| 摘要 |
The invention provides compiler loop optimization techniques to take advantage of a zero overhead loop mechanism (ZOLM) in a processor, e.g., a ZOLM in the form of a zero overhead loop buffer (ZOLB). In an illustrative embodiment, a compiler generates a first set of code, and then applies optimizations to the first set of code so as to generate a second set of code configured to operate efficiently with the ZOLB. The optimizations are designed to increase the number of loops of the first set of code that can be accommodated in the ZOLB, to further reduce the overhead of the loops placed in the ZOLB, and to eliminate redundant loading of the ZOLB. Optimizations for increasing the number of loops that can be accommodated in the ZOLB include, e.g., conditional instructions, loop splitting and function inlining. Optimizations for further reductions in loop overhead include, e.g., loop collapsing and loop interchange. Data flow analysis and loop peeling may be used to avoid redundant loading of the ZOLB.
|
