Please provide the following info (tick the boxes after creating this topic): Software Version
[] DRIVE OS 6.0.6 Target Operating System
[] Linux
Hardware Platform
[*] DRIVE AGX Orin Developer Kit (not sure its number)
SDK Manager Version
[* ] 1.9.2.10884
Host Machine Version
[*] native Ubuntu Linux 20.04 Host installed with DRIVE OS Docker Containers
other
according to cgf_getting_started.html
Currently there is a cgf sample in /usr/local/driveworks-<version>/src/cgf which demostrates many features of the cgf. But we need more details to develop applications based on this new framework.
Due to the fact of closing source of this CGFDemo, we decide to create some minimal testing demo to check the usage of run_cgf.sh, launcher, LoaderLite, ScheduleManager, and so on. And we have many qsts.
is there any development tutorials about dwcgf or is there any plan in your roadmap?
what’s the best practice to launch a cgf app, is it the /bin/launcher?
how to terminate the cgf app cleanly, because we found that the log thread didn’t endup after the launcher ends.
I will add and update this list in this topic.
If you need, we would like to provide our testing code in this forum.
# goto target system which is orin devkit
cd ./nv_driveworks_demo/target/aarch64/install/example/dwcgf_helloworld/bin/
sudo ./run_cgf.sh
the app can run successfully, and output some logs. we can see log files continue to grow in ./LogFolder
the app cannot exit successfully, then Ctrl-Q the terminal, reboot the orin-devkit
Expected behavior
we can run helloworld cgf demo after rebooting the orin-kit.
Actual behavior
The CUDA driver is corrupted and /usr/local/driveworks/bin/sample_hello_world failed.
nvidia@tegra-ubuntu:~/myz/bin$ /usr/local/driveworks/bin/sample_hello_world
*************************************************
Welcome to Driveworks SDK
[30-01-2023 12:22:07] Platform: Detected Drive Orin P3710
[30-01-2023 12:22:07] TimeSource: monotonic epoch time offset is 1675079443679635
[30-01-2023 12:22:07] TimeSourceVibranteLinux: detect valid PTP interface mgbe2_0
[30-01-2023 12:22:07] TimeSource: Could not detect valid PTP time source at nvpps. Fallback to mgbe2_0
[30-01-2023 12:22:07] PTP Time is available from Eth Driver
[30-01-2023 12:22:07] Adding variable DW_Base:DW_Version
[30-01-2023 12:22:07] Added variable DW_Base:DW_Version
libnvrm_gpu.so: NvRmGpuLibOpen failed, error=14
[30-01-2023 12:22:08] Driveworks exception thrown: Platform: cannot retrieve GPU device count.. Error cudaErrorUnknown: unknown error
Cannot init SDK
Additional context
After reflashing the DriveOS SDK using sdk-manager, the sample_hello_world works again.
But running run_cgf.sh actually damanges the cuda driver env. So we must reflash the orin-kit again.
Dear @lizhensheng,
Does that mean, running your sample application corrupt the cuda drivers and reflashing fix the issue? If so, this behavior is strange.
Yes, you are right. This is the minimal cgf demo that I try to run, everything is derived from the CGFDemo Project, which is closed source for now. The behavior of run_cgf.sh is to run launcher which is the cause of the strange behavior.
Could you repeat all behavior of this topic in your env using the helloworld demo?
Dear @lizhensheng,
I escalated the issues related to custom node integration to core team and is under investigation. I will keep you updated. Thank you for brining to our notice.