Dear Barry,
In my previous email, that is a typo.
In the actual code, I did use:
PreLoadStage("A");
PreLoadStage("B"),
Ok, I see the difference:
I use the PetscOptionsSetValue("-log_summary",""); after the preloading
section. And so when I submit the job, I did not include the command line
option -log_summary again. This seems having confused the code.
And this behavior causes some of the statistics missing as I reported above.
After remove the function call PetscOptionsSetValue("-log_summary","") in the
code and use command line option -log_summary. It worked. And all the
information is complete now.
Thanks a lot for the help,
Yan
PreLoadStage((PETSC_TRUE, "B"), doesn't take a PetscBool argument.
src/snes/examples/tutorials/ex19.c uses these preload macros and does produce a table of the operations done. What is different in your code?
Barry
On Jun 30, 2011, at 5:47 PM, Ryan Yan wrote:
> Dear All,
> I am using a developer version of PETSc code to test a solver code.
>
> The code loads a matrix, and solves the matrix by method
> A first and then method B. In each method of A and B, I setup KSP, solve,
> and then destroy KSP. So in this process, I set up KSP twice and destroy
> the KSP object twice.
>
> I am using the PreLoadxxx to do the profiling but seems like part of the
> profiling information is missing.
>
> PreLoadBegin(PETSC_TRUE, "A"),
> PreLoadStage((PETSC_TRUE, "B"),
> PreLoadEnd()
>
> The missing part is:
>
>
> Event Count Time (sec) Flops --- Global --- --- Stage --- Total
> Max Ratio Max Ratio Max Ratio Mess Avg len Reduct %T %F %M %L %R %T %F %M %L %R Mflop/s
> ------------------------------------------------------------------------------------------------------------------------
>
> --- Event Stage 0: Main Stage
>
>
> --- Event Stage 1: A
>
>
> --- Event Stage 2: B
>
> ------------------------------------------------------------------------------------------------------------------------
>
>
> Is there any way to workaround this?
>
> Many thanks,
>
> Yan
>