Recent posts
https://software.intel.com/en-us/recent/952643
enSegmentation fault trying to use mkl_dcscmm from a Matlab mex file
https://software.intel.com/en-us/forums/intel-math-kernel-library/topic/472320
<p>I'm trying to use mkl_dcscmm from inside a Matlab mex file (on 64-bit Linux, have tried Matlab versions 2009b and 2012a), but having trouble avoiding segfaults. I've tried static and dynamic linking to MKL, sequential vs gomp vs iomp, LP64 vs ILP64, tried changing BLAS_VERSION so Matlab would use the more-complete MKL library I'm trying to use instead of its internal one, nothing seems to work. I've attached the source for my mex file, and the mexopts.sh (change the .dat extension to .sh) with the extra compiler and linker flags I used. I was compiling the mex file and running a test from inside Matlab as follows:</p>
<p>mex -v -largeArrayDims -g mkl_sfmult.cpp<br />testA = sprand(10,5,0.5);<br />testB = rand(5,8);<br />testC = mkl_sfmult(testA, testB)</p>
<p>When I start Matlab in gdb, it appears the segmentation fault occurs in mkl_spblas_lp64_mc_dcsr1tg__f__mmout_par (), in mkl_spblas_lp64_dcsr1tg__f__mmout_par (), in mkl_sfmult.mexa64, otherwise just question marks in the stack trace.</p>
<p>I've tried making a standalone executable where I hard-code the data for a few test matrices, and I get the right result with no segfaults outside of Matlab. I'd really prefer to have a Matlab interface if possible, without needing to save the matrix data to disk or generating new source code for each problem instance. Maybe I'll try using loadlibrary instead? I've been searching through the literature and available libraries and MKL is the only multi-threaded implementation of the sparse matrix times dense matrix kernel that I can find.</p>
<p>Any ideas, recommendations, more info you'd need to figure out what's going on here? Thanks.</p>
Fri, 06 Sep 13 06:21:47 -0700Tony K.472320