Bug 59156 - code miscompiled with -O2 -fpeel-loops
Summary: code miscompiled with -O2 -fpeel-loops
Status: RESOLVED INVALID
Alias: None
Product: gcc
Classification: Unclassified
Component: middle-end (show other bugs)
Version: 4.7.2
: P3 normal
Target Milestone: ---
Assignee: Not yet assigned to anyone
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2013-11-16 20:20 UTC by Michael Tokarev
Modified: 2015-03-07 00:35 UTC (History)
0 users

See Also:
Host:
Target:
Build:
Known to work:
Known to fail:
Last reconfirmed:

Attachments
code demonstrating the issue (1.43 KB, text/x-csrc)
2013-11-16 20:20 UTC, Michael Tokarev 		Details
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Michael Tokarev 2013-11-16 20:20:06 UTC 
Created attachment 31231 [details]
code demonstrating the issue

Attached is a code which shows different results when compiled with -O2 (wrong) or -O1 (right).  Initially it was -O2 -funroll-loops which triggered the issue, but after reducing the code further I managed to get the same result with -O1 vs -O2 without any additional options.

Tested with debian gcc 4.7.2-5 x86-65 (wheezy), confirmed with 4.8.1-10 (debian/testing, also 64bits).  32bit code is not affected.

The code does not have any preprocessor directives.

I tried to reduce it further but it is a bit fragile - for example, removing a single call to free() makes the code generated to be correct.
Comment 1 Andrew Pinski 2013-11-16 20:32:47 UTC 
-fno-strict-aliasing fixes it.
Comment 2 Marek Polacek 2015-03-07 00:35:51 UTC 
Not a bug then.  4.7 is obsolete anyway at this point.