-
Notifications
You must be signed in to change notification settings - Fork 140
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Unscalable ASM? #21
Comments
On Mon, Jul 13, 2015 at 1:44 PM, Amneet Bhalla notifications@github.com
Thanks,
What most experimenters take for granted before they begin their |
Yes, this is for MG preconditioning. Basically we are forming level solvers as MG-smoothers. Some processors can be devoid of patches on some levels in SAMRAI, which will trigger an error. |
Matt, can we add in "dummy" subdomains that do not have empty index sets in this case? Or do we need to make sub-communicators that include only MPI ranks with active DOFs? |
On Mon, Jul 13, 2015 at 3:01 PM, Boyce Griffith notifications@github.com
Thanks,
What most experimenters take for granted before they begin their |
I don't think that Amneet has sent in empty IS'es, just noticed that PETSc chokes if there are no subdomains. @amneetb, can you setup code to pass "dummy" empty IS'es for cases in which there are no local DOFs? |
Sure. |
Adding code like:
Gives PETSc error:
|
On Mon, Jul 13, 2015 at 5:41 PM, Amneet Bhalla notifications@github.com
Thanks,
What most experimenters take for granted before they begin their |
Runs but gives a seg fault... Program received signal EXC_BAD_ACCESS, Could not access memory. |
Okay, I will fix it. Thanks,
On Mon, Jul 13, 2015 at 10:53 PM, Amneet Bhalla notifications@github.com
What most experimenters take for granted before they begin their |
I do not see the same error. I am not sure what is different here. Can you take a look at https://bitbucket.org/petsc/petsc/branch/knepley/test-asm-empty-subdomain ? This is my branch which has KSP ex6. It uses both the default and by-hand versions of ASM with an empty subdomain and does not crash. |
@knepley OK; the example helps. So we changed the way the empty IS is created. Before it was
Now following your example it is changed to
This made the code in IBAMR to work. Probably you want to make even the first call to succeed in the ASM code for the case of empty subdomain (?). |
On Wed, Jul 29, 2015 at 1:18 PM, Amneet Bhalla notifications@github.com
Thanks,
What most experimenters take for granted before they begin their |
Ref:
IBAMR/ibtk/src/solvers/impls/PETScLevelSolver.cpp
Line 309 in 68a0c2d
This seems like an unscalable approach. All processors must own atleast one patch on a particular level.
This is not guaranteed in SAMRAI(?).
@knepley Is there a workaround in PETSc?
The text was updated successfully, but these errors were encountered: