Awesome! MatZeroRows is very useful and simplified the code logic. 

On Wed, Nov 29, 2023 at 5:14 PM Matthew Knepley <knepley@gmail.com> wrote:
On Wed, Nov 29, 2023 at 7:27 PM Amneet Bhalla <mail2amneet@gmail.com> wrote:
Ah, I also tried without step 2 (i.e., manually doing MPI_allgatherv for Dirichlet rows), and that also works. So it seems that each processor needs to send in their own Dirichlet rows, and not a union of them. Is that correct? 

Yes, that is correct.

  Thanks,

    Matt
 
On Wed, Nov 29, 2023 at 3:48 PM Amneet Bhalla <mail2amneet@gmail.com> wrote:
Thanks Barry! I tried that and it seems to be working. This is what I did. It would be great if you could take a look at it and let me know if this is what you had in mind.

1. Collected Dirichlet rows locally

On Wed, Nov 29, 2023 at 11:32 AM Barry Smith <bsmith@petsc.dev> wrote:


On Nov 29, 2023, at 2:11 PM, Matthew Knepley <knepley@gmail.com> wrote:

On Wed, Nov 29, 2023 at 1:55 PM Amneet Bhalla <mail2amneet@gmail.com> wrote:
So the code logic is after the matrix is assembled, I iterate over all distributed patches in the domain to see which of the patch is abutting a Dirichlet boundary. Depending upon which patch abuts a physical and Dirichlet boundary, a processor will call this routine. However, that same processor is “owning” that DoF, which would be on its diagonal. 

I think Barry already mentioned this is not going to work unless I use the flag to not communicate explicitly. However, that flag is not working as it should over here for some reason.

Oh, I do not think that is right.

Barry, when I look at the code, MPIU_Allreduce is always going to be called to fix up the nonzero_state. Am I wrong about that?

  No, you are correct. I missed that in my earlier look. Setting those flags reduce the number of MPI reductions but does not eliminate them completely.

  MatZeroRows is collective (as its manual page indicates) so you have to do the second thing I suggested. Inside your for loop construct an array containing all the local
rows being zeroed and then make a single call by all MPI processes to MatZeroRows().  Note this is a small change of just a handful of lines of code.

   Barry


  Thanks,

    Matt
 
I can always change the matrix coefficients for Dirichlet rows during MatSetValues. However, that would lengthen my code and I was trying to avoid that. 

On Wed, Nov 29, 2023 at 10:02 AM Matthew Knepley <knepley@gmail.com> wrote:
On Wed, Nov 29, 2023 at 12:30 PM Amneet Bhalla <mail2amneet@gmail.com> wrote:
Ok, I added both, but it still hangs. Here, is bt from all three tasks:

It looks like two processes are calling AllReduce, but one is not. Are all procs not calling MatZeroRows?

  Thanks,

     Matt
 
Task 1:

