Arm NN compute library build failure on Raspberry PI 4

[pauljainta]

Output of 'strings libarm_compute.so | grep arm_compute_version': arm_compute_version=v24.05 Build options: {'extra_cxx_flags': '-fPIC', 'Werror': '0', 'debug': '0', 'asserts': '0', 'neon': '1', 'opencl': '0', 'os': 'linux', 'arch': 'armv8a', 'examples': '1'} Git hash=b'a53ffdc87869399eb1ff2f9a20c69c8b962e27d2'

Platform: Raspberry PI 4

Operating System: Debian 12

Problem description: I have been trying to follow this documentation to build arm nn in my Raspberry PI 4, but it fails at the step

scons extra_cxx_flags="-fPIC" benchmark_tests=0 validation_tests=0 neon=1

I have attached an image of the failure logs.

[morgolock]

Hi @pauljainta

Could you please let us know what version of gcc you are running? Could you please share the full error message as text rather than an image?

[pauljainta]

Hi @morgolock

1. gcc (Debian 12.2.0-14) 12.2.0

2. I tried to reproduce the error and now it says :

scons: Reading SConscript files ...
Mkdir("build")
Using compilers:
CC  arm-linux-gnueabihf-gcc
CXX  arm-linux-gnueabihf-g++
ERROR: Compiler ' arm-linux-gnueabihf-g++' not found

In both of the cases wget https://dl.bintray.com/boostorg/release/1.64.0/source/boost_1_64_0.tar.bz2 was not working since the link was I guess expired. So I used wget https://archives.boost.io/release/1.64.0/source/boost_1_64_0.tar.bz2

[pauljainta]

A followup : I used sudo apt-get install g++-arm-linux-gnueabihf command to install g++-arm-linux-gnueabihf. Then I ran the scons command and it gave me the following error log

