>> And I have a quite fast algo that I believe is simpler than [Budimlic02] >> and I can share it with you :) > > Do you have a paper on this? I'd be interested in seeing it. >Yes, I have a tech report on this page: http://compilers/fernando/projects/soc/ and I have submitted a paper to SAS, and now I am waiting for the review. The coalescing algorithm is described in sec 4.3. It takes about 10% of the time used in Live Variables analysis in the built in LLVM compiler: 0.0846 ( 1.2%) 0.0009 ( 2.1%) 0.0855 ( 1.2%) 0.0855 ( 1.2%) Live Variable Analysis 0.0737 ( 1.0%) 0.0009 ( 2.1%) 0.0746 ( 1.1%) 0.0746 ( 1.0%) Interval Analysis - Fernando. 0.0361 ( 0.5%) 0.0007 ( 1.6%) 0.0368 ( 0.5%) 0.0368 ( 0.5%) Loop Strength Reduction 0.0146 ( 0.2%) 0.0003 ( 0.6%) 0.0149 ( 0.2%) 0.0149 ( 0.2%) Canonicalize natural loops 0.0134 ( 0.2%) 0.0000 ( 0.1%) 0.0135 ( 0.2%) 0.0135 ( 0.1%) Natural Loop Construction 0.0134 ( 0.2%) 0.0000 ( 0.1%) 0.0135 ( 0.1%) 0.0135 ( 0.1%) Phi mem coalescer - Fernando. <-- **** is this pass here. Fernando
> And I have a quite fast algo that I believe is simpler than [Budimlic02] > and I can share it with you :)Do you have a paper on this? I'd be interested in seeing it. -Tanya> > Fernando > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev >
Hi, Fernando wrote:> Yes, I have a tech report on this page: > > http://compilers/fernando/projects/soc/That's http://compilers.cs.ucla.edu/fernando/projects/soc/ in case anyone had trouble finding it. Cheers, Ralph.
Fernando Magno Quintao Pereira wrote:> Yes, I have a tech report on this page: > > http://compilers/fernando/projects/soc/Cool. I'm also interested in the Chordal Graph allocator. Are you planning to integrate it into the main llvm repository? It would make an interesting research project to compare allocation algorithms on real-world machines. -Dave
Who's your advisor? -scooter (aka "Dr. B. Scott Michel, UCLA CS 2004" :-) On Apr 12, 2007, at 4:39 PM, Fernando Magno Quintao Pereira wrote:> >>> And I have a quite fast algo that I believe is simpler than >>> [Budimlic02] >>> and I can share it with you :) >> >> Do you have a paper on this? I'd be interested in seeing it. >> > > Yes, I have a tech report on this page: > > http://compilers/fernando/projects/soc/ > > and I have submitted a paper to SAS, and now I am waiting for the > review. > The coalescing algorithm is described in sec 4.3. It takes about > 10% of > the time used in Live Variables analysis in the built in LLVM > compiler: > > 0.0846 ( 1.2%) 0.0009 ( 2.1%) 0.0855 ( 1.2%) 0.0855 > ( 1.2%) > Live Variable Analysis > 0.0737 ( 1.0%) 0.0009 ( 2.1%) 0.0746 ( 1.1%) 0.0746 > ( 1.0%) > Interval Analysis - Fernando. > 0.0361 ( 0.5%) 0.0007 ( 1.6%) 0.0368 ( 0.5%) 0.0368 > ( 0.5%) > Loop Strength Reduction > 0.0146 ( 0.2%) 0.0003 ( 0.6%) 0.0149 ( 0.2%) 0.0149 > ( 0.2%) > Canonicalize natural loops > 0.0134 ( 0.2%) 0.0000 ( 0.1%) 0.0135 ( 0.2%) 0.0135 > ( 0.1%) > Natural Loop Construction > 0.0134 ( 0.2%) 0.0000 ( 0.1%) 0.0135 ( 0.1%) 0.0135 > ( 0.1%) > Phi mem coalescer - Fernando. <-- **** is this pass here. > > Fernando > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev