square root crash - Page 2

sleigher · 2010-04-08, 20:28

Quote:

Originally Posted by bsquared

Save the data! As far as I know Jason is still working on a fix for this, and he may need/want the .mat, .dep and other files...

I have all the data.

sleigher · 2010-04-08, 22:05

Done. Thanks Paul!

frmky · 2010-04-08, 23:07

Quote:

Originally Posted by fivemack

This is compiler-dependent: the build on an opteron machine with gcc-4.1.0 fails, a build on a core2 machine with gcc-4.3.2 and exactly the same makefile works.

To add one more data point, my compiles with
gcc version 4.1.2 20080704 (Red Hat 4.1.2-46)
have never crashed. Perhaps this is simply a bug in gcc 4.1.0 that has been fixed?

sleigher · 2010-04-09, 01:02

My gcc compiler is 4.1.2 in redhat 5.3. Actually CentOS but the same nevertheless. I was gonna try to use 4.2.3 and see if made a difference.

--

frmky · 2010-04-09, 05:02

In that case, I dunno. I'm using CentOS as well. I have never experienced this sqrt crash in literally hundreds of factorizations. Here are the details of my compiles.

gcc version 4.1.2 20080704 (Red Hat 4.1.2-46)

Linux physicstitan 2.6.18-53.1.14.el5 #1 SMP Wed Mar 5 11:37:38 EST 2008 x86_64 x86_64 x86_64 GNU/Linux

CentOS release 5 (Final)

OPT_FLAGS = -O3 -fomit-frame-pointer -march=amdfam10 -DNDEBUG
Even though it is built with march=amdfam10, I also run this binary on Core2-based machines.

Anything else useful?

jasonp · 2010-04-09, 12:06

Note that whatever is going wrong could still be a buffer overflow that some versions of gcc don't fail on because they shuffle stack variables around differently...

fivemack · 2010-04-09, 14:11

I'm running an example that crashes, under valgrind, but it's slow enough that I may well not have an answer by the time I go home for the weekend

Edit: after a couple of hours it reported

Code:

==22358== Warning: set address range perms: large range 102400000, a 1, v 1
==22358== Warning: client switching stacks?  SP change: 0x7FEFFB430 --> 0x7FEDC46E0
==22358==          to suppress, use: --max-stackframe=2321744 or greater

and then started reporting invalid writes performed by libgmp itself.

This made me wonder whether the segfault comes from gmp's habit of using up to very large amounts of stack space.

So I did 'ulimit -s unlimited' and then reran (not under valgrind, it takes too long); and the job has got past the point where it crashed last time. I'd be interested to know if this fixes the problem for anyone else.

2010-04-09, 14:11	#18
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 6419₁₀ Posts	I'm running an example that crashes, under valgrind, but it's slow enough that I may well not have an answer by the time I go home for the weekend Edit: after a couple of hours it reported Code: ==22358== Warning: set address range perms: large range 102400000, a 1, v 1 ==22358== Warning: client switching stacks? SP change: 0x7FEFFB430 --> 0x7FEDC46E0 ==22358== to suppress, use: --max-stackframe=2321744 or greater and then started reporting invalid writes performed by libgmp itself. This made me wonder whether the segfault comes from gmp's habit of using up to very large amounts of stack space. So I did 'ulimit -s unlimited' and then reran (not under valgrind, it takes too long); and the job has got past the point where it crashed last time. I'd be interested to know if this fixes the problem for anyone else. Last fiddled with by fivemack on 2010-04-09 at 17:20

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
All square roots failed	chris2be8	Msieve	13	2020-10-14 07:08
Square Root Days	davar55	Lounge	0	2016-03-16 20:19
NFS Square root problems	paul0	Factoring	10	2015-01-19 12:25
Square root of 3	Damian	Math	3	2010-01-01 01:56
Divisible up to Square Root	davar55	Puzzles	3	2007-09-05 15:59

2010-04-08, 22:05	#13
sleigher Mar 2010 3×17 Posts	Done. Thanks Paul!

2010-04-09, 01:02	#15
sleigher Mar 2010 110011₂ Posts	My gcc compiler is 4.1.2 in redhat 5.3. Actually CentOS but the same nevertheless. I was gonna try to use 4.2.3 and see if made a difference. --

2010-04-09, 05:02	#16
frmky Jul 2003 So Cal 2106₁₀ Posts	In that case, I dunno. I'm using CentOS as well. I have never experienced this sqrt crash in literally hundreds of factorizations. Here are the details of my compiles. gcc version 4.1.2 20080704 (Red Hat 4.1.2-46) Linux physicstitan 2.6.18-53.1.14.el5 #1 SMP Wed Mar 5 11:37:38 EST 2008 x86_64 x86_64 x86_64 GNU/Linux CentOS release 5 (Final) OPT_FLAGS = -O3 -fomit-frame-pointer -march=amdfam10 -DNDEBUG Even though it is built with march=amdfam10, I also run this binary on Core2-based machines. Anything else useful?

2010-04-09, 12:06	#17
jasonp Tribal Bullet Oct 2004 3,541 Posts	Note that whatever is going wrong could still be a buffer overflow that some versions of gcc don't fail on because they shuffle stack variables around differently...