Small change required to successfully use XTL with nvcc CUDA compiler #183
Description
Hi,
I was hoping that since XTL works with any C++14 compiler, the compilation of host only CUDA 10 code should work fine with XTL. Normally, I would not attempt this and instead find a replacement that is out-of-the-box compatible with CUDA. However, I need to write some CUDA as part of a large codebase which uses xtensor, openmc. It does seem that xtensor isn't far off from being able to compile host code with nvcc, I only obtain the following errors when compiling the most basic of xtensor programs (the first one on its readthedocs):
gavin@gpad:~/scratch$ nvcc xtensor_example.cu
/usr/local/include/xtl/xsequence.hpp:136:87: error: expected primary-expression before ‘unsigned’
struct sequence_forwarder_impl<R, A, void_t<decltype(std::declval<R>().resize(std::size_t()))>>
^~~~~~~~
/usr/local/include/xtl/xvariant_impl.hpp: In function ‘constexpr decltype(auto) mpark::visit(Visitor&&, Vs&& ...)’:
/usr/local/include/xtl/xvariant_impl.hpp:2676:96: error: parameter packs not expanded with ‘...’:
return (detail::all({!vs.valueless_by_exception()...})
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ^
/usr/local/include/xtl/xvariant_impl.hpp:2676:96: note: ‘vs’
/usr/local/include/xtensor/xstrided_view_base.hpp: In instantiation of ‘void xt::detail::strided_view_args<adj_strides_policy>::fill_args(const S&, ST&&, std::size_t, xt::layout_type, const V&) [with S = xt::svector<long unsigned int, 4, std::allocator<long unsigned int>, true>; ST = const xt::svector<long int, 4, std::allocator<long int>, true>&; V = std::vector<mpark::variant<long int, xt::xrange_adaptor<xt::placeholders::xtuph, long int, long int>, xt::xrange_adaptor<long int, xt::placeholders::xtuph, long int>, xt::xrange_adaptor<long int, long int, xt::placeholders::xtuph>, xt::xrange_adaptor<long int, xt::placeholders::xtuph, xt::placeholders::xtuph>, xt::xrange_adaptor<xt::placeholders::xtuph, long int, xt::placeholders::xtuph>, xt::xrange_adaptor<xt::placeholders::xtuph, xt::placeholders::xtuph, long int>, xt::xrange_adaptor<long int, long int, long int>, xt::xrange_adaptor<xt::placeholders::xtuph, xt::placeholders::xtuph, xt::placeholders::xtuph>, xt::xrange<long int>, xt::xstepped_range<long int>, xt::xall_tag, xt::xellipsis_tag, xt::xnewaxis_tag> >; adj_strides_policy = xt::detail::no_adj_strides_policy; std::size_t = long unsigned int]’:
/usr/local/include/xtensor/xstrided_view.hpp:593:1: required from ‘auto xt::strided_view(E&&, const xstrided_slice_vector&) [with E = const xt::xarray_container<xt::uvector<double, std::allocator<double> >, (xt::layout_type)1, xt::svector<long unsigned int, 4, std::allocator<long unsigned int>, true>, xt::xtensor_expression_tag>&; xt::xstrided_slice_vector = std::vector<mpark::variant<long int, xt::xrange_adaptor<xt::placeholders::xtuph, long int, long int>, xt::xrange_adaptor<long int, xt::placeholders::xtuph, long int>, xt::xrange_adaptor<long int, long int, xt::placeholders::xtuph>, xt::xrange_adaptor<long int, xt::placeholders::xtuph, xt::placeholders::xtuph>, xt::xrange_adaptor<xt::placeholders::xtuph, long int, xt::placeholders::xtuph>, xt::xrange_adaptor<xt::placeholders::xtuph, xt::placeholders::xtuph, long int>, xt::xrange_adaptor<long int, long int, long int>, xt::xrange_adaptor<xt::placeholders::xtuph, xt::placeholders::xtuph, xt::placeholders::xtuph>, xt::xrange<long int>, xt::xstepped_range<long int>, xt::xall_tag, xt::xellipsis_tag, xt::xnewaxis_tag> >]’
/usr/local/include/xtensor/xio.hpp:260:31: required from ‘void xt::detail::recurser_run(F&, const E&, xt::xstrided_slice_vector&, std::size_t) [with F = xt::detail::printer<xt::xarray_container<xt::uvector<double, std::allocator<double> >, (xt::layout_type)1, xt::svector<long unsigned int, 4, std::allocator<long unsigned int>, true>, xt::xtensor_expression_tag>, void>; E = xt::xarray_container<xt::uvector<double, std::allocator<double> >, (xt::layout_type)1, xt::svector<long unsigned int, 4, std::allocator<long unsigned int>, true>, xt::xtensor_expression_tag>; xt::xstrided_slice_vector = std::vector<mpark::variant<long int, xt::xrange_adaptor<xt::placeholders::xtuph, long int, long int>, xt::xrange_adaptor<long int, xt::placeholders::xtuph, long int>, xt::xrange_adaptor<long int, long int, xt::placeholders::xtuph>, xt::xrange_adaptor<long int, xt::placeholders::xtuph, xt::placeholders::xtuph>, xt::xrange_adaptor<xt::placeholders::xtuph, long int, xt::placeholders::xtuph>, xt::xrange_adaptor<xt::placeholders::xtuph, xt::placeholders::xtuph, long int>, xt::xrange_adaptor<long int, long int, long int>, xt::xrange_adaptor<xt::placeholders::xtuph, xt::placeholders::xtuph, xt::placeholders::xtuph>, xt::xrange<long int>, xt::xstepped_range<long int>, xt::xall_tag, xt::xellipsis_tag, xt::xnewaxis_tag> >; std::size_t = long unsigned int]’
/usr/local/include/xtensor/xio.hpp:763:21: required from ‘std::ostream& xt::pretty_print(const xt::xexpression<D>&, std::ostream&) [with E = xt::xarray_container<xt::uvector<double, std::allocator<double> >, (xt::layout_type)1, xt::svector<long unsigned int, 4, std::allocator<long unsigned int>, true>, xt::xtensor_expression_tag>; std::ostream = std::basic_ostream<char>]’
/usr/local/include/xtensor/xio.hpp:776:20: required from ‘std::ostream& xt::operator<<(std::ostream&, const xt::xexpression<D>&) [with E = xt::xarray_container<xt::uvector<double, std::allocator<double> >, (xt::layout_type)1, xt::svector<long unsigned int, 4, std::allocator<long unsigned int>, true>, xt::xtensor_expression_tag>; std::ostream = std::basic_ostream<char>]’
xtensor_example.cu:18:15: required from here
/usr/local/include/xtensor/xstrided_view_base.hpp:897:6: error: ‘void info’ has incomplete type
auto info = xtl::visit(slice_getter, slices[i]);It seems that a relatively small number of changes could make this work. While I do know some stuff about templates, my knowledge isn't good enough to identify the origin of these errors. I imagine there is some C++17 feature being used here that nvcc doesn't recognize. Since xtensor is supposed to work with C++14, and nvcc is supposed to work with C++14 as well, I thought perhaps this would merit raising an issue.
Any help would be MASSIVELY appreciated.