<html><head></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; ">Dmitry,<div><br></div><div>thanks! For the moment, I can use the sorted IS and I just need to pass through user context</div><div>some AO or ISLocalToGlobalMapping.</div><div><br></div><div>Hope you will get the unsorted IS case work soon!</div><div><br></div><div>Thanks!</div><div><br></div><div><br><div><div>On May 22, 2012, at 8:04 PM, Dmitry Karpeev wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite">Hui,<div>There appears to be a bug in MatGetSubMatrices() which inserts submatrix column indices in the same order</div><div>as they appeared in the original matrix. If a column IS has an inversion -- IS[2] = 8 and IS[6] = 4, </div>
<div>as in your case -- then the submatrix column index 2 will appear after the submatrix column index 4, since that's </div><div>how they were ordered in the original matrix (4 < 8, and the column indices were properly sorted). This violates </div>
<div>the ordering of column indices, which prevents binary search from finding column index 2. That results in an </div><div>apparent new nonzero insertion, which is in turn caught, since by default no new nonzeros are allowed.</div>
<div><br></div><div>I'm looking into this, and I'm not sure that there is an easy workaround for you at this time: if you sort the IS indices, </div><div>it will violate the implicit local-to-global submatrix mapping you are using in your ModifySubMatrices().</div>
<div><br></div><div>Thanks.</div><div>Dmitry.</div><div><br><div class="gmail_quote">On Mon, May 21, 2012 at 2:33 PM, Hui Zhang <span dir="ltr"><<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word">Hi, <div><br></div><div>It might be unclear. The error here is that the submat[1] does have a (1,2) entry,</div>
<div>but MatSetValues at (1,2) gives an error message. GASM and ASM give the same error.</div><div><br></div><div>Thanks!</div><div><br><div><div class="im"><div>On May 21, 2012, at 9:11 PM, Hui Zhang wrote:</div><br></div>
<blockquote type="cite"><div class="im"><div style="word-wrap:break-word"><div>Sorry, I uses PCASMSetSortIndices for GASM. Now, I correct it and the output becomes the</div><div>same as PCASM. See attached.</div><div><br>
</div><div></div></div></div><span><GASM_test.zip></span><div style="word-wrap:break-word"><div></div><div><br></div><br><div><div class="im"><div>On May 21, 2012, at 8:59 PM, Hui Zhang wrote:</div><br></div><blockquote type="cite">
<div><div class="h5"><div style="word-wrap:break-word"><div>Dmitry,</div><div><br></div><div>finally, I am able to make a simple program and reproduce the error. Since I had moved to</div><div>petsc-dev (before I used petsc-3.2), the error is a little different.</div>
<div><br></div><div>Download the program attached and run the run.sh. You will see the output from PCASM.</div><div>There is a submatrix (submat[1] on rank 0) which has a row like</div><div><br></div><div> row 1: (0, -1) (1, 0 + 0.4 i) (4, -1) (2, -1) </div>
<div><br></div><div>Note that the column 2 is placed in the last and the column numbers are not in the normal order.</div><div>Following that, we can also see the error message like</div><div><br></div><div> [0]PETSC ERROR: New nonzero at (1,2) caused a malloc!</div>
<div><br></div><div>You can also try with GASM, use</div><div><br></div><div> mpirun -np 2 ./gasm_test -dd_type gasm -n 4 -px 1 -py 2 -sx 2 -sy 1 </div><div><br></div><div>which partitions the domain to 1 * 2 and mapped to the two processors, and for each processor</div>
<div>we further partition its local domain to 2 * 1 subdomains. </div><div><br></div><div>Then, you will see that the row 1 of submat[1] on rank 1 becomes</div><div><br></div><div> row 1: (0, -1) (1, 0 + 0.4 i) (3, -1) (6, -1) </div>
