Recent posts
https://software.intel.com/en-us/recent/240886
enPARDISO - Phase 33
https://software.intel.com/en-us/forums/topic/534606
<p>Hello</p>
<p>I am solving (relatively) a big system with PARDISO (6810921 nonzero terms, 64080 equations, and 508 columns in load vector). I am calling Phase 11, 22 and 33.</p>
<p>I just noticed that PARDISO tries to allocate significant amount of memory when "Phase 33" is called, and then returns me memory allocation.</p>
<p>I understand that Phase 11 and 22 needs to allocate memory for factorization, but I was not expecting high memory demand for Phase 33 as well.</p>
<p>Is this expected? Is there a way to know how much memory needed for Phase 33? Anyway to reduce this memory demand (out-of-core option is not helping me for this).</p>
<p>Regards</p>
<p>Bulent</p>
<p> </p>
<p> </p>
<p> </p>
<p> </p>
Wed, 29 Oct 14 09:44:12 -0700bulent534606pardiso_getdiag
https://software.intel.com/en-us/forums/topic/532338
<p>Hello</p>
<p>I am calling "pardiso_getdiag( pt, df, da, mnum, error)" after PARDISO call (phase = 22) and it returns diagonal terms of initial and final (factorized) matrix.</p>
<p>As noticed, these results returned are not in the same order of the original matrix (they are permuted).</p>
<p> </p>
<p>Is there a way to get these results in the same order of original matrix.</p>
<p> </p>
<p>Bulent</p>
<p> </p>
<p> </p>
<p> </p>
Fri, 03 Oct 14 08:42:23 -0700bulent532338KMP Affinity Issue
https://software.intel.com/en-us/forums/topic/301473
<p>Hi<br />We are using INTEL-MKL Pardiso solver in our programs and recently one of our clients reported that they are seeing the following message when the program runs on an computer with AMD microprocessor:</p>
<p>OMP: Warning #72: KMP_Affinity: affinity only supported for Intel processors. OMP: Warning #71: KMP_Affinity: affinity not supported, using "disabled"</p>
<p>and then, it aborts the program.</p>
<p>Recently we moved from MKL 10.3 Update 3 to MKL 10.3 Update 9 (Windows). The client also said that they did not see this happening with the previous version (that was using MKL 10.3.3.).</p>
<p>In the current version, we also provided the option of using "MKL_NUM_THREADS=1" or<br />"MKL_NUM_THREADS = All Physical Cores" (but we do not use OMP_NUM_THREADS or any other related parameters). It seems like "MKL_NUM_THREADS=1" is not helping much.</p>
<p>ANy help is appreciated.</p>
<p>Regards<br />Bulent</p>
Thu, 23 Aug 12 09:53:18 -0700bulent301473PARDISO + ILL CONDITIONED MATRIX
https://software.intel.com/en-us/forums/topic/279398
<p>Hi All</p>
<p>I am using PARDISO to solve a system of KX=F (K: real symmetric definite matrix). </p>
<p>I have these two cases: K1 and K2 (both matrices are given below). PARDISO returns a success for K1 and it gives an error for K2 (instability in the system).</p>
<p>In fact, both K1 and K2 are not well-conditioned matrices (i.e., they are ill-conditioned) but interestingly the PARDISO returns success for K1. The condition numbers I calculated are something like 10^16 for K1 and 10^17 for K2. </p>
<p>Is there a way to avoid this kind of problems? Any way to set some parameters for PARDISO that catches this? We have other (in-house) solvers that catches this, and I am suprised that PARDISO fails to detect it !..</p>
<p>Regards</p>
<p>Bulent</p>
<p>P.S. My system configuration: Windows7, 64bit, VS2010 C++</p>
<p>K1:</p>
<p>%%MatrixMarket matrix coordinate real symmetric </p>
<p>8 8 16 </p>
<p>1 1 14098.547767242315000000 </p>
<p>1 3 3485.242905845857900000 </p>
<p>1 5 6812.909667832833300000 </p>
<p>2 2 14098.547767242315000000 </p>
<p>2 6 3485.242905845857900000 </p>
<p>2 8 6812.909667832833300000 </p>
<p>3 3 82377.582791945460000000 </p>
<p>3 5 3485.242905845857900000 </p>
<p>3 6 -81215.835156663510000000 </p>
<p>4 4 81215.835156663510000000 </p>
<p>4 7 -0.000000000000113687 </p>
<p>5 5 14098.547767242315000000 </p>
<p>6 6 82377.582791945460000000 </p>
<p>6 8 3485.242905845857900000 </p>
