gcc-4.6.0 optimization bug with cliquer-1.2.p7.spkg

[mariah]

On various 64-bit x86_64 Linux machines (for example, the Skynet machine eno), sage-4.7.alpha5 built with gcc-4.6.0 fails various tests with a timeout:

sage -t -long  -force_lib devel/sage/sage/graphs/graph_generators.py # Time out
sage -t -long  -force_lib devel/sage/sage/graphs/graph_coloring.py # Time out
sage -t -long  -force_lib devel/sage/sage/graphs/graph.py # Time out
sage -t -long  -force_lib devel/sage/sage/graphs/cliquer.pyx # Time out

The problem can be seen by executing the following command, which should take milliseconds but simply hangs instead:

sage: clique_number(Graph({0:[1,2,3], 1:[2], 3:[0,1]}))

New spkg adding flag -fno-ivopts: ​http://boxen.math.washington.edu/home/jdemeyer/spkg/cliquer-1.2.p9.spkg

Fixed upstream: ​http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48774

For other gcc 4.6.0 specific problems see #11216.

[mariah]

The following code triggers the problem (code does not return):

sage: g = graphs.BidiakisCube();
sage: g.clique_number()

If cliquer-1.2.p7.spkg is rebuilt with -O1 rather than -O2, then the problem goes away.

[mariah]

If src/reorder.c is built with -O1 rather than -O2, then the problem goes away.

[drkirkby]

On what platform is this problem?

Dave

[mariah]

Replying to drkirkby:

On what platform is this problem?

Dave

Apologies for not mentioning this. I see the problem on skynet:eno (x86_64-Linux-core2). I have not checked to see if the problem exists on other architectures.

[jdemeyer]

-fno-ivopts also fixes this (just like #11168).

[jdemeyer]

Diff for the cliquer spkg, for reviewing only

[mariah]

Reported upstream to gcc bugzilla, see ​| #48774

Tested on sage-4.7.rc0 on skynet/eno (x86_64-Linux-core2). All tests passed. Positive review.

[jdemeyer]

Diff for the cliquer spkg, for reviewing only

[mariah]

cliquer-1.2.p9 applied to sage-4.7.rc1 and built with 'make testlong' on skynet/eno. All tests passed. Positive review.