MATLAB Coder: How do I setup the environment variables on ARM targets to point to the ARM Compute Library?

I see a few deep learning networks supported for code generation using MATLAB Coder:

• Deep Learning Networks and Layers Supported for C++ Code Generation

And I've seen the question here about building the ARM Compute Library:

• MATLAB Coder: How do I build the ARM Compute Library for Deep Learning C++ code generation and deployment?

The next question I have is, how do I set the environment variables on ARM targets to point to the ARM Compute Library?

Answer 1

[Bill Chou]

To avoid build failures on ARM hardware targets, such as the Raspberry Pi and Hikey960, you must set the necessary environment variables non-interactively.

For example, with the ComputeLibrary folder installed under ~, the user home directory, and the ARM Compute libraries stored at ~/ComputeLibrary/lib, you can add this code block to the file ~/.bashrc:

#Set below block, if you use your hardware non-interactively (through ssh from remote host or tty/new terminal onto local machine)
case $- in 
    *i*) ;; 
      *) 
      export ARM_COMPUTELIB=~/ComputeLibrary 
      export PATH=${PATH}:~/ComputeLibrary 
      export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:~/ComputeLibrary/lib
      return;; 
esac

[Davide Buellis]

Hi Bill!

Could you please help me on an issue I am encountering? I am performing the example "Code Generation for Deep Learning on Raspberry Pi" and I followed your tutorials on how to install the Arm Compute Library and the environmental variables.

Compiling, though, an error comes out:

Error executing command "touch -c /home/pi/remoteBuildDir/MATLAB_ws/R2019b/C/Users/Buellis/Documents/MATLAB/Examples/R2019a/deeplearning_shared/CodeGenerationForDeepLearningOnRaspberryPiExample/codegen/exe/squeezenet_raspi_predict/*.*;make -f squeezenet_raspi_predict_rtw.mk all MATLAB_WORKSPACE="/home/pi/remoteBuildDir/MATLAB_ws/R2019b" -C /home/pi/remoteBuildDir/MATLAB_ws/R2019b/C/Users/Buellis/Documents/MATLAB/Examples/R2019a/deeplearning_shared/CodeGenerationForDeepLearningOnRaspberryPiExample/codegen/exe/squeezenet_raspi_predict". Details:

STDERR: In file included from /home/pi/remoteBuildDir/MATLAB_ws/R2019b/C/Users/Buellis/Documents/MATLAB/Examples/R2019a/deeplearning_shared/CodeGenerationForDeepLearningOnRaspberryPiExample/codegen/exe/squeezenet_raspi_predict/MWDepthConcatenationLayerImpl.hpp:6,

from /home/pi/remoteBuildDir/MATLAB_ws/R2019b/C/Users/Buellis/Documents/MATLAB/Examples/R2019a/deeplearning_shared/CodeGenerationForDeepLearningOnRaspberryPiExample/codegen/exe/squeezenet_raspi_predict/MWDepthConcatenationLayer.cpp:4:

/home/pi/remoteBuildDir/MATLAB_ws/R2019b/C/Users/Buellis/Documents/MATLAB/Examples/R2019a/deeplearning_shared/CodeGenerationForDeepLearningOnRaspberryPiExample/codegen/exe/squeezenet_raspi_predict/MWCNNLayerImpl.hpp:11:10: fatal error: arm_compute/runtime/NEON/NEFunctions.h: No such file or directory

#include "arm_compute/runtime/NEON/NEFunctions.h"

Seems it doesn't find the file "arm_compute/runtime/NEON/NEFunctions.h", but if I go in that directory in my RPi: /home/pi/ComputeLibrary/arm_compute/runtime/NEON, the file is present!

Do you have any idea what's causing it?

[Torbjörn Olsson]

Hi Davide,

Did you ever find out what was wrong?

[Praneet Kala]

Having the same issue. I have the entire arm_compute folder on the same path as the example which i am trying to deploy and i still get this error. It was happening with the MemoryRegion.h file and then i added the entire folder (arm_compute/runtime) to the path in MATLAB and i get this error.

Any help would be appreciated

Answer 2

[Hariprasad Ravishankar]

Hi all,

The build error "fatal error: arm_compute/runtime/NEON/NEFunctions.h: No such file or directory" suggests that the ARM_COMPUTELIB environment variable may be incorrectly set.

