[BUG] Random op can't be casted to tensor_t

[HugoPhibbs]

Describe the Bug
Build and IDE complains when I try put in a RandomOp into a function that expects a tensor_t type

To Reproduce

template <typename T>
        inline auto  performProjectionsMatX(matx::tensor_t<T, 2> X, int D) {
            int d = X.Shape()[1];
            auto Y = matx::random<T>({d, D}, matx::UNIFORM);
            auto res = matx::make_tensor<T>({X.shape(0), D});
            (res = matx::matmul(X, Y)).run();
            return res;
        }

auto X = matx::random<float>({10, 3}, matx::UNIFORM);

auto projections = GsDBSCAN::projections::performProjectionsMatX(X, 3);

Tells me:

error: no instance of function template "GsDBSCAN::projections::performProjectionsMatX" matches the argument list
            argument types are: (matx::detail::RandomOp<float, cuda::std::__4::array<matx::index_t, 2UL>>, int)
      auto projections = GsDBSCAN::projections::performProjectionsMatX(X, 3);
                         ^

I'm honestly also a little lost as to how to copy an operator to a tensor such that this issue may be circumvented.

Expected Behavior
No error being thrown

System Details:

• OS: Ubuntu 20.04
• CUDA version: 12.6
• g++ version: 9.3.4

[cliffburdick]

Hi @HugoPhibbs, a random operator, or any operator in general, cannot be cast to a tensor_t. The reason is that a random operator is a generator, and the values in that generator don't actually exist until they're fetched. A tensor_t needs memory backing it, and the random operator doesn't necessarily have anything backing it. To accomplish this you need a temporary tensor to hold the random numbers if you're trying to get a tensor with random data, or you can modify your function like the following to take an operator instead of a tensor:

template <typename Op>
        inline auto  performProjectionsMatX(Op X, int D) {
            int d = Shape(X)[1];
            auto Y = matx::random<typename X::value_type>({d, D}, matx::UNIFORM);
            auto res = matx::make_tensor<typename X::value_type>({Shape(X)[0], D});
            (res = matx::matmul(X, Y)).run();
            return res;
        }

[HugoPhibbs]

ok thx for your help