amneetb@APSB-MacBook-Pro-16:~$ lldb  -p 44691 
(lldb) process attach --pid 44691
Process 44691 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
    frame #0: 0x000000018a2d750c libsystem_kernel.dylib`__semwait_signal + 8
libsystem_kernel.dylib`:
->  0x18a2d750c <+8>:  b.lo   0x18a2d752c               ; <+40>
    0x18a2d7510 <+12>: pacibsp 
    0x18a2d7514 <+16>: stp    x29, x30, [sp, #-0x10]!
    0x18a2d7518 <+20>: mov    x29, sp
Target 0: (fo_acoustic_streaming_solver_2d) stopped.
Executable module set to "/Users/amneetb/Softwares/IBAMR-Git/objs-dbg/tests/IBTK/fo_acoustic_streaming_solver_2d".
Architecture set to: arm64-apple-macosx-.
(lldb) cont
Process 44691 resuming
Process 44691 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
    frame #0: 0x000000010ba40b60 libpmpi.12.dylib`MPIDI_POSIX_mpi_release_gather_release + 752
libpmpi.12.dylib`MPIDI_POSIX_mpi_release_gather_release:
->  0x10ba40b60 <+752>: add    w8, w8, #0x1
    0x10ba40b64 <+756>: ldr    w9, [x22]
    0x10ba40b68 <+760>: cmp    w8, w9
    0x10ba40b6c <+764>: b.lt   0x10ba40b4c               ; <+732>
Target 0: (fo_acoustic_streaming_solver_2d) stopped.
(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
  * frame #0: 0x000000010ba40b60 libpmpi.12.dylib`MPIDI_POSIX_mpi_release_gather_release + 752
    frame #1: 0x000000010ba48528 libpmpi.12.dylib`MPIDI_POSIX_mpi_allreduce_release_gather + 1088
    frame #2: 0x000000010ba47964 libpmpi.12.dylib`MPIDI_Allreduce_intra_composition_gamma + 368
    frame #3: 0x000000010ba35e78 libpmpi.12.dylib`MPIR_Allreduce + 1588
    frame #4: 0x0000000103f587dc libmpi.12.dylib`MPI_Allreduce + 2280
    frame #5: 0x0000000106d67650 libpetsc.3.17.dylib`MatZeroRows_MPIAIJ(A=0x0000000105846470, N=1, rows=0x000000016dbfa9f4, diag=1, x=0x0000000000000000, b=0x0000000000000000) at mpiaij.c:827:3
    frame #6: 0x0000000106aadfac libpetsc.3.17.dylib`MatZeroRows(mat=0x0000000105846470, numRows=1, rows=0x000000016dbfa9f4, diag=1, x=0x0000000000000000, b=0x0000000000000000) at matrix.c:5935:3
    frame #7: 0x00000001023952d0 fo_acoustic_streaming_solver_2d`IBAMR::AcousticStreamingPETScMatUtilities::constructPatchLevelFOAcousticStreamingOp(mat=0x000000016dc04168, omega=1, sound_speed=1, rho_idx=3, mu_idx=2, lambda_idx=4, u_bc_coefs=0x000000016dc04398, data_time=NaN, num_dofs_per_proc=size=3, u_dof_index_idx=27, p_dof_index_idx=28, patch_level=Pointer<SAMRAI::hier::PatchLevel<2> > @ 0x000000016dbfcec0, mu_interp_type=VC_HARMONIC_INTERP) at AcousticStreamingPETScMatUtilities.cpp:799:36
    frame #8: 0x00000001023acb8c fo_acoustic_streaming_solver_2d`IBAMR::FOAcousticStreamingPETScLevelSolver::initializeSolverStateSpecialized(this=0x000000016dc04018, x=0x000000016dc05778, (null)=0x000000016dc05680) at FOAcousticStreamingPETScLevelSolver.cpp:149:5
    frame #9: 0x000000010254a2dc fo_acoustic_streaming_solver_2d`IBTK::PETScLevelSolver::initializeSolverState(this=0x000000016dc04018, x=0x000000016dc05778, b=0x000000016dc05680) at PETScLevelSolver.cpp:340:5
    frame #10: 0x0000000102202e5c fo_acoustic_streaming_solver_2d`main(argc=11, argv=0x000000016dc07450) at fo_acoustic_streaming_solver.cpp:400:22
    frame #11: 0x0000000189fbbf28 dyld`start + 2236
(lldb) 


Task 2:

