mirrors/gst-plugins-rs
mirrors/gst-plugins-rs
1
1
Fork 0
mirror of https://gitlab.freedesktop.org/gstreamer/gst-plugins-rs.git synced 2024-06-10 18:29:22 +00:00
Code Issues Projects Releases Wiki Activity

webrtcsink: NVIDIA V4L2 encoders always require NVMM memory

Browse source 
And if the input is not like that then a corresponding converter must be
inserted.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-rs/-/merge_requests/1330>
This commit is contained in:
Sebastian Dröge 2023-07-18 15:18:08 +03:00
parent 8f3fed8462
commit 23e7226c94
1 changed files with 28 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

46
net/webrtc/src/webrtcsink/imp.rs
View file

@ -92,6 +92,10 @@ impl Codec {
self.encoder
.has_type(gst::ElementFactoryType::VIDEO_ENCODER)
}
fn encoder_factory(&self) -> &gst::ElementFactory {
&self.encoder
}
}
/// Wrapper around our sink pads
@ -314,7 +318,7 @@ impl Default for State {
}
}
fn make_converter_for_video_caps(caps: &gst::Caps) -> Result<gst::Element, Error> {
fn make_converter_for_video_caps(caps: &gst::Caps, codec: &Codec) -> Result<gst::Element, Error> {
assert!(caps.is_fixed());
let video_info = gst_video::VideoInfo::from_caps(caps)?;
@ -323,7 +327,27 @@ fn make_converter_for_video_caps(caps: &gst::Caps) -> Result<gst::Element, Error
let (head, mut tail) = {
if let Some(feature) = caps.features(0) {
if feature.contains(CUDA_MEMORY_FEATURE) {
if feature.contains(NVMM_MEMORY_FEATURE)
// NVIDIA V4L2 encoders require NVMM memory as input and that requires using the
// corresponding converter
|| codec
.encoder_factory()
.name().starts_with("nvv4l2")
{
let queue = make_element("queue", None)?;
let nvconvert = if let Ok(nvconvert) = make_element("nvvideoconvert", None) {
nvconvert.set_property_from_str("compute-hw", "Default");
nvconvert.set_property_from_str("nvbuf-memory-type", "nvbuf-mem-default");
nvconvert
} else {
make_element("nvvidconv", None)?
};
ret.add_many(&[&queue, &nvconvert])?;
gst::Element::link_many(&[&queue, &nvconvert])?;
(queue, nvconvert)
} else if feature.contains(CUDA_MEMORY_FEATURE) {
if let Some(convert_factory) = gst::ElementFactory::find("cudaconvert") {
let cudaupload = make_element("cudaupload", None)?;
let cudaconvert = convert_factory.create().build()?;
@ -357,20 +381,6 @@ fn make_converter_for_video_caps(caps: &gst::Caps) -> Result<gst::Element, Error
gst::Element::link_many(&[&glupload, &glconvert, &glscale])?;
(glupload, glscale)
} else if feature.contains(NVMM_MEMORY_FEATURE) {
let queue = make_element("queue", None)?;
let nvconvert = if let Ok(nvconvert) = make_element("nvvideoconvert", None) {
nvconvert.set_property_from_str("compute-hw", "Default");
nvconvert.set_property_from_str("nvbuf-memory-type", "nvbuf-mem-default");
nvconvert
} else {
make_element("nvvidconv", None)?
};
ret.add_many(&[&queue, &nvconvert])?;
gst::Element::link_many(&[&queue, &nvconvert])?;
(queue, nvconvert)
} else {
let convert = make_element("videoconvert", None)?;
let scale = make_element("videoscale", None)?;
@ -524,7 +534,7 @@ fn setup_encoding(
twcc: bool,
) -> Result<(gst::Element, gst::Element, gst::Element), Error> {
let conv = match codec.is_video() {
true => make_converter_for_video_caps(input_caps)?.upcast(),
true => make_converter_for_video_caps(input_caps, codec)?.upcast(),
false => gst::parse_bin_from_description("audioresample ! audioconvert", true)?.upcast(),
};
@ -2178,7 +2188,7 @@ impl WebRTCSink {
let mut elements = vec![src.clone()];
if codec.is_video() {
elements.push(make_converter_for_video_caps(caps)?);
elements.push(make_converter_for_video_caps(caps, &codec)?);
}
gst::debug!(