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

outer_product_acc

def outer_product_acc(res: LayoutTensor[address_space=res.address_space, element_layout=res.element_layout, layout_int_type=res.layout_int_type, linear_idx_type=res.linear_idx_type, masked=res.masked, alignment=res.alignment], lhs: LayoutTensor[address_space=lhs.address_space, element_layout=lhs.element_layout, layout_int_type=lhs.layout_int_type, linear_idx_type=lhs.linear_idx_type, masked=lhs.masked, alignment=lhs.alignment], rhs: LayoutTensor[address_space=rhs.address_space, element_layout=rhs.element_layout, layout_int_type=rhs.layout_int_type, linear_idx_type=rhs.linear_idx_type, masked=rhs.masked, alignment=rhs.alignment])

Updates result tensor with the outer product of two vectors.

Computes res += outer(lhs, rhs) where lhs and rhs are vectors and res is a matrix.

Constraints:

All tensors must have statically known shapes. res must be rank 2. lhs and rhs must be rank 1. res.shape[0] == lhs.shape[0] and res.shape[1] == rhs.shape[0].

Args:

• ​res (LayoutTensor[address_space=res.address_space, element_layout=res.element_layout, layout_int_type=res.layout_int_type, linear_idx_type=res.linear_idx_type, masked=res.masked, alignment=res.alignment]): The result matrix to accumulate into, shape (M, N).
• ​lhs (LayoutTensor[address_space=lhs.address_space, element_layout=lhs.element_layout, layout_int_type=lhs.layout_int_type, linear_idx_type=lhs.linear_idx_type, masked=lhs.masked, alignment=lhs.alignment]): The left-hand side vector, shape (M,).
• ​rhs (LayoutTensor[address_space=rhs.address_space, element_layout=rhs.element_layout, layout_int_type=rhs.layout_int_type, linear_idx_type=rhs.linear_idx_type, masked=rhs.masked, alignment=rhs.alignment]): The right-hand side vector, shape (N,).