| 摘要 |
A technique for reducing the latency associated with a memory read request. A bypass path is provided to direct the address of a corresponding request to a memory controller. The memory controller initiates a speculative read request to the corresponding address location. In the meantime, the original request is decoded and directed to the targeted area of the system. If the request is a read request, the memory controller will receive the request, and after comparing the request address to the address received via the bypass path, the memory controller will cancel the request since the speculative read has already been issued. If the request is directed elsewhere or is not a read request, the speculative read request is cancelled.
|
