gstreamer/docs/pwg/advanced-scheduling.xml

<chapter id="chapter-scheduling" xreflabel="Different scheduling modes">
  <title>Different scheduling modes</title>
  <para>
    Scheduling is, in short, a method for making sure that every element gets
    called once in a while to process data and prepare data for the next
    element. Likewise, a kernel has a scheduler to for processes, and your
    brain is a very complex scheduler too in a way.
    Randomly calling elements' chain functions won't bring us far, however, so
    you'll understand that the schedulers in &GStreamer; are a bit more complex
    than this. However, as a start, it's a nice picture.
  </para>
  <para>
    So far, we have only discussed <function>_chain ()</function>-operating
    elements, i.e. elements that have a chain-function set on their sinkpad
    and push buffers on their sinkpad. Pads (or elements) can also operate
    in two other scheduling modes, however. In this chapter, we will discuss
    what those scheduling modes are, how they can be enabled and in what
    cases they are useful. The other two scheduling modes are random access
    (<function>_getrange ()</function>-based) or task-runner (which means
    that this element is the driving force in the pipeline) mode.
  </para>

  <sect1 id="section-scheduling-activation"
      xreflabel="The pad actication stage">
    <title>The pad activation stage</title>
    <para>
      The stage in which &GStreamer; decides in what scheduling mode the
      various elements will operate, is called the pad-activation stage. In
      this stage, &GStreamer; will query the scheduling capabilities (i.e.
      it will see in what modes each particular element/pad can operate) and
      decide on the optimal scheduling composition for the pipeline. Next,
      each pad will be notified of the scheduling mode that was assigned to
      it, and after that the pipeline will start running.
    </para>
    <para>
      Pads can be assigned one of three modes, each mode putting several
      prerequisites on the pads. Pads should implement a notification
      function (<function>gst_pad_set_activatepull_function ()</function> and
      <function>gst_pad_set_activatepush_function ()</function>) to be
      notified of the scheduling mode assignment. Also, sinkpads assigned
      to do pull-based scheduling mode should start and stop their task
      in this function.
    </para>
    <itemizedlist>
      <listitem>
        <para>
          If all pads of an element are assigned to do
          <quote>push</quote>-based scheduling, then this means that data
          will be pushed by upstream elements to this element using the
          sinkpads <function>_chain ()</function>-function. Pprerequisites
          for this scheduling mode are that a chain-function was set for
          each sinkpad using<function>gst_pad_set_chain_function ()</function>
          and that all downstream elements operate in the same mode. Pads are
          assigned to do push-based scheduling in sink-to-source element
          order, and within an element first sourcepads and then sinkpads.
          Sink elements can operate in this mode if their sinkpad is activated
          for push-based scheduling. Source elements cannot be chain-based.
        </para>
      </listitem>
      <listitem>
        <para>
          Alternatively, sinkpads can be the driving force behind a pipeline
          by operating in <quote>pull</quote>-based mode, while the sourcepads
          of the element still operate in push-based mode. In order to be the
          driving force, those pads start a <classname>GstTask</classname>
          when their pads are being activated. This task is a thread, which
          will call a function specified by the element. When called, this
          function will have random data access (through
          <function>gst_pad_get_range ()</function>) over all sinkpads, and
          can push data over the sourcepads, which effectively means that
          this element controls dataflow in the pipeline. Prerequisites for
          this mode are that all downstream elements can act in chain-based
          mode, and that all upstream elements allow random access (see below).
          Source elements can be told to act in this mode if their sourcepads
          are activated in push-based fashion. Sink elements can be told to
          act in this mode when their sinkpads are activated in pull-mode.
        </para>
      </listitem>
      <listitem>
        <para>
          lastly, all pads in an element can be assigned to act in pull-mode.
          too. However, contrary to the above, this does not mean that they
          start a task on their own. Rather, it means that they are pull
          slave for the downstream element, and have to provide random data
          access to it from their <function>_get_range ()</function>-function.
          Requiremenents are that the a <function>_get_range
          ()</function>-function was set on this pad using the function
          <function>gst_pad_set_getrange_function ()</function>. Also, if
          the element has any sinkpads, all those pads (and thereby their
          peers) need to operate in random access mode, too. Note that the
          element is supposed to activate those elements itself! &GStreamer;
          will not do that for you.
        </para>
      </listitem>
    </itemizedlist>
    <para>
      In the next two sections, we will go closer into pull-based scheduling
      (elements/pads driving the pipeline, and elements/pads providing random
      access), and some specific use cases will be given.
    </para>
  </sect1>

