WIZ-level parallelism - page 2

But now we have functions running, not in devices behind registers on a single WIZ bus, but in independent WIZes in a WIZ chip.

For example, to compute a = sine(x), b = cosine(y), and c = sqrt(z), we have the same two choices for coding it:

Sequentially, like this:

  x -> sine   -> a
  y -> cosine -> b
  z -> sqrt   -> c

or in parallel, like this:

  x -> sine
  y -> cosine
  z -> sqrt
  sine   -> a
  cosine -> b
  sqrt   -> c

As with ILP, these functions, each running in a separate WIZ, also run in parallel. The instruction "sine -> a" waits for the sine to complete, but by then the cosine and the sqrt will also have completed and the final two instructions add no wait time.