[petsc-users] Question about LogSummary

[Blaise A Bourdin]

On Mar 24, 2013, at 12:28 PM, Matthew Knepley <knepley at gmail.com<mailto:knepley at gmail.com>>
 wrote:

On Mon, Mar 25, 2013 at 12:42 AM, Blaise A Bourdin <bourdin at lsu.edu<mailto:bourdin at lsu.edu>> wrote:

On Mar 23, 2013, at 11:50 PM, Matthew Knepley <knepley at gmail.com<mailto:knepley at gmail.com>>
 wrote:

On Sun, Mar 24, 2013 at 12:24 PM, Blaise A Bourdin <bourdin at lsu.edu<mailto:bourdin at lsu.edu>> wrote:
Hi,

I am trying to do some scaling tests but can't figure out something about LogSummary: When no custom stages / events are defined, how come "Main Stage" is not 100% of the time? What is not counted? Here is a example of what I get:

Library loading?
Really? Is there anything I can do to test if it really is library loading or to reduce loading time? I am attaching the full log summary
Blaise


   Matt

Summary of Stages:   ----- Time ------  ----- Flops -----  --- Messages ---  -- Message Lengths --  -- Reductions --
                        Avg     %Total     Avg     %Total   counts   %Total     Avg         %Total   counts   %Total
 0:      Main Stage: 2.5763e+03  65.1%  3.2380e+15 100.0%  1.230e+11 100.0%  8.493e+02      100.0%  7.013e+06  97.0%

Where is the missing 35% of the time? I call PetscLogBegin right after PetscInitialize.

Okay, the total time is the time between PetscLogBegin() and the first several lines of PetscLogView(), before all the reductions.
Thus it does not look like library loading is responsible.

In this log, the main routine seems to be KSPSolve which takes 1.5e3. It looks like some time in this app is just not
being logged. That would take popping the current stage and then pushing it again afterwards I think. We would have
to know more about the code.

That's what bugs me... I removed all custom logging and custom events from the code. The code does not call any external library. Basically, it iterates assembling a matrix, calling KSPSolve, assembling another matrix, calling TAOSolve, and writing results in exodus format.

Is there a command line options that would give me some informations on the log level operations?

Blaise


--
Department of Mathematics and Center for Computation & Technology
Louisiana State University, Baton Rouge, LA 70803, USA
Tel. +1 (225) 578 1612, Fax  +1 (225) 578 4276 http://www.math.lsu.edu/~bourdin







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20130324/b2919be6/attachment.html>