Proceedings of the 29th annual ACM/IEEE international symposium on Microarchitecture
Proceedings of the ACM SIGPLAN 1999 conference on Programming language design and implementation
Dynamo: a transparent dynamic optimization system
PLDI '00 Proceedings of the ACM SIGPLAN 2000 conference on Programming language design and implementation
Rapid profiling via stratified sampling
ISCA '01 Proceedings of the 28th annual international symposium on Computer architecture
Catching Accurate Profiles in Hardware
HPCA '03 Proceedings of the 9th International Symposium on High-Performance Computer Architecture
Extending Path Profiling across Loop Backedges and Procedure Boundaries
Proceedings of the international symposium on Code generation and optimization: feedback-directed and runtime optimization
YAWARA: A Meta-Level Optimizing Computer System
IWIA '04 Proceedings of the Innovative Architecture for Future Generation High-Performance Processors and Systems
A Programmable Hardware Path Profiler
Proceedings of the international symposium on Code generation and optimization
The road not taken: Estimating path execution frequency statically
ICSE '09 Proceedings of the 31st International Conference on Software Engineering
Hi-index | 0.00 |
Aggressive execution-path-based compiler and architecture optimizations require a path profiler that can specify several hottest paths at low overheads. Based on our observation that a limited number of paths are executed frequently in hot loops, we have designed a two-level hot path detector to specify such hottest paths within hot loops. The detector consists of two tables: a filter table and an accumulator table. The filter table captures the behavior of locally hot paths. This hot path information is then sent to the accumulator table to capture the behavior of globally hot paths. The path profiler also provides us with the information on hot loops as well as problematic instructions contained in the hot paths. They may be effectively utilized for the optimizations. We evaluate the profiler using SPEC CINT2000. The results show that the two-level organization has an effect of filtering out less frequent paths, and that the top 5 frequent paths and their order can be adequately detected.