BLAS LEVEL 2 routines for Symmetric AR(1) correlation matrices

[may_ka]

Hi,

I am wondering whether MKL has any provision for BLAS LEVEL 2 routines on first-order autoregressive matrices .

All possible values in that matrix can be represented in a single vector and the matrix can be recovered from shift-in shift-out operations on that vector.

BLAS LEVEL 2 operations on such matrix can be either carried out "brute force" by constructing that matrix and using standard "symv".

However, it is also possible by operating only on the above vector using "axpy" and a shift-in shift-out operation on x, which is maybe much more efficient.

I am wondering whether MKL has the above available?

Cheers

[ShanmukhS_Intel]

Hi Karl.

Thanks for posting on Intel Communities.

I am wondering whether MKL has the above available?

>>Could you please refer to the below link and let us know if it's useful or else could you please provide some additional information, so that it helps us in triaging the issue better?

https://www.intel.com/content/www/us/en/develop/documentation/onemkl-developer-reference-c/top/blas-and-sparse-blas-routines/blas-routines/blas-level-2-routines.html

Best Regards,

Shanmukh.SS

[may_ka]

Hi,

there is nothing in the link.

As explained in the link I provided, the matrix n x n in question is symmetric, and all possible values an be stored in a vector of size n. An example would be

a b c d
b a b c
c b a b
d c b a

where it is sufficient to store vector

a b c d

Therefore, for multiplying a vector with such matrix it is sufficient to know the first matrix row, that is, there is no necessity to ever construct the entire matrix. I have certainly home brew routines which do this, but I am wondering whether mkl maybe has a better solution which I have overlooked so far.

[ShanmukhS_Intel]

Hi Karl,

Thanks for elaborating on your requirement.

We are discussing your issue internally. We will get back to you soon with an update.

Best Regards,

Shanmukh.SS

[ShanmukhS_Intel]

Hi Karl,

This is to kindly inform you that oneMKL BLAS does not support this type of structure.

Could you please let us know in what scenarios/applications you work with this kind of matrix?

Best Regards,

Shanmukh.SS

[ShanmukhS_Intel]

Hi Karl,

A gentle reminder:

As we have not heard back from you, Could you please get back to us with your requirement which helps us assist you further?

Best Regards,

Shanmukh.SS

[ShanmukhS_Intel]

Hi Karl,

We have not heard back from you. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.

Best Regards,

Shanmukh.SS