[Pw_forum] About compiling cd.x and fpmd.x

[Huiqun Zhou]

dear list users,

I have successfully compiled all programs of espresso 2.1.5 on a P4 cluster with very aggressive optimization
options (-O3 -tpp7 -xN -unroll -r8 -pc64). pw.x works very well both in serial and parallel runs but cp.x and fpmd.x
don't. Although cp.x works in serial run, it failed in parallel execution. fpmd.x is the worst, it fails in both serial
and parallel runs.

fpmd.x (example21)
both serial and parallel runs (serial is run on local computer) got same error message as below (memory 
addresses are different):

... some results of calculation (mostly initialization, I think) ...
running the calculation with 32 H2O molecules...
*** glibc detected *** free(): invalid next size (fast) : 0x1bebfad8 ***
forrtl: error (76): IOT trap signal

cp.x (example18)
serial run is OK. parallel run got error below:

... some results of calculation ...
forrtl: No such file or directory
forrtl: severe (29): file not found, unit 93, file /my/home/espresso-2.1.5/examples/example18/results/ .//fort.93
......

Note: the serial and parallel runs use same parallel version of cp.x and fpmd.x.

I also tried to compile a less optimized version with option -O2 -tpp7 -r8 -pc64, but got same problem.
In both aggressive and less optimized versions, Intel MKL libraries were linked statically (-i-static option). Other 
libraries may be linked dynamically. But my compute nodes are homogenous architecture with precisely equivalent
OS and other stuff, including Intel icc, ifort, and MKL.


Thanks in advance.


Huiqun Zhou


Can anybody give me a pointer?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/attachments/20050930/19d831d8/attachment.htm