From a712d725c0d466cc3672d626883319ef828ca8d6 Mon Sep 17 00:00:00 2001 From: Michael Niedermayer Date: Fri, 15 Feb 2008 02:57:19 +0000 Subject: [PATCH] Header elision muxing support. This only removes 2 bytes from MP3 and MP2 currently. Up to 4 could be removed from MP3/MP2 though this might need a 2pass muxer. Primitive code to remove headers from MPEG-1/2/4 is there too but for the single file I tried it on (the one in the regression tests), it was a loss because all video frames were >4096 byte, so that it is disabled ATM. Originally committed as revision 11936 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavformat/nutenc.c | 165 ++++++++++++++++++++++++++++++++++++- tests/libav.regression.ref | 4 +- tests/seek.regression.ref | 54 ++++++------ 3 files changed, 191 insertions(+), 32 deletions(-) diff --git a/libavformat/nutenc.c b/libavformat/nutenc.c index 9a17c7759f..f3f8ef9fd7 100644 --- a/libavformat/nutenc.c +++ b/libavformat/nutenc.c @@ -21,6 +21,108 @@ #include "nut.h" #include "tree.h" +#include "mpegaudiodata.h" + +static int find_expected_header(AVCodecContext *c, int size, int key_frame, uint8_t out[64]){ + int sample_rate= c->sample_rate; + + if(size>4096) + return 0; + + AV_WB24(out, 1); + + if(c->codec_id == CODEC_ID_MPEG4){ + if(key_frame){ + return 3; + }else{ + out[3]= 0xB6; + return 4; + } + }else if(c->codec_id == CODEC_ID_MPEG1VIDEO || c->codec_id == CODEC_ID_MPEG2VIDEO){ + return 3; + }else if(c->codec_id == CODEC_ID_H264){ + return 3; + }else if(c->codec_id == CODEC_ID_MP3 || c->codec_id == CODEC_ID_MP2){ + int lsf, mpeg25, sample_rate_index, bitrate_index, frame_size; + int layer= c->codec_id == CODEC_ID_MP3 ? 3 : 2; + unsigned int header= 0xFFF00000; + + lsf = sample_rate < (24000+32000)/2; + mpeg25 = sample_rate < (12000+16000)/2; + sample_rate <<= lsf + mpeg25; + if (sample_rate < (32000 + 44100)/2) sample_rate_index=2; + else if(sample_rate < (44100 + 48000)/2) sample_rate_index=0; + else sample_rate_index=1; + + sample_rate= ff_mpa_freq_tab[sample_rate_index] >> (lsf + mpeg25); + + for(bitrate_index=2; bitrate_index<30; bitrate_index++){ + frame_size = ff_mpa_bitrate_tab[lsf][layer-1][bitrate_index>>1]; + frame_size = (frame_size * 144000) / (sample_rate << lsf) + (bitrate_index&1); + + if(frame_size == size) + break; + } + + header |= (!lsf)<<19; + header |= (4-layer)<<17; + header |= 1<<16; //no crc + AV_WB32(out, header); + if(size <= 0) + return 2; //we guess theres no crc, if there is one the user clearly doesnt care about overhead + if(bitrate_index == 30) + return -1; //something is wrong ... + + header |= (bitrate_index>>1)<<12; + header |= sample_rate_index<<10; + header |= (bitrate_index&1)<<9; + + return 2; //FIXME actually put the needed ones in build_elision_headers() + return 3; //we guess that the private bit isnt set +//FIXME the above asumtations should be checked, if these turn out false too often something should be done + } + return 0; +} + +static int find_header_idx(AVFormatContext *s, AVCodecContext *c, int size, int frame_type){ + NUTContext *nut = s->priv_data; + uint8_t out[64]; + int i; + int len= find_expected_header(c, size, frame_type, out); + +//av_log(NULL, AV_LOG_ERROR, "expected_h len=%d size=%d codec_id=%d\n", len, size, c->codec_id); + + for(i=1; iheader_count; i++){ + if( len == nut->header_len[i] + && !memcmp(out, nut->header[i], len)){ +// av_log(NULL, AV_LOG_ERROR, "found %d\n", i); + return i; + } + } +// av_log(NULL, AV_LOG_ERROR, "nothing found\n"); + return 0; +} + +static void build_elision_headers(AVFormatContext *s){ + NUTContext *nut = s->priv_data; + int i; + //FIXME this is lame + //FIXME write a 2pass mode to find the maximal headers + const static uint8_t headers[][5]={ + {3, 0x00, 0x00, 0x01}, + {4, 0x00, 0x00, 0x01, 0xB6}, + {2, 0xFF, 0xFA}, //mp3+crc + {2, 0xFF, 0xFB}, //mp3 + {2, 0xFF, 0xFC}, //mp2+crc + {2, 0xFF, 0xFD}, //mp2 + }; + + nut->header_count= 7; + for(i=1; iheader_count; i++){ + nut->header_len[i]= headers[i-1][0]; + nut->header [i]= &headers[i-1][1]; + } +} static void build_frame_code(AVFormatContext *s){ NUTContext *nut = s->priv_data; @@ -63,6 +165,8 @@ static void build_frame_code(AVFormatContext *s){ ft->flags|= FLAG_SIZE_MSB | FLAG_CODED_PTS; ft->stream_id= stream_id; ft->size_mul=1; + if(is_audio) + ft->header_idx= find_header_idx(s, codec, -1, key_frame); start2++; } } @@ -80,6 +184,7 @@ static void build_frame_code(AVFormatContext *s){ ft->size_mul=frame_bytes + 2; ft->size_lsb=frame_bytes + pred; ft->pts_delta=pts; + ft->header_idx= find_header_idx(s, codec, frame_bytes + pred, key_frame); start2++; } } @@ -123,6 +228,8 @@ static void build_frame_code(AVFormatContext *s){ ft->size_mul= end3-start3; ft->size_lsb= index - start3; ft->pts_delta= pred_table[pred]; + if(is_audio) + ft->header_idx= find_header_idx(s, codec, -1, key_frame); } } } @@ -212,7 +319,8 @@ static void put_packet(NUTContext *nut, ByteIOContext *bc, ByteIOContext *dyn_bc } static void write_mainheader(NUTContext *nut, ByteIOContext *bc){ - int i, j, tmp_pts, tmp_flags, tmp_stream, tmp_mul, tmp_size, tmp_fields; + int i, j, tmp_pts, tmp_flags, tmp_stream, tmp_mul, tmp_size, tmp_fields, tmp_head_idx; + int64_t tmp_match; put_v(bc, 3); /* version */ put_v(bc, nut->avf->nb_streams); @@ -227,6 +335,8 @@ static void write_mainheader(NUTContext *nut, ByteIOContext *bc){ tmp_pts=0; tmp_mul=1; tmp_stream=0; + tmp_match= 1-(1LL<<62); + tmp_head_idx= 0; for(i=0; i<256;){ tmp_fields=0; tmp_size=0; @@ -236,6 +346,7 @@ static void write_mainheader(NUTContext *nut, ByteIOContext *bc){ if(tmp_stream != nut->frame_code[i].stream_id) tmp_fields=3; if(tmp_size != nut->frame_code[i].size_lsb ) tmp_fields=4; // if(tmp_res != nut->frame_code[i].res ) tmp_fields=5; + if(tmp_head_idx!=nut->frame_code[i].header_idx)tmp_fields=8; tmp_pts = nut->frame_code[i].pts_delta; tmp_flags = nut->frame_code[i].flags; @@ -243,6 +354,7 @@ static void write_mainheader(NUTContext *nut, ByteIOContext *bc){ tmp_mul = nut->frame_code[i].size_mul; tmp_size = nut->frame_code[i].size_lsb; // tmp_res = nut->frame_code[i].res; + tmp_head_idx= nut->frame_code[i].header_idx; for(j=0; i<256; j++,i++){ if(i == 'N'){ @@ -255,6 +367,7 @@ static void write_mainheader(NUTContext *nut, ByteIOContext *bc){ if(nut->frame_code[i].size_mul != tmp_mul ) break; if(nut->frame_code[i].size_lsb != tmp_size+j) break; // if(nut->frame_code[i].res != tmp_res ) break; + if(nut->frame_code[i].header_idx!