internal error 0_1374 when compiling with OpenMP

[wellmann_christiangm]

When compiling a source file with -openmp and optimization (-O1 / -O2 / -O3) I get an "internal error: 0_1374". When compiling without optimization or without -openmp there are no warnings and the code compiles fine.

The code uses STL vector containers combined with a special allocator class (numa_allocator.h) which enables the correct placement of the vector data on NUMA architectures. This is necessary to get a reasonable speedup when using OpenMP on these architectures. As said above the code compiles either without optimization or without OpenMP (both of which is not a solution for me). The gnu compiler compiles the code and the executable shows the expected performance. I attached two slim versions of my code where the "base.tgz" version shows the internal error while the even more reduced "reduced.tgz" version doesn't show the error anymore.

My compiler: icpc (ICC) 11.0 20090318
glibc 2.8
kernel 2.6.25
x86_64

Test:
tar -zxvf base.tgz
cd base/
icpc -O2 -openmp -c *.cpp
icpc -O2 -openmp *.o

Thanks for any hints,
Christian

[JD_Patel]

I've been able to reproduce the problem you've submitted.
I'll work on isolating and resolving this issue.
I'll keep you posted with the progress made.
Thanks, especially for the test-case :-)

[TimP · Accepted solution]

As usual, any internal error such as this justifies a bug report on premier.intel.com.
I don't find any way to suppress the error by reducing optimization level.
It can be run with g++ 4.1.2, but it segfaults immediately in atoi(), possibly as a result of not knowing (and not checking) proper command line input.

[JenniferJ]

Quoting - tim18

As usual, any internal error such as this justifies a bug report on premier.intel.com.

Because testcase is already created, the problem is duplicated with 11.0 and 11.1 beta.
So it's not needed to submit to Premier Support.

Once the issue is fixed in the compiler or some code changes could work-around the problem, we will post the news here.

Thanks,
Jennifer

[JD_Patel]

Hi Christian,

Status update: the issue is now being worked on by engineering - will update you again as soon as the fix is available...

Thanks for your patience!
- J.D. Patel

[dghyams]

Quoting - J.D. Patel (Intel)

Hi Christian,

Status update: the issue is now being worked on by engineering - will update you again as soon as the fix is available...

Thanks for your patience!
- J.D. Patel

Same problem for me; here's hoping that the fix is forthcoming!

[Om_S_Intel]

The issue is fixed in Intel C++ compiler 11.1.038. The compiler is available for download from https://registrationcenter.intel.com/RegCenter/Download.aspx?productid=1188.

[JD_Patel]

Christian,

I verified recompiling your project (base) using the latest 11.1.038 compiler and the issue seems to be resolved.
Hopefully, this lets you move forward towards parallel programming using OpenMP ;-)

I'll mark this as a resolved & closed issue.

Thanks for your patience!

- J.D. Patel

[wellmann_christiangm]

Thanks for your help. I will try the new compiler version with my original code and see how it works.