Intel Fortran Compiler for Mac OS 10.5 Tiger?

[thomaspatko]

I have some programs that need to be compiled from source using a good Fortran compiler for Mac OS (computational chemistry programs specifically) and I understand that Intel is currently the best option available.

Has anyone tested the current version of the Intel Fotran compiler 10 for Mac OS on a beta (developer) release of Tiger with the new version of XCode?

I do not want to purchase the current student version (as it does not come with updates per my understanding) unless it will work on Tiger as I plan to upgrade shortly.

Any thoughts would be appreciated.

--Thomas

[Steven_L_Intel1]

10.5 is Leopard, isn't it? Apple has asked us not to make any public statements about support for Leopard until they actually announce it, never mind that they've been talking it up and giving away tens of thousands of copies for months.

We of course have been doing our own testing each time a new build is made available, but I would strongly recommend waiting for maybe a month before purchasing. You can download and install the free 30-day trial version while you're waiting.

[Po_D_]

Hi Steve,

I tried version 10.0.017 on Leopard and it looks like I have all kinds of error-> libSystem.B.dylib seems to be a problem ...

Is there a beta version of the compiler for Leopard ?

Best regards,

PO Dallaire

[Steven_L_Intel1]

The current 10.0 (10.0.020) supports Leopard. We can say that now. If you have problems, please report them to Intel Premier Support.

[Po_D_]

Thanks Steve for your reply.

I tried 10.0.020 this morning -> still no success. I reported the bug to Intel Premier Support and they said that I should wait for version 10.1 that will come out next month.

PODallaire

edit : I found my problem, it is with the -ipo flag, looks like there is a slash that breaks the path :

ipo: warning #11041: unresolved _fegetenv
Referenced in /usr/lib//libdl.dylib
ipo: warning #11041: unresolved _fegetround
Referenced in /usr/lib//libdl.dylib
ipo: warning #11041: unresolved _fesetenv
Referenced in /usr/lib//libdl.dylib
ipo: warning #11041: unresolved _nan
Referenced in /usr/lib//libdl.dylib
ipo: warning #11041: unresolved _nanf
Referenced in /usr/lib//libdl.dylib
ipo: warning #11041: unresolved _nextafterf
Referenced in /usr/lib//libdl.dylib

[a_e_pace]

I am running OS X 10.5, and have errors compiling. in both 10.0.016 and 10.0.020, I get the following errors when running my code:

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
a.out 000585D1 Unknown Unknown Unknown
a.out 00057A73 Unknown Unknown Unknown
a.out 0002E95C Unknown Unknown Unknown
a.out 00010ABE Unknown Unknown Unknown
a.out 00017438 Unknown Unknown Unknown
libSystem.B.dylib 91B0F97B Unknown Unknown Unknown
Unknown FFFFFFFF Unknown Unknown Unknown
a.out 0000256C Unknown Unknown Unknown
a.out 00002499 Unknown Unknown Unknown
Unknown 00000001 Unknown Unknown Unknown

I did not get any of these errors using the previous version of the compiler in OS X 10.4. Also, this is really basic code (it's for a numerical methods for PDE's class), so there is nothing complex or exotic in it.

Thank you,
Alex Pace

[asafaein]

I am using a 9.0 version and I get the same type of error compiling a program that worked well with 10.4.9. So something has changed. I wonder if Intel is going to have some updates on this?

[asafaein]

You can overcome this problem by downloading version 10 and using the /opt/intel/fce version instead of the /opt/intel/fc/ version. /fce is the 64 bit version.

[Steven_L_Intel1]

asafaein:

I am using a 9.0 version and I get the same type of error compiling a program that worked well with 10.4.9. So something has changed. I wonder if Intel is going to have some updates on this?

There will be no more 9.0 updates. Please use version 10.0. Some users have experienced problems with 10.0 on Leopard - an update coming in the next few weeks should help with that. Please contact Intel Premier Support for more assistance.

[mcneale]

I am getting exactly the same problem as a.e.pace. Using either Fortran version 9 or 10 compiled under Tiger does not run under Leopard. Downloading 10 on the leopard system compiles ok (but only works with sudo due to not being able to create tmp file error, that's another issue) but with a wide variety of compiler/linker options generates the same SIGSEGV segmentation fault error. Every binary shows the problem.

It would be great if Intel support could sort this out soon. For now I have to discourage users from upgrading to Leopard.

[Ron_Green]

Are you saying that you compile a binary under Tiger and move it onto a Leopard system to run? I would not recommend that combo.

for 10.0.020 native on Leopard:
avoid -ipo option
source /opt/intel/fce/10.0.020/bin/ifortvars.sh prior to build and run.
Use the -heap-arrays option
And try -m64 as a last resort.

10.1 is our first compiler that supports Leopard. Again, we all have to wait a little while. The above may help, but no guarantees.

ron

[Intel_C_Intel]

I am facing problem under Leopard even, with latest intel compilers 10.1 (downloaded this morning). I still have problem when using lam-mpi

1) lam-mpi can't be configured new icc and icpc 10.1.006 but can be configred using gcc and ifort 10.1.006.

