Skip to content

5.6 Device-Callable APIs and Intrinsics ​

待翻译

本节已建立目录框架,正文尚未翻译。

5.6.1 Memory Barrier Primitives Interface ​

5.6.1.1 Data Types ​

5.6.1.2 Memory Barrier Primitives API ​

5.6.2 Pipeline Primitives Interface ​

5.6.2.1 memcpy_async Primitive ​

5.6.2.2 Commit Primitive ​

5.6.2.3 Wait Primitive ​

5.6.2.4 Arrive On Barrier Primitive ​

5.6.3 Cooperative Groups API ​

5.6.3.1 cooperative_groups.h ​

5.6.3.1.1 class thread_block ​

5.6.3.1.2 class cluster_group ​

5.6.3.1.3 class grid_group ​

5.6.3.1.4 class thread_block_tile ​

5.6.3.1.5 class coalesced_group ​

5.6.3.2 cooperative_groups/async.h ​

5.6.3.2.1 memcpy_async ​

5.6.3.2.2 wait and wait_prior ​

5.6.3.3 cooperative_groups/partition.h ​

5.6.3.3.1 tiled_partition ​

5.6.3.3.2 labeled_partition ​

5.6.3.3.3 binary_partition ​

5.6.3.4 cooperative_groups/reduce.h ​

5.6.3.4.1 Reduce Operators ​

5.6.3.4.2 reduce ​

5.6.3.5 cooperative_groups/scan.h ​

5.6.3.5.1 inclusive_scan and exclusive_scan ​

5.6.3.6 cooperative_groups/sync.h ​

5.6.3.6.1 barrier_arrive and barrier_wait ​

5.6.3.6.2 sync ​

5.6.3.6.2.1 Grid Synchronization ​

5.6.4 CUDA Device Runtime ​

5.6.4.1 Including Device Runtime API in CUDA Code ​

5.6.4.2 Memory in the CUDA Device Runtime ​

5.6.4.2.1 Configuration Options ​

5.6.4.2.2 Allocation and Lifetime ​

5.6.4.2.2.1 Memory Declarations ​

5.6.4.2.2.2 Device and Constant Memory ​

5.6.4.2.2.3 Textures and Surfaces ​

5.6.4.2.2.4 Shared Memory Variable Declarations ​

5.6.4.2.2.5 Constant Memory ​

5.6.4.2.2.6 Symbol Addresses ​

5.6.4.3 SM Id and Warp Id ​

5.6.4.4 Launch Setup APIs ​

5.6.4.5 Device Management ​

5.6.4.6 API Reference ​

5.6.4.7 API Errors and Launch Failures ​

5.6.4.8 Device Runtime Streams ​

5.6.4.8.1 The Implicit (NULL) Stream ​

5.6.4.8.2 The Fire-and-Forget Stream ​

5.6.4.8.3 The Tail Launch Stream ​

5.6.4.9 ECC Errors ​