Problem running w/ calculating tensors?

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Ira *Wolfson
Posts: 69
Joined: January 24 2013
Affiliation: MPA
Contact:

Problem running w/ calculating tensors?

Post by Ira *Wolfson » April 20 2017

Hi all,

I am having problems running MCMC with:

Code: Select all

compute_tensors = T
param[r] = 0.03 0 2 0.004 0.004
All tests pass (Healpix,wmap, test.ini, test_planck.ini)

as well as running the .ini file with commented lines (i.e.):

Code: Select all

#compute_tensors = T
#param[r] = 0.03 0 2 0.004 0.004

The other .ini files used are:

Code: Select all

DEFAULT(batch2/BKPlanck.ini)
DEFAULT(batch2/common.ini)

The error I get is the following:

Code: Select all

read CMB data sets
forrtl: severe (190): for allocate(source=), source needs to be allocated
Image              PC                Routine            Line        Source             
cosmomc            000000000069D50E  for_alloc_copy        Unknown  Unknown
cosmomc            0000000000436766  Unknown               Unknown  Unknown
cosmomc            000000000043894F  Unknown               Unknown  Unknown
cosmomc            0000000000451BB6  Unknown               Unknown  Unknown
cosmomc            0000000000450C2E  Unknown               Unknown  Unknown
cosmomc            00000000005995C5  Unknown               Unknown  Unknown
cosmomc            000000000059F992  Unknown               Unknown  Unknown
cosmomc            000000000049BC46  Unknown               Unknown  Unknown
cosmomc            00000000005A2E9E  Unknown               Unknown  Unknown
cosmomc            000000000041076E  Unknown               Unknown  Unknown
libc-2.23.so       00007F535D0B2830  __libc_start_main     Unknown  Unknown
cosmomc            0000000000410669  Unknown               Unknown  Unknown
 read CMB data sets
forrtl: severe (190): for allocate(source=), source needs to be allocated
Image              PC                Routine            Line        Source             
cosmomc            000000000069D50E  for_alloc_copy        Unknown  Unknown
cosmomc            0000000000436766  Unknown               Unknown  Unknown
cosmomc            000000000043894F  Unknown               Unknown  Unknown
cosmomc            0000000000451BB6  Unknown               Unknown  Unknown
cosmomc            0000000000450C2E  Unknown               Unknown  Unknown
cosmomc            00000000005995C5  Unknown               Unknown  Unknown
cosmomc            000000000059F992  Unknown               Unknown  Unknown
cosmomc            000000000049BC46  Unknown               Unknown  Unknown
cosmomc            00000000005A2E9E  Unknown               Unknown  Unknown
cosmomc            000000000041076E  Unknown               Unknown  Unknown
libc-2.23.so       00007FC2D6FD5830  __libc_start_main     Unknown  Unknown
cosmomc            0000000000410669  Unknown               Unknown  Unknown
I see the fault is reading CMB data, but I can't see how that's even relevant...

Any clues?

Ira *Wolfson
Posts: 69
Joined: January 24 2013
Affiliation: MPA
Contact:

Problem running w/ calculating tensors?

Post by Ira *Wolfson » April 20 2017

Forgot to mention:

Ubuntu 16.04 + intel composer 2017/3 +MKL

Antony Lewis
Posts: 1936
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Problem running w/ calculating tensors?

Post by Antony Lewis » April 21 2017

Build and run cosmomc_debug - you might then get a more helpful stack trace.

Ira *Wolfson
Posts: 69
Joined: January 24 2013
Affiliation: MPA
Contact:

Problem running w/ calculating tensors?

Post by Ira *Wolfson » April 23 2017

After compiling cosmomc_debug I get this output:

Code: Select all

Number of MPI processes:          12
 Random seeds:  5220, 30035 rand_inst:  10
 Random seeds:  5416, 30035 rand_inst:  12
 Random seeds:  5124, 30036 rand_inst:   9
 Random seeds:  5325, 30036 rand_inst:  11
 Random seeds:  4738, 30037 rand_inst:   5
 Random seeds:  4838, 30037 rand_inst:   6
 file_root:test
 Random seeds:  4356, 30039 rand_inst:   1
 Random seeds:  4440, 30037 rand_inst:   2
 Random seeds:  4541, 30037 rand_inst:   3
 Random seeds:  4641, 30037 rand_inst:   4
 Random seeds:  4938, 30037 rand_inst:   7
 Random seeds:  5038, 30037 rand_inst:   8
 read CMB data sets
forrtl: severe (190): for allocate(source=), source needs to be allocated
Image              PC                Routine            Line        Source             
cosmomc_debug      0000000000BAFB1E  Unknown               Unknown  Unknown
cosmomc_debug      0000000000466C3C  objectlists_mp_ad         217  ObjectLists.f90
cosmomc_debug      000000000047E2B8  objectlists_mp_to         680  ObjectLists.f90
cosmomc_debug      00000000004881F1  objectlists_mp_ad         971  ObjectLists.f90
cosmomc_debug      00000000004F5897  paramnames_mp_par         217  ObjectParamNames.f90
cosmomc_debug      00000000004F28C0  paramnames_mp_par         172  ObjectParamNames.f90
cosmomc_debug      00000000008CA773  cosmologyparamete          68  CosmologyParameterizations.f90
cosmomc_debug      00000000008ECDAA  cosmologyconfig_m          75  CosmologyConfig.f90
cosmomc_debug      00000000005AA301  generaltypes_mp_t         408  GeneralTypes.f90
cosmomc_debug      00000000008F632C  MAIN__                    162  driver.F90
cosmomc_debug      0000000000410CAE  Unknown               Unknown  Unknown
libc-2.23.so       00007FB16D807830  __libc_start_main     Unknown  Unknown
cosmomc_debug      0000000000410BA9  Unknown               Unknown  Unknown
The only changes to test.ini are:

