Memory access method, crossbar and computer system

Abstract

The embodiment of the invention provides a memory access method, a crossbar switch and a computer system. The method includes: receiving an access request message, which is sent by a GPU (Graphic Processing Unit), by the crossbar switch, wherein the access request message includes an initial address corresponding to to-be-accessed memory space; and determining a mapping address, which corresponds to the initial address, by the crossbar switch to enable the GPU to access memory space, which starts from the mapping address, if the initial address is greater than or equal to a first address, wherein the first address is a highest address corresponding to memory space which the GPU can actually access before the crossbar switch determines the mapping address, the mapping address is between a second address and a third address, the second address is a highest address corresponding to memory space which the GPU can theoretically access, the third address is a highest address of a memory, the first address is less than the second address, and the second address is less than the third address. Therefore, a memory requirement of the GPU is met.

Description

technical field [0001] The embodiment of the present invention relates to memory access technology, in particular to a memory access method, a crossbar switch and a computer system. Background technique [0002] With the continuous development of computer technology, many computers currently include not only a central processing unit (Central Processing Unit, referred to as CPU), but also a graphics processing unit (Graphic Processing Unit, referred to as GPU). [0003] When the number of binary digits that the GPU can process once per unit time is less than the number of binary digits that the CPU can process once per unit time, the GPU can be determined according to the number of binary digits that the GPU can process once per unit time. Theoretically accessible memory space, since part of the memory space in the memory is used to store page tables, therefore, the GPU can only remove the memory space used to store page tables on the basis of theoretically accessible memory...

AI Technical Summary

Problems solved by technology

For example: based on the microprocessor (Microprocessor Without interlocked Piped Stages, referred to as: MIPS) 64 architecture without internal interlocking pipeline stage, GPU32 can theoretically access 2 32 = 4G memory space, the corresponding address range is 0-0xffffffff, however, only 256MB of memory space in the 4G memory space can not be used to store page tables, therefore, the GPU can only compete with other devices for the 256MB memory space , however, the memory requirement of the GPU is 4G, so 256MB cannot meet the memory requirement of the GPU

[0004] In summary, the memory access methods of the prior art cannot meet the memory requirements of the GPU

Images