complex workaround does not work for TensorFlow

[iotamudelta]

The complex workaround branch does not work for complex reduction kernel in TensorFlow. Using it in the TF make system and enabling either reduction_ops_gpu_complex64 or complex128 ends with compilation errors like

In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:754:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:870:5: note: in instantiation of function template specialization 'tensorflow::functor::Launch3DXZReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    Launch3DXZReduction(ctx, out, in, in_dim0, in_dim1, in_dim2, op, init,
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:896:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:110:17: error: no matching constructor for initialization of 'output_type [4]'
    output_type values[ItemsPerThread];
                ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel_impl<256, 4, true, std::complex<float>, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    block_reduce_kernel_impl<BlockSize, ItemsPerThread, WithInitialValue, ResultType>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:181:37: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel<256, 4, true, std::complex<float>, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
            HIP_KERNEL_NAME(detail::block_reduce_kernel<
                                    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:302:20: note: in instantiation of function template specialization 'rocprim::detail::reduce_impl<256, 4, true, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::reduce_impl<block_size, items_per_thread, true>(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_reduce.hpp:91:27: note: in instantiation of function template specialization 'rocprim::reduce<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        return ::rocprim::reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:556:43: note: in instantiation of function template specialization 'hipcub::DeviceReduce::Reduce<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceReduce::Reduce(
                                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:858:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchScalarReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *>' requested here
    LaunchScalarReduction(ctx, out, in, in_size, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:931:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, TransformOutputIterator<T, T, DividesBy<T>>, T*,
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:111:17: error: no matching constructor for initialization of 'output_type' (aka 'std::complex<float>')
    output_type output_value;
                ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:596:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:861:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchRowReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *>' requested here
    LaunchRowReduction(ctx, out, in, in_dim0, in_dim1, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:931:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, TransformOutputIterator<T, T, DividesBy<T>>, T*,
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:754:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:870:5: note: in instantiation of function template specialization 'tensorflow::functor::Launch3DXZReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *>' requested here
    Launch3DXZReduction(ctx, out, in, in_dim0, in_dim1, in_dim2, op, init,
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:931:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, TransformOutputIterator<T, T, DividesBy<T>>, T*,
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:110:17: error: no matching constructor for initialization of 'output_type [4]'
    output_type values[ItemsPerThread];
                ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel_impl<256, 4, false, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    block_reduce_kernel_impl<BlockSize, ItemsPerThread, WithInitialValue, ResultType>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:147:37: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel<256, 4, false, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
            HIP_KERNEL_NAME(detail::block_reduce_kernel<
                                    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:302:20: note: in instantiation of function template specialization 'rocprim::detail::reduce_impl<256, 4, true, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    return detail::reduce_impl<block_size, items_per_thread, true>(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_reduce.hpp:91:27: note: in instantiation of function template specialization 'rocprim::reduce<std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
        return ::rocprim::reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:556:43: note: in instantiation of function template specialization 'hipcub::DeviceReduce::Reduce<std::complex<float> *, std::complex<float> *, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceReduce::Reduce(
                                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:858:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchScalarReduction<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchScalarReduction(ctx, out, in, in_size, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:1036:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Prod<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:111:17: error: no matching constructor for initialization of 'output_type' (aka 'std::complex<float>')
    output_type output_value;
                ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:110:17: error: no matching constructor for initialization of 'output_type [4]'
    output_type values[ItemsPerThread];
                ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel_impl<256, 4, true, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    block_reduce_kernel_impl<BlockSize, ItemsPerThread, WithInitialValue, ResultType>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:181:37: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel<256, 4, true, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
            HIP_KERNEL_NAME(detail::block_reduce_kernel<
                                    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:302:20: note: in instantiation of function template specialization 'rocprim::detail::reduce_impl<256, 4, true, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    return detail::reduce_impl<block_size, items_per_thread, true>(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_reduce.hpp:91:27: note: in instantiation of function template specialization 'rocprim::reduce<std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
        return ::rocprim::reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:556:43: note: in instantiation of function template specialization 'hipcub::DeviceReduce::Reduce<std::complex<float> *, std::complex<float> *, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceReduce::Reduce(
                                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:858:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchScalarReduction<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchScalarReduction(ctx, out, in, in_size, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:1036:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Prod<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:111:17: error: no matching constructor for initialization of 'output_type' (aka 'std::complex<float>')
    output_type output_value;
                ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:596:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:861:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchRowReduction<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchRowReduction(ctx, out, in, in_dim0, in_dim1, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:1036:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Prod<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:754:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:870:5: note: in instantiation of function template specialization 'tensorflow::functor::Launch3DXZReduction<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    Launch3DXZReduction(ctx, out, in, in_dim0, in_dim1, in_dim2, op, init,
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:1036:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Prod<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
1 warning and 16 errors generated.