= tmp_head_idx) break; } if(j != tmp_mul - tmp_size) tmp_fields=6; @@ -266,6 +379,13 @@ static void write_mainheader(NUTContext *nut, ByteIOContext *bc){ if(tmp_fields>3) put_v(bc, tmp_size); if(tmp_fields>4) put_v(bc, 0 /*tmp_res*/); if(tmp_fields>5) put_v(bc, j); + if(tmp_fields>6) put_v(bc, tmp_match); + if(tmp_fields>7) put_v(bc, tmp_head_idx); + } + put_v(bc, nut->header_count-1); + for(i=1; iheader_count; i++){ + put_v(bc, nut->header_len[i]); + put_buffer(bc, nut->header[i], nut->header_len[i]); } } @@ -419,6 +539,7 @@ static int write_header(AVFormatContext *s){ } nut->max_distance = MAX_DISTANCE; + build_elision_headers(s); build_frame_code(s); assert(nut->frame_code['N'].flags == FLAG_INVALID); @@ -444,17 +565,40 @@ static int get_needed_flags(NUTContext *nut, StreamContext *nus, FrameCode *fc, if(pkt->size > 2*nut->max_distance ) flags |= FLAG_CHECKSUM; if(FFABS(pkt->pts - nus->last_pts) > nus->max_pts_distance) flags |= FLAG_CHECKSUM; + if( pkt->size < nut->header_len[fc->header_idx] + || (pkt->size > 4096 && fc->header_idx) + || memcmp(pkt->data, nut->header[fc->header_idx], nut->header_len[fc->header_idx])) + flags |= FLAG_HEADER_IDX; return flags | (fc->flags & FLAG_CODED); } +static int find_best_header_idx(NUTContext *nut, AVPacket *pkt){ + int i; + int best_i = 0; + int best_len= 0; + + if(pkt->size > 4096) + return 0; + + for(i=1; iheader_count; i++){ + if( pkt->size >= nut->header_len[i] + && nut->header_len[i] > best_len + && !memcmp(pkt->data, nut->header[i], nut->header_len[i])){ + best_i= i; + best_len= nut->header_len[i]; + } + } + return best_i; +} + static int write_packet(AVFormatContext *s, AVPacket *pkt){ NUTContext *nut = s->priv_data; StreamContext *nus= &nut->stream[pkt->stream_index]; ByteIOContext *bc = s->pb, *dyn_bc; FrameCode *fc; int64_t coded_pts; - int best_length, frame_code, flags, needed_flags, i; + int best_length, frame_code, flags, needed_flags, i, header_idx, best_header_idx; int key_frame = !!(pkt->flags & PKT_FLAG_KEY); int store_sp=0; int ret; @@ -503,6 +647,8 @@ static int write_packet(AVFormatContext *s, AVPacket *pkt){ if(ff_lsb2full(nus, coded_pts) != pkt->pts) coded_pts= pkt->pts + (1<msb_pts_shift); + best_header_idx= find_best_header_idx(nut, pkt); + best_length=INT_MAX; frame_code= -1; for(i=0; i<256; i++){ @@ -539,6 +685,17 @@ static int write_packet(AVFormatContext *s, AVPacket *pkt){ if(flags & FLAG_CODED_PTS) length += get_length(coded_pts); + if( (flags & FLAG_CODED) + && nut->header_len[best_header_idx] > nut->header_len[fc->header_idx]+1){ + flags |= FLAG_HEADER_IDX; + } + + if(flags & FLAG_HEADER_IDX){ + length += 1 - nut->header_len[best_header_idx]; + }else{ + length -= nut->header_len[fc->header_idx]; + } + length*=4; length+= !(flags & FLAG_CODED_PTS); length+= !(flags & FLAG_CHECKSUM); @@ -552,6 +709,7 @@ static int write_packet(AVFormatContext *s, AVPacket *pkt){ fc= &nut->frame_code[frame_code]; flags= fc->flags; needed_flags= get_needed_flags(nut, nus, fc, pkt); + header_idx= fc->header_idx; init_checksum(bc, ff_crc04C11DB7_update, 0); put_byte(bc, frame_code); @@ -562,11 +720,12 @@ static int write_packet(AVFormatContext *s, AVPacket *pkt){ if(flags & FLAG_STREAM_ID) put_v(bc, pkt->stream_index); if(flags & FLAG_CODED_PTS) put_v(bc, coded_pts); if(flags & FLAG_SIZE_MSB) put_v(bc, pkt->size / fc->size_mul); + if(flags & FLAG_HEADER_IDX) put_v(bc, header_idx= best_header_idx); if(flags & FLAG_CHECKSUM) put_le32(bc, get_checksum(bc)); else get_checksum(bc); - put_buffer(bc, pkt->data, pkt->size); + put_buffer(bc, pkt->data + nut->header_len[header_idx], pkt->size - nut->header_len[header_idx]); nus->last_flags= flags; //FIXME just store one per syncpoint diff --git a/tests/libav.regression.ref