  <sect1 id="section-scheduling-loop" xreflabel="Pads driving the pipeline">
    <title>Pads driving the pipeline</title>
    <para>
      Sinkpads assigned to operate in pull-based mode, while none of its
      sourcepads operate in pull-based mode (or it has no sourcepads), can
      start a task that will drive the pipeline dataflow. Within this
      function, those elements have random access over all of their sinkpads,
      and push data over their sourcepads. This can come in useful for
      several different kinds of elements:
    </para>
    <itemizedlist>
      <listitem>
        <para>
          Demuxers, parsers and certain kinds of decoders where data comes
          in unparsed (such as MPEG-audio or video streams), since those will
          prefer byte-exact (random) access from their input. If possible,
          however, such elements should be prepared to operate in chain-based
          mode, too.
        </para>
      </listitem>
      <listitem>
        <para>
          Certain kind of audio outputs, which require control over their
          input dataflow, such as the Jack sound server.
        </para>
      </listitem>
    </itemizedlist>
    <para>
      In order to start this task, you will need to create it in the
      activation function.
    </para>
    <programlisting><!-- example-begin task.c a -->
#include "filter.h"
#include &lt;string.h&gt;

static gboolean	gst_my_filter_activate	(GstPad      * pad);
static gboolean	gst_my_filter_activate_pull (GstPad  * pad,
					 gboolean      active);
static void	gst_my_filter_loop	(GstMyFilter * filter);

GST_BOILERPLATE (GstMyFilter, gst_my_filter, GstElement, GST_TYPE_ELEMENT);
<!-- example-end task.c a -->
<!-- example-begin task.c b --><!--
static gboolean	gst_my_filter_setcaps	(GstPad  *pad,
					 GstCaps *caps);
static GstCaps *gst_my_filter_getcaps	(GstPad  *pad);

static void
gst_my_filter_base_init (gpointer klass)
{
  GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
  static GstElementDetails my_filter_details = {
    "An example plugin",
    "Example/FirstExample",
    "Shows the basic structure of a plugin",
    "your name <your.name@your.isp>"
  };
  static GstStaticPadTemplate sink_factory =
  GST_STATIC_PAD_TEMPLATE (
    "sink",
    GST_PAD_SINK,
    GST_PAD_ALWAYS,
    GST_STATIC_CAPS ("ANY")
  );
  static GstStaticPadTemplate src_factory =
  GST_STATIC_PAD_TEMPLATE (
    "src",
    GST_PAD_SRC,
    GST_PAD_ALWAYS,
    GST_STATIC_CAPS ("ANY")
  );

  gst_element_class_set_details (element_class, &my_filter_details);
  gst_element_class_add_pad_template (element_class,
        gst_static_pad_template_get (&src_factory));
  gst_element_class_add_pad_template (element_class,
        gst_static_pad_template_get (&sink_factory));
}

static void
gst_my_filter_class_init (GstMyFilterClass * klass)
{
}
--><!-- example-begin task.c b -->
<!-- example-begin task.c c -->
static void
gst_my_filter_init (GstMyFilter * filter)
{
<!-- example-end task.c c -->
[..]<!-- example-begin task.c d --><!--
  GstElementClass *klass = GST_ELEMENT_GET_CLASS (filter);

