Detailed information about MKL bug fixes.

[Ken_Thornton]

Is it possible to get detailed information about bug fixes, e.g. the circumstances in which they occur so that we can add tests to confirm that the bug is fixed in future releases?

For example, we recently have had an issue with cblas_gemv when calling it from an OpenMP loop when MKL 2018 threaded is used. This may have been fixed in MKLD-3786.

Our solution is to upgrade to MKL 2023 and link with MKL sequential.

You publish short statements about fixed bugs, but can we get details about the fixes through this forum or by contacting Intel support?

[Gennady_F_Intel]

Ken,

I am not sure, what kind of details you are expecting to see? 

This case has been fixed in MKL 2019.1 ( 6 years  ago ). The cause of the problem is the bug into PBLAS code and it was fixed.

--Gennady

[Ken_Thornton]

Is the one-line bug statement all we can expect?

In our specific instance, MKL failed only in a threaded MKL build (sequential build worked). Your bug statement just says "returns incorrect results under specific conditions". Obviously, would be nice to know what the "incorrect results are like" (in our case the returned return was 2x the correct result) and the "specific conditions" (in our case it was only with multithreaded MKL with > 8 threads).

[Ken_Thornton]

Can you tell me whether the MKLD-3786 bug was present in the sequential library? We only saw the issue in the MKL threaded library. That kind of information for other bugs would be very useful.