| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
| |
Change access modifiers and add overridable methods in HLS stack.
Change-Id: Iae8e77246cc6643735af18617717fba713d0038c
|
|
|
|
|
|
|
|
| |
Do not remove audio only playlist. This will enable
the player to switch to Audio only from Audio Video and
vice versa.
Change-Id: I56b9245f3d28ef9f8e31651cc59b494b763f3feb
|
|
|
|
|
|
|
| |
`onChangeConfiguration2` is only called during seek.
Bug: 22698650
Change-Id: I715fa51d04d503f49d678eaea08f2b63dce4e01e
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- abort high bandwidth immediately when bandwidth is fluctuating
- use short-term bandwidth estimate for downswitch if bandwidth
is not stable
- discard bandwidth samples that's too old in absolute time
- if already underflow, switch to lowest bandwidth to catch up
- if error happened during bandwidth switch (likely due to new
variant link is broken), switch to lowest bandwidth to catch up
bug: 21754330
Change-Id: Ifd16d75e261cefb93b989829bf35a36783142ae0
|
|
|
|
|
| |
Bug: 21151892
Change-Id: I6a243b0edbbb445df0caf65f395f81926fd515f0
|
|
|
|
|
|
|
|
|
|
| |
- pause after the current block when select track
- ignore metadata timestamps as they're too sparse
- use smaller range when searching for next segment to prevent
resumeUntil from downloading too much data
bug: 20500732
Change-Id: Ibda57a39ec86efd96a8dd0db95adeb92d076697a
|
|
|
|
|
|
|
| |
disconnect HTTP connection when we absolutely won't resume
bug: 19890444
Change-Id: Idee36b48741f6f8eb1d65bca32156e9e18349c67
|
|
|
|
|
|
|
|
|
| |
The new getBufferedDurationUs implementation obsoletes the purpose of
getEstimatedDurationUs; remove getEstimatedDurationUs and its
associated member variables. Finally replace calls to
getEstimatedDurationUs with getBufferedDurationUs.
Change-Id: I38f20df8e177ffbfe299b203d99076fc98dcd274
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- fix a 'possible video time jump' after seek, don't update
renderer anchor time for 0-sized audio buffers
- fix another 'possible video time jump' caused by some states
not reset in mStreams
- bandwidth estimator changes to not remove more than 10%
of total transfer duration at a time to avoid jumping up
too quickly
bug: 20267388
related-to-bug: 19864613
related-to-bug: 20138395
Change-Id: I8812332cd1e26bf562acfaf086fd679a3549debc
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- account for playlist age in live streaming when calculating
segment time
- be more conservative on downswitching if bandwidth is unstable
- adjust forward or backward if guessed wrong seq number
- code refactor
bug: 19567254
Change-Id: I0b61cea888fdffd1b3ee2446747ed10152e9e7d7
|
|
|
|
| |
Change-Id: I00a8a786b3f4b74742c34770edd94e937abe20a8
|
|
|
|
|
|
| |
bug: 20160436
Change-Id: Ic3adb84d3c65cc65f62fc509a99d09602db862a1
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
to avoid having to immediately down switch (and pause)
after playback starts.
do not count "discard" packet when estimating duration.
bug: 19567254
Change-Id: I0cdd37a06ca800dd81a91cca5eb9b46a1eab7b20
|
|/
|
|
|
|
|
| |
Only consume reply ID when actually handling the seek.
Bug: 20123914
Change-Id: I2112ee1b89f8193b487ea2b0b3b7050ba3413864
|
|
|
|
|
| |
bug: 19567254
Change-Id: I4305d37cb74279ccd435f99483231cd1dcf42fc9
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- fix no target-duration case
- fix for audio-only <=> audio/video switching
- disable audio-only variants if there is at least
one variant with video
- fix mpeg2ts PTS wraparound when bandwidth adapting
- tweak up/down switch marks
bug: 19567254
Change-Id: Ib46144203c56dfc96eccd6ddaa3867e8a4f2c6a9
|
|
|
|
| |
Change-Id: Iff2ca5d1e800d30943de12191bfe6c43d6a2c7f6
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- when upswitching, discard excessive buffering on low
bandwidth variant, switch to new variant earlier
- when downswitching, report newly found IDR positions
continuously, and switch as soon as new fetcher passes
playback position. This allows us to skip time-consuming
resumeUntil() of old fetcher most of the time
- implement pause/resume on low buffering, and notify
buffering percentage
- buffering parameter tuning, separate pause/resume/ready
buffer level and up/down switch buffer level, boost up
fetcher buffering significantly
bug: 19567254
Change-Id: I750dfcc6f861d78d16a71f501beb86d8129cb048
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- when down switching, decide whether to finish current segment
based on bandwidth settings, abort current segment if needed.
