AI Architecture & Systems

Data Movement and Asynchronous Execution

1.11.1Load/Store Unit and Address Generators#

1.11.2DMA, Descriptors, and Asynchronous Copy#

1.11.3Gather/Scatter and Non-Contiguous Data Movement#

1.11.4Memory-Level Parallelism and Outstanding Requests#

1.11.5Producer–Consumer Pipelines and Double Buffering#

1.11.6Barrier, Fence, Semaphore, and Data Visibility#

1.11.7Overlap Among Compute, Data Movement, and the Memory Hierarchy#