CP2K Intel Compiler Build Issues
Zeke P
Summary of the regression tester run from 2018-03-21_17-21-50 using Linux-x86-64-intel-ku popt
Number of FAILED tests 165
Number of WRONG tests 1003
Number of CORRECT tests 1524
Number of NEW tests 10
Total number of tests 2702
-----
We have compiled CP2K/5.1 using the following arch makefile (created by our cluster support)
-----
INTEL_INC = $(MKL_ROOT)/include/fftw
INTEL_LIB = $(MKL_ROOT)/lib/intel64
CC = mpiicc
FC = mpiifort
LD = mpiifort
AR = xiar -r
DFLAGS = -D__INTEL -D__FFTSG -D__FFTW3 -D__FFTMKL -D__parallel -D__BLACS -D__SCALAPACK
CPPFLAGS = -C -traditional $(DFLAGS) -I$(INTEL_INC)
FCFLAGS = $(DFLAGS) -I$(INTEL_INC) -O2 -xHost -heap-arrays 64 -fpp -free -unroll -pc 64 -convert big_endian -nofor_main
LDFLAGS = $(FCFLAGS)
LIBS = -L$(INTEL_LIB) -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lmkl_scalapack_lp64 -lmkl_blacs_intelmpi_lp64
OBJECTS_ARCHITECTURE = machine_intel.o
-----
That being said - we aren't having much success. During compilation we are getting remark 8291 which states "Recommended relationship between field width 'W' and the number of fractional digits 'D' in this edit descriptor is 'W>=D+7'." Could this be causing the problem? I am attaching the error summary output by the regtest in case that would help.
I would appreciate any suggestions you may have - it is very likely that we are missing something stupid that is causing the compilation trouble.
Thanks so much,
Zeke Piskulich
Nico Holmberg
Which version of the Intel compiler suite/MKL are you using? Only version 17.0.4 20170411 currently works without issues, see https://www.cp2k.org/dev:compiler_support and earlier discussions on the Google groups forum.
I am no expert on Intel compiling but I think there are some issues with the arch-file, compare yours e.g. with this one https://sourceforge.net/p/cp2k/code/18346/tree//trunk/cp2k/arch/Linux-x86-64-intel.popt from the CP2K repository. For instance, I don't think the -D__FFTMKL flag exists (?), use -D__MKL for MKL linear algebra/FFT. Also, setting -fp-model precise to FCFLAGS might help.
The error/note you mentioned is harmless.
- Nico
Zeke P
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
cp2k.popt 00000000061CEDD4 Unknown Unknown Unknown
libpthread-2.12.s 00002AFB6C88A710 Unknown Unknown Unknown
cp2k.popt 00000000060C5B41 mkl_lapack_ps_avx Unknown Unknown
cp2k.popt 0000000005FE4287 Unknown Unknown Unknown
cp2k.popt 0000000002F9F719 Unknown Unknown Unknown
cp2k.popt 0000000002DC42D4 Unknown Unknown Unknown
cp2k.popt 0000000002A9B519 Unknown Unknown Unknown
cp2k.popt 0000000000CBE12C Unknown Unknown Unknown
cp2k.popt 0000000000787E2A Unknown Unknown Unknown
cp2k.popt 00000000007842BC Unknown Unknown Unknown
cp2k.popt 0000000000599F37 Unknown Unknown Unknown
cp2k.popt 00000000005999F3 Unknown Unknown Unknown
cp2k.popt 000000000045B876 Unknown Unknown Unknown
cp2k.popt 0000000000450C38 Unknown Unknown Unknown
cp2k.popt 0000000006232E02 Unknown Unknown Unknown
libc-2.12.so 00002AFB6E219D5D __libc_start_main Unknown Unknown
cp2k.popt 0000000000450229 Unknown Unknown Unknown
CC = icc
CPP =
FC = mpiifort
LD = mpiifort
AR = ar -r
MKLROOT=/panfs/pfs.local/software/install/intel/17.4/mkl
INTEL_LIB = $(MKLROOT)/lib/intel64
DFLAGS = -D__INTEL -D__FFTSG -D__parallel -D__BLACS -D__SCALAPACK -D__FFTSG -D__MAX_CONTR=3 -D__HAS_NO_GETLOG -D__STATM_RESIDENT -D__HAS_NO_SHARED_GLIBC
FCFLAGS = -D__INTEL -D__FFTSG -D__parallel -D__BLACS -D__SCALAPACK -D__FFTSG -D__MAX_CONTR=3 -D__HAS_NO_GETLOG -D__STATM_RESIDENT -D__HAS_NO_SHARED_GLIBC
FCFLAGS = $(DFLAGS) -I$ -O3 -msse2 -heap-arrays 64 -funroll-loops -fpp -free -nofor_main
FCFLAGS2 = $(DFLAGS) -O1 -msse2 -heap-arrays 64 -fpp -free
CPPFLAGS = $(FCFLAGS)
LDFLAGS = $(FCFLAGS)
LIBS = -L$(INTEL_LIB) -lstdc++ ${MKLROOT}/lib/intel64/libmkl_blas95_ilp64.a ${MKLROOT}/lib/intel64/libmkl_lapack95_ilp64.a ${MKLROOT}/lib/intel64/libmkl_scalapack_ilp64.a -Wl,--start-group ${MKLROOT}/lib/intel64/libmkl_intel_ilp64.a ${MKLROOT}/lib/intel64/libmkl_intel_thread.a ${MKLROOT}/lib/intel64/libmkl_core.a ${MKLROOT}/lib/intel64/libmkl_blacs_intelmpi_ilp64.a -Wl,--end-group -liomp5 -lpthread -lm -ldl
OBJECTS_ARCHITECTURE = machine_intel.o
graphcon.o: graphcon.F
$(FC) -c $(FCFLAGS2) $<
mao_methods.o: mao_methods.F
$(FC) -c $(FCFLAGS2) $<
-----
Do you have any thoughts on where the issue might be?
Would it be better if I gave up on the intel compiler for now and switched to trying to compile with gcc?
I really appreciate your response.
Thanks,
Zeke
Alfio Lazzaro
Hans Pabst
you may also have a look at http://xconfigure.readthedocs.io/cp2k/README/.
It appears to focus on CP2K/intel branch, but the recipe applies to the regular/official CP2K as well (and this is also explicitly covered). If you don't mind the outgrown ARCH file(s), you can simply copy the suggested ARCH files and proceed. There are no changes/edits needed since these ARCH files pick-up the usual environment variables (as set by the Intel development tools). Please note, there is also a list of "good compilers" as well as 2018 suite excluded from being good (at least for SMP builds). Key for correct regressions is to set the -fp-model if you also set any kind of target flag (aka -xHost or -xCORE-AVX2, etc.). The ARCH files also maintain a list of overridden builds rules for miscompiled translation units. However, the suggested 2017 compilers (u4 and later) are fairly ok (I am not even sure if they need anything to be worked around). I am only trimming these rules from time to time. Please note, the recipe only does not apply (or has not been tested) to CRAY systems. On CRAY, you may be better to start with an ARCH file as recently added to CP2K/master (https://raw.githubusercontent.com/cp2k/cp2k/master/cp2k/arch/Linux-x86-64-intel.*).
Hans