Mojo function

apple_mma_load

apple_mma_load[dtype: DType](ptr: UnsafePointer[Scalar[dtype], ptr.origin, address_space=ptr.address_space], row_stride: Int, col_stride: Int = 1) -> SIMD[dtype, 8]

Loads a 16x16 matrix fragment for the current simdgroup thread.

Parameters:

dtype (DType): Element type of the matrix.

Args:

ptr (UnsafePointer): Pointer to the top-left corner of the 16x16 tile.
row_stride (Int): Distance between consecutive rows in the buffer.
col_stride (Int): Distance between consecutive columns within a row.

Returns:

SIMD: SIMD vector of 8 elements for this thread's fragment.