Programming models

On any processor, coding optimizations are achieved by exploiting the unique features of the hardware. In the case of the Cell Broadband Engine, the large number of SPEs, their large register file, and their ability to hide main-storage latency with concurrent computation and DMA transfers support many interesting programming models.

With the computational efficiency of the SPEs, software developers can create programs that manage dataflow as opposed to leaving dataflow to a compiler or to later optimizations.

Many of the unique features of the SPE are handled by the compiler, although programmers looking for the best performance can take advantage of the features independently of the compiler. It is almost never necessary to program the SPE in assembly language. C intrinsics provide a convenient way to program the efficient movement and buffering of data.

Application partitioning introduced some concepts for application programming.

1. Function-Offload Model
   In the Function-Offload Model, the SPEs are used as accelerators for performance-critical procedures.
2. Device-Extension Model
   The Device Extension Model is a special case of the Function-Offload Model in which the SPEs act like I/O devices.
3. Computation-Acceleration Model
   The Computation-Acceleration Model is an SPE-centric model that provides a smaller-grained and more integrated use of SPEs.
4. Streaming model
   In the Streaming Model, each SPE, in either a serial or parallel pipeline, computes data that streams through.
5. Shared-Memory Multiprocessor Model
   The Cell Broadband Engine can be programmed as a shared-memory multiprocessor, using two different instruction sets. The SPEs and the PPE fully interoperate in a cache-coherent Shared-Memory Multiprocessor Model.
6. Asymmetric-Thread Runtime Model
   Threads can be scheduled to run on either the PPE or on the SPEs, and threads interact with one another in the same way they do in a conventional symmetric multiprocessor. The Asymmetric-Thread Runtime Model extends thread task models and lightweight task models to include the different instruction sets supported by the PPE and SPE.
7. User-mode thread model
   The User-Mode Thread Model refers to one SPE thread managing a set of user-level functions running in parallel.
8. Cell application frameworks
   The complexity of implementing many of these programming models is significantly reduced by using application frameworks.
9. SPE overlays
   When code does not fit in an SPE's local store, overlays can be useful.