摘要 |
A flow control method for active pipelines segments macro processes into tasks. Each time a coprocessing device becomes available, a scheduling algorithm is run to determine which task is dispatched to the available coprocessing device for execution. The scheduling algorithm determines a slack time to meet deadline for each of the tasks and ranks the tasks accordingly, with the tasks having the shorter slack times having the higher ranking. The highest ranked task that has a buffer available to read data from and a buffer available to write data into is dispatched for execution by the available coprocessing device. <IMAGE> |