[llvm-bugs] [Bug 43058] New: [DebugInfo at O2] LiveDebugValues: Spill restores don't check location discriminator

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=43058

            Bug ID: 43058
           Summary: [DebugInfo at O2] LiveDebugValues: Spill restores don't
                    check location discriminator
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Keywords: wrong-debug
          Severity: normal
          Priority: P
         Component: Common Code Generator Code
          Assignee: unassignedbugs at nondot.org
          Reporter: jeremy.morse.llvm at gmail.com
                CC: aprantl at apple.com, chackz0x12 at gmail.com,
                    greg.bedwell at sony.com, llvm-bugs at lists.llvm.org,
                    orlando.hyams at sony.com, paul.robinson at am.sony.com,
                    stephen.tozer at sony.com, vsk at apple.com
            Blocks: 31268

Carrot Wei writes in the llvm-commits discussion for r368339, here:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190819/685357.html

  There are following code in
  function LiveDebugValues::transferSpillOrRestoreInst,

   854     } else if (TKind == TransferKind::TransferRestore &&
   855                VarLocIDs[ID].Loc.SpillLocation == *Loc) {
   856       LLVM_DEBUG(dbgs() << "Restoring Register " << printReg(Reg,
  TRI) << '('
   857                         << VarLocIDs[ID].Var.getVar()->getName() <<
")\n");

  There is no checking of VarLocIDs[ID].Kind before the usage of
  VarLocIDs[ID].Loc.SpillLocation, is this another bug?

This is certainly a bug, and one we should fix: the VarLoc discriminator should
be checked to ensure the "Loc" union contains data of the expected type. As
it's a one-liner and obviously broken, I was just going to drop a commit in
that fixed this; however comparing a clang-3.4 build with/without that change
based on r369117, the "scope bytes covered" statistic from `llvm-dwarfdump
--statistics` jumps from 46% to 54%, which is too good to be true.

My hunch is that this bug is masking the effects of bug 42772, which would
explain the sudden increase in code coverage; only 15% confidence though.
Additional investigation of what effect this change has is needed IMO.


Referenced Bugs:

https://bugs.llvm.org/show_bug.cgi?id=31268
[Bug 31268] Umbrella: debug info for optimized code
-- 
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/20190820/f3b0a8c3/attachment-0001.html>