[NNStreamer] Gstreamer 기본 튜토리얼 7
Goal
- How to create new threads of execution for some parts of the pipeline
- What is the Pad Availability
- How to replicate streams
Introduction
Multithreading
GStreamer is a multithreaded framework. This means that, internally, it creates and destroys threads as it needs them, for example, to decouple streaming from the application thread. Moreover, plugins are also free to create threads for their own processing, for example, a video decoder could create 4 threads to take full advantage of a CPU with 4 cores.
On top of this, when building the pipeline an application can specify explicitly that a branch (a part of the pipeline) runs on a different thread (for example, to have the audio and video decoders executing simultaneously).
This is accomplished using the queue element, which works as follows. The sink pad just enqueues대기열에 넣다 data and returns control. On a different thread, data is dequeued대기열에서 제외 and pushed downstream. This element is also used for buffering, as seen later in the streaming tutorials. The size of the queue can be controlled through properties.
The source is a synthetic audio signal (a continuous tone) which is split using a tee element
(it sends through its source pads everything it receives through its sink pad). One branch then sends the signal to the audio card, and the other renders a video of the waveform and sends it to the screen.
As seen in the picture, queues create a new thread, so this pipeline runs in 3 threads. Pipelines with more than one sink usually need to be multithreaded, because, to be synchronized, sinks usually block execution until all other sinks are ready, and they cannot get ready if there is only one thread, being blocked by the first sink.
Request pads
In Basic tutorial 3: Dynamic pipelines we saw an element (uridecodebin
) which had no pads to begin with, and they appeared as data started to flow and the element learned about the media. These are called Sometimes Pads
, and contrast with the regular pads which are always available and are called Always Pads
.
The third kind of pad is the Request Pad
, which is created on demand. The classical example is the tee element, which has one sink pad and no initial source pads: they need to be requested복제 and then tee
adds them. In this way, an input stream can be replicated any number of times. The disadvantage is that linking elements with Request Pads is not as automatic, as linking Always Pads, as the walkthrough for this example will show.
Also, to request (or release) pads in the PLAYING or PAUSED states, you need to take additional cautions (Pad blocking) which are not described in this tutorial. It is safe to request (or release) pads in the NULL or READY states, though.
Simple multithreaded example
//basic-tutorial-7.c
#include <gst/gst.h>
int main(int argc, char *argv[]) {
GstElement *pipeline, *audio_source, *tee, *audio_queue, *audio_convert, *audio_resample, *audio_sink;
GstElement *video_queue, *visual, *video_convert, *video_sink;
GstBus *bus;
GstMessage *msg;
GstPad *tee_audio_pad, *tee_video_pad;
GstPad *queue_audio_pad, *queue_video_pad;
/* Initialize GStreamer */
gst_init (&argc, &argv);
/* Create the elements */
audio_source = gst_element_factory_make ("audiotestsrc", "audio_source");
tee = gst_element_factory_make ("tee", "tee");
audio_queue = gst_element_factory_make ("queue", "audio_queue");
audio_convert = gst_element_factory_make ("audioconvert", "audio_convert");
audio_resample = gst_element_factory_make ("audioresample", "audio_resample");
audio_sink = gst_element_factory_make ("autoaudiosink", "audio_sink");
video_queue = gst_element_factory_make ("queue", "video_queue");
visual = gst_element_factory_make ("wavescope", "visual");
video_convert = gst_element_factory_make ("videoconvert", "csp");
video_sink = gst_element_factory_make ("autovideosink", "video_sink");
/* Create the empty pipeline */
pipeline = gst_pipeline_new ("test-pipeline");
