Fixes ticket 1070.
See also
https://obsproject.com/forum/threads/ffmpeg-recording.77378/#post-330473
(related bugs).
The ffmpeg constant AVFMT_RAWPICTURE was deprecated in october 2015
and marked for removal at avformat major bump to version 58
(ffmpeg commit 34ed5c2 , oct 12, 2015).
The bump occured with commit 69b5ce6 (oct21, 2017).
The constant was subsequently removed (commit 693a11b, oct 26 2017).
It was removed from obs-studio with commit d670d7b (from me).
But the code block which was executed with this constant was not
removed, causing issues with ffmpeg output.
The commit fixes the issue for old ffmpeg builds as well as new ones.
The constant is reintegrated for avformat major version < 58 and removed
for version >= 58 (along with its accompanying code).
Thanks to J Lowe for help in solving the bug.
(tested on win 10, macos 10.13, ubuntu 17.10 with ffmpeg head & ffmpeg
3.4.1)
If the target process re-creates its D3D context, the game capture tick
can trigger before the capture is setup, in which case OBS gets a
CAPTURE_RETRY message. However with the memory capture method, it
continues to try and copy from the shared memory pointer which is no
longer valid, resulting in a crash. The fix uses the old texture until
the next tick at which point the new capture should be ready for use.
Code Changes:
- Removed flag BUILD_AMF_ENCODER.
- AMF SDK is now a submodule of the plugin, no longer requiring extra steps.
Plugin Changes:
- 'Bitrate.Target' has been renamed to 'bitrate' internally, improving support for "New Networking Code" and "Replay Buffer" which incorrectly rely on this value instead of taking the average bitrate of packets sent over the last second.
- Drivers with a runtime older than 1.4.6.0 are blacklisted now.
- All hidden and experimental options have been removed.
Windows SDK 10.0.16299.0 defines these structures as part of winternl.h
but using different types and names. Unfortunately there's no macro to
detect the SDK version, so to avoid conflicting with newer / older SDKs
the OBS structs have been renamed.
When this was being fixed up, the incorrect function name was used --
however it still compiled because the author was using the newer FFmpeg
version at the time.
Common multi-channel setup is 5.1 and 7.1 with rear speakers.
Thus only setups that include SPEAKER_SIDE_LEFT and SPEAKER_SIDE_RIGHT
needs the marking as not common (or "side" use), while it stays the
true Side setup (with side speakers) by its internal meaning.
This "side" is named "surround" by Microsoft. To not confuse users and
translators, it is wise to use "Side" mark next to format name.
Certain NVIDIA GPUs don't support NVENC, but ship with the NVENC
library, causing OBS to mistakenly think that NVENC is available when it
actually isn't.
Closesjp9000/obs-studio#1087
Implement automatic audio device selection for devices that use two
separate DirectShow filters for audio and video instead of having a
single filter with audio and video output pins.
Please note that this fix is currently only active for Elgato USB and
PCIe devices (e.g. Cam Link, HD60 S, HD60 Pro, 4K60 Pro) to avoid
unintentionally changing the behavior for any other devices (e.g.
webcams).
(Jim edit: This fixes an issue with newer Elgato devices where the
devices would not automatically have their audio coupled with the video;
users would have to manually select the audio device in order to get
audio functioning.)
Closesjp9000/obs-studio#1081
(This commit also modifies the following modules: UI,
deps/media-playback, coreaudio-encoder, decklink, linux-alsa,
linux-pulseaudio, mac-capture, obs-ffmpeg, obs-filters, obs-libfdk,
obs-outputs, win-dshow, and win-wasapi)
Adds surround sound audio support to the core, core plugins, and user
interface.
Compatible streaming services: Twitch, FB 360 live
Compatible protocols: rtmp / mpeg-ts tcp udp
Compatible file formats: mkv mp4 ts (others untested)
Compatible codecs: ffmpeg aac, fdk_aac, CoreAudio aac,
opus, vorbis, pcm (others untested).
Tested streaming servers: wowza, nginx
HLS, mpeg-dash : surround passthrough
Html5 players tested with live surround:
videojs, mediaelement, viblast (hls+dash), hls.js
Decklink: on win32, swap channels order for 5.1 7.1
(due to different channel mapping on wav, mpeg, ffmpeg)
Audio filters: surround working.
