[LLVMbugs] [Bug 17525] New: Without late-vectorize, clang can run forever and uses ridiculous amounts of memory

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=17525

            Bug ID: 17525
           Summary: Without late-vectorize, clang can run forever and uses
                    ridiculous amounts of memory
           Product: clang
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: -New Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: dimitry at andric.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Created attachment 11353
  --> http://llvm.org/bugs/attachment.cgi?id=11353&action=edit
Testcase for ns_core memory exhaustion problem

I recently received a sample source file from a FreeBSD user, who observed
clang running forever, until all memory was exhausted and the process was
killed, while compiling a Mozilla WebRTC source file.  This occurred with clang
3.3 release on the i386 arch, and *only* when compiling with -march=pentium3.

With more recent trunk versions I could not reproduce the problem, so I did
some bisecting, and the revision that 'fixed' it turned out to be r189858
("Enable late-vectorization by default").  However, this just changes the
default setting for late-vectorize, and when you turn it off, the problem can
still be reproduced with trunk r192062.

I reduced the sample to the attached testcase, which should be compiled with
the following flags:

clang -O3 -m32 -march=pentium3 -mllvm -late-vectorize=false -c

This makes clang run forever and eats up all RAM, until it crashes.  Other
-march= settings do not exhibit this problem, e.g. pentium2, pentium4, core2,
etc all compile it quickly as expected, and use very little RAM.

-- 
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/20131009/3ceef336/attachment.html>