LLVM Bugzilla is read-only and represents the historical archive of all LLVM issues filled before November 26, 2021. Use github to submit LLVM bugs

Bug 4 - LLC Produced executables randomly crash
Summary: LLC Produced executables randomly crash
Status: RESOLVED FIXED
Alias: None
Product: libraries
Classification: Unclassified
Component: Backend: X86 (show other bugs)
Version: trunk
Hardware: PC Linux
: P normal
Assignee: Brian R. Gaeke
URL:
Keywords:
Depends on: 41
Blocks:
  Show dependency tree
 
Reported: 2003-10-07 13:43 PDT by Chris Lattner
Modified: 2010-02-22 12:42 PST (History)
1 user (show)

See Also:
Fixed By Commit(s):


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Chris Lattner 2003-10-07 13:43:59 PDT
For an unknown reason, LLC produced executables randomly segfault.  This does
not effect the JIT.
Comment 1 Brian R. Gaeke 2003-10-07 16:21:25 PDT
I am not seeing this behavior on x86 linux. Could you supply a testcase?
Comment 2 Brian R. Gaeke 2003-10-07 20:57:03 PDT
Assuming this was just a test bug. Moving right along...
Comment 3 Chris Lattner 2003-10-08 00:51:41 PDT
This is not a test bug.  See the nightly tester for examples.  254.gap is an
example that fairly consistently fails with LLC, but works fine with the JIT.
Comment 4 Brian R. Gaeke 2003-10-08 10:17:27 PDT
oh, ok. Submit a better bug report next time. :-P
Comment 5 Brian R. Gaeke 2003-10-08 10:20:33 PDT
chris says this happens mostly on fhourstones and gap
Comment 6 Brian R. Gaeke 2003-10-08 12:59:20 PDT
Confirmed, I am seeing this on fhourstones (llc build), about 30% of the time.
Comment 7 Brian R. Gaeke 2003-10-14 13:15:39 PDT
This may have to do with bug 41, which is currently being looked at.

Comment 8 Chris Lattner 2003-10-14 13:54:03 PDT
One half of the fix:
http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20031013/008071.html
Comment 9 Chris Lattner 2003-10-14 14:14:16 PDT
The fix for Bug 41 seems to have fixed fhourstones, but GAP still fails with LLC
and not the JIT.
Comment 10 Chris Lattner 2003-10-14 14:16:49 PDT
The nondeterminism seems gone for good, so I'm closing this bug.  However, there
are still problems.  *sigh*