ffmpeg源码分析与应用示例（一）——H.264解码与QP提取
本文包含以下内容
1、H.264解码流程详述与对应ffmpeg源码解析
2、针对一个应用实例介绍通过修改ffmpeg源码解决问题的方案
具有较强的综合性。
先介绍一下在第二部分中将要解决的实际问题：自ffmpeg 1.2版本之后，用于描述解码后的视音频原始数据相关信息的AVFrame结构体被移出了avcodec库，转而加入了avutil库之中，这样的改变本来是合理的，但是也带来了一些问题。例如，以前在AVFrame中可以直接提取解码得到的QP table（int8_t *qscale_table）和QP stride（int qstride），但是因为这两个变量需要利用avcodec库进行解码后才能获得，于是在AVFrame独立之后，这两个变量（以及包括mbskip_table在内的很多其他变量）也被标注为弃用的（attribute_deprecated），无法再进行提取。在进行诸如码流分析或质量评价之类的工作时，QP table是经常要用到的，于是这篇文章里就介绍一下如何通过分析ffmpeg源码并进行修改来重新获得QP table和QP stride。
简单介绍一下ffmpeg源码调试与修改的基本方法：
首先是基本的conifigure和make，configure命令不用太复杂，像下面这个这么简单的就可以，需要注意的是不要--enable-shared，否则调试时不能step into。如果是windows的话，需要mingw。
# ./configure --enable-gpl --enable-debug --disable-optimizations --disable-asm --enable-ffplay然后打开eclipse+cdt，选File->new->Makefile Project with Existing Code，接下来的对话框中的Toolchain for Indexer Settings中选Linux GCC，完成项目的新建之后右键点击ffmpeg_g选debug即可开始调试。如果进行修改的话，在每次修改了源码之后重新进行make即可。
H.264解码流程详述与对应ffmpeg源码解析
这里的分析基于20150630的ffmpeg版本。
ffmpeg中通过av_register_all注册了诸多的编解码器、封装格式、协议等，由此，在进行针对相应内容进行操作时会调用对应的函数。如下，在ffmpeg的API函数avcodec_decode_video2中，通过调用AVCodec的decode函数完成解码操作的关键部分
int attribute_align_arg avcodec_decode_video2(AVCodecContext *avctx, AVFrame *picture,
int *got_picture_ptr,
const AVPacket *avpkt)
{
AVCodecInternal *avci = avctx->internal;
int ret;
// copy to ensure we do not change avpkt
AVPacket tmp = *avpkt;
if (!avctx->codec)
return AVERROR(EINVAL);
if (avctx->codec->type != AVMEDIA_TYPE_VIDEO) {
av_log(avctx, AV_LOG_ERROR, "Invalid media type for video\n");
return AVERROR(EINVAL);
}
*got_picture_ptr = 0;
if ((avctx->coded_width || avctx->coded_height) && av_image_check_size(avctx->coded_width, avctx->coded_height, 0, avctx))
return AVERROR(EINVAL);
av_frame_unref(picture);
if ((avctx->codec->capabilities & CODEC_CAP_DELAY) || avpkt->size || (avctx->active_thread_type & FF_THREAD_FRAME)) {
int did_split = av_packet_split_side_data(&tmp);
ret = apply_param_change(avctx, &tmp);
if (ret < 0) {
av_log(avctx, AV_LOG_ERROR, "Error applying parameter changes.\n");
if (avctx->err_recognition & AV_EF_EXPLODE)
goto fail;
}
avctx->internal->pkt = &tmp;
if (HAVE_THREADS && avctx->active_thread_type & FF_THREAD_FRAME)
ret = ff_thread_decode_frame(avctx, picture, got_picture_ptr,
&tmp);
else {
ret = avctx->codec->decode(avctx, picture, got_picture_ptr,
&tmp);//解码功能的核心部分
picture->pkt_dts = avpkt->dts;
if(!avctx->has_b_frames){
av_frame_set_pkt_pos(picture, avpkt->pos);
}
//FIXME these should be under if(!avctx->has_b_frames)
/* get_buffer is supposed to set frame parameters */
if (!(avctx->codec->capabilities & CODEC_CAP_DR1)) {
if (!picture->sample_aspect_ratio.num) picture->sample_aspect_ratio = avctx->sample_aspect_ratio;
if (!picture->width) picture->width = avctx->width;
if (!picture->height) picture->height = avctx->height;
if (picture->format == AV_PIX_FMT_NONE) picture->format = avctx->pix_fmt;
}
}
add_metadata_from_side_data(avctx, picture);
fail:
emms_c(); //needed to avoid an emms_c() call before every return;
avctx->internal->pkt = NULL;
if (did_split) {
av_packet_free_side_data(&tmp);
if(ret == tmp.size)
ret = avpkt->size;
}
if (*got_picture_ptr) {
if (!avctx->refcounted_frames) {
int err = unrefcount_frame(avci, picture);
if (err < 0)
return err;
}
avctx->frame_number++;
av_frame_set_best_effort_timestamp(picture,
guess_correct_pts(avctx,
picture->pkt_pts,
picture->pkt_dts));
} else
av_frame_unref(picture);
} else
ret = 0;
/* many decoders assign whole AVFrames, thus overwriting extended_data;
* make sure it's set correctly */
av_assert0(!picture->extended_data || picture->extended_data == picture->data);
#if FF_API_AVCTX_TIMEBASE
if (avctx->framerate.num > 0 && avctx->framerate.den > 0)
avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
#endif
return ret;
}
而注册的H.264具体解码功能包含在结构体ff_h264_decoder中，如下
AVCodec ff_h264_decoder = {
.name = "h264",
.long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
.type = AVMEDIA_TYPE_VIDEO,
.id = AV_CODEC_ID_H264,
.priv_data_size = sizeof(H264Context),
.init = ff_h264_decode_init,
.close = h264_decode_end,
.decode = h264_decode_frame,
.capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
CODEC_CAP_FRAME_THREADS,
.flush = flush_dpb,
.init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
.update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
.profiles = NULL_IF_CONFIG_SMALL(profiles),
.priv_class = &h264_class,
};可以看到它就是一个AVCodec结构体，里面包含了H264解码的初始化、实际解码、关闭等操作。因此，要分析ffmpeg源码中的H264解码部分，就要重点看h264_decode_frame
先给出一个标准的H.264解码流程图，如下
解码器从NAL中接收压缩的比特流，经过对码流进行熵解码获得量化系数等参数，量化参数X经过反量化和反变换后得到参数数据D‘，解码器使用从码流中解码得到的头信息创建一个预测块P，P与D'求和得到图像块数据uF'，最后每个uF’通过去块效应滤波器得到重建图像的解码块F。
下面结合此流程图的各部分进行分析
首先是从NAL中接收压缩的比特流并进行解码，见ff_h264_decode_frame中的decode_nal_units
static int h264_decode_frame(AVCodecContext *avctx, void *data,
int *got_frame, AVPacket *avpkt)
{
……
if (h->is_avc && av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, NULL)) {
int side_size;
uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, &side_size);
if (is_extra(side, side_size))
ff_h264_decode_extradata(h, side, side_size);
}
if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
if (is_extra(buf, buf_size))
return ff_h264_decode_extradata(h, buf, buf_size);
}
buf_index = decode_nal_units(h, buf, buf_size, 0);
if (buf_index < 0)
return AVERROR_INVALIDDATA;
if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
av_assert0(buf_index <= buf_size);
goto out;
}
……
av_assert0(pict->buf[0] || !*got_frame);
ff_h264_unref_picture(h, &h->last_pic_for_ec);
return get_consumed_bytes(buf_index, buf_size);
}
进一步深入decode_nal_units，可见其首先调用了ff_h264_decode_nal对NAL单元进行解析，然后根据NAL单元类型的不同调用不同的处理函数，例如，对于SPS，调用ff_h264_decode_seq_parameter_set对SPS进行解析。在经过解析之后，会调用ff_h264_execute_decode_slices进行真正的解码slice的操作
static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
int parse_extradata)
{
AVCodecContext *const avctx = h->avctx;
H264Context *hx; ///< thread context
……
for (;;) {
……
ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
&consumed, next_avc - buf_index);
if (!ptr || dst_length < 0) {
ret = -1;
goto end;
}
……
switch (hx->nal_unit_type) {
case NAL_IDR_SLICE:
if ((ptr[0] & 0xFC) == 0x98) {
av_log(h->avctx, AV_LOG_ERROR, "Invalid inter IDR frame\n");
h->next_outputed_poc = INT_MIN;
ret = -1;
goto end;
}
if (h->nal_unit_type != NAL_IDR_SLICE) {
av_log(h->avctx, AV_LOG_ERROR,
"Invalid mix of idr and non-idr slices\n");
ret = -1;
goto end;
}
if(!idr_cleared)
idr(h); // FIXME ensure we don't lose some frames if there is reordering
idr_cleared = 1;
h->has_recovery_point = 1;
case NAL_SLICE:
init_get_bits(&hx->gb, ptr, bit_length);
hx->intra_gb_ptr =
hx->inter_gb_ptr = &hx->gb;
if ((err = ff_h264_decode_slice_header(hx, h)))
break;
if (h->sei_recovery_frame_cnt >= 0) {
if (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I)
h->valid_recovery_point = 1;
if ( h->recovery_frame < 0
|| ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt) {
h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) &
((1 << h->sps.log2_max_frame_num) - 1);
if (!h->valid_recovery_point)
h->recovery_frame = h->frame_num;
}
}
h->cur_pic_ptr->f.key_frame |=
(hx->nal_unit_type == NAL_IDR_SLICE);
if (hx->nal_unit_type == NAL_IDR_SLICE ||
h->recovery_frame == h->frame_num) {
h->recovery_frame = -1;
h->cur_pic_ptr->recovered = 1;
}
// If we have an IDR, all frames after it in decoded order are
// "recovered".
