<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Thu, Apr 9, 2015 at 2:12 PM, Steena M <span dir="ltr"><<a href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div><div style="color:#000;background-color:#fff;font-family:HelveticaNeue,Helvetica Neue,Helvetica,Arial,Lucida Grande,sans-serif;font-size:16px"><div><div><div style="color:#000;background-color:#fff;font-family:HelveticaNeue,Helvetica Neue,Helvetica,Arial,Lucida Grande,sans-serif;font-size:16px"><div dir="ltr">Thanks, Barry! I patched the master, modified src/mat/examples/tests/ex190.c to suit my data, and the fix works.</div><div dir="ltr"><br></div><div dir="ltr">I need to execute MatMult and I am assigning vectors on each rank using MatCreateVecs(A, &x, &y). However, I don't think multiplication is happening at all. (I inserted a few printf statements inside MatMult_MPIBAIJ and inside MatMult_SeqBAIJ1 to check). VecSum(&ysum) is always zero. Maybe I'm assigning data incorrectly? </div><div dir="ltr"><br></div><div dir="ltr">Code snippet in ex190.c after partitioning the matrix unequally</div><div dir="ltr"><br></div><div dir="ltr"><div dir="ltr"> ierr = MatLoad(A,fd);CHKERRQ(ierr);</div><div dir="ltr"> ierr = MatCreateVecs(A, &x, &y);CHKERRQ(ierr);</div><div dir="ltr"> </div><div dir="ltr"> ierr = VecSet(x,one);CHKERRQ(ierr); //PetscScalar one = 1.0;</div><div dir="ltr"> ierr = VecSet(y,zero); CHKERRQ(ierr); //PetscScalar zero = 0.0;</div></div></div></div></div></div></div></blockquote><div><br></div><div>MatView(A, 0) gives you the matrix, so you can see what Ax should be.</div><div> <br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div><div style="color:#000;background-color:#fff;font-family:HelveticaNeue,Helvetica Neue,Helvetica,Arial,Lucida Grande,sans-serif;font-size:16px"><div><div><div style="color:#000;background-color:#fff;font-family:HelveticaNeue,Helvetica Neue,Helvetica,Arial,Lucida Grande,sans-serif;font-size:16px"><div dir="ltr"><div dir="ltr"> ierr = MatMult(A,x,y); CHKERRQ(ierr);</div></div></div></div></div></div></div></blockquote><div><br></div><div>VecView(y, 0) gives you the output</div><div><br></div><div> Matt</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div><div style="color:#000;background-color:#fff;font-family:HelveticaNeue,Helvetica Neue,Helvetica,Arial,Lucida Grande,sans-serif;font-size:16px"><div><div><div style="color:#000;background-color:#fff;font-family:HelveticaNeue,Helvetica Neue,Helvetica,Arial,Lucida Grande,sans-serif;font-size:16px"><div dir="ltr"><div dir="ltr"> ierr = VecSum(y,&ysum);CHKERRQ(ierr); //ysum is always zero. </div><div dir="ltr"><br></div><div dir="ltr">Data is a 20x20 matrix and block size is 1.</div><div dir="ltr"><br></div><div dir="ltr">Any thoughts?</div><div dir="ltr"><br></div><div dir="ltr">Thanks,</div><div dir="ltr">Steena</div><div dir="ltr"> </div><div dir="ltr"> </div></div><div dir="ltr">. <br></div> <div dir="ltr"> </div><br><div><br><br></div></div></div></div><div> <div style="font-family:HelveticaNeue,Helvetica Neue,Helvetica,Arial,Lucida Grande,sans-serif;font-size:16px"> <div style="font-family:HelveticaNeue,Helvetica Neue,Helvetica,Arial,Lucida Grande,sans-serif;font-size:16px"> <div dir="ltr"> <font face="Arial"> On Friday, April 3, 2015 4:17 PM, Barry Smith <<a href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>> wrote:<br> </font> </div> <br><br> <div><br> Steena,<br><br> Sorry for all the delays. Our code was just totally wrong for a user provided decomposition for two reasons 1) the logic in MatLoad_MPIXAIJ() was too convoluted to work in all cases and 2) we didn't have a single test case.