omxh26xdec: videodecoder support subframe

Use of subframe API from videodecoder base class.
This subframe allows to decode subframe instead of
waiting for a whole frame.
The subframe uses the same frame over the whole
subframe passing process and will wait
for a signal to know the last subframe.
In this implementation it will use
GST_VIDEO_BUFFER_FLAG_MARKER as the
end of batch of subframes.

This implement subframe mode negotation for the Zynq based on caps
negotation. This mode can be combined with low-latency mode, in order to
reach the lowest possible latency (assuming the stream is within the
low-latency constraints for the HW).

... ! video/x-h264,alignment=nal ! omxh264dec ! ...

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-omx/-/merge_requests/49>

omx/gstomx.c

@@ -3462,7 +3462,8 @@ gst_omx_port_set_subframe (GstOMXPort * port, gboolean enabled)
 #ifdef USE_OMX_TARGET_ZYNQ_USCALE_PLUS
   OMX_ALG_VIDEO_PARAM_SUBFRAME subframe_mode;
   OMX_ERRORTYPE err;
-
+  GST_DEBUG_OBJECT (port->comp->parent, "%s subframe mode for Zynq",
+      enabled ? "Enable" : "Disable");
   GST_OMX_INIT_STRUCT (&subframe_mode);
   subframe_mode.nPortIndex = port->index;
 

omx/gstomxh264dec.c

@@ -50,6 +50,21 @@ enum
 G_DEFINE_TYPE_WITH_CODE (GstOMXH264Dec, gst_omx_h264_dec,
     GST_TYPE_OMX_VIDEO_DEC, DEBUG_INIT);
 
+#define MAKE_CAPS(alignment) \
+  "video/x-h264, " \
+      "alignment=(string) " alignment ", " \
+      "stream-format=(string) byte-stream, " \
+      "width=(int) [1,MAX], height=(int) [1,MAX]"
+
+/* The Zynq supports decoding subframes, though we want "au" to be the
+ * default, so we keep it prepended. This is the only way that it works with
+ * rtph264depay. */
+#ifdef USE_OMX_TARGET_ZYNQ_USCALE_PLUS
+#define SINK_CAPS MAKE_CAPS ("au") ";" MAKE_CAPS ("nal")
+#else
+#define SINK_CAPS MAKE_CAPS ("au")
+#endif
+
 static void
 gst_omx_h264_dec_class_init (GstOMXH264DecClass * klass)
 {
@@ -60,10 +75,7 @@ gst_omx_h264_dec_class_init (GstOMXH264DecClass * klass)
       GST_DEBUG_FUNCPTR (gst_omx_h264_dec_is_format_change);
   videodec_class->set_format = GST_DEBUG_FUNCPTR (gst_omx_h264_dec_set_format);
 
-  videodec_class->cdata.default_sink_template_caps = "video/x-h264, "
-      "alignment=(string) au, "
-      "stream-format=(string) byte-stream, "
-      "width=(int) [1,MAX], " "height=(int) [1,MAX]";
+  videodec_class->cdata.default_sink_template_caps = SINK_CAPS;
 
   gst_element_class_set_static_metadata (element_class,
       "OpenMAX H.264 Video Decoder",
@@ -86,7 +98,8 @@ gst_omx_h264_dec_is_format_change (GstOMXVideoDec * dec,
   GstCaps *old_caps = NULL;
   GstCaps *new_caps = state->caps;
   GstStructure *old_structure, *new_structure;
-  const gchar *old_profile, *old_level, *new_profile, *new_level;
+  const gchar *old_profile, *old_level, *old_alignment, *new_profile,
+      *new_level, *new_alignment;
 
   if (dec->input_state) {
     old_caps = dec->input_state->caps;
@@ -100,11 +113,14 @@ gst_omx_h264_dec_is_format_change (GstOMXVideoDec * dec,
   new_structure = gst_caps_get_structure (new_caps, 0);
   old_profile = gst_structure_get_string (old_structure, "profile");
   old_level = gst_structure_get_string (old_structure, "level");
+  old_alignment = gst_structure_get_string (old_structure, "alignment");
   new_profile = gst_structure_get_string (new_structure, "profile");
   new_level = gst_structure_get_string (new_structure, "level");
+  new_alignment = gst_structure_get_string (new_structure, "alignment");
 
   if (g_strcmp0 (old_profile, new_profile) != 0
-      || g_strcmp0 (old_level, new_level) != 0) {
+      || g_strcmp0 (old_level, new_level) != 0
+      || g_strcmp0 (old_alignment, new_alignment) != 0) {
     return TRUE;
   }
 