if (!pipeline || !audio_source || !tee || !audio_queue || !audio_convert || !audio_resample || !audio_sink ||
!video_queue || !visual || !video_convert || !video_sink) {
g_printerr ("Not all elements could be created.\n");
return -1;
}
/* Configure elements */
g_object_set (audio_source, "freq", 215.0f, NULL);
g_object_set (visual, "shader", 0, "style", 1, NULL);
/* Link all elements that can be automatically linked because they have "Always" pads */
gst_bin_add_many (GST_BIN (pipeline), audio_source, tee, audio_queue, audio_convert, audio_resample, audio_sink,
video_queue, visual, video_convert, video_sink, NULL);
if (gst_element_link_many (audio_source, tee, NULL) != TRUE ||
gst_element_link_many (audio_queue, audio_convert, audio_resample, audio_sink, NULL) != TRUE ||
gst_element_link_many (video_queue, visual, video_convert, video_sink, NULL) != TRUE) {
g_printerr ("Elements could not be linked.\n");
gst_object_unref (pipeline);
return -1;
}
/* Manually link the Tee, which has "Request" pads */
tee_audio_pad = gst_element_get_request_pad (tee, "src_%u");
g_print ("Obtained request pad %s for audio branch.\n", gst_pad_get_name (tee_audio_pad));
queue_audio_pad = gst_element_get_static_pad (audio_queue, "sink");
tee_video_pad = gst_element_get_request_pad (tee, "src_%u");
g_print ("Obtained request pad %s for video branch.\n", gst_pad_get_name (tee_video_pad));
queue_video_pad = gst_element_get_static_pad (video_queue, "sink");
if (gst_pad_link (tee_audio_pad, queue_audio_pad) != GST_PAD_LINK_OK ||
gst_pad_link (tee_video_pad, queue_video_pad) != GST_PAD_LINK_OK) {
g_printerr ("Tee could not be linked.\n");
gst_object_unref (pipeline);
return -1;
}
gst_object_unref (queue_audio_pad);
gst_object_unref (queue_video_pad);
/* Start playing the pipeline */
gst_element_set_state (pipeline, GST_STATE_PLAYING);
/* Wait until error or EOS */
bus = gst_element_get_bus (pipeline);
msg = gst_bus_timed_pop_filtered (bus, GST_CLOCK_TIME_NONE, GST_MESSAGE_ERROR | GST_MESSAGE_EOS);
/* Release the request pads from the Tee, and unref them */
gst_element_release_request_pad (tee, tee_audio_pad);
gst_element_release_request_pad (tee, tee_video_pad);
gst_object_unref (tee_audio_pad);
gst_object_unref (tee_video_pad);
/* Free resources */
if (msg != NULL)
gst_message_unref (msg);
gst_object_unref (bus);
gst_element_set_state (pipeline, GST_STATE_NULL);
gst_object_unref (pipeline);
return 0;
}
$ gcc basic-tutorial-7.c -o basic-tutorial-7 `pkg-config --cflags --libs gstreamer-1.0`
Walkthrough
/* Create the elements */
audio_source = gst_element_factory_make ("audiotestsrc", "audio_source");
tee = gst_element_factory_make ("tee", "tee");
audio_queue = gst_element_factory_make ("queue", "audio_queue");
audio_convert = gst_element_factory_make ("audioconvert", "audio_convert");
audio_resample = gst_element_factory_make ("audioresample", "audio_resample");
audio_sink = gst_element_factory_make ("autoaudiosink", "audio_sink");
video_queue = gst_element_factory_make ("queue", "video_queue");
visual = gst_element_factory_make ("wavescope", "visual");
video_convert = gst_element_factory_make ("videoconvert", "video_convert");
video_sink = gst_element_factory_make ("autovideosink", "video_sink");
All the elements in the above picture are instantiated here:
audiotestsrc
produces a synthetic tone. wavescope
consumes an audio signal and renders a waveform as if it was an (admittedly당연히 cheap) oscilloscope. We have already worked with the autoaudiosink and autovideosink.
The conversion elements (audioconvert
, audioresample
and videoconvert
) are necessary to guarantee that the pipeline can be linked. Indeed, the Capabilities of the audio and video sinks depend on the hardware, and you do not know at design time if they will match the Caps produced by the audiotestsrc and wavescope. If the Caps matched, though, these elements act in “pass-through” mode and do not modify the signal, having negligible무시할만한 impact on performance.
