5

I have been trying to run a Python program on a system with RISC-V architecture. The program involves runs a simple artificial neural network model, and involves the use of numpy.

However, the operating system running on the RISC-V archtecture (currently running on an FPGA) is a very basic Linux system that cannot install software such as Python3 or Pip. But it can compile C/C++ programs and run them. So I thought there could still be a way to run my Python program on this system, by converting it to C/C++ on my host system, transferring it to the RISC-V system, and compile it there.

The problem is that I am not able to find a good Python to C/C++ converter software anywhere. I found that there are some software tools available to convert Python code to binary executables, such as py2exe and pyinstaller. However, it is irrelevant if I am trying to run machine code, only compatible only with my host system architecture, using RISC-V architecture.

I also came across Nuitka, which does convert Python code into C/C++ and stores it in a portable distribution folder, but then also generates shared object files (.so) for libraries which I have imported in my program, which again can't be executed on the RISC-V system. Furthermore, based on what I have seen in the Nuitka documentation, there are no guidelines for compiling of the C/C++ code in the distribution.

So I have a couple of questions :

  1. Based on the constraints I have mentioned regarding my RISC-V system, is there any other procedure I could follow to run my Python program on the system?
  2. If not, is there any software tools that I could use to convert Python to C/C++, other than Nuitka?
  3. If not, can you provide a clear procedure as to how to convert Python code to C/C++ using Nuitka, and then compiling the given C/C++ code on the RISC-V system?
Bharath Kinnal
  • 113
  • 8
  • That task is not as easy as you may think; it is somehow hard to manage. The best is to have some cross-compiler to compile to native RISC-V platform on another one and then deploy the generated executable on your target. But this is beyond SO purpose. First have a look a cross-compiler for your dev platform. – Jean-Baptiste Yunès Jun 10 '20 at 08:15
  • 2
    If it is a simple neural network model, why not directly writing a C++ code for it, starting from this model and not from the Python code? – Damien Jun 10 '20 at 08:15
  • @Damien, the program that I had mentioned was just a test program. My main aim is to run any Python program, which would include those importing Tensorflow, PyTorch, etc., on the RISC-V system. – Bharath Kinnal Jun 10 '20 at 13:40
  • Based on what you've described, I'd go for patching Nuitka and have it use a RISC-V cross compiler to generate binaries, instead of the host compiler – valiano Jun 11 '20 at 14:42
  • @BharathKinnal how about progress in your project? – Adam Mierzwiak Aug 21 '20 at 12:37
  • try https://chocopy.org – Delfino Jun 29 '21 at 01:38
  • @BharathKinnal Have you got the answer to your question? It's my question too – engineer1155 Feb 03 '22 at 17:00

1 Answers1

0

You say your system can compile C/C++ programs and run them. Most python interpreters, such as the reference implementation CPython, are C programs, so perhaps the best course of action is to try to compile python itself on your system and then run the python scripts using that. https://github.com/python/cpython#build-instructions CPython's build instructions claim it will work for Linux, although it may fail if you don't have other programs used in the build process (eg make). Good luck!

Wyatt Carpenter
  • 176
  • 4