发明名称 Rendergraph compilation method and use thereof for low-latency execution
摘要 A graph is compiled that defines a data flow from input(s) to output(s) for images. The data flow includes one or more filters to be applied to the images. Compiling the graph includes forming an assemblage of kernel invocations for the data flow and forming a mapping between kernel invocations in code for the one or more filters and the assemblage of kernel invocations. For multiple ones of a number of frames of images, code in the one or more filters is executed, data is passed into the assemblage to indicate which execution path in the assemblage should be chosen from among a plurality of possible execution paths for one of the filters, wherein the data is determined using at least the mapping and the executing code, and kernel invocations in the indicated execution path are executed. Methods, apparatus, and computer program products are disclosed.
申请公布号 US9223551(B1) 申请公布日期 2015.12.29
申请号 US201414337855 申请日期 2014.07.22
申请人 HERE GLOBAL B.V. 发明人 Heirich Alan Bryant
分类号 G06F9/45;G06T15/00;G06T11/20 主分类号 G06F9/45
代理机构 Harrington & Smith 代理人 Harrington & Smith
主权项 1. A method, comprising: compiling a render graph that defines a data flow from one or more inputs to one or more outputs for images, wherein the data flow includes one or more filters to be applied to the images, and wherein compiling the render graph comprises: forming an assemblage of kernel invocations for the data flow andforming a mapping between kernel invocations in code for the one or more filters and the assemblage of kernel invocations; andfor multiple ones of a plurality of frames of images, rewriting code in the one or more filters prior to execution of the code so that kernel invocations use hashed values of kernel names and wherein a finite state automaton maps hash-encoded kernel sequences in the code to execution sequences in the assemblage,executing code in the one or more filters,passing data into the assemblage to indicate which execution path in the assemblage should be chosen from among a plurality of possible execution paths for one of the filters, wherein the data is determined using at least the mapping and the executing code, and executing kernel invocations in the indicated execution path.
地址 Veldhoven NL