|
|||||||||||||||||||||||||
|
|||||||||||||||||||||||||
ABSTRACT
This paper is motivated by trends in processor models of which the Cell BE is an exemplar, and by the need to reliably apply multi-level code optimizations in safety-critical code. A MultiLoop is a loop specification construct designed to expose in a structured way details of instruction scheduling needed for performance-enhancing transformations. For a representative collection of examples from scientific computation, we have shown that MultiLoops can be used to express software branch prediction, completely eliminating branch misses. For some examples, transformations so enabled reduce code size by a factor of two, and in other examples reduce execution time by a factor of two. REFERENCES
Note: OCR errors may be found in this Reference List extracted from the full text article. ACM has opted to expose the complete List rather than only correct and linked references.
|
|||||||||||||||||||||||||