[llvm-bugs] [Bug 26382] New: Long compile time due to domain building

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=26382

            Bug ID: 26382
           Summary: Long compile time due to domain building
           Product: Polly
           Version: unspecified
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: normal
          Priority: P
         Component: Optimizer
          Assignee: polly-dev at googlegroups.com
          Reporter: chrisj at codeaurora.org
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Created attachment 15758
  --> https://llvm.org/bugs/attachment.cgi?id=15758&action=edit
Test case

Command:
opt  -polly-process-unprofitable -polly-code-generator=isl -polly-scops
-analyze  long_compile.ll 

The attached test is taking a very long time to compile and I noticed it was in
in the below set of complements and coalescing in buildConditionSets
(ScopInfo.cpp)

Function signature:
void buildConditionSets(Scop &S, Value *Condition, TerminatorInst *TI, Loop *L,
                   __isl_keep isl_set *Domain,
                   SmallVectorImpl<__isl_give isl_set *> &ConditionSets)
..
..
  isl_set *AlternativeCondSet =
      isl_set_complement(isl_set_copy(ConsequenceCondSet));

  ConditionSets.push_back(isl_set_coalesce(
      isl_set_intersect(ConsequenceCondSet, isl_set_copy(Domain))));
  ConditionSets.push_back(isl_set_coalesce(
      isl_set_intersect(AlternativeCondSet, isl_set_copy(Domain))));
..
..



The test finally got over after ~132 minutes and the number of sets in
ConsequenceCondSet grew to 2187. In buildDomainsWithBranchConstraints, size of
SuccDomain does cross the limit of 20 (MaxConjunctsInDomain) and we do make it
empty. I was wondering if we could safely bail out from domain building and
invalidate the current Scop in such a case rather than continuing with the
algorithm.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160129/8a2db306/attachment.html>