jewettaij / ndautocrr Goto Github PK

Last night, I tried compiling a parallel (OpenMP) version of the program using the "setup_gcc_parallel.sh" compiler settings, using a recent version of the code. (I forget which version, but it was last night.) When I did this, I noticed a bug when using the "-L" argument. (The other arguments I tried were working.) Fortunately the bug was not subtle. It was hard to miss. This bug caused the correlation function to lack some entries and be truncated at the wrong location. Example:

./ndautocrr -L 50 < sample_data.txt

resulted in:

ndautocrr, v0.12.0 <2021-3-24>
#  processing data set #1
#----- delta  C(delta) -----

0 0.99961896143152
1 0.95984282518274
2 0.89531720755809
3 0.79452364560286
13 -0.30364017102635

#--------------------------------------
# correlation length = 3.65069

It did not happen if I omitted the "-L" argument, but I can't be sure that the bug was specific to the feature associated with that argument. Unfortunately, I haven't been able to reproduce the error since then. (It's a "Heisenbug".) I don't think I've changed anything that would have fixed the bug. I have fixed several OpenMP specific errors recently, so I suspect it has something to do with the fact that I compiled the code with parallel (OpenMP) support.

If you run into this error, try recompiling the program without parallel (OpenMP) support.

To do that, try using one of the other compiler options such which disable OpenMP, such as "setup_gcc.sh", "setup_clang.sh", or "setup_gcc_dbg.sh". (The program may be slightly slower, but it is more likely to work.)

Alternatively, you can try reverting to version v0.10.0 of the code and see if that helps. (I don't know if it will. That version has its own set of bugs.)

Meanwhile, please report the issue, and please mention which version of the code you are using (commit hash number), compiler options you used, the arguments you used to invoke the program. If possible, please attach the file (if it's not too big) that you applied the program to when you ran into the issue. Debugging OpenMP problems can be painful, so your help is appreciated.

-Andrew