[petsc-dev] malloc_dump noise from threadcomm

[Barry Smith]

  Jed,

    Stop, I've decided I don't like this approach, it mixed the low-level malloc() stuff with the higher level PETSc object stuff. 

    We already maintain a list of PetscObjects in the file inherit.c   So we can use the function PetscObjectsView() to print out all the non-destroyed objects. All we need to do is a add support to PetscObjectsView() to snag the the stack list for the first malloc in each object (which is created with PetscMalloc()) and print that out with the object info.  Note that PetscTrMallocDefault() uses PetscStackCopy() which somehow keeps the entire stack at the malloc, this could be accessed somehow in PetscObjectsView() to get the stack where the object was malloced.

   Barry







On Dec 8, 2012, at 6:20 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> On Sat, Dec 8, 2012 at 4:17 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
>    I was thinking of just marking the first memory allocation for each object, like the PetscNew(struct tp,&(h) with a bool (another record in TRSPACE), then -malloc_dump_objects would be just like -malloc_dump except only dump memory with that mark and hence one dump for each object. Nothing fancy but get's the job done.   Could have PetscNewMark() used for this purpose.
> 
> So just mark it with a Bool rather than the classid. That's fine, but the extensible malloc interfaces need to be extended to support the extra routine.