amneetb@APSB-MacBook-Pro-16:~$ lldb  -p 44692 
(lldb) process attach --pid 44692
Process 44692 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
    frame #0: 0x000000018a2d750c libsystem_kernel.dylib`__semwait_signal + 8
libsystem_kernel.dylib`:
->  0x18a2d750c <+8>:  b.lo   0x18a2d752c               ; <+40>
    0x18a2d7510 <+12>: pacibsp 
    0x18a2d7514 <+16>: stp    x29, x30, [sp, #-0x10]!
    0x18a2d7518 <+20>: mov    x29, sp
Target 0: (fo_acoustic_streaming_solver_2d) stopped.
Executable module set to "/Users/amneetb/Softwares/IBAMR-Git/objs-dbg/tests/IBTK/fo_acoustic_streaming_solver_2d".
Architecture set to: arm64-apple-macosx-.
(lldb) cont
Process 44692 resuming
Process 44692 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
    frame #0: 0x000000010e5a022c libpmpi.12.dylib`MPIDI_POSIX_mpi_barrier_release_gather + 516
libpmpi.12.dylib`MPIDI_POSIX_mpi_barrier_release_gather:
->  0x10e5a022c <+516>: ldr    x10, [x19, #0x4e8]
    0x10e5a0230 <+520>: cmp    x9, x10
    0x10e5a0234 <+524>: b.hs   0x10e5a0254               ; <+556>
    0x10e5a0238 <+528>: add    w8, w8, #0x1
Target 0: (fo_acoustic_streaming_solver_2d) stopped.
(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
  * frame #0: 0x000000010e5a022c libpmpi.12.dylib`MPIDI_POSIX_mpi_barrier_release_gather + 516
    frame #1: 0x000000010e59fd14 libpmpi.12.dylib`MPIDI_SHM_mpi_barrier + 224
    frame #2: 0x000000010e59fb60 libpmpi.12.dylib`MPIDI_Barrier_intra_composition_alpha + 44
    frame #3: 0x000000010e585490 libpmpi.12.dylib`MPIR_Barrier + 900
    frame #4: 0x0000000106ac5030 libmpi.12.dylib`MPI_Barrier + 684
    frame #5: 0x0000000108e62638 libpetsc.3.17.dylib`PetscCommDuplicate(comm_in=1140850688, comm_out=0x00000001408ae4b0, first_tag=0x00000001408ae4e4) at tagm.c:235:5
    frame #6: 0x0000000108e6a910 libpetsc.3.17.dylib`PetscHeaderCreate_Private(h=0x00000001408ae470, classid=1211228, class_name="KSP", descr="Krylov Method", mansec="KSP", comm=1140850688, destroy=(libpetsc.3.17.dylib`KSPDestroy at itfunc.c:1418), view=(libpetsc.3.17.dylib`KSPView at itcreate.c:113)) at inherit.c:62:3
    frame #7: 0x000000010aa28010 libpetsc.3.17.dylib`KSPCreate(comm=1140850688, inksp=0x000000016b0a4160) at itcreate.c:679:3
    frame #8: 0x00000001050aa2f4 fo_acoustic_streaming_solver_2d`IBTK::PETScLevelSolver::initializeSolverState(this=0x000000016b0a4018, x=0x000000016b0a5778, b=0x000000016b0a5680) at PETScLevelSolver.cpp:344:12
    frame #9: 0x0000000104d62e5c fo_acoustic_streaming_solver_2d`main(argc=11, argv=0x000000016b0a7450) at fo_acoustic_streaming_solver.cpp:400:22
    frame #10: 0x0000000189fbbf28 dyld`start + 2236
(lldb) 


Task 3:

