compilation errors on Raspberry Pi

[amahoneyLIT]

I'm trying to use XTENSOR_USE_XSIMD in my project, which otherwise compiles and runs fine on a Raspberry Pi 3B+ with up to date Raspbian Stretch

Using

• xtensor master
• xtl 0.6.4
• xsimd 7.2.3

The first error I get is

In file included from /usr/local/include/xsimd/types/xsimd_traits.hpp:15:0,
                 from /usr/local/include/xsimd/xsimd.hpp:14,
                 from /usr/local/include/xtensor/xtensor_config.hpp:64,
                 from /usr/local/include/xtensor/xlayout.hpp:15,
                 from /usr/local/include/xtensor/xshape.hpp:21,
                 from /usr/local/include/xtensor/xstrides.hpp:20,
                 from /usr/local/include/xtensor/xaccessible.hpp:13,
                 from /usr/local/include/xtensor/xview.hpp:24,
                 from ../src_daq/fft_functions.cpp:4:
/usr/local/include/xsimd/types/xsimd_types_include.hpp:74:4: warning: #warning "Please compile with SIMD instructions enabled or activate the fallback mode. (e.g. for x86 -march=native or for ARM -mfpu=neon)" [-Wcpp]
   #warning "Please compile with SIMD instructions enabled or activate the fallback mode. (e.g. for x86 -march=native or for ARM -mfpu=neon)"
    ^~~~~~~

which is strange because I've tried both -mfpu=neon and -mfpu=neon-vfpv4

Any ideas?

[wolfv]

are you cross-compiling? If not, -march=native should probably also work.

[amahoneyLIT]

I am getting the same error trying -march=native with and without -mfpu=neon

I am not cross-compiling. Compiling directly on the RPi. Using this same project on an arm64 dev kit with 64bit OS and compiler works. The RPi's Raspbian is 32bit

[amahoneyLIT]

getting closer by using flags from https://gist.github.com/fm4dd/c663217935dc17f0fc73c9c81b0aa845

-mcpu=cortex-a53 -mfloat-abi=hard -mfpu=neon-fp-armv8 -mneon-for-64bits

and I get only 3 errors now

/usr/local/include/xtensor/xassign.hpp:539: error: invalid use of incomplete type ‘xt::xcontainer<xt::xarray_container<xt::uvector<long unsigned int, xsimd::aligned_allocator<long unsigned int, 16u> >, (xt::layout_type)1, xt::svector<unsigned int, 4u, std::allocator<unsigned int>, true>, xt::xtensor_expression_tag> >::simd_return_type<long unsigned int> {aka class xsimd::batch<long unsigned int, 4u>}’
             e1.template store_simd<lhs_align_mode>(i, e2.template load_simd<rhs_align_mode, value_type>(i));
             ^~

/usr/local/include/xtensor/xcontainer.hpp:714: error: return type ‘xt::xcontainer<xt::xarray_container<xt::uvector<long unsigned int, xsimd::aligned_allocator<long unsigned int, 16u> >, (xt::layout_type)1, xt::svector<unsigned int, 4u, std::allocator<unsigned int>, true>, xt::xtensor_expression_tag> >::simd_return_type<long unsigned int> {aka class xsimd::batch<long unsigned int, 4u>}’ is incomplete
     inline auto xcontainer<D>::load_simd(size_type i) const
                 ^~~~~~~~~~~~~

/usr/local/include/xtensor/xcontainer.hpp:718: error: invalid use of incomplete type ‘xsimd::simd_return_type<long unsigned int, long unsigned int, 4u> {aka class xsimd::batch<long unsigned int, 4u>}’
         return xt_simd::load_simd<value_type, requested_type>(&(storage()[i]), align_mode());
                ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

will poke at these flags some more

[wolfv]

This might now be a problem from our side, if the type or type conversion is not correctly implemented...

Can you try with only integer for example?

[amahoneyLIT]

Based on those errors sounding integer-ish, maybe you meant the opposite - I tried removing all the integer arrays from my code, including

• xarray<unsigned long>
• xarray<unsigned long long>
• xarray<size_t>

and now it compiles.

[wolfv]

hey, cool, that's good to hear.

we should keep this issue open to investigate in the future. it might be related to 64-bit integers and the NEON version.

[JohanMabille]

Out of curiosity, since you're compiling on a 32bit system, can you print the results of the following:

#include <type_traits>
#include <cstdint>
#include <iostream>

int main(int argc, char* argv[])
{
    using std::uint32_t;
    using std::uint64_t;

    std::cout << "unsigned long == uint32_t? ";
    std::cout << std::is_same<unsigned long, uint32_t>::value << std::endl;

    std::cout << "unsigned long long == uint32_t? ";
    std::cout << std::is_same<unsigned long long, uint32_t>::value << std::end;

    std::cout << "size_t == uint32_t? ";
    std::cout << std::is_same<size_t, uint32_t>::value << std::endl;

    std::cout << "unsigned long == uint64_t? ";
    std::cout << std::is_same<unsigned long, uint64_t>::value << std::endl;

    std::cout << "unsigned long long == uint64_t? ";
    std::cout << std::is_same<unsigned long long, uint64_t>::value << std::endl;

    std::cout << "size_t == uint64_t? ";
    std::cout << std::is_same<size_t, uint64_t>::value << std::endl;

    return 0;
}

[amahoneyLIT]

@JohanMabille the second check does not compile

/home/pi/dev/cpp-scratch/main.cpp:14: error: no match for ‘operator<<’ (operand types are ‘std::basic_ostream<char>::__ostream_type {aka std::basic_ostream<char>}’ and ‘<unresolved overloaded function type>’)
     std::cout << std::is_same<unsigned long long, uint32_t>::value << std::end;
     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~

commenting that out, the rest of the output is

unsigned long == uint32_t? 0
size_t == uint32_t? 1
unsigned long == uint64_t? 0
unsigned long long == uint64_t? 1
size_t == uint64_t? 0