| 摘要 |
A system and method for queuing, control and transfer of data between a host processor and a peripheral processor using an architecture and a data flow strategy of one or more virtual FIFO data structures stored in main memory and a hardware FIFO under control of the host and peripheral processors. One virtual FIFO at a time drives the data FIFO with data. In turn, the data FIFO drives a peripheral device with this data. The host software running on the digital processor controls the loading of data for each process (context) into its associated virtual FIFO. The host processor controls the operation of the peripheral processor and the virtual FIFOs. The peripheral processor controls the flow of data from the data FIFO to the peripheral device, and under control of the host software, the flow of data from the driving virtual FIFO to the data FIFO. Start and end address pointers for each virtual FIFO stored in associated memory block indicate the memory location in the virtual FIFO where data is stored. The peripheral processor also keeps a start address pointer of the memory location of the last unused data then read out of the data FIFO to the peripheral device. These address pointers allow the hardware FIFO to be flushed when a context switch occurs, which can take place before all of the data in the driving virtual FIFO is supplied to the data FIFO, and supplied by the data FIFO to the peripheral device. A strategy for updating the address pointers improves effective memory bandwidth. The present invention allows for main memory latency, minimizes main memory bandwidth utilization, allows for relatively fast context switching, and reduces hardware cost because the storage capacity of the hardware FIFO is small relative to the storage capacity of each virtual FIFO. Total storage capacity of each virtual FIFO, data FIFO combination is large, and the hardware data FIFO can be fast, which reduces latency. |
