<div dir="ltr"><div dir="ltr">On Tue, Oct 22, 2019 at 12:43 PM Thibaut Appel via petsc-users <<a href="mailto:petsc-users@mcs.anl.gov">petsc-users@mcs.anl.gov</a>> wrote:<br></div><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<div bgcolor="#FFFFFF">
<p>Hi Hong,</p>
<p>Thank you for having a look, I copied/pasted your code snippet
into ex28.c and the error indeed appears if you change that
col[0]. That's because you did not allow a new non-zero location
in the matrix with the option MAT_NEW_NONZERO_LOCATION_ERR.<br>
</p>
<p>I spent the day debugging the code and already checked my calls
to MatSetValues:</p>
<p>For all MatSetValues calls corresponding to the row/col location
in the error messages in the subsequent assembly, the numerical
value associated with that row/col was exactly (0.0,0.0) (complex
arithmetic) so it shouldn't be inserted w.r.t. the option
MAT_IGNORE_ZERO_ENTRIES. It seems MatSetValues still did it
anyway.<br>
</p>
<p></p></div></blockquote><div>Okay, lets solve this problem first. You say that</div><div><br></div><div> - You called MatSetOption(A, MAT_IGNORE_ZERO_ENTRIES, PETSC_TRUE)<br></div><div> - You called MatSetValues(A, ,,,, ADD_VALUES, ..., val) and val had a complex 0 in it</div><div> - PETSc tried to insert the complex 0</div><div><br></div><div>This should be really easy to test in a tiny example. Do you mind making it? If its broken, I will fix it.</div><div><br></div><div> Thanks,</div><div><br></div><div> Matt</div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div bgcolor="#FFFFFF"><p>I was able to solve the problem by adding
MatSetOption(L,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE) after my
first assembly. However I don't know why it fixed it as the manual
seems to say this is just for efficiency purposes</p>
<p>It "should be specified after the first matrix has been fully
assembled. This option ensures that certain data structures and
communication information will be reused (instead of regenerated
during successive steps, thereby increasing efficiency"</p>
<p>So I'm still puzzled by why I got that error in the first place.
Unless "regenerated" implies resetting some attributes of the
preallocated non-zero structure / first assembly?</p>
<p><br>
</p>
<p>Thibaut</p>
<p><br>
</p>
<div>On 22/10/2019 17:06, Zhang, Hong wrote:<br>
</div>
<blockquote type="cite">
<div dir="ltr">
<div dir="ltr">Thibaut:<br>
</div>
<div>Check your code on MatSetValues(), likely you set a value
"to a new nonzero at global row/column (200, 160) into matrix"
L.</div>
<div>I tested petsc/src/mat/examples/tests/ex28.c by adding </div>
<div>@@ -95,6 +95,26 @@ int main(int argc,char **args)<br>
/* Compute numeric factors using same F, then solve */<br>
for (k = 0; k < num_numfac; k++) {<br>
/* Get numeric factor of A[k] */<br>
+ if (k==0) {<br>
+ ierr = MatZeroEntries(A[0]);CHKERRQ(ierr);<br>
+ for (i=rstart; i<rend; i++) {<br>
+ col[0] = i-1; col[1] = i; col[2] = i+1;<br>
+ if (i == 0) {<br>
+ ierr =
MatSetValues(A[k],1,&i,2,col+1,value+1,INSERT_VALUES);CHKERRQ(ierr);<br>
+ } else if (i == N-1) {<br>
+ ierr =
MatSetValues(A[k],1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);<br>
+ } else {<br>
+ ierr =
MatSetValues(A[k],1,&i,3,col,value,INSERT_VALUES);CHKERRQ(ierr);<br>
+ }<br>
+ }<br>
+ if (!rank) {<br>
+ i = N - 1; col[0] = N - 1;<br>
+ ierr =
MatSetValues(A[k],1,&i,1,col,value,INSERT_VALUES);CHKERRQ(ierr);<br>
+ }<br>
+ ierr =
MatAssemblyBegin(A[k],MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);<br>
+ ierr =
MatAssemblyEnd(A[k],MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);<br>
+ }<br>
+<br>
</div>
<div><br>
</div>
<div>It works in both sequential and parallel. If I set col[0] =
0, then I get the same error as yours.</div>
<div>Hong</div>
<div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<div bgcolor="#FFFFFF">
<p>Dear PETSc developers,</p>
<p>I'm extending a validated matrix preallocation/assembly
part of my code to solve multiple linear systems with
MUMPS at each iteration of a main loop, following the
example src/mat/examples/tests/ex28.c that Hong Zhang
added a few weeks ago. The difference is that I'm using
just 1 matrix to solve different systems.<br>
</p>
<p>I'm trying to investigate a nasty bug arising when I
try to assemble "for a second time" that MPIAIJ matrix.