This may perhaps be if you have setup the envrionment variables at the end of the ~/.bashrc script instead of the beginning.

In order to execute commands non-interactively, such as the codegen command from MATLAB on a host to compile on the Raspberry Pi, the environment variables need to be setup inside the

case $- in
    *i*) ;;
      *)
...
    return;;
esac

block in the ~/.bashrc script. This section can be found in the first few lines of the ~/.bashrc script for non-interactive,non-login shells.

Here is how it is setup for me for reference

# ~/.bashrc: executed by bash(1) for non-login shells.
# see /usr/share/doc/bash/examples/startup-files (in the package bash-doc)
# for examples

# If not running interactively
case $- in
    *i*) ;;
      *) export ARM_COMPUTELIB=~/ComputeLibrary
         export PATH=${PATH}:~/ComputeLibrary
         export LD_LIBRARY_PATH=~/opencv-3.2.0/build/lib:~/ComputeLibrary/lib:$LD_LIBRARY_PATH
         return;;
esac

# don't put duplicate lines or lines starting with space in the history.
# See bash(1) for more options
HISTCONTROL=ignoreboth

# append to the history file, don't overwrite it
shopt -s histappend

Alternatively, you can setup environment variables in the first few lines of the ~/.bashrc before the return;; statement

# ~/.bashrc: executed by bash(1) for non-login shells.
# see /usr/share/doc/bash/examples/startup-files (in the package bash-doc)
# for examples

export ARM_COMPUTELIB=~/ComputeLibrary
export PATH=${PATH}:~/ComputeLibrary
export LD_LIBRARY_PATH=~/opencv-3.2.0/build/lib:~/ComputeLibrary/lib:$LD_LIBRARY_PATH

# If not running interactively
case $- in
    *i*) ;;
      *) 
         return;;
esac

[Xiaoxing Wang]

Dear Hariprasad,

Thank you for your solution. It worked perfectly!

After the modification, I can now run the Segmentation Deep Network on my Raspberry Pi 3B+.

https://www.mathworks.com/help/signal/ug/code-generation-for-ecg-segmentation-using-deep-learning-on-raspberry-pi.html

Though it is redundant, below is my set up as a reference.

# ~/.bashrc: executed by bash(1) for non-login shells.
# see /usr/share/doc/bash/examples/startup-files (in the package bash-doc)
# for examples

export ARM_COMPUTELIB=~/ComputeLibrary
export PATH=${PATH}:~/ComputeLibrary
export LD_LIBRARY_PATH=~/opencv-3.2.0/build/lib:~/ComputeLibrary/lib:$LD_LIBRARY_PATH

# If not running interactively, don't do anything
case $- in
    *i*) ;;
      *) return;;
esac

...

[Xisheng Yu]

Hi,

I am trying the case you provided : Deploy Signal Segmentation Deep Network on Raspberry Pi. the configuration I am currently using is MATLAB2023b, Raspberry Pi 4B. arm compute library is 20.02.1. when doing a PIL, everything is fine. But when generating the executable file using MATLAB Coder, it compiles with an error, the error is a missing header file with the name“ rt_nonfinite.h”. Have you ever encountered this problem while using it? How did you solve it? Looking forward to your reply.

[Hariprasad Ravishankar]

Hi @Xisheng Yu, I'll look into it.

[Hariprasad Ravishankar]

Hi @Xisheng Yu,

I was able to reproduce the error. It seems that there are two issues with the ecgsegmentation_main.cpp shipping with the example. The #include "rt_nonfinite.h" is not requried, and the datatype of the variable 'out' should be cell_wrap_4.

I have fixed both issues and verified that the example works in R2023b. Please use the cpp file attached as part of the ZIP file to fix the issues.

I will ensure we fix this issue.

Hari

[Xisheng Yu]

Hi @Hariprasad Ravishankar,

Very happy to hear from you and that the problem was solved with the new file. Thank you very much and best wishes to you.

Answer 3

[Torbjörn Olsson]

Hi Davide,

while you are awaiting an answer from a cunning person, I thought I draw your attention to a minor issue that perhaps could be at least part of the compiling error. The paths seems to differ by one character between the one in the error message and the one you mention in the latter part of your query, i.e. "arm_compute/runtime/" vs. "arm_compute_runtime/".

[Davide Buellis]

Hi Torbjorn!

Thanks for your help! Sorry I made a typo, it is not "arm_compute_runtime/". I corrected the question.