WACCM-X SD run-time error

[inos@bas_ac_uk]

Hi all,I am trying to run a specified dynamics simulation with WACCM-X. Joe McInerney provided me with a version of the CESM code to do this, which is CESM version 1.3 beta 18. I am using the Intel compiler to build the code and am running on a Cray cluster in the UK (Archer; http://www.archer.ac.uk/about-archer/). I am getting the following run-time error (see log-file attached for more information): “Fatal error in PMPI_Allgatherv: Invalid buffer pointer, error stack:
PMPI_Allgatherv(1235): MPI_Allgatherv(sbuf=0x7fffffeabb20, scount=3, MPI_DOUBLE_PRECISION, rbuf=0x7fffffeabb20, rcounts=0x7fffffeaa7a0, displs=0x7fffffeaaaa0, MPI_DOUBLE_PRECISION, comm=0xc4000002) failed
PMPI_Allgatherv(1183): Buffers must not be aliased. Consider using MPI_IN_PLACE or setting MPICH_NO_BUFFER_ALIAS_CHECK”. I am not sure where in one of the CESM scripts I would have to add the suggestions mentioned in the error message (MPI_IN_PLACE or MPICH_NO_BUFFER_ALIAS_CHECK). Do you have any suggestions?The same error seems to have been reported also by someone else on the forum (https://bb.cgd.ucar.edu/node/1002351), though it appears to be related to a different version of the model, and probably run on a different system. Still, a few files were posted there to reach a solution, which are now longer available (I logged in to Yellowstone to check), but perhaps they would help me. Does someone still has these files, so I could try?Any other ideas on how I might fix the problem?Thanks,Ingrid  

 

[jedwards]

Try Setting MPICH_NO_BUFFER_ALIAS_CHECK in your env_mach_specific file.

 

[inos@bas_ac_uk]

Ah, that was easier than I thought, and seems to have done the trick. Thanks!I'm now back to what looks like another error with a missing input file (see my post https://bb.cgd.ucar.edu/problems-downloading-data-repository). I'll post details there, as the problem fits better under that topic. -Ingrid

 

[wu_ker_wang@gmail_com]

Hi Ingrid,I am trying to run a specified dynamics simulation with cesm_1.2.2, and have the same problem as you posted. Have your problem been fixed? I am trying to set MPICH_NO_BUFFER_ALIAS_CHECK in my env_mach_specific file (by adding "setenv MPICH_NO_BUFFER_ALIAS_CHECK TRUE" in the env_mach_specific file ), however, it doesn't work. Could you let me know how did you fix the problem finally? Thanks a lot!Wuke

 

[inos@bas_ac_uk]

Hi Wuke,I don't actually remember, but based on the thread it seems that setting MPICH_NO_BUFFER_ALIAS_CHECK in my env_mach_specific file worked for me. If that's not doing the trick for you, I'm not sure what to suggest - sorry. I hope someone else can help you.
Ingrid

 

[wu_ker_wang@gmail_com]

Hi Ingrid,Thank you very much for your reply! I have figured out the problem. MPICH_NO_BUFFER_ALIAS_CHECK works for MPICH, but I am using Intel MPI, which should set I_MPI_COMPATIBILITY=4 instead.Best,
Wuke