Scott Brozell wrote:
> The simplest fix is to use an older intel compiler.
> I stay away from the latest from intel because I have found their
> quality assurance to be lacking.
> On the machines I use the latest versions are
> ia32 intel-compilers-9.0-20050916
> ia64 intel-compilers-9.1-20070111
> but I have not built amber with these versions (using 8.1 instead).
> I could not quickly find details about which intel versions
> have been used to build Amber at the web site.
> Other posters may be helpful here.
I agree with Scott here, I've been burnt lots of times with random intel
ifort compiler bugs. I notice they they are generally related to
vectorisation type optimisations. Hence, a while ago, I started to
compile a list of working / non-working intel ifort releases. I have not
tested the compiler version you're using (l_fc_p_9.1.032), but the list
may be useful to others.
All tests done on an i686 (P4 2.8GHz) with RHEL3-U4 . Tested on AMBER 9
with all patches up to patch level 17 (unoffically termed AMBER 9.17 by
me mjw99). This list is by no means exhaustive, I've not even looked at
the x86_64 or mpi situations.
config.h for all the tests was created using ./configure -p4 ifort_ia32
* 8.1.031 Build 20051007Z Package ID: l_fc_pc_8.1.031
NO significant diffs in make test
* 8.1.034 Build 20060606Z Package ID: l_fc_pc_8.1.034
NO significant diffs in make test
* 9.0.033 Build 20060222Z Package ID: l_fc_c_9.0.033
NO significant diffs in make test
* 9.1.036 Build 20060706Z Package ID: l_fc_c_9.1.036
Same fail as 9.1.037 (see next entry)
* 9.1.037 Build 20060816Z Package ID: l_fc_c_9.1.037
Fail on crambin_qmmmnmr test
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
sander 08582345 Unknown Unknown Unknown
sander 08500E85 Unknown Unknown Unknown
sander 08466E2B Unknown Unknown Unknown
sander 0843410E Unknown Unknown Unknown
sander 08418A8E Unknown Unknown Unknown
sander 0840DDA7 Unknown Unknown Unknown
sander 08413E51 Unknown Unknown Unknown
sander 0811FC61 Unknown Unknown Unknown
sander 0810E0B0 Unknown Unknown Unknown
sander 0810AC3D Unknown Unknown Unknown
sander 0804A44E Unknown Unknown Unknown
libc.so.6 00D3479D Unknown Unknown Unknown
sander 0804A381 Unknown Unknown Unknown
./Run.cnmr: Program error
make: *** [test.sander.DIVCON] Error 1
Recompiled with "FC= ifort -g" in config.h and debugged with idb
idb ../../exe/sander
run -O -i mdin -p cram_am1.top -c cram_am1.crd -o mdout.cnmr
(idb) run -O -i mdin -p cram_am1.top -c cram_am1.crd -o mdout.cnmr
Thread received signal SEGV
stopped at [subroutine rdnmr(logical*4, integer*4):232 0x08587f63]
232 if ( .not. calcMemRequirements) then
(idb) where
>0 0x08587f63 in rdnmr(calcmemrequirements=.TRUE. (4294967295), ierror=0)
>"../obj/rdnmr.f":232
#1 0x08506809 in nmrrdtail(line=(...), calcmemrequirements=.TRUE.
(4294967295), ierror=0) "../obj/nmrRdTail.f":21
#2 0x0846c7ef in rdtail(calcmemrequirements=.TRUE. (4294967295), ierror=0)
"../obj/rdtail.f":125
#3 0x08439a5a in initmem2_.J(...) in /tmp/amber9/exe/sander
#4 0x0841e38a in evb_amber_dealloc() in /tmp/amber9/exe/sander
#5 0x0841361a in qm_div_.J(...) in /tmp/amber9/exe/sander
#6 0x08419711 in force_.J(...) "_force.f":1028
#7 0x08121045 in runmin(xx=(...), ix=(...), ih=(...), ipairs=(...),
x=(...),
fg=(...), w=(...), ib=(...), jb=(...), conp=(...), winv=(...), igrp=(...),
skips=(...), ene=(...), carrms=-5.6871717779221343e-42, qsetup=.FALSE. (0),
qpsander=.FALSE. (0)) "_runmin.f":690
#8 0x0810f1b8 in sander() "_sander.f":1434
#9 0x0810bd39 in multisander() "_multisander.f":307
#10 0x0804a44e in main(...) in /tmp/amber9/exe/sander
#11 0x004f579d in __libc_start_main(...) in /lib/tls/libc-2.3.2.so
#12 0x0804a381 in _start(...) in /tmp/amber9/exe/sander
Works WITHOUT -p4 option on both 9.1.036 and 9.1.037 . Hence must be a
compiler bug related to " -tpp7 -axNP" flags.
* 9.1.039 Build 20060927Z Package ID: l_fc_c_9.1.039
Same fail as 9.1.037
* 9.1.040 Build 20061103Z Package ID: l_fc_c_9.1.040
NO significant diffs in make test
* 9.1.045 Build 20070320Z Package ID: l_fc_c_9.1.045
NO significant diffs in make test
regards,
Mark
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Sun May 20 2007 - 06:07:33 PDT
