[LLVMbugs] [Bug 23301] New: InstCombine: objectsize of alloca optimized to undef returns unknown

bugzilla-daemon at llvm.org bugzilla-daemon at llvm.org
Mon Apr 20 17:32:25 PDT 2015


https://llvm.org/bugs/show_bug.cgi?id=23301

            Bug ID: 23301
           Summary: InstCombine: objectsize of alloca optimized to undef
                    returns unknown
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: ahmed.bougacha at gmail.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Reported by Daniel Micay in PR23280:

  #include <stdio.h>
  #include <stdlib.h>

  static inline __attribute__((always_inline)) size_t object_size(char *p) {
    return __builtin_object_size(p, 0);
  }

  int main(int argc, char **argv) {
    char buf[2];
    printf("%zu\n", object_size(buf));
    return 0;
  }


This returns -1 because buf is optimized to undef by InstCombine.

When InstCombine tries to erase an alloca, the isAllocSiteRemovable logic is
smarter than the RAUW-undef logic: when there's a (objectsize (GEP/bitcast
(alloca))), we should look through the alloca users to properly lower the
objectsize calls.  We currently just do something like alloca->RAUW(undef).

Perhaps the RAUW-undef logic and isAllocSiteRemovable should match?  This would
let the above example print 2 instead of the current -1.

-- 
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/20150421/ee6b65b2/attachment.html>


More information about the llvm-bugs mailing list