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:20:
In file included from ./third_party/eigen3/unsupported/Eigen/CXX11/Tensor:1:
In file included from external/eigen_archive/unsupported/Eigen/CXX11/Tensor:14:
In file included from external/eigen_archive/unsupported/Eigen/CXX11/../../../Eigen/Core:69:
In file included from /opt/rocm/include/hip/math_functions.h:32:
In file included from /opt/rocm/include/hip/hcc_detail/math_functions.h:31:
In file included from /opt/rocm/include/hip/hip_runtime.h:53:
/opt/rocm/include/hip/hip_common.h:30:9: warning: '__HIP_PLATFORM_HCC__' macro redefined [-Wmacro-redefined]
#define __HIP_PLATFORM_HCC__
        ^
<command line>:20:9: note: previous definition is here
#define __HIP_PLATFORM_HCC__ 1
        ^
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:110:17: error: no matching constructor for initialization of 'output_type [4]'
    output_type values[ItemsPerThread];
                ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel_impl<256, 4, false, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    block_reduce_kernel_impl<BlockSize, ItemsPerThread, WithInitialValue, ResultType>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:147:37: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel<256, 4, false, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
            HIP_KERNEL_NAME(detail::block_reduce_kernel<
                                    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:302:20: note: in instantiation of function template specialization 'rocprim::detail::reduce_impl<256, 4, true, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::reduce_impl<block_size, items_per_thread, true>(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_reduce.hpp:91:27: note: in instantiation of function template specialization 'rocprim::reduce<std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        return ::rocprim::reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:556:43: note: in instantiation of function template specialization 'hipcub::DeviceReduce::Reduce<std::complex<float> *, std::complex<float> *, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceReduce::Reduce(
                                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:858:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchScalarReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchScalarReduction(ctx, out, in, in_size, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:896:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:111:17: error: no matching constructor for initialization of 'output_type' (aka 'std::complex<float>')
    output_type output_value;
                ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:110:17: error: no matching constructor for initialization of 'output_type [4]'
    output_type values[ItemsPerThread];
                ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel_impl<256, 4, true, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    block_reduce_kernel_impl<BlockSize, ItemsPerThread, WithInitialValue, ResultType>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:181:37: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel<256, 4, true, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
            HIP_KERNEL_NAME(detail::block_reduce_kernel<
                                    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:302:20: note: in instantiation of function template specialization 'rocprim::detail::reduce_impl<256, 4, true, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::reduce_impl<block_size, items_per_thread, true>(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_reduce.hpp:91:27: note: in instantiation of function template specialization 'rocprim::reduce<std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        return ::rocprim::reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:556:43: note: in instantiation of function template specialization 'hipcub::DeviceReduce::Reduce<std::complex<float> *, std::complex<float> *, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceReduce::Reduce(
                                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:858:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchScalarReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchScalarReduction(ctx, out, in, in_size, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:896:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:111:17: error: no matching constructor for initialization of 'output_type' (aka 'std::complex<float>')
    output_type output_value;
                ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:596:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:861:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchRowReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchRowReduction(ctx, out, in, in_dim0, in_dim1, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:896:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:754:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:870:5: note: in instantiation of function template specialization 'tensorflow::functor::Launch3DXZReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    Launch3DXZReduction(ctx, out, in, in_dim0, in_dim1, in_dim2, op, init,
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:896:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:110:17: error: no matching constructor for initialization of 'output_type [4]'
    output_type values[ItemsPerThread];
                ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel_impl<256, 4, true, std::complex<float>, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    block_reduce_kernel_impl<BlockSize, ItemsPerThread, WithInitialValue, ResultType>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:181:37: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel<256, 4, true, std::complex<float>, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
