使用 CEVA 的并行处理框架将 AI 和视觉应用程序高效映射到多核 AI 处理器上.pdf

《使用 CEVA 的并行处理框架将 AI 和视觉应用程序高效映射到多核 AI 处理器上.pdf》由会员分享，可在线阅读，更多相关《使用 CEVA 的并行处理框架将 AI 和视觉应用程序高效映射到多核 AI 处理器上.pdf（25页珍藏版）》请在三个皮匠报告上搜索。

1、Efficiently Map AI and Vision Applications onto Multi-Core AI Processors Using CEVAs Parallel Processing FrameworkRami DruckerSW ArchitectCEVAExecutive Summary Next generation AI and CV applications require higher than ever computing power.Edge devices use multi-core processors to deliver high perfo

2、rmance.However,developers must efficiently map their applications onto the multiple cores,which can be difficult.CEVA has introduced the Architecture Planner tool as a new element of CDNN,CEVAs comprehensive AI SDK.In this talk,well show how the Architecture Planner tool analyzes the network model a

3、nd maps the workload onto the multiple cores in an efficient manner.Well explain key techniques used by the Tool,including symmetrical and asymmetrical multi-processing paradigms.2NeuPro-M A Family of AI ProcessorsA Full System SolutionNeuPro-M AI CoreNeuPro-M Common SubsystemMultiEnginerData&Weight

4、s CompressionAXI MatrixHostInterfaceRealtimeInterfaceFunctionalSafetySecurity logicCore SharedMemoryNeuPro-M Master Controller-Optimized for performance,control and code sizeAXIEngine#8Engine#.Engine#2Engine#1NeuPro-M EngineAXIMixedPrecisionNeural EngineComplementaryActivationUnitUnstructuredSparsit

5、yEngine(Data/Weights)Winograd Transform EngineVectorProcessorUnitEngine Local ControllerMatrix DecompositionEngine localSharedMemory4KMAC3CDNN Open Development PlatformCDNN SDKNeuPro-M Core HWNPM ControllerNPM Sub-systemCDNNFrontendCDNNBackendOptimized Layers/Operators Multi-thread CDNNAndroid NNCEV

6、AsHost/OS InterfaceCEVAsParallel Programing SolutionNPM EngineSensPro DSP HWSP ControllerHalideSPUVCU4CDNN Toolchain Workflow1Neural Network Training Optimizer ToolReal-Time Multi-Network Inferencing4CDNN Offline Optimization32Architecture Planner ToolsTraining FrameworkImageDatabaseOUTPUTBW to DDRL