Code: Select all

[uncomment] compute_tensors = T
[uncomment] param[r] = 0.03 0 2 0.04 0.04
.
.
.

action = 4 --> 0
.
.
.
feedback=1 -->2

Antony Lewis
Posts: 1936
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Problem running w/ calculating tensors?

Post by Antony Lewis » April 23 2017

Can you load a slightly older ifort version (e.g. see "module avail") to compare with? It seems OK in the end-2016 version I have to hand. (could be a new compiler bug)

Ira *Wolfson
Posts: 69
Joined: January 24 2013
Affiliation: MPA
Contact:

Problem running w/ calculating tensors?

Post by Ira *Wolfson » April 23 2017

This might indeed be a new compiler issue.

The initial 2017 release present this bug, as well as every version upto and including 2017 update 3.

The older versions 2016 3 and higher are not compatible (it seems) with Ubuntu 16.04.

I already know that 14.04 X 2016 3 works fine. I was just thinking it might be a good idea to upgrade to 16.04 since 14.04 has almost run its course.


I will probably downgrade my setup if nothing else works :(

Antony Lewis
Posts: 1936
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Problem running w/ calculating tensors?

Post by Antony Lewis » April 24 2017

Actually I wonder if it's to do with a change in the way optional parameters are handled. Try today's commit on github?

Ira *Wolfson
Posts: 69
Joined: January 24 2013
Affiliation: MPA
Contact:

Problem running w/ calculating tensors?

Post by Ira *Wolfson » April 25 2017

Thank you!
It works just fine now,

what was the problem?

Ira *Wolfson
Posts: 69
Joined: January 24 2013
Affiliation: MPA
Contact:

Problem running w/ calculating tensors?

Post by Ira *Wolfson » May 10 2017

Apparently I spoke too soon.

I have now a segmentation fault, while trying to run w/ tensors.
This seg. fault does not appear while running the tests.

Code: Select all

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image              PC                Routine            Line        Source             
cosmomc_debug      0000000000C9AA01  Unknown               Unknown  Unknown
cosmomc_debug      0000000000C98B3B  Unknown               Unknown  Unknown
cosmomc_debug      0000000000C3B4F4  Unknown               Unknown  Unknown
cosmomc_debug      0000000000C3B306  Unknown               Unknown  Unknown
cosmomc_debug      0000000000BC3779  Unknown               Unknown  Unknown
cosmomc_debug      0000000000BCD076  Unknown               Unknown  Unknown
libpthread-2.23.s  00007F723BFD9390  Unknown               Unknown  Unknown
libiomp5.so        00007F723D7A17B5  Unknown               Unknown  Unknown
cosmomc_debug      0000000000C03FE5  Unknown               Unknown  Unknown
cosmomc_debug      000000000095F313  modeldata_mp_free        1085  modules.f90
cosmomc_debug      0000000000B25494  camb_mp_camb_free          79  camb.f90
cosmomc_debug      000000000070B23B  calculator_camb_m         939  Calculator_CAMB.f90
cosmomc_debug      00000000005A6E5D  generaltypes_mp_t         229  GeneralTypes.f90
cosmomc_debug      00000000005A7542  generaltypes_mp_t         247  GeneralTypes.f90
cosmomc_debug      0000000000651903  montecarlo_mp_tme         291  MCMC.f90
cosmomc_debug      00000000006545D0  montecarlo_mp_tfa         353  MCMC.f90
cosmomc_debug      000000000064CF24  montecarlo_mp_tch         144  MCMC.f90
cosmomc_debug      00000000006A7AA2  generalsetup_mp_t         137  GeneralSetup.f90
cosmomc_debug      00000000008FBCD4  MAIN__                    268  driver.F90
cosmomc_debug      0000000000410BAE  Unknown               Unknown  Unknown
libc-2.23.so       00007F723B916830  __libc_start_main     Unknown  Unknown
cosmomc_debug      0000000000410AA9  Unknown               Unknown  Unknown


I looked in the older posts, and I think 90% of the time seg. fault no. 174 is linked to cfitsio.

However all of the other tests run smoothly, so I'm a little puzzled by this...


Any clue?

-- Ira

Post Reply