@@ -176,6 +192,7 @@ gst_omx_h264_dec_set_format (GstOMXVideoDec * dec, GstOMXPort * port,
   GstOMXVideoDecClass *klass = GST_OMX_VIDEO_DEC_GET_CLASS (dec);
   OMX_PARAM_PORTDEFINITIONTYPE port_def;
   OMX_ERRORTYPE err;
+  const GstStructure *s;
 
   gst_omx_port_get_port_definition (port, &port_def);
   port_def.format.video.eCompressionFormat = OMX_VIDEO_CodingAVC;
@@ -188,5 +205,12 @@ gst_omx_h264_dec_set_format (GstOMXVideoDec * dec, GstOMXPort * port,
       return FALSE;
   }
 
+  /* Enable subframe mode if NAL aligned */
+  s = gst_caps_get_structure (state->caps, 0);
+  if (!g_strcmp0 (gst_structure_get_string (s, "alignment"), "nal")
+      && gst_omx_port_set_subframe (dec->dec_in_port, TRUE)) {
+    gst_video_decoder_set_subframe_mode (GST_VIDEO_DECODER (dec), TRUE);
+  }
+
   return TRUE;
 }

omx/gstomxh265dec.c

@@ -51,6 +51,21 @@ enum
 G_DEFINE_TYPE_WITH_CODE (GstOMXH265Dec, gst_omx_h265_dec,
     GST_TYPE_OMX_VIDEO_DEC, DEBUG_INIT);
 
+#define MAKE_CAPS(alignment) \
+   "video/x-h265, " \
+      "alignment=(string) " alignment ", " \
+      "stream-format=(string) byte-stream, " \
+      "width=(int) [1,MAX], height=(int) [1,MAX]"
+
+/* The Zynq MPSoC supports decoding subframes though we want "au" to be the
+ * default, so we keep it prepended. This is the only way that it works with
+ * rtph265depay. */
+#ifdef USE_OMX_TARGET_ZYNQ_USCALE_PLUS
+#define SINK_CAPS MAKE_CAPS ("au") ";" MAKE_CAPS ("nal");
+#else
+#define SINK_CAPS MAKE_CAPS ("au")
+#endif
+
 static void
 gst_omx_h265_dec_class_init (GstOMXH265DecClass * klass)
 {
@@ -61,10 +76,7 @@ gst_omx_h265_dec_class_init (GstOMXH265DecClass * klass)
       GST_DEBUG_FUNCPTR (gst_omx_h265_dec_is_format_change);
   videodec_class->set_format = GST_DEBUG_FUNCPTR (gst_omx_h265_dec_set_format);
 
-  videodec_class->cdata.default_sink_template_caps = "video/x-h265, "
-      "alignment=(string) au, "
-      "stream-format=(string) byte-stream, "
-      "width=(int) [1,MAX], " "height=(int) [1,MAX]";
+  videodec_class->cdata.default_sink_template_caps = SINK_CAPS;
 
   gst_element_class_set_static_metadata (element_class,
       "OpenMAX H.265 Video Decoder",
@@ -87,8 +99,8 @@ gst_omx_h265_dec_is_format_change (GstOMXVideoDec * dec,
   GstCaps *old_caps = NULL;
   GstCaps *new_caps = state->caps;
   GstStructure *old_structure, *new_structure;
-  const gchar *old_profile, *old_level, *old_tier, *new_profile, *new_level,
-      *new_tier;
+  const gchar *old_profile, *old_level, *old_tier, *old_alignment,
+      *new_profile, *new_level, *new_tier, *new_alignment;
 
