Building ESP32-S3 Firmware VS Code Mac Terminal

[TGiles1998]

The following commands were run on macOS Monterey - version 12.1 Terminal on VS Code

git clone https://github.com/mocleiri/tensorflow-micropython-examples.git

Then Following the build_esp32s3.yml in the .github/workflows folder.

cd /tensorflow-micropython-examples

git submodule init git submodule update --recursive cd micropython git submodule update --init lib/axtls git submodule update --init lib/berkeley-db-1.xx cd ..

IDF_COMMIT=142bb32c50fa9875b8b69fa539a2d59559460d72 echo "esp-idf-commit=$IDF_COMMIT" >> $GITHUB_ENV TFLM_COMMIT=$(git submodule status tensorflow | awk '{print ($1)}') echo "tflm-commit=$TFLM_COMMIT" >> $GITHUB_ENV

source ./micropython/tools/ci.sh && ci_esp32_idf44_setup

source ./esp-idf/export.sh

pip3 install Pillow pip3 install Wave

echo "Regenerating microlite/tfm directory" rm -rf ./micropython-modules/microlite/tflm

cd ./tensorflow

../micropython-modules/microlite/prepare-tflm-esp.sh

cd ..

source ./esp-idf/export.sh

cd ./micropython

echo "make -C mpy-cross V=1 clean all"

make -C mpy-cross V=1 clean all

cd ..

echo "cd ./boards/esp32/MICROLITE_S3" cd ./boards/esp32/MICROLITE_S3

echo "Building MICROLITE_S3" rm -rf build idf.py clean build

FAILED:

/tensorflow-micropython-examples/micropython/ports/esp32/main.c:38:10: fatal error: soc/cpu.h: No such file or directory

include "soc/cpu.h"

      ^~~~~~~~~~~

compilation terminated. /tensorflow-micropython-examples/micropython/ports/esp32/gccollect.c:37:10: fatal error: soc/cpu.h: No such file or directory

include "soc/cpu.h"

      ^~~~~~~~~~~

compilation terminated. /tensorflow-micropython-examples/micropython/ports/esp32/modmachine.c:46:10: fatal error: esp32s3/clk.h: No such file or directory

include "esp32s3/clk.h"

      ^~~~~~~~~~~~~~~

compilation terminated.

returned non-zero exit status 1. ninja: build stopped: subcommand failed. ninja failed with exit code 1

I can find all of these files by hand in the downloaded ESP-IDF in the repository but the compiler just can't seem to pick them up. any suggestions would be fantastic thank you!

[TGiles1998]

Copying and pasting the missing individual folder and files into the board root folder does seem to partially solve it but then produces further errors:

[1116/1606] Building C object esp-idf/main/CMakeFiles/idf_main.dir/...ensorflow-micropython-examples/micropython/extmod/moduwebsocket.c.ob cc1: warning: command line option '-fno-rtti' is valid for C++/ObjC++ but not for C [1118/1606] Building C object esp-idf/main/CMakeFiles/idf_main.dir/.../tensorflow-micropython-examples/micropython/extmod/uos_dupterm.c.ob cc1: warning: command line option '-fno-rtti' is valid for C++/ObjC++ but not for C [1120/1606] Building C object esp-idf/main/CMakeFiles/idf_main.dir/...1/tensorflow-micropython-examples/micropython/extmod/modwebrepl.c.ob cc1: warning: command line option '-fno-rtti' is valid for C++/ObjC++ but not for C [1121/1606] Building C object esp-idf/main/CMakeFiles/idf_main.dir/.../tensorflow-micropython-examples/micropython/extmod/utime_mphal.c.ob cc1: warning: command line option '-fno-rtti' is valid for C++/ObjC++ but not for C ...

etc for all of the remaining build lines

[mocleiri]

My development environment is windows but I'm doing most work in visual studio code running in windows subsystem for linux so I'm unable to help directly.

However if you are able to figure out the steps to build on MacOS please feel free to contribute the steps and we can improve the manual build instructions to include for mac os.

