Mojo struct
RuntimeLayout
@register_passable(trivial)
struct RuntimeLayout[layout: Layout, /, *, bitwidth: Int = bitwidthof[AnyTrivialRegType,__mlir_type.!kgen.target]()]
A runtime-configurable layout that uses RuntimeTuple for storage.
This struct provides a layout implementation that can be modified at runtime,
unlike the static Layout type. It
uses RuntimeTuple for
shape and stride storage.
The layout must have statically known dimensions at compile time, but the actual shape and stride values can be modified during execution.
Parameters
- layout (
Layout): The staticLayouttype to base this runtime layout on. - bitwidth (
Int): The bit width to use for shape storage, defaults to systemInt.
Fields
- shape (
RuntimeTuple[layout.shape, element_bitwidth=bitwidth, unsigned=True]): The shape of the layout as a runtime tuple. - stride (
RuntimeTuple[layout.stride, unsigned=True]): The stride of the layout as a runtime tuple.
Implemented traits
AnyType,
Stringable,
UnknownDestructibility,
Writable
Methods
__init__
__init__() -> Self
Initialize a RuntimeLayout with default values.
Creates a new RuntimeLayout instance with default shape and stride
values. Requires that the static layout has known dimensions at compile
time.
Constraints:
The static layout that this runtime layout is based on must have all dimensions known.
__init__(shape: RuntimeTuple[layout.shape, element_bitwidth=bitwidth, unsigned=True], stride: RuntimeTuple[layout.stride, unsigned=True]) -> Self
Initialize a RuntimeLayout with specified shape and stride.
Args:
- shape (
RuntimeTuple[layout.shape, element_bitwidth=bitwidth, unsigned=True]): ARuntimeTuplecontaining the dimensions of each axis. - stride (
RuntimeTuple[layout.stride, unsigned=True]): ARuntimeTuplecontaining the stride values for each axis.
__call__
__call__(self, idx: Int) -> Int
Convert a single index to a flat linear index.
Args:
- idx (
Int): The one-dimensional index to convert.
Returns:
The corresponding flat linear index in the layout.
__call__[: ImmutableOrigin, //, t: IntTuple[$0]](self, idx: RuntimeTuple[t, element_bitwidth=element_bitwidth, unsigned=unsigned]) -> Int
Convert a multi-dimensional index to a flat linear index.
Parameters:
- t (
IntTuple[$0]): TheIntTupletype for the index.
Args:
- idx (
RuntimeTuple[t, element_bitwidth=element_bitwidth, unsigned=unsigned]): ARuntimeTuplecontaining the multi-dimensional coordinates.
Returns:
The corresponding flat linear index in the layout.
size
size(self) -> Int
Calculate the total number of elements in the layout.
Returns:
The product of all dimensions in the shape, representing the total number of elements that can be addressed by this layout.
bound_check_required
bound_check_required(self) -> Bool
Determine if bounds checking is required for this layout.
Returns:
True if any dimension in the shape differs from the static layout's shape, False otherwise.
cast
cast[type: DType](self) -> RuntimeLayout[layout, bitwidth=bitwidthof[::DType,__mlir_type.!kgen.target]()]
Cast the layout to use a different element bitwidth.
Parameters:
- type (
DType): The target data type that determines the new bitwidth.
Returns:
A new RuntimeLayout with the shape cast to the specified type's bitwidth.
__str__
__str__(self) -> String
Convert the layout to a string representation.
Returns:
A string representation of the layout.
row_major
static row_major[rank: Int, //](shape: Index[rank, element_bitwidth=element_bitwidth, unsigned=unsigned]) -> RuntimeLayout[layout, bitwidth=element_bitwidth]
Create a row-major layout from the given shape.
In row-major layout, elements with adjacent rightmost indices are adjacent in memory.
Parameters:
- rank (
Int): The number of dimensions in the layout.
Args:
- shape (
Index[rank, element_bitwidth=element_bitwidth, unsigned=unsigned]): AnIndexListcontaining the dimensions of each axis.
Returns:
A RuntimeLayout with row-major stride ordering.
col_major
static col_major[rank: Int, //](shape: Index[rank, element_bitwidth=element_bitwidth, unsigned=unsigned]) -> RuntimeLayout[layout, bitwidth=element_bitwidth]
Create a column-major layout from the given shape.
In column-major layout, elements with adjacent leftmost indices are adjacent in memory.
Parameters:
- rank (
Int): The number of dimensions in the layout.
Args:
- shape (
Index[rank, element_bitwidth=element_bitwidth, unsigned=unsigned]): AnIndexListcontaining the dimensions of each axis.
Returns:
A RuntimeLayout with column-major stride ordering.
write_to
write_to[W: Writer](self, mut writer: W)
Write a string representation of the layout to a writer.
Parameters:
- W (
Writer): TheWritertype.
Args:
- writer (
W): TheWriterobject to write the layout representation to.
sublayout
sublayout[i: Int](self) -> RuntimeLayout[layout.__getitem__(i), bitwidth=bitwidth]
Extract a nested sublayout at the specified index.
Parameters:
- i (
Int): The index of the nested layout to extract.
Returns:
A RuntimeLayout representing the nested layout at index i.
dim
dim(self, i: Int) -> Int
Get the size of the dimension at the specified index.
Args:
- i (
Int): The index of the dimension to retrieve.
Returns:
The size of the dimension at index i.
__len__
static __len__() -> Int
Get the number of dimensions in the layout.
Returns:
The number of dimensions (rank) of the layout.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!