Dear Nek users, Currently, I am working on a simulation with a decent size (11760 * 12 elements) employing ‘lpmin = 32’. When I compile, I received an error saying: “relocation overflows omitted from the output”. I googled this error and it turned out this error comes from the ‘COMMON’ block; Static data, such as COMMON variables, and all of the code together is still limited to 2GB under any mcmodel. The suggested solution is to move the data out of a COMMON block and into a module and make the data allocatable. The most straightforward way to avoid this would be increase the ‘lpmin’ (for example if lpmin = 64, then it works), but this does not solve this issue, which is a big constraint. I am wondering has anyone has run into this and got any experience in solving it? Thanks a lot in advance
When you define your object in the common block, do you define it so that it stays the same size regardless of how many threads you use? Presumably, it should look something like this: common /mydata/ data(lx1*ly1*lz1*lelt) So that when you double the number of threads, you halve the value of lelt On Mon, Oct 8, 2018 at 8:25 PM <[email protected]> wrote:
Dear Nek users,
Currently, I am working on a simulation with a decent size (11760 * 12 elements) employing ‘lpmin = 32’. When I compile, I received an error saying: “relocation overflows omitted from the output”. I googled this error and it turned out this error comes from the ‘COMMON’ block; Static data, such as COMMON variables, and all of the code together is still limited to 2GB under any mcmodel. The suggested solution is to move the data out of a COMMON block and into a module and make the data allocatable.
The most straightforward way to avoid this would be increase the ‘lpmin’ (for example if lpmin = 64, then it works), but this does not solve this issue, which is a big constraint. I am wondering has anyone has run into this and got any experience in solving it?
Thanks a lot in advance _______________________________________________ Nek5000-users mailing list [email protected] https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
To me it looks like the number of elements are to large to fit on just 32 ranks. Increase lpmin! On 7 Nov 2018, at 17:11, nek5000-users--- via Nek5000-users <[email protected] <mailto:[email protected]> > wrote: When you define your object in the common block, do you define it so that it stays the same size regardless of how many threads you use? Presumably, it should look something like this: common /mydata/ data(lx1*ly1*lz1*lelt) So that when you double the number of threads, you halve the value of lelt On Mon, Oct 8, 2018 at 8:25 PM <[email protected] <mailto:[email protected]> > wrote: Dear Nek users, Currently, I am working on a simulation with a decent size (11760 * 12 elements) employing ‘lpmin = 32’. When I compile, I received an error saying: “relocation overflows omitted from the output”. I googled this error and it turned out this error comes from the ‘COMMON’ block; Static data, such as COMMON variables, and all of the code together is still limited to 2GB under any mcmodel. The suggested solution is to move the data out of a COMMON block and into a module and make the data allocatable. The most straightforward way to avoid this would be increase the ‘lpmin’ (for example if lpmin = 64, then it works), but this does not solve this issue, which is a big constraint. I am wondering has anyone has run into this and got any experience in solving it? Thanks a lot in advance _______________________________________________ Nek5000-users mailing list [email protected] <mailto:[email protected]> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users _______________________________________________ Nek5000-users mailing list [email protected] <mailto:[email protected]> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
you exceed 2 GB memory. Either compile with more processors or use mcmodel=big flags (something like big-memory for Nek, I forgot). On 2018-10-09 05:25, [email protected] wrote:
Dear Nek users,
Currently, I am working on a simulation with a decent size (11760 * 12 elements) employing �lpmin = 32�. When I compile, I received an error saying: �relocation overflows omitted from the output�. �I googled this error and it turned out this error comes from the �COMMON� block; Static data, such as COMMON variables, and all of the code together is still limited to 2GB under any mcmodel.� The suggested solution is to �move the data out of a COMMON block and into a module and make the data allocatable.
The most straightforward way to avoid this would be increase the �lpmin� (for example if lpmin = 64, then it works), but this does not solve this issue, which is a big constraint. I am wondering has anyone has run into this and got any experience in solving it?
Thanks a lot in advance
_______________________________________________ Nek5000-users mailing list [email protected] https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
Check our FAQs: https://nek5000.github.io/NekDoc/faq.html -----Original message-----
From:[email protected] <[email protected]> Sent: Tuesday 9th October 2018 5:25 To: NEK5000 <[email protected]> Subject: [Nek5000-users] 'COMMON' block issue
Dear Nek users, Currently, I am working on a simulation with a decent size (11760 * 12 elements) employing ‘lpmin = 32’. When I compile, I received an error saying: “relocation overflows omitted from the output”. I googled this error and it turned out this error comes from the ‘COMMON’ block; Static data, such as COMMON variables, and all of the code together is still limited to 2GB under any mcmodel. The suggested solution is to move the data out of a COMMON block and into a module and make the data allocatable. The most straightforward way to avoid this would be increase the ‘lpmin’ (for example if lpmin = 64, then it works), but this does not solve this issue, which is a big constraint. I am wondering has anyone has run into this and got any experience in solving it? Thanks a lot in advance _______________________________________________ Nek5000-users mailing list [email protected] https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
participants (1)
-
nek5000-users@lists.mcs.anl.gov