Getting nvc++ on the Nvidia Jetson Orin Nano Dev board

Issue Overview

Users are experiencing difficulties in obtaining and using the nvc++ compiler on the Nvidia Jetson Orin Nano Dev board. The main concerns include:

• The Jetpack SDK includes nvcc but not nvc++
• Users want to utilize C++20 features, particularly for GPU parallelism
• There’s confusion about whether the HPC SDK, which includes nvc++, is compatible with Jetson devices
• Attempts to use the HPC SDK Docker container on Jetson have led to memory binding issues

The problem impacts users who wish to leverage advanced C++ features and GPU acceleration techniques on their Jetson devices, particularly those following NVIDIA’s course on "Scaling GPU-Accelerated Application with the C++ standard library".

Possible Causes

1. Incompatibility: The HPC SDK, which includes nvc++, is not officially supported on Jetson devices.

2. Hardware Limitations: The Jetson’s ARM-based architecture may not be fully compatible with all features of the HPC SDK, which is primarily designed for x86-64 systems.

3. Docker Configuration Issues: The Docker container for HPC SDK may not be properly configured for the Jetson’s hardware, leading to memory binding problems.

4. Software Version Mismatch: The current software stack on the Jetson may not meet all the requirements for running the HPC SDK effectively.

5. Compiler Availability: The specific compiler (nvc++) required for certain GPU parallelism features is not included in the standard Jetpack SDK for Jetson devices.

Troubleshooting Steps, Solutions & Fixes

1. Understand Limitations:

   • Recognize that nvc++ is part of the HPC SDK, which is not officially supported on Jetson devices.
   • Be aware that attempting to use unsupported software may lead to unexpected issues and is not recommended for production environments.

2. Explore Alternative Compilers:

   • Consider using the available nvcc compiler for CUDA-specific development on Jetson.
   • Investigate the capabilities of the g++ compiler included in Jetpack for C++20 features.

3. Upgrade G++ Version:

   • Jetson supports upgrading the G++ version via apt, which may provide access to most C++20 features.
   • To upgrade, use the following command:

     sudo apt-get update
     sudo apt-get install g++-10
     

   • After installation, you can use g++-10 for compiling with C++20 features.

4. Check C++20 Feature Support:

   • Refer to the GCC C++20 status page to understand which features are supported in your G++ version:
     https://gcc.gnu.org/onlinedocs/libstdc++/manual/status.html#status.iso.2020

5. Adapt Course Materials:

   • For users following the NVIDIA course on GPU acceleration, try to adapt the examples to use nvcc and CUDA directly, rather than relying on nvc++.
   • Explore CUDA-specific parallelism techniques that are supported on Jetson devices.

6. Monitor Future Updates:

   • Keep an eye on NVIDIA’s announcements for potential future support of nvc++ or equivalent functionality on Jetson devices.
   • Consider participating in NVIDIA developer forums to express interest in expanded compiler support for Jetson.

7. Docker Troubleshooting (if attempting to use HPC SDK despite limitations):

   • Investigate the memory binding issues by checking the system’s NUMA configuration and Docker’s resource allocation settings.
   • Run hwloc-info --support on the host system to compare with the Docker container’s output.
   • Consider using --privileged flag or adjusting cgroup settings when running the Docker container, but be aware of security implications.

8. Explore Jetson-specific Development Tools:

   • Familiarize yourself with Jetson-specific development tools and libraries that may offer similar functionality to what you’re seeking from nvc++.
   • Consult the Jetson documentation for recommended practices in GPU-accelerated application development.

Remember that while these steps may help in exploring alternatives or working around limitations, the core issue of nvc++ availability on Jetson devices remains unresolved due to platform incompatibility. It’s crucial to align development goals with the officially supported tools and capabilities of the Jetson platform.