Monitoring: surround working (win macOs linux (pulse-audio)).
VST: stereo plugins keep in general only the first two channels.
surround plugins should work (e.g. mcfx does).
OS: win, macOs, linux (alsa, pulse-audio).
Misc: larger audio bitrates unlocked to accommodate more channels
NB: mf-aac only supports mono and stereo + 5.1 on win 10
(not implemented due to lack of usefulness)
Closesjp9000/obs-studio#968
When device timing is used, it shouldn't be modifying the timestamp.
Fixes an issue where certain devices with large audio segments would
seem a bit out of sync.
(This commit also modifies the deps/media-playback, obs-ffmpeg, and
win-dshow modules)
More fixes due to ffmpeg renaming some constants and deprecating
AVFMT_RAWPICTURE and AV_PIX_FMT_VDA_VLD.
Latter replaced by AV_PIX_FMT_VIDEOTOOLBOX per ffmpeg dev advice.
Closesjp9000/obs-studio#1061
The new code in 3032535f56 would signal that the output has stopped to
the back-end and front-end, but the event used in the outputs themselves
to shut down the send thread would still be signaled, causing the next
connection to immediately stop as soon as it had started. This fixes it
so that the event does not get signaled unless the thread is active.
Use unbuffered async mode by default, and when in unbuffered mode,
decouple audio/video so that audio plays as soon as it's received.
This is a workaround for decklink device drivers having unreliable
video/audio timestamps (audio/video sync drifting over time). From
testing, it seems that the handling of video and audio is completely
separate in the driver; along with the timestamp calculations. For
example, when the thread of the decklink audio callback is stalled, it
would cause the timestamps of the audio alone to go out of sync, which
indicates timestamps are calculated more or less on the spot independent
of what video is doing (which is how we replicated the issue fixed by
b63e4b055e). Because decklink drivers treats the audio and video as
essentially decoupled, we must also treat it as decoupled. This is what
was causing video/audio to drift out of sync over time.
(This commit also modifies UI)
Instead of pinging Twitch every time the program starts up, only pings
for new servers when the ingests are actually being used, and when the
UI uses the auto-configuration dialog.
If ingests have not been cached when using the "Auto" server, it will
wait for 3 seconds max to query the Twitch ingest API. If it takes
longer than 3 seconds or fails, it will defer to SF. If ingests were
already cached, then it will use the existing cache immediately.
Video playback doesn't work if the default format is MJPEG and there are
other formats to use; this is because the useDefaultConfig variable is
still set to true, which overrides the format value that would normally
tell it to convert to RGB.
"Life is Feudal: Your Own" will use Direct3D to render the game, then
OpenGL to render its in-game menus, which causes a conflict with itself.
This specifically blacklists the game from capturing OpenGL to prevent
that from happening.
It would appear that the timestamp values returned by devices are not
perfectly accurate, and in some cases may be calculated on the spot --
to combat that, it's best to subtract the audio segment's duration from
the current audio segment's timestamp to ensure the timing is as
accurate as possible.
The end of an FLV tag would contain the size of the tag, but the code
was erroneously including the end size value in addition, which it's not
supposed to do normally.
(This commit also modifies the obs-outputs module)
The first video packet video offset (the value used to set the starting
point of video data) would be set to the DTS value of the first video
packet. However, when b-frames are used, the first DTS value will be
negative. This was originally done because FLV muxing requires that the
first packet's DTS start from 0. Unfortunately, this would also
effectively cause the first packet's PTS/DTS value to be shifted forward
by the negative amount, which would cause video sync to be off by a
video frame or two.
This fixes it to start at the PTS value instead and preserve any
negative offsets. Additionally, the FLV muxing code has been fixed to
ensure that it adjusts the starting video DTS to 0, and now correctly
adjusts the first audio packet's timestamp according to that DTS as well
(which it didn't do before).
The "Custom" device entry was added inside the loop that runs for each
device; this causes "Custom" to be added once for each device detected,
or not at all if no device was detected. Moving this outside the loop
(as would appear to have been the original intent, judging by the
indentation) fixes this by adding the entry only after all devices have
been processed.