            HIP_KERNEL_NAME(detail::block_reduce_kernel<
                                    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:302:20: note: in instantiation of function template specialization 'rocprim::detail::reduce_impl<256, 4, true, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::reduce_impl<block_size, items_per_thread, true>(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_reduce.hpp:91:27: note: in instantiation of function template specialization 'rocprim::reduce<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        return ::rocprim::reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:556:43: note: in instantiation of function template specialization 'hipcub::DeviceReduce::Reduce<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceReduce::Reduce(
                                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:858:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchScalarReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *>' requested here
    LaunchScalarReduction(ctx, out, in, in_size, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:931:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, TransformOutputIterator<T, T, DividesBy<T>>, T*,
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:111:17: error: no matching constructor for initialization of 'output_type' (aka 'std::complex<float>')
    output_type output_value;
                ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:596:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<std::complex<float> *, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:861:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchRowReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *>' requested here
    LaunchRowReduction(ctx, out, in, in_dim0, in_dim1, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:931:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, TransformOutputIterator<T, T, DividesBy<T>>, T*,
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Sum<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:754:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Sum<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:870:5: note: in instantiation of function template specialization 'tensorflow::functor::Launch3DXZReduction<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *>' requested here
    Launch3DXZReduction(ctx, out, in, in_dim0, in_dim1, in_dim2, op, init,
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:931:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Sum<std::complex<float> >, tensorflow::TransformOutputIterator<std::complex<float>, std::complex<float>, tensorflow::functor::DividesBy<std::complex<float>, std::complex<float> >, long>, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Sum<T>, TransformOutputIterator<T, T, DividesBy<T>>, T*,
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:110:17: error: no matching constructor for initialization of 'output_type [4]'
    output_type values[ItemsPerThread];
                ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel_impl<256, 4, false, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    block_reduce_kernel_impl<BlockSize, ItemsPerThread, WithInitialValue, ResultType>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:147:37: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel<256, 4, false, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
            HIP_KERNEL_NAME(detail::block_reduce_kernel<
                                    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:302:20: note: in instantiation of function template specialization 'rocprim::detail::reduce_impl<256, 4, true, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    return detail::reduce_impl<block_size, items_per_thread, true>(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_reduce.hpp:91:27: note: in instantiation of function template specialization 'rocprim::reduce<std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
        return ::rocprim::reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:556:43: note: in instantiation of function template specialization 'hipcub::DeviceReduce::Reduce<std::complex<float> *, std::complex<float> *, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceReduce::Reduce(
                                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:858:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchScalarReduction<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchScalarReduction(ctx, out, in, in_size, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:1036:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Prod<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:111:17: error: no matching constructor for initialization of 'output_type' (aka 'std::complex<float>')
    output_type output_value;
                ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:110:17: error: no matching constructor for initialization of 'output_type [4]'
    output_type values[ItemsPerThread];
                ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel_impl<256, 4, true, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    block_reduce_kernel_impl<BlockSize, ItemsPerThread, WithInitialValue, ResultType>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:181:37: note: in instantiation of function template specialization 'rocprim::detail::block_reduce_kernel<256, 4, true, std::complex<float>, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
            HIP_KERNEL_NAME(detail::block_reduce_kernel<
