Name
Abstract | ||
|---|---|---|
Memory references in an instruction stream often pose a challenge for performance improvements on high-per- formance microprocessors. One common scenario is that a load and a sequence of dependent instructions are on a critical path, but scheduling these instructions early to reduce the critical path length is often hindered by a preceding store, which may write to the same memory location as the load. A traditional memory disambigua- tion approach employed by a compiler may break such a potential dependence only if it can successfully disam- biguate the memory locations referenced by the store and the load. This conservative treatment in memory disambiguation is greatly alleviated by a recent architec- tural support for data speculation. With such a support, a load can be freely moved up across an aliasing store, and if the alias does occur, it is detected by a run time check and a recovery code will be invoked to re-execute the speculated instructions. Data speculation opens up exciting new opportunities for performance improve- ments, but it also demands a new memory disambigua- tion technique to guide this transformation with aliasing probabilities as the performance of unguided specula- tions will be greatly penalized by the cost of running recovery code. This paper presents a probabilistic memory disam- biguation (PMD) framework to statically derive the symbolic probabilities of the aliases among array refer- ences in application programs. We show how to apply this framework to build a profitability cost model to guide data speculation in a compiler. We also provide a set of heuristics to quickly approximate the aliasing probabilities in many common cases. Experimental results show that data speculation can often give a 1.1 to 1.2 performance speedup, but unguided data speculation may also cause a two to three fold slowdown. Therefore, it is important to guide data speculation with a cost model like the PMD framework to maximize program performance. |
Year | DOI | Venue |
|---|---|---|
1999 | 10.1145/309758.309769 | SIGARCH Computer Architecture News |
Keywords | DocType | Volume |
critical path,profitability | Journal | 27 |
Issue | Citations | PageRank |
1 | 11 | 1.06 |
References | Authors | |
6 | 3 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Roy Dz-ching Ju | 1 | 326 | 21.37 |
| Jean-François Collard | 2 | 247 | 21.24 |
| Karim Oukbir | 3 | 63 | 6.60 |