The -fno-rtti error is not an error per-se. I need to improve the microlite cmake file to seperate the C++ build from the C build. CMake is passing that option to gcc which is complaining that its not a C option. But I added for the C++ code where it is needed.

That line is emitted right now for all esp32 builds in linux.

If you get a firmware generated at the end of the build then I think you have a working build.

[TGiles1998]

Thank you for your reply! I will focus on the mac build and reply back to the thread so it can be added. Thanks again!

[TGiles1998]

Solved: Ive solved it using latest IDF-4.4 and making sure all of the toolchain had been properly added to the bash Profile. Ive added a pull request with a .md document detailing the process.

The Process is as pasted below:

Build Tensorflow Micropython on MAC OS using Bash Terminal

Tested on: macOS Monterey Version 12.1

Prerequisites

Properly Installing ESP-IDF (Taken from: https://docs.espressif.com/projects/esp-idf/en/v3.3/get-started-cmake/macos-setup.html)

sudo easy_install pip

pip install --user pyserial

brew install cmake ninja

brew install ccache

Download: https://dl.espressif.com/dl/xtensa-esp32-elf-osx-1.22.0-80-g6c4433a-5.2.0.tar.gz

mkdir -p ~/esp cd ~/esp tar -xzf ~/Downloads/xtensa-esp32-elf-osx-1.22.0-80-g6c4433a-5.2.0.tar.gz

Make Toolchain Accessible from Bash Terminal

Open your /bash.profile file by navigating to usr directory and pressing down the keys "Command, Shift, FullStop" it shoudl appear.

Add to your profile:

export PATH=$HOME/esp/xtensa-esp32-elf/bin:$PATH alias get_esp32="export PATH=$HOME/esp/xtensa-esp32-elf/bin:$PATH"

Git Clone ESP-IDF Version 4.4

cd ~/esp git clone -b v4.4 --recursive https://github.com/espressif/esp-idf.git esp-idf

Update the Submodules

cd esp-idf git submodule update --init

Reopen your /bash.profile and add the following:

export IDF_PATH=~/esp/esp-idf export PATH="$IDF_PATH/tools:$PATH"

Install the required python packages

python3 -m pip install --user -r $IDF_PATH/requirements.txt

Download the Tensorflow Micropython Examples Repo

git clone https://github.com/mocleiri/tensorflow-micropython-examples.git

Building ESP32S3

Building the TensorFlow Firmware

cd /tensorflow-micropython-examples

git submodule init git submodule update --recursive cd micropython git submodule update --init lib/axtls git submodule update --init lib/berkeley-db-1.xx cd ..

Get Cache Keys

IDF_COMMIT=142bb32c50fa9875b8b69fa539a2d59559460d72 TFLM_COMMIT=$(git submodule status tensorflow | awk '{print ($1)}')

Setup IDF within Repo (Another Option)

source ./micropython/tools/ci.sh && ci_esp32_idf44_setup

source ./esp-idf/export.sh

pip3 install Pillow pip3 install Wave

rm -rf ./micropython-modules/microlite/tflm

cd ./tensorflow

../micropython-modules/microlite/prepare-tflm-esp.sh

cd ..

Build Micropython Cross-Compiler

source ./esp-idf/export.sh cd ./micropython make -C mpy-cross V=1 clean all

cd ..

Build Generic ESP32S3 8MB FLASH

source ./esp-idf/export.sh

(OR: get_esp32 get_idf)

cd ./boards/esp32/MICROLITE_S3

rm -rf build idf.py clean build

To build other boards e.g. GENERIC_SPIRAM replace with Board definition

cd ./boards/esp32/MICROLITE_S3_SPIRAM

If succesful you should see a reult similar to this:

/micropython.bin or run 'idf.py -p (PORT) flash' bootloader @0x000000 18640 ( 14128 remaining) partitions @0x008000 3072 ( 1024 remaining) application @0x010000 1191120 ( 840496 remaining) total 1256656

[mocleiri]

The corresponding pull request for this was #66 it was applied in d083f23ad32a16bf40b7383a69bf98ab1c6c9281.