The issue arises only in parallel, serial works fine.</p>
<p>Creating 1 MPIAIJ matrix, preallocating it "perfectly"
with the case where I have the fewest zero entries in
the non-zero structure, before getting its symbolic
factorization.</p>
<p>Further in the main loop, I'm solely changing its
entries <b>retaining the non-zero structure</b>.<br>
</p>
<p>Here is the simplified Fortran code I'm using:<br>
</p>
<p><tt>! Fill (M,N) case to ensure all non-zero entries
are preallocated</tt><tt><br>
</tt><tt>CALL set_equations(M,N)</tt><tt><br>
</tt><tt><br>
</tt><tt>CALL alloc_matrix(L)</tt><tt><br>
</tt><tt> ! --> Call
MatSeqAIJSetPreallocation/MatMPIAIJSetPreallocation</tt><tt><br>
</tt><tt> ! --> Sets MAT_IGNORE_ZERO_ENTRIES,
MAT_NEW_NONZERO_ALLOCATION_ERR,
MAT_NO_OFF_PROC_ENTRIES to true</tt><tt><br>
</tt><tt><br>
</tt><tt>CALL assemble_matrix(L)</tt><tt><br>
</tt><tt> ! --> Calls MatSetValues with ADD_VALUES</tt><tt><br>
</tt><tt> ! --> Call MatAssemblyBegin/MatAssemblyEnd</tt><tt><br>
</tt><tt><br>
</tt><tt>! Tell PETSc that new non-zero insertions in
matrix are forbidden</tt><tt><br>
</tt><tt>CALL
MatSetOption(L,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE,ierr)</tt><tt><br>
</tt><tt><br>
</tt><tt>CALL set_mumps_parameters()</tt><tt><br>
</tt><tt><br>
</tt><tt>! Get symbolic LU factorization using MUMPS</tt><tt><br>
</tt><tt>CALL
MatGetFactor(L,MATSOLVERMUMPS,MAT_FACTOR_LU,F,ierr)</tt><tt><br>
</tt><tt>CALL
MatGetOrdering(L,MATORDERINGNATURAL,rperm,cperm,ierr)</tt><tt><br>
</tt><tt>CALL
MatLUFactorSymbolic(F,L,rperm,cperm,info,ierr)</tt><tt><br>
</tt><tt><br>
</tt><tt>CALL initialize_right_hand_sides()</tt><tt><br>
</tt><tt><br>
</tt><tt>! Zero matrix entries</tt><tt><br>
</tt><tt>CALL MatZeroEntries(L,ierr)</tt><tt><br>
</tt><tt><br>
</tt><tt>! Main loop</tt><tt><br>
</tt><tt>DO itr=1, maxitr</tt><tt><br>
</tt><tt><br>
</tt><tt> DO m = 1, M</tt><tt><br>
</tt><tt> DO n = 1, N</tt><tt><br>
</tt><tt><br>
</tt><tt> CALL set_equations(m,n)</tt><tt><br>
</tt><tt> CALL assemble_matrix(L) ! ERROR HERE when
m=1, n=1, CRASH IN MatSetValues call</tt><tt><br>
</tt><tt><br>
</tt><tt> ! Solving the linear system associated with
(m,n)</tt><tt><br>
</tt><tt> CALL MatLUFactorNumeric(F,L,info,ierr)</tt><tt><br>
</tt><tt> CALL
MatSolve(F,v_rhs(m,n),v_sol(m,n),ierr) </tt><tt><br>
</tt><tt><br>
</tt><tt> ! Process v_rhs's from v_sol's for next
iteration</tt><tt><br>
</tt><tt><br>
</tt><tt> CALL MatZeroEntries(L,ierr)</tt><tt><br>
</tt><tt><br>
</tt><tt> END DO</tt><tt><br>
</tt><tt> END DO</tt><tt><br>
</tt><tt><br>
</tt><tt>END DO</tt></p>
<p><tt><br>
</tt></p>
<p>Testing on a small case, the error I get is</p>
<tt>[1]PETSC ERROR: --------------------- Error Message
--------------------------------------------------------------</tt><tt><br>
