GPU multi-granularity memory access request access fine-grained cache method and device
CN114661352BActive Publication Date: 2024-11-19南京达道电子科技有限公司
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- CN202210288631.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-23
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-03-23
Smart Images

Figure CN114661352B_ABST
Abstract
The present invention discloses a GPU multi-granularity memory access request access fine-grained cache method and device, including: when there is a memory access request to access the cache, and the memory access request is 128 bytes, the current memory access request is split into four consecutive 32-byte sub-memory access requests for parallel processing; it is determined whether all four sub-memory access requests are hit, if yes, the data storage is accessed and the hit data is returned, and the process ends; if not, the 128-byte memory access request has data missing, only the data of the sub-memory access request that has hit is returned, and the four sub-memory access requests are merged into a new memory access request and sent to the next level memory; when the new memory access request carries the missing data and returns from the next level memory, the new memory access request is split into four 32-byte sub-memory access requests again, and then the missing data required by the sub-memory access request that has previously been missing is saved in the cache, and the process ends. The present invention reduces the congestion caused by too many memory access requests.
Need to check novelty before this filing date? Find Prior Art
