[llvm-bugs] [Bug 30629] New: Indvars not respecting "nuw" property on add, leading to infinite loop

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=30629

            Bug ID: 30629
           Summary: Indvars not respecting "nuw" property on add, leading
                    to infinite loop
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: mikael.holmen at ericsson.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Created attachment 17414
  --> https://llvm.org/bugs/attachment.cgi?id=17414&action=edit
Reproducer

It seems like indvars can produce wrong output.

In the input we have the following loop counter update and test

  %_tmp4 = icmp ult i16 %uc.1.0, 65535
  %_tmp2 = add nuw i16 %uc.1.0, 1

indvars rewrites this to

  %_tmp2 = add nuw i16 %uc.1.0, 1
  %exitcond = icmp ne i16 %_tmp2, 0

Note that the input add instruction was "nuw", and the "nuw" property is still
left on the add after the transformation, but now the add will indeed
overflow!

If running e.g. instcombine afterwards the icmp is simply replaced with i1 true
and we get an infinite loop.

Reproduce with:
opt -S ./indvar-bug.ll -indvars

-- 
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/20161007/024e6106/attachment.html>