I'm working in C++ with Visual Studio 2010 and Intel ComposerXE-2011
In my program, I do a symbolic and a numeric factorization of a matrix A. After that, I do a lot of solve (Ax=b) with different b (Using intel mkl Pardiso).
Previously, my program was using Jet OLEDB 4.0 do to other job and all work find.
1 week ago, we change from Jet OLEDB 4.0 to ACE OLEDB 12.0. After this change, solve step of Ax=b (phase =33) take 3 times longer to execute. If I run my test many times, I saw some little difference in results of Ax=b. All were correct when using Jet OLEDB
Anyone else have this problem?
Anyone have a idea of what is happening? Conflict with OpenMP? Conflict with other libraries?
this is an unknown issue and MKL doesn't deal with OLEDB at all. Some performance degradation has been reported into the previous versions of MKL at the solving stage, but it was caused by another issue. Could you check the latest 2017 update 1 ( released a few days ago ) and check how it works?