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

Mojo function

packA_i8mm

def packA_i8mm[a_type: DType](t0: Int, t1: Int, k: Int, a_ptr: Pointer[Scalar[a_type], address_space=a_ptr.address_space, _safe=False], a_packed_ptr: Pointer[Scalar[a_type], address_space=a_packed_ptr.address_space, _safe=False])

Packs a range of rows of matrix A for the i8mm kernel layout.

Parameters:

a_type (DType): The dtype of the A tensor.

Args:

t0 (Int): The starting row index of the range to pack.
t1 (Int): The ending row index of the range to pack (exclusive).
k (Int): The K dimension of the matmul.
a_ptr (Pointer[Scalar[a_type], address_space=a_ptr.address_space, _safe=False]): Pointer to the source A matrix in row-major layout.
a_packed_ptr (Pointer[Scalar[a_type], address_space=a_packed_ptr.address_space, _safe=False]): Pointer to the destination packed A buffer.