gstreamer 1.19.3 CPU overhead for simple pipeline

Hi folks,

I'm running the same pipeline with two different versions of gstreamer

gst-launch-1.0 v4l2src ! videoconvert ! video/x-raw,format=I420, width=640, height=480, framerate=30/1 ! x264enc ! fakesink

And I'm getting very different CPU overhead.

The base system is ubuntu 20.04 with intel i7 processor.

The system also has an NVIDIA GPU:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.86       Driver Version: 470.86       CUDA Version: 11.4     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA T500         Off  | 00000000:01:00.0 Off |                  N/A |
| N/A   47C    P0    N/A /  N/A |   1307MiB /  3914MiB |     36%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

Version A is from apt:

gst-inspect-1.0 version 1.16.2
GStreamer 1.16.2
https://launchpad.net/distros/ubuntu/+source/gstreamer1.0

Version B was compiled from sources:

gst-inspect-1.0 version 1.19.3
GStreamer 1.19.3
Unknown package origin

Using the following script:

meson \
  --buildtype=release \
  -Dprefix=/usr \
  -Dgpl=enabled \
  -Dbad=enabled \
  -Dugly=enabled \
  -Dgst-plugins-base:gl=enabled \
  -Dgst-plugins-bad:gl=enabled \
  -Dgst-plugins-bad:nvcodec=enabled \
  -Dtests=disabled \
  -Dexamples=disabled \
  -Dgst-examples=disabled \
  build_dir

ninja -C build_dir
ninja -C build_dir install

Running the above pipeline has an overhead of 100% when B is used compared to A (i.e. B is using more CPUs).

Some hypothesis:

we're building it without some optimisations;
regression in 1.19.3;
different default params for x264enc (although I checked and I can't see anything suspicious)