From: albert@mustatilhi.cs.tut.fi   
      
   On 2008-02-01, Pertti Kellomdki  wrote:   
   > So the trick that e.g. the Multiflow compiler used is to guess based   
   > on the profiling information which basic blocks will usually be   
   > executed in sequence, and do instruction scheduling (ordering of   
   > instructions) on this larger region.   
      
   We have a DSP with a 3-stage pipeline, and the backend optimizer   
   /parallelizer /scheduler can use profiling information for just   
   that. For each conditional jump either the fall-through or jump target   
   is preferred when moving instructions across basic blocks.   
      
   It does help, if you have representative test data.   
      
   The main problem is that as code modules get larger, generating   
   suitable profiling information gets proportionally harder.   
      
   -Pasi   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)