Thanks for the information, that clarifies quite a bit. Unfortunately I probably have a number of memory issues that are colliding that I need to clean up.
Thanks On Mon, Nov 21, 2022 at 3:18 PM Barry Smith <[email protected]> wrote: > > My understanding of Fortran bounds checking is that before each array > access in Fortran it checks to see if the index is valid for the array you > are accessing; that is it is from start to end if you had declared the > array as > > double precision, dimension (start:end) :: A > > It should also work if the array is a Fortran allocatable array or if you > obtain the array from PETSc with VecGetArrayF90() and all its friends and > relations. > > PETSc should not change the behavior above. > > Now if there is memory corruption (or some other error) somewhere else > (like in PETSc or a more subtle problem in your code) then simply array out > of bounds then yes you can get more complicated error messages that would > usually include the PETSc stack trace. > > Instead of using valgrind you can also try running the PETSc program > with -malloc_debug, this is sort of a poor person's version of valgrind but > can sometimes provide more useful information than valgrind. > > When debugging always make sure PETSc was NOT ./configure with > --with-debugging=0 > > You can send specific error messages that are cryptic to > [email protected] and we may be able to help decipher them. > > Barry > > > > > > On Nov 21, 2022, at 2:16 PM, Nicholas Arnold-Medabalimi < > [email protected]> wrote: > > Hi Petsc users > > I'm working on an integration of Petsc into an existing fortran code. Most > of my memory debugging is very primitive and is usually accomplished using > the -check bounds option in the compiler. However with Petsc attached the > stack trace becomes much more opaque compared to the original code. At > least as far as I can tell the error becomes much harder to pin down (just > pointing to libpetsc.so). Any assistance in getting more informative error > messages or checks would be much appreciated. > > Sincerely > Nicholas > > -- > Nicholas Arnold-Medabalimi > > Ph.D. Candidate > Computational Aeroscience Lab > University of Michigan > > > -- Nicholas Arnold-Medabalimi Ph.D. Candidate Computational Aeroscience Lab University of Michigan