</tt><tt>[1]PETSC ERROR: Argument out of range</tt><tt><br>
</tt><tt>[1]PETSC ERROR: Inserting a new nonzero at global
row/column (200, 160) into matrix</tt><tt><br>
</tt><tt>[1]PETSC ERROR: See <a href="https://www.mcs.anl.gov/petsc/documentation/faq.html" target="_blank">
https://www.mcs.anl.gov/petsc/documentation/faq.html</a>
for trouble shooting.</tt><tt><br>
</tt><tt>[1]PETSC ERROR: Petsc Release Version 3.12.0,
unknown </tt><tt><br>
</tt><tt>[1]PETSC ERROR: Configure options
--PETSC_ARCH=cplx_gcc_debug --with-scalar-type=complex
--with-precision=double --with-debugging=1
--with-valgrind=1 --with-debugger=gdb
--with-fortran-kernels=1 --download-mpich
--download-fblaslapack --download-scalapack
--download-metis --download-parmetis --download-ptscotch
--download-mumps --download-slepc --COPTFLAGS="-O0 -g"
--CXXOPTFLAGS="-O0 -g" --FOPTFLAGS="-O0 -g -fbacktrace"</tt><tt><br>
</tt><tt>[1]PETSC ERROR: #1 MatSetValues_MPIAIJ() line 634
in
/home/thibaut/Packages/petsc/src/mat/impls/aij/mpi/mpiaij.c</tt><tt><br>
</tt><tt>[1]PETSC ERROR: #2 MatSetValues() line 1375 in
/home/thibaut/Packages/petsc/src/mat/interface/matrix.c</tt><tt><br>
</tt><tt>[1]PETSC ERROR: #3 User provided function() line
0 in User file</tt><tt><br>
</tt><tt>application called MPI_Abort(MPI_COMM_SELF, 63) -
process 0</tt><tt><br>
</tt>
<p> <br>
</p>
<p>which I don't understand. That element was not in the
non-zero structure and wasn't preallocated. I printed
the value to be inserted at this location (200,160) and
it is exactly (0.0000000000000000,0.0000000000000000) so
this entry should not be inserted due to
MAT_IGNORE_ZERO_ENTRIES, however it seems it is. I'm
using ADD_VALUES in MatSetValues but it is the only call
where (200,160) is inserted.<br>
</p>
<p><br>
</p>
<p> - I zero the matrix entries with MatZeroEntries
which retains the non-zero structure (checked when I
print the matrix) but tried to comment the corresponding
calls.<br>
</p>
<p> - I tried to set MAT_NEW_NONZERO_LOCATION_ERR AND
MAT_NEW_NONZERO_ALLOCATION_ERR to PETSC_FALSE without
effect.</p>
<p><br>
</p>
<p>Perhaps there's something fundamentally wrong in my
approach, in any case would you have any suggestions to
identify the exact problem?
<br>
</p>
<p>Using PETSc 3.12.0. Thanks for your support,</p>
<p><br>
</p>
<p>Thibaut<br>
</p>
</div>
</blockquote>
</div>
</div>
</blockquote>
</div>
</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr" class="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div>What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.<br>-- Norbert Wiener</div><div><br></div><div><a href="http://www.cse.buffalo.edu/~knepley/" target="_blank">https://www.cse.buffalo.edu/~knepley/</a><br></div></div></div></div></div></div></div></div>