X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fdv.c;h=5253b7f88e2ce3ebf005e3dcdcf74e79c568ecbd;hb=f099d3d1d5466bd63f4ab36270d169ff9ea613b8;hp=47f6afcf4835821509d239d2493f656c0a6c558f;hpb=b45ba151cf36d5bdaff4778b631280eeef232eb9;p=ffmpeg diff --git a/libavcodec/dv.c b/libavcodec/dv.c index 47f6afcf483..5253b7f88e2 100644 --- a/libavcodec/dv.c +++ b/libavcodec/dv.c @@ -16,99 +16,62 @@ * Many thanks to Dan Dennedy for providing wealth * of DV technical info. * - * This file is part of FFmpeg. + * This file is part of Libav. * - * FFmpeg is free software; you can redistribute it and/or + * Libav 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 2.1 of the License, or (at your option) any later version. * - * FFmpeg is distributed in the hope that it will be useful, + * Libav 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 FFmpeg; if not, write to the Free Software + * License along with Libav; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ /** - * @file dv.c + * @file * DV codec. */ -#define ALT_BITSTREAM_READER + +#include "libavutil/internal.h" +#include "libavutil/pixdesc.h" #include "avcodec.h" -#include "dsputil.h" -#include "bitstream.h" +#include "get_bits.h" +#include "internal.h" +#include "put_bits.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]; - - 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; - -#define TEX_VLC_BITS 9 - -#if CONFIG_SMALL -#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 //FIXME sign was removed so this should be /2 but needs check -#endif +#include "dv_tablegen.h" /* XXX: also include quantization */ -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_RUN_SIZE][DV_VLC_MAP_LEV_SIZE]; - -static inline int dv_work_pool_size(const DVprofile *d) -{ - int size = d->n_difchan*d->difseg_size*27; - if (DV_PROFILE_IS_1080i50(d)) - size -= 3*27; - if (DV_PROFILE_IS_720p50(d)) - size -= 4*27; - return size; -} +RL_VLC_ELEM ff_dv_rl_vlc[1184]; static inline void dv_calc_mb_coordinates(const DVprofile *d, int chan, int seq, int slot, uint16_t *tbl) { - const static uint8_t off[] = { 2, 6, 8, 0, 4 }; - const static uint8_t shuf1[] = { 36, 18, 54, 0, 72 }; - const static uint8_t shuf2[] = { 24, 12, 36, 0, 48 }; - const static uint8_t shuf3[] = { 18, 9, 27, 0, 36 }; + static const uint8_t off[] = { 2, 6, 8, 0, 4 }; + static const uint8_t shuf1[] = { 36, 18, 54, 0, 72 }; + static const uint8_t shuf2[] = { 24, 12, 36, 0, 48 }; + static const uint8_t shuf3[] = { 18, 9, 27, 0, 36 }; - const static uint8_t l_start[] = {0, 4, 9, 13, 18, 22, 27, 31, 36, 40}; - const static uint8_t l_start_shuffled[] = { 9, 4, 13, 0, 18 }; + static const uint8_t l_start[] = {0, 4, 9, 13, 18, 22, 27, 31, 36, 40}; + static const uint8_t l_start_shuffled[] = { 9, 4, 13, 0, 18 }; - const static uint8_t serpent1[] = {0, 1, 2, 2, 1, 0, + static const uint8_t serpent1[] = {0, 1, 2, 2, 1, 0, 0, 1, 2, 2, 1, 0, 0, 1, 2, 2, 1, 0, 0, 1, 2, 2, 1, 0, 0, 1, 2}; - const static uint8_t serpent2[] = {0, 1, 2, 3, 4, 5, 5, 4, 3, 2, 1, 0, + static const uint8_t serpent2[] = {0, 1, 2, 3, 4, 5, 5, 4, 3, 2, 1, 0, 0, 1, 2, 3, 4, 5, 5, 4, 3, 2, 1, 0, 0, 1, 2, 3, 4, 5}; - const static uint8_t remap[][2] = {{ 0, 0}, { 0, 0}, { 0, 0}, { 0, 0}, /* dummy */ + static const uint8_t remap[][2] = {{ 0, 0}, { 0, 0}, { 0, 0}, { 0, 0}, /* dummy */ { 0, 0}, { 0, 1}, { 0, 2}, { 0, 3}, {10, 0}, {10, 1}, {10, 2}, {10, 3}, {20, 0}, {20, 1}, {20, 2}, {20, 3}, {30, 