<br><br> I have attached a patch that fixes it for petsc-3.5.3 <br>. The fix is also in the branch barry/fix-matload-uneven-rows<br>and I'll put it into next then after testing it will got into maint and master and the next PETSc patch release.<br><br> Please let us know if the patch doesn't work for you<br><br> Thanks for your patience,<br><br> Barry<br><br>> On Apr 1, 2015, at 3:10 PM, Steena M <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>> wrote:<br>> <br>> Thanks Barry. Attached is the driver program, the binary mat file, and the corresponding mtx file.<br>> <br>> Runtime command used:<br>> <br>> <a rel="nofollow" href="mailto:sierra324@monteiro" target="_blank">sierra324@monteiro</a>:time srun -n 2 -ppdebug ./petsc-mpibaij-unequalrows -fin trefethen.dat -matload_block_size 1<br>> <br>> <br>> <br>> <br>> On Wednesday, April 1, 2015 12:28 PM, Barry Smith <<a rel="nofollow" href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>> wrote:<br>> <br>> <br>> <br>> Send a data file you generated and your reader program and we'll debug it.<br>> <br>> Barry<br>> <br>> > On Apr 1, 2015, at 2:18 PM, Steena M <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>> wrote:<br>> > <br>> > Thanks Barry. I removed the Preallocation calls. It is still complaining about the malloc and incorrect data in the matrix file. I generate binary matrix files using PETSc's pythonscript to loop through a set of UFL sparse matrices. For this use case:<br>> > <br>> > mtx_mat = scipy.io.mmread('trefethen.mtx')<br>> > PetscBinaryIO.PetscBinaryIO().writeMatSciPy(open('trefnew.dat','w'), mtx_mat)<br>> > <br>> > <br>> > <br>> > <br>> > <br>> > <br>> > On Tuesday, March 31, 2015 9:15 PM, Barry Smith <<a rel="nofollow" href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>> wrote:<br>> > <br>> > <br>> > <br>> > You should not need to call any preallocation routines when using MatLoad()<br>> > <br>> > <br>> > How did you generate the file? Are you sure it has the correct information for the matrix? <br>> > <br>> > Barry<br>> > <br>> > <br>> > <br>> > > On Mar 31, 2015, at 11:05 PM, Steena M <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>> wrote:<br>> > > <br>> > > Thanks Matt. I'm still getting the malloc error <br>> > > <br>> > > [0]PETSC ERROR: Argument out of range!<br>> > > [0]PETSC ERROR: New nonzero at (2,18) caused a malloc!<br>> > > <br>> > > and <br>> > > <br>> > > a new incorrect matrix file error:<br>> > > <br>> > > [0]PETSC ERROR: Unexpected data in file!<br>> > > [0]PETSC ERROR: not matrix object!<br>> > > <br>> > > Maybe the order of calls is mixed up. This is the code snippet:<br>> > > <br>> > > if (rank ==0)<br>> > > {<br>> > > PetscPrintf (PETSC_COMM_WORLD,"\n On rank %d ", rank);<br>> > > <br>> > > CHKERRQ(MatSetSizes(A, 15, PETSC_DETERMINE, 20, 20));<br>> > > CHKERRQ(MatSetType(A, MATMPIBAIJ));<br>> > > CHKERRQ( MatMPIBAIJSetPreallocation(A,1,1,NULL,1,NULL));<br>> > > CHKERRQ( MatLoad(A,fd)); <br>> > > CHKERRQ(MatSetOption(A,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_FALSE));<br>> > > }<br>> > > <br>> > > else<br>> > > {<br>> > > PetscPrintf (PETSC_COMM_WORLD,"\n On rank %d ", rank);<br>> > > <br>> > > CHKERRQ( MatSetSizes(A, 5, PETSC_DETERMINE, 20, 20) );<br>> > > CHKERRQ(MatSetType(A, MATMPIBAIJ));<br>> > > CHKERRQ( MatMPIBAIJSetPreallocation(A,1,1,NULL,1,NULL));<br>> > > CHKERRQ(MatLoad(A,fd));<br>> > > CHKERRQ(MatSetOption(A,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_FALSE));<br>> > > }<br>> > > <br>> > > Is there something I'm missing? <br>> > > <br>> > > Thanks,<br>> > > Steena<br>> > > <br>> > > <br>> > > <br>> > > <br>> > > <br>> > > On Tuesday, March 31, 2015 6:10 PM, Matthew Knepley <<a rel="nofollow" href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>> wrote:<br>> > > <br>> > > <br>> > > On Tue, Mar 31, 2015 at 6:51 PM, Steena M <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>> wrote:<br>> > > Thanks Barry. I'm still getting the malloc error with NULL. Is there a way to distribute the matrix without explicit preallocation? Different matrices will be loaded during runtime and assigning preallocation parameters would mean an additional preprocessing step.