Cross-compiling with cmake for embedded systems QNX

shyney7 · May 13, 2020, 12:52pm

@glaringlee
Thank you so much for replying and trying to help! I really appreciate this!
This is also what I found in another repo on github while searching the web for similar errors.
So I have changed the CMakelists.txt file as follows:

set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=gnu++14 -stdlib=libstdc++ ${TORCH_CXX_FLAGS}")

But now I’m getting the following compiler errors:

github.com

shyney7/libtorch-cross-RTOS-test/blob/fa8f42964be0b4e194f402f8296909f0e438d470/Compilerlogs/comperr_with-stdlib=libstdc++_flag.md

```console
marcel@ubuntu:~/Documents/libtorch_cross_test/build$ cmake -DCMAKE_TOOLCHAIN_FILE=../cmake/qnx_7.0.0_linux_x86_64.cmake ..
-- The C compiler identification is GNU 5.4.0
-- The CXX compiler identification is GNU 5.4.0
-- Check for working C compiler: /home/marcel/qnx700/host/linux/x86_64/usr/bin/ntox86_64-gcc
-- Check for working C compiler: /home/marcel/qnx700/host/linux/x86_64/usr/bin/ntox86_64-gcc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /home/marcel/qnx700/host/linux/x86_64/usr/bin/ntox86_64-c++
-- Check for working CXX compiler: /home/marcel/qnx700/host/linux/x86_64/usr/bin/ntox86_64-c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Looking for pthread.h
-- Looking for pthread.h - found
-- Looking for pthread_create
-- Looking for pthread_create - found

This file has been truncated. show original

I have also opened an issue on github describing everything and with all the files from a repo I created.

github.com/pytorch/pytorch

Compiling errors when trying to cross-compile the C++ API for RTOS (QNX)

opened 09:12AM - 12 May 20 UTC

shyney7

module: build module: cpp triaged

I am currently trying to implement the Pytorch C++ API into production. RTOS are… indispensable in production environments so I'm trying to cross-compile with cmake for QNX Neutrino 7.0.0 (x86_64). However, when compiling a simple test program (empty main() function just including "#include <torch/torch.h>") I get the following compiler errors: https://github.com/shyney7/libtorch-cross-RTOS-test/blob/master/Compilerlogs/comperr-without-stdlib-flag.md I had to set the following command so that cmake finds the torch library in the host path. ` SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY BOTH)` This is how my **cmake Toolchain File** looks like: https://github.com/shyney7/libtorch-cross-RTOS-test/blob/master/cmake/qnx_7.0.0_linux_x86_64.cmake This is my **CMakeLists.txt** file: https://github.com/shyney7/libtorch-cross-RTOS-test/blob/master/CMakeLists.txt Unfortunately I am not a computer scientist but a mechanical engineer working at a research lab of a german university trying to implement libtorch into production. So I am not sure what the compiler errors mean. Thats why I searched the web for similar errors and found an issue of another repo here: https://github.com/eProsima/Fast-RTPS/issues/300 There it is mentioned that it is also needed to prevent the qcc compiler from using the LLVM Compiler Infrastructure from v1/memory by using the `-stdlib=libstdc++` flag. I also tried this by changing my CMakeLists.txt file to use the following flags: `set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=gnu++14 -stdlib=libstdc++ ${TORCH_CXX_FLAGS}")` But then I'm getting the following **compiler errors**: https://github.com/shyney7/libtorch-cross-RTOS-test/blob/master/Compilerlogs/comperr_with-stdlib%3Dlibstdc%2B%2B_flag.md In the announcement videos where Libtorch and the C++ API were introduced, it was always said that the C++ API is specifically for production environments. The fact is that in production environments you mostly work with real-time operating systems like QNX or FreeRTOS, so I'm sure that there is somehow a way to get libtorch working on QNX. Here is the repo with all the files: https://github.com/shyney7/libtorch-cross-RTOS-test Here is also a link to my first post on the discussion forums where I faced some issues with setting up the cmake toolchain file correctly: https://discuss.pytorch.org/t/cross-compiling-with-cmake-for-embedded-systems-qnx/79681 ## Environment - Libtorch Version : 1.5 Release - OS : cross-compiling on Ubuntu 18.04 (VM) for **QNX 7.0.0 (x84_64)** - How you installed PyTorch : libtorch 1.5 zip from pytorch.org (for linux tried with both Pre & cxx11 ABI) - Build command you used : `cmake -DCMAKE_TOOLCHAIN_FILE=../cmake/qnx_7.0.0_linux_x86_64.cmake ..` `make` - GNU version: 5.4.0 - CUDA/cuDNN version: none - GPU models and configuration: none - cmake version: 3.10.2 cc @malfet @yf225 @glaringlee

I’ve tried this by moving the libtorch folder to the QNX Target Environment as stated here (QNX Docs):

/home/marcel/qnx700/target/qnx7/x86_64/lib/libtorch/

And changed the line in the CMakelists.txt as follows:
list(APPEND CMAKE_PREFIX_PATH "${QNX_TARGET}/x86_64/lib/libtorch")

But this change did not lead to a succesful build and I’m getting the same compiler errors as shown above.
Compiler Errors