dfference between ifc and ifort

[Sukru_Besiktepe]

I have a code that I compile on Linux with ifc V7.1. Just got a Mac and ifort12.1.2.269 and now trying to compile the same programs with makefile on the new machine. These programs also link to netcdf library. I get the following error

Undefined symbols:

"_set_edges_", referenced from:

_step_ in step.o

_step_ in step.o

"_press_bias_", referenced from:

_clinic_ in clinic.o

ld: symbol(s) not found

in linux with ifc I use following options: ifc -w -L/opt/intel -lc-2.2.93 -u -O3

on Mac I use only ifort without any compiler flags.

Thanks

[Sukru_Besiktepe]

I would like to add one more thing. I now tried to compile with gfortran together with netcdf libraries build for gfortran and I got the same errors.

Any idea will be appreciated.

Sukru

[mecej4]

It is not feasible to reply until you provide the makefile, or at least those lines from the makefile that show the command used for linking. After all, you have a link-time error.

Simply saying that the makefile worked on an Machine-A and does not on Machine-B is not enough. We were not standing next to you to observe what worked and what did not.

[Sukru_Besiktepe]

I dd not want to create mess in the forum with bunch of files. But you are right. Here is the my Makefilefile. Editor does not allow to send more that I sent. Thats why I sent only the part I play with.

NETCDF_ROOT=/Users/mac/Software/intel64/netcdf-3.6.3

SRCDIR =/Users/mac/izmir/HOPS/PE/Src

PARAMDIR = ./

NCDIR =$(NETCDF_ROOT)/include

BINDIR =/Users/mac/Bin

BIN = pe_krfz_PBI

CPPFLAGS = -Ddblsigma -Drmdenbar -Dcubspl -Dvbart \

-Dunesco -Dppvmix -Dshapiro -Dsecondmean -Dvel_conv -Drmdocinc \

-Dcoast -Dresetjulian -Dpressbias -Dshapmean -Dbotfrc -Dcstfrc \

-Dpressinbias

FFLAGS =

#************************************************************************

#******************** End of USER's tunable macros **********************

#************************************************************************

#------------------------------------------------------------------------

#-------- Define string macro to parse C Pre-Processing Flags. ----------

##------------------------------------------------------------------------

find_flag = $(findstring $(flag),$(CPPFLAGS))

#------------------------------------------------------------------------

#--------------------- Internal macro definitions. ----------------------

#------------------------------------------------------------------------

RMBLKLINES = rmblklines

SHELL = /bin/sh

RM = rm -f

ECHO = echo

FALSE = false

LIB = -L$(NETCDF_ROOT)/lib -lnetcdff -lnetcdf

CPP = cpp -P -traditional-cpp -I. -I$(PARAMDIR) -I$(SRCDIR) -I$(NCDIR)

FC = ifort

wk_flags := nest2larger nest2smaller

ifneq (,$(strip $(foreach flag,$(wk_flags),$(find_flag))))

CPP := $(CPP) -I$(PVM_ROOT)/include

LIB := $(LIB) -L$(PVM_ROOT)/lib/$(PVM_ARCH) -lfpvm3 -lpvm3 -lsocket

endif

LIB := $(LIB)

#------------------------------------------------------------------------

#---------------- Look for bad C Pre-Processing Flags. ------------------

#------------------------------------------------------------------------

VALID_CPP_OPT = -Dadd_tide -Dadvtide -Dadvtide0 -Daixdate -Danalytical \

-DAsselinFilt -DAsselinFilt_cod -Dbarotropic -DbioAnder \

-DbioDuse -DbioFasham -DbioMcGillic -Dbio_12_A -Dbioadjloc \

-Dbioadjvert -Dbndy_rlx -Dbotfrc -Dbottom -Dclose_tracers \

-Dcoast -Dcoastedge -Dcod_ing -Dcod_square -Dcodlim -Dcodunlim \

-Dcodvadvect -Dcodvmix -Dcraydate -Dcstfrc -Dcubspl -Dcyclic \

-Ddblprec -Ddblsigma -Ddecdate -DdusDroop -Dexplicitvmix \

-Dext_tide -Dextraction -Dfcsterr -Dforcing -Dfrozentrc \

-Dgendbg -Dgridold -Dgrids -Dhpg4 -Dimp_bnd_rlx -Dislands \

-Dldrifters -Dleqstate -Dlinear_physics -Dmixtide \

-Dnest_ext2lrgr -Dnest_ext2smlr -Dnest2larger -Dnest2smaller \

-Dnestnultest -Dnesttime -Dnkfix -Dnocfl -Dnotadvm \

-Dnotadvt -Dnowcfl -Dnoxtrp -Doias -Dpeprf -Dpindex -Dposmxtid \

-Dppvmix -Dpressbias -Dpressinbias -Dpttrcsrc -Dresetjulian \

-Drivsrc -Drmcomments -Drmdenbar -Drmdocinc -Dsecondmean \

-Dshapiro -Dshapmean -Dshapnocoastflux -Dsimpramp -Dsor5pt \

-Dsundate -Dsunflush -Dsunfpe -Dtide_zero -Dunesco \

-Dusrdiagnostic -Dvbart -Dvel_conv -Dwgterr -Dzoo_ing

INVALID_CPP_OPT = $(strip $(filter-out $(VALID_CPP_OPT),$(CPPFLAGS)))

#------------------------------------------------------------------------

#--------------------- Redefine compiling macros. -----------------------

#------------------------------------------------------------------------

MAKEFLAGS := -r -$(MAKEFLAGS)