if (hx->nal_unit_type == NAL_IDR_SLICE)
h->frame_recovered |= FRAME_RECOVERED_IDR;
h->frame_recovered |= 3*!!(avctx->flags2 & CODEC_FLAG2_SHOW_ALL);
h->frame_recovered |= 3*!!(avctx->flags & CODEC_FLAG_OUTPUT_CORRUPT);
#if 1
h->cur_pic_ptr->recovered |= h->frame_recovered;
#else
h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
#endif
if (h->current_slice == 1) {
if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS))
decode_postinit(h, nal_index >= nals_needed);
if (h->avctx->hwaccel &&
(ret = h->avctx->hwaccel->start_frame(h->avctx, NULL, 0)) < 0)
return ret;
if (CONFIG_H264_VDPAU_DECODER &&
h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
ff_vdpau_h264_picture_start(h);
}
if (hx->redundant_pic_count == 0) {
if (avctx->hwaccel) {
ret = avctx->hwaccel->decode_slice(avctx,
&buf[buf_index - consumed],
consumed);
if (ret < 0)
return ret;
} else if (CONFIG_H264_VDPAU_DECODER &&
h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0],
start_code,
sizeof(start_code));
ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0],
&buf[buf_index - consumed],
consumed);
} else
context_count++;
}
break;
case NAL_DPA:
case NAL_DPB:
case NAL_DPC:
avpriv_request_sample(avctx, "data partitioning");
ret = AVERROR(ENOSYS);
goto end;
break;
case NAL_SEI:
init_get_bits(&h->gb, ptr, bit_length);
ret = ff_h264_decode_sei(h);
if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
goto end;
break;
case NAL_SPS:
init_get_bits(&h->gb, ptr, bit_length);
if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? nalsize : 1)) {
av_log(h->avctx, AV_LOG_DEBUG,
"SPS decoding failure, trying again with the complete NAL\n");
if (h->is_avc)
av_assert0(next_avc - buf_index + consumed == nalsize);
if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
break;
init_get_bits(&h->gb, &buf[buf_index + 1 - consumed],
8*(next_avc - buf_index + consumed - 1));
ff_h264_decode_seq_parameter_set(h);
}
break;
case NAL_PPS:
init_get_bits(&h->gb, ptr, bit_length);
ret = ff_h264_decode_picture_parameter_set(h, bit_length);
if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
goto end;
break;
case NAL_AUD:
case NAL_END_SEQUENCE:
case NAL_END_STREAM:
case NAL_FILLER_DATA:
case NAL_SPS_EXT:
case NAL_AUXILIARY_SLICE:
break;
case NAL_FF_IGNORE:
break;
default:
av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
hx->nal_unit_type, bit_length);
}
if (context_count == h->max_contexts) {
ret = ff_h264_execute_decode_slices(h, context_count);
if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
goto end;
context_count = 0;
}
if (err < 0 || err == SLICE_SKIPED) {
if (err < 0)
av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
h->ref_count[0] = h->ref_count[1] = h->list_count = 0;
} else if (err == SLICE_SINGLETHREAD) {
/* Slice could not be decoded in parallel mode, copy down
* NAL unit stuff to context 0 and restart. Note that
* rbsp_buffer is not transferred, but since we no longer
* run in parallel mode this should not be an issue. */
h->nal_unit_type = hx->nal_unit_type;
h->nal_ref_idc = hx->nal_ref_idc;
hx = h;
goto again;
}
}
}
if (context_count) {
ret = ff_h264_execute_decode_slices(h, context_count);
if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
goto end;
}
ret = 0;
end:
/* clean up */
if (h->cur_pic_ptr && !h->droppable) {
ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
h->picture_structure == PICT_BOTTOM_FIELD);
}
return (ret < 0) ? ret : buf_index;
}
再深入去看ff_h264_execute_decode_slices，会发现它直接调用了decode_slice来进行slice的解码。
/**
* Call decode_slice() for each context.