2) once configured, I am facing problem to link program with lam-mpi

for instance, the program example.f90

program qui_je_suis
implicit none
include 'mpif.h'
integer :: nb_procs,rang,code

call mpi_init(code)
call mpi_comm_size(MPI_COMM_WORLD,nb_procs,code)
call mpi_comm_rank(MPI_COMM_WORLD,rang,code)
print *,'Je suis le processus ',rang,' parmi ',nb_procs
call mpi_finalize(code)
end program qui_je_suis

if I compile it:

mpif77 example.f90

gives:

Undefined symbols:
"_lam_f77init", referenced from:
_lam_f77init$non_lazy_ptr in liblamf77mpi.a(init_f.o)
"_lam_darwin_malloc_linker_hack", referenced from:
"_lam_F_handles", referenced from:
_lam_F_handles$non_lazy_ptr in liblamf77mpi.a(crank_f.o)
_lam_F_handles$non_lazy_ptr in liblamf77mpi.a(csize_f.o)
ld: symbol(s) not found

[Po_D_]

Yes, same thing here, mpi does not work with ifort and Leopard. I recompiled openmpi, tried to re-run my codes and several errors occured.

[Ron_Green]

I can try a build with 10.1 and OpenMPI on Leopard. I'll report back on my findings.

ron

[mcneale]

Ron

I have tried it both ways. Compile under Tiger doesn't work except on Tiger. Compile under Leopard also a non-starter yesterday. I will retry with your suggested options when I get back to my Leopard system on Sunday.

Thank you for your help!

Cheers
Mike

[Ron_Green]

I think I'm hot on the trail of this bug. Yes, it does only seem to affect Leopard systems.

In the configure script, LANGUAGE env var is set to LANGUAGE=C. This in a loop around line 105.

If you insert a new line at 116, and put in

unset LANGUAGE

the configure will complete and life is good. I suspect lam-mpi is doing the same setting of LANGUAGE, since Open MPI evolved from lam.

The final failure comes about in line 5533, where the ICC compiler attempts to compile a simple source file.

You can also try setting LANGUAGE=C at the command line and get the same failure with icc when you try to compile source files.

So use this work around whilst I try to find out what broke with LANGUAGE=C on Leopard.

cheers

ron

[Ron_Green]

Another interesting discovery, this seems to only fail for root. Were you logged in as root when you tried your openmpi build? I was.

[atomicham]

I have a similar problem trying to get netcdf-3.6.2 to work on MacOSX 10.5 with ifort 10.1.006. If I compile it normally and then try out a test f90 program, I get the following at runtime:

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
test 000E52B7 Unknown Unknown Unknown
test 000E4753 Unknown Unknown Unknown
test 000BB430 Unknown Unknown Unknown
test 000992CA Unknown Unknown Unknown
test 0009FC30 Unknown Unknown Unknown
libSystem.B.dylib 93C4797B Unknown Unknown Unknown
Unknown FFFFFFFF Unknown Unknown Unknown
test 00003212 Unknown Unknown Unknown
test 00002382 Unknown Unknown Unknown
test 0000233C Unknown Unknown Unknown
test 00002269 Unknown Unknown Unknown
Unknown 00000001 Unknown Unknown Unknown

If I recompile netcdf and the test program (in nf_test/tst_f90.f90) using debug (-g), then it all works just fine. there is a problem with the runtime of 10.1.006 it appears.

I have submitted a support ticket to see if we can't track it down.

[atomicham]

Doing some more digging, I created a very simple test program. Copy the following into a file, ifc_test.f90. Compile using:

ifort ifc_test.f90

Run a.out, and it will crash with the errors I listed in the previous message at TEST_FIVE. It appears when assigning a vector array.

program ifc_test

implicit none
integer :: test_one
integer, dimension(2) :: test_two
integer, dimension(2) :: test_three
real*4, dimension(2) :: test_four
real*4, dimension(4) :: test_five

print *,'WELCOME'
test_one=4
print *,'TEST_ONE=',test_one
test_two(1)=5
test_two(2)=7
print *,'TEST_TWO=',test_two
test_three=(/ 6, 8 /)
print *,'TEST_THREE=',test_three
test_four=(/ -4.2, 7.6 /)
print *,'TEST_FOUR=',test_four
test_five = (/ -90.0, -87.5, -85.0, -82.5 /)
print *,'TEST_FIVE=',test_five

end program ifc_test

[Steven_L_Intel1]

Please report this issue to Intel Premier Support. I'm aware that there's a problem recently discovered for applications running on Leopard that you may be encountering. Support can tell you if this is the same issue and when an update is available.

[Bernard_K_]

I encounter exactly the same problem.

Did you find a workaround aside from compiling with -g

Thanks,
Bernard.