<br>> > > <br>> > > 1) MatSetOption(MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_FALSE)<br>> > > <br>> > > 2) Note that this is never ever ever more efficient than making another pass and preallocating<br>> > > <br>> > > Thanks,<br>> > > <br>> > > Matt<br>> > > --------------------------------------------<br>> > > On Sun, 3/29/15, Barry Smith <<a rel="nofollow" href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>> wrote:<br>> > > <br>> > > Subject: Re: [petsc-users] Unequal sparse matrix row distribution for MPI MatMult<br>> > > To: "Steena M" <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>><br>> > > Cc: "Matthew Knepley" <<a rel="nofollow" href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>>, <a rel="nofollow" href="mailto:petsc-users@mcs.anl.gov" target="_blank">petsc-users@mcs.anl.gov</a><br>> > > Date: Sunday, March 29, 2015, 9:26 PM<br>> > > <br>> > > <br>> > > > On<br>> > > Mar 29, 2015, at 11:05 PM, Steena M <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>><br>> > > wrote:<br>> > > ><br>> > > > Thanks<br>> > > Matt. I used PETSC_DETERMINE but I'm now getting an<br>> > > allocation-based error:<br>> > > ><br>> > > > [0]PETSC ERROR: ---------------------<br>> > > Error Message ------------------------------------<br>> > > > [0]PETSC ERROR: Argument out of range!<br>> > > > [0]PETSC ERROR: New nonzero at (2,18)<br>> > > caused a malloc!<br>> > > > [0]PETSC ERROR:<br>> > > ------------------------------------------------------------------------<br>> > > ><br>> > > > I tried<br>> > > preallocating on each rank for the diagonal and off diagonal<br>> > > section of the matrix as the next step My current<br>> > > approximations for preallocation<br>> > > ><br>> > > > CHKERRQ(<br>> > > MatMPIBAIJSetPreallocation(A,1,5,PETSC_DEFAULT,5,PETSC_DEFAULT)); <br>> > > <br>> > > <br>> > > These<br>> > > arguments where you pass PETSC_DEFAULT are expecting a<br>> > > pointer not an integer. You can pass NULL in those<br>> > > locations. Though it is better to provide the correct<br>> > > preallocation rather than some defaults.<br>> > > <br>> > > Barry<br>> > > <br>> > > ><br>> > > > are throwing segmentation errors.<br>> > > ><br>> > > > [0]PETSC ERROR: <br>> > > Caught signal number 11 SEGV: Segmentation Violation,<br>> > > probably memory access out of range<br>> > > ><br>> > > > Any insights into what I'm doing<br>> > > wrong?<br>> > > ><br>> > > > Thanks,<br>> > > > Steena<br>> > > ><br>> > > ><br>> > > ><br>> > > > On Sun, 3/29/15, Matthew Knepley <<a rel="nofollow" href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>><br>> > > wrote:<br>> > > ><br>> > > > Subject:<br>> > > Re: [petsc-users] Unequal sparse matrix row distribution for<br>> > > MPI MatMult<br>> > > > To: "Steena M"<br>> > > <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>><br>> > > > Cc: "Barry Smith" <<a rel="nofollow" href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>>,<br>> > > <a rel="nofollow" href="mailto:petsc-users@mcs.anl.gov" target="_blank">petsc-users@mcs.anl.gov</a><br>> > > > Date: Sunday, March 29, 2015, 10:02 PM<br>> > > ><br>> > > > On Sun, Mar 29, 2015<br>> > > at<br>> > > > 9:56 PM, Steena M <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>><br>> > > > wrote:<br>> > > > Hi<br>> > > > Barry,<br>> > > ><br>> > > ><br>> > > ><br>> > > > I