Specialization impls conflicting

help
1

I'm trying to basically use const generics to compile time check tensor shapes. I have a pretty simple example where I need to custom specify different impls of Mul for tensors of 2 dims times 2 dims, 1 x 1 dim, 2 x 1 dim, and 1 x 2 dim. So I wrote this:

struct Tensor<const D1: usize, const D2: usize> {}

// 2 x 2
impl <const D1: usize, const D2: usize>Mul<&Tensor<D1, D2>> for Tensor<D1, D2> {
    default type Output = Tensor<D1, D2>;

    default fn mul(self, rhs: &Tensor<D1, D2>) -> Self::Output {
        todo!()
    }
}

// 2 x 1
impl <const D1: usize, const D2: usize>Mul<&Tensor<D1, 1>> for Tensor<D1, D2> {
    type Output = Tensor<D1, 1>;

    fn mul(self, rhs: &Tensor<D1, 1>) -> Self::Output {
        todo!()
    }
}

but I get an error on the 2x1 impl: conflicting implementations of trait std::ops::Mul<&Tensor<{_: usize}, 1_usize>> for type Tensor<{_: usize}, 1_usize> conflicting implementation for Tensor<{_: usize}, 1_usize>

I also tried to mark the 2x2 impl with default:

default impl <const D1: usize, const D2: usize>Mul<&Tensor<D1, D2>> for Tensor<D1, D2> {
    default type Output = Tensor<D1, D2>;

    default fn mul(self, rhs: &Tensor<D1, D2>) -> Self::Output {
        todo!()
    }
}

but this is recursing a bunch so I get this error:
overflow evaluating the requirement Tensor<{_: usize}, 1_usize>: std::ops::Mul<&Tensor<{_: usize}, 1_usize>>

I'm not sure why this is conflicting since one implementation is more specific than the other (D2 is set to 1 in the 2x1 impl)

Any help is much appreciated!

2

This topic was automatically closed 90 days after the last reply. We invite you to open a new topic if you have further questions or comments.