(This commit also modifies deps/media-playback)
Before, the media-playback library would detect whether something was
seekable by checking the filename for "://", which is unideal because
there are other cases where targets may not be seekable. So instead, an
explicit "seekable" property (off by default) is now in place in the
media source when not in "local file" mode. Seeking will only be
enabled if local file mode is on, or if "seekable" is explicitly checked
by the user.
Closesjp9000/obs-studio#1022
Add option in properties that let you choose how audio is mixed during
transition:
- Fade Out/Fade In (existing behavior, default)
- Crossfade
Closesjp9000/obs-studio#1028
Some encoders such as the AMD AMF encoder don't set their bitrate
property, so it gets returned as 0. This causes the new network code to
allocate a tiny buffer, resulting in output starvation and throttling
to the point of uselessness if low latency mode is also enabled.
The ftl output gets very spammy (3-6 lines every 5 seconds),
this commit changes the loglevel of those messages to debug
so they do not clutter up the logs of users.
In case these messages are needed they can be re-enabled using
the '--verbose' flag.
(also fixes minor formatting issues)
Some window managers like i3 don't issue Expose events when switching
between workspaces, but send VisibilityNotify events. This causes the
stream to freeze even when the workspace is switched back.
Related issue: #863
VFR encoding mode was originally added experimentally, but in truth,
it's nothing but a detriment because it can cause wild bitrate
fluctuations. It's still a constant framerate even when VFR mode is on
anyway.
When the slideshow is deactivated (not shown to the stream), makes it so
that it can either continue playing regardless, pause, or restart the
slideshow when the source is activated (shown to the stream) again.
Split from jp9000/obs-studio#976
With the newly released Driver 17.7.2, AMD fixed many reported issues and added some much wanted features:
- Pre-Pass and VBAQ are finally working,
- H265/HEVC content can distinguish between IDR- and I- Frames
- and also supports Color Range and Color Space now,
- and the 32-bit only text log crash was fixed.
And now it's time for the plugin to also be updated! This release primarily aims at adding support for AMF 1.4.4 and improving support for older drivers, with some of the recently discovered bugs and crashes fixed.
Hotfix 1: Fixed VBAQ and Pre-Pass being used with Constant QP, causing significant corruption issues and changed some default values for H265.
## Changelog
### 2.2.1
* Fixed a frame corruption bug with Constant QP caused by toggling VBAQ and Pre-Pass after it had already been applied.
* Changed the default value of 'Enforce HRD' to 'Enabled' for H265/HEVC encoding.
* Changed the default value of 'Deblocking Filter' to 'Disabled' for H265/HEVC encoding to fix corruption in strict decoders. (Premiere Pro, After Effects, Vegas, VLC (Software))
### 2.2.0
* Added support for AMF 1.4.4 and improved support for older drivers like 16.11.5 and 16.12.1.
* Fixed a memory corruption crash due to misuse of managed DirectX 11 objects.
* Fixed a frame corruption bug caused by applying 'VBAQ' and 'Pre-Pass' when using 'Constant QP' Rate Control Method.
* Fixed a string comparison bug causing Simple Output Mode to always use the Profile 'Constrained Baseline'
* Fixed a crash when enabling 'Debug' on 32-bit OBS caused by different type sizes.
* Fixed a crash with H265/HEVC caused by use of the undocumented Intra-Refresh properties.
* Changed the default value for 'Profile' to 'High'.
* Changed the default value for 'VBAQ' to 'Disabled' for improved performance.
* Changed the default value for 'Queue Size' to 8 for improved stability in spontaneous load situations.
* Changed the log crash work around to only apply to older AMF versions than 1.4.4.
* Removed unnecessary configuration changes.
* Removed unsupported Pre-Pass options 'Half Size' and 'Quarter Size'.
Updates ingests via Twitch's ingest API. The servers are ordered by
closest to farthest from the user via the API. It's probably still good
to keep the services.json Twitch ingests updated in case there are
issues with the Twitch ingest API, but otherwise the ingests will update
from Twitch itself every time the program is opened automatically.
