Charm++ parallel programming system

[Ronak Buch <ronakbuch AT gmail.com>]

Hi Jozsef,

redn_wrapper_advance(CkReductionMsg*impl_msg) just denotes that advance was called via a reduction. It includes the unpacking of variables from the message after it's been received, but it doesn't include communication or synchronization costs.

Since advance() calls rhs() directly (as in not via a proxy), the time spent inside that call to rhs() is also attributed to it.

Thanks,

Ronak

On Wed, Dec 18, 2019 at 1:07 PM Jozsef Bakosi <jbakosi AT lanl.gov> wrote:

Hi folks,

I have a projections question: I have the following function names when I hover over "Usage Profile" or "Timeline" plots:

DiagCG::DiagCG_rhs and
DiagCG::redn_wrapper_advance(CkReductionMsg*impl_msg)

The corresponding code is the following (DiagCG is a chare array):

void
DiagCG::dt()
{
  // Contribute to minimum dt across all chares the advance to next step
  contribute( sizeof(tk::real), &mindt, CkReduction::min_double,
              CkCallback(CkReductionTarget(DiagCG,advance), thisProxy) );
}

void
DiagCG::advance( tk::real newdt )
{
  // Compute rhs for next time step
  rhs();
}

void
DiagCG::rhs()
{
  ...
}

I would like to know how to interpret the usage (percentage) associated to DiagCG::DiagCG_rhs and DiagCG::redn_wrapper_advance(CkReductionMsg*impl_msg). More specifically, what does the latter include in this case? The communication/synchronization costs of the all-to-all? Does it also involve what advance() calls, e.g., rhs()?

Please let me know if you need more information.

Thanks,
Jozsef