Synchronization of concurrent computation engines
摘要
說(shuō)明書(shū)
Block 820 of the method identifies a first node in the input data set with a data or resource dependency. The first node on the graph may include a first operation to be performed by the first execution engine. The first operation may include a first memory operation involving a memory location in the integrated circuit device. The first operation may include a computation or a data transfer operation. The first node may have a graph edge indicating the dependency from another node on the graph. For example, a computation or a memory transfer operation may need to be completed by the first execution engine to enable an operation performed by the second execution engine to begin.
At block 830, a second node in the input data set that has a connection indicating a dependency from the first node may be identified. The second node may include a second operation to be performed by the second execution engine. The second operation may use the memory location used by the first memory operation. Thus, the second operation may require the first memory operation performed by the first execution engine to be completed before the second operation begins.