- when switching, pause new fetcher after the first 47K chunk,
and go back to resume old fethcer to stop point immediately.
- when old fetcher reaches stop point, swap packet sources and
resume new fetcher.
- mark switching as done as soon as old fecther reaches stop
point. This allows us to resume bandwidth monitoring earlier,
and do subsequent switches sooner.
bug: 19567254
Change-Id: Iba4b5fb9b06541bb1e49592536648f5d4cbc69ab
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- separate bandwidth estimator from HTTPBase, so that we have
better control on which samples to use, it also allows bandiwdth
history across multiple HTTPBase objects (which we'll use later).
- use min buffer duration among the streams to decide whether to
download next segment.
- maintain constant buffer level, time next download to happen
when buffer just goes below kMinBufferedDurationUs.
bug: 19567254
Change-Id: I5c481ad1f7ff3f084d57ec68856e12ae6b40ce41
|
|
|
|
| |
Change-Id: I2239d875420f6926918c1a0dcab31b71c8329d1f
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- keep old fetcher when seeking, unless the URI is changing.
- when restarting after a seek, check discontinuity seq, and
queue format change if it's changed.
- add a simple kill switch to abort when stop (or pause for seek).
- when seeking, if searching for start time goes into 2nd segment,
do not signal time discontinuity or reset first PTS.
- use setFormat() to set format in AnotherPacketSource, otherwise
video/audio flags are not updated and format are not cleared on
discontinuities.
- do not start queueing video access unit until first IDR after start
bug: 19656539
Change-Id: I79108d26964f59ea00d2eeac8f5f9318747f8541
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
remove unnecessary time discontinuity
move fetcher to separate looper so that download won't
block LiveSession
poll buffering at 1 sec interval in LiveSession, and
switch bandwidth if necessary
use fixed 100ms threshold for resumeUntil
bug: 19567254
Change-Id: I911e5041364f0858b43f2312756e173db5870a1e
|
|/
|
|
|
|
|
|
| |
Change replyID-s from uint32_t to an object
Move reply handling into the loopers (to reuse a common mutex)
Bug: 19607784
Change-Id: Iaa035b846c424c5687ed17ce1079b325e86c54be
|
|
|
|
|
| |
Bug: 19607784
Change-Id: I94cddcb81f671422ad4982a23dc4acfe57a9f1aa
|
|\
| |
| |
| |
| |
| |
| | |
time in conjunction." into lmp-mr1-dev
* commit 'a291dabcab10cafc1749d1d9493d269049502256':
httplive: Set start time and segment start time in conjunction.
|
| |\
| | |
| | |
| | | |
lmp-mr1-dev
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Also add comments describing how start time and segment start time are
used.
Based on AOSP CL https://android-review.googlesource.com/127653
by Joakim Johansson <joakim.c.johansson@sonymobile.com> but uses the
lowest segment start time instead of highest.
Bug: 18821145
Change-Id: I14cf1186d0daf517a24e8423c3a708b4c9ba06c4
|
|\ \ \
| | | |
| | | |
| | | |
| | | | |
* commit '6a025acb630a3ac4a84715d188aeb48f1946bc3f':
Move AString's StringPrintf out of the way.
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
We should come back and replace AString with std::string and switch to the
"real" StringPrintf family, but this fixes the ODR violation that was
preventing us from booting.
Bug: 19265750
Change-Id: I798eb9ca5dd634e44625af5e583439ef9f0cdc35
|
|\ \ \ \
| | |/ /
| |/| |
| | | |
| | | | |
* commit 'e3ada5d580a32b0133ac3db881e1574af57cb4fc':
httplive: Defer switch down if a switch is in progress
|
| | |/
| |/|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Bandwidth switch down is triggered if the buffered duration in
any of the current packet sources is below a threshold. When a
switch is in progress, all the packet sources are drained until
they are empty or until stop time is dequeued. Hence buffered
duration keeps going down during switch. Defering check switch
down will avoid unnecessary switches.
Do not switch down if estimated bandwidth index is more than
the current one.
Bug: 18821145
Change-Id: I655a308462503cf9df10672ecd904a51b2cba691
|
|\ \ \
| |/ /
| | |
| | |
| | |
| | |
| | | |
bandwidth estimate
* commit 'ce25d85ad22e6df4b861d17e9e67cb6d0e62c363':
stagefright: httplive: Decouple block size from bandwidth estimate
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
A very small block size in PlaylistFetcher can lead to framework
overhead and difficulty streaming high bitrate content, but since
HTTPBase keeps a constant history of the past 100 HTTP reads, the
block size directly affects bandwidth estimation and in turn,
switching latency.