<div><br></div><div>which is totally wrong because from the IS for this overlapping subdomain and the big matrix A,</div><div>row 1 should have non-zeros like the output from PCASM.</div><div><br></div><div>I guess the problem is due to that we set PCASMSetSortIndices to FALSE and something goes wrong</div>
<div>in asm.c and gasm.c.</div><div><br></div><div>Thanks!</div><div>Hui</div><div><br></div><div><br></div><div></div></div></div></div><span><GASM_test.zip></span><div><div class="h5"><div style="word-wrap:break-word">
<div></div><div><br></div><div><br></div><div><br></div><div><br></div><div><br><blockquote type="cite"><div class="gmail_quote">On Sun, May 20, 2012 at 5:06 AM, Hui Zhang <span dir="ltr"><<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word"><div>I have the following MatView output: </div><div><br></div><div>row 0: (0, -0.0351719 - 0.638814 i) (1, -0.517586 - 0.617823 i) (4, -0.517586 - 0.617823 i) (1, -1.03517) (5, -0.508793) </div>
<div><br></div><div>Note that there are two values for the column 1, is this normal?</div><div><br></div><div>The above matrix is obtained from PCASMGetSubKSP(pc,subksp) and KSPGetOperators(subksp[0], ...).</div></div></blockquote>
<div><br></div><div>This is odd. Can you explain a bit more what leads up to this, so we can try and reproduce the problem?</div><div>Thanks.</div><div>Dmitry.</div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word"><div><br></div><div>Thanks!</div><div><div><div><br></div><div><br></div><br><div><div>On May 11, 2012, at 7:17 PM, Dmitry Karpeev wrote:</div><br><blockquote type="cite">You can call PCSetUp(pc) on either ASM or GASM, and that will destroy and recreate the matrices (including calling <div>
your modification subroutine), but not the subdomains or the subdomain solvers. </div><div>If you just want to modify the submatrices, you can call PC(G)ASMGetSubmatrices() and modify the matrices it returns</div>
<div>(in the same order as the subdomains were set). That's a bit of a hack, since you will essentially be modifying the PC's internal data structures. As long as you are careful, you should be okay, since you already effectively have the same type of access to the submatrices through the Modify callback.</div>
<div><br></div><div>Dmitry.</div><div><div><br><div class="gmail_quote">On Fri, May 11, 2012 at 11:52 AM, Hui Zhang <span dir="ltr"><<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word">I just have a question about reuse of PCASM or PCGASM.<div>Suppose I have seted up the PCASM and related KSP and I solved one time.</div>
<div>Next for the same linear system (matrix and RHS), I just want PCASM modify the submatrices (PCSetModifySubmatrices) in a different way, using the same routine for modifying but with</div><div>different user context for the modifying routine. </div>
<div><br></div><div>What can I do for this task? Currently, I destroy the KSP and re-construct it. I guess</div><div>even for PCASM I can re-use it because the partition of subdomains remain the same.</div><div><br></div>
<div>Thanks!</div><div><div><div><br><div><br><div><div>On May 10, 2012, at 6:37 PM, Dmitry Karpeev wrote:</div><br><blockquote type="cite">Hui,<div>There've been several changes to PCGASM ahead of the new release.</div>
<div>Let me go back and see if it affected the convergence problem.</div><div>Dmitry.<br><br><div class="gmail_quote">On Thu, May 10, 2012 at 4:16 AM, Hui Zhang <span dir="ltr"><<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word">Hi Dmitry,<div><br></div><div>is there any news about PCGASM? </div><div><br></div><div>
thanks,</div><div>Hui</div><div><div><div><br><div><div>On Feb 20, 2012, at 6:38 PM, Dmitry Karpeev wrote:</div><br><blockquote type="cite">Okay, thanks.<div>I'll take a look.</div><div><br></div><div>Dmitry.<br>
<br><div class="gmail_quote">On Mon, Feb 20, 2012 at 11:30 AM, Hui Zhang <span dir="ltr"><<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word">For reference, my results are attached.<div><br></div><div>asm1.txt for asm with 1 process,</div>
