Skip to content

4.20 CUDA Dynamic Parallelism ​

待翻译

本节已建立目录框架,正文尚未翻译。

4.20.1 Introduction ​

4.20.1.1 Overview ​

4.20.2 Execution Environment ​

4.20.2.1 Parent and Child Grids ​

4.20.2.2 Scope of CUDA Primitives ​

4.20.2.3 Streams and Events ​

4.20.2.4 Ordering and Concurrency ​

4.20.3 Memory Coherence and Consistency ​

4.20.3.1 Global Memory ​

4.20.3.2 Mapped Memory ​

4.20.3.3 Shared and Local Memory ​

4.20.3.4 Local Memory ​

4.20.3.4.1 Texture Memory ​

4.20.4 Programming Interface ​

4.20.4.1 Basics ​

4.20.4.2 C++ Language Interface for CDP ​

4.20.4.2.1 Device-Side Kernel Launch ​

4.20.4.2.1.1 Launches are Asynchronous ​

4.20.4.2.1.2 Launch Environment Configuration ​

4.20.4.2.2 Events ​

4.20.4.2.3 Synchronization ​

4.20.4.2.4 Device Management ​

4.20.5 Programming Guidelines ​

4.20.5.1 Performance ​

4.20.5.1.1 Dynamic-parallelism-enabled Kernel Overhead ​

4.20.5.2 Implementation Restrictions and Limitations ​

4.20.5.2.1 Runtime ​

4.20.5.2.1.1 Memory Footprint ​

4.20.5.2.1.2 Pending Kernel Launches ​

4.20.5.3 Compatibility and Interoperability ​

4.20.6 Device-side Launch from PTX ​

4.20.6.1 Kernel Launch APIs ​

4.20.6.1.1 cudaLaunchDevice ​

4.20.6.1.2 cudaGetParameterBuffer ​

4.20.6.2 Parameter Buffer Layout ​