0}, {30, 1}, {30, 2}, @@ -174,19 +137,19 @@ static inline void dv_calc_mb_coordinates(const DVprofile *d, int chan, int seq, break; case 720: switch (d->pix_fmt) { - case PIX_FMT_YUV422P: + case AV_PIX_FMT_YUV422P: x = shuf3[m] + slot/3; y = serpent1[slot] + ((((seq + off[m]) % d->difseg_size)<<1) + chan)*3; tbl[m] = (x<<1)|(y<<8); break; - case PIX_FMT_YUV420P: + case AV_PIX_FMT_YUV420P: x = shuf3[m] + slot/3; y = serpent1[slot] + ((seq + off[m]) % d->difseg_size)*3; tbl[m] = (x<<1)|(y<<9); break; - case PIX_FMT_YUV411P: + case AV_PIX_FMT_YUV411P: i = (seq + off[m]) % d->difseg_size; k = slot + ((m==1||m==2)?3:0); @@ -203,7 +166,16 @@ static inline void dv_calc_mb_coordinates(const DVprofile *d, int chan, int seq, } } -static int dv_init_dynamic_tables(const DVprofile *d) +/* quantization quanta by QNO for DV100 */ +static const uint8_t dv100_qstep[16] = { + 1, /* QNO = 0 and 1 both have no quantization */ + 1, + 2, 3, 4, 5, 6, 7, 8, 16, 18, 20, 22, 24, 28, 52 +}; + +static const uint8_t dv_quant_areas[4] = { 6, 21, 43, 64 }; + +int ff_dv_init_dynamic_tables(const DVprofile *d) { int j,i,c,s,p; uint32_t *factor1, *factor2; @@ -231,12 +203,12 @@ static int dv_init_dynamic_tables(const DVprofile *d) factor1 = &d->idct_factor[0]; factor2 = &d->idct_factor[DV_PROFILE_IS_HD(d)?4096:2816]; if (d->height == 720) { - iweight1 = &dv_iweight_720_y[0]; - iweight2 = &dv_iweight_720_c[0]; + iweight1 = &ff_dv_iweight_720_y[0]; + iweight2 = &ff_dv_iweight_720_c[0]; } else { - iweight1 = &dv_iweight_1080_y[0]; - iweight2 = &dv_iweight_1080_c[0]; - } + iweight1 = &ff_dv_iweight_1080_y[0]; + iweight2 = &ff_dv_iweight_1080_c[0]; + } if (DV_PROFILE_IS_HD(d)) { for (c = 0; c < 4; c++) { for (s = 0; s < 16; s++) { @@ -247,24 +219,24 @@ static int dv_init_dynamic_tables(const DVprofile *d) } } } else { - iweight1 = &dv_iweight_88[0]; - for (j = 0; j < 2; j++, iweight1 = &dv_iweight_248[0]) { + iweight1 = &ff_dv_iweight_88[0]; + for (j = 0; j < 2; j++, iweight1 = &ff_dv_iweight_248[0]) { for (s = 0; s < 22; s++) { for (i = c = 0; c < 4; c++) { for (; i < dv_quant_areas[c]; i++) { - *factor1 = iweight1[i] << (dv_quant_shifts[s][c] + 1); + *factor1 = iweight1[i] << (ff_dv_quant_shifts[s][c] + 1); *factor2++ = (*factor1++) << 1; - } - } + } + } + } } } } -} return 0; } -static av_cold int dvvideo_init(AVCodecContext *avctx) +av_cold int ff_dvvideo_init(AVCodecContext *avctx) { DVVideoContext *s = avctx->priv_data; DSPContext dsp; @@ -317,58 +289,18 @@ static av_cold int dvvideo_init(AVCodecContext *avctx) run = new_dv_vlc_run [code] + 1; level = new_dv_vlc_level[code]; } - dv_rl_vlc[i].len = len; - dv_rl_vlc[i].level = level; - dv_rl_vlc[i].run = run; - } - free_vlc(&dv_vlc); - - for (i = 0; i < NB_DV_VLC - 1; i++) { - if (dv_vlc_run[i] >= DV_VLC_MAP_RUN_SIZE) - continue; -#if CONFIG_SMALL - 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++) { -#if CONFIG_SMALL - 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; - } -#endif + ff_dv_rl_vlc[i].len = len; + ff_dv_rl_vlc[i].level = level; + ff_dv_rl_vlc[i].run = run; } + ff_free_vlc(&dv_vlc); } /* Generic DSP setup */ - dsputil_init(&dsp, avctx); + ff_dsputil_init(&dsp, avctx); + ff_set_cmp(&dsp, dsp.ildct_cmp, avctx->ildct_cmp); s->get_pixels = dsp.get_pixels; + s->ildct_cmp = dsp.ildct_cmp[5]; /* 88DCT setup */ s->fdct[0] = dsp.fdct; @@ -379,321 +311,42 @@ static av_cold int dvvideo_init(AVCodecContext *avctx) /* 248DCT setup */ s->fdct[1] = dsp.fdct248; 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); + memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64); + avcodec_get_frame_defaults(&s->picture); avctx->coded_frame = &s->picture; s->avctx = avctx; + avctx->chroma_sample_location = AVCHROMA_LOC_TOPLEFT; return 0; } -// #define VLC_DEBUG -// #define printf(...) av_log(NULL, AV_LOG_ERROR, __VA_ARGS__) +static av_cold int dvvideo_init_encoder(AVCodecContext *avctx) +{ + if (!avpriv_dv_codec_profile(avctx)) { + av_log(avctx, AV_LOG_ERROR, "Found no DV profile for %ix%i %s video. " + "Valid DV profiles are:\n", + avctx->width, avctx->height, av_get_pix_fmt_name(avctx->pix_fmt)); + ff_dv_print_profiles(avctx, AV_LOG_ERROR); + return AVERROR(EINVAL); + } -typedef struct BlockInfo { - const uint32_t *factor_table; - const uint8_t *scan_table; - uint8_t pos; /* position in block */ - 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; + dv_vlc_map_tableinit(); + + return ff_dvvideo_init(avctx); +} /* 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 }; -static inline int get_bits_left(GetBitContext *s) -{ - return s->size_in_bits - get_bits_count(s); -} - static inline int put_bits_left(PutBitContext* s) { return (s->buf_end - s->buf) * 8 - put_bits_count(s); } -/* decode ac coefficients */ -static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block) -{ - int last_index = gb->size_in_bits; - const uint8_t *scan_table = mb->scan_table; - const uint32_t *factor_table = mb->factor_table; - int pos = mb->pos; - int partial_bit_count = mb->partial_bit_count; - 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; - } - - /* get the AC coefficients until last_index is reached */ - for (;;) { -#ifdef VLC_DEBUG - printf("%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16), re_index); -#endif - /* our own optimized GET_RL_VLC */ - index = NEG_USR32(re_cache, TEX_VLC_BITS); - 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; - -#ifdef VLC_DEBUG - printf("run=%d level=%d\n", run, level); -#endif - 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); - } - CLOSE_READER(re, gb); - mb->pos = pos; -} - -static inline void bit_copy(PutBitContext *pb, GetBitContext *gb) -{ - int bits_left = get_bits_left(gb); - 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)); - } -} - -static inline void dv_calculate_mb_xy(DVVideoContext *s, DVwork_chunk *work_chunk, int m, int *mb_x, int *mb_y) -{ - *mb_x = work_chunk->mb_coordinates[m] & 0xff; - *mb_y = work_chunk->mb_coordinates[m] >> 8; - - /* 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]&0x0C)) { - *mb_y -= (*mb_y>17)?18:-72; /* shifting the Y coordinate down by 72/2 macro blocks */ - } -} - -/* mb_x and mb_y are in units of 8 pixels */ -static int dv_decode_video_segment(AVCodecContext *avctx, DVwork_chunk *work_chunk) -{ - DVVideoContext *s = avctx->priv_data; - int quant, dc, dct_mode, class1, j; - int mb_index, mb_x, mb_y, last_index; - int y_stride, linesize; - DCTELEM *block, *block1; - int c_offset; - uint8_t *y_ptr; - const uint8_t *buf_ptr; - PutBitContext pb, vs_pb; - GetBitContext gb; - 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 */ - buf_ptr = &s->buf[work_chunk->buf_offset*80]; - 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 += 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; - 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_sbits(&gb, 9); - dct_mode = get_bits1(&gb); - class1 = get_bits(&gb, 2); - 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->sys->idct_factor[(j >= 4)*4*16*64 + class1*16*64 + quant*64]; - 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->sys->idct_factor[(class1 == 3)*2*22*64 + dct_mode*22*64 + - (quant + dv_quant_offset[class1])*64]; - } - dc = dc << 2; - /* convert to unsigned because 128 is not added in the - standard IDCT */ - dc += 1024; - block[0] = dc; - buf_ptr += last_index >> 3; - mb->pos = 0; - mb->partial_bit_count = 0; - -#ifdef VLC_DEBUG - printf("MB block: %d, %d ", mb_index, j); -#endif - dv_decode_ac(&gb, mb, block); - - /* write the remaining bits in a new buffer only if the - 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); -#endif - block = block1; - mb = mb1; - init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb)); - 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 */ - if (mb->pos < 64) - break; - } - } - /* all blocks are finished, so the extra bytes can be used at - the video segment level */ - if (j >= s->sys->bpm) - bit_copy(&vs_pb, &gb); - } - - /* we need a pass other the whole video segment */ -#ifdef VLC_DEBUG - printf("***pass 3 size=%d\n", put_bits_count(&vs_pb)); -#endif - block = &sblock[0][0]; - mb = mb_data; - 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 < 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); - block += 64; - mb++; - } - } - - /* compute idct and place blocks */ - block = &sblock[0][0]; - mb = mb_data; - for (mb_index = 0; mb_index < 5; mb_index++) { - dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y); - - /* 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)); - } else { - y_stride = (2 << log2_blocksize); - } - y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x) << log2_blocksize); - linesize = s->picture.linesize[0] << is_field_mode[mb_index]; - mb[0] .idct_put(y_ptr , linesize, block + 0*64); - if (s->sys->video_stype == 4) { /* SD 422 */ - mb[2].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 2*64); - } else { - mb[1].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 1*64); - mb[2].idct_put(y_ptr + y_stride, linesize, block + 2*64); - mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3*64); - } - mb += 4; - block += 4*64; - - /* idct_put'ting chrominance */ - 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))) << log2_blocksize); - for (j = 2; j; j--) { - uint8_t *c_ptr = s->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 << log2_blocksize); y++, c_ptr += s->picture.linesize[j], pixels += 8) { - ptr1 = pixels + (1 << (log2_blocksize - 1)); - c_ptr1 = c_ptr + (s->picture.linesize[j] << log2_blocksize); - for (x = 0; x < (1 << (log2_blocksize - 1)); x++) { - c_ptr[x] = pixels[x]; - c_ptr1[x] = ptr1[x]; - } - } - block += 64; mb++; - } else { - y_stride = (mb_y == 134) ? (1 << log2_blocksize) : - s->picture.linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize); - linesize = s->picture.linesize[j] << is_field_mode[mb_index]; - (mb++)-> idct_put(c_ptr , linesize, block); block += 64; - if (s->sys->bpm == 8) { - (mb++)->idct_put(c_ptr + y_stride, linesize, block); block += 64; - } - } - } - } - return 0; -} - #if CONFIG_SMALL -/* Converts run and level (where level != 0) pair into vlc, returning bit size */ +/* Converts run and level (where level != 0) pair into VLC, returning bit size */ static av_always_inline int dv_rl2vlc(int run, int level, int