s->picture.quality*s->picture.quality) >> (2*FF_LAMBDA_SHIFT); /* figure out the acceptable level thresholds in advance */ threshold[5] = QUALITY_THRESHOLD; @@ -1459,146 +1105,205 @@ av_log(s->avctx, AV_LOG_INFO, "********* frame #%d\n", frame++); block_width = (width + 15) / 16; block_height = (height + 15) / 16; - for (y = 0; y < block_height; y++) { + if(s->picture.pict_type == P_TYPE){ + s->m.avctx= s->avctx; + s->m.current_picture_ptr= &s->m.current_picture; + s->m.last_picture_ptr = &s->m.last_picture; + s->m.last_picture.data[0]= ref_plane; + s->m.linesize= + s->m.last_picture.linesize[0]= + s->m.new_picture.linesize[0]= + s->m.current_picture.linesize[0]= stride; + s->m.width= width; + s->m.height= height; + s->m.mb_width= block_width; + s->m.mb_height= block_height; + s->m.mb_stride= s->m.mb_width+1; + s->m.b8_stride= 2*s->m.mb_width+1; + s->m.f_code=1; + s->m.pict_type= s->picture.pict_type; + s->m.me_method= s->avctx->me_method; + s->m.me.scene_change_score=0; + s->m.flags= s->avctx->flags; +// s->m.out_format = FMT_H263; +// s->m.unrestricted_mv= 1; + + s->m.lambda= s->picture.quality; + s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7); + s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT; + + if(!s->motion_val8[plane]){ + s->motion_val8 [plane]= av_mallocz((s->m.b8_stride*block_height*2 + 2)*2*sizeof(int16_t)); + s->motion_val16[plane]= av_mallocz((s->m.mb_stride*(block_height + 2) + 1)*2*sizeof(int16_t)); + } - for (x = 0; x < block_width; x++) { - uint8_t reorder_buffer[6][7*32]; + s->m.mb_type= s->mb_type; -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, "* level 5 vector @ %d, %d:\n", x * 16, y * 16); -#endif + //dummies, to avoid segfaults + s->m.current_picture.mb_mean= (uint8_t *)s->dummy; + s->m.current_picture.mb_var= (uint16_t*)s->dummy; + s->m.current_picture.mc_mb_var= (uint16_t*)s->dummy; + s->m.current_picture.mb_type= s->dummy; - /* copy the block into the current work buffer */ - left_edge = (y * 16 * stride) + (x * 16); - - for(i=0; i<6; i++){ - init_put_bits(&s->reorder_pb[i], reorder_buffer[i], 7*32); - } - encode_block(s, &plane[left_edge], stride, 5, 256, (s->picture.quality*s->picture.quality) >> (2*FF_LAMBDA_SHIFT)); - for(i=5; i>=0; i--){ - int count= put_bits_count(&s->reorder_pb[i]); - - flush_put_bits(&s->reorder_pb[i]); - ff_copy_bits(&s->pb, s->reorder_pb[i].buf, count); - } - -#if 0 - for (i = 0; i < 256; i += 16) { - memcpy(&buffer0[i], &plane[left_edge], 16); - left_edge += stride; - } - current_buffer = 1; /* this will toggle to 0 immediately */ - - /* perform a breadth-first tree encoding for each vector level */ - subvector_count = 1; /* one subvector at level 5 */ - for (level = 5; level >= 0; level--) { - - vector_count = subvector_count; - subvector_count = 0; - - if (current_buffer == 0) { - current_buffer = 1; - vector = buffer1; - subvectors = buffer0; - } else { - current_buffer = 0; - vector = buffer0; - subvectors = buffer1; - } + s->m.current_picture.motion_val[0]= s->motion_val8[plane] + 2; + s->m.p_mv_table= s->motion_val16[plane] + s->m.mb_stride + 1; + s->m.dsp= s->dsp; //move + ff_init_me(&s->m); - /* iterate through each vector in the list */ - for (i = 0; i < vector_count; i++) { + s->m.me.dia_size= s->avctx->dia_size; + s->m.first_slice_line=1; + for (y = 0; y < block_height; y++) { + uint8_t src[stride*16]; - if (encode_vector(s, vector, level, threshold[level])) { + s->m.new_picture.data[0]= src - y*16*stride; //ugly + s->m.mb_y= y; -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " split to level %d\n", level - 1); -#endif - /* subdivide into 2 subvectors for later processing */ - subvector_count += 2; - - if (level - 1 == 3) { - /* subdivide 16x8 -> 2 8x8 */ - for (j = 0; j < 8; j++) { - /* left half */ - memcpy(subvectors + j * 8, vector + j * 16, 8); - /* right half */ - memcpy(subvectors + 64 + j * 8, - vector + 8 + j * 16, 8); - } - subvectors += 128; - } else if (level - 1 == 1) { - /* subdivide 8x4 -> 2 4x4 */ - for (j = 0; j < 4; j++) { - /* left half */ - memcpy(subvectors + j * 4, vector + j * 8, 4); - /* right half */ - memcpy(subvectors + 16 + j * 4, - vector + 4 + j * 8, 4); - } - subvectors += 32; - } else { - /* first half */ - memcpy(subvectors, vector, level_sizes[level - 1]); - subvectors += level_sizes[level - 1]; - /* second half */ - memcpy(subvectors, vector + level_sizes[level - 1], - level_sizes[level - 1]); - subvectors += level_sizes[level - 1]; - } - } - - vector += level_sizes[level]; - } + for(i=0; i<16 && i + 16*ym.mb_x= x; + ff_init_block_index(&s->m); + ff_update_block_index(&s->m); + + ff_estimate_p_frame_motion(&s->m, x, y); } -#endif + s->m.first_slice_line=0; } + + ff_fix_long_p_mvs(&s->m); + ff_fix_long_mvs(&s->m, NULL, 0, s->m.p_mv_table, s->m.f_code, CANDIDATE_MB_TYPE_INTER, 0); } -} -/* output a plane with a constant mean value; good for debugging and for - * greyscale encoding but only valid for intra frames */ -static void svq1_output_intra_constant_mean(SVQ1Context *s, int block_width, - int block_height, unsigned char mean) -{ - int i; + s->m.first_slice_line=1; + for (y = 0; y < block_height; y++) { + uint8_t src[stride*16]; - /* for each level 5 vector, output the specified mean value */ - for (i = 0; i < block_width * block_height; i++) { + for(i=0; i<16 && i + 16*ypb, 1, 0); + s->m.mb_y= y; + for (x = 0; x < block_width; x++) { + uint8_t reorder_buffer[3][6][7*32]; + int count[3][6]; + int offset = y * 16 * stride + x * 16; + uint8_t *decoded= decoded_plane + offset; + uint8_t *ref= ref_plane + offset; + int score[4]={0,0,0,0}, best; + uint8_t temp[16*stride]; + + if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < 3000){ //FIXME check size + av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); + return -1; + } - /* output a 0 indicating mean-only encoding; use index 1 as that - * maps to code 0 */ - put_bits(&s->pb, svq1_intra_multistage_vlc[5][1][1], - svq1_intra_multistage_vlc[5][1][0]); + s->m.mb_x= x; + ff_init_block_index(&s->m); + ff_update_block_index(&s->m); - /* output a constant mean */ - put_bits(&s->pb, svq1_intra_mean_vlc[mean][1], - svq1_intra_mean_vlc[mean][0]); -#ifdef DEBUG_SVQ1 -av_log(s->avctx, AV_LOG_INFO, " const L5 %d/%d: multistage = 0 (0x%X, %d), mean = %d (0x%X, %d)\n", - i, block_width * block_height, - svq1_intra_multistage_vlc[5][1][0], - svq1_intra_multistage_vlc[5][1][1], - mean, - svq1_intra_mean_vlc[mean][0], - svq1_intra_mean_vlc[mean][1]); -#endif + if(s->picture.pict_type == I_TYPE || (s->m.mb_type[x + y*s->m.mb_stride]&CANDIDATE_MB_TYPE_INTRA)){ + for(i=0; i<6; i++){ + init_put_bits(&s->reorder_pb[i], reorder_buffer[0][i], 7*32); + } + if(s->picture.pict_type == P_TYPE){ + const uint8_t *vlc= svq1_block_type_vlc[SVQ1_BLOCK_INTRA]; + put_bits(&s->reorder_pb[5], vlc[1], vlc[0]); + score[0]= vlc[1]*lambda; + } + score[0]+= encode_block(s, src+16*x, NULL, temp, stride, 5, 64, lambda, 1); + for(i=0; i<6; i++){ + count[0][i]= put_bits_count(&s->reorder_pb[i]); + flush_put_bits(&s->reorder_pb[i]); + } + }else + score[0]= INT_MAX; + + best=0; + + if(s->picture.pict_type == P_TYPE){ + const uint8_t *vlc= svq1_block_type_vlc[SVQ1_BLOCK_INTER]; + int mx, my, pred_x, pred_y, dxy; + int16_t *motion_ptr; + + motion_ptr= h263_pred_motion(&s->m, 0, 0, &pred_x, &pred_y); + if(s->m.mb_type[x + y*s->m.mb_stride]&CANDIDATE_MB_TYPE_INTER){ + for(i=0; i<6; i++) + init_put_bits(&s->reorder_pb[i], reorder_buffer[1][i], 7*32); + + put_bits(&s->reorder_pb[5], vlc[1], vlc[0]); + + s->m.pb= s->reorder_pb[5]; + mx= motion_ptr[0]; + my= motion_ptr[1]; + assert(mx>=-32 && mx<=31); + assert(my>=-32 && my<=31); + assert(pred_x>=-32 && pred_x<=31); + assert(pred_y>=-32 && pred_y<=31); + ff_h263_encode_motion(&s->m, mx - pred_x, 1); + ff_h263_encode_motion(&s->m, my - pred_y, 1); + s->reorder_pb[5]= s->m.pb; + score[1] += lambda*put_bits_count(&s->reorder_pb[5]); + + dxy= (mx&1) + 2*(my&1); + + s->dsp.put_pixels_tab[0][dxy](temp+16, ref + (mx>>1) + stride*(my>>1), stride, 16); + + score[1]+= encode_block(s, src+16*x, temp+16, decoded, stride, 5, 64, lambda, 0); + best= score[1] <= score[0]; + + vlc= svq1_block_type_vlc[SVQ1_BLOCK_SKIP]; + score[2]= s->dsp.sse[0](NULL, src+16*x, ref, stride, 16); + score[2]+= vlc[1]*lambda; + if(score[2] < score[best] && mx==0 && my==0){ + best=2; + s->dsp.put_pixels_tab[0][0](decoded, ref, stride, 16); + for(i=0; i<6; i++){ + count[2][i]=0; + } + put_bits(&s->pb, vlc[1], vlc[0]); + } + } + + if(best==1){ + for(i=0; i<6; i++){ + count[1][i]= put_bits_count(&s->reorder_pb[i]); + flush_put_bits(&s->reorder_pb[i]); + } + }else{ + motion_ptr[0 ] = motion_ptr[1 ]= + motion_ptr[2 ] = motion_ptr[3 ]= + motion_ptr[0+2*s->m.b8_stride] = motion_ptr[1+2*s->m.b8_stride]= + motion_ptr[2+2*s->m.b8_stride] = motion_ptr[3+2*s->m.b8_stride]=0; + } + } + + s->rd_total += score[best]; + + for(i=5; i>=0; i--){ + ff_copy_bits(&s->pb, reorder_buffer[best][i], count[best][i]); + } + if(best==0){ + s->dsp.put_pixels_tab[0][0](decoded, temp, stride, 16); + } + } + s->m.first_slice_line=0; } + return 0; } static int svq1_encode_init(AVCodecContext *avctx) { SVQ1Context * const s = avctx->priv_data; - int i; - unsigned char least_bits_value = 0; - int least_bits; dsputil_init(&s->dsp, avctx); avctx->coded_frame= (AVFrame*)&s->picture; @@ -1612,73 +1317,89 @@ static int svq1_encode_init(AVCodecContext *avctx) s->c_block_width = (s->frame_width / 4 + 15) / 16; s->c_block_height = (s->frame_height / 4 + 15) / 16; -av_log(s->avctx, AV_LOG_INFO, " Hey: %d x %d, %d x %d, %d x %d\n", - s->frame_width, s->frame_height, - s->y_block_width, s->y_block_height, - s->c_block_width, s->c_block_height); - - /* allocate a plane for the U & V planes (color, or C, planes) and - * initialize them to the value that is represented by the fewest bits - * in the mean table; the reasoning behind this is that when the border - * vectors are operated upon and possibly subdivided, the mean will be - * removed resulting in a perfect deviation score of 0 and encoded with - * the minimal possible bits */ - s->c_plane = av_malloc(s->c_block_width * s->c_block_height * 16 * 16); - least_bits = 10000; - for (i = 0; i < 256; i++) - if (svq1_intra_mean_vlc[i][1] < least_bits) { - least_bits = svq1_intra_mean_vlc[i][1]; - least_bits_value = i; - } - memset(s->c_plane, least_bits_value, - s->c_block_width * s->c_block_height * 16 * 16); + s->avctx= avctx; + s->m.avctx= avctx; + s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t)); + s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); + s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); + s->mb_type = av_mallocz((s->y_block_width+1)*s->y_block_height*sizeof(int16_t)); + s->dummy = av_mallocz((s->y_block_width+1)*s->y_block_height*sizeof(int32_t)); + h263_encode_init(&s->m); //mv_penalty return 0; } -static int svq1_encode_frame(AVCodecContext *avctx, unsigned char *buf, +static int svq1_encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data) { SVQ1Context * const s = avctx->priv_data; AVFrame *pict = data; AVFrame * const p= (AVFrame*)&s->picture; + AVFrame temp; + int i; + + if(avctx->pix_fmt != PIX_FMT_YUV410P){ + av_log(avctx, AV_LOG_ERROR, "unsupported pixel format\n"); + return -1; + } + + if(!s->current_picture.data[0]){ + avctx->get_buffer(avctx, &s->current_picture); + avctx->get_buffer(avctx, &s->last_picture); + } + + temp= s->current_picture; + s->current_picture= s->last_picture; + s->last_picture= temp; init_put_bits(&s->pb, buf, buf_size); *p = *pict; - p->pict_type = I_TYPE; - p->key_frame = 1; + p->pict_type = avctx->gop_size && avctx->frame_number % avctx->gop_size ? P_TYPE : I_TYPE; + p->key_frame = p->pict_type == I_TYPE; svq1_write_header(s, p->pict_type); - svq1_encode_plane(s, s->picture.data[0], s->frame_width, s->frame_height, - s->picture.linesize[0]); -// if (avctx->flags & CODEC_FLAG_GRAY) { -if (1) { - svq1_output_intra_constant_mean(s, s->c_block_width * 2, - s->c_block_height * 2, 128); - } else { - svq1_encode_plane(s, s->picture.data[1], s->frame_width / 4, - s->frame_height / 4, s->picture.linesize[1]); - svq1_encode_plane(s, s->picture.data[2], s->frame_width / 4, - s->frame_height / 4, s->picture.linesize[2]); + for(i=0; i<3; i++){ + if(svq1_encode_plane(s, i, + s->picture.data[i], s->last_picture.data[i], s->current_picture.data[i], + s->frame_width / (i?4:1), s->frame_height / (i?4:1), + s->picture.linesize[i], s->current_picture.linesize[i]) < 0) + return -1; } // align_put_bits(&s->pb); while(put_bits_count(&s->pb) & 31) put_bits(&s->pb, 1, 0); + flush_put_bits(&s->pb); + return (put_bits_count(&s->pb) / 8); } static int svq1_encode_end(AVCodecContext *avctx) { SVQ1Context * const s = avctx->priv_data; + int i; + + av_log(avctx, AV_LOG_DEBUG, "RD: %f\n", s->rd_total/(double)(avctx->width*avctx->height*avctx->frame_number)); + + av_freep(&s->m.me.scratchpad); + av_freep(&s->m.me.map); + av_freep(&s->m.me.score_map); + av_freep(&s->mb_type); + av_freep(&s->dummy); - av_free(s->c_plane); + for(i=0; i<3; i++){ + av_freep(&s->motion_val8[i]); + av_freep(&s->motion_val16[i]); + } return 0; } +#endif //CONFIG_ENCODERS + +#ifdef CONFIG_DECODERS AVCodec svq1_decoder = { "svq1", CODEC_TYPE_VIDEO, @@ -1690,7 +1411,9 @@ AVCodec svq1_decoder = { svq1_decode_frame, CODEC_CAP_DR1, .flush= ff_mpeg_flush, + .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV410P, -1}, }; +#endif #ifdef CONFIG_ENCODERS @@ -1702,6 +1425,7 @@ AVCodec svq1_encoder = { svq1_encode_init, svq1_encode_frame, svq1_encode_end, + .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV410P, -1}, }; #endif //CONFIG_ENCODERS