Bug 712480

Summary: Incorrect common subexpression elimination in 4.6.0
Product: [Fedora] Fedora Reporter: Tom Lane <tgl>
Component: gccAssignee: Jakub Jelinek <jakub>
Status: CLOSED UPSTREAM QA Contact: Fedora Extras Quality Assurance <extras-qa>
Severity: unspecified Docs Contact:
Priority: unspecified    
Version: 15CC: hhorak, jakub
Target Milestone: ---   
Target Release: ---   
Hardware: x86_64   
OS: Unspecified   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2011-06-13 11:27:32 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
gzip'd intermediate file. none

Description Tom Lane 2011-06-10 18:11:59 UTC 
Created attachment 504181 [details]
gzip'd intermediate file.

Description of problem:
4.6.0 with -O2 incorrectly optimizes code of the following form:

   struct foo fooTmp;
   struct foo *fooPtr = &fooTmp;

   ...
   expression involving fooPtr->fld
   fooTmp.fld += something;
   same expression involving fooPtr->fld

The generated code assumes that the value fetched through fooPtr is still good after the update to the referenced variable.

The same code has worked fine in all previous gcc versions (it's out of postgresql FWIW)

Version-Release number of selected component (if applicable):
gcc-4.6.0-9.fc15.x86_64

How reproducible:
100%

Steps to Reproduce:
1.  Compile with /usr/bin/gcc -O2 -fno-strict-aliasing -fwrapv -g -I../../../../src/include -D_GNU_SOURCE  -c xlog.c
2. Inspect code for lines 3772, 3780, 3798

Additional info:
Unsurprisingly, -fno-gcse makes the bug go away.

I attach the .i file for the source code.
Comment 1 Jakub Jelinek 2011-06-13 11:27:32 UTC 
I've so far distilled a small self-contained testcase from this, tracking upstream.