*
* @param h h264 master context
* @param context_count number of contexts to execute
*/
int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
{
AVCodecContext *const avctx = h->avctx;
H264Context *hx;
int i;
av_assert0(h->mb_y < h->mb_height);
if (h->avctx->hwaccel ||
h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
return 0;
if (context_count == 1) {
return decode_slice(avctx, &h);
} else {
av_assert0(context_count > 0);
for (i = 1; i < context_count; i++) {
hx = h->thread_context[i];
if (CONFIG_ERROR_RESILIENCE) {
hx->er.error_count = 0;
}
hx->x264_build = h->x264_build;
}
avctx->execute(avctx, decode_slice, h->thread_context,
NULL, context_count, sizeof(void *));
/* pull back stuff from slices to master context */
hx = h->thread_context[context_count - 1];
h->mb_x = hx->mb_x;
h->mb_y = hx->mb_y;
h->droppable = hx->droppable;
h->picture_structure = hx->picture_structure;
if (CONFIG_ERROR_RESILIENCE) {
for (i = 1; i < context_count; i++)
h->er.error_count += h->thread_context[i]->er.error_count;
}
}
return 0;
}那就再来看看decode_slice
static int decode_slice(struct AVCodecContext *avctx, void *arg)
{
H264Context *h = *(void **)arg;
……
if (h->pps.cabac) {
/* realign */
align_get_bits(&h->gb);
/* init cabac */
ff_init_cabac_decoder(&h->cabac,
h->gb.buffer + get_bits_count(&h->gb) / 8,
(get_bits_left(&h->gb) + 7) / 8);
ff_h264_init_cabac_states(h);
for (;;) {
// START_TIMER
int ret = ff_h264_decode_mb_cabac(h);
int eos;
// STOP_TIMER("decode_mb_cabac")
if (ret >= 0)
ff_h264_hl_decode_mb(h);
// FIXME optimal? or let mb_decode decode 16x32 ?
if (ret >= 0 && FRAME_MBAFF(h)) {
h->mb_y++;
ret = ff_h264_decode_mb_cabac(h);
if (ret >= 0)
ff_h264_hl_decode_mb(h);
h->mb_y--;
}
eos = get_cabac_terminate(&h->cabac);
if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
h->cabac.bytestream > h->cabac.bytestream_end + 2) {
er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
h->mb_y, ER_MB_END);
if (h->mb_x >= lf_x_start)
loop_filter(h, lf_x_start, h->mb_x + 1);
return 0;
}
if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", h->cabac.bytestream_end - h->cabac.bytestream);
if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
av_log(h->avctx, AV_LOG_ERROR,
"error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
h->mb_x, h->mb_y,
h->cabac.bytestream_end - h->cabac.bytestream);
er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
h->mb_y, ER_MB_ERROR);
return AVERROR_INVALIDDATA;
}
if (++h->mb_x >= h->mb_width) {
loop_filter(h, lf_x_start, h->mb_x);
h->mb_x = lf_x_start = 0;
decode_finish_row(h);
++h->mb_y;
if (FIELD_OR_MBAFF_PICTURE(h)) {
++h->mb_y;
if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
predict_field_decoding_flag(h);
}
}
if (eos || h->mb_y >= h->mb_height) {
tprintf(h->avctx, "slice end %d %d\n",
get_bits_count(&h->gb), h->gb.size_in_bits);
er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
h->mb_y, ER_MB_END);
if (h->mb_x > lf_x_start)
loop_filter(h, lf_x_start, h->mb_x);
return 0;
}
}
} else {
for (;;) {
int ret = ff_h264_decode_mb_cavlc(h);
if (ret >= 0)
ff_h264_hl_decode_mb(h);
// FIXME optimal? or let mb_decode decode 16x32 ?
