[BUG] Function "all" and "any" can not work normally

[DerekLin919]

Describe the Bug

When the dimension of the input tensor is 3 or more, the functions all and any can't work normally.

To Reproduce

Expected Behavior

Code Snippets

System Details (please complete the following information):

• OS: [e.g. Ubuntu 18.04]
• CUDA version: [e.g CUDA 11.4]
• g++ version: [e.g. 9.3.0]

Additional Context

[cliffburdick]

Hi @DerekLin919 , can you please give an example of this not working? Our unit tests have a 4D tensor and it passes, and I just tried on a 3D tensor and it seems to be correct:

    auto a = make_tensor<float>({10, 10, 10});
    auto b = make_tensor<bool>({});
    (a = random<float>({10, 10, 10}, NORMAL)).run(exec);
    (b = all(a)).run(exec);
    print(b);
    a(1,1,1) = 0.0f;
    (b = all(a)).run(exec);
    print(b);

Output:

tensor_0_b: Tensor{bool} Rank: 0, Sizes:[], Strides:[]
 1
tensor_0_b: Tensor{bool} Rank: 0, Sizes:[], Strides:[]
 0

[DerekLin919]

Thank you for your prompt reply! I will try the the unit test you provide on several days later. Thank you again！

[DerekLin919]

I have tried the example you provided and it does work correctly. However, I reproduced the example provided in the documentation for converting a 4-dimensional tensor to a 2-dimensional tensor, and it does not run correctly. The program is as follows.
int main() {
auto a = make_tensor({10, 10, 10, 10});
auto b = make_tensor({10, 10});
(a = random({10, 10, 10}, NORMAL)).run();
(b = all(a, {0, 1})).run();
print(b);
a(1,1,1,1) = 0.0f;
(b = all(a, {0, 1})).run();
print(b);
}