In file included from /usr/arm-linux-gnueabihf/include/c++/12/vector:63,
                 from include/CL/opencl.hpp:596,
                 from ./arm_compute/core/CL/OpenCL.h:44,
                 from ./arm_compute/core/CL/ICLArray.h:27,
                 from ./arm_compute/core/CL/CLTypes.h:27,
                 from ./arm_compute/core/CL/ICLTensor.h:27,
                 from ./arm_compute/runtime/CL/CLTensor.h:27,
                 from ./src/gpu/cl/operators/ClWinogradConv2d.h:27,
                 from src/gpu/cl/operators/ClWinogradConv2d.cpp:24:
In function ‘_ForwardIterator std::__do_uninit_copy(_InputIterator, _InputIterator, _ForwardIterator) [with _InputIterator = const pair<pair<int, int>, pair<int, int> >*; _ForwardIterator = pair<pair<int, int>, pair<int, int> >*]’,
    inlined from ‘static _ForwardIterator std::__uninitialized_copy<_TrivialValueTypes>::__uninit_copy(_InputIterator, _InputIterator, _ForwardIterator) [with _InputIterator = const std::pair<std::pair<int, int>, std::pair<int, int> >*; _ForwardIterator = std::pair<std::pair<int, int>, std::pair<int, int> >*; bool _TrivialValueTypes = false]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/stl_uninitialized.h:137:32,
    inlined from ‘_ForwardIterator std::uninitialized_copy(_InputIterator, _InputIterator, _ForwardIterator) [with _InputIterator = const pair<pair<int, int>, pair<int, int> >*; _ForwardIterator = pair<pair<int, int>, pair<int, int> >*]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/stl_uninitialized.h:185:15,
    inlined from ‘_ForwardIterator std::__uninitialized_copy_a(_InputIterator, _InputIterator, _ForwardIterator, allocator<_Tp>&) [with _InputIterator = const pair<pair<int, int>, pair<int, int> >*; _ForwardIterator = pair<pair<int, int>, pair<int, int> >*; _Tp = pair<pair<int, int>, pair<int, int> >]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/stl_uninitialized.h:372:37,
    inlined from ‘void std::vector<_Tp, _Alloc>::_M_range_initialize(_ForwardIterator, _ForwardIterator, std::forward_iterator_tag) [with _ForwardIterator = const std::pair<std::pair<int, int>, std::pair<int, int> >*; _Tp = std::pair<std::pair<int, int>, std::pair<int, int> >; _Alloc = std::allocator<std::pair<std::pair<int, int>, std::pair<int, int> > >]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/stl_vector.h:1690:33,
    inlined from ‘std::vector<_Tp, _Alloc>::vector(std::initializer_list<_Tp>, const allocator_type&) [with _Tp = std::pair<std::pair<int, int>, std::pair<int, int> >; _Alloc = std::allocator<std::pair<std::pair<int, int>, std::pair<int, int> > >]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/stl_vector.h:677:21,
    inlined from ‘bool arm_compute::opencl::{anonymous}::check_support_fast_math(const arm_compute::Size2D&, const arm_compute::Size2D&)’ at src/gpu/cl/operators/ClWinogradConv2d.cpp:94:84,
    inlined from ‘arm_compute::Status arm_compute::opencl::{anonymous}::validate_arguments(const arm_compute::ITensorInfo*, const arm_compute::ITensorInfo*, const arm_compute::ITensorInfo*, const arm_compute::ITensorInfo*, const arm_compute::PadStrideInfo&, const arm_compute::ActivationLayerInfo&, bool)’ at src/gpu/cl/operators/ClWinogradConv2d.cpp:131:9:
/usr/arm-linux-gnueabihf/include/c++/12/bits/stl_uninitialized.h:119:48: error: pointer used after ‘void operator delete(void*, std::size_t)’ [-Werror=use-after-free]
  119 |           for (; __first != __last; ++__first, (void)++__cur)
      |                                                ^~~~~~~~~~~~~
In file included from /usr/arm-linux-gnueabihf/include/c++/12/arm-linux-gnueabihf/bits/c++allocator.h:33,
                 from /usr/arm-linux-gnueabihf/include/c++/12/bits/allocator.h:46,
                 from /usr/arm-linux-gnueabihf/include/c++/12/string:41,
                 from ./arm_compute/core/CL/OpenCL.h:27:
In member function ‘void std::__new_allocator<_Tp>::deallocate(_Tp*, size_type) [with _Tp = std::pair<std::pair<int, int>, std::pair<int, int> >]’,
    inlined from ‘static void std::allocator_traits<std::allocator<_CharT> >::deallocate(allocator_type&, pointer, size_type) [with _Tp = std::pair<std::pair<int, int>, std::pair<int, int> >]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/alloc_traits.h:496:23,
    inlined from ‘void std::_Vector_base<_Tp, _Alloc>::_M_deallocate(pointer, std::size_t) [with _Tp = std::pair<std::pair<int, int>, std::pair<int, int> >; _Alloc = std::allocator<std::pair<std::pair<int, int>, std::pair<int, int> > >]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/stl_vector.h:387:19,
    inlined from ‘std::_Vector_base<_Tp, _Alloc>::~_Vector_base() [with _Tp = std::pair<std::pair<int, int>, std::pair<int, int> >; _Alloc = std::allocator<std::pair<std::pair<int, int>, std::pair<int, int> > >]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/stl_vector.h:366:15,
    inlined from ‘std::vector<_Tp, _Alloc>::~vector() [with _Tp = std::pair<std::pair<int, int>, std::pair<int, int> >; _Alloc = std::allocator<std::pair<std::pair<int, int>, std::pair<int, int> > >]’ at /usr/arm-linux-gnueabihf/include/c++/12/bits/stl_vector.h:733:7,
    inlined from ‘bool arm_compute::opencl::{anonymous}::check_support_fast_math(const arm_compute::Size2D&, const arm_compute::Size2D&)’ at src/gpu/cl/operators/ClWinogradConv2d.cpp:100:1,
    inlined from ‘arm_compute::Status arm_compute::opencl::{anonymous}::validate_arguments(const arm_compute::ITensorInfo*, const arm_compute::ITensorInfo*, const arm_compute::ITensorInfo*, const arm_compute::ITensorInfo*, const arm_compute::PadStrideInfo&, const arm_compute::ActivationLayerInfo&, bool)’ at src/gpu/cl/operators/ClWinogradConv2d.cpp:131:9:
/usr/arm-linux-gnueabihf/include/c++/12/bits/new_allocator.h:158:33: note: call to ‘void operator delete(void*, std::size_t)’ here
  158 |         _GLIBCXX_OPERATOR_DELETE(_GLIBCXX_SIZED_DEALLOC(__p, __n));
      |                                 ^
cc1plus: all warnings being treated as errors
scons: *** [build/src/gpu/cl/operators/ClWinogradConv2d.o] Error 1
scons: building terminated because of errors

[pauljainta]

logs.txt I have uploaded the full logs file here

[pauljainta]

@morgolock Did u find any solution?

[JuanMARM]

Hi @pauljainta , we will get back to you early next week when @morgolock is back in the office. I hope that is ok?

[morgolock]

Hi @pauljainta

It looks like you're trying to build ArmNN, if this is so I'd recommend downloading the ArmNN prebuilt binaries: https://github.com/ARM-software/armnn/releases/download/v24.05/MULTI_ISA-GCC11-ArmNN+ACL-linux-armv8a.tar.gz You could also get ACL prebuilt binaries https://github.com/ARM-software/ComputeLibrary/releases/download/v24.06/arm_compute-v24.06-bin-linux-arm64-v8a-cl.tar.gz

The error you get it seems to be a toolchain problem, could you please try building with scons extra_cxx_flags="-fPIC" benchmark_tests=0 validation_tests=0 neon=1 opencl=0 ?

You could also try downloading a different version of gcc from https://developer.arm.com/downloads/-/arm-gnu-toolchain-downloads