if (ret >= 0 && FRAME_MBAFF(h)) {
h->mb_y++;
ret = ff_h264_decode_mb_cavlc(h);
if (ret >= 0)
ff_h264_hl_decode_mb(h);
h->mb_y--;
}
if (ret < 0) {
av_log(h->avctx, AV_LOG_ERROR,
"error while decoding MB %d %d\n", h->mb_x, h->mb_y);
er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
h->mb_y, ER_MB_ERROR);
return ret;
}
if (++h->mb_x >= h->mb_width) {
loop_filter(h, lf_x_start, h->mb_x);
h->mb_x = lf_x_start = 0;
decode_finish_row(h);
++h->mb_y;
if (FIELD_OR_MBAFF_PICTURE(h)) {
++h->mb_y;
if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
predict_field_decoding_flag(h);
}
if (h->mb_y >= h->mb_height) {
tprintf(h->avctx, "slice end %d %d\n",
get_bits_count(&h->gb), h->gb.size_in_bits);
if ( get_bits_left(&h->gb) == 0
|| get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
h->mb_x - 1, h->mb_y, ER_MB_END);
return 0;
} else {
er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
h->mb_x, h->mb_y, ER_MB_END);
return AVERROR_INVALIDDATA;
}
}
}
if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
tprintf(h->avctx, "slice end %d %d\n",
get_bits_count(&h->gb), h->gb.size_in_bits);
if (get_bits_left(&h->gb) == 0) {
er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
h->mb_x - 1, h->mb_y, ER_MB_END);
if (h->mb_x > lf_x_start)
loop_filter(h, lf_x_start, h->mb_x);
return 0;
} else {
er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
h->mb_y, ER_MB_ERROR);
return AVERROR_INVALIDDATA;
}
}
}
}
}至此就已经很清晰了，首先通过PPS判断是CABAC熵编码还是CAVLC熵编码，然后针对每一个宏块调用对应的熵解码函数进行解码，即ff_h264_decode_mb_cabac和ff_h264_decode_mb_cavlc，然后再进行宏块的解码，使用的是ff_h264_hl_decode_mb，帧内预测和帧间预测的内容也包含在其中，最后使用loop_filter进行去块效应滤波，这里还多了一个错误隐藏功能，使用的是er_add_slice。
对于本文的需求，分析到这一步就够了。再往里面深入的话，就需要各位静下心来，对照标准一点一点看了，网上也有很多相关的文章供参考。
ffmpeg源码修改与QP提取
通过前面对ffmpeg中H.264解码部分的源码分析，现在我们可以解决在文章最开始提到的实际问题了。
每一个宏块都有自己的QP值，将一帧中所有宏块的QP值集合到一起即得到QP table。直接定位到与帧解码对应的函数h264_decode_frame（libavcodec/h264.c），当成功解码得到一帧数据的时候，会将got_frame赋值为1，此时即可将解码得到的QP table赋值给作为输出的AVFrame。于是，修改后的代码如下，非常简单。需要注意的是，在flush buffer阶段也需要进行对应的QP table赋值操作。
static int h264_decode_frame(AVCodecContext *avctx, void *data,
int *got_frame, AVPacket *avpkt)
{
const uint8_t *buf = avpkt->data;
int buf_size = avpkt->size;
H264Context *h = avctx->priv_data;
AVFrame *pict = data;
int buf_index = 0;
H264Picture *out;
int i, out_idx;
int ret;
h->flags = avctx->flags;
ff_h264_unref_picture(h, &h->last_pic_for_ec);
/* end of stream, output what is still in the buffers */
if (buf_size == 0) {
out:
h->cur_pic_ptr = NULL;
h->first_field = 0;
// FIXME factorize this with the output code below
out = h->delayed_pic[0];
out_idx = 0;
for (i = 1;
h->delayed_pic[i] &&
!h->delayed_pic[i]->f.key_frame &&
!h->delayed_pic[i]->mmco_reset;
i++)
if (h->delayed_pic[i]->poc < out->poc) {
out = h->delayed_pic[i];
out_idx = i;
}
for (i = out_idx; h->delayed_pic[i]; i++)
h->delayed_pic[i] = h->delayed_pic[i + 1];
if (out) {
out->reference &= ~DELAYED_PIC_REF;
ret = output_frame(h, pict, out);
if (ret < 0)
return ret;
//get the qscale table, modified by zhanghui
if(out->qscale_table)
{
pict->qscale_table=out->qscale_table;
