direct sparse solvers and memory allocation

Petros_Mamales — Tue, 17 Jan 2012 21:20:36 GMT

Hi,
In all linear eqtn solvers there is an initial stage where memory allocations take place given the "dimensions" of the problem. This is very beneficial for using the same memory (as in a PDE setting) for reusing the same solver without paying the penalty of memory allocations.
With direct sparse solvers though, I see that the very rhs and sparse matrix are used for initialization.
Since I have overwritten the mkl allocator with another one (tbb), I tried to see, if any memory allocations happen at all, by putting a breakpoint, but got nothing. Since I never had an opportunity to check my overwritting of the MKL allocators-still looking for a function totest it-, though, I cannot be sure.
Can you please confirm, if this is the case, that is, that no memory allocations happen internally in the direct sparse solver ?
TIA for your help,
Petros

Update: it's been a little while I asked this. anyone noticed ? ;- ))

direct sparse solvers and memory allocation

Murat_G_Intel — Sun, 04 Mar 2012 11:36:59 GMT

Hi Petros,

Direct sparse solvers typically require more memory than the original sparse matrix. Therefore, PARDISO should do internal memory allocation. Can you explain how you are overwriting MKL allocators?

Thank you,

Efe

topic direct sparse solvers and memory allocation in Intel® oneAPI Math Kernel Library

direct sparse solvers and memory allocation

direct sparse solvers and memory allocation