IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /docs/manual/basics.md). For the complete Mojo documentation index, see llms.txt.
Skip to main content
Version: Nightly
For the complete Mojo documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /docs/manual/basics.md).

stack_allocation

stack_allocation[LayoutType: TensorLayout, //, dtype: DType, address_space: AddressSpace = AddressSpace.GENERIC, alignment: Int = align_of[dtype]()](var layout: LayoutType) -> TileTensor[dtype, LayoutType, MutExternalOrigin, address_space=address_space] where LayoutType.all_dims_known

Allocate a TileTensor on the stack with the given layout.

Creates a stack-allocated buffer sized for the layout and returns a TileTensor pointing to it. The layout must have all dimensions known at compile time.

Constraints:

All layout dimensions must be statically known.

Parameters:

  • ​LayoutType (TensorLayout): The layout type (inferred from layout argument).
  • ​dtype (DType): The data type of tensor elements.
  • ​address_space (AddressSpace): Memory address space (default: GENERIC).
  • ​alignment (Int): Allocation alignment in bytes (default: natural type alignment from align_of[dtype]()). Pass an explicit value when downstream loads/stores require larger alignment (for example, AMD ds_read_b128 requires 16 B, and sub-block swizzles can require alignment up to the sub-block size). Forwarded to the underlying std.memory.stack_allocation.

Args:

  • ​layout (LayoutType): The layout instance defining shape and strides.

Returns:

TileTensor[dtype, LayoutType, MutExternalOrigin, address_space=address_space]: A mutable TileTensor backed by stack-allocated memory.

View source