Skip to content

4.12 Asynchronous Data Copies ​

待翻译

本节已建立目录框架,正文尚未翻译。

4.12.1 Using LDGSTS ​

4.12.1.1 Batching Loads in Conditional Code ​

4.12.1.2 Prefetching Data ​

4.12.1.3 Producer-Consumer Pattern Through Warp Specialization ​

4.12.2 Using the Tensor Memory Accelerator (TMA) ​

4.12.2.1 Using TMA to transfer one-dimensional arrays ​

4.12.2.1.1 Prefetching Data ​

4.12.2.2 Using TMA to transfer multi-dimensional arrays ​

4.12.2.2.1 Encoding a Tensor Map on Device ​

4.12.2.2.2 Device-side Encoding and Modification of a Tensor Map ​

4.12.2.2.3 Usage of a Modified Tensor Map ​

4.12.2.2.4 Creating a Template Tensor Map Value Using the Driver API ​

4.12.2.2.5 Shared-Memory Bank Swizzling ​

4.12.3 Using STAS ​