<html><head><meta http-equiv="content-type" content="text/html; charset=us-ascii"></head><body style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;">Hi all,<div><br></div><div>I am trying to multiply two Petsc matrices as C = A * B, where A is a tall matrix and B is a relatively small matrix.</div><div><br></div><div>I have taken the decision to create A as (row-)partitioned matrix and B as a non-partitioned matrix that it is entirely shared by all procs (to avoid unnecessary communication).</div><div><br></div><div>Here is my code:</div><div><br></div><div style="color: rgb(29, 33, 38); background-color: rgb(255, 255, 255); font-family: Menlo, Monaco, "Courier New", monospace; line-height: 18px; white-space: pre;"><div><span style="color: rgb(168, 3, 20);">import</span> <span style="color: rgb(72, 31, 148);">numpy</span> <span style="color: rgb(168, 3, 20);">as</span> <span style="color: rgb(72, 31, 148);">np</span></div><div><span style="color: rgb(168, 3, 20);">from</span> <span style="color: rgb(72, 31, 148);">firedrake</span> <span style="color: rgb(168, 3, 20);">import</span> <span style="color: rgb(13, 92, 181);">COMM_WORLD</span></div><div><span style="color: rgb(168, 3, 20);">from</span> <span style="color: rgb(72, 31, 148);">firedrake</span>.<span style="color: rgb(72, 31, 148);">petsc</span> <span style="color: rgb(168, 3, 20);">import</span> <span style="color: rgb(72, 31, 148);">PETSc</span></div><div><span style="color: rgb(168, 3, 20);">from</span> <span style="color: rgb(72, 31, 148);">numpy</span>.<span style="color: rgb(72, 31, 148);">testing</span> <span style="color: rgb(168, 3, 20);">import</span> <span style="color: rgb(72, 31, 148);">assert_array_almost_equal</span></div><br><div>nproc <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(13, 92, 181);">COMM_WORLD</span>.size</div><div>rank <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(13, 92, 181);">COMM_WORLD</span>.rank</div><br><div><span style="color: rgb(168, 3, 20);">def</span> <span style="color: rgb(72, 31, 148);">create_petsc_matrix_non_partitioned</span>(<span style="color: rgb(166, 61, 0);">input_array</span>):</div><div> <span style="color: rgb(10, 66, 131);">"""Building a mpi non-partitioned petsc matrix from an array</span></div><br><div><span style="color: rgb(10, 66, 131);"> Args:</span></div><div><span style="color: rgb(10, 66, 131);"> input_array (np array): Input array</span></div><div><span style="color: rgb(10, 66, 131);"> sparse (bool, optional): Toggle for sparese or dense. Defaults to True.</span></div><br><div><span style="color: rgb(10, 66, 131);"> Returns:</span></div><div><span style="color: rgb(10, 66, 131);"> mpi mat: PETSc matrix</span></div><div><span style="color: rgb(10, 66, 131);"> """</span></div><div> <span style="color: rgb(168, 3, 20);">assert</span> <span style="color: rgb(72, 31, 148);">len</span>(<span style="color: rgb(166, 61, 0);">input_array</span>.shape) <span style="color: rgb(168, 3, 20);">==</span> <span style="color: rgb(13, 92, 181);">2</span></div><br><div> m, n <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(166, 61, 0);">input_array</span>.shape</div><br><div> matrix <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(72, 31, 148);">PETSc</span>.Mat().createAIJ(<span style="color: rgb(166, 61, 0);">size</span><span style="color: rgb(168, 3, 20);">=</span>((m, n), (m, n)), <span style="color: rgb(166, 61, 0);">comm</span><span style="color: rgb(168, 3, 20);">=</span><span style="color: rgb(13, 92, 181);">COMM_WORLD</span>)</div><br><div> <span style="color: rgb(164, 176, 188);"># Set the values of the matrix</span></div><div> matrix.setValues(<span style="color: rgb(72, 31, 148);">range</span>(m), <span style="color: rgb(72, 31, 148);">range</span>(n), <span style="color: rgb(166, 61, 0);">input_array</span>[:, :], <span style="color: rgb(166, 61, 0);">addv</span><span style="color: rgb(168, 3, 20);">=</span><span style="color: rgb(13, 92, 181);">False</span>)</div><br><div> <span style="color: rgb(164, 176, 188);"># Assembly the matrix to compute the final structure</span></div><div> matrix.assemblyBegin()</div><div> matrix.assemblyEnd()</div><br><div> <span style="color: rgb(168, 3, 20);">return</span> matrix</div><br><br><div><span style="color: rgb(168, 3, 20);">def</span> <span style="color: rgb(72, 31, 148);">create_petsc_matrix</span>(<span style="color: rgb(166, 61, 0);">input_array</span>, <span style="color: rgb(166, 61, 0);">partition_like</span><span style="color: rgb(168, 3, 20);">=</span><span style="color: rgb(13, 92, 181);">None</span>):</div><div> <span style="color: rgb(10, 66, 131);">"""Create a PETSc matrix from an input_array</span></div><br><div><span style="color: rgb(10, 66, 131);"> Args:</span></div><div><span style="color: rgb(10, 66, 131);"> input_array (np array): Input array</span></div><div><span style="color: rgb(10, 66, 131);"> partition_like (petsc mat, optional): Petsc matrix. Defaults to None.