am trying to partition a 20 row and 20<br>> > > col sparse matrix<br>> > > > between two procs<br>> > > such that proc 0 has 15 rows and 20 cols<br>> > > > and proc 1 has 5 rows and 20 cols. The<br>> > > code snippet:<br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > > <br>> > > CHKERRQ(MatCreate(PETSC_COMM_WORLD,&A));<br>> > > //<br>> > > > at runtime: -matload_block_size 1<br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > > <br>> > > if (rank ==0)<br>> > > ><br>> > > > {<br>> > > ><br>> > > > <br>> > > CHKERRQ( MatSetSizes(A, 15, 20, 20,<br>> > > > 20) ); //rank 0 gets 75% of the rows<br>> > > ><br>> > > > <br>> > > CHKERRQ( MatSetType(A, MATMPIBAIJ)<br>> > > > );<br>> > > ><br>> > > > CHKERRQ(<br>> > > MatLoad(A,fd) );<br>> > > ><br>> > > > }<br>> > > ><br>> > > ><br>> > > ><br>> > > > else<br>> > > ><br>> > > > <br>> > > {<br>> > > ><br>> > > > <br>> > > CHKERRQ( MatSetSizes(A, 5,<br>> > > 20, 20,<br>> > > > 20) ); //rank 1 gets 25% of the<br>> > > rows<br>> > > ><br>> > > > <br>> > > CHKERRQ( MatSetType(A, MATMPIBAIJ)<br>> > > > );<br>> > > ><br>> > > > CHKERRQ(<br>> > > MatLoad(A,fd) );<br>> > > ><br>> > > > }<br>> > > ><br>> > > <br>> > > ><br>> > > ><br>> > > > This throws the following error (probably<br>> > > from psplit.c):<br>> > > ><br>> > > ><br>> > > [1]PETSC ERROR: --------------------- Error Message<br>> > > > ------------------------------------<br>> > > ><br>> > > > [1]PETSC ERROR:<br>> > > Nonconforming object sizes!<br>> > > ><br>> > > > [1]PETSC ERROR: Sum of local lengths 40<br>> > > does not equal<br>> > > > global length 20, my<br>> > > local length 20<br>> > > ><br>> > > > likely a call to<br>> > > VecSetSizes() or MatSetSizes() is<br>> > > ><br>> > > wrong.<br>> > > ><br>> > > > See<br>> > > <a href="http://www.mcs.anl.gov/petsc/documentation/faq.html#split" target="_blank">http://www.mcs.anl.gov/petsc/documentation/faq.html#split</a>!<br>> > > ><br>> > > ><br>> > > ><br>> > > > This error printout<br>> > > doesn't quite make sense to me.<br>> > > ><br>> > > I'm trying to specify a total matrix size of 20x20...<br>> > > I<br>> > > > haven't yet figured out where the<br>> > > '40' comes<br>> > > > from in the error<br>> > > message.<br>> > > ><br>> > > ><br>> > > ><br>> > > > Any thoughts on what<br>> > > might be going wrong?<br>> > > ><br>> > > > Its the column specification. Just<br>> > > > use PETSC_DETERMINE for the local columns<br>> > > since all our<br>> > > > sparse matrixformats are<br>> > > row divisions<br>> > > > anyway.<br>> > > > <br>> > > > Thanks,<br>> > > > <br>> > > > Matt<br>> > > > Thanks in advance,<br>> > > ><br>> > > <br>> > > > Steena<br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > --------------------------------------------<br>> > > ><br>> > > > On Sun, 3/22/15,<br>> > > Barry Smith <<a rel="nofollow" href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>><br>> > > > wrote:<br>> > > ><br>> > > ><br>> > > ><br>> > > > Subject: Re: [petsc-users] Unequal<br>> > > sparse matrix row<br>> > > > distribution for MPI<br>> > > MatMult<br>> > > ><br>> > > > To:<br>> > > "Steena M" <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>><br>> > > ><br>> > > > Cc: <a rel="nofollow" href="mailto:petsc-users@mcs.anl.gov" target="_blank">petsc-users@mcs.anl.gov</a><br>> > > ><br>> > > > Date: Sunday,<br>> > > March 22, 2015, 3:58 PM<br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > > <br>> > > ><br>> > > > Steena,<br>> > > ><br>> > > ><br>> > > ><br>> > > > I am<br>> > > ><br>> > > <br>> > > > a little unsure of your question.