sign, uint32_t* vlc) { int size; @@ -754,7 +407,7 @@ typedef struct EncBlockInfo { int cur_ac; int cno; int dct_mode; - DCTELEM mb[64]; + int16_t mb[64]; uint8_t next[64]; uint8_t sign[64]; uint8_t partial_bit_count; @@ -804,43 +457,52 @@ static av_always_inline PutBitContext* dv_encode_ac(EncBlockInfo* bi, return pb; } -//FIXME replace this by dsputil -#define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7)) -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); - 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); - s += 8; +static av_always_inline int dv_guess_dct_mode(DVVideoContext *s, uint8_t *data, int linesize) { + if (s->avctx->flags & CODEC_FLAG_INTERLACED_DCT) { + int ps = s->ildct_cmp(NULL, data, NULL, linesize, 8) - 400; + if (ps > 0) { + int is = s->ildct_cmp(NULL, data , NULL, linesize<<1, 4) + + s->ildct_cmp(NULL, data + linesize, NULL, linesize<<1, 4); + return ps > is; + } } - return (score88 - score248 > -10); + return 0; } +static const int dv_weight_bits = 18; +static const int dv_weight_88[64] = { + 131072, 257107, 257107, 242189, 252167, 242189, 235923, 237536, + 237536, 235923, 229376, 231390, 223754, 231390, 229376, 222935, + 224969, 217965, 217965, 224969, 222935, 200636, 218652, 211916, + 212325, 211916, 218652, 200636, 188995, 196781, 205965, 206433, + 206433, 205965, 196781, 188995, 185364, 185364, 200636, 200704, + 200636, 185364, 185364, 174609, 180568, 195068, 195068, 180568, + 174609, 170091, 175557, 189591, 175557, 170091, 165371, 170627, + 170627, 165371, 160727, 153560, 160727, 144651, 144651, 136258, +}; +static const int dv_weight_248[64] = { + 131072, 242189, 257107, 237536, 229376, 200636, 242189, 223754, + 224969, 196781, 262144, 242189, 229376, 200636, 257107, 237536, + 211916, 185364, 235923, 217965, 229376, 211916, 206433, 180568, + 242189, 223754, 224969, 196781, 211916, 185364, 235923, 217965, + 200704, 175557, 222935, 205965, 200636, 185364, 195068, 170627, + 229376, 211916, 206433, 180568, 200704, 175557, 222935, 205965, + 175557, 153560, 188995, 174609, 165371, 144651, 200636, 185364, + 195068, 170627, 175557, 153560, 188995, 174609, 165371, 144651, +}; + static av_always_inline int dv_init_enc_block(EncBlockInfo* bi, uint8_t *data, int linesize, DVVideoContext *s, int bias) { const int *weight; const uint8_t* zigzag_scan; - DECLARE_ALIGNED_16(DCTELEM, blk[64]); + LOCAL_ALIGNED_16(int16_t, blk, [64]); int i, area; /* 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 + component is greater than 36. Libav'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 @@ -848,7 +510,7 @@ static av_always_inline int dv_init_enc_block(EncBlockInfo* bi, uint8_t *data, i #if 0 /* SMPTE spec method */ static const int classes[] = {12, 24, 36, 0xffff}; -#else /* improved FFmpeg method */ +#else /* improved Libav method */ static const int classes[] = {-1, -1, 255, 0xffff}; #endif int max = classes[0]; @@ -861,14 +523,13 @@ static av_always_inline int dv_init_enc_block(EncBlockInfo* bi, uint8_t *data, i bi->partial_bit_buffer = 0; bi->cur_ac = 0; if (data) { + bi->dct_mode = dv_guess_dct_mode(s, data, linesize); s->get_pixels(blk, data, linesize); - bi->dct_mode = (s->avctx->flags & CODEC_FLAG_INTERLACED_DCT) && - dv_guess_dct_mode(blk); s->fdct[bi->dct_mode](blk); } else { /* 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(blk, 0, sizeof(blk)); + memset(blk, 0, 64*sizeof(*blk)); bi->dct_mode = 0; } bi->mb[0] = blk[0]; @@ -884,7 +545,7 @@ static av_always_inline int dv_init_enc_block(EncBlockInfo* bi, uint8_t *data, i if (level + 15 > 30U) { bi->sign[i] = (level >> 31) & 1; - /* weigh it and and shift down into range, adding for rounding */ + /* weight 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); @@ -942,7 +603,7 @@ static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos) 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]) { + if (b->area_q[a] != ff_dv_quant_shifts[qnos[i] + ff_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]; @@ -994,15 +655,15 @@ static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos) } } -static int dv_encode_video_segment(AVCodecContext *avctx, DVwork_chunk *work_chunk) +static int dv_encode_video_segment(AVCodecContext *avctx, void *arg) { DVVideoContext *s = avctx->priv_data; + DVwork_chunk *work_chunk = arg; int mb_index, i, j; - int mb_x, mb_y, c_offset, linesize; + int mb_x, mb_y, c_offset, linesize, y_stride; uint8_t* y_ptr; - uint8_t* data; uint8_t* dif; - uint8_t scratch[64]; + LOCAL_ALIGNED_8(uint8_t, scratch, [128]); EncBlockInfo enc_blks[5*DV_MAX_BPM]; PutBitContext pbs[5*DV_MAX_BPM]; PutBitContext* pb; @@ -1015,56 +676,58 @@ static int dv_encode_video_segment(AVCodecContext *avctx, DVwork_chunk *work_chu enc_blk = &enc_blks[0]; for (mb_index = 0; mb_index < 5; mb_index++) { dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y); + + /* initializing luminance blocks */ + if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) || + (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) || + (s->sys->height >= 720 && mb_y != 134)) { + y_stride = s->picture.linesize[0] << 3; + } else { + y_stride = 16; + } 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); - for (j = 0; j < 6; j++) { - 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 { - /* j=1 and j=3 are "dummy" blocks, used for AC data only */ - data = NULL; - linesize = 0; - } - } 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)) { - uint8_t* d; - uint8_t* b = scratch; - 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; - } - data = scratch; - linesize = 8; - } + linesize = s->picture.linesize[0]; + + if (s->sys->video_stype == 4) { /* SD 422 */ + vs_bit_size += + dv_init_enc_block(enc_blk+0, y_ptr , linesize, s, 0) + + dv_init_enc_block(enc_blk+1, NULL , linesize, s, 0) + + dv_init_enc_block(enc_blk+2, y_ptr + 8 , linesize, s, 0) + + dv_init_enc_block(enc_blk+3, NULL , linesize, s, 0); + } else { + vs_bit_size += + dv_init_enc_block(enc_blk+0, y_ptr , linesize, s, 0) + + dv_init_enc_block(enc_blk+1, y_ptr + 8 , linesize, s, 0) + + dv_init_enc_block(enc_blk+2, y_ptr + y_stride, linesize, s, 0) + + dv_init_enc_block(enc_blk+3, y_ptr + 8 + y_stride, linesize, s, 0); + } + enc_blk += 4; + + /* initializing chrominance blocks */ + c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->picture.linesize[1] + + (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << 3); + for (j = 2; j; j--) { + uint8_t *c_ptr = s->picture.data[j] + c_offset; + linesize = s->picture.linesize[j]; + y_stride = (mb_y == 134) ? 