Difference between revisions of "Stream Programming Model"
From esoterum.org
| (2 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
*William Thies. [http://groups.csail.mit.edu/commit/papers/09/thies-phd-thesis.pdf Language and Compiler Support for Stream Programs]. Ph.D. Dissertation, MIT. February 2009. | *William Thies. [http://groups.csail.mit.edu/commit/papers/09/thies-phd-thesis.pdf Language and Compiler Support for Stream Programs]. Ph.D. Dissertation, MIT. February 2009. | ||
| + | :#"Computer architectures are becoming multicore." | ||
| + | :#"Computer applications are becoming embedded and data-centric." | ||
| + | :#"Finally, there are rare cases in which the structured primitives in StreamIt have been inadequate | ||
| + | for representing a streaming communication pattern. Figure 2-20 illustrates an example | ||
| + | from video compression, where each parallel filter performs a motion prediction for a fixed | ||
| + | area of the screen. Between successive frames, each filters shares its prediction with its neighbors | ||
| + | on either side. While this could be represented with a feedback loop around the entire | ||
| + | computation, there would be complicated interleaving involved. This case reflects a broader | ||
| + | shortcoming, discussed in Section 2.7, that StreamIt is not designed for multidimensional data | ||
| + | processing." (p.49) | ||
Latest revision as of 19:18, 25 April 2010
- William Thies. Language and Compiler Support for Stream Programs. Ph.D. Dissertation, MIT. February 2009.
- "Computer architectures are becoming multicore."
- "Computer applications are becoming embedded and data-centric."
- "Finally, there are rare cases in which the structured primitives in StreamIt have been inadequate
for representing a streaming communication pattern. Figure 2-20 illustrates an example from video compression, where each parallel filter performs a motion prediction for a fixed area of the screen. Between successive frames, each filters shares its prediction with its neighbors on either side. While this could be represented with a feedback loop around the entire computation, there would be complicated interleaving involved. This case reflects a broader shortcoming, discussed in Section 2.7, that StreamIt is not designed for multidimensional data processing." (p.49)
