support for armv6l?

[prettyflyforabeeguy]

Is there support for armv6l (i.e. Raspberry Pi Zero W) If no support exists currently, is there a way I can build this natively to support this ARM32 device?

More about the device I'm attempting to use the tflite_runtime on: $ python3 --version Python 3.7.3

$ cat /etc/os-release PRETTY_NAME="Raspbian GNU/Linux 10 (buster)" NAME="Raspbian GNU/Linux" VERSION_ID="10" VERSION="10 (buster)" VERSION_CODENAME=buster ID=raspbian ID_LIKE=debian HOME_URL="http://www.raspbian.org/" SUPPORT_URL="http://www.raspbian.org/RaspbianForums" BUG_REPORT_URL="http://www.raspbian.org/RaspbianBugs"

$ cat /proc/cpuinfo processor : 0 model name : ARMv6-compatible processor rev 7 (v6l) BogoMIPS : 697.95 Features : half thumb fastmult vfp edsp java tls CPU implementer : 0x41 CPU architecture: 7 CPU variant : 0x0 CPU part : 0xb76 CPU revision : 7

Hardware        : BCM2835
Revision        : 9000c1
Serial          : 00000000fdca881b
Model           : Raspberry Pi Zero W Rev 1.1

$ uname -a Linux experimentalpiZero 5.4.79+ #1373 Mon Nov 23 13:18:15 GMT 2020 armv6l GNU/Linux

$ uname -m armv6l

$ gcc -v Using built-in specs. COLLECT_GCC=gcc COLLECT_LTO_WRAPPER=/usr/lib/gcc/arm-linux-gnueabihf/8/lto-wrapper Target: arm-linux-gnueabihf Configured with: ../src/configure -v --with-pkgversion='Raspbian 8.3.0-6+rpi1' --with-bugurl=file:///usr/share/doc/gcc- 8/README.Bugs --enable-languages=c,ada,c++,go,d,fortran,objc,obj-c++ --prefix=/usr --with-gcc-major-version-only -- program-suffix=-8 --program-prefix=arm-linux-gnueabihf- --enable-shared --enable-linker-build-id --libexecdir=/usr/lib -- without-included-gettext --enable-threads=posix --libdir=/usr/lib --enable-nls --enable-bootstrap --enable-clocale=gnu -- enable-libstdcxx-debug --enable-libstdcxx-time=yes --with-default-libstdcxx-abi=new --enable-gnu-unique-object -- disable-libitm --disable-libquadmath --disable-libquadmath-support --enable-plugin --with-system-zlib --with-target- system-zlib --enable-objc-gc=auto --enable-multiarch --disable-sjlj-exceptions --with-arch=armv6 --with-fpu=vfp --with- float=hard --disable-werror --enable-checking=release --build=arm-linux-gnueabihf --host=arm-linux-gnueabihf -- target=arm-linux-gnueabihf Thread model: posix gcc version 8.3.0 (Raspbian 8.3.0-6+rpi1)

[Namburger]

Hello, this issue have been addressed before and users have been success in doing this. You can take a look at this thread for more info: https://github.com/google-coral/edgetpu/issues/229#issuecomment-704576309

[prettyflyforabeeguy]

Thank you for the quick reply, but I'm am not using docker and I am not trying to use libedgetpu, however I'm willing to try anything if I can get tflite working on the armv6l I'm trying to follow the instructions https://www.tensorflow.org/lite/guide/build_rpi and https://www.tensorflow.org/lite/guide/python I can import tensorflow just fine, but trying to use the tflite_runtime is where things fall appart. I get an illegal instruction the moment I try import tflite_runtime.interpreter as tflite

~/tensorflow_src $ python3 -c 'print(import("tflite_runtime").__git_version__)' 0.6.0-76902-gd855adfc5a

~/tensorflow_src $ git rev-parse 0.6.0-76902-gd855adfc5a d855adfc5a0195788bf5f92c3c7352e638aa1109

I'm trying to piece the conversation together from the #229 comment, but I'm not quite sure i'm followling the necessary steps. I appreciate any addtional guidence you may have!

[Namburger]

@prettyflyforabeeguy hello, docker is specifically and it is very important piece for building and you won't be able to do anything without libedgetpu.so. Basically you'll need to compile the needed libraries on an x86_64 machineand cross compile it for armv6l. We already provided compatible docker containers to get this going so without that it'll only make it harder for you to get thins going!!

Here is the gist:

1) The 3 peaces that is necessary for you to use our HW from here on out is :

• libedgetpu
• tflite_runtime
• tensorflow-lite.a

2) Since tensorlfow-lite.a is statically linked to both libedgetpu AND tflite_runtime, you won't need that for your application it self, however you'll NEED to make sure that libedgetpu and the tflite_runtime are both linked against the same exact tensorflow commit (hence we use git rev-parse to find out the exact tensorflow commit).

Okay, so here is where the fun begins :)

I get an illegal instruction the moment I try import tflite_runtime.interpreter as tflite

That is because the tflite_runtime package was not built to supports the RPI0 with armv6l architecture. So you'll need to build your own package, the instruction to build it is [here](https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/tools/pip_package. You'll need to dig into those build fles to find out which instruction is causing the illegal instructions on your pi. We don't support this arch anymore so we can't fix those built file to bring support back, unfortunately..... HOWEVER, in #229, @rhadnum already mentioned that he found a known working tflite_runtime package that is working with the pi0 and even include the download link. Therefore my suggestion is to remove the currently installed package and start with that link. Then get the tensorflow commit version (same command you posted).

~/tensorflow_src $ python3 -c 'print(__import__("tflite_runtime").git_version)'
should be different from what you posted since this is a different package version

~/tensorflow_src $ git rev-parse 0.6.0-76902-gd855adfc5a
should be different from what you posted since this is a different package version

Now that you have a working tflite_runtime package, you'll need a libedgetpu.so. That's why you'll need to build it from our libedgetpu repo. Just make sure that when you build, you are linking against the same tensorflow version by replacing that line with the commit you got from that above git-revparse command. Then build like this:

$ DOCKER_CPUS="armv6" DOCKER_TARGETS=libedgetpu make docker-build

A couple of libedgetpu.so* binaries will be in the out directory. You'll need to install them to your library paths and that should be it as far as libraries installation.

One last step, which may not be necessary but just for assurance. Sometimes, linux doesn't allows user to have plugdev group by default, and this is necessary for interacting with the edgetpu. So I'd also do this:

$ sudo usermod -aG plugdev $USER
$ sudo reboot now

That is all!

[prettyflyforabeeguy]

@Namburger you are my new best friend whom I've never met!!! Somehow I missed the link to the .whl file that rhadnum had shared. Thank you thank you thank you for taking the time to write up such a wonderful explanation but in the end I didn't need to do any of that. I simply downloaded the tflite_runtime-2.3.1-cp37-cp37m-linux_armv6l.whl and installed it and everything works. This is so awesome!

[Namburger]

Nicee and no probs :)