Compillation breaks with clang8: no matching constructor for initialization of 'fbgemm::ReQuantizeOutput<false, fbgemm::QuantizationGranularity::TENSOR, float>'

[yurivict]

/usr/ports/science/py-pytorch/work/pytorch-1.3.0/aten/src/ATen/native/quantized/cpu/qlinear.cpp:155:15: error: no matching constructor for initialization of 'fbgemm::ReQuantizeOutput<false, fbgemm::QuantizationGranularity::TENSOR, float>'
              outputProcObj(
              ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/aten/src/ATen/native/quantized/cpu/qlinear.cpp:337:14: note: in instantiation of member function 'at::native::(anonymous namespace)::QLinearInt8<false>::fbgemm_linear' requested here
      return fbgemm_linear(
             ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/aten/src/ATen/core/boxing/kernel_functor.h:194:12: note: in instantiation of member function 'at::native::(anonymous namespace)::QLinearInt8<false>::operator()' requested here
    return (*functor)(ivalue_to_arg<guts::remove_cv_t<guts::remove_reference_t<guts::typelist::element_t<ivalue_arg_indices, IValueArgTypes>>>, AllowDeprecatedTypes>(
           ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/aten/src/ATen/core/boxing/kernel_functor.h:202:12: note: in instantiation of function template specialization 'c10::detail::call_functor_with_args_from_stack_<at::native::(anonymous namespace)::QLinearInt8<false>, false, 0, 1, 2, 3>' requested here
    return call_functor_with_args_from_stack_<Functor, AllowDeprecatedTypes>(functor, stack, guts::make_index_sequence<num_ivalue_args>());
           ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/aten/src/ATen/core/boxing/kernel_functor.h:234:21: note: in instantiation of function template specialization 'c10::detail::call_functor_with_args_from_stack<at::native::(anonymous namespace)::QLinearInt8<false>, false>' requested here
      auto output = call_functor_with_args_from_stack<KernelFunctor, AllowDeprecatedTypes>(functor_, stack);
                    ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/aten/src/ATen/core/boxing/KernelFunction.h:205:76: note: in instantiation of member function 'c10::detail::wrap_kernel_functor_boxed<at::native::(anonymous namespace)::QLinearInt8<false>, false, void>::call' requested here
      &detail::wrap_kernel_functor_boxed<KernelFunctor, AllowLegacyTypes>::call,
                                                                           ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/aten/src/ATen/core/op_registration/op_registration.h:154:25: note: in instantiation of function template specialization 'c10::KernelFunction::makeFromUnboxedFunctorFactory<at::native::(anonymous namespace)::QLinearInt8<false>, false>' requested here
        KernelFunction::makeFromUnboxedFunctorFactory<KernelFunctor>(detail::KernelFactory<KernelFunctor, guts::decay_t<ConstructorParameters>...>(std::forward<ConstructorParameters>(constructorParameters)...)),
                        ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/aten/src/ATen/native/quantized/cpu/qlinear.cpp:357:49: note: in instantiation of function template specialization 'c10::RegisterOperators::Options::kernel<at::native::(anonymous namespace)::QLinearInt8<false>>' requested here
            torch::RegisterOperators::options().kernel<QLinearInt8<false>>(
                                                ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/third_party/fbgemm/include/fbgemm/Fbgemm.h:1158:3: note: candidate constructor not viable: requires at most 10 arguments, but 11 were provided
  ReQuantizeOutput(
  ^
/usr/ports/science/py-pytorch/work/pytorch-1.3.0/third_party/fbgemm/include/fbgemm/Fbgemm.h:1134:18: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 11 were provided
class FBGEMM_API ReQuantizeOutput {
                 ^

Version 1.3.0
clang8
FreeBSD 12

cc @jerryzh168 @jianyuh @dzhulgakov @raghuramank100

[lly-zero-one]

@dskhudia @jianyuh Could you help to take a look? Seems to be building issue.

[yurivict]

This issue stops me from creating a FreeBSD port for PyTorch.

[jianyuh]

@lly-zero-one @dskhudia We currently don't have a FreeBSD 12 for reproduce. It might take some time to look at this issue. @ezyang Do we have the OSS CI test for FreeBSD?

For now, you might disable FBGEMM for building PyTorch on FreeBSD:
USE_FBGEMM=0 python setup.py install .

[ezyang]

We don't have any FreeBSD CI, and there are no plans in our roadmap for it.