Hi, everyone
We have a CFD code contains about 20000 lines of Fortran code. We are implementing MPI, OpenMP and CUDA parallel computing using PGI 11.8.
But when we compile the program with -O2 or -fast flags, the compiler will report something like “PGI internal error 127, terminated by signal 11”. The error is occured while compiling a file contains some CUDA FORTRAN codes. I have writen and compiled a huge mount of CUDA FORTRAN code using PGI 11.8 and never encounted such an error. If i remove -O2 or -fast flags, it will ok.
Ocassionaly i also encounter another internal compiler error: missing case in pr_lili with -O2 flags.
I checked all our codes and just couldnot find anything wrong. Anyone could help me about this error? I think the compiler surely reports too litlle useful information that are helpful to me
Internal compiler errors are generated by the compiler and is an indication that something went wrong within the compiler during compilation. There is usually nothing wrong with the users code, rather a bug in the compiler. There are a couple of things that you could do.
*) Try upgrading to the latest PGI compiler release, 12.5. Since 11.8 there have been
numerous bug fixes, so its possible that the bug you are encountering is fixed.
*) Send a test case that reproduces the problem to trs@pgroup.com.
Thanks toepfer.
Since the code is huge, it is not easy for me to make a test case. It is possible that there will not be such an internal error if i extract some part of our codes.
I donot know how to update our 11.8 PGI to 12.5. Could you help me with this??
thanks
It is possible that trying to extract a test case will perturb it enough which will cause the internal compiler error to disappear. Typically, what we need is the file that is causing the error and any include files or module files(if the code is Fortran) that this file depends on.
Also we need the exact compile line used as well.
As far as upgrading the compilers from the 11.8 release to the 12.5 release will require you to download the 12.5 release from the pgroup.com website. This will require that you have a login and a valid subscription which will allow you to upgrade.
I faced with the same behaviour.
arom@cuda:~$ pgfortran -c m_buffer.f90
arom@cuda:~$ pgfortran -c -g m_buffer.f90
pgfortran-Fatal-/home/opt/pgi/linux86/12.5/bin/pgf901 TERMINATED by signal 11
Arguments to /home/opt/pgi/linux86/12.5/bin/pgf901
/home/opt/pgi/linux86/12.5/bin/pgf901 m_buffer.f90 -debug -x 120 0x200 -opt 0 -terse 1 -inform warn -nohpf -nostatic -x 119 0x100000 -x 19 0x400000 -x 59 4 -x 15 2 -x 49 0x400004 -x 51 0x20 -x 57 0x4c -x 58 0x10000 -x 124 0x1000 -x 57 0xfb0000 -x 58 0x78031040 -x 48 3328 -stdinc /home/opt/pgi/linux86/12.5/include:/usr/local/include:/usr/lib/gcc/i486-linux-gnu/4.4.5/include:/usr/lib/gcc/i486-linux-gnu/4.4.5/include:/usr/include -def unix -def __unix -def __unix__ -def linux -def __linux -def __linux__ -def i386 -def __i386 -def __i386__ -def __NO_MATH_INLINES -def linux86 -def __THROW= -def __extension__= -def __SSE__ -def __MMX__ -def __SSE2__ -def __SSE3__ -def __SSSE3__ -freeform -vect 48 -y 54 1 -x 49 0x1000 -modexport /tmp/pgfortranWYpeq34b4O27.cmod -modindex /tmp/pgfortranWYpeqX-i6EJS.cmdx -output /tmp/pgfortranWYpeqn-5mQHd.ilm
So, my compiler is PGI 12.5
-g options result in PGI killed be signal 11
Hi Alexey,
Thank you for the report and example code (sent to PGI Customer Support).
I’m unable to recreate the error in our 12.5 compiler and our usual diagnostic utilities don’t show any issues. However, I am able to reproduce an error in our development compiler which may or may not be related. Though most likely, they are the same issue since my error also only occurs in 32-bits with “-g”.
To confirm, can you please comment out the two “STOP” statements and recompile? In my case, the error is being caused these statements.
- Mat