X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fdv.c;h=71022463af2c34a4cc0bbb2d6ded66903cf47b0f;hb=d3b4b74c32cf302d36a4c4d2cce08027f0a22560;hp=5f1eaaa3b4b47f378a6deb90f29c953f1e7e7c99;hpb=901ee2bea3e51ec4b0755f434c11ef1554acb189;p=ffmpeg diff --git a/libavcodec/dv.c b/libavcodec/dv.c index 5f1eaaa3b4b..71022463af2 100644 --- a/libavcodec/dv.c +++ b/libavcodec/dv.c @@ -3,97 +3,124 @@ * Copyright (c) 2002 Fabrice Bellard. * Copyright (c) 2004 Roman Shaposhnik. * - * DV encoder + * DV encoder * Copyright (c) 2003 Roman Shaposhnik. * + * 50 Mbps (DVCPRO50) support + * Copyright (c) 2006 Daniel Maas + * + * 100 Mbps (DVCPRO HD) support + * Initial code by Daniel Maas (funded by BBC R&D) + * Final code by Roman Shaposhnik + * * Many thanks to Dan Dennedy for providing wealth * of DV technical info. * - * 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 */ /** * @file dv.c * DV codec. */ +#define ALT_BITSTREAM_READER #include "avcodec.h" #include "dsputil.h" -#include "mpegvideo.h" +#include "bitstream.h" #include "simple_idct.h" #include "dvdata.h" +//#undef NDEBUG +//#include + typedef struct DVVideoContext { const DVprofile* sys; AVFrame picture; + AVCodecContext *avctx; uint8_t *buf; - + uint8_t dv_zigzag[2][64]; - uint8_t dv_idct_shift[2][2][22][64]; - + uint32_t dv_idct_factor[2][2][22][64]; + uint32_t dv100_idct_factor[4][4][16][64]; + void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size); void (*fdct[2])(DCTELEM *block); void (*idct_put[2])(uint8_t *dest, int line_size, DCTELEM *block); } DVVideoContext; +/* MultiThreading - dv_anchor applies to entire DV codec, not just the avcontext */ +/* one element is needed for each video segment in a DV frame */ +/* at most there are 4 DIF channels * 12 DIF sequences * 27 video segments (1080i50) */ +#define DV_ANCHOR_SIZE (4*12*27) + +static void* dv_anchor[DV_ANCHOR_SIZE]; + #define TEX_VLC_BITS 9 #ifdef DV_CODEC_TINY_TARGET #define DV_VLC_MAP_RUN_SIZE 15 #define DV_VLC_MAP_LEV_SIZE 23 #else -#define DV_VLC_MAP_RUN_SIZE 64 -#define DV_VLC_MAP_LEV_SIZE 512 +#define DV_VLC_MAP_RUN_SIZE 64 +#define DV_VLC_MAP_LEV_SIZE 512 //FIXME sign was removed so this should be /2 but needs check #endif -/* MultiThreading */ -static uint8_t** dv_anchor; - /* XXX: also include quantization */ -static RL_VLC_ELEM *dv_rl_vlc; +static RL_VLC_ELEM dv_rl_vlc[1184]; /* VLC encoding lookup table */ static struct dv_vlc_pair { uint32_t vlc; uint8_t size; -} (*dv_vlc_map)[DV_VLC_MAP_LEV_SIZE] = NULL; +} dv_vlc_map[DV_VLC_MAP_RUN_SIZE][DV_VLC_MAP_LEV_SIZE]; static void dv_build_unquantize_tables(DVVideoContext *s, uint8_t* perm) { - int i, q, j; + int i, q, a; /* NOTE: max left shift is 6 */ for(q = 0; q < 22; q++) { /* 88DCT */ - for(i = 1; i < 64; i++) { - /* 88 table */ - j = perm[i]; - s->dv_idct_shift[0][0][q][j] = - dv_quant_shifts[q][dv_88_areas[i]] + 1; - s->dv_idct_shift[1][0][q][j] = s->dv_idct_shift[0][0][q][j] + 1; + i=1; + for(a = 0; a<4; a++) { + for(; i < dv_quant_areas[a]; i++) { + /* 88 table */ + s->dv_idct_factor[0][0][q][i] = dv_iweight_88[i]<<(dv_quant_shifts[q][a] + 1); + s->dv_idct_factor[1][0][q][i] = s->dv_idct_factor[0][0][q][i]<<1; + + /* 248 table */ + s->dv_idct_factor[0][1][q][i] = dv_iweight_248[i]<<(dv_quant_shifts[q][a] + 1); + s->dv_idct_factor[1][1][q][i] = s->dv_idct_factor[0][1][q][i]<<1; + } } - - /* 248DCT */ - for(i = 1; i < 64; i++) { - /* 248 table */ - s->dv_idct_shift[0][1][q][i] = - dv_quant_shifts[q][dv_248_areas[i]] + 1; - s->dv_idct_shift[1][1][q][i] = s->dv_idct_shift[0][1][q][i] + 1; + } + + for(a = 0; a < 4; a++) { + for(q = 0; q < 16; q++) { + for(i = 1; i < 64; i++) { + s->dv100_idct_factor[0][a][q][i]= (dv100_qstep[q]<<(a+9))*dv_iweight_1080_y[i]; + s->dv100_idct_factor[1][a][q][i]= (dv100_qstep[q]<<(a+9))*dv_iweight_1080_c[i]; + s->dv100_idct_factor[2][a][q][i]= (dv100_qstep[q]<<(a+9))*dv_iweight_720_y[i]; + s->dv100_idct_factor[3][a][q][i]= (dv100_qstep[q]<<(a+9))*dv_iweight_720_c[i]; + } } } } -static int dvvideo_init(AVCodecContext *avctx) +static av_cold int dvvideo_init(AVCodecContext *avctx) { DVVideoContext *s = avctx->priv_data; DSPContext dsp; @@ -109,54 +136,40 @@ static int dvvideo_init(AVCodecContext *avctx) done = 1; - dv_vlc_map = av_mallocz(DV_VLC_MAP_LEV_SIZE*DV_VLC_MAP_RUN_SIZE*sizeof(struct dv_vlc_pair)); - if (!dv_vlc_map) - return -ENOMEM; - - /* dv_anchor lets each thread know its Id */ - dv_anchor = av_malloc(12*27*sizeof(void*)); - if (!dv_anchor) { - av_free(dv_vlc_map); - return -ENOMEM; - } - for (i=0; i<12*27; i++) - dv_anchor[i] = (void*)(size_t)i; - - /* it's faster to include sign bit in a generic VLC parsing scheme */ - for (i=0, j=0; i= DV_VLC_MAP_RUN_SIZE || dv_vlc_level[i] >= DV_VLC_MAP_LEV_SIZE) - continue; - - if (dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size != 0) - continue; - - dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].vlc = dv_vlc_bits[i] << - (!!dv_vlc_level[i]); - dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size = dv_vlc_len[i] + - (!!dv_vlc_level[i]); - } - for (i = 0; i < DV_VLC_MAP_RUN_SIZE; i++) { + free_vlc(&dv_vlc); + + for (i = 0; i < NB_DV_VLC - 1; i++) { + if (dv_vlc_run[i] >= DV_VLC_MAP_RUN_SIZE) + continue; +#ifdef DV_CODEC_TINY_TARGET + if (dv_vlc_level[i] >= DV_VLC_MAP_LEV_SIZE) + continue; +#endif + + if (dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size != 0) + continue; + + dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].vlc = dv_vlc_bits[i] << + (!!dv_vlc_level[i]); + dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size = dv_vlc_len[i] + + (!!dv_vlc_level[i]); + } + for (i = 0; i < DV_VLC_MAP_RUN_SIZE; i++) { #ifdef DV_CODEC_TINY_TARGET - for (j = 1; j < DV_VLC_MAP_LEV_SIZE; j++) { - if (dv_vlc_map[i][j].size == 0) { - dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc | - (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size)); - dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size + - dv_vlc_map[0][j].size; - } - } + for (j = 1; j < DV_VLC_MAP_LEV_SIZE; j++) { + if (dv_vlc_map[i][j].size == 0) { + dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc | + (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size)); + dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size + + dv_vlc_map[0][j].size; + } + } #else - for (j = 1; j < DV_VLC_MAP_LEV_SIZE/2; j++) { - if (dv_vlc_map[i][j].size == 0) { - dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc | - (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size)); - dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size + - dv_vlc_map[0][j].size; - } - dv_vlc_map[i][((uint16_t)(-j))&0x1ff].vlc = - dv_vlc_map[i][j].vlc | 1; - dv_vlc_map[i][((uint16_t)(-j))&0x1ff].size = - dv_vlc_map[i][j].size; - } + for (j = 1; j < DV_VLC_MAP_LEV_SIZE/2; j++) { + if (dv_vlc_map[i][j].size == 0) { + dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc | + (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size)); + dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size + + dv_vlc_map[0][j].size; + } + dv_vlc_map[i][((uint16_t)(-j))&0x1ff].vlc = + dv_vlc_map[i][j].vlc | 1; + dv_vlc_map[i][((uint16_t)(-j))&0x1ff].size = + dv_vlc_map[i][j].size; + } #endif - } + } } /* Generic DSP setup */ @@ -221,23 +238,21 @@ static int dvvideo_init(AVCodecContext *avctx) /* 248DCT setup */ s->fdct[1] = dsp.fdct248; - s->idct_put[1] = simple_idct248_put; // FIXME: need to add it to DSP - memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64); + s->idct_put[1] = ff_simple_idct248_put; // FIXME: need to add it to DSP + if(avctx->lowres){ + for (i=0; i<64; i++){ + int j= ff_zigzag248_direct[i]; + s->dv_zigzag[1][i] = dsp.idct_permutation[(j&7) + (j&8)*4 + (j&48)/2]; + } + }else + memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64); /* XXX: do it only for constant case */ dv_build_unquantize_tables(s, dsp.idct_permutation); - /* FIXME: I really don't think this should be here */ - if (dv_codec_profile(avctx)) - avctx->pix_fmt = dv_codec_profile(avctx)->pix_fmt; avctx->coded_frame = &s->picture; - - return 0; -} + s->avctx= avctx; -static int dvvideo_end(AVCodecContext *avctx) -{ - avcodec_default_free_buffers(avctx); return 0; } @@ -245,38 +260,25 @@ static int dvvideo_end(AVCodecContext *avctx) // #define printf(...) av_log(NULL, AV_LOG_ERROR, __VA_ARGS__) typedef struct BlockInfo { - const uint8_t *shift_table; + const uint32_t *factor_table; const uint8_t *scan_table; uint8_t pos; /* position in block */ - uint8_t dct_mode; + void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block); uint8_t partial_bit_count; uint16_t partial_bit_buffer; int shift_offset; } BlockInfo; -/* block size in bits */ -static const uint16_t block_sizes[6] = { - 112, 112, 112, 112, 80, 80 -}; /* bit budget for AC only in 5 MBs */ static const int vs_total_ac_bits = (100 * 4 + 68*2) * 5; /* see dv_88_areas and dv_248_areas for details */ -static const int mb_area_start[5] = { 1, 6, 21, 43, 64 }; - -#ifndef ALT_BITSTREAM_READER -#warning only works with ALT_BITSTREAM_READER -#endif +static const int mb_area_start[5] = { 1, 6, 21, 43, 64 }; static inline int get_bits_left(GetBitContext *s) { return s->size_in_bits - get_bits_count(s); } -static inline int get_bits_size(GetBitContext *s) -{ - return s->size_in_bits; -} - static inline int put_bits_left(PutBitContext* s) { return (s->buf_end - s->buf) * 8 - put_bits_count(s); @@ -285,22 +287,22 @@ static inline int put_bits_left(PutBitContext* s) /* decode ac coefs */ static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block) { - int last_index = get_bits_size(gb); + int last_index = gb->size_in_bits; const uint8_t *scan_table = mb->scan_table; - const uint8_t *shift_table = mb->shift_table; + const uint32_t *factor_table = mb->factor_table; int pos = mb->pos; int partial_bit_count = mb->partial_bit_count; - int level, pos1, run, vlc_len, index; - + int level, run, vlc_len, index; + OPEN_READER(re, gb); UPDATE_CACHE(re, gb); - + /* if we must parse a partial vlc, we do it here */ if (partial_bit_count > 0) { re_cache = ((unsigned)re_cache >> partial_bit_count) | - (mb->partial_bit_buffer << (sizeof(re_cache)*8 - partial_bit_count)); - re_index -= partial_bit_count; - mb->partial_bit_count = 0; + (mb->partial_bit_buffer << (sizeof(re_cache)*8 - partial_bit_count)); + re_index -= partial_bit_count; + mb->partial_bit_count = 0; } /* get the AC coefficients until last_index is reached */ @@ -310,35 +312,33 @@ static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block) #endif /* our own optimized GET_RL_VLC */ index = NEG_USR32(re_cache, TEX_VLC_BITS); - vlc_len = dv_rl_vlc[index].len; + vlc_len = dv_rl_vlc[index].len; if (vlc_len < 0) { index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) + dv_rl_vlc[index].level; vlc_len = TEX_VLC_BITS - vlc_len; } level = dv_rl_vlc[index].level; - run = dv_rl_vlc[index].run; - - /* gotta check if we're still within gb boundaries */ - if (re_index + vlc_len > last_index) { - /* should be < 16 bits otherwise a codeword could have been parsed */ - mb->partial_bit_count = last_index - re_index; - mb->partial_bit_buffer = NEG_USR32(re_cache, mb->partial_bit_count); - re_index = last_index; - break; - } - re_index += vlc_len; + run = dv_rl_vlc[index].run; + + /* gotta check if we're still within gb boundaries */ + if (re_index + vlc_len > last_index) { + /* should be < 16 bits otherwise a codeword could have been parsed */ + mb->partial_bit_count = last_index - re_index; + mb->partial_bit_buffer = NEG_USR32(re_cache, mb->partial_bit_count); + re_index = last_index; + break; + } + re_index += vlc_len; #ifdef VLC_DEBUG - printf("run=%d