   if (dec->input_state) {
     old_caps = dec->input_state->caps;
@@ -103,13 +115,16 @@ gst_omx_h265_dec_is_format_change (GstOMXVideoDec * dec,
   old_profile = gst_structure_get_string (old_structure, "profile");
   old_level = gst_structure_get_string (old_structure, "level");
   old_tier = gst_structure_get_string (old_structure, "tier");
+  old_alignment = gst_structure_get_string (old_structure, "alignment");
   new_profile = gst_structure_get_string (new_structure, "profile");
   new_level = gst_structure_get_string (new_structure, "level");
   new_tier = gst_structure_get_string (new_structure, "tier");
+  new_alignment = gst_structure_get_string (new_structure, "alignment");
 
   if (g_strcmp0 (old_profile, new_profile) != 0
       || g_strcmp0 (old_level, new_level) != 0
-      || g_strcmp0 (old_tier, new_tier)) {
+      || g_strcmp0 (old_tier, new_tier) != 0
+      || g_strcmp0 (old_alignment, new_alignment) != 0) {
     return TRUE;
   }
 
@@ -184,6 +199,7 @@ gst_omx_h265_dec_set_format (GstOMXVideoDec * dec, GstOMXPort * port,
   GstOMXVideoDecClass *klass = GST_OMX_VIDEO_DEC_GET_CLASS (dec);
   OMX_PARAM_PORTDEFINITIONTYPE port_def;
   OMX_ERRORTYPE err;
+  const GstStructure *s;
 
   gst_omx_port_get_port_definition (port, &port_def);
   port_def.format.video.eCompressionFormat =
@@ -197,5 +213,12 @@ gst_omx_h265_dec_set_format (GstOMXVideoDec * dec, GstOMXPort * port,
       return FALSE;
   }
 
+  /* Enable subframe mode if NAL aligned */
+  s = gst_caps_get_structure (state->caps, 0);
+  if (!g_strcmp0 (gst_structure_get_string (s, "alignment"), "nal")
+      && gst_omx_port_set_subframe (dec->dec_in_port, TRUE)) {
+    gst_video_decoder_set_subframe_mode (GST_VIDEO_DECODER (dec), TRUE);
+  }
+
   return TRUE;
 }

omx/gstomxvideodec.c

@@ -2976,10 +2976,17 @@ gst_omx_video_dec_handle_frame (GstVideoDecoder * decoder,
   gboolean done = FALSE;
   gboolean first_ouput_buffer = TRUE;
   guint memory_idx = 0;         /* only used in dynamic buffer mode */
+  gboolean last_subframe = GST_BUFFER_FLAG_IS_SET (frame->input_buffer,
+      GST_VIDEO_BUFFER_FLAG_MARKER);
+  gboolean header =
+      GST_BUFFER_FLAG_IS_SET (frame->input_buffer, GST_BUFFER_FLAG_HEADER);
+  gboolean subframe_mode = gst_video_decoder_get_subframe_mode (decoder);
 
   self = GST_OMX_VIDEO_DEC (decoder);
 
-  GST_DEBUG_OBJECT (self, "Handling frame");
+  GST_DEBUG_OBJECT (self,
+      "Handling frame %p last_subframe=%d header %d subframes %d", frame,
+      last_subframe, header, frame->abidata.ABI.num_subframes);
 
   if (self->downstream_flow_ret != GST_FLOW_OK) {
     gst_video_codec_frame_unref (frame);
@@ -3203,8 +3210,18 @@ gst_omx_video_dec_handle_frame (GstVideoDecoder * decoder,
      *     the segment
      */
 
-    if (done)
+    if (done) {
+      /* If the input buffer is a subframe mark the OMX buffer as such */
+      if (subframe_mode && !last_subframe) {
+#ifdef OMX_BUFFERFLAG_ENDOFSUBFRAME
+        buf->omx_buf->nFlags |= OMX_BUFFERFLAG_ENDOFSUBFRAME;
+#endif
+      } else {
         buf->omx_buf->nFlags |= OMX_BUFFERFLAG_ENDOFFRAME;
+        if (subframe_mode && last_subframe)
+          gst_video_decoder_have_last_subframe (decoder, frame);
+      }
+    }
 
     self->started = TRUE;
     err = gst_omx_port_release_buffer (port, buf);