TDA4VH-Q1: support on concurrent video encode on TDA4VH

Hi,

A quick question, How the customer is logging the time perframe encode ? Is it through gstreamer ?

As you might be aware, TDA4VH has 2 instances of codec on it. When using v4l2h265enc, which is a gst element for /dev/video1. For /dev/video3, which is the second H265 encoder instance of codec, it will be v4l2video3h265enc. You can find the elements by " gst-inspect-1.0 | grep v4l2 ". Using the second instance, will extend the performance as, both the instances run in parallel.

root@j784s4-evm:/opt/edgeai-gst-apps# gst-inspect-1.0 | grep v4l2
video4linux2: v4l2src: Video (video4linux2) Source
video4linux2: v4l2sink: Video (video4linux2) Sink
video4linux2: v4l2radio: Radio (video4linux2) Tuner
video4linux2: v4l2deviceprovider (GstDeviceProviderFactory)
video4linux2: v4l2h264dec: V4L2 H264 Decoder
video4linux2: v4l2h265dec: V4L2 H265 Decoder
video4linux2: v4l2h264enc: V4L2 H.264 Encoder
video4linux2: v4l2h265enc: V4L2 H.265 Encoder
video4linux2: v4l2video2h264dec: V4L2 H264 Decoder
video4linux2: v4l2video2h265dec: V4L2 H265 Decoder
video4linux2: v4l2video3h264enc: V4L2 H.264 Encoder
video4linux2: v4l2video3h265enc: V4L2 H.265 Encoder

First Instance : gst-launch-1.0 videotestsrc ! v4l2h265enc  ! fakevideosink

Second Instance : gst-launch-1.0 videotestsrc ! v4l2video3h265enc  ! fakevideosink 

These two command running concurrently, runs in parallel eliminating the overhead seen in serial processing.

Regards,

Prasanth

Thanks Reply! But we still have two questions here

1. We see in the manual that the codec supports 4k60 frames, so is the encoding/decoding performance of each channel 4k60 frames? Or is it 4k60 frames in total?

2. We have 6 channels of encoding and 7 channels of decoding. The video resolutions of these channels are also different. Can they run at the same time?

3. I also saw that you said on 8.6 that running 8-way decoding memory at the same time will not be enough. Is there any statistics on how much memory will be used by using one codec? We found that using one 1080p encoding will use 60M memory. using gstreamer

4. We test it down. The decoding performance seems to be relatively weak, 1080p needs more than 50ms, have you tested the performance? using gstreamer

Regarding the fourth question. Do you have test cases for gstreamer? I feel more like a problem with gstreamer

Your observation is correct. See this section: https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-j721s2/08_06_00_10/exports/docs/linux/Foundational_Components_Multimedia_wave5.html#memory-requirement

Is the memory usage of this test CMA memory?  vmstat would read user memory

1. We see in the manual that the codec supports 4k60 frames, so is the encoding/decoding performance of each channel 4k60 frames? Or is it 4k60 frames in total?

Sorry, I want to confirm again, it is a total of 4k60 frames for encoding and decoding. Or encoding and decoding can run at the same time, both can achieve this performance