 |
Memory Reference Aggregation and Classification
Data dependence analysis requires precise information on the relative order of write memory references with respect to all other references. Memory reference aggregation and classification computes, for each program context such as loop body, the following mutually disjoint sets of references:
- RO summary set records all memory locations only read (not written) within a section of code,
- WF summary set records all memory locations that are written first and then possibly read and written,
- RW summary set records all other memory locations referenced from within a context.
The power of Hybrid Analysis lies in its capacity to tolerate static analysis failures. It does so by representing the RO, WF, and RW as Uniform Sets of References (USRs).
This classification scheme is biased towards collecting information useful to automatic parallelization. Other hybrid techniques could define other aggregation or/and classification methods using USRs.
Publications and Presentations
Silvius Rus, Lawrence Rauchwerger, "Hybrid Dependence Analysis for Automatic Parallelization," Technical Report, TR05-013, Parasol Laboratory, Department of Computer Science, Texas A&M University, Nov 2005. |