Trait TensorReduction 

pub trait TensorReduction {
    // Required methods
    fn reduce_sum_read(
        &mut self,
        input: TensorRead<'_>,
        axes: &[usize],
    ) -> Result<Tensor>;
    fn reduce_prod_read(
        &mut self,
        input: TensorRead<'_>,
        axes: &[usize],
    ) -> Result<Tensor>;
    fn reduce_max_read(
        &mut self,
        input: TensorRead<'_>,
        axes: &[usize],
    ) -> Result<Tensor>;
    fn reduce_min_read(
        &mut self,
        input: TensorRead<'_>,
        axes: &[usize],
    ) -> Result<Tensor>;
}

Reduction operations.

Reducing over an axis whose extent is zero returns an error for every reduction operation. Passing an empty axes slice is a no-op for the public reductions and returns the input values unchanged. Internal mapped reductions document their own empty-axis semantics.

Examples

use tenferro_tensor::TensorReduction;

fn accepts_reduction<B: TensorReduction>(_backend: &mut B) {}

Required Methods

fn reduce_sum_read( &mut self, input: TensorRead<'_>, axes: &[usize], ) -> Result<Tensor>

Sum elements across axes from an owned tensor or borrowed view.

Examples

use tenferro_tensor::{Tensor, TensorRead, TensorReduction};

fn sum_owned<B: TensorReduction>(
    backend: &mut B,
    input: &Tensor,
) -> tenferro_tensor::Result<Tensor> {
    backend.reduce_sum_read(TensorRead::from_tensor(input), &[0])
}

Errors

Returns crate::Error::Validation with a typed ValidationError source for invalid shapes, ranks, axes, dtypes, or output metadata. It returns crate::Error::BackendFailure or crate::Error::BackendSource when backend execution or storage access cannot provide the requested result.

fn reduce_prod_read( &mut self, input: TensorRead<'_>, axes: &[usize], ) -> Result<Tensor>

Multiply elements across axes from an owned tensor or borrowed view.

Overflow policy

Backends may reassociate a floating-point or complex product across lanes, chunks and threads, and the CPU and CUDA kernels do so differently. Rounding and range behaviour therefore depend on the backend, the layout and the thread count: a partial product can overflow to inf or underflow to 0, and combining the partials can then give inf, 0 or NaN even where the sequential product is representable. The #2150 alternating 2.0, 0.5 axis is one example: an f32 sequential product is exactly 1.0, while a reassociated one may return NaN. This is the same trade-off torch.prod makes. Integer products are exact and wrap on overflow.

Examples

use tenferro_tensor::{Tensor, TensorRead, TensorReduction};

fn prod_owned<B: TensorReduction>(
    backend: &mut B,
    input: &Tensor,
) -> tenferro_tensor::Result<Tensor> {
    backend.reduce_prod_read(TensorRead::from_tensor(input), &[0])
}

Errors

Returns crate::Error::Validation with a typed ValidationError source for invalid shapes, ranks, axes, dtypes, or output metadata. It returns crate::Error::BackendFailure or crate::Error::BackendSource when backend execution or storage access cannot provide the requested result.

fn reduce_max_read( &mut self, input: TensorRead<'_>, axes: &[usize], ) -> Result<Tensor>

Take maximum values across axes from an owned tensor or borrowed view.

Examples

use tenferro_tensor::{Tensor, TensorRead, TensorReduction};

fn max_owned<B: TensorReduction>(
    backend: &mut B,
    input: &Tensor,
) -> tenferro_tensor::Result<Tensor> {
    backend.reduce_max_read(TensorRead::from_tensor(input), &[0])
}

Errors

Returns crate::Error::Validation with a typed ValidationError source for invalid shapes, ranks, axes, dtypes, or output metadata. It returns crate::Error::BackendFailure or crate::Error::BackendSource when backend execution or storage access cannot provide the requested result.

fn reduce_min_read( &mut self, input: TensorRead<'_>, axes: &[usize], ) -> Result<Tensor>

Take minimum values across axes from an owned tensor or borrowed view.

Examples

use tenferro_tensor::{Tensor, TensorRead, TensorReduction};

fn min_owned<B: TensorReduction>(
    backend: &mut B,
    input: &Tensor,
) -> tenferro_tensor::Result<Tensor> {
    backend.reduce_min_read(TensorRead::from_tensor(input), &[0])
}

Errors

Returns crate::Error::Validation with a typed ValidationError source for invalid shapes, ranks, axes, dtypes, or output metadata. It returns crate::Error::BackendFailure or crate::Error::BackendSource when backend execution or storage access cannot provide the requested result.

Dyn Compatibility

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors