Using Nsight for building C++ project for TX1 tegra ARM

Autonomous Machines Jetson & Embedded Systems Jetson TX1
1

I am trying to use the NVIDIA Nsight IDE to build my CUDA and C++ applications for the TX1. I am able to build a CUDA Application but when I create a new C++ project and try to get the compiler to generate an ARM executable it fails to do so. The GUI isn’t letting me select the CPU type for the remote as it doesn’t even post the widget. What is the proper sequence for generating a standard C++ project on the ubuntu host that will generate an ARM executable for the TX1? I would like to get a “hello world” to work first. Thanks

2

Hi Sherrick,

Could you try if https://devblogs.nvidia.com/parallelforall/remote-application-development-nvidia-nsight-eclipse-edition/ helps?

3

Hi Sherrick,

Have this issue be clarified and resolved?
Any further suggestion required?

Thanks

4

not really. The link talks about building a CUDA project for the target system. I would like to simply build a non CUDA C++ project. Just a C++ project. I’m trying to get a simple Hello World .cpp file to compile for the ARM.

5

when I create a C++ project and use the Linux GCC toolchain. I can compile a hello world example, however when I try to run as a remote app on the target ARM jetson TX1 I get an error with the message

syntax error: “(” expected

Something with the execution line when it trys to run. I’d like the steps to simply generate a C++ non CUDA app to run on the CPU built with the g++ compiler.

6

I just tried 3 different builds on Nsight for a simple Hello World example.

#include

int main()
{
std::cout << “Hello World” << std::endl;
}

If I go to the project properties and select the CUDA tool chain and build for the TX1 ( Ubuntu@tegra-ubuntu ) and run the project then it runs.

If I go to the project properties and select Cross GCC and build and run it gives me an error like the exe is not in an ARM exe format.

If I go to the project properties and select the Linux GCC and build and run I get a similar error as above.

So the only way it seems to build a valid ARM exe is if I select the CUDA tool chain. I was hoping to be able not have to do this as my code has no CUDA dependencies.

Thanks,
Steve

7

I would like to be able to cross compile a simple C++ program with the native TX1 g++ compiler and be able to run the executable via the Nsight tool on my Ubuntu 14.04 host

8

The Nsight GUI is not provided for ARM to natively run on Jetson.

Please see more update from below thread:
https://devtalk.nvidia.com/default/topic/985352/jetson-tx1/can-nsight-run-directly-on-tx1-to-do-native-development/post/5048821/#5048821

Thanks