topic OK, so I tried 2017 (with no in Intel® oneAPI Math Kernel Library

BUG in P?GEMM remains

Ron_Cohen — Mon, 19 Sep 2016 08:51:37 GMT

The bug in Scalapack P?GEMM seems not to be fixed even in the 2017 beta3 library , although it has apparently been known since at least June . It seems to be present since mkl 11.3 . It works fine in mkl 11.2 . See here for the previous discussion.

Ron, I don't know what MKL

Gennady_F_Intel — Mon, 19 Sep 2016 15:21:55 GMT

Ron, I don't know what MKL 2017 beta3 is... Actually we had only MKL 2017 beta 1 :). Nevertheless, checking the internal database, I see this issue is targeted to MKL 2017 ( released Sep 6th 2016). Could you please check your case with this version and let us know how it work? wbr, Gennady

Yes I tried that and it did

Ron_Cohen — Mon, 19 Sep 2016 19:38:19 GMT

Yes I tried that and it did not work. It was the same as 11.3 . On the superMUC machine here in Germany there is 2017 beta1 beta2 and beta3 . It doesn't work for any of those.

here are the dates that I

Ron_Cohen — Mon, 19 Sep 2016 19:52:27 GMT

here are the dates that I find:

Apr 6 00:57 studio2017_beta1

Jun 3 12:46 studio2017_beta2

Jul 14 12:28 studio2017_beta3

Sep 6 20:07 studio2017

It sounds like I should be using studio2017 although no module points to it I will try it.

Thanks!

OK, so I tried 2017 (with no

Ron_Cohen — Mon, 19 Sep 2016 21:03:29 GMT

OK, so I tried 2017 (with no beta) and it does work. It is slower than 11.2 for my test case, but it could be because my test case is too small. So it seems like it is fixed in 2017. Thank you!

quote - "It is slower than 11

Gennady_F_Intel — Tue, 20 Sep 2016 02:25:19 GMT

quote - "It is slower than 11.2 for my test case..." . Ron, in the case if p?gemm from 2017 version is slower vs 11.2 for reasonable problem sizes, please let us know. thanks, Gennady