[llvm-bugs] [Bug 25388] New: Non-affine region scalar store generation dominance problem

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=25388

            Bug ID: 25388
           Summary: Non-affine region scalar store generation dominance
                    problem
           Product: Projects
           Version: unspecified
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Polly
          Assignee: polly-dev at googlegroups.com
          Reporter: doerfert at cs.uni-saarland.de
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Created attachment 15206
  --> https://llvm.org/bugs/attachment.cgi?id=15206&action=edit
Testcase

I think this bug is caused by the recent change in the scalar code generation,
namely the reload/store of scalar at the beginning/end of statements (r250625).
In non-affine regions it can happen that the scalar accesses are traversed (in
order to store them) in a way that conflicts with the dominance relation inside
the non-affine region. As we only use one BBMap for the whole statement this
will cause us to reuse values that do not dominate the current position. I
tried using multiple BBMaps for the scalar codegeneration in non-affine regions
but that does not work as we loose the mappings from the region that are
actually valid. It might work to (1) use the initial mappings but no new ones
for the scalar writes or (2) to go back to the old way of scalar code
generation.

-- 
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/20151103/d0cf6dff/attachment.html>