Skip to content

4.3 Stream-Ordered Memory Allocator ​

待翻译

本节已建立目录框架,正文尚未翻译。

4.3.1 Introduction ​

4.3.2 Memory Management ​

4.3.2.1 Allocating Memory ​

4.3.2.2 Freeing Memory ​

4.3.3 Memory Pools ​

4.3.3.1 Default/Implicit Pools ​

4.3.3.2 Explicit Pools ​

4.3.3.3 Device Accessibility for Multi-GPU Support ​

4.3.3.4 Enabling Memory Pools for IPC ​

4.3.3.4.1 Creating and Sharing IPC Memory Pools ​

4.3.3.4.2 Set Access in the Importing Process ​

4.3.3.4.3 Creating and Sharing Allocations from an Exported Pool ​

4.3.3.4.4 IPC Export Pool Limitations ​

4.3.3.4.5 IPC Import Pool Limitations ​

4.3.4 Best Practices and Tuning ​

4.3.4.1 Query for Support ​

4.3.4.2 Physical Page Caching Behavior ​

4.3.4.3 Resource Usage Statistics ​

4.3.4.4 Memory Reuse Policies ​

4.3.4.4.1 cudaMemPoolReuseFollowEventDependencies ​

4.3.4.4.2 cudaMemPoolReuseAllowOpportunistic ​

4.3.4.4.3 cudaMemPoolReuseAllowInternalDependencies ​

4.3.4.4.4 Disabling Reuse Policies ​

4.3.4.5 Synchronization API Actions ​

4.3.5 Addendums ​

4.3.5.1 cudaMemcpyAsync Current Context/Device Sensitivity ​

4.3.5.2 cudaPointerGetAttributes Query ​

4.3.5.3 cudaGraphAddMemsetNode ​

4.3.5.4 Pointer Attributes ​

4.3.5.5 CPU Virtual Memory ​