  filter-&gt;sinkpad = gst_pad_new_from_template (
        gst_element_class_get_pad_template (klass, "sink"), "sink");
  gst_pad_set_setcaps_function (filter-&gt;sinkpad, gst_my_filter_setcaps);
  gst_pad_set_getcaps_function (filter-&gt;sinkpad, gst_my_filter_getcaps);
--><!-- example-end task.c d -->
<!-- example-begin task.c e -->
  gst_pad_set_activate_function (filter-&gt;sinkpad, gst_my_filter_activate);
  gst_pad_set_activatepull_function (filter-&gt;sinkpad,
      gst_my_filter_activate_pull);
<!-- example-end task.c e -->
<!-- example-begin task.c f --><!--
  gst_element_add_pad (GST_ELEMENT (filter), filter-&gt;sinkpad);

  filter-&gt;srcpad = gst_pad_new_from_template (
        gst_element_class_get_pad_template (klass, "src"), "src");
  gst_element_add_pad (GST_ELEMENT (filter), filter-&gt;srcpad);
--><!-- example-end task.c f -->
[..]<!-- example-begin task.c g -->
}
<!-- example-end task.c g -->
[..]<!-- example-begin task.c h --><!--
#include "caps.func"
--><!-- example-end task.c h -->
<!-- example-begin task.c i -->
static gboolean
gst_my_filter_activate (GstPad * pad)
{
  if (gst_pad_check_pull_range (pad)) {
    return gst_pad_activate_pull (pad, TRUE);
  } else {
    return FALSE;
  }
}

static gboolean
gst_my_filter_activate_pull (GstPad  *pad,
			     gboolean active)
{
  GstMyFilter *filter = GST_MY_FILTER (GST_OBJECT_PARENT (pad));

  if (active) {
    filter->offset = 0;
    return gst_pad_start_task (pad,
        (GstTaskFunction) gst_my_filter_loop, filter);
  } else {
    return gst_pad_stop_task (pad);
  }
}
    <!-- example-end task.c i --></programlisting>
    <para>
      Once started, your task has full control over input and output. The
      most simple case of a task function is one that reads input and pushes
      that over its source pad. It's not all that useful, but provides some
      more flexibility than the old chain-based case that we've been looking
      at so far.
    </para>
    <programlisting><!-- example-begin task.c j -->
#define BLOCKSIZE 2048

static void
gst_my_filter_loop (GstMyFilter * filter)
{
  guint64 len;
  GstFormat fmt = GST_FORMAT_BYTES;
  GstBuffer *buf = NULL;

  if (!gst_pad_query_position (filter-&gt;sinkpad, &amp;fmt, NULL, &amp;len)) {
    goto stop;
  } else if (filter-&gt;offset >= len) {
    gst_pad_push_event (filter-&gt;sinkpad, gst_event_new (GST_EVENT_EOS));
  } else if (gst_pad_pull_range (filter-&gt;sinkpad, filter-&gt;offset,
				 BLOCKSIZE, &amp;buf) != GST_FLOW_OK ||
	     gst_pad_push (filter-&gt;sinkpad, buf) != GST_FLOW_OK) {
    goto stop;
  } else {
    filter-&gt;offset += BLOCKSIZE;
    return;
  }

stop:
  gst_pad_pause_task (filter-&gt;sinkpad);
}
<!-- example-end task.c j -->
<!-- example-begin task.c k --><!--
#include "register.func"
    --><!-- example-end task.c k --></programlisting>
  </sect1>