8 : (s->picture.linesize[j] << 3); + if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) { + uint8_t* d; + uint8_t* b = scratch; + for (i = 0; i < 8; i++) { + d = c_ptr + (linesize << 3); + b[0] = c_ptr[0]; b[1] = c_ptr[1]; b[2] = c_ptr[2]; b[3] = c_ptr[3]; + b[4] = d[0]; b[5] = d[1]; b[6] = d[2]; b[7] = d[3]; + c_ptr += linesize; + b += 16; } + c_ptr = scratch; + linesize = 16; } - vs_bit_size += dv_init_enc_block(enc_blk, data, linesize, s, j>>2); - - ++enc_blk; + vs_bit_size += dv_init_enc_block( enc_blk++, c_ptr , linesize, s, 1); + if (s->sys->bpm == 8) { + vs_bit_size += dv_init_enc_block(enc_blk++, c_ptr + y_stride, linesize, s, 1); + } } } @@ -1083,7 +746,7 @@ static int dv_encode_video_segment(AVCodecContext *avctx, DVwork_chunk *work_chu int sz = s->sys->block_sizes[i]>>3; init_put_bits(&pbs[j], dif, sz); - put_bits(&pbs[j], 9, (uint16_t)(((enc_blks[j].mb[0] >> 3) - 1024 + 2) >> 2)); + put_sbits(&pbs[j], 9, ((enc_blks[j].mb[0] >> 3) - 1024 + 2) >> 2); put_bits(&pbs[j], 1, enc_blks[j].dct_mode); put_bits(&pbs[j], 2, enc_blks[j].cno); @@ -1105,58 +768,23 @@ static int dv_encode_video_segment(AVCodecContext *avctx, DVwork_chunk *work_chu if (enc_blks[j].partial_bit_count) pb = dv_encode_ac(&enc_blks[j], pb, &pbs[s->sys->bpm*5]); if (enc_blks[j].partial_bit_count) - av_log(NULL, AV_LOG_ERROR, "ac bitstream overflow\n"); + av_log(avctx, AV_LOG_ERROR, "ac bitstream overflow\n"); } - for (j=0; j<5*s->sys->bpm; j++) + for (j=0; j<5*s->sys->bpm; j++) { + int pos; + int size = pbs[j].size_in_bits >> 3; flush_put_bits(&pbs[j]); - - return 0; -} - -#if CONFIG_DVVIDEO_DECODER -/* NOTE: exactly one frame must be given (120000 bytes for NTSC, - 144000 bytes for PAL - or twice those for 50Mbps) */ -static int dvvideo_decode_frame(AVCodecContext *avctx, - void *data, int *data_size, - const uint8_t *buf, int buf_size) -{ - DVVideoContext *s = avctx->priv_data; - - s->sys = dv_frame_profile(buf); - if (!s->sys || buf_size < s->sys->frame_size || dv_init_dynamic_tables(s->sys)) - 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->time_base = s->sys->time_base; - 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; + pos = put_bits_count(&pbs[j]) >> 3; + if (pos > size) { + av_log(avctx, AV_LOG_ERROR, "bitstream written beyond buffer size\n"); + return -1; + } + memset(pbs[j].buf + pos, 0xff, size - pos); } - s->picture.interlaced_frame = 1; - s->picture.top_field_first = 0; - - s->buf = buf; - avctx->execute(avctx, dv_decode_video_segment, s->sys->work_chunks, NULL, - dv_work_pool_size(s->sys), sizeof(DVwork_chunk)); - - emms_c(); - /* return image */ - *data_size = sizeof(AVFrame); - *(AVFrame*)data = s->picture; - - return s->sys->frame_size; + return 0; } -#endif /* CONFIG_DVVIDEO_DECODER */ - static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c, uint8_t* buf) @@ -1179,11 +807,10 @@ static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c, * 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); + int apt = (c->sys->pix_fmt == AV_PIX_FMT_YUV420P ? 0 : 1); uint8_t aspect = 0; - if ((int)(av_q2d(c->avctx->sample_aspect_ratio) * c->avctx->width / c->avctx->height * 10) == 17) /* 16:9 */ + 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; @@ -1210,7 +837,7 @@ static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c, 0xf; /* reserved -- always 1 */ buf[3] = (3 << 6) | /* reserved -- always 1 */ (c->sys->dsf << 5) | /* system: 60fields/50fields */ - stype; /* signal type video compression */ + c->sys->video_stype; /* signal type video compression */ buf[4] = 0xff; /* VISC: 0xff -- no information */ break; case dv_video_control: @@ -1232,6 +859,41 @@ static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c, } #if CONFIG_DVVIDEO_ENCODER +static inline int