[LLVMbugs] [Bug 20811] New: [GVN, EarlyCSE, Missed Opt] Load from memory after invariant.end is undef

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=20811

            Bug ID: 20811
           Summary: [GVN, EarlyCSE, Missed Opt] Load from memory after
                    invariant.end is undef
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: listmail at philipreames.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Created attachment 12960
  --> http://llvm.org/bugs/attachment.cgi?id=12960&action=edit
Failing test case for GVN

Currently, GVN has handling to remove a load following a invariant.start with
no intervening stores since the memory is undefined and the load can be
replaced with undef.  This is an analogous case for invariant.end which is not
implemented.  Once the lifetime of a memory region has ended, the value of the
memory becomes undefined.  As a result, a load from such memory should be
replaced with undef.

Here's the example case (also attached):
define i8 @test(i8* %P) nounwind {
entry:
  call void @llvm.lifetime.end(i64 32, i8* %P)
  %0 = load i8* %P <-- this load is undefined
  ret i8 %0
}

The failing test case I've attached is for GVN, but EarlyCSE has the same
issue.  (In fact, this load is not currently removed by an pass in -O3.)  As
mentioned in 20809, it would be nice if the logic in question could be
refactored to be sharable between the two passes.

-- 
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/20140829/1fa1a4e7/attachment.html>