Hi
I'm experiencing an issue with Intel MKL in a very specific setting.
The lapack solver for banded matrices is very slow on my ThinkPad P14s with an Intel Core Ultra 9 285H when using Intel MKL. We happen to have two laptops with the same configuration and it's the same on both of them. However, it's much faster on other computers with different CPU's (Intel and AMD), it's faster on WSL (Windows Subsystem for Linux), it's faster if I use OpenBLAS, and it's faster if I disable AVX2.
I've attached a small benchmark which takes approximately 2 to 2.5 seconds with SSE4.2. On other computers we've measured various values between 2 and 4 seconds. The fastest was WSL on my computer which was slightly faster than 2 seconds.
The same benchmark with AVX2 or AVX2 E1 takes more than 20 seconds on my computer.
The vectorization can be set in main.cpp on lines 24-27.
I'd be glad if anyone could tell me whether I'm doing something wrong, if this is to be expected for some reason, or if this is and issue in Intel MKL.
Many thanks in advance!
