| 摘要 |
A dataflow graph is split into sub-graphs referred to as configurations, each configuration comprising computational hardware containing elements that operate on operand sets. A configuration executes by consuming completed operand sets from a designated input tag space (e.g., in a content addressable memory) until the operand sets are exhausted. At that point, the configuration is replaced by another configuration. The execution of a configuration may be triggered by system events, including by the completion of one or more other configurations. Each configuration has a list of inputs on which it depends to form complete operand sets. As other configurations that provide an input complete, a dependency flag is set in each dependent configuration. As each flag is set, the complete set of flags is checked for that configuration; if all the input flags for any configuration are set, then that configuration is scheduled for execution.
|