</span></div><div><span style="color: rgb(10, 66, 131);"> sparse (bool, optional): Toggle for sparese or dense. Defaults to True.</span></div><br><div><span style="color: rgb(10, 66, 131);"> Returns:</span></div><div><span style="color: rgb(10, 66, 131);"> petsc mat: PETSc matrix</span></div><div><span style="color: rgb(10, 66, 131);"> """</span></div><div> <span style="color: rgb(164, 176, 188);"># Check if input_array is 1D and reshape if necessary</span></div><div> <span style="color: rgb(168, 3, 20);">assert</span> <span style="color: rgb(72, 31, 148);">len</span>(<span style="color: rgb(166, 61, 0);">input_array</span>.shape) <span style="color: rgb(168, 3, 20);">==</span> <span style="color: rgb(13, 92, 181);">2</span>, <span style="color: rgb(10, 66, 131);">"Input array should be 2-dimensional"</span></div><div> global_rows, global_cols <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(166, 61, 0);">input_array</span>.shape</div><br><div> comm <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(13, 92, 181);">COMM_WORLD</span></div><div> <span style="color: rgb(168, 3, 20);">if</span> <span style="color: rgb(166, 61, 0);">partition_like</span> <span style="color: rgb(168, 3, 20);">is</span> <span style="color: rgb(168, 3, 20);">not</span> <span style="color: rgb(13, 92, 181);">None</span>:</div><div> local_rows_start, local_rows_end <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(166, 61, 0);">partition_like</span>.getOwnershipRange()</div><div> local_rows <span style="color: rgb(168, 3, 20);">=</span> local_rows_end <span style="color: rgb(168, 3, 20);">-</span> local_rows_start</div><br><div> <span style="color: rgb(164, 176, 188);"># No parallelization in the columns, set local_cols = None to parallelize</span></div><div> size <span style="color: rgb(168, 3, 20);">=</span> ((local_rows, global_rows), (global_cols, global_cols))</div><div> <span style="color: rgb(168, 3, 20);">else</span>:</div><div> size <span style="color: rgb(168, 3, 20);">=</span> ((<span style="color: rgb(13, 92, 181);">None</span>, global_rows), (global_cols, global_cols))</div><br><div> matrix <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(72, 31, 148);">PETSc</span>.Mat().createAIJ(<span style="color: rgb(166, 61, 0);">size</span><span style="color: rgb(168, 3, 20);">=</span>size, <span style="color: rgb(166, 61, 0);">comm</span><span style="color: rgb(168, 3, 20);">=</span>comm)</div><div> matrix.setUp()</div><br><div> local_rows_start, local_rows_end <span style="color: rgb(168, 3, 20);">=</span> matrix.getOwnershipRange()</div><br><div> <span style="color: rgb(168, 3, 20);">for</span> counter, i <span style="color: rgb(168, 3, 20);">in</span> <span style="color: rgb(72, 31, 148);">enumerate</span>(<span style="color: rgb(72, 31, 148);">range</span>(local_rows_start, local_rows_end)):</div><div> <span style="color: rgb(164, 176, 188);"># Calculate the correct row in the array for the current process</span></div><div> row_in_array <span style="color: rgb(168, 3, 20);">=</span> counter <span style="color: rgb(168, 3, 20);">+</span> local_rows_start</div><div> matrix.setValues(</div><div> i, <span style="color: rgb(72, 31, 148);">range</span>(global_cols), <span style="color: rgb(166, 61, 0);">input_array</span>[row_in_array, :], <span style="color: rgb(166, 61, 0);">addv</span><span style="color: rgb(168, 3, 20);">=</span><span style="color: rgb(13, 92, 181);">False</span></div><div> )</div><br><div> <span style="color: rgb(164, 176, 188);"># Assembly the matrix to compute the final structure</span></div><div> matrix.assemblyBegin()</div><div> matrix.assemblyEnd()</div><br><div> <span style="color: rgb(168, 3, 20);">return</span> matrix</div><br><br><div>m, k <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(13, 92, 181);">10</span>, <span style="color: rgb(13, 92, 181);">3</span></div><div><span style="color: rgb(164, 176, 188);"># Generate the random numpy matrices</span></div><div><span style="color: rgb(72, 31, 148);">np</span>.