<div>asm2.txt for asm with 2 processes,</div><div>gasm1.txt for gasm with 1 process, (with the iteration numbers different from others)</div><div>gasm2.txt for gasm with 2 processes</div><div><br></div><div></div></div><br>
<div style="word-wrap:break-word"><div></div></div><br><div style="word-wrap:break-word"><div></div></div><br><div style="word-wrap:break-word"><div></div></div><br><div style="word-wrap:break-word"><div></div><div><br></div>
<div>thank you,</div><div>Hui<br><div><br><div><div>On Feb 20, 2012, at 3:06 PM, Dmitry Karpeev wrote:</div><br><blockquote type="cite"><br><br><div class="gmail_quote">On Mon, Feb 20, 2012 at 12:59 AM, Hui Zhang <span dir="ltr"><<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word"><br><div><div><div>On Feb 20, 2012, at 12:41 AM, Dmitry Karpeev wrote:</div><br><blockquote type="cite"><br><br><div class="gmail_quote">On Sun, Feb 19, 2012 at 3:08 PM, Hui Zhang <span dir="ltr"><<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word">I have a new problem: the results from ASM and GASM are different and it seems<div>GASM has something wrong with SetModifySubMatrices. Numerical tests are with </div><div>each subdomain supported only by one subdomain. There are no problems when</div>
<div>I did not modify submatrices. But when I modify submatrices, there are problems</div><div>with GASM but no problems with ASM. </div><div><br></div><div>For example, I use two subdomains. In the first case each subdomain is supported by</div>
<div>one processor and there seems no problem with GASM. But when I use run my program </div><div>with only one proc. so that it supports both of the two subdomains, the iteration </div><div>number is different from the first case and is much larger. On the other hand</div>
<div>ASM has no such problem.</div></div></blockquote><div> </div><div>Are the solutions the same?</div><div>What problem are you solving?</div></div></blockquote><div><br></div></div><div>Yes, the solutions are the same. That's why ASM gives the same results with one or</div>
<div>two processors. But GASM did not. </div></div></div></blockquote><div>Sorry, I wasn't clear: ASM and GASM produced different solutions in the case of two domains per processor?</div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word"><div><div>I'm solving the Helmholtz equation. Maybe </div><div>I can prepare a simpler example to show this difference.</div></div></div></blockquote><div>That would be helpful. </div>
<div>Thanks.</div><div><br></div><div>Dmitry. </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word"><div><div><div><br><blockquote type="cite">
<div class="gmail_quote"><div><br></div><div>Dmitry. </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word"><div><div><div><div><br><div><br><div><div>On Feb 15, 2012, at 6:46 PM, Dmitry Karpeev wrote:</div><br><blockquote type="cite"><div bgcolor="#FFFFFF"><div>You should be able to. </div>
<div>This behavior is the same as in PCASM,</div><div>except in GASM the matrices live on subcommunicators.</div><div>I am in transit right now, but I can take a closer look in Friday.</div><div><br></div><div>Dmitry<br>
<br>
<br></div><div><br>On Feb 15, 2012, at 8:07, Hui Zhang <<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>> wrote:<br><br></div><div></div><blockquote type="cite"><div><div>On Feb 15, 2012, at 11:19 AM, Hui Zhang wrote:</div>
<div><br><blockquote type="cite"><div style="word-wrap:break-word">Hi Dmitry,<div><br></div><div>thanks a lot! Currently, I'm not using ISColoring. Just comes another question</div><div>on PCGASMSetModifySubMatrices(). The user provided function has the prototype</div>
