For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Mojo function

max

def max[axis: Int](inp: LayoutTensor[address_space=inp.address_space, element_layout=inp.element_layout, layout_int_type=inp.layout_int_type, linear_idx_type=inp.linear_idx_type, masked=inp.masked, alignment=inp.alignment], outp: LayoutTensor[address_space=outp.address_space, element_layout=outp.element_layout, layout_int_type=outp.layout_int_type, linear_idx_type=outp.linear_idx_type, masked=outp.masked, alignment=outp.alignment])

Computes maximum reduction along specified axis.

Reduces the input tensor by taking maximum elements along the specified axis and stores the result in the output tensor.

Constraints:

All tensors must have statically known shapes. outp.rank must equal inp.rank - 1. Non-reduction dimensions must match between inp and outp. Currently only supports rank-2 inputs.

Parameters:

axis (Int): The axis to take maximum along.

Args:

inp (LayoutTensor[address_space=inp.address_space, element_layout=inp.element_layout, layout_int_type=inp.layout_int_type, linear_idx_type=inp.linear_idx_type, masked=inp.masked, alignment=inp.alignment]): The input tensor to reduce.
outp (LayoutTensor[address_space=outp.address_space, element_layout=outp.element_layout, layout_int_type=outp.layout_int_type, linear_idx_type=outp.linear_idx_type, masked=outp.masked, alignment=outp.alignment]): The output tensor to store maximum results.

def max[axis: Int](inp: LayoutTensor[address_space=inp.address_space, element_layout=inp.element_layout, layout_int_type=inp.layout_int_type, linear_idx_type=inp.linear_idx_type, masked=inp.masked, alignment=inp.alignment]) -> LayoutTensor[inp.dtype, _reduce_res_row_major_shape(axis, inp.layout), MutAnyOrigin, address_space=inp.address_space, element_layout=inp.element_layout, layout_int_type=inp.layout_int_type, linear_idx_type=inp.linear_idx_type]

Computes maximum reduction along specified axis, returning a new tensor.

Reduces the input tensor by taking maximum elements along the specified axis and returns a new tensor with the results.

Constraints:

All tensors must have statically known shapes. Result will have rank equal to inp.rank - 1. Non-reduction dimensions in the result match the input. Currently only supports rank-2 inputs.

Parameters:

axis (Int): The axis to take maximum along.

Args:

inp (LayoutTensor[address_space=inp.address_space, element_layout=inp.element_layout, layout_int_type=inp.layout_int_type, linear_idx_type=inp.linear_idx_type, masked=inp.masked, alignment=inp.alignment]): The input tensor to reduce.

Returns:

LayoutTensor[inp.dtype, _reduce_res_row_major_shape(axis, inp.layout), MutAnyOrigin, address_space=inp.address_space, element_layout=inp.element_layout, layout_int_type=inp.layout_int_type, linear_idx_type=inp.linear_idx_type]: A new tensor containing the maximum values along the specified axis.

def max[dtype: DType, layout: Layout](x: LayoutTensor[dtype, layout, address_space=x.address_space, element_layout=x.element_layout, layout_int_type=x.layout_int_type, linear_idx_type=x.linear_idx_type, masked=x.masked, alignment=x.alignment], y: LayoutTensor[dtype, layout, address_space=y.address_space, element_layout=y.element_layout, layout_int_type=y.layout_int_type, linear_idx_type=y.linear_idx_type, masked=y.masked, alignment=y.alignment]) -> LayoutTensor[dtype, layout, x.origin, address_space=x.address_space, element_layout=x.element_layout, layout_int_type=x.layout_int_type, linear_idx_type=x.linear_idx_type, masked=x.masked, alignment=x.alignment].MutableAnyType

Computes element-wise maximum of two tensors.

Returns a new tensor containing the element-wise maximum between the input tensors.

Constraints:

Input tensors must have statically known shapes and matching layouts.

Parameters:

dtype (DType): The data type of the input tensors.
layout (Layout): The layout of the input tensors.

Args:

Returns:

LayoutTensor[dtype, layout, x.origin, address_space=x.address_space, element_layout=x.element_layout, layout_int_type=x.layout_int_type, linear_idx_type=x.linear_idx_type, masked=x.masked, alignment=x.alignment].MutableAnyType: A new tensor containing the element-wise maximum.