<p>7 7 81215.835156663510000000 </p>
<p>8 8 14098.547767242315000000 </p>
<p>K2:</p>
<p>%%MatrixMarket matrix coordinate real symmetric </p>
<p>8 8 17 </p>
<p>1 1 14571.276198818963000000 </p>
<p>1 3 3642.819049704741200000 </p>
<p>1 5 7285.638099409481600000 </p>
<p>2 2 14571.276198818963000000 </p>
<p>2 6 3642.819049704741200000 </p>
<p>2 8 7285.638099409481600000 </p>
<p>3 3 82430.108173231754000000 </p>
<p>3 5 3642.819049704741200000 </p>
<p>3 6 -81215.835156663510000000 </p>
<p>4 4 81215.835156663510000000 </p>
<p>4 5 -0.000000000000454747 </p>
<p>4 7 0.000000000000113687 </p>
<p>5 5 14571.276198818963000000 </p>
<p>6 6 82430.108173231754000000 </p>
<p>6 8 3642.819049704741200000 </p>
<p>7 7 81215.835156663510000000 </p>
<p>8 8 14571.276198818963000000 </p>
Mon, 05 Mar 12 12:42:19 -0800bulent279398Visual Studio 2010 (C++) + PARDISO Issue
https://software.intel.com/en-us/forums/topic/281872
<p>Hi</p>
<p>I recently noticed this unusual behavior in one of our big products . I am solving the following system repeatedly: K U = F and </p>
<p>each time I call PARDISO to get U. I confirmed that it is the same K and F matrices passed into PARDISO at each call (size of these matrices are really small). </p>
<p>I follow the following steps at each PARDISO call: phase 11, phase 22 and phase 33 and then, -1 to release all memories allocated.</p>
<p>If I run the model from VS2010 (C++ and DEBUG mode) and if I use MKL 10.3.3, the first try to solve system is a success. But the 2nd try to solve the same system always stuck in phase 22 (I see all 8 cores hits 100%, and it keeps going like that and never bails out. I had to kill the process myself. Not sure what is happening but I might speculate that it i srelated todata racing orparallel running)</p>
<p>If I run the model from VS2010 and if I use MKL 10.2.5, it is always successfull on each call.</p>
<p>Quite interestingly: If I run my program directly (not from VS2010), I do not see any problem with MKL 10.3.3.</p>
<p>I tried to create a small project to duplicate this unusual behavior, but I couldnt make it. Somehow, it is only happening with my full program when running with VS2010 + MKL 10.3.3. </p>
<p>Does anyone see a similar problem with use of VS2010 + MKL 10.3.3?</p>
<p>Bulent</p>
</p>
<p>Computer: Windows 7, 64-bit, 12 GB RAM, Intel Core i7, 960 @ 3.20 GHz</p>
<p>PARDISO Settings (C++):</p>
<p>Matrix type : 2 (sym. Pos. definite)</p>
<p>m_iparm[0] = 1;</p>
<p>m_iparm[1] = 3;</p>
<p>m_iparm[5] = 1;</p>
<p>m_iparm[7] = 2;</p>
<p>m_iparm[9] = 13;</p>
<p>m_iparm[10] = 1; </p>
<p>m_iparm[17] = -1;</p>
<p>m_iparm[18] = -1;</p>
<p>m_iparm[24] = 1;</p>
<p>In my project, I do a static linking and I compile my projects against these files:</p>
<p>mkl_intel_c.lib</p>
<p>mkl_intel_thread.lib</p>
<p>mkl_core.lib</p>
<p>libiomp5md.lib</p>
Fri, 23 Sep 11 11:31:29 -0700bulent281872PARDISO ERROR (error_num = 2578)
https://software.intel.com/en-us/forums/topic/282543
<p>HI<br />I am getting this error when I solve a system of Ax = B (#equations= 14791, #non-zeros in A= 167019.B is a matrix with 14791 rows and 799 columns. I use mtype = 2):</p>
<p>*** Error in PARDISO (numerical_factorization> error_num = 2578<br />PARDISO Internationalization error! Message 2578 is unknown.</p>
<p>This is happening in Phase =22, and PARDISO returns error = -4, indicating negative pivoting error, but I also see error 2578 when I print out more statistical information about the solver.</p>
<p>I saw error = -4 before but I never saw "PARDISO Internationalization" error.</p>
<p>Does anyone knowsmore about this?</p>
<p>Thanks <br />Bulent</p>
<p></p>
Mon, 01 Aug 11 17:32:47 -0700bulent282543Evaluating Intel Math Kernel Library
https://software.intel.com/en-us/forums/topic/285266
<p>Hello, wehave developed a software for structural analysis and design.The software will use finite element analsysis for the purpose of analysisng structures. We would like to check the feasibility of the Intel Math Kernel library for finite element calculation purposes.</p>