The user agent for the obsproject.com ingest update request would have
the version string baked in to the plugin, so if the plugin library
wasn't updated with the rest of the project, it would return the
incorrect version. This makes sure it returns the correct version by
requesting the version string from libobs itself.
Allows using a video file as a means of transitioning, transitioning two
targets at a specific time during the video playback. Audio for target
A fades out to the transition point, and then audio for target B fades
in after the transition point.
The invalid format version warning for service files is an unnecessary
warning which isn't necessary to display because it'll automatically
fall back to the distributed version over the cached remote version.
It's erroneously including an entire directory, including a file that no
longer exists, and the obs-outputs_HEADERS variable had a typo when used
in the add_library call.
Allows the ability for services to specify a different output if needed.
NOTE: This should probably be considered temporary, and services within
this file that use this functionality should probably be moved out of
the file and in to a separate service.
Delayed reading the text file till the update after the timestamp
changes, this should fix a minor issue where if the file updates twice
in one second, only the first change it displayed.
There's no need to display this property at the moment, the default
amount is more than sufficient for most cases. That and most people
wouldn't know what to do with it anyway.
This version fixes the code applying the streaming services settings, which would incorrectly always be applied due to left over obs_data_set_* lines. It also adds a safety check into SetIDRPeriod for H264 that fixes being able to set an IDR Period of 0 which is simply not supported (we would never be able to stop streaming or recording).
In the media source, do not pre-cache frames if the source is set to
close the file when inactive because that setting is designed to allow
the file to be replaced by the user. If it's replaced, it can
unintentionally keep the old precache frame, playing the frame from the
older video when it starts up.
When hooking DXGI-based graphics programs in the hook, the first hook
point is IDXGISwapChain::Present. To be able to initiate a capture, a
pointer to the device context that created the swap chain is required,
which can be retrieved via IDXGISwapChain::GetDevice. Determining
whether the device context was D3D10 or D3D11 has always been somewhat
of an issue due to D3D10 and D3D11 being nearly identical, as well as
their interoperability/interchangeability. The GetDevice function would
first be called with the UUID of ID3D10Device, then if that failed,
the UUID of ID3D11Device.
However, with certain specific D3D11 games, GetDevice would for some
unknown reason succeed with the UUID of ID3D10Device, which would cause
capture to fail. (Conversely, attempting to call GetDevice with the
UUID of ID3D11Device on a device that's actually ID3D10Device would
always succeed, so reversing the order of the test was not an option).
There were originally three known D3D11 games that would erroneously
succeed when querying a D3D10 device interface: Call of Duty: Ghosts,
Just Cause 3, and theHunter: Call of the Wild. All other known D3D11
games would work correctly. Because it was only these three games, a
hack was originally implemented in the form of an executable exception
list that would force them to capture D3D11.
Unfortunately, Oculus games are now failing under the same circumstance
as well, so a simple hack will no longer work. To fix this, a more
reliable method of detecting which context it is had to be discovered:
simply check the feature level using ID3D11Device::GetFeatureLevel. If
the feature level is D3D_FEATURE_LEVEL_11_0 or D3D_FEATURE_LEVEL_11_1,
the device is definitely a D3D11 device. Otherwise, continue the tests
as they were before. Successfully tested with many D3D10 games, many
D3D11 games, and especially those three D3D11 games that previously were
detected as D3D10 erroneously.
Preloading is designed to overwrite the current internal render texture
of the source so it'll be ready to play the first frame right when the
source is first displayed.
However, When the media source is set to pause on the last frame, it
keeps that current render texture visible, so preloading it with the
first frame will essentially overwrite the current render, causing it to
inadvertently show and pause on the first frame rather than the last
frame. So instead, just disable preloading when the user has it set to
pause on the last frame.
(Note: This commit also modifies deps/media-playback)
Allows buffering network-based media sources where supported. Default
is two megabytes of buffering.
Due to the recent renaming of hitbox to smashcast, old settings for
hitbox would not be shown in the service UI. This change preserves the
user's service settings even if the service name has been changed, shows
the setting and marks it as invalid to the user, and prevents the UI
from selecting a value that doesn't match the user's last selection.