level=%d\n", run, level); + printf("run=%d level=%d\n", run, level); #endif - pos += run; - if (pos >= 64) - break; - - if (level) { - pos1 = scan_table[pos]; - block[pos1] = level << shift_table[pos1]; - } + pos += run; + if (pos >= 64) + break; + + level = (level*factor_table[pos] + (1 << (dv_iweight_bits-1))) >> dv_iweight_bits; + block[scan_table[pos]] = level; UPDATE_CACHE(re, gb); } @@ -349,9 +349,9 @@ static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block) static inline void bit_copy(PutBitContext *pb, GetBitContext *gb) { int bits_left = get_bits_left(gb); - while (bits_left >= 16) { - put_bits(pb, 16, get_bits(gb, 16)); - bits_left -= 16; + while (bits_left >= MIN_CACHE_BITS) { + put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS)); + bits_left -= MIN_CACHE_BITS; } if (bits_left > 0) { put_bits(pb, bits_left, get_bits(gb, bits_left)); @@ -359,24 +359,29 @@ static inline void bit_copy(PutBitContext *pb, GetBitContext *gb) } /* mb_x and mb_y are in units of 8 pixels */ -static inline void dv_decode_video_segment(DVVideoContext *s, - uint8_t *buf_ptr1, +static inline void dv_decode_video_segment(DVVideoContext *s, + const uint8_t *buf_ptr1, const uint16_t *mb_pos_ptr) { int quant, dc, dct_mode, class1, j; int mb_index, mb_x, mb_y, v, last_index; + int y_stride, i; DCTELEM *block, *block1; int c_offset; uint8_t *y_ptr; - void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block); - uint8_t *buf_ptr; + const uint8_t *buf_ptr; PutBitContext pb, vs_pb; GetBitContext gb; - BlockInfo mb_data[5 * 6], *mb, *mb1; - DCTELEM sblock[5*6][64] __align8; - uint8_t mb_bit_buffer[80 + 4]; /* allow some slack */ - uint8_t vs_bit_buffer[5 * 80 + 4]; /* allow some slack */ - + BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1; + DECLARE_ALIGNED_16(DCTELEM, sblock[5*DV_MAX_BPM][64]); + DECLARE_ALIGNED_8(uint8_t, mb_bit_buffer[80 + 4]); /* allow some slack */ + DECLARE_ALIGNED_8(uint8_t, vs_bit_buffer[5 * 80 + 4]); /* allow some slack */ + const int log2_blocksize= 3-s->avctx->lowres; + int is_field_mode[5]; + + assert((((int)mb_bit_buffer)&7)==0); + assert((((int)vs_bit_buffer)&7)==0); + memset(sblock, 0, sizeof(sblock)); /* pass 1 : read DC and AC coefficients in blocks */ @@ -384,26 +389,33 @@ static inline void dv_decode_video_segment(DVVideoContext *s, block1 = &sblock[0][0]; mb1 = mb_data; init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80); - for(mb_index = 0; mb_index < 5; mb_index++, mb1 += 6, block1 += 6 * 64) { + for(mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) { /* skip header */ quant = buf_ptr[3] & 0x0f; buf_ptr += 4; init_put_bits(&pb, mb_bit_buffer, 80); mb = mb1; block = block1; - for(j = 0;j < 6; j++) { - last_index = block_sizes[j]; - init_get_bits(&gb, buf_ptr, last_index); - + is_field_mode[mb_index] = 0; + for(j = 0;j < s->sys->bpm; j++) { + last_index = s->sys->block_sizes[j]; + init_get_bits(&gb, buf_ptr, last_index); + /* get the dc */ - dc = get_bits(&gb, 9); - dc = (dc << (32 - 9)) >> (32 - 9); + dc = get_sbits(&gb, 9); dct_mode = get_bits1(&gb); - mb->dct_mode = dct_mode; - mb->scan_table = s->dv_zigzag[dct_mode]; class1 = get_bits(&gb, 2); - mb->shift_table = s->dv_idct_shift[class1 == 3][dct_mode] - [quant + dv_quant_offset[class1]]; + if (DV_PROFILE_IS_HD(s->sys)) { + mb->idct_put = s->idct_put[0]; + mb->scan_table = s->dv_zigzag[0]; + mb->factor_table = s->dv100_idct_factor[((s->sys->height == 720)<<1)&(j < 4)][class1][quant]; + is_field_mode[mb_index] |= !j && dct_mode; + } else { + mb->idct_put = s->idct_put[dct_mode && log2_blocksize==3]; + mb->scan_table = s->dv_zigzag[dct_mode]; + mb->factor_table = s->dv_idct_factor[class1 == 3][dct_mode] + [quant + dv_quant_offset[class1]]; + } dc = dc << 2; /* convert to unsigned because 128 is not added in the standard IDCT */ @@ -422,11 +434,11 @@ static inline void dv_decode_video_segment(DVVideoContext *s, block is finished */ if (mb->pos >= 64) bit_copy(&pb, &gb); - + block += 64; mb++; } - + /* pass 2 : we can do it just after */ #ifdef VLC_DEBUG printf("***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index); @@ -434,8 +446,8 @@ static inline void dv_decode_video_segment(DVVideoContext *s, block = block1; mb = mb1; init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb)); - flush_put_bits(&pb); - for(j = 0;j < 6; j++, block += 64, mb++) { + flush_put_bits(&pb); + for(j = 0;j < s->sys->bpm; j++, block += 64, mb++) { if (mb->pos < 64 && get_bits_left(&gb) > 0) { dv_decode_ac(&gb, mb, block); /* if still not finished, no need to parse other blocks */ @@ -445,8 +457,8 @@ static inline void dv_decode_video_segment(DVVideoContext *s, } /* all blocks are finished, so the extra bytes can be used at the video segment level */ - if (j >= 6) - bit_copy(&vs_pb, &gb); + if (j >= s->sys->bpm) + bit_copy(&vs_pb, &gb); } /* we need a pass other the whole video segment */ @@ -458,20 +470,20 @@ static inline void dv_decode_video_segment(DVVideoContext *s, init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb)); flush_put_bits(&vs_pb); for(mb_index = 0; mb_index < 5; mb_index++) { - for(j = 0;j < 6; j++) { + for(j = 0;j < s->sys->bpm; j++) { if (mb->pos < 64) { #ifdef VLC_DEBUG printf("start %d:%d\n", mb_index, j); #endif dv_decode_ac(&gb, mb, block); } - if (mb->pos >= 64 && mb->pos < 127) - av_log(NULL, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos); + if (mb->pos >= 64 && mb->pos < 127) + av_log(NULL, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos); block += 64; mb++; } } - + /* compute idct and place blocks */ block = &sblock[0][0]; mb = mb_data; @@ -479,227 +491,260 @@ static inline void dv_decode_video_segment(DVVideoContext *s, v = *mb_pos_ptr++; mb_x = v & 0xff; mb_y = v >> 8; - y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8); - if (s->sys->pix_fmt == PIX_FMT_YUV411P) - c_offset = (mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8); - else - c_offset = ((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8); - for(j = 0;j < 6; j++) { - idct_put = s->idct_put[mb->dct_mode]; - if (j < 4) { - if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) { - /* NOTE: at end of line, the macroblock is handled as 420 */ - idct_put(y_ptr + (j * 8), s->picture.linesize[0], block); - } else { - idct_put(y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]), - s->picture.linesize[0], block); - } + /* We work with 720p frames split in half. The odd half-frame (chan==2,3) is displaced :-( */ + if (s->sys->height == 720 && ((s->buf[1]>>2)&0x3) == 0) { + mb_y -= (mb_y>17)?18:-72; /* shifting the Y coordinate down by 72/2 macro blocks */ + } + + /* idct_put'ting luminance */ + if ((s->sys->pix_fmt == PIX_FMT_YUV420P) || + (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) || + (s->sys->height >= 720 && mb_y != 134)) { + y_stride = (s->picture.linesize[0]<<((!is_field_mode[mb_index])*log2_blocksize)) - (2<picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x)<sys->pix_fmt == PIX_FMT_YUV422P && s->sys->width == 720 && i) + y_ptr -= (1<idct_put(y_ptr, s->picture.linesize[0]<>(s->sys->pix_fmt == PIX_FMT_YUV420P)) * s->picture.linesize[1] + + (mb_x>>((s->sys->pix_fmt == PIX_FMT_YUV411P)?2:1)))<picture.data[j] + c_offset; + if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) { + uint64_t aligned_pixels[64/8]; + uint8_t *pixels = (uint8_t*)aligned_pixels; + uint8_t *c_ptr1, *ptr1; + int x, y; + mb->idct_put(pixels, 8, block); + for(y = 0; y < (1<picture.linesize[j], pixels += 8) { + ptr1= pixels + (1<<(log2_blocksize-1)); + c_ptr1 = c_ptr + (s->picture.linesize[j]<sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) { - uint64_t aligned_pixels[64/8]; - uint8_t *pixels= (uint8_t*)aligned_pixels; - uint8_t *c_ptr, *c_ptr1, *ptr; - int y, linesize; - /* NOTE: at end of line, the macroblock is handled as 420 */ - idct_put(pixels, 8, block); - linesize = s->picture.linesize[6 - j]; - c_ptr = s->picture.data[6 - j] + c_offset; - ptr = pixels; - for(y = 0;y < 8; y++) { - /* convert to 411P */ - c_ptr1 = c_ptr + 8*linesize; - c_ptr[0]= ptr[0]; c_ptr1[0]= ptr[4]; - c_ptr[1]= ptr[1]; c_ptr1[1]= ptr[5]; - c_ptr[2]= ptr[2]; c_ptr1[2]= ptr[6]; - c_ptr[3]= ptr[3]; c_ptr1[3]= ptr[7]; - c_ptr += linesize; - ptr += 8; - } - } else { - /* don't ask me why they inverted Cb and Cr ! */ - idct_put(s->picture.data[6 - j] + c_offset, - s->picture.linesize[6 - j], block); - } + y_stride = (mb_y == 134) ? (1<picture.linesize[j]<<((!is_field_mode[mb_index])*log2_blocksize); + for (i=0; i<(1<<(s->sys->bpm==8)); i++, block += 64, mb++, c_ptr += y_stride) + mb->idct_put(c_ptr, s->picture.linesize[j]<> 8; - int level = (l ^ sign) - sign; int size; - - sign = (sign & 1); - if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) { *vlc = dv_vlc_map[run][level].vlc | sign; - size = dv_vlc_map[run][level].size; + size = dv_vlc_map[run][level].size; } - else { + else { if (level < DV_VLC_MAP_LEV_SIZE) { - *vlc = dv_vlc_map[0][level].vlc | sign; - size = dv_vlc_map[0][level].size; - } else { + *vlc = dv_vlc_map[0][level].vlc | sign; + size = dv_vlc_map[0][level].size; + } else { *vlc = 0xfe00 | (level << 1) | sign; - size = 16; - } - if (run) { - *vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc : - (0x1f80 | (run - 1))) << size; - size += (run < 16) ? dv_vlc_map[run-1][0].size : 13; - } + size = 16; + } + if (run) { + *vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc : + (0x1f80 | (run - 1))) << size; + size += (run < 16) ? dv_vlc_map[run-1][0].size : 13; + } } - + return size; } -static always_inline int dv_rl2vlc_size(int run, int l) +static av_always_inline int dv_rl2vlc_size(int run, int level) { - int level = (l ^ (l >> 8)) - (l >> 8); int size; - + if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) { - size = dv_vlc_map[run][level].size; + size = dv_vlc_map[run][level].size; } - else { - size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16; - if (run) { - size += (run < 16) ? dv_vlc_map[run-1][0].size : 13; - } + else { + size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16; + if (run) { + size += (run < 16) ? dv_vlc_map[run-1][0].size : 13; + } } return size; } #else -static always_inline int dv_rl2vlc(int run, int l, uint32_t* vlc) +static av_always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t* vlc) { - *vlc = dv_vlc_map[run][((uint16_t)l)&0x1ff].vlc; - return dv_vlc_map[run][((uint16_t)l)&0x1ff].size; + *vlc = dv_vlc_map[run][l].vlc | sign; + return dv_vlc_map[run][l].size; } -static always_inline int dv_rl2vlc_size(int run, int l) +static av_always_inline int dv_rl2vlc_size(int run, int l) { - return dv_vlc_map[run][((uint16_t)l)&0x1ff].size; + return dv_vlc_map[run][l].size; } #endif typedef struct EncBlockInfo { int area_q[4]; int bit_size[4]; - int prev_run[4]; + int prev[5]; int cur_ac; int cno; int dct_mode; - DCTELEM *mb; + DCTELEM mb[64]; + uint8_t next[64]; + uint8_t sign[64]; uint8_t partial_bit_count; uint32_t partial_bit_buffer; /* we can't use uint16_t here */ } EncBlockInfo; -static always_inline void dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool, - int pb_size) +static av_always_inline PutBitContext* dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool, + PutBitContext* pb_end) { - int run; + int prev; int bits_left; PutBitContext* pb = pb_pool; int size = bi->partial_bit_count; uint32_t vlc = bi->partial_bit_buffer; - + bi->partial_bit_count = bi->partial_bit_buffer = 0; -vlc_loop: + for(;;){ /* Find suitable storage space */ for (; size > (bits_left = put_bits_left(pb)); pb++) { if (bits_left) { size -= bits_left; - put_bits(pb, bits_left, vlc >> size); - vlc = vlc & ((1<partial_bit_count = size; - bi->partial_bit_buffer = vlc; - return; - } - --pb_size; + put_bits(pb, bits_left, vlc >> size); + vlc = vlc & ((1<= pb_end) { + bi->partial_bit_count = size; + bi->partial_bit_buffer = vlc; + return pb; + } } - + /* Store VLC */ put_bits(pb, size, vlc); - + + if(bi->cur_ac>=64) + break; + /* Construct the next VLC */ - run = 0; - for (; bi->cur_ac < 64; bi->cur_ac++, run++) { - if (bi->mb[bi->cur_ac]) { - size = dv_rl2vlc(run, bi->mb[bi->cur_ac], &vlc); - bi->cur_ac++; - goto vlc_loop; - } - } - - if (bi->cur_ac == 64) { + prev= bi->cur_ac; + bi->cur_ac = bi->next[prev]; + if(bi->cur_ac < 64){ + size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac], bi->sign[bi->cur_ac], &vlc); + } else { size = 4; vlc = 6; /* End Of Block stamp */ - bi->cur_ac++; - goto vlc_loop; } + } + return pb; } -static always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi, - const uint8_t* zigzag_scan, int bias) +static av_always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi, + const uint8_t* zigzag_scan, const int *weight, int bias) { int i, area; - int run; - int classes[] = {12, 24, 36, 0xffff}; + /* We offer two different methods for class number assignment: the + method suggested in SMPTE 314M Table 22, and an improved + method. The SMPTE method is very conservative; it assigns class + 3 (i.e. severe quantization) to any block where the largest AC + component is greater than 36. ffmpeg's DV encoder tracks AC bit + consumption precisely, so there is no need to bias most blocks + towards strongly lossy compression. Instead, we assign class 2 + to most blocks, and use class 3 only when strictly necessary + (for blocks whose largest AC component exceeds 255). */ + +#if 0 /* SMPTE spec method */ + static const int classes[] = {12, 24, 36, 0xffff}; +#else /* improved ffmpeg method */ + static const int classes[] = {-1, -1, 255, 0xffff}; +#endif + int max=classes[0]; + int prev=0; + + bi->mb[0] = blk[0]; - run = 0; - bi->mb[0] = blk[0]; - bi->cno = 0; for (area = 0; area < 4; area++) { - bi->prev_run[area] = run; - bi->bit_size[area] = 0; + bi->prev[area] = prev; + bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :) for (i=mb_area_start[area]; imb[i] = (blk[zigzag_scan[i]] / 16); - while ((bi->mb[i] ^ (bi->mb[i] >> 8)) > classes[bi->cno]) - bi->cno++; - - if (bi->mb[i]) { - bi->bit_size[area] += dv_rl2vlc_size(run, bi->mb[i]); - run = 0; - } else - ++run; + int level = blk[zigzag_scan[i]]; + + if (level+15 > 30U) { + bi->sign[i] = (level>>31)&1; + /* weigh it and and shift down into range, adding for rounding */ + /* the extra division by a factor of 2^4 reverses the 8x expansion of the DCT + AND the 2x doubling of the weights */ + level = (FFABS(level) * weight[i] + (1<<(dv_weight_bits+3))) >> (dv_weight_bits+4); + bi->mb[i] = level; + if(level>max) max= level; + bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, level); + bi->next[prev]= i; + prev= i; + } } } - bi->bit_size[3] += 4; /* EOB marker */ + bi->next[prev]= i; + for(bi->cno = 0; max > classes[bi->cno]; bi->cno++); + bi->cno += bias; - - if (bi->cno >= 3) { /* FIXME: we have to recreate bit_size[], prev_run[] */ + + if (bi->cno >= 3) { bi->cno = 3; - for (i=1; i<64; i++) - bi->mb[i] /= 2; + prev=0; + i= bi->next[prev]; + for (area = 0; area < 4; area++) { + bi->prev[area] = prev; + bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :) + for (; inext[i]) { + bi->mb[i] >>=1; + + if (bi->mb[i]) { + bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]); + bi->next[prev]= i; + prev= i; + } + } + } + bi->next[prev]= i; } } +//FIXME replace this by dsputil #define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7)) -static always_inline int dv_guess_dct_mode(DCTELEM *blk) { +static av_always_inline int dv_guess_dct_mode(DCTELEM *blk) { DCTELEM *s; int score88 = 0; int score248 = 0; int i; - + /* Compute 8-8 score (small values give a better chance for 8-8 DCT) */ s = blk; for(i=0; i<7; i++) { - score88 += SC(0, 8) + SC(1, 9) + SC(2, 10) + SC(3, 11) + - SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15); + score88 += SC(0, 8) + SC(1, 9) + SC(2, 10) + SC(3, 11) + + SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15); s += 8; } /* Compute 2-4-8 score (small values give a better chance for 2-4-8 DCT) */ s = blk; for(i=0; i<6; i++) { score248 += SC(0, 16) + SC(1,17) + SC(2, 18) + SC(3, 19) + - SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23); + SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23); s += 8; } @@ -709,132 +754,187 @@ static always_inline int dv_guess_dct_mode(DCTELEM *blk) { static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos) { int size[5]; - int i, j, k, a, run; + int i, j, k, a, prev, a2; EncBlockInfo* b; - + + size[0] = size[1] = size[2] = size[3] = size[4] = 1<<24; do { b = blks; for (i=0; i<5; i++) { if (!qnos[i]) - continue; - - qnos[i]--; - size[i] = 0; + continue; + + qnos[i]--; + size[i] = 0; for (j=0; j<6; j++, b++) { - for (a=0; a<4; a++) { - if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) { - b->bit_size[a] = (a==3)?4:0; - b->area_q[a]++; - run = b->prev_run[a]; - for (k=mb_area_start[a]; kmb[k] /= 2; - if (b->mb[k]) { - b->bit_size[a] += dv_rl2vlc_size(run, b->mb[k]); - run = 0; - } else - ++run; - } - } - size[i] += b->bit_size[a]; - } - } + for (a=0; a<4; a++) { + if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) { + b->bit_size[a] = 1; // 4 areas 4 bits for EOB :) + b->area_q[a]++; + prev= b->prev[a]; + assert(b->next[prev] >= mb_area_start[a+1] || b->mb[prev]); + for (k= b->next[prev] ; knext[k]) { + b->mb[k] >>= 1; + if (b->mb[k]) { + b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]); + prev= k; + } else { + if(b->next[k] >= mb_area_start[a+1] && b->next[k]<64){ + for(a2=a+1; b->next[k] >= mb_area_start[a2+1]; a2++) + b->prev[a2] = prev; + assert(a2<4); + assert(b->mb[b->next[k]]); + b->bit_size[a2] += dv_rl2vlc_size(b->next[k] - prev - 1, b->mb[b->next[k]]) + -dv_rl2vlc_size(b->next[k] - k - 1, b->mb[b->next[k]]); + assert(b->prev[a2]==k && (a2+1 >= 4 || b->prev[a2+1]!