Adding sparse matrices

javier_d · March 11, 2010, 10:41am

Hello,

I have a private implementation of sparse matrix which works fine but would like to move to ROOT.

I need to add large sparse matrices which are of the same size but not compatible in ROOT sense, i.e., there are couple of elements filled in one matrix but not the other. How can I do that?

Thanks.
Dano

Eddy_Offermann · March 11, 2010, 2:57pm

Hi Dano,

Have a look at

root.cern.ch/root/html/MATH_MATRIX_Index.html

and in particular the sparse matrix class

root.cern.ch/root/html/TMatrixTS … uble_.html

Eddy

javier_d · March 11, 2010, 3:10pm

Hello,

yes, that’s what I’ve done and I didn’t manage to find a solution to my problem. And that’s why I’m writing to this forum. Maybe I am doing something wrong.

I define 2 TMatrixDSparse matrices as

A:
1 0 0
1 0 0
0 0 0

B:
0 0 0
1 0 0
1 0 0

And when doing A+B I get
1 0 0
2 0 0
0 0 0

whereas I’d expect
1 0 0
2 0 0
1 0 0

I understood from the code that this is not a bug but rather a feature. It comes from the fact that the two sparse matrices are not compatible in ROOT sense.

My question is, in which way can I get the normal behavior for the addition, i.e., get the correctly summed matrix?

Thanks for your help.
Cheers,
Dano

Eddy_Offermann · March 11, 2010, 7:57pm

No that is a bug. Quite a few problems got resolved in the latest
pro release.
Which version are you using ?

javier_d · March 13, 2010, 5:51pm

Hello,

I tried 5.22 and 5.24.
With 5.25 on MAC it wouldn’t give out the result of addition but crash.

When were the fixes done? I’ll try to get access to 5.26.

Cheers,
Dano

Eddy_Offermann · March 14, 2010, 1:00pm

revision 30103 in svn solved the issue. It is contained in
release 5.26, see

root.cern.ch/root/v526/Version526.news.html#math

javier_d · March 16, 2010, 10:15am

Hello,

I tried 5.26 binary release on MAC OS X and I see that for some matrices it works ok but for others I still see problems. I attach an example script.

Here’s an output for one case where I see an error message while the summing is still done. However, the result is incorrect.

root [0] 
Processing testSparse.C...
Matrix A:

3x3 matrix is as follows

     |      0    |      1    |      2    |
--------------------------------------------
   0 |          1           0           0 
   1 |          1           0           0 
   2 |          0           0           0 

Matrix B:

3x3 matrix is as follows

     |      0    |      1    |      2    |
--------------------------------------------
   0 |          0           0           0 
   1 |          1           0           0 
   2 |          1           0           0 

now do the addition
root.exe(3946) malloc: *** error for object 0x147d194: incorrect checksum for freed object - object was probably modified after being freed.
*** set a breakpoint in malloc_error_break to debug
Matrix C = A + B:

3x3 matrix is as follows

     |      0    |      1    |      2    |
--------------------------------------------
   0 |          1           0           0 
   1 |          2           0           0 
   2 |          0           0           0 

Matrix A += B:

3x3 matrix is as follows

     |      0    |      1    |      2    |
--------------------------------------------
   0 |          1           0           0 
   1 |          2           0           0 
   2 |          0           0           0 

root [1]

I have found at least another 3 matrix configurations for this simple example, which behave in the same way.

For another example the script simply crashes before doing the sum.

root [0] 
Processing testSparse.C...
Matrix A:

3x3 matrix is as follows

     |      0    |      1    |      2    |
--------------------------------------------
   0 |          0           1           1 
   1 |          0           0           0 
   2 |          0           0           0 

Matrix B:

3x3 matrix is as follows

     |      0    |      1    |      2    |
--------------------------------------------
   0 |          0           0           0 
   1 |          1           0           0 
   2 |          1           0           0 

now do the addition
root.exe(3952) malloc: *** error for object 0x147d1d4: incorrect checksum for freed object - object was probably modified after being freed.
*** set a breakpoint in malloc_error_break to debug

 *** Break *** bus error

===========================================================
There was a crash (#4 0x000de038 in SigHandler ()).
This is the entire stack trace of all threads:
===========================================================
...

I can provide the full stack trace if needed.

