topic 引文：Voronin, Kirill (Intel) 写道 in Intel® oneAPI Math Kernel Library

is there any MKL function can outputs L matrix of cholesky decomposition

lin__ziije — Tue, 03 Dec 2019 06:53:46 GMT

I want to solve some linear equations（Ax=b） with same matrix A and different vector b. So i decided to decompose matrix A with cholesky decomposition to get matrix L (A= LT*L).

Is there any MKL function can outputs L matrix of cholesky decomposition, i have tried some other library, but most of them can not directly output matrix L.

Hello,

Kirill_V_Intel — Wed, 04 Dec 2019 00:15:30 GMT

Hello,

Is the matrix sparse or dense? Intel MKL provides a way in both cases to solve one system with multiple right-hand sides but functionality is different. If you just want to solve with multiple right-hand sides, you don't actually need the matrix L. So, is there something else which you didn't mention in your description?

Thanks,
Kirill

引文：Voronin, Kirill (Intel) 写道

lin__ziije — Wed, 04 Dec 2019 04:06:19 GMT

Voronin, Kirill (Intel) wrote:
Hello,
Is the matrix sparse or dense? Intel MKL provides a way in both cases to solve one system with multiple right-hand sides but functionality is different. If you just want to solve with multiple right-hand sides, you don't actually need the matrix L. So, is there something else which you didn't mention in your description?
Thanks,
Kirill

thanks for your answer. Matrix A is a 5000x5000 matrix with about 60,000 (0.25%) none zeros elements. I actually need the matrix L for follow-up work.

so， MKL can not output matrix L driectly?

I see. For such a tiny matrix

Kirill_V_Intel — Wed, 04 Dec 2019 04:25:49 GMT

I see. For such a tiny matrix and such sparsity I believe there is no need to use sparse matrices. Have a look at https://software.intel.com/en-us/onemkl-developer-reference-c-getrf from LAPACK.

Best,
Kirill

引文：Voronin, Kirill (Intel) 写道

lin__ziije — Wed, 04 Dec 2019 06:41:25 GMT

Voronin, Kirill (Intel) wrote:
I see. For such a tiny matrix and such sparsity I believe there is no need to use sparse matrices. Have a look at https://software.intel.com/en-us/onemkl-developer-reference-c-getrf from LAPACK.
Best,
Kirill

I will try it, by the way ,have you ever try to solve problem like mine, do you know how long will take to do cholesky or LU decomposition. my program is run on a server with 2 Intel(R) Xeon(R) CPU E5-2609 v3 @ 1.90GHz.

you may take a look at the

Gennady_F_Intel — Fri, 06 Dec 2019 04:15:03 GMT

you may take a look at the published perf results for some of mkl's factorizations routines from intel mkl product page - https://software.intel.com/en-us/mkl/features/benchmarks