I am trying to do a remote build on my Jetson AGX Xavier from Nsignt Eclipse Edition in Debian Linux. I am following this post Remote Application Development using NVIDIA Nsight Eclipse Edition | NVIDIA Technical Blog for the remote profiling. In the post, it is mentioned to use the “Target Systems” option to build on the target but I could not find this option. Is there a new post that helps for remote build on the target for the newer versions of the tools?
MRK
Please check 1. Introduction — Nsight Eclipse Plugins Guide 12.3 documentation
It is a plugin install in Eclipse.
I installed the required plugins using the below command
/usr/local/cuda/bin/nsight_ee_plugins_manage.sh install <eclipse exe path>
I am not able to do a remote build on jetson but cross-compilation workflow is available.
Is remote build workflow removed?
Yes. Please use cross-compile or build in the target directly and do remote debug from eclipse.
Thanks Viraj, for the quick response.
I started using the cross-compile workflow but seeing the below error.
2004hubuntu@orin-1:~$ /bin/sh -c “cd "/tmp/cuda-debug";export LD_LIBRARY_PATH="/usr/local/cuda":"/usr/local/cuda/lib64":${LD_LIBRARY_PATH};"/tmp/cuda-debug/simple_add"”;exit
2004l[Vector addition of 50000 elements]
Copy input data from the host memory to the CUDA device
CUDA kernel launch with 196 blocks of 256 threads
Failed to launch vectorAdd kernel (error code the provided PTX was compiled with an unsupported toolchain.)!
I checked the CUDA version on both host and target, it is 12.3. I also ensured that I installed the CUDA-cross toolkit.
The sample I am trying to run on the target is ‘simple_add’.
I have provided necessary target paths in the “Remote systems” section.
Is there anything I am missing here?
Thanks
error code the provided PTX was compiled with an unsupported toolchain.
—It shows the sample is not built for target arch. Which arch have you chosen during build ?
Yes, Aarch64 and selected the right toolchain aarch64-linux-gnu-g++10 as my CPU compiler. Am I missing anything here?
Seems nothing wrong. You just need to double confirm the sample you run on target is copied from host, and it is generated by cross-build.
Yes, it is, when looked at the details of the executable on target, it showed correct attributes. I am pasting them here.
simple_add: ELF 64-bit LSB pie executable, ARM aarch64, version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux-aarch64.so.1, BuildID[sha1]=a94db816cd6fda4f34d25c889cfb90e6e855bd18, for GNU/Linux 3.7.0, with debug_info, not stripped
My host machine is debian 11 and target is Orin AGX.
Hi, @rmandala
I found an Orin internally and check the file there. You file details matches mine. So actually there is no issue with the sample build.
I searched internally and found this error may be due to mismatch with the driver. Are you using the BSP that support CUDA12.3 ?
Hi Veraj,
Yes, I flashed the board using SDK manager. The BSP components are intact since the flash. Here si the information about the L4T kernel.
TARGET_USERSPACE_LIB_DIR=nvidia
TARGET_USERSPACE_LIB_DIR_PATH=usr/lib/aarch64-linux-gnu/nvidia
do let me know if you want to know any information on the board.
Hi Veraj,
Thanks for the support. I figured out the issue. In the properties I should use the target GPU’s architecture to generate PTX code.
Project->Properties->C/C++ Build->Settings->Tool Settings->CUDA-> select appropriate architecture for PTX code generation.
MRK
Great. Thanks for letting me know !
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.