The following compilation error is displayed.
/tmp/tmp.w1IQLiweob/MatX/include/matx/transforms/reduce.h(975): error: class "matx::detail::reduceOpAll<__nv_bool>" has no member "Reduce"
detected during:
instantiation of "T matx::detail::warpReduceOp(T, Op, uint32_t) [with T=__nv_bool, Op=matx::detail::reduceOpAll<__nv_bool>]"
(1162): here
instantiation of "void matx::detail::matxReduceKernel(OutType, InType, ReduceOp, matx::index_t) [with OutType=matx::detail::tensor_impl_t<__nv_bool, 2, matx::tensor_desc_t<cuda::std::__4::array<matx::index_t, 2UL>, cuda::std::__4::array<matx::index_t, 2UL>, 2>, matx::detail::DenseTensorData<__nv_bool>>, InType=matx::detail::tensor_impl_t<float, 4, matx::tensor_desc_t<cuda::std::__4::array<long long, 4UL>, cuda::std::__4::array<long long, 4UL>, 4>, matx::detail::DenseTensorData>, ReduceOp=matx::detail::reduceOpAll<__nv_bool>]"
(1391): here
instantiation of "void matx::reduce(OutType, TensorIndexType, const InType &, ReduceOp, cudaStream_t, __nv_bool) [with OutType=matx::detail::tensor_impl_t<__nv_bool, 2, matx::tensor_desc_t<cuda::std::__4::array<matx::index_t, 2UL>, cuda::std::__4::array<matx::index_t, 2UL>, 2>, matx::detail::DenseTensorData<__nv_bool>>, TensorIndexType=std::nullopt_t, InType=matx::detail::tensor_impl_t<float, 4, matx::tensor_desc_t<cuda::std::__4::array<long long, 4UL>, cuda::std::__4::array<long long, 4UL>, 4>, matx::detail::DenseTensorData>, ReduceOp=matx::detail::reduceOpAll<__nv_bool>, =true]"
(1452): here
instantiation of "void matx::reduce(OutType, const InType &, ReduceOp, cudaStream_t, __nv_bool) [with OutType=matx::detail::tensor_impl_t<__nv_bool, 2, matx::tensor_desc_t<cuda::std::__4::array<matx::index_t, 2UL>, cuda::std::__4::array<matx::index_t, 2UL>, 2>, matx::detail::DenseTensorData<__nv_bool>>, InType=matx::detail::tensor_impl_t<float, 4, matx::tensor_desc_t<cuda::std::__4::array<long long, 4UL>, cuda::std::__4::array<long long, 4UL>, 4>, matx::detail::DenseTensorData>, ReduceOp=matx::detail::reduceOpAll<__nv_bool>]"
(2460): here
instantiation of "void matx::all_impl(OutType, const InType &, matx::cudaExecutor) [with OutType=matx::detail::tensor_impl_t<__nv_bool, 2, matx::tensor_desc_t<cuda::std::__4::array<matx::index_t, 2UL>, cuda::std::__4::array<matx::index_t, 2UL>, 2>, matx::detail::DenseTensorData<__nv_bool>>, InType=matx::detail::tensor_impl_t<float, 4, matx::tensor_desc_t<cuda::std::__4::array<long long, 4UL>, cuda::std::__4::array<long long, 4UL>, 4>, matx::detail::DenseTensorData>]"
/tmp/tmp.w1IQLiweob/MatX/include/matx/operators/all.h(77): here
instantiation of "void matx::detail::AllOp<OpA, ORank>::Exec(Out &&, Executor &&) const [with OpA=matx::tensor_t<float, 4, matx::basic_storage<matx::raw_pointer_buffer<float, matx::matx_allocator>>, matx::tensor_desc_t<cuda::std::__4::array<long long, 4UL>, cuda::std::__4::array<long long, 4UL>, 4>>, ORank=2, Out=cuda::std::__4::tuple<matx::detail::tensor_impl_t<__nv_bool, 2, matx::tensor_desc_t<cuda::std::__4::array<matx::index_t, 2UL>, cuda::std::__4::array<matx::index_t, 2UL>, 2>, matx::detail::DenseTensorData<__nv_bool>>>, Executor=matx::cudaExecutor &]"
/tmp/tmp.w1IQLiweob/MatX/include/matx/operators/set.h(179): here
instantiation of "void matx::detail::set<T, Op>::TransformExec(ShapeType &&, Executor &&) const noexcept [with T=matx::tensor_t<__nv_bool, 2, matx::basic_storage<matx::raw_pointer_buffer<__nv_bool, matx::matx_allocator<__nv_bool>>>, matx::tensor_desc_t<cuda::std::__4::array<matx::index_t, 2UL>, cuda::std::__4::array<matx::index_t, 2UL>, 2>>, Op=matx::detail::AllOp<matx::tensor_t<float, 4, matx::basic_storage<matx::raw_pointer_buffer<float, matx::matx_allocator>>, matx::tensor_desc_t<cuda::std::__4::array<long long, 4UL>, cuda::std::__4::array<long long, 4UL>, 4>>, 2>, ShapeType=cuda::std::__4::array<matx::index_t, 2UL>, Executor=matx::cudaExecutor &]"
/tmp/tmp.w1IQLiweob/MatX/include/matx/operators/base_operator.h(77): here
instantiation of "void matx::BaseOp::run(Ex &&) [with T=matx::detail::set<matx::tensor_t<__nv_bool, 2, matx::basic_storage<matx::raw_pointer_buffer<__nv_bool, matx::matx_allocator<__nv_bool>>>, matx::tensor_desc_t<cuda::std::__4::array<matx::index_t, 2UL>, cuda::std::__4::array<matx::index_t, 2UL>, 2>>, matx::detail::AllOp<matx::tensor_t<float, 4, matx::basic_storage<matx::raw_pointer_buffer<float, matx::matx_allocator>>, matx::tensor_desc_t<cuda::std::__4::array<long long, 4UL>, cuda::std::__4::array<long long, 4UL>, 4>>, 2>>, Ex=matx::cudaExecutor]"
/tmp/tmp.w1IQLiweob/MatX/include/matx/operators/base_operator.h(112): here
instantiation of "void matx::BaseOp::run(cudaStream_t) [with T=matx::detail::set<matx::tensor_t<__nv_bool, 2, matx::basic_storage<matx::raw_pointer_buffer<__nv_bool, matx::matx_allocator<__nv_bool>>>, matx::tensor_desc_t<cuda::std::__4::array<matx::index_t, 2UL>, cuda::std::__4::array<matx::index_t, 2UL>, 2>>, matx::detail::AllOp<matx::tensor_t<float, 4, matx::basic_storage<matx::raw_pointer_buffer<float, matx::matx_allocator>>, matx::tensor_desc_t<cuda::std::__4::array<long long, 4UL>, cuda::std::__4::array<long long, 4UL>, 4>>, 2>>]"
/tmp/tmp.w1IQLiweob/test/test_matx.cu(10): here
And there are a lot of similar errors in the back.

[cliffburdick]

Thanks, I was able to reproduce it now. We will look into it.