  <sect1 id="section-scheduling-randomxs" xreflabel="Providing random access">
    <title>Providing random access</title>
    <para>
      In the previous section, we have talked about how elements (or pads)
      that are assigned to drive the pipeline using their own task, have
      random access over their sinkpads. This means that all elements linked
      to those pads (recursively) need to provide random access functions.
      Requesting random access is done using the function
      <function>gst_pad_pull_range ()</function>, which requests a buffer of
      a specified size and offset. Source pads implementing and assigned to
      do random access will have a <function>_get_range ()</function>-function
      set using <function>gst_pad_set_getrange_function ()</function>, and
      that function will be called when the peer pad requests some data. The
      element is then responsible for seeking to the right offset and
      providing the requested data. Several elements can implement random
      access:
    </para>
    <itemizedlist>
      <listitem>
        <para>
          Data sources, such as a file source, that can provide data from any
          offset with reasonable low latency.
        </para>
      </listitem>
      <listitem>
        <para>
          Filters that would like to provide a pull-based-like scheduling
          mode over the whole pipeline. Note that elements assigned to do
          random access-based scheduling are themselves responsible for
          assigning this scheduling mode to their upstream peers! &GStreamer;
          will not do that for you.
        </para>
      </listitem>
      <listitem>
        <para>
          Parsers who can easily provide this by skipping a small part of
          their input and are thus essentially "forwarding" random access
          requests literally without any own processing involved. Examples
          include tag readers (e.g. ID3) or single output parsers, such as
          a WAVE parser.
        </para>
      </listitem>
    </itemizedlist>
    <para>
      The following example will show how a <function>_get_range
      ()</function>-function can be implemented in a source element:
    </para>
    <programlisting><!-- example-begin range.c a -->
#include "filter.h"
static GstFlowReturn
		gst_my_filter_get_range	(GstPad     * pad,
					 guint64      offset,
					 guint        length,
					 GstBuffer ** buf);

GST_BOILERPLATE (GstMyFilter, gst_my_filter, GstElement, GST_TYPE_ELEMENT);
<!-- example-end range.c a -->
<!-- example-begin range.c b --><!--
static void
gst_my_filter_base_init (gpointer klass)
{
  GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
  static GstElementDetails my_filter_details = {
    "An example plugin",
    "Example/FirstExample",
    "Shows the basic structure of a plugin",
    "your name <your.name@your.isp>"
  };
  static GstStaticPadTemplate src_factory =
  GST_STATIC_PAD_TEMPLATE (
    "src",
    GST_PAD_SRC,
    GST_PAD_ALWAYS,
    GST_STATIC_CAPS ("ANY")
  );

  gst_element_class_set_details (element_class, &my_filter_details);
  gst_element_class_add_pad_template (element_class,
        gst_static_pad_template_get (&src_factory));
}

static void
gst_my_filter_class_init (GstMyFilterClass * klass)
{
}
--><!-- example-begin range.c b -->
<!-- example-begin range.c c -->
static void
gst_my_filter_init (GstMyFilter * filter)
{
  GstElementClass *klass = GST_ELEMENT_GET_CLASS (filter);

  filter-&gt;srcpad = gst_pad_new_from_template (
        gst_element_class_get_pad_template (klass, "src"), "src");
  gst_pad_set_getrange_function (filter-&gt;srcpad,
      gst_my_filter_get_range);
  gst_element_add_pad (GST_ELEMENT (filter), filter-&gt;srcpad);
<!-- example-end range.c c -->
[..]<!-- example-begin range.c d -->
}

static gboolean
gst_my_filter_get_range (GstPad     * pad,
			 guint64      offset,
			 guint        length,
			 GstBuffer ** buf)
{
<!-- example-end range.c d -->
  GstMyFilter *filter = GST_MY_FILTER (GST_OBJECT_PARENT (pad));

  [.. here, you would fill *buf ..]
<!-- example-begin range.c e -->
  return GST_FLOW_OK;
}
<!-- example-end range.c e -->
<!-- example-begin range.c f --><!--
#include "register.func"
    --><!-- example-end range.c f --></programlisting>
    <para>
      In practice, many elements that could theoretically do random access,
      may in practice often be assigned to do push-based scheduling anyway,
      since there is no downstream element able to start its own task.
      Therefore, in practice, those elements should implement both a
      <function>_get_range ()</function>-function and a <function>_chain
      ()</function>-function (for filters and parsers) or a <function>_get_range
      ()</function>-function and be prepared to start their own task by
      providing <function>_activate_* ()</function>-functions (for
      source elements), so that &GStreamer; can decide for the optimal
      scheduling mode and have it just work fine in practice.
    </para>
  </sect1>
</chapter>