amneetb@APSB-MacBook-Pro-16:~$ lldb  -p 44693 
(lldb) process attach --pid 44693
Process 44693 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
    frame #0: 0x000000018a2d750c libsystem_kernel.dylib`__semwait_signal + 8
libsystem_kernel.dylib`:
->  0x18a2d750c <+8>:  b.lo   0x18a2d752c               ; <+40>
    0x18a2d7510 <+12>: pacibsp 
    0x18a2d7514 <+16>: stp    x29, x30, [sp, #-0x10]!
    0x18a2d7518 <+20>: mov    x29, sp
Target 0: (fo_acoustic_streaming_solver_2d) stopped.
Executable module set to "/Users/amneetb/Softwares/IBAMR-Git/objs-dbg/tests/IBTK/fo_acoustic_streaming_solver_2d".
Architecture set to: arm64-apple-macosx-.
(lldb) cont
Process 44693 resuming
Process 44693 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
    frame #0: 0x000000010e59c68c libpmpi.12.dylib`MPIDI_POSIX_mpi_release_gather_gather + 952
libpmpi.12.dylib`MPIDI_POSIX_mpi_release_gather_gather:
->  0x10e59c68c <+952>: ldr    w9, [x21]
    0x10e59c690 <+956>: cmp    w8, w9
    0x10e59c694 <+960>: b.lt   0x10e59c670               ; <+924>
    0x10e59c698 <+964>: bl     0x10e59ce64               ; MPID_Progress_test
Target 0: (fo_acoustic_streaming_solver_2d) stopped.
(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
  * frame #0: 0x000000010e59c68c libpmpi.12.dylib`MPIDI_POSIX_mpi_release_gather_gather + 952
    frame #1: 0x000000010e5a44bc libpmpi.12.dylib`MPIDI_POSIX_mpi_allreduce_release_gather + 980
    frame #2: 0x000000010e5a3964 libpmpi.12.dylib`MPIDI_Allreduce_intra_composition_gamma + 368
    frame #3: 0x000000010e591e78 libpmpi.12.dylib`MPIR_Allreduce + 1588
    frame #4: 0x0000000106ab47dc libmpi.12.dylib`MPI_Allreduce + 2280
    frame #5: 0x00000001098c3650 libpetsc.3.17.dylib`MatZeroRows_MPIAIJ(A=0x0000000136862270, N=1, rows=0x000000016b09e9f4, diag=1, x=0x0000000000000000, b=0x0000000000000000) at mpiaij.c:827:3
    frame #6: 0x0000000109609fac libpetsc.3.17.dylib`MatZeroRows(mat=0x0000000136862270, numRows=1, rows=0x000000016b09e9f4, diag=1, x=0x0000000000000000, b=0x0000000000000000) at matrix.c:5935:3
    frame #7: 0x0000000104ef12d0 fo_acoustic_streaming_solver_2d`IBAMR::AcousticStreamingPETScMatUtilities::constructPatchLevelFOAcousticStreamingOp(mat=0x000000016b0a8168, omega=1, sound_speed=1, rho_idx=3, mu_idx=2, lambda_idx=4, u_bc_coefs=0x000000016b0a8398, data_time=NaN, num_dofs_per_proc=size=3, u_dof_index_idx=27, p_dof_index_idx=28, patch_level=Pointer<SAMRAI::hier::PatchLevel<2> > @ 0x000000016b0a0ec0, mu_interp_type=VC_HARMONIC_INTERP) at AcousticStreamingPETScMatUtilities.cpp:799:36
    frame #8: 0x0000000104f08b8c fo_acoustic_streaming_solver_2d`IBAMR::FOAcousticStreamingPETScLevelSolver::initializeSolverStateSpecialized(this=0x000000016b0a8018, x=0x000000016b0a9778, (null)=0x000000016b0a9680) at FOAcousticStreamingPETScLevelSolver.cpp:149:5
    frame #9: 0x00000001050a62dc fo_acoustic_streaming_solver_2d`IBTK::PETScLevelSolver::initializeSolverState(this=0x000000016b0a8018, x=0x000000016b0a9778, b=0x000000016b0a9680) at PETScLevelSolver.cpp:340:5
    frame #10: 0x0000000104d5ee5c fo_acoustic_streaming_solver_2d`main(argc=11, argv=0x000000016b0ab450) at fo_acoustic_streaming_solver.cpp:400:22
    frame #11: 0x0000000189fbbf28 dyld`start + 2236
(lldb) 


On Wed, Nov 29, 2023 at 7:22 AM Barry Smith <bsmith@petsc.dev> wrote:


On Nov 29, 2023, at 1:16 AM, Amneet Bhalla <mail2amneet@gmail.com> wrote:

BTW, I think you meant using MatSetOption(mat, MAT_NO_OFF_PROC_ZERO_ROWS, PETSC_TRUE)

Yes

 instead ofMatSetOption(mat, MAT_NO_OFF_PROC_ENTRIES, PETSC_TRUE) ??

  Please try setting both flags.

 However, that also did not help to overcome the MPI Barrier issue. 

  If there is still a problem please trap all the MPI processes when they hang in the debugger and send the output from using bt on all of them. This way
we can see the different places the different MPI processes are stuck at.



