Arithmetic processing apparatus and control method of arithmetic processing apparatus

摘要

In a multicore system in which a plurality of CPUs each including a cache memory share one main memory, a write buffer having a plurality of stages of buffers each holding data to be written to the main memory and an address of a write destination is provided between the cache memory and the main memory, and at the time of a write to the write buffer from the cache memory, an address of a write destination and the addresses stored in the buffers are compared, and when any of the buffers has an agreeing address, data is overwritten to this buffer, and the buffer is logically moved to a last stage.

主权项

1. An arithmetic processing apparatus comprising
a plurality of arithmetic processing parts which access an external memory via a bus, the arithmetic processing parts each comprising: a cache memory; a processing part which executes arithmetic processing by using data held in the cache memory; a write buffer which is provided between an own cache memory and the external memory and has a plurality of stages of buffers each holding a set of data to be written to the external memory and an address of a write destination; a first control part which controls a write from the own cache memory to own write buffer in write processing from the own cache memory to the external memory via the own write buffer; and a second control part which controls a write from the own write buffer to the external memory in the write processing from the own cache memory to the external memory via the own write buffer, wherein the first control part which compares an address of a write destination and the addresses stored in the plurality of stages of buffers, and when any of the plurality of stages of buffers has an address agreeing with the address of the write destination, overwrites data to a buffer having the agreeing address out of the plurality of stages of buffers and logically moves the buffer overwriting data to a last stage out of the plurality of stages of buffers holding the data to be written to the external memory and the address of the write destination, and wherein; when a certain arithmetic processing part receives a read request of data held in the plurality of stages of buffers included in the own write buffer from another arithmetic processing part, the first control part of the certain arithmetic processing part supplies the data in a buffer holding requested data out of the plurality of stages of buffers and holds, in the buffer, information indicating that a read from another arithmetic processing part is executed, without accessing the own cache memory; and after the certain arithmetic processing part writes the data in the buffer holding the requested data to the external memory, the first control part of the certain arithmetic processing part accesses the own cache memory and updates information stored in the own cache memory indicating that the data is shared with another arithmetic processing part.