search

mpip / pfft

Parallel fast Fourier transforms
GNU General Public License v3.0
54 stars 23 forks source link

Memory leak? #24

Closed rainwoodman closed 8 years ago

rainwoodman commented 8 years ago

Do you run valgrind on pfft to check for leaks?

mpip commented 8 years ago

Yes, I'm currently looking for some existing memory leaks in PFFT with valgrind. However, there are a lot of false positives due to MPI.

alanvbs commented 8 years ago

Maybe this information can be useful.
Compiling the mpich library (I have tested this with the version 3.2) with these configure options: /configure --prefix=PATH --enable-shared --enable-threads=multiple --enable-g=all --enable-error-checking=all --enable-error-messages=all I am being able to use the valgrind and also the -fsanitize=address -fsanitize=leak options from newer versions of the gcc without false alarms regarding the MPI library (I have tested these options only in personal projects. I don't known if some MPI functions can still trigger false alarms). To use the option -fsanitize=threads (useful for detecting data races) I had to compile the gcc with this configure option --disable-linux-futex.

It somewhat cumbersome to setup but for me was very useful, specially since using the fsanitize options is way faster than using the valgrind.

mpip commented 8 years ago

Thank you very much. I will give it a try.

mpip commented 8 years ago

Thanks again. I got a version of mpich working with valgrind and got no more false positives. For my test programs, PFFT seems to be free of memory leaks.