On Tue, Nov 28, 2023 at 9:57 PM Amneet Bhalla <mail2amneet@gmail.com> wrote:
I added that option but the code still gets stuck at the same call MatZeroRows with 3 processors. 

On Tue, Nov 28, 2023 at 7:23 PM Amneet Bhalla <mail2amneet@gmail.com> wrote:


On Tue, Nov 28, 2023 at 6:42 PM Barry Smith <bsmith@petsc.dev> wrote:

  for (int comp = 0; comp < 2; ++comp)
                {
                .......
                    for (Box<NDIM>::Iterator bc(bc_coef_box); bc; bc++)
                    {
                       ......
                        if (IBTK::abs_equal_eps(b, 0.0))
                        {
                            const double diag_value = a;
                            ierr = MatZeroRows(mat, 1, &u_dof_index, diag_value, NULL, NULL);
                            IBTK_CHKERRQ(ierr);
                        }
                    }
                }

In general, this code will not work because each process calls MatZeroRows a different number of times, so it cannot match up with all the processes.

If u_dof_index is always local to the current process, you can call MatSetOption(mat, MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE) above the for loop and 
the MatZeroRows will not synchronize across the MPI processes (since it does not need to and you told it that).

Yes, u_dof_index is going to be local and I put a check on it a few lines before calling MatZeroRows.

Can MatSetOption() be called after the matrix has been assembled?


If the u_dof_index will not always be local, then you need, on each process, to list all the u_dof_index for each process in an array and then call MatZeroRows()
once after the loop so it can exchange the needed information with the other MPI processes to get the row indices to the right place.

Barry




On Nov 28, 2023, at 6:44 PM, Amneet Bhalla <mail2amneet@gmail.com> wrote:


Hi Folks, 

I am using MatZeroRows() to set Dirichlet boundary conditions. This works fine for the serial run and the solver produces correct results (verified through analytical solution). However, when I run the case in parallel, the simulation gets stuck at MatZeroRows(). My understanding is that this function needs to be called after the MatAssemblyBegin{End}() has been called, and should be called by all processors. Here is that bit of the code which calls MatZeroRows() after the matrix has been assembled


I ran the parallel code (on 3 processors) in the debugger (-start_in_debugger). Below is the call stack from the processor that gets stuck

amneetb@APSB-MBP-16:~$ lldb  -p 4307 
(lldb) process attach --pid 4307
Process 4307 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
    frame #0: 0x000000018a2d750c libsystem_kernel.dylib`__semwait_signal + 8
libsystem_kernel.dylib`:
->  0x18a2d750c <+8>:  b.lo   0x18a2d752c               ; <+40>
    0x18a2d7510 <+12>: pacibsp 
    0x18a2d7514 <+16>: stp    x29, x30, [sp, #-0x10]!
    0x18a2d7518 <+20>: mov    x29, sp
Target 0: (fo_acoustic_streaming_solver_2d) stopped.
Executable module set to "/Users/amneetb/Softwares/IBAMR-Git/objs-dbg/tests/IBTK/fo_acoustic_streaming_solver_2d".
Architecture set to: arm64-apple-macosx-.
(lldb) cont
Process 4307 resuming
Process 4307 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
    frame #0: 0x0000000109d281b8 libpmpi.12.dylib`MPIDI_POSIX_mpi_barrier_release_gather + 400
libpmpi.12.dylib`MPIDI_POSIX_mpi_barrier_release_gather:
->  0x109d281b8 <+400>: ldr    w9, [x24]
    0x109d281bc <+404>: cmp    w8, w9
    0x109d281c0 <+408>: b.lt   0x109d281a0               ; <+376>
    0x109d281c4 <+412>: bl     0x109d28e64               ; MPID_Progress_test
Target 0: (fo_acoustic_streaming_solver_2d) stopped.
(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGSTOP
  * frame #0: 0x0000000109d281b8 libpmpi.12.dylib`MPIDI_POSIX_mpi_barrier_release_gather + 400
    frame #1: 0x0000000109d27d14 libpmpi.12.dylib`MPIDI_SHM_mpi_barrier + 224