/* Configure elements */
g_object_set (audio_source, "freq", 215.0f, NULL);
g_object_set (visual, "shader", 0, "style", 1, NULL)
Small adjustments for better demonstration: The “freq
” property of audiotestsrc
controls the frequency of the wave (215Hz makes the wave appear almost stationary in the window), and this style and shader
for wavescope make the wave continuous. Use the gst-inspect-1.0 tool described in Basic tutorial 10: GStreamer tools to learn all the properties of these elements.
/* Link all elements that can be automatically linked because they have "Always" pads */
gst_bin_add_many (GST_BIN (pipeline), audio_source, tee, audio_queue, audio_convert, audio_sink,
video_queue, visual, video_convert, video_sink, NULL);
if (gst_element_link_many (audio_source, tee, NULL) != TRUE ||
gst_element_link_many (audio_queue, audio_convert, audio_sink, NULL) != TRUE ||
gst_element_link_many (video_queue, visual, video_convert, video_sink, NULL) != TRUE) {
g_printerr ("Elements could not be linked.\n");
gst_object_unref (pipeline);
return -1;
}
This code block adds all elements to the pipeline and then links the ones that can be automatically linked (the ones with Always Pads, as the comment says).
Primary Notice:
gst_element_link_many()
can actually link elements with Request Pads. It internally requests the Pads so you do not have worry about the elements being linked having Always or Request Pads. Strange as it might seem, this is actually inconvenient, because you still need to release the requested Pads afterwards, and, if the Pad was requested automatically by gst_element_link_many(), it is easy to forget. Stay out of trouble by always requesting Request Pads manually, as shown in the next code block.
/* Manually link the Tee, which has "Request" pads */
tee_audio_pad = gst_element_get_request_pad (tee, "src_%u");
g_print ("Obtained request pad %s for audio branch.\n", gst_pad_get_name (tee_audio_pad));
queue_audio_pad = gst_element_get_static_pad (audio_queue, "sink");
tee_video_pad = gst_element_get_request_pad (tee, "src_%u");
g_print ("Obtained request pad %s for video branch.\n", gst_pad_get_name (tee_video_pad));
queue_video_pad = gst_element_get_static_pad (video_queue, "sink");
if (gst_pad_link (tee_audio_pad, queue_audio_pad) != GST_PAD_LINK_OK ||
gst_pad_link (tee_video_pad, queue_video_pad) != GST_PAD_LINK_OK) {
g_printerr ("Tee could not be linked.\n");
gst_object_unref (pipeline);
return -1;
}
gst_object_unref (queue_audio_pad);
gst_object_unref (queue_video_pad);
To link Request Pads, they need to be obtained by “requesting” them to the element. An element might be able to produce different kinds of Request Pads, so, when requesting them, the desired Pad Template name must be provided. In the documentation for the tee
element we see that it has two pad templates named “sink” (for its sink Pads) and “src_%u” (for the Request Pads). We request two Pads from the tee (for the audio and video branches) with gst_element_get_request_pad()
.
We then obtain the Pads from the downstream elements to which these Request Pads need to be linked. These are normal Always Pads, so we obtain them with gst_element_get_static_pad()
.
Finally, we link the pads with gst_pad_link()
. This is the function that gst_element_link()
and gst_element_link_many()
use internally.
The sink Pads we have obtained need to be released with gst_object_unref()
. The Request Pads will be released when we no longer need them, at the end of the program.
The only thing left to so is cleanup the requested Pads:
/* Release the request pads from the Tee, and unref them */
gst_element_release_request_pad (tee, tee_audio_pad);
gst_element_release_request_pad (tee, tee_video_pad);
gst_object_unref (tee_audio_pad);
gst_object_unref (tee_video_pad);
Conclusion
- How to make parts of a pipeline run on a different thread by using queue elements.
- What is a Request Pad and how to link elements with request pads, with
gst_element_get_request_pad()
,gst_pad_link()
andgst_element_release_request_pad()
. - How to have the same stream available in different branches by using tee elements.