.SUFFIXES:

.SUFFIXES: .o .f .F .FF

%.o: %.f

-$(FC) -c $(FFLAGS) -o $*.o $*.f

%.o: %.F

%.o: %.FF

%.f: %.F

-$(CPP) $(CPPFLAGS) $< ./$*.cpp

-./$(RMBLKLINES) < $*.cpp > $*.f

-$(RM) $*.cpp

%.f: %.FF

-$(CPP) $(CPPFLAGS) $< ./$*.f

#------------------------------------------------------------------------

#-------------- Define alternate paths for source codes. ----------------

#------------------------------------------------------------------------

#--------------------------

#--- Define source path ---

#--------------------------

testpath = .

pathstr =

ifeq (,$(strip $(filter $(SRCDIR),$(subst :, ,$(testpath)))))

testpath := $(testpath):$(SRCDIR)

pathstr := $(SRCDIR)

endif

ifneq (,$(pathstr))

vpath %.F $(pathstr)

vpath %.FF $(pathstr)

endif

#---------------------------

#--- Define include path ---

#---------------------------

ifeq (,$(strip $(filter $(PARAMDIR),$(subst :, ,$(testpath)))))

testpath := $(PARAMDIR):$(testpath)

ifneq (,$(pathstr))

pathstr := $(PARAMDIR):$(pathstr)

else

pathstr := $(PARAMDIR)

endif

endif

ifeq (,$(strip $(filter $(NCDIR),$(subst :, ,$(testpath)))))

testpath := $(testpath):$(NCDIR)

ifneq (,$(pathstr))

pathstr := $(pathstr):$(NCDIR)

else

pathstr := $(NCDIR)

endif

endif

wk_flags := nest2larger nest2smaller

ifneq (,$(strip $(foreach flag,$(wk_flags),$(find_flag))))

ifeq (,$(strip $(filter $(PVM_ROOT)/include,$(subst :, ,$(testpath)))))

testpath := $(pathstr):$(PVM_ROOT)/include

ifneq (,$(pathstr))

pathstr := $(pathstr):$(PVM_ROOT)/include

else

pathstr := $(PVM_ROOT)/include

endif

endif

endif

ifneq (,$(pathstr))

vpath %.h $(pathstr)

vpath %.inc $(pathstr)

endif

#------------------------------------------------------------------------

#------- Preserve intermediate source codes only when debugging. --------

#------------------------------------------------------------------------

dbg_flags = sunfpe gendbg

ifneq (,$(strip $(foreach flag,$(dbg_flags),$(find_flag))))

.PRECIOUS: %.f

endif

#------------------------------------------------------------------------

#--- Declare "phony" targets. GNUmake can always run these targets. ----

#------------------------------------------------------------------------

ifeq (,$(INVALID_CPP_OPT))

.PHONY: clean

else

.PHONY: clean $(BIN)

endif

[mecej4]

The missing externals, _set_edges_ and _press_bias_, are not part of NetCDF or HDF5.

Are they routines that are in your source files, or in user libraries that you forgot to include in the list of libraries to search?

[Sukru_Besiktepe]

They are routines as part of my source fies. Actually this is the ocean model which consists of several fortran files. According to different requirements I configure the model with C-preprocessor options. Normally model works with all cases excep when I used one specific configuration which includes press_bias0.F. But error shows up in other subroutines (press_bias.F and set_edges.F). I tried diffent options if compiler did not like the underscores etc. Now, I am wondering if the order of linking object files are importnt or not?

[mecej4]

Now, I am wondering if the order of linking object files are importnt or not?

The order in which .o files are listed should not matter. If, however, these files are in a library file, the order of the library files on the command line does matter.

Have you tried

nm press_bias0.o | grep -i press_bias

yet? That is sure to show up mismatches in underscores or case.

[Sukru_Besiktepe]

nm press_bias0.o | grep -i press_bias gives

0000000000000000 T _press_bias0_

0000000000000640 d _press_bias0_$FIRST.0.1

000000000000da40 b _press_bias0_$PBID.0.1

000000000000da44 b _press_bias0_$SHID.0.1

0000000000000840 b _press_bias0_$TP2.0.1

00000000000006f0 S _press_bias0_.eh

00000000000005e0 d press_bias0_$format_pack.0.1

when I tried

nm clinic.o | grep -i press_bias which compilation was giving an error

Undefined symbols:

.

.

.

"_press_bias_", referenced from:

_clinic_ in clinic.o

it gives

U _press_bias_

[mecej4]

0000000000000000 T _press_bias0_

is the important output. It shows that there is a routine press_bias0 that can be called, rather than the expected press_bias. Check the other routine name similarly.

Check the subroutine names at the calling point as well as where they are declared.

[Sukru_Besiktepe]

I guess I fiexd the problem. After cheking subroutine names at the calling points as you suggested, I found very strange problem. When I list the directory I can not see the clinic.f. But when I gave the command ls clinic.f it shows up. clinic.f was remained from previous configuration which uses press_bias. But in the new configuration I need press_bias0. When I gave rm clinic.f command, I was able to remove clinic.f. Then, I recompiled the program and it works now.

I do not know if I explained clearly, but this is really strange and dangereous and it may be due to the MAC file system.

Thanks for all your help.

[mecej4]

I understand. Such thing happen regularly. GUIs are nice for end users, but they create some problems such as this when used for program development.