How is the value of vector width used to calculate specific operations?

[lovemyyuyu]

For the computing units mentioned in your paper, they mainly include pulsating arrays and vector units. Pulsating arrays are used to handle matrix multiplication operations, but how are vector units related to vector width when performing specific operations such as element add, softmax, etc. For example, when vector_width=8, how do I obtain cycles when performing elementadded or softmax operations on a matrix with a shape of (256, 512)