b/tests/libav.regression.ref index 4791dc09b4..edee7aec09 100644 --- a/tests/libav.regression.ref +++ b/tests/libav.regression.ref @@ -30,8 +30,8 @@ f8ad5bd78f4d012a8ce9570aa395ac54 *./tests/data/b-libav.flv 2ba41cf880aa5cb87e8a08fdb735398d *./tests/data/b-libav.gxf 814640 ./tests/data/b-libav.gxf ./tests/data/b-libav.gxf CRC=0xa376c09e -868101eb2d9f8fe128f84399d450633e *./tests/data/b-libav.nut -329785 ./tests/data/b-libav.nut +7e8f5cfb9c1c0ac60260a10e2f72f3ea *./tests/data/b-libav.nut +329347 ./tests/data/b-libav.nut ./tests/data/b-libav.nut CRC=0x400c29e9 43238b15f1012aa27536c2be20bb2383 *./tests/data/b-libav.mkv 329854 ./tests/data/b-libav.mkv diff --git a/tests/seek.regression.ref b/tests/seek.regression.ref index 5bd80d3508..f1d9f97a8c 100644 --- a/tests/seek.regression.ref +++ b/tests/seek.regression.ref @@ -3193,59 +3193,59 @@ ret: 0 st:-1 ts:-0.645825 flags:1 ret: 0 st: 1 dts:0.500000 pts:0.500000 pos:-1 size:208 flags:1 ---------------- tests/data/b-libav.nut -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st:-1 ts:-1.000000 flags:0 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st:-1 ts:1.894167 flags:1 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 0 ts:0.800000 flags:0 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 0 ts:-0.320000 flags:1 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st: 1 ts:2.586122 flags:0 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 1 ts:1.462857 flags:1 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st:-1 ts:0.365002 flags:0 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st:-1 ts:-0.740831 flags:1 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st: 0 ts:2.160000 flags:0 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 0 ts:1.040000 flags:1 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 1 ts:-0.052245 flags:0 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st: 1 ts:2.847347 flags:1 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st:-1 ts:1.730004 flags:0 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st:-1 ts:0.624171 flags:1 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 0 ts:-0.480000 flags:0 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st: 0 ts:2.400000 flags:1 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 1 ts:1.306122 flags:0 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 1 ts:0.208980 flags:1 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st:-1 ts:-0.904994 flags:0 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st:-1 ts:1.989173 flags:1 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 0 ts:0.880000 flags:0 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 0 ts:-0.240000 flags:1 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ret: 0 st: 1 ts:2.664490 flags:0 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st: 1 ts:1.567347 flags:1 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st:-1 ts:0.460008 flags:0 -ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149135 size:27412 flags:1 +ret: 0 st: 0 dts:0.480000 pts:0.480000 pos:149133 size:27412 flags:1 ret: 0 st:-1 ts:-0.645825 flags:1 -ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:243 size:28088 flags:1 +ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:279 size:28088 flags:1 ---------------- tests/data/b-libav.ogg ret: 0 st: 0 dts:0.000000 pts:0.000000 pos:-1 size:2352 flags:1