=k)); + b->prev[a2] = prev; + } + b->next[prev] = b->next[k]; + } + } + b->prev[a+1]= prev; + } + size[i] += b->bit_size[a]; + } + } + if(vs_total_ac_bits >= size[0] + size[1] + size[2] + size[3] + size[4]) + return; } - } while ((vs_total_ac_bits < size[0] + size[1] + size[2] + size[3] + size[4]) && - (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4])); + } while (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4]); + + + for(a=2; a==2 || vs_total_ac_bits < size[0]; a+=a){ + b = blks; + size[0] = 5*6*4; //EOB + for (j=0; j<6*5; j++, b++) { + prev= b->prev[0]; + for (k= b->next[prev]; k<64; k= b->next[k]) { + if(b->mb[k] < a && b->mb[k] > -a){ + b->next[prev] = b->next[k]; + }else{ + size[0] += dv_rl2vlc_size(k - prev - 1, b->mb[k]); + prev= k; + } + } + } + } } -/* - * This is a very rough initial implementaion. The performance is - * horrible and the weighting is missing. But it's missing from the - * decoding step also -- so at least we're on the same page with decoder ;-) - */ -static inline void dv_encode_video_segment(DVVideoContext *s, - uint8_t *dif, +static inline void dv_encode_video_segment(DVVideoContext *s, + uint8_t *dif, const uint16_t *mb_pos_ptr) { int mb_index, i, j, v; - int mb_x, mb_y, c_offset, linesize; + int mb_x, mb_y, c_offset, linesize; uint8_t* y_ptr; uint8_t* data; uint8_t* ptr; int do_edge_wrap; - DCTELEM block[64] __align8; - DCTELEM sblock[5*6][64] __align8; + DECLARE_ALIGNED_16(DCTELEM, block[64]); EncBlockInfo enc_blks[5*6]; PutBitContext pbs[5*6]; - PutBitContext* pb; + PutBitContext* pb; EncBlockInfo* enc_blk; int vs_bit_size = 0; int qnos[5]; - + + assert((((int)block) & 15) == 0); + enc_blk = &enc_blks[0]; pb = &pbs[0]; for(mb_index = 0; mb_index < 5; mb_index++) { v = *mb_pos_ptr++; mb_x = v & 0xff; mb_y = v >> 8; - y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8); - c_offset = (s->sys->pix_fmt == PIX_FMT_YUV411P) ? - ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8)) : - (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8)); - do_edge_wrap = 0; - qnos[mb_index] = 15; /* No quantization */ + y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x)<<3); + c_offset = (((mb_y>>(s->sys->pix_fmt == PIX_FMT_YUV420P)) * s->picture.linesize[1] + + (mb_x>>((s->sys->pix_fmt == PIX_FMT_YUV411P)?2:1)))<<3); + do_edge_wrap = 0; + qnos[mb_index] = 15; /* No quantization */ ptr = dif + mb_index*80 + 4; for(j = 0;j < 6; j++) { - if (j < 4) { /* Four Y blocks */ - /* NOTE: at end of line, the macroblock is handled as 420 */ - if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) { - data = y_ptr + (j * 8); + int dummy = 0; + if (s->sys->pix_fmt == PIX_FMT_YUV422P) { /* 4:2:2 */ + if (j == 0 || j == 2) { + /* Y0 Y1 */ + data = y_ptr + ((j>>1) * 8); + linesize = s->picture.linesize[0]; + } else if (j > 3) { + /* Cr Cb */ + data = s->picture.data[6 - j] + c_offset; + linesize = s->picture.linesize[6 - j]; } else { - data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]); + /* j=1 and j=3 are "dummy" blocks, used for AC data only */ + data = 0; + linesize = 0; + dummy = 1; } - linesize = s->picture.linesize[0]; - } else { /* Cr and Cb blocks */ - /* don't ask Fabrice why they inverted Cb and Cr ! */ - data = s->picture.data[6 - j] + c_offset; - linesize = s->picture.linesize[6 - j]; - if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) - do_edge_wrap = 1; - } - - /* Everything is set up -- now just copy data -> DCT block */ - if (do_edge_wrap) { /* Edge wrap copy: 4x16 -> 8x8 */ - uint8_t* d; - DCTELEM *b = block; - for (i=0;i<8;i++) { - d = data + 8 * linesize; - b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3]; + } else { /* 4:1:1 or 4:2:0 */ + if (j < 4) { /* Four Y blocks */ + /* NOTE: at end of line, the macroblock is handled as 420 */ + if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) { + data = y_ptr + (j * 8); + } else { + data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]); + } + linesize = s->picture.linesize[0]; + } else { /* Cr and Cb blocks */ + /* don't ask Fabrice why they inverted Cb and Cr ! */ + data = s->picture.data[6 - j] + c_offset; + linesize = s->picture.linesize[6 - j]; + if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) + do_edge_wrap = 1; + } + } + + /* Everything is set up -- now just copy data -> DCT block */ + if (do_edge_wrap) { /* Edge wrap copy: 4x16 -> 8x8 */ + uint8_t* d; + DCTELEM *b = block; + for (i=0;i<8;i++) { + d = data + 8 * linesize; + b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3]; b[4] = d[0]; b[5] = d[1]; b[6] = d[2]; b[7] = d[3]; - data += linesize; - b += 8; - } - } else { /* Simple copy: 8x8 -> 8x8 */ - s->get_pixels(block, data, linesize); - } - - enc_blk->dct_mode = dv_guess_dct_mode(block); - enc_blk->mb = &sblock[mb_index*6+j][0]; - enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0; - enc_blk->partial_bit_count = 0; - enc_blk->partial_bit_buffer = 0; - enc_blk->cur_ac = 1; - - s->fdct[enc_blk->dct_mode](block); - - dv_set_class_number(block, enc_blk, - enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct, - j/4*(j%2)); - - init_put_bits(pb, ptr, block_sizes[j]/8); - put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024) >> 2)); - put_bits(pb, 1, enc_blk->dct_mode); - put_bits(pb, 2, enc_blk->cno); - - vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] + - enc_blk->bit_size[2] + enc_blk->bit_size[3]; - ++enc_blk; - ++pb; - ptr += block_sizes[j]/8; + data += linesize; + b += 8; + } + } else { /* Simple copy: 8x8 -> 8x8 */ + if (!dummy) + s->get_pixels(block, data, linesize); + } + + if(s->avctx->flags & CODEC_FLAG_INTERLACED_DCT) + enc_blk->dct_mode = dv_guess_dct_mode(block); + else + enc_blk->dct_mode = 0; + enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0; + enc_blk->partial_bit_count = 0; + enc_blk->partial_bit_buffer = 0; + enc_blk->cur_ac = 0; + + if (dummy) { + /* We rely on the fact that encoding all zeros leads to an immediate EOB, + which is precisely what the spec calls for in the "dummy" blocks. */ + memset(block, 0, sizeof(block)); + } else { + s->fdct[enc_blk->dct_mode](block); + } + + dv_set_class_number(block, enc_blk, + enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct, + enc_blk->dct_mode ? dv_weight_248 : dv_weight_88, + j/4); + + init_put_bits(pb, ptr, s->sys->block_sizes[j]/8); + put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024 + 2) >> 2)); + put_bits(pb, 1, enc_blk->dct_mode); + put_bits(pb, 2, enc_blk->cno); + + vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] + + enc_blk->bit_size[2] + enc_blk->bit_size[3]; + ++enc_blk; + ++pb; + ptr += s->sys->block_sizes[j]/8; } } @@ -847,18 +947,24 @@ static inline void dv_encode_video_segment(DVVideoContext *s, /* First pass over individual cells only */ for (j=0; j<5*6; j++) - dv_encode_ac(&enc_blks[j], &pbs[j], 1); + dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j+1]); /* Second pass over each MB space */ - for (j=0; j<5*6; j++) { - if (enc_blks[j].cur_ac < 65 || enc_blks[j].partial_bit_count) - dv_encode_ac(&enc_blks[j], &pbs[(j/6)*6], 6); + for (j=0; j<5*6; j+=6) { + pb= &pbs[j]; + for (i=0; i<6; i++) { + if (enc_blks[i+j].partial_bit_count) + pb=dv_encode_ac(&enc_blks[i+j], pb, &pbs[j+6]); + } } /* Third and final pass over the whole vides segment space */ + pb= &pbs[0]; for (j=0; j<5*6; j++) { - if (enc_blks[j].cur_ac < 65 || enc_blks[j].partial_bit_count) - dv_encode_ac(&enc_blks[j], &pbs[0], 6*5); + if (enc_blks[j].partial_bit_count) + pb=dv_encode_ac(&enc_blks[j], pb, &pbs[6*5]); + if (enc_blks[j].partial_bit_count) + av_log(NULL, AV_LOG_ERROR, "ac bitstream overflow\n"); } for (j=0; j<5*6; j++) @@ -869,44 +975,72 @@ static int dv_decode_mt(AVCodecContext *avctx, void* sl) { DVVideoContext *s = avctx->priv_data; int slice = (size_t)sl; - dv_decode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80], - &s->sys->video_place[slice*5]); + + /* which DIF channel is this? */ + int chan = slice / (s->sys->difseg_size * 27); + + /* slice within the DIF channel */ + int chan_slice = slice % (s->sys->difseg_size * 27); + + /* byte offset of this channel's data */ + int chan_offset = chan * s->sys->difseg_size * 150 * 80; + + /* DIF sequence */ + int seq = chan_slice / 27; + + /* in 1080i50 and 720p50 some seq are unused */ + if ((DV_PROFILE_IS_1080i50(s->sys) && chan != 0 && seq == 11) || + (DV_PROFILE_IS_720p50(s->sys) && seq > 9)) + return 0; + + dv_decode_video_segment(s, &s->buf[(seq*6+(chan_slice/3)+chan_slice*5+7)*80 + chan_offset], + &s->sys->video_place[slice*5]); return 0; } +#ifdef CONFIG_DVVIDEO_ENCODER static int dv_encode_mt(AVCodecContext *avctx, void* sl) { DVVideoContext *s = avctx->priv_data; int slice = (size_t)sl; - dv_encode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80], - &s->sys->video_place[slice*5]); + + /* which DIF channel is this? */ + int chan = slice / (s->sys->difseg_size * 27); + + /* slice within the DIF channel */ + int chan_slice = slice % (s->sys->difseg_size * 27); + + /* byte offset of this channel's data */ + int chan_offset = chan * s->sys->difseg_size * 150 * 80; + + dv_encode_video_segment(s, &s->buf[((chan_slice/27)*6+(chan_slice/3)+chan_slice*5+7)*80 + chan_offset], + &s->sys->video_place[slice*5]); return 0; } +#endif +#ifdef CONFIG_DECODERS /* NOTE: exactly one frame must be given (120000 bytes for NTSC, - 144000 bytes for PAL) */ -static int dvvideo_decode_frame(AVCodecContext *avctx, + 144000 bytes for PAL - or twice those for 50Mbps) */ +static int dvvideo_decode_frame(AVCodecContext *avctx, void *data, int *data_size, - uint8_t *buf, int buf_size) + const uint8_t *buf, int buf_size) { DVVideoContext *s = avctx->priv_data; - - *data_size=0; - /* special case for last picture */ - if(buf_size==0) - return 0; - + s->sys = dv_frame_profile(buf); if (!s->sys || buf_size < s->sys->frame_size) return -1; /* NOTE: we only accept several full frames */ if(s->picture.data[0]) avctx->release_buffer(avctx, &s->picture); - + s->picture.reference = 0; + s->picture.key_frame = 1; + s->picture.pict_type = FF_I_TYPE; avctx->pix_fmt = s->sys->pix_fmt; - avctx->width = s->sys->width; - avctx->height = s->sys->height; + avctx->time_base = (AVRational){s->sys->frame_rate_base, s->sys->frame_rate}; + avcodec_set_dimensions(avctx, s->sys->width, s->sys->height); if(avctx->get_buffer(avctx, &s->picture) < 0) { av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n"); return -1; @@ -915,38 +1049,183 @@ static int dvvideo_decode_frame(AVCodecContext *avctx, s->picture.top_field_first = 0; s->buf = buf; - avctx->execute(avctx, dv_decode_mt, (void**)&dv_anchor[0], NULL, - s->sys->difseg_size * 27); - + avctx->execute(avctx, dv_decode_mt, (void**)&dv_anchor[0], NULL, + s->sys->n_difchan * s->sys->difseg_size * 27); + emms_c(); /* return image */ *data_size = sizeof(AVFrame); *(AVFrame*)data= s->picture; - + return s->sys->frame_size; } +#endif + + +static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c, uint8_t* buf) +{ + /* + * Here's what SMPTE314M says about these two: + * (page 6) APTn, AP1n, AP2n, AP3n: These data shall be identical + * as track application IDs (APTn = 001, AP1n = + * 001, AP2n = 001, AP3n = 001), if the source signal + * comes from a digital VCR. If the signal source is + * unknown, all bits for these data shall be set to 1. + * (page 12) STYPE: STYPE defines a signal type of video signal + * 00000b = 4:1:1 compression + * 00100b = 4:2:2 compression + * XXXXXX = Reserved + * Now, I've got two problems with these statements: + * 1. it looks like APT == 111b should be a safe bet, but it isn't. + * It seems that for PAL as defined in IEC 61834 we have to set + * APT to 000 and for SMPTE314M to 001. + * 2. It is not at all clear what STYPE is used for 4:2:0 PAL + * compression scheme (if any). + */ + int apt = (c->sys->pix_fmt == PIX_FMT_YUV420P ? 