Patent · US Active

Preloading translation and data caches using on-chip DMA engine with fast data discard

US10198849B1 · kind B1 · utility

3Cited by
4References
20Claims
0Family size

Assignee

Inventors

Key dates

Filing dateOct 26, 2017
Grant dateFeb 5, 2019
Priority date
Expiry dateOct 26, 2037

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG09G2360/18
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

Systems, apparatuses, and methods for preloading caches using a direct memory access (DMA) engine with a fast discard mode are disclosed. In one embodiment, a processor includes one or more compute units, a DMA engine, and one or more caches. When a shader program is detected in a sequence of instructions, the DMA engine is programmed to utilize a fast discard mode to prefetch the shader program from memory. By prefetching the shader program from memory, the one or more caches are populated with address translations and the shader program. Then, the DMA engine discards the shader program rather than writing the shader program to another location. Accordingly, when the shader program is invoked on the compute unit(s), the shader program and its translations are already preloaded in the cache(s).

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.