CosmoCoffee

[Alex Conley]

Has anybody had any luck getting this to run? On a RH Linux box using the
Intel MKL version 10.0.1.014, the test code seg faults, although it compiles
successfully. The problem seems to be the OPTIMIZE compiler flag.
I experience the same problem with MKL version 9.1. Turning off
the new Gibbs sampler does not address the problem, although it
appears the crash is in a different place.

Turning off OPTIMIZE makes the code run in all cases, although
(presumably) more slowly.

Initializing WMAP likelihood, version v3
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
test 08071B20 wmap_teeebb_lowl_ 452 WMAP_5yr_teeebb_pixlike.F90
test 08082D6C wmap_likelihood_5 401 WMAP_5yr_likelihood.F90
test 0805A32F MAIN__ 95 test.F90

The crash, incidentally, is in the WMAP_5yr_teeebb_pixlike.F90 file.

[Antony Lewis]

Two fixes:

1. ulimit -s unlimited

2. (better for cluster) change the code so both declarations of ninvplninv2 are changed to real(8), and change dble(ninvplninv2) to ninvplninv2 in the arguments of DGEMV. [this prevents a large temporary array being created on the stack]

I have a probably-working cosmomc test build if anyone wants to try it.

[Alex Conley]

Yup, both solutions fixed it. The latter is clearly better.
Thanks, Antony.

[Vimal Simha]

[Michael Drinkwater]

Runtime errors on WMAP5 likelihood_v3

I was getting the same seg faults running the test program as above, despite using the release that includes Antony's fixes (V3P1 RELEASE, E.Komatsu, March 5, 2009). My system is a Linux cluster with Intel compilers, version 10.1.018. I found two solutions for this setup.

(A) As a novice I used the Intel MKL link hints page to get suggested options: http://software.intel.com/en-us/articles/intel-mkl-link-line-advisor/

After lots of trial and error I found the following worked (chosing 32-bit integers on the web page made the difference). The options I set on the Intel page were: Linux, Intel(R)64, Intel, Dynamic, 32-bit, Multi-thread, Intel, None. The (working) link command it gave (plus cfitsio) was:

LIBS = -L. -L$(CFITSIO)/lib -lcfitsio -L/opt/intel/mkl/10.0.5.025/lib/em64t -Wl,--start-group -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -Wl,--end-group -openmp -lpthread

(B) The Netlib Lapack library is installed on our system and linking to its lapack library alone gives working code too:

LIBS = -L. -L$(CFITSIO)/lib -lcfitsio -llapack

Hopefully this is of some use.

[Karsten Jedamzik]

Hi,

I, unfortunately, also get segmentation faults though cosmomc has compiled.

For some reason,

ulimit -s unlimited

does not help. When I try the second fix proposed by you Antony I can find
the function 'DGEMV' but nowhere can I find a variable 'ninvplninv2'. Is this
due to changed versions ? or in which files should this variable be ?

Thanks for helping

Karsten

PS: Working on a Dell with Intel processor Intel(R) Core(TM)2 CPU
T5500 @ 1.66GHz and gfortran version 4.4

[Karsten Jedamzik]

Sorry, found the variables 'ninvplninv2' but the problem seemed to have been
now fixed by the WMAP team.

However, my segmentation faults still persist !

Karsten