opencollab
MKL ERROR: Parameter 4 was incorrect on entry to DLASCL
Expected behavior
Should runs at least a few iterations of the Arnoldi process even if it won't converged.
Actual behavior
Say 0 iteraton were done, print a MKL error saying some parameter is incorrect in a LAPACK routine use for computing a ratio times a matrix (cf https://software.intel.com/content/www/us/en/develop/documentation/onemkl-developer-reference-c/top/lapack-routines/lapack-auxiliary-routines/lascl.html).
Where/how to reproduce the problem
Difficult to reproduce the problem since my data is generated on the fly on a cluster, so you'll need access to the cluster AND the code. But I found online some similar problem with data that "should" reproduce the problem : https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/MKL-ERROR-Parameter-4-was-incorrect-on-entry-to-DLASCL/td-p/907349 http://icl.cs.utk.edu/lapack-forum/viewtopic.php?f=2&t=529&p=1766&hilit=dgelsd#p1766
- arpack-ng: it's arpack-ng 3.7.0
- OS: it's on CORI from NERSC, which a Cray system and I run it on the haswell nodes. You can find more info here : https://docs.nersc.gov/systems/cori/
Error message
I get INFO=-9999, and my matrix is a Toeplitz block diagonal matrix of size 19 198 080. So the problem could be that it's too big... And it prints this : MKL ERROR: Parameter 4 was incorrect on entry to DLASCL.
Do you have warnings / errors at link time ?
Do you link with MKL using build sys (cmake, autotools), or, "by hand" ? If by hand, did you follow exactly intel advisor : https://software.intel.com/content/www/us/en/develop/tools/oneapi/components/onemkl/link-line-advisor.html ?
Try with static linking (if silent [runtime] errors with dynamic linking : they may show up at link time) and/or use something like gcc -v -Wl,--verbose -Wl,--fatal-warnings (or equivalent - check in man) to make sure link is clean.
@sylvestre : just noticed CI is KO on master e3bdd3b8db727a10551c06959e45e866a6b8d8fc