Due to a noticeable frequency of crashes inside of FFmpeg when using
hardware encoding on mac, this feature is going to be disabled for now
pending more investigation at a later time.
Adds an option that allows the user to specify the size (or aspect) of
the bounding source size which all the images are then scaled in to.
Acts similarly to "bounding box" of scene items, but for the slide show
and its sub-sources.
(This commit also modifies the decklink, linux-v4l2, mac-avcapture,
obs-ffmpeg, and win-dshow modules)
Originally, async buffering for sources was supposed to be a
user-controllable flag. However, that turned out to be less than ideal
because sources (such as the win-dshow plugin) were programmed with
automatic control over their buffering (such as automatically detecting
USB 2.0 capture devices and then enabling in those cases).
The fact that it was a flag caused a design flaw to where buffering
values would be overwritten when a source is loaded from save data.
Because of that, this flag is being deprecated and replaced with a
specific function to enable unbuffered mode instead.
Remove two unused variables that are causing a compiler warning on gcc.
The size was probably added in the past to calculate the the packet
size, but in the end was not actually needed.
Fix a warning from gcc about one of the const keywords being ignored.
This happens because the returned type is not a reference contrary to
the methods above.
x264 on ultrafast won't drop any b-frames, so it might buffer packets a
bit too much in that case.
The frame drop duration calculation now searches for the first valid
video frame within the packets buffer. Additionally, it no longer
prevents frame dropping for a specific time after dropping frames,
because the first case eliminates the need for that.
On macs, some fonts can be in sub-folders of the font folder, and may
not properly be found. This fixes that to detect directories and make
the lookup recursive.
* Supports both H264/AVC and H265/HEVC encoding.
* Fixed several crashes and bugs that plagued users.
* Massively reduced CPU usage caused by the plugin.
* Added settings for custom I/P/B Interval.
* Added settings for custom Frame Skipping.
Clears/reinitializes RTMP structure to defaults, preventing data from
potentially persisting between different connections.
Closesjp9000/obs-studio#878
Changes "class" prioritization to attempt to find the window either with
the same title, or the next window of the same window class (window
type), changes "title" prioritization to only find the window based
upon its title, and changes "executable" prioritization to attempt to
find the window with the same title, or the next window of the same
executable.
Additionally changes the text associated with these selections to
clarify that functionality to users.
(Note: This commits also modifies the linux-pulseaudio, mac-capture, and
win-wasapi plugins)
Do not prevent the targeted output device from being monitored if the
selected monitor output device is a different one.
Closesjp9000/obs-studio#872
This happens because the enum had the incorrect name, and microsoft
automatically treats all enums as integers in C, regardless of whether
they actually exist or not. Microsoft makes terrible compilers and
whoever decided this was a good idea should be fired.
When the windows video device source source is set to only activate when
showing, it would still activate on first startup of the program even if
it was in another scene and not showing anywhere to the user. This
fixes that issue.
The "default" preset is not an actual default, but something Nvidia
decided to just call that way. It yields the worst quality per bitrate
out of all the presets, for no actual benefits. The actual FFmpeg
default is the hq one, which yields the best quality, especially when
twopass mode is enabled.
I can't think of a way to keep the "default" preset in a non-confusing
way, and as it gives no known benefits, might as well just remove it
entirely.
(Jim edit: Also made it so that if the settings have it set to
"default", it automatically treats it as "hq")
Closesjp9000/obs-studio#865
The code to process GIF animations contains a return statement that
skiped the file modification check, if the source is not in an active
scene. Like that images in the preview scene of studio mode would never
update, when they get modified on disk.
Also reloading the image, if an animation file has been modified before
processing the old animation that will be replaced later anyway, seems
like a better way to do it.
The LGP issue is caused by the device drivers returning two or more
packets in a single segment of audio data. This fixes it by detecting
that and decoding subsequent packets.
When the FFmpeg audio decoder returns, it returns how many bytes of data
was decoded. To have it decode multiple packets in a single segment,
just subtract the return value from the expected size, and if that size
is still larger than zero, then there are more packets in the segment to
decode. Otherwise, stop.