Re: [petsc-users] (percent time in this phase)

> On Oct 5, 2021, at 2:17 PM, Karthikeyan Chockalingam - STFC UKRI > wrote: > > Thank you very much! > KPS Setup does appear in the flame login it in the far left (and you have to > zoom to see it using the browser) > I will look into the debugger option. > > You said “.

Re: [petsc-users] (percent time in this phase)

Just to clarify, I am referring to the below -log_view output, where KPSSetUp is called two times KSPSetUp 2 1.0 1.3278e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 6.0e+00 0 0 0 0 1 0 0 0 0 1 0 By using the following option: mpirun -n 6 ……… -pc_bjacobi_local_blocks 1 •

Re: [petsc-users] (percent time in this phase)

Thank you very much! KPS Setup does appear in the flame login it in the far left (and you have to zoom to see it using the browser) I will look into the debugger option. You said “. -pc_bjacobi_local_blocks 2 indicates you want to 2 blocks per rank” does it imply on mpi -n 2 that KSPSetUP

Re: [petsc-users] (percent time in this phase)

> On Oct 5, 2021, at 12:28 PM, Karthikeyan Chockalingam - STFC UKRI > wrote: > > Thanks Barry. > > Please find the attached screen shoot (flamegraph) and foo.txt which > generated that graph (using https://www.speedscope.app > ). > If you look at the far right

Re: [petsc-users] (percent time in this phase)

The graph was generated using the flag -log_view :foo.txt:ascii_flamegraph From: "Chockalingam, Karthikeyan (STFC,DL,HC)" Date: Tuesday, 5 October 2021 at 17:28 To: Barry Smith Cc: Matthew Knepley , "petsc-users@mcs.anl.gov" Subject: Re: [petsc-users] (percent time in this phase) Thanks

Re: [petsc-users] (percent time in this phase)

PCSetUpOnBlocks() calls KSPSetUp() on each block which calls PCSetUp() on each block. This is why you are getting the confusing logging results. So the " extra" KSPSetUp() is the one on the blocks. > On Oct 5, 2021, at 12:02 PM, Karthikeyan Chockalingam - STFC UKRI > wrote: > > Hi Matt, >

Re: [petsc-users] (percent time in this phase)

Hi Matt, I have a couple of questions; 1. Weather I run on single core or on multiple cores I find that KSPSetUP (ksp/tutorial/ex.45.c) is always called twice. But why? Is setup not done once? 2. I find PCSetUpOnBlocks is calling PCSetup and not the other way around. Can you shed some

Re: [petsc-users] SLEPc: smallest eigenvalues

Nonlinear eigenvalue problems can still be considered a research topic. The NEP package is more or less "finished", cf. https://doi.org/10.1145/3447544 , but your use case may require changes. I would suggest that you write another email to me (not the list) and we can discuss the details.

Re: [petsc-users] SLEPc: smallest eigenvalues

Hi Jose, I have now gotten the quadratic problem working decently using the PEP package with appropriate scaling and preconditioning, so thanks for all the suggestions! For the case where K is a shell matrix, I used a scaling based on an approximation of K, and that seems to be working well. So