dv_write_dif_id(enum dv_section_type t, uint8_t chan_num, + uint8_t seq_num, uint8_t dif_num, + uint8_t* buf) +{ + buf[0] = (uint8_t)t; /* Section type */ + buf[1] = (seq_num << 4) | /* DIF seq number 0-9 for 525/60; 0-11 for 625/50 */ + (chan_num << 3) | /* FSC: for 50Mb/s 0 - first channel; 1 - second */ + 7; /* reserved -- always 1 */ + buf[2] = dif_num; /* DIF block number Video: 0-134, Audio: 0-8 */ + return 3; +} + + +static inline int dv_write_ssyb_id(uint8_t syb_num, uint8_t fr, uint8_t* buf) +{ + if (syb_num == 0 || syb_num == 6) { + buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ + (0 << 4) | /* AP3 (Subcode application ID) */ + 0x0f; /* reserved -- always 1 */ + } + else if (syb_num == 11) { + buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ + 0x7f; /* reserved -- always 1 */ + } + else { + buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ + (0 << 4) | /* APT (Track application ID) */ + 0x0f; /* reserved -- always 1 */ + } + buf[1] = 0xf0 | /* reserved -- always 1 */ + (syb_num & 0x0f); /* SSYB number 0 - 11 */ + buf[2] = 0xff; /* reserved -- always 1 */ + return 3; +} + static void dv_format_frame(DVVideoContext* c, uint8_t* buf) { int chan, i, j, k; @@ -1282,68 +944,50 @@ static void dv_format_frame(DVVideoContext* c, uint8_t* buf) } -static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size, - void *data) +static int dvvideo_encode_frame(AVCodecContext *c, AVPacket *pkt, + const AVFrame *frame, int *got_packet) { DVVideoContext *s = c->priv_data; + int ret; - s->sys = dv_codec_profile(c); - if (!s->sys || buf_size < s->sys->frame_size || dv_init_dynamic_tables(s->sys)) + s->sys = avpriv_dv_codec_profile(c); + if (!s->sys || ff_dv_init_dynamic_tables(s->sys)) return -1; + if ((ret = ff_alloc_packet(pkt, s->sys->frame_size)) < 0) { + av_log(c, AV_LOG_ERROR, "Error getting output packet.\n"); + return ret; + } c->pix_fmt = s->sys->pix_fmt; - s->picture = *((AVFrame *)data); + s->picture = *frame; s->picture.key_frame = 1; - s->picture.pict_type = FF_I_TYPE; + s->picture.pict_type = AV_PICTURE_TYPE_I; - s->buf = buf; + s->buf = pkt->data; c->execute(c, dv_encode_video_segment, s->sys->work_chunks, NULL, dv_work_pool_size(s->sys), sizeof(DVwork_chunk)); emms_c(); - dv_format_frame(s, buf); + dv_format_frame(s, pkt->data); - 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); + pkt->flags |= AV_PKT_FLAG_KEY; + *got_packet = 1; return 0; } - -#if CONFIG_DVVIDEO_ENCODER -AVCodec dvvideo_encoder = { - "dvvideo", - CODEC_TYPE_VIDEO, - CODEC_ID_DVVIDEO, - sizeof(DVVideoContext), - dvvideo_init, - dvvideo_encode_frame, - .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)"), +AVCodec ff_dvvideo_encoder = { + .name = "dvvideo", + .type = AVMEDIA_TYPE_VIDEO, + .id = AV_CODEC_ID_DVVIDEO, + .priv_data_size = sizeof(DVVideoContext), + .init = dvvideo_init_encoder, + .encode2 = dvvideo_encode_frame, + .capabilities = CODEC_CAP_SLICE_THREADS, + .pix_fmts = (const enum AVPixelFormat[]) { + AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE + }, + .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"), }; #endif // CONFIG_DVVIDEO_ENCODER - -#if CONFIG_DVVIDEO_DECODER -AVCodec dvvideo_decoder = { - "dvvideo", - CODEC_TYPE_VIDEO, - CODEC_ID_DVVIDEO, - sizeof(DVVideoContext), - dvvideo_init, - NULL, - dvvideo_close, - dvvideo_decode_frame, - CODEC_CAP_DR1, - NULL, - .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"), -}; -#endif