Global Circulation Merged

The gap between CPU and memory is becoming greater and greater due to theirunmatched developing speed. So multi-level Caches have emerged in Memory System. How tooptimally use the precious buffer? How to smash the Memory Wall? It has been a research hotspot for decades. Global Loop Fusion, as a new data locality optimization, is very effective toimprove the data locality of some typical scientific computation applications and need to beimplemented urgently.The dissertation implements the Global Loop Fusion phase in Godson-2 Compiler andanalyzes the significance and effectivity of the data locality optimization.1. Global Loop Fusion, as an advanced loop optimization, avoids the necessaryoptimizing condition of common loop fusion – the two fusion candidates must beadjacent, and digs up more opportunities for loop fusion. 171.swim and 172.mgrid inSPECfp2000 are two benchmarks applied by the new optimization.2. Reuse Distance is a quantitative description for locality. The dissertation givesobvious impressions of the improvement by Global Loop Fusion on 171.swim and172.mgrid with Reuse Distance Analysis.3. The future work such as Array Contraction and Data Regrouping is proposed. Andthe two data locality optimizations are examined by Reuse Distance Analysis on171.swim.