FEAP User Forum
FEAP => Installation => Topic started by: JStorm on September 19, 2017, 10:26:19 PM
-
My try to compile Feap 8.5 with big integers as default ('-fdefault-integer-8' / '-i8', ipr=1) has failed. I have noticed that in contrast to Feap 8.4 a lot of the internal arrays are forced to standard integer size (32bit). But it seems to be that this data-type change was not changed everywhere. Hence, I get compilation errors caused by data-type mismatch.
- Was the new Feap also testet with big integers?
- Why and which variables are forced to standard integers?
-
Thank you for alerting us to this problem. We had not tested the code in this fashion.
We will repair this in an upcoming update. In the meantime, you can perform your correction to the code
with a few scripts.
Run: grep kind=4 $FEAPHOME8_5 | grep integer
to find all the locations where we have forced integers to be 32-bit. Option (1) is to just zap the (kind=4) specification, option (2) would be be to change the 4 to a variable, set that variable in feap85.f, and pass that variable to all the routines via a module/common block.
-
Does that mean, that in the upcoming update the internal arrays will not be forced to integer (kind=4) in general?
My colleagues and me are really interested to have the internal arrays (node list, element connectivity, ....) as big integers like it is possible in FEAP 8.4.
-
Yes, that is the current plan.
-
Considering the parfeap 8.5, do you know about the integer kind of the PETSC routines? Is there additional effort necessary to set big integers there too?
-
PETSc has its own internal mechanisms for selecting kind. Probably you can force 64 bit integers during the configuration of PETSc, then use -i8 with FEAP, but we have never tested this combination.