<p><strong>Type of matrices</strong></p>
<p>We deal with matrices which are sparse i.e. many of its eleemnts are zero. </p>
<p>Operations required to be carried out using the matrices include;</p>
<p>i. Matrix multiplication</p>
<p>ii. Matrix addition</p>
<p>iii. Matrix subtraction</p>
<p>iv. Solving a system of linear simultaneous equations</p>
<p>Both iterative and non-iterative techniques are required to solve the system of linear simultaneous equations.</p>
<p><strong>Exercise</strong></p>
<p>We have the FE analysis calculations written in MathCAD and now would like to use the Intel Math Kernel Library to perform the above operations.</p>
<p>We are interested in using the Intel Math Kernel library to carry out the above stated calculations of matrices .We would like to like to have a demonstration/ evaluation version of the Intel Math Kernel library and would like to know how to go about using the functions/ sub-routines. </p>
<p><strong>Specific questions;</strong></p>
<p>i. We deal with matrices which are sparse (for carrying out finite element analysis).As you see these matrices, have a large number of elements 0. Does the Intel Math Kernel library have functions/ sub-routines that will convert a full matrix to sparse form by squeezing out any zero elements? OR Will the user of the Intel Math Kernel Library have to provide such a matrix (i.e. a matrix wherein the 0s are already squeezed out)?</p>
<p>ii. As we have the finite element calculations written in MathCAD, how do we go about using the Intel Math Kernel Library to evaluate its efficiency/ feasibility during the evaluation period? Any suggestions for these will be gratefully appreciated.</p>
Wed, 09 Feb 11 08:29:49 -0800atee1234285266MKL 10.2 Update 7 versus MKL 10.3 Update 1
https://software.intel.com/en-us/forums/topic/286112
<p>HI<br />Just got an email today, saying v10.2 Update 7 is available for download. Last month,I also got an email for the availability of v10.3 Update 1. </p>
<p>Which version is the latest? we are planning to update our program to the latest MKL version, so I need to know.</p>
<p>Thx.<br />bulent</p>
<p></p>
Tue, 14 Dec 10 15:50:39 -0800bulent286112(symmetric) Sparse - Sparse Multiplication
https://software.intel.com/en-us/forums/topic/286854
<p>I am looking for a solution for C = A * B, where A and B are both symmetric sparse matrices (CSR with 3array variation), and the results is also a symmetric sparse matrix (CSR with 3array variation).</p>
<p>I found "mkl_csrmultcsr" but as I understand both A and B must include full storage of data (upper + lower diagonal terms) whereas I have A & B and only upper diagonals stored.</p>
<p>Thanks<br />Bulent</p>
Wed, 03 Nov 10 15:15:11 -0700bulent286854Use of iparam(30) for PARDISO
https://software.intel.com/en-us/forums/topic/293750
<p>Hi<br />I just started using v10.2 update 2 and found that for mytpe =2, PARDISO reports the number of equation when it detects a zero:</p>
<p><em>"iparm(30) - the number of equation where PARDISO detects zero or negative pivot for MTYPE=2 (real positive<br />definite matrix) and MTYPE=4 (complex and Hermitian positive definite matrices). If the solver detects a zero or negative pivot for these matrix types, the factorization is stopped, PARDISO returns immediately with an error (error = -4) and iparm(30) contains the number of the equation where the first zero or negative pivot is detected."</p>
<p></em>As far as I remember, PARDISO does an internal equation ordering for the given matrix. I just want to know the reported Equation # is the equation number associated with PARDISO's internal ordering or the equation number referring to my original matrix. The former really deos not help since we have no idea how the PARDISO internally orders equations. Thus, what I really need is the one that refers to the order given with the original matrix.</p>
<p>Bulent</p>
Fri, 16 Oct 09 10:36:14 -0700bulent293750