[AMBER] Amber16+AT17: Several checks failing w/Intel Parallel Studio XE 2018.1

From: Patrice Peterson <patrice.peterson.itz.uni-halle.de>
Date: Tue, 23 Jan 2018 21:18:30 +0100

Hi,

I'm trying to build Amber16+AT17 on CentOS 7.4.1708 with Intel Parallel Studio
XE 2018.1. I have applied the patches that were supposed to fix the problem w/
r/t Intel compilers and CentOS 7.4's glibc version [1].

The build goes through fine, but I'm seeing lots of failures in the test
suite. I've decided to start with the AMD pmemd tests because they segfault.
Interestingly, the segfaults go away when I:

a) change PMEMD_FOPTFLAGS from using -O3 to -O1, or
b) remove "-ipo" from PMEMD_FOPTFLAGS.

Here are some tracebacks (I have added "-traceback" to PMEMD_FOPTFLAGS, but
that doesn't seem to be enough for full tracebacks):

Testing AMD with PME
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
libifcoremt.so.5 00002B5879C83F0C for__signal_handl Unknown Unknown
libpthread-2.17.s 00002B587C1815E0 Unknown Unknown Unknown
pmemd 00000000004C584E cit_mod_mp_setup_ 202 cit.F90
pmemd 00000000004BE601 pme_direct_mod_mp 232
pme_direct.F90
pmemd 00000000004B683C pme_force_mod_mp_ 1555
pme_force.F90
pmemd 000000000048218F runmd_mod_mp_runm 1036 runmd.F90
pmemd 0000000000405026 MAIN__ 577 pmemd.F90
pmemd 0000000000404AFE Unknown Unknown Unknown
libc-2.17.so 00002B587C5B3C05 __libc_start_main Unknown Unknown
pmemd 0000000000404A09 Unknown Unknown Unknown
  ./Run.pme.amd1: Program error
make: [pme] Error 1 (ignored)
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
libifcoremt.so.5 00002B3794BCAF0C for__signal_handl Unknown Unknown
libpthread-2.17.s 00002B37970C85E0 Unknown Unknown Unknown
pmemd 00000000004C584E cit_mod_mp_setup_ 202 cit.F90
pmemd 00000000004BE601 pme_direct_mod_mp 232
pme_direct.F90
pmemd 00000000004B683C pme_force_mod_mp_ 1555
pme_force.F90
pmemd 000000000048218F runmd_mod_mp_runm 1036 runmd.F90
pmemd 0000000000405026 MAIN__ 577 pmemd.F90
pmemd 0000000000404AFE Unknown Unknown Unknown
libc-2.17.so 00002B37974FAC05 __libc_start_main Unknown Unknown
pmemd 0000000000404A09 Unknown Unknown Unknown
  ./Run.pme.amd2: Program error
make: [pme] Error 1 (ignored)
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
libifcoremt.so.5 00002B7B709A2F0C for__signal_handl Unknown Unknown
libpthread-2.17.s 00002B7B72EA05E0 Unknown Unknown Unknown
pmemd 00000000004C584E cit_mod_mp_setup_ 202 cit.F90
pmemd 00000000004BE601 pme_direct_mod_mp 232
pme_direct.F90
pmemd 00000000004B683C pme_force_mod_mp_ 1555
pme_force.F90
pmemd 000000000048218F runmd_mod_mp_runm 1036 runmd.F90
pmemd 0000000000405026 MAIN__ 577 pmemd.F90
pmemd 0000000000404AFE Unknown Unknown Unknown
libc-2.17.so 00002B7B732D2C05 __libc_start_main Unknown Unknown
pmemd 0000000000404A09 Unknown Unknown Unknown
  ./Run.pme.amd3: Program error
make: [pme] Error 1 (ignored)
Testing AMD with IPS
diffing pmemd.ips.amd1.save with pmemd.ips.amd1
PASSED
==============================================================
diffing pmemd.ips.amd2.save with pmemd.ips.amd2
PASSED
==============================================================
diffing pmemd.ips.amd3.save with pmemd.ips.amd3
PASSED
==============================================================
Testing AMD with GB
diffing pmemd.gb.amd1.save with pmemd.gb.amd1
PASSED
==============================================================
diffing pmemd.gb.amd2.save with pmemd.gb.amd2
PASSED
==============================================================
diffing pmemd.gb.amd3.save with pmemd.gb.amd3
PASSED
==============================================================

If I leave "-O3" in place but omit "-ipo", the segfaults go away, but the
subsequent IPS tests show "possible FAILURE".

Has anyone seen that before? Does anyone have any pointers as to what's going
on?

Thanks,
Patrice

[1] https://software.intel.com/en-us/articles/intel-compiler-not-compatible-with-glibc-224-9-and-newer

-- 
Patrice Peterson
Referent für High Performance Computing
Martin-Luther-Universität Halle-Wittenberg
☎ (+49) 345-55 21864



_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber

Received on Tue Jan 23 2018 - 12:30:03 PST
Custom Search