There are several advantages to using one or more Streaming Engines and/or Streaming
Address Generators:
- Both hardware features allow you
to specify the count and offset for each dimension of data, up to six
dimensions.
- Both hardware features compute
addresses automatically. There is no need for the compiler to generate address
calculation instructions, which typically reduce loop performance. Using one or
more Streaming Engines can allow the compiler to software pipeline more code,
leading to improved performance.
- Streaming Engines prefetch data from above the L2 cache into memory that is
close to the CPU.
- Streaming Engines provide several additional data formatting and pattern access
features.