vdso_clock_gettime undefined in libpthread.a on fc8

[talziary]

Hi

I am using fc8 with glibc 2.7-2 and I have the following problem :

ifort -fpp -fast -openmp openmp_sample.f90
ipo: warning #11043: unresolved __vdso_clock_gettime
Referenced in libpthread.a(pthread_cond_timedwait.o)
ipo: remark #11001: performing single-file optimizations
ipo: remark #11005: generating object file /tmp/ipo_ifort0iT9Mp.o
openmp_sample.f90(90): (col. 7) remark: OpenMP DEFINED LOOP WAS PARALLELIZED.
openmp_sample.f90(81): (col. 7) remark: OpenMP DEFINED REGION WAS PARALLELIZED.
/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/libpthread.a(pthread_cond_timedwait.o): In function `pthread_cond_timedwait':
(.text+0xa7): undefined reference to `__vdso_clock_gettime'

The problem is due to the -static option and can be bypassed by replacing -fast by -ipo -O3 -no-prec-div

Is it a bug in libpthread.a ???

I should perhaps have tried nptl-devel but unfortunately I did not find a version of nptl-devel compatible with glibc 2.7-2

[TimP]

In case any of these are relevant, I'll point out:

Released Intel compilers haven't been tested with glibc 2.7 (as far as I can find out). Initial FC8 was supposed to be supported by ifort 10.1, but significant patches to FC8 have come out since. It's possible static libpthread wasn't tested.

I agree that the supported glibc version range for ifort is rather limiting, but it's not feasible for ifort to be tested fully against all recent variations.

glibc debian bug report r2769 reports a patch submitted last January against such a problem in x86_64 support. DId you determine whether that was relevant?

[talziary]

Quoting - tim18

In case any of these are relevant, I'll point out:

Released Intel compilers haven't been tested with glibc 2.7 (as far as I can find out). Initial FC8 was supposed to be supported by ifort 10.1, but significant patches to FC8 have come out since. It's possible static libpthread wasn't tested.

I agree that the supported glibc version range for ifort is rather limiting, but it's not feasible for ifort to be tested fully against all recent variations.

glibc debian bug report r2769 reports a patch submitted last January against such a problem in x86_64 support. DId you determine whether that was relevant?

Thank you

Most likely the r2769 bug is the same

[Sharatt_C_]

Quoting - talziary

Quoting - tim18

In case any of these are relevant, I'll point out:

Released Intel compilers haven't been tested with glibc 2.7 (as far as I can find out). Initial FC8 was supposed to be supported by ifort 10.1, but significant patches to FC8 have come out since. It's possible static libpthread wasn't tested.

I agree that the supported glibc version range for ifort is rather limiting, but it's not feasible for ifort to be tested fully against all recent variations.

glibc debian bug report r2769 reports a patch submitted last January against such a problem in x86_64 support. DId you determine whether that was relevant?

Thank you

Most likely the r2769 bug is the same

What is the current status of this bug? I am trying to compile my application in static mode for use on a cluster which does not have any nfs and so all the libraries are not shared across all the nodes. Hence it is necessary to compile the code as a static binary, but I keep on getting the following error:

"undefined reference to `__vdso_clock_gettime'"

I have tried to compile with both ifort 10 and 11. My system is Fedora 8 with glibc 2.7.

I would be very grateful if any one would be able to help.

Regards

[TimP]

If you want help from Intel, you would submit your problem report on your premier.intel.com account. As Fedora 8 is designated as supported, you might get a report about whether it seems feasible to get a glibc correction for FC, assuming you applied all the updates. Unfortunately, glibc support is not as transparent as other gnu software, and this is not directly under control of anyone reading this list. You might still need to test the suggested patch yourself.