Inefficient algorithms work with packed matrices in the Intel MKL

[yuriisig]

Intel MKL has long been aware of this (see paragraphs 1.2 and 3.5 http://www.thesa-store.com/products/english.php ). The new versions of Intel MKL results are somewhat better, but good to them is still very far away. Why is this happening? When using large matrices is very important.

[Shane_S_Intel]

Hi Yurii, it would be helpful if you could clarify yourrequest and do your comparisons against a more recent version of MKL. The current version available is 10.2.5; your comparison uses version 8.0 which shipped 4 or more years ago.Also, are you running sequentially or in parallel?If there is a performance issue in an existing MKLfunction, we will add it to our to-do list and prioritizeaddressing itaccordingly. Ifthis is intended as an MKL feature request,please let us know - we can add it to our system and prioritize working on it among the many other feature requests we receive.

[yuriisig]

My algorithms DSPTRD and DOPMTR faster appropriate algorithms DSYTRD Intel MKL 10.2.5 and DORMTR Intel MKL 10.2.5. See also http://software.intel.com/en-us/forums/showthread.php?t=73653&o=d&s=lr

[Shane_S_Intel]

Hi Yurii, you appear to be asking us to further optimize the MKL functions DSYTRD and DORMTR. We can certainly look into if thereare performanceissues hereandif so, consider optimizationsfor a future MKL release. Weappreciate you taking the time to bring thistopic to our attention.

[yuriisig]

I do not think that Intel will cope with the Intel MKL DSYTRD and DORMTR, andIntel MKL DSPTRD and DOPMTR they can improve.