Remove VideoDecoderConfig.coded_size |
|||
Issue descriptionDetection of coded size needs to be done elsewhere other than container parsers, especially now that (in M56 ffmpeg roll) ffmpeg no longer allows access to the parsed-from-codec internal AVCodecContext in an AVStream. Rather than spinning up a temporary ffmpeg s/w decoder just to extract that information again in OnFindStreamInfoDone, we now (https://codereview.chromium.org/2506533002) default coded_size to just width and height as *hints* to the decoder. Outside of src=, ote that MSE MP4 parser has always just used the width, height for coded size in similar construction of VideoDecoderConfigs in MP4StreamParser. MSE mp2t and WebM stream parsers appear to populate coded size more precisely. This bug tracks removing the VideoDecoderConfig.coded_size field, since it's now containing less precise information than what decoders might need.
,
Nov 15 2016
From offline chat, I'll take this for now.
,
Nov 16 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/29241f214adfd993990b409c8f42dca858775dc9 commit 29241f214adfd993990b409c8f42dca858775dc9 Author: wolenetz <wolenetz@chromium.org> Date: Wed Nov 16 00:07:26 2016 Use width,height for coded_{width,height} in AVStreamToVideoDecoderConfig Anticipating AVStream.codec->coded_{width,height} will be inaccessible in ffmpeg soon, this change uses just the width and height values as hints of the coded size in AVStreamToVideoDecoderConfig (used only in FFmpegDemuxerStream creation and tests). Once AVStream.codec becomes deprecated, coded size fields are not externally populated unless we spin up a temporary ffmpeg decoder to obtain that information. It also drops the metadata logging in OnFindStreamInfoDone for coded_size, since that metadata is no longer different than just width and height, which are already logged. Note that MSE MP4 parser has always just used the width, height for coded size in similar construction of VideoDecoderConfigs in MP4StreamParser. MSE mp2t and WebM stream parsers appear to populate coded size more precisely. Marks VideoDecoderConfig.coded_size() and related histograms.xml entries as deprecated. BUG= 591845 , 665539 TEST=No media_unittests, ffmpeg_regression_tests regression under LSAN on linux Review-Url: https://codereview.chromium.org/2506533002 Cr-Commit-Position: refs/heads/master@{#432300} [modify] https://crrev.com/29241f214adfd993990b409c8f42dca858775dc9/media/base/video_decoder_config.h [modify] https://crrev.com/29241f214adfd993990b409c8f42dca858775dc9/media/ffmpeg/ffmpeg_common.cc [modify] https://crrev.com/29241f214adfd993990b409c8f42dca858775dc9/media/filters/ffmpeg_demuxer.cc [modify] https://crrev.com/29241f214adfd993990b409c8f42dca858775dc9/tools/metrics/histograms/histograms.xml
,
Jan 23 2018
|
|||
►
Sign in to add a comment |
|||
Comment 1 by wolenetz@chromium.org
, Nov 15 2016