Copying the data output from the omxh264dec will result in high CPU consumption

Autonomous Machines Jetson & Embedded Systems Jetson TK1
1

I use Nvidia GPU for hardware decoding on TK1, and then i found that copying the data output from the omxh264dec will result in high CPU consumption(overall CPU consumption about 55%) and if display directly, the CPU consumption is low(overall CPU consumption about 10%), so my question is that how to copy the data output from the omxh264dec without incleasing CPU consumption?

Gstreamer pipeline as below:
1.copy the data output from the omxh264dec
gst-launch-1.0 -v rtspsrc user-id=admin user-pw=ms1234 location=rtsp://192.168.200.122/main ! rtph264depay ! h264parse ! omxh264dec ! nvvidconv ! ‘video/x-raw, format=(string)I420’ ! filesink location=./test.yuv

  1. display directly the data output from the omxh264dec
    gst-launch-1.0 -v rtspsrc user-id=admin user-pw=ms1234 location=rtsp://192.168.200.122/main ! rtph264depay ! h264parse ! omxh264dec ! nvhdmioverlaysink -e

Thanks a lot!

2

Hi Kevin,
The output buffertype of omxh264dec is in video/x-raw(memory:NVMM). It requires memcpy() to make it become video/x-raw.

For rendering out, all buffers are in video/x-raw(memory:NVMM) and zero memcpy(). This is where the difference of CPU consumption comes form.

3

Hi Dane,
Thanks for your reply.

Will consume so much CPU((overall CPU consumption about 55%)) for making video/x-raw(memory:NVMM) become video/x-raw is normal? Because i am worried that i use the wrong way and there is a way that copying the data output from the omxh264dec consume low CPU consumption?

Thanks a lot!

4

Hi Kevin,
Here is the result profiled via /home/ubuntu/tegrastats on r21.5
Source: bourne_ultimatum_trailer.zip @ H.264 Demo Clips | H264info.com

Set CPU to performance mode

echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

gst-launch-1.0 filesrc location=Bourne_trailer.mp4 ! qtdemux ! h264parse ! omxh264dec ! nvvidconv ! ‘video/x-raw, format=(string)I420’ ! filesink location=test.yuv
RAM 513/1892MB (lfb 9x4MB) cpu [98%,off,off,off]@2065 EMC 10%@924 AVP 0%@204 VDE 120 GR3D 0%@72 EDP limit 0
RAM 514/1892MB (lfb 9x4MB) cpu [94%,off,off,off]@2065 EMC 12%@924 AVP 0%@204 VDE 120 GR3D 0%@252 EDP limit 0
RAM 513/1892MB (lfb 9x4MB) cpu [95%,off,off,off]@2065 EMC 12%@924 AVP 0%@204 VDE 120 GR3D 0%@72 EDP limit 0

gst-launch-1.0 filesrc location=Bourne_Trailer.mp4 ! qtdemux ! h264parse ! omxh264dec ! nvhdmioverlaysink
RAM 498/1892MB (lfb 128x4MB) cpu [4%,off,off,off]@2065 EMC 9%@924 AVP 0%@204 VDE 120 GR3D 0%@72 EDP limit 0
RAM 498/1892MB (lfb 128x4MB) cpu [3%,off,off,off]@2065 EMC 9%@924 AVP 0%@204 VDE 276 GR3D 0%@72 EDP limit 0
RAM 498/1892MB (lfb 128x4MB) cpu [4%,off,off,off]@2065 EMC 9%@924 AVP 0%@204 VDE 444 GR3D 0%@72 EDP limit 0

Mostly dumping buffer out is for debugging. If you have a real usecase of dumping buffer to file, inevitably it suffers the high CPU consumption.

5

Hi Dane,
Thanks for your help.

Setting CPU to performance mode is very helpful, after doing this, the CPU consumption from 55% to 20%, so i think this is the root cause of high CPU consumption.

PS: In fact my purpose is not to save as file but callback HW decoding data

6

Hello DaneLLL!Sorry to bother you.I want to know if I set CPU to performance mode,is there any problem when I run my qt program on the TX1?