<div><br></div><div><span style="font-family:monospace;white-space:pre-wrap"> func (</span><span style="font-family:monospace;white-space:pre-wrap"><a>PC</a></span><span style="font-family:monospace;white-space:pre-wrap"> pc,</span><span style="font-family:monospace;white-space:pre-wrap"><a>PetscInt</a></span><span style="font-family:monospace;white-space:pre-wrap"> nsub,</span><span style="font-family:monospace;white-space:pre-wrap"><a>IS</a></span><span style="font-family:monospace;white-space:pre-wrap"> *row,</span><span style="font-family:monospace;white-space:pre-wrap"><a>IS</a></span><span style="font-family:monospace;white-space:pre-wrap"> *col,</span><span style="font-family:monospace;white-space:pre-wrap"><a>Mat</a></span><span style="font-family:monospace;white-space:pre-wrap"> *submat,void *ctx);</span></div>
<div><span style="font-family:monospace;white-space:pre-wrap"><br></span></div><div><span style="white-space:pre-wrap">I think the coloumns from the parameter 'col' are always the same <font face="monospace">as the rows </font></span></div>
<div><span style="white-space:pre-wrap"><font face="monospace">from the parameter 'row'. Because PCGASMSetLocalSubdomains() only accepts </font></span></div><div><span style="white-space:pre-wrap"><font face="monospace">index sets but not rows and columns. Has I misunderstood something?</font></span></div>
</div></blockquote><div><br></div><div>As I tested, the row and col are always the same. </div><div><br></div><div>I have a new question. Am I allowed to SetLocalToGlobalMapping() for the submat's</div><div>in the above func()?</div>
<div><br></div><div>thanks,</div><div>Hui</div><br><blockquote type="cite"><div style="word-wrap:break-word"><span style="white-space:pre-wrap"><font face="monospace"><br></font></span></div><div><span style="white-space:pre-wrap"><font face="monospace">thanks,</font></span></div>
<div><span style="white-space:pre-wrap"><font face="monospace">Hui</font></span></div><div><br></div><div><br><div><div>On Feb 11, 2012, at 3:36 PM, Dmitry Karpeev wrote:</div><br><blockquote type="cite">Yes, that's right.<div>
There is no good way to help the user assemble the subdomains at the moment beyond the 2D stuff.</div><div>It is expected that they are generated from mesh subdomains.</div><div>Each IS does carry the subdomains subcomm.</div>
<div><br></div><div>There is ISColoringToList() that is supposed to convert a "coloring" of indices to an array of ISs,</div><div>each having the indices with the same color and the subcomm that supports that color. It is</div>
<div>largely untested, though. You could try using it and give us feedback on any problems you encounter.</div><div><br></div><div>Dmitry.</div><div><br><br><div class="gmail_quote">On Sat, Feb 11, 2012 at 6:06 AM, Hui Zhang <span dir="ltr"><<a href="mailto:mike.hui.zhang@hotmail.com" target="_blank"></a><a href="mailto:mike.hui.zhang@hotmail.com" target="_blank">mike.hui.zhang@hotmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">About PCGASMSetLocalSubdomains(), in the case of one subdomain supported by<br>
multiple processors, shall I always create the arguments 'is[s]' and 'is_local[s]'<br>
in a subcommunicator consisting of processors supporting the subdomain 's'?<br>
<br>
The source code of PCGASMCreateSubdomains2D() seemingly does so.<br>
<br>
Thanks,<br>
Hui<br>
<br>
</blockquote></div><br></div>
</blockquote></div><br></div></blockquote></div><br></div></blockquote></div></blockquote></div><br></div></div></div></div></div></div></blockquote></div><br>
</blockquote></div></div></div><br></div></blockquote></div><br>
</blockquote></div><br></div></div></div><br></blockquote></div><br></div>
</blockquote></div><br></div></div></div></div></blockquote></div><br></div>
</blockquote></div><br></div></div></div></div></div></blockquote></div><br></div></div>
</blockquote></div><br></div></div></div></blockquote></div><br>
</blockquote></div><br></div></div></div></blockquote></div><br></div></blockquote></div><br></div></div></blockquote></div><br></div>
</blockquote></div><br></div></body></html>