0 : 1); + int stype = (c->sys->pix_fmt == PIX_FMT_YUV422P ? 4 : 0); + + uint8_t aspect = 0; + if((int)(av_q2d(c->avctx->sample_aspect_ratio) * c->avctx->width / c->avctx->height * 10) == 17) /* 16:9 */ + aspect = 0x02; + + buf[0] = (uint8_t)pack_id; + switch (pack_id) { + case dv_header525: /* I can't imagine why these two weren't defined as real */ + case dv_header625: /* packs in SMPTE314M -- they definitely look like ones */ + buf[1] = 0xf8 | /* reserved -- always 1 */ + (apt & 0x07); /* APT: Track application ID */ + buf[2] = (0 << 7) | /* TF1: audio data is 0 - valid; 1 - invalid */ + (0x0f << 3) | /* reserved -- always 1 */ + (apt & 0x07); /* AP1: Audio application ID */ + buf[3] = (0 << 7) | /* TF2: video data is 0 - valid; 1 - invalid */ + (0x0f << 3) | /* reserved -- always 1 */ + (apt & 0x07); /* AP2: Video application ID */ + buf[4] = (0 << 7) | /* TF3: subcode(SSYB) is 0 - valid; 1 - invalid */ + (0x0f << 3) | /* reserved -- always 1 */ + (apt & 0x07); /* AP3: Subcode application ID */ + break; + case dv_video_source: + buf[1] = 0xff; /* reserved -- always 1 */ + buf[2] = (1 << 7) | /* B/W: 0 - b/w, 1 - color */ + (1 << 6) | /* following CLF is valid - 0, invalid - 1 */ + (3 << 4) | /* CLF: color frames id (see ITU-R BT.470-4) */ + 0xf; /* reserved -- always 1 */ + buf[3] = (3 << 6) | /* reserved -- always 1 */ + (c->sys->dsf << 5) | /* system: 60fields/50fields */ + stype; /* signal type video compression */ + buf[4] = 0xff; /* VISC: 0xff -- no information */ + break; + case dv_video_control: + buf[1] = (0 << 6) | /* Copy generation management (CGMS) 0 -- free */ + 0x3f; /* reserved -- always 1 */ + buf[2] = 0xc8 | /* reserved -- always b11001xxx */ + aspect; + buf[3] = (1 << 7) | /* Frame/field flag 1 -- frame, 0 -- field */ + (1 << 6) | /* First/second field flag 0 -- field 2, 1 -- field 1 */ + (1 << 5) | /* Frame change flag 0 -- same picture as before, 1 -- different */ + (1 << 4) | /* 1 - interlaced, 0 - noninterlaced */ + 0xc; /* reserved -- always b1100 */ + buf[4] = 0xff; /* reserved -- always 1 */ + break; + default: + buf[1] = buf[2] = buf[3] = buf[4] = 0xff; + } + return 5; +} + +#ifdef CONFIG_DVVIDEO_ENCODER +static void dv_format_frame(DVVideoContext* c, uint8_t* buf) +{ + int chan, i, j, k; + + for (chan = 0; chan < c->sys->n_difchan; chan++) { + for (i = 0; i < c->sys->difseg_size; i++) { + memset(buf, 0xff, 80 * 6); /* First 6 DIF blocks are for control data */ + + /* DV header: 1DIF */ + buf += dv_write_dif_id(dv_sect_header, chan, i, 0, buf); + buf += dv_write_pack((c->sys->dsf ? dv_header625 : dv_header525), c, buf); + buf += 72; /* unused bytes */ + + /* DV subcode: 2DIFs */ + for (j = 0; j < 2; j++) { + buf += dv_write_dif_id(dv_sect_subcode, chan, i, j, buf); + for (k = 0; k < 6; k++) + buf += dv_write_ssyb_id(k, (i < c->sys->difseg_size/2), buf) + 5; + buf += 29; /* unused bytes */ + } + + /* DV VAUX: 3DIFS */ + for (j = 0; j < 3; j++) { + buf += dv_write_dif_id(dv_sect_vaux, chan, i, j, buf); + buf += dv_write_pack(dv_video_source, c, buf); + buf += dv_write_pack(dv_video_control, c, buf); + buf += 7*5; + buf += dv_write_pack(dv_video_source, c, buf); + buf += dv_write_pack(dv_video_control, c, buf); + buf += 4*5 + 2; /* unused bytes */ + } + + /* DV Audio/Video: 135 Video DIFs + 9 Audio DIFs */ + for (j = 0; j < 135; j++) { + if (j%15 == 0) { + memset(buf, 0xff, 80); + buf += dv_write_dif_id(dv_sect_audio, chan, i, j/15, buf); + buf += 77; /* audio control & shuffled PCM audio */ + } + buf += dv_write_dif_id(dv_sect_video, chan, i, j, buf); + buf += 77; /* 1 video macro block: 1 bytes control + 4 * 14 bytes Y 8x8 data + 10 bytes Cr 8x8 data + 10 bytes Cb 8x8 data */ + } + } + } +} + -static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size, +static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size, void *data) { DVVideoContext *s = c->priv_data; s->sys = dv_codec_profile(c); if (!s->sys) - return -1; - + return -1; + if(buf_size < s->sys->frame_size) + return -1; + c->pix_fmt = s->sys->pix_fmt; s->picture = *((AVFrame *)data); + s->picture.key_frame = 1; + s->picture.pict_type = FF_I_TYPE; s->buf = buf; - c->execute(c, dv_encode_mt, (void**)&dv_anchor[0], NULL, - s->sys->difseg_size * 27); + c->execute(c, dv_encode_mt, (void**)&dv_anchor[0], NULL, + s->sys->n_difchan * s->sys->difseg_size * 27); emms_c(); + + dv_format_frame(s, buf); + return s->sys->frame_size; } +#endif + +static int dvvideo_close(AVCodecContext *c) +{ + DVVideoContext *s = c->priv_data; + if(s->picture.data[0]) + c->release_buffer(c, &s->picture); + + return 0; +} + + +#ifdef CONFIG_DVVIDEO_ENCODER AVCodec dvvideo_encoder = { "dvvideo", CODEC_TYPE_VIDEO, @@ -954,12 +1233,12 @@ AVCodec dvvideo_encoder = { sizeof(DVVideoContext), dvvideo_init, dvvideo_encode_frame, - dvvideo_end, - NULL, - CODEC_CAP_DR1, - NULL + .pix_fmts = (enum PixelFormat[]) {PIX_FMT_YUV411P, PIX_FMT_YUV422P, PIX_FMT_YUV420P, PIX_FMT_NONE}, + .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"), }; +#endif // CONFIG_DVVIDEO_ENCODER +#ifdef CONFIG_DVVIDEO_DECODER AVCodec dvvideo_decoder = { "dvvideo", CODEC_TYPE_VIDEO, @@ -967,8 +1246,10 @@ AVCodec dvvideo_decoder = { sizeof(DVVideoContext), dvvideo_init, NULL, - dvvideo_end, + dvvideo_close, dvvideo_decode_frame, CODEC_CAP_DR1, - NULL + NULL, + .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"), }; +#endif