[LLVMdev] SCEV Alias Analysis

Hal Finkel hfinkel at anl.gov
Thu Apr 30 06:01:19 PDT 2015


----- Original Message -----
> From: "Raphael Ernani Rodrigues" <raphael at dcc.ufmg.br>
> To: "LLVM Developers Mailing List" <llvmdev at cs.uiuc.edu>
> Sent: Thursday, April 30, 2015 7:45:09 AM
> Subject: [LLVMdev] SCEV Alias Analysis
> 
> 
> 
> Dear LLVMers,
> 
> 
> I am looking into the passes that clang uses by default (using only
> the flags -O0, -O1, -O2, -O3) and I noticed that the pass scev-aa
> (ScalarEvolution-based Alias Analysis) is not being used in any of
> those options.The pass scev, however, is executed with the
> optimization levels -O1, -O2, and -O3.
> 
> 
> Does anybody know the reason why scev-aa is not included? Does it
> make the compilation too slow? Is the pass buggy?

Because of current pass manager limitations, it is not possible to preserve the pass when SCEV is invalidated. I don't believe it has known bugs, but it might be slow (it does no caching, as I recall, so I suspect some performance work is in order). The pass manager is being rewritten, and once the new framework is in place, using scev-aa might be practical.

 -Hal

> 
> 
> Thanks in advance,
> 
> 
> Raphael Ernani Rodrigues
> PhD student, University of Alberta
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> 

-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory



More information about the llvm-dev mailing list