<br>> > > <br>> > > ><br>> > > ><br>> > > ><br>> > > > 1)<br>> > > you can create a MPIBAIJ<br>> > > ><br>> > > > matrix with any distribution of block<br>> > > rows per process<br>> > > > you<br>> > > ><br>> > > > want, just set the<br>> > > local row size for each process to<br>> > > ><br>> > > be<br>> > > ><br>> > > > what you<br>> > > like. Use MatCreateVecs() to get<br>> > > ><br>> > > correspondingly<br>> > > ><br>> > > > <br>> > > laid out vectors.<br>> > > ><br>> > > ><br>> > > <br>> > > ><br>> > > > <br>> > > or 2) if you have a MPIBAIJ<br>> > > ><br>> > > > matrix with<br>> > > "equal" row layout and you want a<br>> > > > new<br>> > > ><br>> > > > one with uneven row layout you can<br>> > > simply use<br>> > > ><br>> > > > <br>> > > MatGetSubMatrix() to create that new matrix.<br>> > > ><br>> > > ><br>> > > ><br>> > > > Barry<br>> > > ><br>> > > ><br>> > > ><br>> > > > Unless you have<br>> > > another reason to have the<br>> > > ><br>> > > > matrix with an equal number row layout I<br>> > > would just<br>> > > > generate<br>> > > ><br>> > > > the matrix with<br>> > > the layout you want.<br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > > > On Mar 22, 2015, at 5:50 PM, Steena<br>> > > M<br>> > > ><br>> > > > <<a rel="nofollow" href="mailto:stm8086@yahoo.com" target="_blank">stm8086@yahoo.com</a>><br>> > > ><br>> > > > wrote:<br>> > > ><br>> > > > ><br>> > > ><br>> > > > > Hello,<br>> > > ><br>> > > > ><br>> > > ><br>> > > > > I need to<br>> > > distribute<br>> > > ><br>> > > > a<br>> > > sparse matrix such that each proc owns an unequal<br>> > > > number<br>> > > ><br>> > > > of blocked rows before I proceed with<br>> > > MPI MatMult. My<br>> > > ><br>> > > > <br>> > > initial thoughts on doing this:<br>> > > ><br>> > > > ><br>> > > ><br>> > > > > 1) Use MatGetSubMatrices() on the<br>> > > test<br>> > > ><br>> > > > MATMPIBAIJ<br>> > > matrix to produce a new matrix where each<br>> > > > proc<br>> > > ><br>> > > > has an unequal number of rows.<br>> > > ><br>> > > > ><br>> > > ><br>> > > > > 2) Provide<br>> > > scatter context for vector X<br>> > > ><br>> > > > (for MatMult )using IS iscol from<br>> > > MatGetSubMatrices()<br>> > > > while<br>> > > ><br>> > > > creating the<br>> > > vector X.<br>> > > ><br>> > > > <br>> > > ><br>> > > ><br>> > > > > 3)<br>> > > Call MatMult()<br>> > > ><br>> > > > <br>> > > ><br>> > > ><br>> > > > > Will<br>> > > MatMult_MPIBAIJ continue to scatter<br>> > > ><br>> > > > this matrix and vector such that each<br>> > > proc will own an<br>> > > > equal<br>> > > ><br>> > > > number of matrix<br>> > > rows and corresponding diagonal vector<br>> > > ><br>> > > <br>> > > > elements? Should I write my own<br>> > > MPIMatMult function to<br>> > > ><br>> > > > retain my redistribution of the matrix<br>> > > and vector?<br>> > > ><br>> > > > <br>> > > ><br>> > > ><br>> > > > ><br>> > > Thanks in<br>> > > ><br>> > > > <br>> > > advance,<br>> > > ><br>> > > > ><br>> > > Steena<br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > ><br>> > > > --<br>> > > > What most<br>> > > experimenters<br>> > > > take for granted before<br>> > > they begin their experiments is<br>> > > ><br>> > > infinitely more interesting than any results to which<br>> > > their<br>> > > > experiments lead.<br>> > > > -- Norbert<br>> > > > Wiener<br>> > > <br>> > > ><br>> > > <br>> > > <br>> > > <br>> > > <br>> > > -- <br>> > > 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<br>> > > <br>> > > <br>> > <br>> > <br>> <br>> <br>> <trefethen.mtx><trefethen.dat><petsc-unequalrows-mpibaij.c><br><br><br></div> </div> </div> </div></div></div></blockquote></div><br><br clear="all"><div><br></div>-- <br><div class="gmail_signature">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></div>