                                    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:302:20: note: in instantiation of function template specialization 'rocprim::detail::reduce_impl<256, 4, true, std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    return detail::reduce_impl<block_size, items_per_thread, true>(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_reduce.hpp:91:27: note: in instantiation of function template specialization 'rocprim::reduce<std::complex<float> *, std::complex<float> *, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
        return ::rocprim::reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:556:43: note: in instantiation of function template specialization 'hipcub::DeviceReduce::Reduce<std::complex<float> *, std::complex<float> *, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceReduce::Reduce(
                                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:858:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchScalarReduction<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchScalarReduction(ctx, out, in, in_size, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:1036:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Prod<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:75:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_reduce.hpp:111:17: error: no matching constructor for initialization of 'output_type' (aka 'std::complex<float>')
    output_type output_value;
                ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:596:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<std::complex<float> *, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:861:5: note: in instantiation of function template specialization 'tensorflow::functor::LaunchRowReduction<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    LaunchRowReduction(ctx, out, in, in_dim0, in_dim1, op, init, cu_stream);
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:1036:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Prod<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
In file included from tensorflow/core/kernels/reduction_ops_gpu_complex64.cu.cc:20:
In file included from ./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp:29:
In file included from ./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/../../config.hpp:38:
In file included from external/rocprim_archive/rocprim/include/rocprim/rocprim.hpp:80:
In file included from external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:31:
external/rocprim_archive/rocprim/include/rocprim/device/detail/device_segmented_reduce.hpp:81:16: error: no matching constructor for initialization of 'std::complex<float>'
    ResultType result;
               ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:58:5: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    segmented_reduce<BlockSize, ItemsPerThread>(
    ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:121:25: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_kernel<256, 8, tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
        HIP_KERNEL_NAME(segmented_reduce_kernel<block_size, items_per_thread>),
                        ^
external/rocprim_archive/rocprim/include/rocprim/device/device_segmented_reduce_hip.hpp:248:20: note: in instantiation of function template specialization 'rocprim::detail::segmented_reduce_impl<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, std::complex<float>, tensorflow::functor::Prod<std::complex<float> > >' requested here
    return detail::segmented_reduce_impl(
                   ^
./external/rocprim_archive/hipcub/include/hipcub/rocprim/device/device_segmented_reduce.hpp:55:27: note: in instantiation of function template specialization 'rocprim::segmented_reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
        return ::rocprim::segmented_reduce(
                          ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:754:52: note: in instantiation of function template specialization 'hipcub::DeviceSegmentedReduce::Reduce<tensorflow::PermutationInputIterator<std::complex<float>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::GatherOp, int>, long>, std::complex<float> *, rocprim::transform_iterator<rocprim::counting_iterator<int, long>, tensorflow::functor::RowOffset, int>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> >' requested here
    auto success = gpuprim::DeviceSegmentedReduce::Reduce(
                                                   ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:870:5: note: in instantiation of function template specialization 'tensorflow::functor::Launch3DXZReduction<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *>' requested here
    Launch3DXZReduction(ctx, out, in, in_dim0, in_dim1, in_dim2, op, init,
    ^
./tensorflow/core/kernels/reduction_gpu_kernels.cu.h:1036:5: note: in instantiation of function template specialization 'tensorflow::functor::ReduceImpl<std::complex<float>, tensorflow::functor::Prod<std::complex<float> >, std::complex<float> *, std::complex<float> *, Eigen::array<long, 1> >' requested here
    ReduceImpl<T, Prod<T>, T*, T*, ReductionAxes>(
    ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
    struct complex<float>
           ^
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 0 were provided
/usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/complex:1053:12: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 0 were provided
1 warning and 16 errors generated.
Died at /opt/rocm/bin/hipcc line 496.

[jszuppe]

I don't know if there's anything we can/should do about it in rocPRIM since std::complex<> is just not supported on device (that's independent from rocPRIM). no matching constructor error is there because std::complex<> doesn't have default constructor with __device__ attribute. Why it works on nvcc is not entirely clear to me. For example, see https://llvm.org/docs/CompileCudaWithLLVM.html#std-complex.