Add comparison check so that NaN is caught.
Comparisons with NaN always result in false.
NaNs cause problems later in audio encoder.
Note: may break with -ffast-math compiler flag.
Fixes#4885.
Since [1], avcodec/version.h is not included anymore in codec.h and
therefore is not included any more in avformat.h.
As a result, LIBAVCODEC_VERSION_INT is no longer defined. This commit
fixes the include.
Since obviously we can't ifdef the avcodec include by referring to an
avcodec version, we ifdef it with the avformat version which was bumped
at the same time [2].
[1] libavcodec: Split version.h
f2da2e1458
[2] doc: Add an entry to APIchanges about changes to version.h and
version_major.h
f3a0e2ee2b
Signed-off-by: pkv <pkv@obsproject.com>
Prior to this change, the audio thread roughly waits the time of
`AUDIO_OUTPUT_FRAMES`, and consecutively processes twice or more if the
latency is accumulated. This behavior causes fluctuation of timing to
output audio packets. This change introduces `os_sleepto_ns_fast` so
that the audio packets will be periodically output.
Rec. 2020 is really an SDR spec, but I think HDR10 made it okay to slap
PQ on it, call it an HDR spec. Rec. 2100 came along after and formally
allowed the use of PQ/HLG, so we should use 2100 instead.
When playing video, OBS can overflow and crash, or render video edges
incorrectly if the video resolution does not divide into 2 for trivial
chroma subsampling. This is fixed by rounding chroma plane sizes up
instead of down, and maintaining that through the video pipeline.
Similar to 315fbd6 and 6071098, ignore non-fatal errors during the
remux process. It's possible OBS itself writes a file that generates
these errors when remuxed.
As os_gettime_ns() gets large the current scaling methods, mostly by casting
to uint64_t, may lead to numerical overflows. Sweep the code and use
util_mul_div64() where applicable.
Signed-off-by: Hans Petter Selasky <hps@selasky.org>
sse-intrin.h is a required header now, but the implicit path
breaks building addons once the headers are installed.
Fix this by making the path explicit.
Code submissions have continually suffered from formatting
inconsistencies that constantly have to be addressed. Using
clang-format simplifies this by making code formatting more consistent,
and allows automation of the code formatting so that maintainers can
focus more on the code itself instead of code formatting.
Fixes handling of the `obs_source_frame::full_range` member variable,
which is often set to false by default by many plugins even when using
RGB, which would cause RGB to be marked as "partial range". This change
is crucial for when partial range RBG support is implemented.
Adds `obs_source_frame2` structure that replaces the `full_range` member
variable with a `range` variable, which uses the `video_range_type` enum
to allow handling default range values. This member variable treats
VIDEO_RANGE_DEFAULT as full range if the format is RGB, and partial
range if the format is YUV.
Also adds `obs_source_output_video2` and `obs_source_preload_video2`
functions which use the `obs_source_frame2` structure instead of the
`obs_source_frame` structure.
When using the original `obs_source_frame`, `obs_source_output_video`,
and `obs_source_preload_video` functions, RGB will always be full range
by default for backward compatibility purposes.
A mono source is currently upmixed by swresampler in the following way:
- for stereo output, FL=FR=input/sqrt(2)
- for other speaker layouts of the outputs, FC=input, other channels
are zeroed.
In the case of stereo output, this leads to a 3dB level decrease which
users have issue with [1].
The obvious fix of adding a 3dB gain is reported to be adding distortions
on some setups [2].
Note that the "Downmix to Mono" does not fix this upmix problem, since
it just makes all output channels identical by summing all input channels
and normalizing (by dividing by the number of output channels). This last
normalization step results in a level reduction for a mono input.
[1] This fixes https://obsproject.com/mantis/view.php?id=960.
[2] See also: https://obsproject.com/forum/threads/please-allow-for-mono-recording-of-microphones-ill-explain-why.84834
Normally, the total and skipped frame count for the encoder is performed
in the video-io thread. However, because a new thread is being
introduced for texture-based encoding, the frontend has no way of being
able to query that. So, instead of making the frontend query that data
separately, just make the texture encoder thread increment the values of
video-io. That way, the frontend doesn't need to change any code, and
can continue using the same functions for determining the total/skipped
frame count.
This can cause the frame count to be doubled if both a texture-based
encoder and a raw data encoder is active at the same time, but it's an
acceptable alternative.
(This commit also modifies the obs-ffmpeg module)
The default channel layouts from aac spec are implemented in FFmpeg
native aac encoder as follows:
AV_CH_LAYOUT_MONO,
AV_CH_LAYOUT_STEREO,
AV_CH_LAYOUT_SURROUND,
AV_CH_LAYOUT_4POINT0,
AV_CH_LAYOUT_5POINT0_BACK,
AV_CH_LAYOUT_5POINT1_BACK,
AV_CH_LAYOUT_7POINT1,
The correspondence of speaker layouts to AV_CH_LAYOUT from FFmpeg is
changed to reflect the previous table.
Although FFmpeg native aac encoder can now encode all the layouts listed
in avutil channel_layout.h (on master), there might be issues with older
FFmpeg binaries.
Note that 2.1 speaker layout will be encoded as AV_CH_LAYOUT_SURROUND
(FL FR FC) because it is not listed as the default layout for three
channels.
This just means some optimizations for LFE channel will not be used by
the encoder which will treat it as an SCE (single channel element).
Closesjp9000/obs-studio#1182