Cheers,
Dano
testSparse.C (712 Bytes)

javier_d · March 18, 2010, 9:15am

Hello,

can you please confirm that this is indeed a bug and not just some problem with my installation?

Should I submit bugreport?

Cheers,
Dano

Eddy_Offermann · March 18, 2010, 12:53pm

Hi Dano,

Yes this is a bug, plesae submit it a bug report so that
it does not go off the radar.
I will try to submit a fix asap.

Eddy

Eddy_Offermann · March 28, 2010, 5:48pm

Hi Dano,

A patch has deposited in the trunk.

Thanks for reporting this bug

Eddy

javier_d · April 7, 2010, 11:22am

Hello,

I have tested the trunk (from April 2) and the addition seems to work ok now. But I see a crash when doing a+=b while c=a+b works ok. The log from the error is:

[code]*** Break *** segmentation violation

===========================================================
There was a crash (#4 0x00125048 in SigHandler ()).
This is the entire stack trace of all threads:

Thread 1 (process 2395):
#0 0x918aa711 in wait4 ()
#1 0x919197c3 in system$UNIX2003 ()
#2 0x00127bce in TUnixSystem::StackTrace ()
#3 0x00124f70 in TUnixSystem::DispatchSignals ()
#4 0x00125048 in SigHandler ()
#5
#6 0x009204f3 in Ggetexpr ()
#7 0x009b6433 in Gexec_statement ()
#8 0x0096091e in G__interpret_func ()
#9 0x00945f5c in Ggetfunction ()
#10 0x00913989 in Ggetitem ()
#11 0x00918848 in Ggetexpr ()
#12 0x00927455 in Gcalc_internal ()
#13 0x009cd461 in G__process_cmd ()
#14 0x000fc781 in TCint::ProcessLine ()
#15 0x000fbe8f in TCint::ProcessLineSynch ()
#16 0x00078e05 in TApplication::ExecuteFile ()
#17 0x0007689b in TApplication::ProcessLine ()
#18 0x010f254f in TRint::Run ()
#19 0x00001abe in main ()

[/code]

The crash doesn’t appear when I compile the script first.

I’m running on MAC OS X 10.6.3.

Is it possible to include this fix in one of the 5.22 patches?

Thanks.
Cheers,
Dano

Axel · April 8, 2010, 2:09pm

Hi,

do you have a stand-alone, running script that demonstrates the crash?

Cheers, Axel.

javier_d · April 8, 2010, 2:23pm

Hi,

I see it with the script I uploaded earlier in this thread.

Cheers,
Dano

Eddy_Offermann · April 9, 2010, 1:21am

I confirm that I see this problem too on MacOSX: .L testSparse.C+
works while .x testSparse.C gives a bus error.

Axel · April 13, 2010, 3:00pm

Hi,

with the script you posted (testSparse.C) and the current trunk I do not see any crashes on Linux 64bit. Valgrind doesn’t report any problems either. I.e. I cannot reproduce it.

Cheers, Axel.

javier_d · April 13, 2010, 3:51pm

Well, maybe it is only the MAC OS X issue. However, the addition seems to yield correct results when compiling the script. So I assume it’s OK.

Could you please comment on whether it is possible to add this fix to the next 5.22 patch release?

Thanks a lot.
Cheers,
Dano

Axel · April 14, 2010, 5:01pm

Hi,

your script works for me even on MacOS 10.6.2 32bit and 64bit with the current trunk ; the problem was likely related to the temporary object management code within CINT which was upgraded for v5.26.

We are not putting anything into the v5.22 patches branch unless an experiment officially requests it (e.g. because ROOT crashes its trigger or something as severe Sorry, you’ll have to move to v5.26.

Cheers, Axel.

rdm · April 30, 2010, 10:11am

I also do not see a problem on Mac OS X 64-bit. Is this still reproducible?

Cheers, Fons.

javier_d · April 30, 2010, 1:25pm

I’ll try again with more recent trunk and let you know.

Dano

javier_d · May 8, 2010, 12:55pm

Hello,

I just tried the script on the trunk from today and I still see the same crash.

Cheers,
Dano

Adding sparse matrices

=========================================================== There was a crash (#4 0x00125048 in SigHandler ()). This is the entire stack trace of all threads:

===========================================================
There was a crash (#4 0x00125048 in SigHandler ()).
This is the entire stack trace of all threads: