[Decoder/direct_video] Pipeline may allocate buffer for you already

Do not die when the pipeline has already allocated the exact
size of buffer for you. Just reuse what's already there.

This fixes the 3rd subissue of #700.

Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>

diff --git a/gst/tensor_decoder/tensordec-directvideo.c b/gst/tensor_decoder/tensordec-directvideo.c
index 6072103..7f7588e 100644 (file)
--- a/gst/tensor_decoder/tensordec-directvideo.c
+++ b/gst/tensor_decoder/tensordec-directvideo.c
@@ -108,12 +108,26 @@ dv_getOutputDim (GstTensorDec * self, const GstTensorConfig * config)
   return gst_caps_simplify (caps);
 }
 
+/** @brief get video output buffer size */
+static gsize
+_get_video_xraw_bufsize (tensor_dim dim)
+{
+  /* dim[0] is bpp and there is zeropadding only when dim[0]%4 > 0 */
+  return ((dim[0] * dim[1] - 1) / 4 + 1) * 4 * dim[2];
+}
+
 /** @brief tensordec-plugin's TensorDecDef callback */
 static gsize
 dv_getTransformSize (GstTensorDec * self, GstCaps * caps,
     gsize size, GstCaps * othercaps, GstPadDirection direction)
 {
-  return 0;                     /* I'll allocate. Do not allocate for me */
+  GstTensorConfig *config = &self->tensor_config;
+  uint32_t *dim = &(config->info.dimension[0]);
+
+  if (direction == GST_PAD_SINK)
+    return _get_video_xraw_bufsize (dim);
+  else
+    return 0; /** @todo NYI */
 }
 
 /** @brief tensordec-plugin's TensorDecDef callback */
@@ -125,26 +139,30 @@ dv_decode (GstTensorDec * self, const GstTensorMemory * input,
   GstMemory *out_mem;
   GstTensorConfig *config = &self->tensor_config;
   uint32_t *dim = &(config->info.dimension[0]);
+  size_t size = _get_video_xraw_bufsize (dim);
 
   g_assert (outbuf);
-  g_assert (gst_buffer_get_size (outbuf) == 0);
+  if (gst_buffer_get_size (outbuf) > 0 && gst_buffer_get_size (outbuf) != size) {
+    gst_buffer_set_size (outbuf, size);
+  }
   g_assert (config->info.type == _NNS_UINT8);
 
+  if (gst_buffer_get_size (outbuf) == size) {
+    /* Don't reallocate. Reuse what's already given */
+    out_mem = gst_buffer_get_all_memory (outbuf);
+  } else {
+    out_mem = gst_allocator_alloc (NULL, size, NULL);
+  }
+  g_assert (gst_memory_map (out_mem, &out_info, GST_MAP_WRITE));
+
   if (0 == ((dim[0] * dim[1]) % 4)) {
     /* No Padding Required */
-    out_mem = gst_allocator_alloc (NULL, input->size, NULL);
-    g_assert (gst_memory_map (out_mem, &out_info, GST_MAP_WRITE));
     memcpy (out_info.data, input->data, input->size);
   } else {
     /* Do Padding */
-    size_t size;
     int h;
     uint8_t *ptr, *inp;
 
-    size = ((dim[0] * dim[1] - 1) / 4 + 1) * 4 * dim[2];
-    out_mem = gst_allocator_alloc (NULL, size, NULL);
-    g_assert (gst_memory_map (out_mem, &out_info, GST_MAP_WRITE));
-
     ptr = (uint8_t *) out_info.data;
     inp = (uint8_t *) input->data;
     for (h = 0; h < dim[2]; h++) {
@@ -154,7 +172,9 @@ dv_decode (GstTensorDec * self, const GstTensorMemory * input,
     }
   }
   gst_memory_unmap (out_mem, &out_info);
-  gst_buffer_append_memory (outbuf, out_mem);
+
+  if (gst_buffer_get_size (outbuf) == 0)
+    gst_buffer_append_memory (outbuf, out_mem);
 
   /** @todo Caller of dv_decode in tensordec.c should call gst_memory_unmap to inbuf */
 

diff --git a/gst/tensor_decoder/tensordec.c b/gst/tensor_decoder/tensordec.c
index 745b4ae..7b8fa5b 100644 (file)
--- a/gst/tensor_decoder/tensordec.c
+++ b/gst/tensor_decoder/tensordec.c
@@ -587,6 +587,7 @@ gst_tensordec_class_init (GstTensorDecClass * klass)
 
   /** Refer: https://gstreamer.freedesktop.org/documentation/design/element-transform.html */
   trans_class->passthrough_on_same_caps = FALSE;
+  trans_class->transform_ip_on_passthrough = FALSE;
 
   /** Processing units */
   trans_class->transform = GST_DEBUG_FUNCPTR (gst_tensordec_transform);