<span style="color: rgb(72, 31, 148);">random</span>.seed(<span style="color: rgb(13, 92, 181);">0</span>) <span style="color: rgb(164, 176, 188);"># sets the seed to 0</span></div><div>A_np <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(72, 31, 148);">np</span>.<span style="color: rgb(72, 31, 148);">random</span>.randint(<span style="color: rgb(166, 61, 0);">low</span><span style="color: rgb(168, 3, 20);">=</span><span style="color: rgb(13, 92, 181);">0</span>, <span style="color: rgb(166, 61, 0);">high</span><span style="color: rgb(168, 3, 20);">=</span><span style="color: rgb(13, 92, 181);">6</span>, <span style="color: rgb(166, 61, 0);">size</span><span style="color: rgb(168, 3, 20);">=</span>(m, k))</div><div>B_np <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(72, 31, 148);">np</span>.<span style="color: rgb(72, 31, 148);">random</span>.randint(<span style="color: rgb(166, 61, 0);">low</span><span style="color: rgb(168, 3, 20);">=</span><span style="color: rgb(13, 92, 181);">0</span>, <span style="color: rgb(166, 61, 0);">high</span><span style="color: rgb(168, 3, 20);">=</span><span style="color: rgb(13, 92, 181);">6</span>, <span style="color: rgb(166, 61, 0);">size</span><span style="color: rgb(168, 3, 20);">=</span>(k, k))</div><br><br><div><span style="color: rgb(13, 92, 181);">A</span> <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(72, 31, 148);">create_petsc_matrix</span>(A_np)</div><br><div><span style="color: rgb(13, 92, 181);">B</span> <span style="color: rgb(168, 3, 20);">=</span> <span style="color: rgb(72, 31, 148);">create_petsc_matrix_non_partitioned</span>(B_np)</div><br><div><span style="color: rgb(164, 176, 188);"># Now perform the multiplication</span></div></div><div><span style="font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; color: rgb(13, 92, 181);">C</span><span style="color: rgb(29, 33, 38); font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; background-color: rgb(255, 255, 255);"> </span><span style="font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; color: rgb(168, 3, 20);">=</span><span style="color: rgb(29, 33, 38); font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; background-color: rgb(255, 255, 255);"> </span><span style="font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; color: rgb(13, 92, 181);">A</span><span style="color: rgb(29, 33, 38); font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; background-color: rgb(255, 255, 255);"> </span><span style="font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; color: rgb(168, 3, 20);">*</span><span style="color: rgb(29, 33, 38); font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; background-color: rgb(255, 255, 255);"> </span><span style="font-family: Menlo, Monaco, "Courier New", monospace; white-space: pre; color: rgb(13, 92, 181);">B</span></div><div><br></div><div>The problem with this is that there is a mismatch between the local rows of A (depend on the partitioning) and the global rows of B (3 for all procs), so that the multiplication cannot happen in parallel. Here is the error:</div><div><br></div><div><div>[0]PETSC ERROR: ------------------------------------------------------------------------</div><div>[0]PETSC ERROR: Caught signal number 11 SEGV: Segmentation Violation, probably memory access out of range</div><div>[0]PETSC ERROR: Try option -start_in_debugger or -on_error_attach_debugger</div><div>[0]PETSC ERROR: or see https://petsc.org/release/faq/#valgrind and https://petsc.org/release/faq/</div><div>[0]PETSC ERROR: configure using --with-debugging=yes, recompile, link, and run</div><div>[0]PETSC ERROR: to get more information on the crash.</div><div>[0]PETSC ERROR: Run with -malloc_debug to check if memory corruption is causing the crash.</div><div>application called MPI_Abort(MPI_COMM_WORLD, 59) - process 0</div><div>[unset]: write_line error; fd=-1 buf=:cmd=abort exitcode=59</div><div>:</div><div>system msg for write_line failure : Bad file descriptor</div></div><div><br></div><div><br></div><div>Is there a standard way to achieve this?</div><div><br></div><div>Thanks,</div><div>Thanos</div><div><br></div><div> </div></body></html>