pict->qstride=h->mb_stride;
pict->qscale_type=FF_QSCALE_TYPE_H264;
}
*got_frame = 1;
}
return buf_index;
}
if (h->is_avc && av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, NULL)) {
int side_size;
uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, &side_size);
if (is_extra(side, side_size))
ff_h264_decode_extradata(h, side, side_size);
}
if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
if (is_extra(buf, buf_size))
return ff_h264_decode_extradata(h, buf, buf_size);
}
buf_index = decode_nal_units(h, buf, buf_size, 0);
if (buf_index < 0)
return AVERROR_INVALIDDATA;
if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
av_assert0(buf_index <= buf_size);
goto out;
}
if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
if (avctx->skip_frame >= AVDISCARD_NONREF ||
buf_size >= 4 && !memcmp("Q264", buf, 4))
return buf_size;
av_log(avctx, AV_LOG_ERROR, "no frame!\n");
return AVERROR_INVALIDDATA;
}
if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) ||
(h->mb_y >= h->mb_height && h->mb_height)) {
if (avctx->flags2 & CODEC_FLAG2_CHUNKS)
decode_postinit(h, 1);
ff_h264_field_end(h, 0);
/* Wait for second field. */
*got_frame = 0;
if (h->next_output_pic && (
h->next_output_pic->recovered)) {
if (!h->next_output_pic->recovered)
h->next_output_pic->f.flags |= AV_FRAME_FLAG_CORRUPT;
if (!h->avctx->hwaccel &&
(h->next_output_pic->field_poc[0] == INT_MAX ||
h->next_output_pic->field_poc[1] == INT_MAX)
) {
int p;
AVFrame *f = &h->next_output_pic->f;
int field = h->next_output_pic->field_poc[0] == INT_MAX;
uint8_t *dst_data[4];
int linesizes[4];
const uint8_t *src_data[4];
av_log(h->avctx, AV_LOG_DEBUG, "Duplicating field %d to fill missing\n", field);
for (p = 0; p<4; p++) {
dst_data[p] = f->data[p] + (field^1)*f->linesize[p];
src_data[p] = f->data[p] + field *f->linesize[p];
linesizes[p] = 2*f->linesize[p];
}
av_image_copy(dst_data, linesizes, src_data, linesizes,
f->format, f->width, f->height>>1);
}
ret = output_frame(h, pict, h->next_output_pic);
if (ret < 0)
return ret;
//get the qscale table, modified by zhanghui
if(h->next_output_pic->qscale_table)
{
pict->qscale_table=h->next_output_pic->qscale_table;
pict->qstride=h->mb_stride;
pict->qscale_type=FF_QSCALE_TYPE_H264;
}
*got_frame = 1;
if (CONFIG_MPEGVIDEO) {
ff_print_debug_info2(h->avctx, pict, h->er.mbskip_table,
h->next_output_pic->mb_type,
h->next_output_pic->qscale_table,
h->next_output_pic->motion_val,
&h->low_delay,
h->mb_width, h->mb_height, h->mb_stride, 1);
}
}
}
av_assert0(pict->buf[0] || !*got_frame);
ff_h264_unref_picture(h, &h->last_pic_for_ec);
return get_consumed_bytes(buf_index, buf_size);
}保存修改，重新make即可。也可以利用diff和patch工具把对源码的修改做成一个补丁，这样以后再下载了新的ffmpeg源码也可以直接通过patch将我们曾经的修改应用到新的源码中去，非常方便。
上面的应用实例非常简单，进行其他的修改也可以使用类似的思路，我这里算是抛砖引玉，以后还会讲述如何从ffmpeg中抽取代码等其他的ffmpeg源码应用实例。
本文的示例也存在一个小问题，经观察，在AVFrame中新增了AVBufferRef *qp_table_buf变量，应该也是用于存储QP table的，但是AVBufferRef这个结构体我还不太了解它是干什么用的、怎么用的，按理说如果能将QP table存到这个里面应该是更好的，但是我的尝试都失败了，希望有经验的朋友多多指教，谢谢！
关注公众号，掌握更多多媒体领域知识与资讯
文章帮到你了？可以扫描如下二维码进行打赏~，打赏多少您随意~
