Problem running phosim v5.0.2
I am having trouble running even the simplest phosim example at NERSC. The errors seen are not consistent. Various build options have been explored, including Haswell vs KNL, building in different types of disk areas (/global/common/software, projecta, and $SCRATCH), and using different accounts (descpho and glanzman). In one example,
(Mon 08:32) descpho@cori02 $ ./phosim examples/star -c examples/nobackground
[...intervening, and reasonable looking output omitted]
Atmosphere Creator
Creating layer 0. /bin/sh: line 1: 27746 Illegal instruction /global/common/software/lsst/cori-haswell-intel/phosim/v5.0.2/bin/atmosphere < obsExtra_9999.pars
In another example:
(Mon 08:44) glanzman@cori09 $ ./phosim examples/star -c examples/nobackground
[...intervening, and reasonable looking output omitted]
Number of Sources: 40
Photons: 5.93e+05 Photons/cm^2/s: 1.15e-01
Photon Raytrace
commit none
Type Sources Photons (Sat,Rem,Rej,Acc)% Time (s) Photons/s
Error in `/global/u2/g/glanzman/phosim5/v5.0.2/bin/raytrace': double free or corruption (!prev): 0x000000000c9c8280
======= Backtrace: =========
/lib64/libc.so.6(+0x740ef)[0x2b4eff0b80ef]
/lib64/libc.so.6(+0x79646)[0x2b4eff0bd646]
/lib64/libc.so.6(+0x7a393)[0x2b4eff0be393]
/global/u2/g/glanzman/phosim5/v5.0.2/bin/raytrace[0x4069f9]
/global/u2/g/glanzman/phosim5/v5.0.2/bin/raytrace[0x4055e1]
/lib64/libc.so.6(__libc_start_main+0xf5)[0x2b4eff064725]
/global/u2/g/glanzman/phosim5/v5.0.2/bin/raytrace[0x4051c9]
======= Memory map: ========
00400000-004bd000 r-xp 00000000 00:4b 140941722 /global/u2/g/glanzman/phosim5/v5.0.2/bin/raytrace
006bd000-006be000 r--p 000bd000 00:4b 140941722 /global/u2/g/glanzman/phosim5/v5.0.2/bin/raytrace
[...memory dump omitted]
/bin/sh: line 1: 18167 Aborted /global/u2/g/glanzman/phosim5/v5.0.2/bin/raytrace < raytrace_9999_R22_S11_E000.pars
The build process used for the v5 series is the same as for the successfully built and run v3.7 series. In this case, the builds proceed using the Intel Compiler, and versions of fftw (3.3.6.2) and cfitsio (3.410) built for Cori and provided by NERSC.
Have any of the phosim developers successfully built/run v5 on Cori@NERSC? If so, I would appreciate hearing about your experience there. If a fuller output log or other details are needed, please let me know.
Thank you.
Comments (3)
-
-
closing as it appears to be related to an old version of fftw on this system.
-
- changed status to closed
- Log in to comment
Glenn will look into it. must be something strange with the compiler there.