Add setBandwidthHistorySize() to HTTPBase to allow setting the
history size for bandwidth estimation. Call this within LiveSession
based on the current block size to ensure that the number of bytes
used for estimating bandwidth does not change if the block size is
changed in PlaylistFetcher.
Since a single TCP/IP packet can contain up to 64k of data, increase
the block size in PlaylistFetcher from 2k to lcm(188, 1024) or 47k to
avoid inaccuracies in read timings due to up to a comparable 47 reads
from the same locally-cached packet instead of from the network.
Also make HTTPBase::addBandwidthMeasurement() virtual to allow
bandwidth estimation extensions that do not rely on a history list.
Bug: 18821145
Change-Id: I5f957be01f5346e74cfb7eeb150ca4b397ad5798
|
|\ \ \
| |/ /
| | |
| | |
| | |
| | |
| | | |
chunked content
* commit '0512881b08d03d10d6f164566c9a787d2f56ab6d':
stagefright: httplive: Reduce memcpy calls for chunked content
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Streams using http chunking will not report the segment's total
content-length. In this case, a 64k buffer is allocated and is
increased by 32k each time the buffer is filled again. For high
bitrate content, this can lead to a large number of copies that
affect the HLS framework delay. Increase fetchFile buffer size
exponentially by 50% or at least 32k instead of by 32k each time
to reduce the number of memcpy calls.
Example for a chunked 6 MB 1080p segment (ie ~3s):
Adding 32k:
190 copies with 572.97 MB copied
Increasing by 50%:
12 copies with 16.09 MB copied
Bug: 18821145
Change-Id: Iedf0e4437e96026a58d50bce2660f85ac90d0ada
|
|\ \ \
| |/ /
| | |
| | |
| | |
| | |
| | | |
duration clips
* commit '958a54322ea3ff2ad8ed0ac6e229c90c638f8a7f':
stagefright: httplive: Fix deadlock for low duration clips
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
PlaylistFetcher buffers up to 3 * target-duration bytes of data,
but if a stream is slow (ie due to bad network conditions), a
buffer threshold of 10s is used to resume playback. This results
in an indefinite freeze as PlaylistFetcher has stopped buffering
before this threshold. Reduce the 10s threshold to be more in-sync
with PlaylistFetcher's buffering size.
Bug: 18821145
Change-Id: Ife846e7c5b4f9645895873d08250c4bee0164972
|
|\ \ \
| |/ /
| | |
| | | |
Change-Id: I39fdc2e8895e1e943749b9a2628656a8fa5bb72b
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
This is to restore patch attributions
This reverts commit f580806d893c4631f5324ff0af5c2db68a40ef42.
Bug: 18821145
Change-Id: Idc49385fffccfde2a3915388fe3fe4e2b740d787
|
|\ \ \
| |/ /
| | |
| | | |
Change-Id: I4313941f3561176ce9f6ab055678fb626e570107
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
This commit consists of:
http://go/pag/c/188753 Add NULL check for empty playlist
http://go/pag/c/188754 Fix deadlock for low duration clips
http://go/pag/c/188757 Create a copy of last enqueued metadata
http://go/pag/c/188755 Propagate target duration to LiveSession
http://go/pag/c/188762 Decouple block size from bandwidth estimate
http://go/pag/c/188756 Reduce memcpy calls for chunked content
http://go/pag/c/188758 Dont resume if we have almost fetched till stop time
Bug: 18821145
Change-Id: I7fd650999c6c50bbadffd65adee9020e669dfe62
|
|\ \ \
| |/ /
| | |
| | |
| | |
| | |
| | | |
client when the source is not seekable." into lmp-mr1-dev
* commit 'a0b3a0a46dc42eafe620ffd053604515bbd9ca9a':
NuPlayer: send NOT_SEEKABLE media info to client when the source is not seekable.
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
seekable.
LiveSession: return -1 for duration when it's not available.
Bug: 18599325
Change-Id: Iecd040f48750806f98d1799e2aaab2f90c6f3887
|
|\ \ \
| |/ /
|/| /
| |/ |
Change-Id: Iea12c8a6cabf84584e4a89ad80e298c1f4ea3dd7
|
| |
| |
| |
| |
| |
| |
| | |
For build-system CFLAGS clean-up, remove unused functions and
variables.
Change-Id: Ic3dee56b589ea9a693efa1d72ba394036efff168
|
|\ \ |
|
| |/
| |
| |
| |
| | |
Bug: 17514665
Change-Id: I81c62553f2c5acb4d2436a9d8f04c10fdbe315d0
|